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 sz
= __le32_to_cpu(dev
->size_high
);
1487 sz
+= __le32_to_cpu(dev
->size_low
);
1488 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1489 human_size(sz
* 512));
1490 sz
= blocks_per_member(map
);
1491 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1492 human_size(sz
* 512));
1493 printf(" Sector Offset : %llu\n",
1495 printf(" Num Stripes : %llu\n",
1496 num_data_stripes(map
));
1497 printf(" Chunk Size : %u KiB",
1498 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1500 printf(" <-- %u KiB",
1501 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1503 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1504 printf(" Migrate State : ");
1505 if (dev
->vol
.migr_state
) {
1506 if (migr_type(dev
) == MIGR_INIT
)
1507 printf("initialize\n");
1508 else if (migr_type(dev
) == MIGR_REBUILD
)
1509 printf("rebuild\n");
1510 else if (migr_type(dev
) == MIGR_VERIFY
)
1512 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1513 printf("general migration\n");
1514 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1515 printf("state change\n");
1516 else if (migr_type(dev
) == MIGR_REPAIR
)
1519 printf("<unknown:%d>\n", migr_type(dev
));
1522 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1523 if (dev
->vol
.migr_state
) {
1524 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1526 printf(" <-- %s", map_state_str
[map
->map_state
]);
1527 printf("\n Checkpoint : %u ",
1528 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1529 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1532 printf("(%llu)", (unsigned long long)
1533 blocks_per_migr_unit(super
, dev
));
1536 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1538 printf(" RWH Policy : ");
1539 if (dev
->rwh_policy
== RWH_OFF
)
1541 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1542 printf("PPL distributed\n");
1543 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1544 printf("PPL journaling drive\n");
1546 printf("<unknown:%d>\n", dev
->rwh_policy
);
1549 static void print_imsm_disk(struct imsm_disk
*disk
,
1552 unsigned int sector_size
) {
1553 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1556 if (index
< -1 || !disk
)
1560 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1562 printf(" Disk%02d Serial : %s\n", index
, str
);
1564 printf(" Disk Serial : %s\n", str
);
1565 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1566 is_configured(disk
) ? " active" : "",
1567 is_failed(disk
) ? " failed" : "",
1568 is_journal(disk
) ? " journal" : "");
1569 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1570 sz
= total_blocks(disk
) - reserved
;
1571 printf(" Usable Size : %llu%s\n",
1572 (unsigned long long)sz
* 512 / sector_size
,
1573 human_size(sz
* 512));
1576 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1578 struct migr_record
*migr_rec
= super
->migr_rec
;
1580 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1581 migr_rec
->ckpt_area_pba
/= IMSM_4K_DIV
;
1582 migr_rec
->dest_1st_member_lba
/= IMSM_4K_DIV
;
1583 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1584 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1585 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1586 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1589 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1591 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1594 void convert_to_4k(struct intel_super
*super
)
1596 struct imsm_super
*mpb
= super
->anchor
;
1597 struct imsm_disk
*disk
;
1599 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1601 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1602 disk
= __get_imsm_disk(mpb
, i
);
1604 convert_to_4k_imsm_disk(disk
);
1606 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1607 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1608 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1610 split_ull((join_u32(dev
->size_low
, dev
->size_high
)/IMSM_4K_DIV
),
1611 &dev
->size_low
, &dev
->size_high
);
1612 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1615 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1616 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1617 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1619 if (dev
->vol
.migr_state
) {
1621 map
= get_imsm_map(dev
, MAP_1
);
1622 set_blocks_per_member(map
,
1623 blocks_per_member(map
)/IMSM_4K_DIV
);
1624 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1625 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1629 struct bbm_log
*log
= (void *)mpb
+
1630 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1633 for (i
= 0; i
< log
->entry_count
; i
++) {
1634 struct bbm_log_entry
*entry
=
1635 &log
->marked_block_entries
[i
];
1637 __u8 count
= entry
->marked_count
+ 1;
1638 unsigned long long sector
=
1639 __le48_to_cpu(&entry
->defective_block_start
);
1641 entry
->defective_block_start
=
1642 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1643 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1647 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1650 void examine_migr_rec_imsm(struct intel_super
*super
)
1652 struct migr_record
*migr_rec
= super
->migr_rec
;
1653 struct imsm_super
*mpb
= super
->anchor
;
1656 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1657 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1658 struct imsm_map
*map
;
1661 if (is_gen_migration(dev
) == 0)
1664 printf("\nMigration Record Information:");
1666 /* first map under migration */
1667 map
= get_imsm_map(dev
, MAP_0
);
1669 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1670 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1671 printf(" Empty\n ");
1672 printf("Examine one of first two disks in array\n");
1675 printf("\n Status : ");
1676 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1679 printf("Contains Data\n");
1680 printf(" Current Unit : %u\n",
1681 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1682 printf(" Family : %u\n",
1683 __le32_to_cpu(migr_rec
->family_num
));
1684 printf(" Ascending : %u\n",
1685 __le32_to_cpu(migr_rec
->ascending_migr
));
1686 printf(" Blocks Per Unit : %u\n",
1687 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1688 printf(" Dest. Depth Per Unit : %u\n",
1689 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1690 printf(" Checkpoint Area pba : %u\n",
1691 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1692 printf(" First member lba : %u\n",
1693 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1694 printf(" Total Number of Units : %u\n",
1695 __le32_to_cpu(migr_rec
->num_migr_units
));
1696 printf(" Size of volume : %u\n",
1697 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1698 printf(" Expansion space for LBA64 : %u\n",
1699 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1700 printf(" Record was read from : %u\n",
1701 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1707 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1709 struct migr_record
*migr_rec
= super
->migr_rec
;
1711 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1712 migr_rec
->ckpt_area_pba
*= IMSM_4K_DIV
;
1713 migr_rec
->dest_1st_member_lba
*= IMSM_4K_DIV
;
1714 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1715 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1716 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1717 &migr_rec
->post_migr_vol_cap
,
1718 &migr_rec
->post_migr_vol_cap_hi
);
1721 void convert_from_4k(struct intel_super
*super
)
1723 struct imsm_super
*mpb
= super
->anchor
;
1724 struct imsm_disk
*disk
;
1726 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1728 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1729 disk
= __get_imsm_disk(mpb
, i
);
1731 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1734 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1735 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1736 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1738 split_ull((join_u32(dev
->size_low
, dev
->size_high
)*IMSM_4K_DIV
),
1739 &dev
->size_low
, &dev
->size_high
);
1740 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1743 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1744 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1745 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1747 if (dev
->vol
.migr_state
) {
1749 map
= get_imsm_map(dev
, MAP_1
);
1750 set_blocks_per_member(map
,
1751 blocks_per_member(map
)*IMSM_4K_DIV
);
1752 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1753 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1757 struct bbm_log
*log
= (void *)mpb
+
1758 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1761 for (i
= 0; i
< log
->entry_count
; i
++) {
1762 struct bbm_log_entry
*entry
=
1763 &log
->marked_block_entries
[i
];
1765 __u8 count
= entry
->marked_count
+ 1;
1766 unsigned long long sector
=
1767 __le48_to_cpu(&entry
->defective_block_start
);
1769 entry
->defective_block_start
=
1770 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1771 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1775 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1778 /*******************************************************************************
1779 * function: imsm_check_attributes
1780 * Description: Function checks if features represented by attributes flags
1781 * are supported by mdadm.
1783 * attributes - Attributes read from metadata
1785 * 0 - passed attributes contains unsupported features flags
1786 * 1 - all features are supported
1787 ******************************************************************************/
1788 static int imsm_check_attributes(__u32 attributes
)
1791 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1793 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1795 not_supported
&= attributes
;
1796 if (not_supported
) {
1797 pr_err("(IMSM): Unsupported attributes : %x\n",
1798 (unsigned)__le32_to_cpu(not_supported
));
1799 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1800 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1801 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1803 if (not_supported
& MPB_ATTRIB_2TB
) {
1804 dprintf("\t\tMPB_ATTRIB_2TB\n");
1805 not_supported
^= MPB_ATTRIB_2TB
;
1807 if (not_supported
& MPB_ATTRIB_RAID0
) {
1808 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1809 not_supported
^= MPB_ATTRIB_RAID0
;
1811 if (not_supported
& MPB_ATTRIB_RAID1
) {
1812 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1813 not_supported
^= MPB_ATTRIB_RAID1
;
1815 if (not_supported
& MPB_ATTRIB_RAID10
) {
1816 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1817 not_supported
^= MPB_ATTRIB_RAID10
;
1819 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1820 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1821 not_supported
^= MPB_ATTRIB_RAID1E
;
1823 if (not_supported
& MPB_ATTRIB_RAID5
) {
1824 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1825 not_supported
^= MPB_ATTRIB_RAID5
;
1827 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1828 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1829 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1831 if (not_supported
& MPB_ATTRIB_BBM
) {
1832 dprintf("\t\tMPB_ATTRIB_BBM\n");
1833 not_supported
^= MPB_ATTRIB_BBM
;
1835 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1836 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1837 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1839 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1840 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1841 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1843 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1844 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1845 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1847 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1848 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1849 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1851 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1852 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1853 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1857 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1865 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1867 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1869 struct intel_super
*super
= st
->sb
;
1870 struct imsm_super
*mpb
= super
->anchor
;
1871 char str
[MAX_SIGNATURE_LENGTH
];
1876 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1879 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
1880 str
[MPB_SIG_LEN
-1] = '\0';
1881 printf(" Magic : %s\n", str
);
1882 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1883 printf(" Version : %s\n", get_imsm_version(mpb
));
1884 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1885 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1886 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1887 printf(" Attributes : ");
1888 if (imsm_check_attributes(mpb
->attributes
))
1889 printf("All supported\n");
1891 printf("not supported\n");
1892 getinfo_super_imsm(st
, &info
, NULL
);
1893 fname_from_uuid(st
, &info
, nbuf
, ':');
1894 printf(" UUID : %s\n", nbuf
+ 5);
1895 sum
= __le32_to_cpu(mpb
->check_sum
);
1896 printf(" Checksum : %08x %s\n", sum
,
1897 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1898 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
1899 printf(" Disks : %d\n", mpb
->num_disks
);
1900 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1901 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
1902 super
->disks
->index
, reserved
, super
->sector_size
);
1903 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
1904 struct bbm_log
*log
= super
->bbm_log
;
1907 printf("Bad Block Management Log:\n");
1908 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1909 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1910 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1912 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1914 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1916 super
->current_vol
= i
;
1917 getinfo_super_imsm(st
, &info
, NULL
);
1918 fname_from_uuid(st
, &info
, nbuf
, ':');
1919 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1921 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1922 if (i
== super
->disks
->index
)
1924 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
1925 super
->sector_size
);
1928 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1929 if (dl
->index
== -1)
1930 print_imsm_disk(&dl
->disk
, -1, reserved
,
1931 super
->sector_size
);
1933 examine_migr_rec_imsm(super
);
1936 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1938 /* We just write a generic IMSM ARRAY entry */
1941 struct intel_super
*super
= st
->sb
;
1943 if (!super
->anchor
->num_raid_devs
) {
1944 printf("ARRAY metadata=imsm\n");
1948 getinfo_super_imsm(st
, &info
, NULL
);
1949 fname_from_uuid(st
, &info
, nbuf
, ':');
1950 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1953 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1955 /* We just write a generic IMSM ARRAY entry */
1959 struct intel_super
*super
= st
->sb
;
1962 if (!super
->anchor
->num_raid_devs
)
1965 getinfo_super_imsm(st
, &info
, NULL
);
1966 fname_from_uuid(st
, &info
, nbuf
, ':');
1967 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1968 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1970 super
->current_vol
= i
;
1971 getinfo_super_imsm(st
, &info
, NULL
);
1972 fname_from_uuid(st
, &info
, nbuf1
, ':');
1973 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1974 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1978 static void export_examine_super_imsm(struct supertype
*st
)
1980 struct intel_super
*super
= st
->sb
;
1981 struct imsm_super
*mpb
= super
->anchor
;
1985 getinfo_super_imsm(st
, &info
, NULL
);
1986 fname_from_uuid(st
, &info
, nbuf
, ':');
1987 printf("MD_METADATA=imsm\n");
1988 printf("MD_LEVEL=container\n");
1989 printf("MD_UUID=%s\n", nbuf
+5);
1990 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1993 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1995 /* The second last sector of the device contains
1996 * the "struct imsm_super" metadata.
1997 * This contains mpb_size which is the size in bytes of the
1998 * extended metadata. This is located immediately before
2000 * We want to read all that, plus the last sector which
2001 * may contain a migration record, and write it all
2005 unsigned long long dsize
, offset
;
2007 struct imsm_super
*sb
;
2008 struct intel_super
*super
= st
->sb
;
2009 unsigned int sector_size
= super
->sector_size
;
2010 unsigned int written
= 0;
2012 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
2015 if (!get_dev_size(from
, NULL
, &dsize
))
2018 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
2020 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
2023 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2026 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2027 offset
= dsize
- sectors
* sector_size
;
2028 if (lseek64(from
, offset
, 0) < 0 ||
2029 lseek64(to
, offset
, 0) < 0)
2031 while (written
< sectors
* sector_size
) {
2032 int n
= sectors
*sector_size
- written
;
2035 if (read(from
, buf
, n
) != n
)
2037 if (write(to
, buf
, n
) != n
)
2048 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
2053 getinfo_super_imsm(st
, &info
, NULL
);
2054 fname_from_uuid(st
, &info
, nbuf
, ':');
2055 printf("\n UUID : %s\n", nbuf
+ 5);
2058 static void brief_detail_super_imsm(struct supertype
*st
)
2062 getinfo_super_imsm(st
, &info
, NULL
);
2063 fname_from_uuid(st
, &info
, nbuf
, ':');
2064 printf(" UUID=%s", nbuf
+ 5);
2067 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2068 static void fd2devname(int fd
, char *name
);
2070 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2072 /* dump an unsorted list of devices attached to AHCI Intel storage
2073 * controller, as well as non-connected ports
2075 int hba_len
= strlen(hba_path
) + 1;
2080 unsigned long port_mask
= (1 << port_count
) - 1;
2082 if (port_count
> (int)sizeof(port_mask
) * 8) {
2084 pr_err("port_count %d out of range\n", port_count
);
2088 /* scroll through /sys/dev/block looking for devices attached to
2091 dir
= opendir("/sys/dev/block");
2095 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2106 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2108 path
= devt_to_devpath(makedev(major
, minor
));
2111 if (!path_attached_to_hba(path
, hba_path
)) {
2117 /* retrieve the scsi device type */
2118 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2120 pr_err("failed to allocate 'device'\n");
2124 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2125 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2127 pr_err("failed to read device type for %s\n",
2133 type
= strtoul(buf
, NULL
, 10);
2135 /* if it's not a disk print the vendor and model */
2136 if (!(type
== 0 || type
== 7 || type
== 14)) {
2139 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2140 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2141 strncpy(vendor
, buf
, sizeof(vendor
));
2142 vendor
[sizeof(vendor
) - 1] = '\0';
2143 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2144 while (isspace(*c
) || *c
== '\0')
2148 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2149 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2150 strncpy(model
, buf
, sizeof(model
));
2151 model
[sizeof(model
) - 1] = '\0';
2152 c
= (char *) &model
[sizeof(model
) - 1];
2153 while (isspace(*c
) || *c
== '\0')
2157 if (vendor
[0] && model
[0])
2158 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2160 switch (type
) { /* numbers from hald/linux/device.c */
2161 case 1: sprintf(buf
, "tape"); break;
2162 case 2: sprintf(buf
, "printer"); break;
2163 case 3: sprintf(buf
, "processor"); break;
2165 case 5: sprintf(buf
, "cdrom"); break;
2166 case 6: sprintf(buf
, "scanner"); break;
2167 case 8: sprintf(buf
, "media_changer"); break;
2168 case 9: sprintf(buf
, "comm"); break;
2169 case 12: sprintf(buf
, "raid"); break;
2170 default: sprintf(buf
, "unknown");
2176 /* chop device path to 'host%d' and calculate the port number */
2177 c
= strchr(&path
[hba_len
], '/');
2180 pr_err("%s - invalid path name\n", path
+ hba_len
);
2185 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2186 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2190 *c
= '/'; /* repair the full string */
2191 pr_err("failed to determine port number for %s\n",
2198 /* mark this port as used */
2199 port_mask
&= ~(1 << port
);
2201 /* print out the device information */
2203 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2207 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2209 printf(" Port%d : - disk info unavailable -\n", port
);
2211 fd2devname(fd
, buf
);
2212 printf(" Port%d : %s", port
, buf
);
2213 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2214 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2229 for (i
= 0; i
< port_count
; i
++)
2230 if (port_mask
& (1 << i
))
2231 printf(" Port%d : - no device attached -\n", i
);
2237 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2245 if (hba
->type
!= SYS_DEV_VMD
)
2248 /* scroll through /sys/dev/block looking for devices attached to
2251 dir
= opendir("/sys/bus/pci/drivers/nvme");
2255 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2258 /* is 'ent' a device? check that the 'subsystem' link exists and
2259 * that its target matches 'bus'
2261 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2263 n
= readlink(path
, link
, sizeof(link
));
2264 if (n
< 0 || n
>= (int)sizeof(link
))
2267 c
= strrchr(link
, '/');
2270 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2273 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2275 rp
= realpath(path
, NULL
);
2279 if (path_attached_to_hba(rp
, hba
->path
)) {
2280 printf(" NVMe under VMD : %s\n", rp
);
2289 static void print_found_intel_controllers(struct sys_dev
*elem
)
2291 for (; elem
; elem
= elem
->next
) {
2292 pr_err("found Intel(R) ");
2293 if (elem
->type
== SYS_DEV_SATA
)
2294 fprintf(stderr
, "SATA ");
2295 else if (elem
->type
== SYS_DEV_SAS
)
2296 fprintf(stderr
, "SAS ");
2297 else if (elem
->type
== SYS_DEV_NVME
)
2298 fprintf(stderr
, "NVMe ");
2300 if (elem
->type
== SYS_DEV_VMD
)
2301 fprintf(stderr
, "VMD domain");
2303 fprintf(stderr
, "RAID controller");
2306 fprintf(stderr
, " at %s", elem
->pci_id
);
2307 fprintf(stderr
, ".\n");
2312 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2319 if ((dir
= opendir(hba_path
)) == NULL
)
2322 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2325 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2326 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2328 if (*port_count
== 0)
2330 else if (host
< host_base
)
2333 if (host
+ 1 > *port_count
+ host_base
)
2334 *port_count
= host
+ 1 - host_base
;
2340 static void print_imsm_capability(const struct imsm_orom
*orom
)
2342 printf(" Platform : Intel(R) ");
2343 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2344 printf("Matrix Storage Manager\n");
2346 printf("Rapid Storage Technology%s\n",
2347 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2348 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2349 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2350 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2351 printf(" RAID Levels :%s%s%s%s%s\n",
2352 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2353 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2354 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2355 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2356 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2357 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2358 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2359 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2360 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2361 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2362 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2363 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2364 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2365 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2366 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2367 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2368 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2369 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2370 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2371 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2372 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2373 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2374 printf(" 2TB volumes :%s supported\n",
2375 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2376 printf(" 2TB disks :%s supported\n",
2377 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2378 printf(" Max Disks : %d\n", orom
->tds
);
2379 printf(" Max Volumes : %d per array, %d per %s\n",
2380 orom
->vpa
, orom
->vphba
,
2381 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2385 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2387 printf("MD_FIRMWARE_TYPE=imsm\n");
2388 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2389 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2390 orom
->hotfix_ver
, orom
->build
);
2391 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2392 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2393 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2394 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2395 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2396 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2397 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2398 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2399 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2400 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2401 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2402 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2403 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2404 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2405 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2406 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2407 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2408 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2409 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2410 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2411 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2412 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2413 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2414 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2415 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2416 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2417 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2418 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2421 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2423 /* There are two components to imsm platform support, the ahci SATA
2424 * controller and the option-rom. To find the SATA controller we
2425 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2426 * controller with the Intel vendor id is present. This approach
2427 * allows mdadm to leverage the kernel's ahci detection logic, with the
2428 * caveat that if ahci.ko is not loaded mdadm will not be able to
2429 * detect platform raid capabilities. The option-rom resides in a
2430 * platform "Adapter ROM". We scan for its signature to retrieve the
2431 * platform capabilities. If raid support is disabled in the BIOS the
2432 * option-rom capability structure will not be available.
2434 struct sys_dev
*list
, *hba
;
2439 if (enumerate_only
) {
2440 if (check_env("IMSM_NO_PLATFORM"))
2442 list
= find_intel_devices();
2445 for (hba
= list
; hba
; hba
= hba
->next
) {
2446 if (find_imsm_capability(hba
)) {
2456 list
= find_intel_devices();
2459 pr_err("no active Intel(R) RAID controller found.\n");
2461 } else if (verbose
> 0)
2462 print_found_intel_controllers(list
);
2464 for (hba
= list
; hba
; hba
= hba
->next
) {
2465 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2467 if (!find_imsm_capability(hba
)) {
2469 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2470 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2471 get_sys_dev_type(hba
->type
));
2477 if (controller_path
&& result
== 1) {
2478 pr_err("no active Intel(R) RAID controller found under %s\n",
2483 const struct orom_entry
*entry
;
2485 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2486 if (entry
->type
== SYS_DEV_VMD
) {
2487 print_imsm_capability(&entry
->orom
);
2488 printf(" 3rd party NVMe :%s supported\n",
2489 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2490 for (hba
= list
; hba
; hba
= hba
->next
) {
2491 if (hba
->type
== SYS_DEV_VMD
) {
2493 printf(" I/O Controller : %s (%s)\n",
2494 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2495 if (print_vmd_attached_devs(hba
)) {
2497 pr_err("failed to get devices attached to VMD domain.\n");
2506 print_imsm_capability(&entry
->orom
);
2507 if (entry
->type
== SYS_DEV_NVME
) {
2508 for (hba
= list
; hba
; hba
= hba
->next
) {
2509 if (hba
->type
== SYS_DEV_NVME
)
2510 printf(" NVMe Device : %s\n", hba
->path
);
2516 struct devid_list
*devid
;
2517 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2518 hba
= device_by_id(devid
->devid
);
2522 printf(" I/O Controller : %s (%s)\n",
2523 hba
->path
, get_sys_dev_type(hba
->type
));
2524 if (hba
->type
== SYS_DEV_SATA
) {
2525 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2526 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2528 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2539 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2541 struct sys_dev
*list
, *hba
;
2544 list
= find_intel_devices();
2547 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2552 for (hba
= list
; hba
; hba
= hba
->next
) {
2553 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2555 if (!find_imsm_capability(hba
) && verbose
> 0) {
2557 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2558 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2564 const struct orom_entry
*entry
;
2566 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2567 if (entry
->type
== SYS_DEV_VMD
) {
2568 for (hba
= list
; hba
; hba
= hba
->next
)
2569 print_imsm_capability_export(&entry
->orom
);
2572 print_imsm_capability_export(&entry
->orom
);
2578 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2580 /* the imsm metadata format does not specify any host
2581 * identification information. We return -1 since we can never
2582 * confirm nor deny whether a given array is "meant" for this
2583 * host. We rely on compare_super and the 'family_num' fields to
2584 * exclude member disks that do not belong, and we rely on
2585 * mdadm.conf to specify the arrays that should be assembled.
2586 * Auto-assembly may still pick up "foreign" arrays.
2592 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2594 /* The uuid returned here is used for:
2595 * uuid to put into bitmap file (Create, Grow)
2596 * uuid for backup header when saving critical section (Grow)
2597 * comparing uuids when re-adding a device into an array
2598 * In these cases the uuid required is that of the data-array,
2599 * not the device-set.
2600 * uuid to recognise same set when adding a missing device back
2601 * to an array. This is a uuid for the device-set.
2603 * For each of these we can make do with a truncated
2604 * or hashed uuid rather than the original, as long as
2606 * In each case the uuid required is that of the data-array,
2607 * not the device-set.
2609 /* imsm does not track uuid's so we synthesis one using sha1 on
2610 * - The signature (Which is constant for all imsm array, but no matter)
2611 * - the orig_family_num of the container
2612 * - the index number of the volume
2613 * - the 'serial' number of the volume.
2614 * Hopefully these are all constant.
2616 struct intel_super
*super
= st
->sb
;
2619 struct sha1_ctx ctx
;
2620 struct imsm_dev
*dev
= NULL
;
2623 /* some mdadm versions failed to set ->orig_family_num, in which
2624 * case fall back to ->family_num. orig_family_num will be
2625 * fixed up with the first metadata update.
2627 family_num
= super
->anchor
->orig_family_num
;
2628 if (family_num
== 0)
2629 family_num
= super
->anchor
->family_num
;
2630 sha1_init_ctx(&ctx
);
2631 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2632 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2633 if (super
->current_vol
>= 0)
2634 dev
= get_imsm_dev(super
, super
->current_vol
);
2636 __u32 vol
= super
->current_vol
;
2637 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2638 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2640 sha1_finish_ctx(&ctx
, buf
);
2641 memcpy(uuid
, buf
, 4*4);
2646 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2648 __u8
*v
= get_imsm_version(mpb
);
2649 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2650 char major
[] = { 0, 0, 0 };
2651 char minor
[] = { 0 ,0, 0 };
2652 char patch
[] = { 0, 0, 0 };
2653 char *ver_parse
[] = { major
, minor
, patch
};
2657 while (*v
!= '\0' && v
< end
) {
2658 if (*v
!= '.' && j
< 2)
2659 ver_parse
[i
][j
++] = *v
;
2667 *m
= strtol(minor
, NULL
, 0);
2668 *p
= strtol(patch
, NULL
, 0);
2672 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2674 /* migr_strip_size when repairing or initializing parity */
2675 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2676 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2678 switch (get_imsm_raid_level(map
)) {
2683 return 128*1024 >> 9;
2687 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2689 /* migr_strip_size when rebuilding a degraded disk, no idea why
2690 * this is different than migr_strip_size_resync(), but it's good
2693 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2694 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2696 switch (get_imsm_raid_level(map
)) {
2699 if (map
->num_members
% map
->num_domains
== 0)
2700 return 128*1024 >> 9;
2704 return max((__u32
) 64*1024 >> 9, chunk
);
2706 return 128*1024 >> 9;
2710 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2712 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2713 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2714 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2715 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2717 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2720 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2722 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2723 int level
= get_imsm_raid_level(lo
);
2725 if (level
== 1 || level
== 10) {
2726 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2728 return hi
->num_domains
;
2730 return num_stripes_per_unit_resync(dev
);
2733 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2735 /* named 'imsm_' because raid0, raid1 and raid10
2736 * counter-intuitively have the same number of data disks
2738 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2740 switch (get_imsm_raid_level(map
)) {
2742 return map
->num_members
;
2746 return map
->num_members
/2;
2748 return map
->num_members
- 1;
2750 dprintf("unsupported raid level\n");
2755 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2757 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2758 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2760 switch(get_imsm_raid_level(map
)) {
2763 return chunk
* map
->num_domains
;
2765 return chunk
* map
->num_members
;
2771 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2773 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2774 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2775 __u32 strip
= block
/ chunk
;
2777 switch (get_imsm_raid_level(map
)) {
2780 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2781 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2783 return vol_stripe
* chunk
+ block
% chunk
;
2785 __u32 stripe
= strip
/ (map
->num_members
- 1);
2787 return stripe
* chunk
+ block
% chunk
;
2794 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2795 struct imsm_dev
*dev
)
2797 /* calculate the conversion factor between per member 'blocks'
2798 * (md/{resync,rebuild}_start) and imsm migration units, return
2799 * 0 for the 'not migrating' and 'unsupported migration' cases
2801 if (!dev
->vol
.migr_state
)
2804 switch (migr_type(dev
)) {
2805 case MIGR_GEN_MIGR
: {
2806 struct migr_record
*migr_rec
= super
->migr_rec
;
2807 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2812 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2813 __u32 stripes_per_unit
;
2814 __u32 blocks_per_unit
;
2823 /* yes, this is really the translation of migr_units to
2824 * per-member blocks in the 'resync' case
2826 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2827 migr_chunk
= migr_strip_blocks_resync(dev
);
2828 disks
= imsm_num_data_members(dev
, MAP_0
);
2829 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2830 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2831 segment
= blocks_per_unit
/ stripe
;
2832 block_rel
= blocks_per_unit
- segment
* stripe
;
2833 parity_depth
= parity_segment_depth(dev
);
2834 block_map
= map_migr_block(dev
, block_rel
);
2835 return block_map
+ parity_depth
* segment
;
2837 case MIGR_REBUILD
: {
2838 __u32 stripes_per_unit
;
2841 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2842 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2843 return migr_chunk
* stripes_per_unit
;
2845 case MIGR_STATE_CHANGE
:
2851 static int imsm_level_to_layout(int level
)
2859 return ALGORITHM_LEFT_ASYMMETRIC
;
2866 /*******************************************************************************
2867 * Function: read_imsm_migr_rec
2868 * Description: Function reads imsm migration record from last sector of disk
2870 * fd : disk descriptor
2871 * super : metadata info
2875 ******************************************************************************/
2876 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2879 unsigned int sector_size
= super
->sector_size
;
2880 unsigned long long dsize
;
2882 get_dev_size(fd
, NULL
, &dsize
);
2883 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
2885 pr_err("Cannot seek to anchor block: %s\n",
2889 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
2890 MIGR_REC_BUF_SECTORS
*sector_size
) !=
2891 MIGR_REC_BUF_SECTORS
*sector_size
) {
2892 pr_err("Cannot read migr record block: %s\n",
2897 if (sector_size
== 4096)
2898 convert_from_4k_imsm_migr_rec(super
);
2904 static struct imsm_dev
*imsm_get_device_during_migration(
2905 struct intel_super
*super
)
2908 struct intel_dev
*dv
;
2910 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2911 if (is_gen_migration(dv
->dev
))
2917 /*******************************************************************************
2918 * Function: load_imsm_migr_rec
2919 * Description: Function reads imsm migration record (it is stored at the last
2922 * super : imsm internal array info
2923 * info : general array info
2927 * -2 : no migration in progress
2928 ******************************************************************************/
2929 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2936 struct imsm_dev
*dev
;
2937 struct imsm_map
*map
;
2940 /* find map under migration */
2941 dev
= imsm_get_device_during_migration(super
);
2942 /* nothing to load,no migration in progress?
2948 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2949 /* read only from one of the first two slots */
2950 if ((sd
->disk
.raid_disk
< 0) ||
2951 (sd
->disk
.raid_disk
> 1))
2954 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2955 fd
= dev_open(nm
, O_RDONLY
);
2961 map
= get_imsm_map(dev
, MAP_0
);
2962 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2963 /* skip spare and failed disks
2967 /* read only from one of the first two slots */
2969 slot
= get_imsm_disk_slot(map
, dl
->index
);
2970 if (map
== NULL
|| slot
> 1 || slot
< 0)
2972 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2973 fd
= dev_open(nm
, O_RDONLY
);
2980 retval
= read_imsm_migr_rec(fd
, super
);
2988 /*******************************************************************************
2989 * function: imsm_create_metadata_checkpoint_update
2990 * Description: It creates update for checkpoint change.
2992 * super : imsm internal array info
2993 * u : pointer to prepared update
2996 * If length is equal to 0, input pointer u contains no update
2997 ******************************************************************************/
2998 static int imsm_create_metadata_checkpoint_update(
2999 struct intel_super
*super
,
3000 struct imsm_update_general_migration_checkpoint
**u
)
3003 int update_memory_size
= 0;
3005 dprintf("(enter)\n");
3011 /* size of all update data without anchor */
3012 update_memory_size
=
3013 sizeof(struct imsm_update_general_migration_checkpoint
);
3015 *u
= xcalloc(1, update_memory_size
);
3017 dprintf("error: cannot get memory\n");
3020 (*u
)->type
= update_general_migration_checkpoint
;
3021 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
3022 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3024 return update_memory_size
;
3027 static void imsm_update_metadata_locally(struct supertype
*st
,
3028 void *buf
, int len
);
3030 /*******************************************************************************
3031 * Function: write_imsm_migr_rec
3032 * Description: Function writes imsm migration record
3033 * (at the last sector of disk)
3035 * super : imsm internal array info
3039 ******************************************************************************/
3040 static int write_imsm_migr_rec(struct supertype
*st
)
3042 struct intel_super
*super
= st
->sb
;
3043 unsigned int sector_size
= super
->sector_size
;
3044 unsigned long long dsize
;
3050 struct imsm_update_general_migration_checkpoint
*u
;
3051 struct imsm_dev
*dev
;
3052 struct imsm_map
*map
;
3054 /* find map under migration */
3055 dev
= imsm_get_device_during_migration(super
);
3056 /* if no migration, write buffer anyway to clear migr_record
3057 * on disk based on first available device
3060 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3061 super
->current_vol
);
3063 map
= get_imsm_map(dev
, MAP_0
);
3065 if (sector_size
== 4096)
3066 convert_to_4k_imsm_migr_rec(super
);
3067 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3070 /* skip failed and spare devices */
3073 /* write to 2 first slots only */
3075 slot
= get_imsm_disk_slot(map
, sd
->index
);
3076 if (map
== NULL
|| slot
> 1 || slot
< 0)
3079 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3080 fd
= dev_open(nm
, O_RDWR
);
3083 get_dev_size(fd
, NULL
, &dsize
);
3084 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3086 pr_err("Cannot seek to anchor block: %s\n",
3090 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3091 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3092 MIGR_REC_BUF_SECTORS
*sector_size
) {
3093 pr_err("Cannot write migr record block: %s\n",
3100 if (sector_size
== 4096)
3101 convert_from_4k_imsm_migr_rec(super
);
3102 /* update checkpoint information in metadata */
3103 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3105 dprintf("imsm: Cannot prepare update\n");
3108 /* update metadata locally */
3109 imsm_update_metadata_locally(st
, u
, len
);
3110 /* and possibly remotely */
3111 if (st
->update_tail
) {
3112 append_metadata_update(st
, u
, len
);
3113 /* during reshape we do all work inside metadata handler
3114 * manage_reshape(), so metadata update has to be triggered
3117 flush_metadata_updates(st
);
3118 st
->update_tail
= &st
->updates
;
3129 /* spare/missing disks activations are not allowe when
3130 * array/container performs reshape operation, because
3131 * all arrays in container works on the same disks set
3133 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3136 struct intel_dev
*i_dev
;
3137 struct imsm_dev
*dev
;
3139 /* check whole container
3141 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3143 if (is_gen_migration(dev
)) {
3144 /* No repair during any migration in container
3152 static unsigned long long imsm_component_size_aligment_check(int level
,
3154 unsigned int sector_size
,
3155 unsigned long long component_size
)
3157 unsigned int component_size_alligment
;
3159 /* check component size aligment
3161 component_size_alligment
= component_size
% (chunk_size
/sector_size
);
3163 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
3164 level
, chunk_size
, component_size
,
3165 component_size_alligment
);
3167 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
3168 dprintf("imsm: reported component size alligned from %llu ",
3170 component_size
-= component_size_alligment
;
3171 dprintf_cont("to %llu (%i).\n",
3172 component_size
, component_size_alligment
);
3175 return component_size
;
3178 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3180 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3181 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3183 return pba_of_lba0(map
) +
3184 (num_data_stripes(map
) * map
->blocks_per_strip
);
3187 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3189 struct intel_super
*super
= st
->sb
;
3190 struct migr_record
*migr_rec
= super
->migr_rec
;
3191 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3192 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3193 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3194 struct imsm_map
*map_to_analyse
= map
;
3196 int map_disks
= info
->array
.raid_disks
;
3198 memset(info
, 0, sizeof(*info
));
3200 map_to_analyse
= prev_map
;
3202 dl
= super
->current_disk
;
3204 info
->container_member
= super
->current_vol
;
3205 info
->array
.raid_disks
= map
->num_members
;
3206 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3207 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3208 info
->array
.md_minor
= -1;
3209 info
->array
.ctime
= 0;
3210 info
->array
.utime
= 0;
3211 info
->array
.chunk_size
=
3212 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3213 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3214 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
3215 info
->custom_array_size
<<= 32;
3216 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
3217 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3219 if (is_gen_migration(dev
)) {
3220 info
->reshape_active
= 1;
3221 info
->new_level
= get_imsm_raid_level(map
);
3222 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3223 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3224 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3225 if (info
->delta_disks
) {
3226 /* this needs to be applied to every array
3229 info
->reshape_active
= CONTAINER_RESHAPE
;
3231 /* We shape information that we give to md might have to be
3232 * modify to cope with md's requirement for reshaping arrays.
3233 * For example, when reshaping a RAID0, md requires it to be
3234 * presented as a degraded RAID4.
3235 * Also if a RAID0 is migrating to a RAID5 we need to specify
3236 * the array as already being RAID5, but the 'before' layout
3237 * is a RAID4-like layout.
3239 switch (info
->array
.level
) {
3241 switch(info
->new_level
) {
3243 /* conversion is happening as RAID4 */
3244 info
->array
.level
= 4;
3245 info
->array
.raid_disks
+= 1;
3248 /* conversion is happening as RAID5 */
3249 info
->array
.level
= 5;
3250 info
->array
.layout
= ALGORITHM_PARITY_N
;
3251 info
->delta_disks
-= 1;
3254 /* FIXME error message */
3255 info
->array
.level
= UnSet
;
3261 info
->new_level
= UnSet
;
3262 info
->new_layout
= UnSet
;
3263 info
->new_chunk
= info
->array
.chunk_size
;
3264 info
->delta_disks
= 0;
3268 info
->disk
.major
= dl
->major
;
3269 info
->disk
.minor
= dl
->minor
;
3270 info
->disk
.number
= dl
->index
;
3271 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3275 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3277 if (info
->array
.level
== 5) {
3278 info
->component_size
= num_data_stripes(map_to_analyse
) *
3279 map_to_analyse
->blocks_per_strip
;
3281 info
->component_size
= blocks_per_member(map_to_analyse
);
3284 info
->component_size
= imsm_component_size_aligment_check(
3286 info
->array
.chunk_size
,
3288 info
->component_size
);
3289 info
->bb
.supported
= 1;
3291 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3292 info
->recovery_start
= MaxSector
;
3294 if (info
->array
.level
== 5 && dev
->rwh_policy
== RWH_DISTRIBUTED
) {
3295 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3296 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3297 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
) >> 9;
3298 } else if (info
->array
.level
<= 0) {
3299 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3301 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3304 info
->reshape_progress
= 0;
3305 info
->resync_start
= MaxSector
;
3306 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3307 !(info
->array
.state
& 1)) &&
3308 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3309 info
->resync_start
= 0;
3311 if (dev
->vol
.migr_state
) {
3312 switch (migr_type(dev
)) {
3315 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3317 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3319 info
->resync_start
= blocks_per_unit
* units
;
3322 case MIGR_GEN_MIGR
: {
3323 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3325 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
3326 unsigned long long array_blocks
;
3329 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3331 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
3332 (super
->migr_rec
->rec_status
==
3333 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3336 info
->reshape_progress
= blocks_per_unit
* units
;
3338 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3339 (unsigned long long)units
,
3340 (unsigned long long)blocks_per_unit
,
3341 info
->reshape_progress
);
3343 used_disks
= imsm_num_data_members(dev
, MAP_1
);
3344 if (used_disks
> 0) {
3345 array_blocks
= blocks_per_member(map
) *
3347 info
->custom_array_size
=
3348 round_size_to_mb(array_blocks
,
3354 /* we could emulate the checkpointing of
3355 * 'sync_action=check' migrations, but for now
3356 * we just immediately complete them
3359 /* this is handled by container_content_imsm() */
3360 case MIGR_STATE_CHANGE
:
3361 /* FIXME handle other migrations */
3363 /* we are not dirty, so... */
3364 info
->resync_start
= MaxSector
;
3368 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3369 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3371 info
->array
.major_version
= -1;
3372 info
->array
.minor_version
= -2;
3373 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3374 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3375 uuid_from_super_imsm(st
, info
->uuid
);
3379 for (i
=0; i
<map_disks
; i
++) {
3381 if (i
< info
->array
.raid_disks
) {
3382 struct imsm_disk
*dsk
;
3383 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3384 dsk
= get_imsm_disk(super
, j
);
3385 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3392 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3393 int failed
, int look_in_map
);
3395 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3398 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3400 if (is_gen_migration(dev
)) {
3403 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3405 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3406 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3407 if (map2
->map_state
!= map_state
) {
3408 map2
->map_state
= map_state
;
3409 super
->updates_pending
++;
3414 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3418 for (d
= super
->missing
; d
; d
= d
->next
)
3419 if (d
->index
== index
)
3424 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3426 struct intel_super
*super
= st
->sb
;
3427 struct imsm_disk
*disk
;
3428 int map_disks
= info
->array
.raid_disks
;
3429 int max_enough
= -1;
3431 struct imsm_super
*mpb
;
3433 if (super
->current_vol
>= 0) {
3434 getinfo_super_imsm_volume(st
, info
, map
);
3437 memset(info
, 0, sizeof(*info
));
3439 /* Set raid_disks to zero so that Assemble will always pull in valid
3442 info
->array
.raid_disks
= 0;
3443 info
->array
.level
= LEVEL_CONTAINER
;
3444 info
->array
.layout
= 0;
3445 info
->array
.md_minor
= -1;
3446 info
->array
.ctime
= 0; /* N/A for imsm */
3447 info
->array
.utime
= 0;
3448 info
->array
.chunk_size
= 0;
3450 info
->disk
.major
= 0;
3451 info
->disk
.minor
= 0;
3452 info
->disk
.raid_disk
= -1;
3453 info
->reshape_active
= 0;
3454 info
->array
.major_version
= -1;
3455 info
->array
.minor_version
= -2;
3456 strcpy(info
->text_version
, "imsm");
3457 info
->safe_mode_delay
= 0;
3458 info
->disk
.number
= -1;
3459 info
->disk
.state
= 0;
3461 info
->recovery_start
= MaxSector
;
3462 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3463 info
->bb
.supported
= 1;
3465 /* do we have the all the insync disks that we expect? */
3466 mpb
= super
->anchor
;
3467 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3469 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3470 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3471 int failed
, enough
, j
, missing
= 0;
3472 struct imsm_map
*map
;
3475 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3476 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3477 map
= get_imsm_map(dev
, MAP_0
);
3479 /* any newly missing disks?
3480 * (catches single-degraded vs double-degraded)
3482 for (j
= 0; j
< map
->num_members
; j
++) {
3483 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3484 __u32 idx
= ord_to_idx(ord
);
3486 if (!(ord
& IMSM_ORD_REBUILD
) &&
3487 get_imsm_missing(super
, idx
)) {
3493 if (state
== IMSM_T_STATE_FAILED
)
3495 else if (state
== IMSM_T_STATE_DEGRADED
&&
3496 (state
!= map
->map_state
|| missing
))
3498 else /* we're normal, or already degraded */
3500 if (is_gen_migration(dev
) && missing
) {
3501 /* during general migration we need all disks
3502 * that process is running on.
3503 * No new missing disk is allowed.
3507 /* no more checks necessary
3511 /* in the missing/failed disk case check to see
3512 * if at least one array is runnable
3514 max_enough
= max(max_enough
, enough
);
3516 dprintf("enough: %d\n", max_enough
);
3517 info
->container_enough
= max_enough
;
3520 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3522 disk
= &super
->disks
->disk
;
3523 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3524 info
->component_size
= reserved
;
3525 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3526 /* we don't change info->disk.raid_disk here because
3527 * this state will be finalized in mdmon after we have
3528 * found the 'most fresh' version of the metadata
3530 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3531 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3532 0 : (1 << MD_DISK_SYNC
);
3535 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3536 * ->compare_super may have updated the 'num_raid_devs' field for spares
3538 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3539 uuid_from_super_imsm(st
, info
->uuid
);
3541 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3543 /* I don't know how to compute 'map' on imsm, so use safe default */
3546 for (i
= 0; i
< map_disks
; i
++)
3552 /* allocates memory and fills disk in mdinfo structure
3553 * for each disk in array */
3554 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3556 struct mdinfo
*mddev
;
3557 struct intel_super
*super
= st
->sb
;
3558 struct imsm_disk
*disk
;
3561 if (!super
|| !super
->disks
)
3564 mddev
= xcalloc(1, sizeof(*mddev
));
3568 tmp
= xcalloc(1, sizeof(*tmp
));
3570 tmp
->next
= mddev
->devs
;
3572 tmp
->disk
.number
= count
++;
3573 tmp
->disk
.major
= dl
->major
;
3574 tmp
->disk
.minor
= dl
->minor
;
3575 tmp
->disk
.state
= is_configured(disk
) ?
3576 (1 << MD_DISK_ACTIVE
) : 0;
3577 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3578 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3579 tmp
->disk
.raid_disk
= -1;
3585 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3586 char *update
, char *devname
, int verbose
,
3587 int uuid_set
, char *homehost
)
3589 /* For 'assemble' and 'force' we need to return non-zero if any
3590 * change was made. For others, the return value is ignored.
3591 * Update options are:
3592 * force-one : This device looks a bit old but needs to be included,
3593 * update age info appropriately.
3594 * assemble: clear any 'faulty' flag to allow this device to
3596 * force-array: Array is degraded but being forced, mark it clean
3597 * if that will be needed to assemble it.
3599 * newdev: not used ????
3600 * grow: Array has gained a new device - this is currently for
3602 * resync: mark as dirty so a resync will happen.
3603 * name: update the name - preserving the homehost
3604 * uuid: Change the uuid of the array to match watch is given
3606 * Following are not relevant for this imsm:
3607 * sparc2.2 : update from old dodgey metadata
3608 * super-minor: change the preferred_minor number
3609 * summaries: update redundant counters.
3610 * homehost: update the recorded homehost
3611 * _reshape_progress: record new reshape_progress position.
3614 struct intel_super
*super
= st
->sb
;
3615 struct imsm_super
*mpb
;
3617 /* we can only update container info */
3618 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3621 mpb
= super
->anchor
;
3623 if (strcmp(update
, "uuid") == 0) {
3624 /* We take this to mean that the family_num should be updated.
3625 * However that is much smaller than the uuid so we cannot really
3626 * allow an explicit uuid to be given. And it is hard to reliably
3628 * So if !uuid_set we know the current uuid is random and just used
3629 * the first 'int' and copy it to the other 3 positions.
3630 * Otherwise we require the 4 'int's to be the same as would be the
3631 * case if we are using a random uuid. So an explicit uuid will be
3632 * accepted as long as all for ints are the same... which shouldn't hurt
3635 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3638 if (info
->uuid
[0] != info
->uuid
[1] ||
3639 info
->uuid
[1] != info
->uuid
[2] ||
3640 info
->uuid
[2] != info
->uuid
[3])
3646 mpb
->orig_family_num
= info
->uuid
[0];
3647 } else if (strcmp(update
, "assemble") == 0)
3652 /* successful update? recompute checksum */
3654 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3659 static size_t disks_to_mpb_size(int disks
)
3663 size
= sizeof(struct imsm_super
);
3664 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3665 size
+= 2 * sizeof(struct imsm_dev
);
3666 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3667 size
+= (4 - 2) * sizeof(struct imsm_map
);
3668 /* 4 possible disk_ord_tbl's */
3669 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3670 /* maximum bbm log */
3671 size
+= sizeof(struct bbm_log
);
3676 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3677 unsigned long long data_offset
)
3679 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3682 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3685 static void free_devlist(struct intel_super
*super
)
3687 struct intel_dev
*dv
;
3689 while (super
->devlist
) {
3690 dv
= super
->devlist
->next
;
3691 free(super
->devlist
->dev
);
3692 free(super
->devlist
);
3693 super
->devlist
= dv
;
3697 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3699 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3702 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3706 * 0 same, or first was empty, and second was copied
3707 * 1 second had wrong number
3709 * 3 wrong other info
3711 struct intel_super
*first
= st
->sb
;
3712 struct intel_super
*sec
= tst
->sb
;
3719 /* in platform dependent environment test if the disks
3720 * use the same Intel hba
3721 * If not on Intel hba at all, allow anything.
3723 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3724 if (first
->hba
->type
!= sec
->hba
->type
) {
3726 "HBAs of devices do not match %s != %s\n",
3727 get_sys_dev_type(first
->hba
->type
),
3728 get_sys_dev_type(sec
->hba
->type
));
3731 if (first
->orom
!= sec
->orom
) {
3733 "HBAs of devices do not match %s != %s\n",
3734 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3739 /* if an anchor does not have num_raid_devs set then it is a free
3742 if (first
->anchor
->num_raid_devs
> 0 &&
3743 sec
->anchor
->num_raid_devs
> 0) {
3744 /* Determine if these disks might ever have been
3745 * related. Further disambiguation can only take place
3746 * in load_super_imsm_all
3748 __u32 first_family
= first
->anchor
->orig_family_num
;
3749 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3751 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3752 MAX_SIGNATURE_LENGTH
) != 0)
3755 if (first_family
== 0)
3756 first_family
= first
->anchor
->family_num
;
3757 if (sec_family
== 0)
3758 sec_family
= sec
->anchor
->family_num
;
3760 if (first_family
!= sec_family
)
3765 /* if 'first' is a spare promote it to a populated mpb with sec's
3768 if (first
->anchor
->num_raid_devs
== 0 &&
3769 sec
->anchor
->num_raid_devs
> 0) {
3771 struct intel_dev
*dv
;
3772 struct imsm_dev
*dev
;
3774 /* we need to copy raid device info from sec if an allocation
3775 * fails here we don't associate the spare
3777 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3778 dv
= xmalloc(sizeof(*dv
));
3779 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3782 dv
->next
= first
->devlist
;
3783 first
->devlist
= dv
;
3785 if (i
< sec
->anchor
->num_raid_devs
) {
3786 /* allocation failure */
3787 free_devlist(first
);
3788 pr_err("imsm: failed to associate spare\n");
3791 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3792 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3793 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3794 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3795 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3796 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3802 static void fd2devname(int fd
, char *name
)
3806 char dname
[PATH_MAX
];
3811 if (fstat(fd
, &st
) != 0)
3813 sprintf(path
, "/sys/dev/block/%d:%d",
3814 major(st
.st_rdev
), minor(st
.st_rdev
));
3816 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3821 nm
= strrchr(dname
, '/');
3824 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3828 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3831 char *name
= fd2kname(fd
);
3836 if (strncmp(name
, "nvme", 4) != 0)
3839 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3841 return load_sys(path
, buf
, buf_len
);
3844 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3846 static int imsm_read_serial(int fd
, char *devname
,
3847 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3856 memset(buf
, 0, sizeof(buf
));
3858 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3861 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3863 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3864 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3865 fd2devname(fd
, (char *) serial
);
3871 pr_err("Failed to retrieve serial for %s\n",
3876 /* trim all whitespace and non-printable characters and convert
3879 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3882 /* ':' is reserved for use in placeholder serial
3883 * numbers for missing disks
3894 /* truncate leading characters */
3895 if (len
> MAX_RAID_SERIAL_LEN
) {
3896 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3897 len
= MAX_RAID_SERIAL_LEN
;
3900 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3901 memcpy(serial
, dest
, len
);
3906 static int serialcmp(__u8
*s1
, __u8
*s2
)
3908 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3911 static void serialcpy(__u8
*dest
, __u8
*src
)
3913 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3916 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3920 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3921 if (serialcmp(dl
->serial
, serial
) == 0)
3927 static struct imsm_disk
*
3928 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3932 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3933 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3935 if (serialcmp(disk
->serial
, serial
) == 0) {
3946 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3948 struct imsm_disk
*disk
;
3953 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3955 rv
= imsm_read_serial(fd
, devname
, serial
);
3960 dl
= xcalloc(1, sizeof(*dl
));
3963 dl
->major
= major(stb
.st_rdev
);
3964 dl
->minor
= minor(stb
.st_rdev
);
3965 dl
->next
= super
->disks
;
3966 dl
->fd
= keep_fd
? fd
: -1;
3967 assert(super
->disks
== NULL
);
3969 serialcpy(dl
->serial
, serial
);
3972 fd2devname(fd
, name
);
3974 dl
->devname
= xstrdup(devname
);
3976 dl
->devname
= xstrdup(name
);
3978 /* look up this disk's index in the current anchor */
3979 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3982 /* only set index on disks that are a member of a
3983 * populated contianer, i.e. one with raid_devs
3985 if (is_failed(&dl
->disk
))
3987 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
3994 /* When migrating map0 contains the 'destination' state while map1
3995 * contains the current state. When not migrating map0 contains the
3996 * current state. This routine assumes that map[0].map_state is set to
3997 * the current array state before being called.
3999 * Migration is indicated by one of the following states
4000 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
4001 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
4002 * map1state=unitialized)
4003 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
4005 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
4006 * map1state=degraded)
4007 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4010 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4011 __u8 to_state
, int migr_type
)
4013 struct imsm_map
*dest
;
4014 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4016 dev
->vol
.migr_state
= 1;
4017 set_migr_type(dev
, migr_type
);
4018 dev
->vol
.curr_migr_unit
= 0;
4019 dest
= get_imsm_map(dev
, MAP_1
);
4021 /* duplicate and then set the target end state in map[0] */
4022 memcpy(dest
, src
, sizeof_imsm_map(src
));
4023 if (migr_type
== MIGR_REBUILD
|| migr_type
== MIGR_GEN_MIGR
) {
4027 for (i
= 0; i
< src
->num_members
; i
++) {
4028 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4029 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4033 if (migr_type
== MIGR_GEN_MIGR
)
4034 /* Clear migration record */
4035 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4037 src
->map_state
= to_state
;
4040 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4043 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4044 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4048 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4049 * completed in the last migration.
4051 * FIXME add support for raid-level-migration
4053 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4054 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4055 /* when final map state is other than expected
4056 * merge maps (not for migration)
4060 for (i
= 0; i
< prev
->num_members
; i
++)
4061 for (j
= 0; j
< map
->num_members
; j
++)
4062 /* during online capacity expansion
4063 * disks position can be changed
4064 * if takeover is used
4066 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4067 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4068 map
->disk_ord_tbl
[j
] |=
4069 prev
->disk_ord_tbl
[i
];
4072 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4073 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4076 dev
->vol
.migr_state
= 0;
4077 set_migr_type(dev
, 0);
4078 dev
->vol
.curr_migr_unit
= 0;
4079 map
->map_state
= map_state
;
4082 static int parse_raid_devices(struct intel_super
*super
)
4085 struct imsm_dev
*dev_new
;
4086 size_t len
, len_migr
;
4088 size_t space_needed
= 0;
4089 struct imsm_super
*mpb
= super
->anchor
;
4091 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4092 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4093 struct intel_dev
*dv
;
4095 len
= sizeof_imsm_dev(dev_iter
, 0);
4096 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4098 space_needed
+= len_migr
- len
;
4100 dv
= xmalloc(sizeof(*dv
));
4101 if (max_len
< len_migr
)
4103 if (max_len
> len_migr
)
4104 space_needed
+= max_len
- len_migr
;
4105 dev_new
= xmalloc(max_len
);
4106 imsm_copy_dev(dev_new
, dev_iter
);
4109 dv
->next
= super
->devlist
;
4110 super
->devlist
= dv
;
4113 /* ensure that super->buf is large enough when all raid devices
4116 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4119 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4120 super
->sector_size
);
4121 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4124 memcpy(buf
, super
->buf
, super
->len
);
4125 memset(buf
+ super
->len
, 0, len
- super
->len
);
4131 super
->extra_space
+= space_needed
;
4136 /*******************************************************************************
4137 * Function: check_mpb_migr_compatibility
4138 * Description: Function checks for unsupported migration features:
4139 * - migration optimization area (pba_of_lba0)
4140 * - descending reshape (ascending_migr)
4142 * super : imsm metadata information
4144 * 0 : migration is compatible
4145 * -1 : migration is not compatible
4146 ******************************************************************************/
4147 int check_mpb_migr_compatibility(struct intel_super
*super
)
4149 struct imsm_map
*map0
, *map1
;
4150 struct migr_record
*migr_rec
= super
->migr_rec
;
4153 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4154 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4157 dev_iter
->vol
.migr_state
== 1 &&
4158 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4159 /* This device is migrating */
4160 map0
= get_imsm_map(dev_iter
, MAP_0
);
4161 map1
= get_imsm_map(dev_iter
, MAP_1
);
4162 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4163 /* migration optimization area was used */
4165 if (migr_rec
->ascending_migr
== 0
4166 && migr_rec
->dest_depth_per_unit
> 0)
4167 /* descending reshape not supported yet */
4174 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4176 /* load_imsm_mpb - read matrix metadata
4177 * allocates super->mpb to be freed by free_imsm
4179 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4181 unsigned long long dsize
;
4182 unsigned long long sectors
;
4183 unsigned int sector_size
= super
->sector_size
;
4185 struct imsm_super
*anchor
;
4188 get_dev_size(fd
, NULL
, &dsize
);
4189 if (dsize
< 2*sector_size
) {
4191 pr_err("%s: device to small for imsm\n",
4196 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4198 pr_err("Cannot seek to anchor block on %s: %s\n",
4199 devname
, strerror(errno
));
4203 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4205 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4208 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4210 pr_err("Cannot read anchor block on %s: %s\n",
4211 devname
, strerror(errno
));
4216 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4218 pr_err("no IMSM anchor on %s\n", devname
);
4223 __free_imsm(super
, 0);
4224 /* reload capability and hba */
4226 /* capability and hba must be updated with new super allocation */
4227 find_intel_hba_capability(fd
, super
, devname
);
4228 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4229 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4231 pr_err("unable to allocate %zu byte mpb buffer\n",
4236 memcpy(super
->buf
, anchor
, sector_size
);
4238 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4241 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
4242 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
4243 pr_err("could not allocate migr_rec buffer\n");
4247 super
->clean_migration_record_by_mdmon
= 0;
4250 check_sum
= __gen_imsm_checksum(super
->anchor
);
4251 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4253 pr_err("IMSM checksum %x != %x on %s\n",
4255 __le32_to_cpu(super
->anchor
->check_sum
),
4263 /* read the extended mpb */
4264 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4266 pr_err("Cannot seek to extended mpb on %s: %s\n",
4267 devname
, strerror(errno
));
4271 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4272 super
->len
- sector_size
) != super
->len
- sector_size
) {
4274 pr_err("Cannot read extended mpb on %s: %s\n",
4275 devname
, strerror(errno
));
4279 check_sum
= __gen_imsm_checksum(super
->anchor
);
4280 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4282 pr_err("IMSM checksum %x != %x on %s\n",
4283 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4291 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4293 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4294 static void clear_hi(struct intel_super
*super
)
4296 struct imsm_super
*mpb
= super
->anchor
;
4298 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4300 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4301 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4302 disk
->total_blocks_hi
= 0;
4304 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4305 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4308 for (n
= 0; n
< 2; ++n
) {
4309 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4312 map
->pba_of_lba0_hi
= 0;
4313 map
->blocks_per_member_hi
= 0;
4314 map
->num_data_stripes_hi
= 0;
4320 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4324 err
= load_imsm_mpb(fd
, super
, devname
);
4327 if (super
->sector_size
== 4096)
4328 convert_from_4k(super
);
4329 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4332 err
= parse_raid_devices(super
);
4335 err
= load_bbm_log(super
);
4340 static void __free_imsm_disk(struct dl
*d
)
4352 static void free_imsm_disks(struct intel_super
*super
)
4356 while (super
->disks
) {
4358 super
->disks
= d
->next
;
4359 __free_imsm_disk(d
);
4361 while (super
->disk_mgmt_list
) {
4362 d
= super
->disk_mgmt_list
;
4363 super
->disk_mgmt_list
= d
->next
;
4364 __free_imsm_disk(d
);
4366 while (super
->missing
) {
4368 super
->missing
= d
->next
;
4369 __free_imsm_disk(d
);
4374 /* free all the pieces hanging off of a super pointer */
4375 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4377 struct intel_hba
*elem
, *next
;
4383 /* unlink capability description */
4385 if (super
->migr_rec_buf
) {
4386 free(super
->migr_rec_buf
);
4387 super
->migr_rec_buf
= NULL
;
4390 free_imsm_disks(super
);
4391 free_devlist(super
);
4395 free((void *)elem
->path
);
4401 free(super
->bbm_log
);
4405 static void free_imsm(struct intel_super
*super
)
4407 __free_imsm(super
, 1);
4408 free(super
->bb
.entries
);
4412 static void free_super_imsm(struct supertype
*st
)
4414 struct intel_super
*super
= st
->sb
;
4423 static struct intel_super
*alloc_super(void)
4425 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4427 super
->current_vol
= -1;
4428 super
->create_offset
= ~((unsigned long long) 0);
4430 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4431 sizeof(struct md_bb_entry
));
4432 if (!super
->bb
.entries
) {
4441 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4443 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4445 struct sys_dev
*hba_name
;
4448 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4453 hba_name
= find_disk_attached_hba(fd
, NULL
);
4456 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4460 rv
= attach_hba_to_super(super
, hba_name
);
4463 struct intel_hba
*hba
= super
->hba
;
4465 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4466 " but the container is assigned to Intel(R) %s %s (",
4468 get_sys_dev_type(hba_name
->type
),
4469 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4470 hba_name
->pci_id
? : "Err!",
4471 get_sys_dev_type(super
->hba
->type
),
4472 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4475 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4477 fprintf(stderr
, ", ");
4480 fprintf(stderr
, ").\n"
4481 " Mixing devices attached to different %s is not allowed.\n",
4482 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
4486 super
->orom
= find_imsm_capability(hba_name
);
4493 /* find_missing - helper routine for load_super_imsm_all that identifies
4494 * disks that have disappeared from the system. This routine relies on
4495 * the mpb being uptodate, which it is at load time.
4497 static int find_missing(struct intel_super
*super
)
4500 struct imsm_super
*mpb
= super
->anchor
;
4502 struct imsm_disk
*disk
;
4504 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4505 disk
= __get_imsm_disk(mpb
, i
);
4506 dl
= serial_to_dl(disk
->serial
, super
);
4510 dl
= xmalloc(sizeof(*dl
));
4514 dl
->devname
= xstrdup("missing");
4516 serialcpy(dl
->serial
, disk
->serial
);
4519 dl
->next
= super
->missing
;
4520 super
->missing
= dl
;
4526 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4528 struct intel_disk
*idisk
= disk_list
;
4531 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4533 idisk
= idisk
->next
;
4539 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4540 struct intel_super
*super
,
4541 struct intel_disk
**disk_list
)
4543 struct imsm_disk
*d
= &super
->disks
->disk
;
4544 struct imsm_super
*mpb
= super
->anchor
;
4547 for (i
= 0; i
< tbl_size
; i
++) {
4548 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4549 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4551 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4552 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4553 dprintf("mpb from %d:%d matches %d:%d\n",
4554 super
->disks
->major
,
4555 super
->disks
->minor
,
4556 table
[i
]->disks
->major
,
4557 table
[i
]->disks
->minor
);
4561 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4562 is_configured(d
) == is_configured(tbl_d
)) &&
4563 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4564 /* current version of the mpb is a
4565 * better candidate than the one in
4566 * super_table, but copy over "cross
4567 * generational" status
4569 struct intel_disk
*idisk
;
4571 dprintf("mpb from %d:%d replaces %d:%d\n",
4572 super
->disks
->major
,
4573 super
->disks
->minor
,
4574 table
[i
]->disks
->major
,
4575 table
[i
]->disks
->minor
);
4577 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4578 if (idisk
&& is_failed(&idisk
->disk
))
4579 tbl_d
->status
|= FAILED_DISK
;
4582 struct intel_disk
*idisk
;
4583 struct imsm_disk
*disk
;
4585 /* tbl_mpb is more up to date, but copy
4586 * over cross generational status before
4589 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4590 if (disk
&& is_failed(disk
))
4591 d
->status
|= FAILED_DISK
;
4593 idisk
= disk_list_get(d
->serial
, *disk_list
);
4596 if (disk
&& is_configured(disk
))
4597 idisk
->disk
.status
|= CONFIGURED_DISK
;
4600 dprintf("mpb from %d:%d prefer %d:%d\n",
4601 super
->disks
->major
,
4602 super
->disks
->minor
,
4603 table
[i
]->disks
->major
,
4604 table
[i
]->disks
->minor
);
4612 table
[tbl_size
++] = super
;
4616 /* update/extend the merged list of imsm_disk records */
4617 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4618 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4619 struct intel_disk
*idisk
;
4621 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4623 idisk
->disk
.status
|= disk
->status
;
4624 if (is_configured(&idisk
->disk
) ||
4625 is_failed(&idisk
->disk
))
4626 idisk
->disk
.status
&= ~(SPARE_DISK
);
4628 idisk
= xcalloc(1, sizeof(*idisk
));
4629 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4630 idisk
->disk
= *disk
;
4631 idisk
->next
= *disk_list
;
4635 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4642 static struct intel_super
*
4643 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4646 struct imsm_super
*mpb
= super
->anchor
;
4650 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4651 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4652 struct intel_disk
*idisk
;
4654 idisk
= disk_list_get(disk
->serial
, disk_list
);
4656 if (idisk
->owner
== owner
||
4657 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4660 dprintf("'%.16s' owner %d != %d\n",
4661 disk
->serial
, idisk
->owner
,
4664 dprintf("unknown disk %x [%d]: %.16s\n",
4665 __le32_to_cpu(mpb
->family_num
), i
,
4671 if (ok_count
== mpb
->num_disks
)
4676 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4678 struct intel_super
*s
;
4680 for (s
= super_list
; s
; s
= s
->next
) {
4681 if (family_num
!= s
->anchor
->family_num
)
4683 pr_err("Conflict, offlining family %#x on '%s'\n",
4684 __le32_to_cpu(family_num
), s
->disks
->devname
);
4688 static struct intel_super
*
4689 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4691 struct intel_super
*super_table
[len
];
4692 struct intel_disk
*disk_list
= NULL
;
4693 struct intel_super
*champion
, *spare
;
4694 struct intel_super
*s
, **del
;
4699 memset(super_table
, 0, sizeof(super_table
));
4700 for (s
= *super_list
; s
; s
= s
->next
)
4701 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4703 for (i
= 0; i
< tbl_size
; i
++) {
4704 struct imsm_disk
*d
;
4705 struct intel_disk
*idisk
;
4706 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4709 d
= &s
->disks
->disk
;
4711 /* 'd' must appear in merged disk list for its
4712 * configuration to be valid
4714 idisk
= disk_list_get(d
->serial
, disk_list
);
4715 if (idisk
&& idisk
->owner
== i
)
4716 s
= validate_members(s
, disk_list
, i
);
4721 dprintf("marking family: %#x from %d:%d offline\n",
4723 super_table
[i
]->disks
->major
,
4724 super_table
[i
]->disks
->minor
);
4728 /* This is where the mdadm implementation differs from the Windows
4729 * driver which has no strict concept of a container. We can only
4730 * assemble one family from a container, so when returning a prodigal
4731 * array member to this system the code will not be able to disambiguate
4732 * the container contents that should be assembled ("foreign" versus
4733 * "local"). It requires user intervention to set the orig_family_num
4734 * to a new value to establish a new container. The Windows driver in
4735 * this situation fixes up the volume name in place and manages the
4736 * foreign array as an independent entity.
4741 for (i
= 0; i
< tbl_size
; i
++) {
4742 struct intel_super
*tbl_ent
= super_table
[i
];
4748 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4753 if (s
&& !is_spare
) {
4754 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4756 } else if (!s
&& !is_spare
)
4769 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4770 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4772 /* collect all dl's onto 'champion', and update them to
4773 * champion's version of the status
4775 for (s
= *super_list
; s
; s
= s
->next
) {
4776 struct imsm_super
*mpb
= champion
->anchor
;
4777 struct dl
*dl
= s
->disks
;
4782 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4784 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4785 struct imsm_disk
*disk
;
4787 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4790 /* only set index on disks that are a member of
4791 * a populated contianer, i.e. one with
4794 if (is_failed(&dl
->disk
))
4796 else if (is_spare(&dl
->disk
))
4802 if (i
>= mpb
->num_disks
) {
4803 struct intel_disk
*idisk
;
4805 idisk
= disk_list_get(dl
->serial
, disk_list
);
4806 if (idisk
&& is_spare(&idisk
->disk
) &&
4807 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4815 dl
->next
= champion
->disks
;
4816 champion
->disks
= dl
;
4820 /* delete 'champion' from super_list */
4821 for (del
= super_list
; *del
; ) {
4822 if (*del
== champion
) {
4823 *del
= (*del
)->next
;
4826 del
= &(*del
)->next
;
4828 champion
->next
= NULL
;
4832 struct intel_disk
*idisk
= disk_list
;
4834 disk_list
= disk_list
->next
;
4842 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4843 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4844 int major
, int minor
, int keep_fd
);
4846 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4847 int *max
, int keep_fd
);
4849 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4850 char *devname
, struct md_list
*devlist
,
4853 struct intel_super
*super_list
= NULL
;
4854 struct intel_super
*super
= NULL
;
4859 /* 'fd' is an opened container */
4860 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4862 /* get super block from devlist devices */
4863 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4866 /* all mpbs enter, maybe one leaves */
4867 super
= imsm_thunderdome(&super_list
, i
);
4873 if (find_missing(super
) != 0) {
4879 /* load migration record */
4880 err
= load_imsm_migr_rec(super
, NULL
);
4882 /* migration is in progress,
4883 * but migr_rec cannot be loaded,
4889 /* Check migration compatibility */
4890 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4891 pr_err("Unsupported migration detected");
4893 fprintf(stderr
, " on %s\n", devname
);
4895 fprintf(stderr
, " (IMSM).\n");
4904 while (super_list
) {
4905 struct intel_super
*s
= super_list
;
4907 super_list
= super_list
->next
;
4916 strcpy(st
->container_devnm
, fd2devnm(fd
));
4918 st
->container_devnm
[0] = 0;
4919 if (err
== 0 && st
->ss
== NULL
) {
4920 st
->ss
= &super_imsm
;
4921 st
->minor_version
= 0;
4922 st
->max_devs
= IMSM_MAX_DEVICES
;
4928 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4929 int *max
, int keep_fd
)
4931 struct md_list
*tmpdev
;
4935 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4936 if (tmpdev
->used
!= 1)
4938 if (tmpdev
->container
== 1) {
4940 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4942 pr_err("cannot open device %s: %s\n",
4943 tmpdev
->devname
, strerror(errno
));
4947 err
= get_sra_super_block(fd
, super_list
,
4948 tmpdev
->devname
, &lmax
,
4957 int major
= major(tmpdev
->st_rdev
);
4958 int minor
= minor(tmpdev
->st_rdev
);
4959 err
= get_super_block(super_list
,
4976 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4977 int major
, int minor
, int keep_fd
)
4979 struct intel_super
*s
;
4991 sprintf(nm
, "%d:%d", major
, minor
);
4992 dfd
= dev_open(nm
, O_RDWR
);
4998 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
4999 find_intel_hba_capability(dfd
, s
, devname
);
5000 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5002 /* retry the load if we might have raced against mdmon */
5003 if (err
== 3 && devnm
&& mdmon_running(devnm
))
5004 for (retry
= 0; retry
< 3; retry
++) {
5006 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5012 s
->next
= *super_list
;
5020 if (dfd
>= 0 && !keep_fd
)
5027 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5034 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5038 if (sra
->array
.major_version
!= -1 ||
5039 sra
->array
.minor_version
!= -2 ||
5040 strcmp(sra
->text_version
, "imsm") != 0) {
5045 devnm
= fd2devnm(fd
);
5046 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5047 if (get_super_block(super_list
, devnm
, devname
,
5048 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5059 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5061 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5064 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5066 struct intel_super
*super
;
5070 if (test_partition(fd
))
5071 /* IMSM not allowed on partitions */
5074 free_super_imsm(st
);
5076 super
= alloc_super();
5077 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5080 /* Load hba and capabilities if they exist.
5081 * But do not preclude loading metadata in case capabilities or hba are
5082 * non-compliant and ignore_hw_compat is set.
5084 rv
= find_intel_hba_capability(fd
, super
, devname
);
5085 /* no orom/efi or non-intel hba of the disk */
5086 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5088 pr_err("No OROM/EFI properties for %s\n", devname
);
5092 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5094 /* retry the load if we might have raced against mdmon */
5096 struct mdstat_ent
*mdstat
= NULL
;
5097 char *name
= fd2kname(fd
);
5100 mdstat
= mdstat_by_component(name
);
5102 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5103 for (retry
= 0; retry
< 3; retry
++) {
5105 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5111 free_mdstat(mdstat
);
5116 pr_err("Failed to load all information sections on %s\n", devname
);
5122 if (st
->ss
== NULL
) {
5123 st
->ss
= &super_imsm
;
5124 st
->minor_version
= 0;
5125 st
->max_devs
= IMSM_MAX_DEVICES
;
5128 /* load migration record */
5129 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5130 /* Check for unsupported migration features */
5131 if (check_mpb_migr_compatibility(super
) != 0) {
5132 pr_err("Unsupported migration detected");
5134 fprintf(stderr
, " on %s\n", devname
);
5136 fprintf(stderr
, " (IMSM).\n");
5144 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5146 if (info
->level
== 1)
5148 return info
->chunk_size
>> 9;
5151 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5152 unsigned long long size
)
5154 if (info
->level
== 1)
5157 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5160 static void imsm_update_version_info(struct intel_super
*super
)
5162 /* update the version and attributes */
5163 struct imsm_super
*mpb
= super
->anchor
;
5165 struct imsm_dev
*dev
;
5166 struct imsm_map
*map
;
5169 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5170 dev
= get_imsm_dev(super
, i
);
5171 map
= get_imsm_map(dev
, MAP_0
);
5172 if (__le32_to_cpu(dev
->size_high
) > 0)
5173 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5175 /* FIXME detect when an array spans a port multiplier */
5177 mpb
->attributes
|= MPB_ATTRIB_PM
;
5180 if (mpb
->num_raid_devs
> 1 ||
5181 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5182 version
= MPB_VERSION_ATTRIBS
;
5183 switch (get_imsm_raid_level(map
)) {
5184 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5185 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5186 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5187 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5190 if (map
->num_members
>= 5)
5191 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5192 else if (dev
->status
== DEV_CLONE_N_GO
)
5193 version
= MPB_VERSION_CNG
;
5194 else if (get_imsm_raid_level(map
) == 5)
5195 version
= MPB_VERSION_RAID5
;
5196 else if (map
->num_members
>= 3)
5197 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5198 else if (get_imsm_raid_level(map
) == 1)
5199 version
= MPB_VERSION_RAID1
;
5201 version
= MPB_VERSION_RAID0
;
5203 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5207 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5209 struct imsm_super
*mpb
= super
->anchor
;
5210 char *reason
= NULL
;
5213 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
5214 reason
= "must be 16 characters or less";
5216 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5217 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5219 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5220 reason
= "already exists";
5225 if (reason
&& !quiet
)
5226 pr_err("imsm volume name %s\n", reason
);
5231 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5232 struct shape
*s
, char *name
,
5233 char *homehost
, int *uuid
,
5234 long long data_offset
)
5236 /* We are creating a volume inside a pre-existing container.
5237 * so st->sb is already set.
5239 struct intel_super
*super
= st
->sb
;
5240 unsigned int sector_size
= super
->sector_size
;
5241 struct imsm_super
*mpb
= super
->anchor
;
5242 struct intel_dev
*dv
;
5243 struct imsm_dev
*dev
;
5244 struct imsm_vol
*vol
;
5245 struct imsm_map
*map
;
5246 int idx
= mpb
->num_raid_devs
;
5248 unsigned long long array_blocks
;
5249 size_t size_old
, size_new
;
5250 unsigned long long num_data_stripes
;
5251 unsigned int data_disks
;
5252 unsigned long long size_per_member
;
5254 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5255 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5259 /* ensure the mpb is large enough for the new data */
5260 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5261 size_new
= disks_to_mpb_size(info
->nr_disks
);
5262 if (size_new
> size_old
) {
5264 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5266 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5267 pr_err("could not allocate new mpb\n");
5270 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5271 MIGR_REC_BUF_SECTORS
*
5272 MAX_SECTOR_SIZE
) != 0) {
5273 pr_err("could not allocate migr_rec buffer\n");
5279 memcpy(mpb_new
, mpb
, size_old
);
5282 super
->anchor
= mpb_new
;
5283 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5284 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5285 super
->len
= size_round
;
5287 super
->current_vol
= idx
;
5289 /* handle 'failed_disks' by either:
5290 * a) create dummy disk entries in the table if this the first
5291 * volume in the array. We add them here as this is the only
5292 * opportunity to add them. add_to_super_imsm_volume()
5293 * handles the non-failed disks and continues incrementing
5295 * b) validate that 'failed_disks' matches the current number
5296 * of missing disks if the container is populated
5298 if (super
->current_vol
== 0) {
5300 for (i
= 0; i
< info
->failed_disks
; i
++) {
5301 struct imsm_disk
*disk
;
5304 disk
= __get_imsm_disk(mpb
, i
);
5305 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5306 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5307 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5308 "missing:%d", (__u8
)i
);
5310 find_missing(super
);
5315 for (d
= super
->missing
; d
; d
= d
->next
)
5317 if (info
->failed_disks
> missing
) {
5318 pr_err("unable to add 'missing' disk to container\n");
5323 if (!check_name(super
, name
, 0))
5325 dv
= xmalloc(sizeof(*dv
));
5326 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5327 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
5328 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5329 info
->layout
, info
->chunk_size
,
5330 s
->size
* BLOCKS_PER_KB
);
5331 data_disks
= get_data_disks(info
->level
, info
->layout
,
5333 array_blocks
= round_size_to_mb(array_blocks
, data_disks
);
5334 size_per_member
= array_blocks
/ data_disks
;
5336 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
5337 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
5338 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5340 vol
->migr_state
= 0;
5341 set_migr_type(dev
, MIGR_INIT
);
5342 vol
->dirty
= !info
->state
;
5343 vol
->curr_migr_unit
= 0;
5344 map
= get_imsm_map(dev
, MAP_0
);
5345 set_pba_of_lba0(map
, super
->create_offset
);
5346 set_blocks_per_member(map
, info_to_blocks_per_member(info
,
5349 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5350 map
->failed_disk_num
= ~0;
5351 if (info
->level
> 0)
5352 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5353 : IMSM_T_STATE_UNINITIALIZED
);
5355 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5356 IMSM_T_STATE_NORMAL
;
5359 if (info
->level
== 1 && info
->raid_disks
> 2) {
5362 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5366 map
->raid_level
= info
->level
;
5367 if (info
->level
== 10) {
5368 map
->raid_level
= 1;
5369 map
->num_domains
= info
->raid_disks
/ 2;
5370 } else if (info
->level
== 1)
5371 map
->num_domains
= info
->raid_disks
;
5373 map
->num_domains
= 1;
5375 /* info->size is only int so use the 'size' parameter instead */
5376 num_data_stripes
= size_per_member
/ info_to_blocks_per_strip(info
);
5377 num_data_stripes
/= map
->num_domains
;
5378 set_num_data_stripes(map
, num_data_stripes
);
5380 map
->num_members
= info
->raid_disks
;
5381 for (i
= 0; i
< map
->num_members
; i
++) {
5382 /* initialized in add_to_super */
5383 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5385 mpb
->num_raid_devs
++;
5386 mpb
->num_raid_devs_created
++;
5387 dev
->my_vol_raid_dev_num
= mpb
->num_raid_devs_created
;
5389 if (s
->consistency_policy
<= CONSISTENCY_POLICY_RESYNC
) {
5390 dev
->rwh_policy
= RWH_OFF
;
5391 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5392 dev
->rwh_policy
= RWH_DISTRIBUTED
;
5396 pr_err("imsm does not support consistency policy %s\n",
5397 map_num(consistency_policies
, s
->consistency_policy
));
5402 dv
->index
= super
->current_vol
;
5403 dv
->next
= super
->devlist
;
5404 super
->devlist
= dv
;
5406 imsm_update_version_info(super
);
5411 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5412 struct shape
*s
, char *name
,
5413 char *homehost
, int *uuid
,
5414 unsigned long long data_offset
)
5416 /* This is primarily called by Create when creating a new array.
5417 * We will then get add_to_super called for each component, and then
5418 * write_init_super called to write it out to each device.
5419 * For IMSM, Create can create on fresh devices or on a pre-existing
5421 * To create on a pre-existing array a different method will be called.
5422 * This one is just for fresh drives.
5424 struct intel_super
*super
;
5425 struct imsm_super
*mpb
;
5429 if (data_offset
!= INVALID_SECTORS
) {
5430 pr_err("data-offset not supported by imsm\n");
5435 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5439 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5441 mpb_size
= MAX_SECTOR_SIZE
;
5443 super
= alloc_super();
5445 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5450 pr_err("could not allocate superblock\n");
5453 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5454 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5455 pr_err("could not allocate migr_rec buffer\n");
5460 memset(super
->buf
, 0, mpb_size
);
5462 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5466 /* zeroing superblock */
5470 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5472 version
= (char *) mpb
->sig
;
5473 strcpy(version
, MPB_SIGNATURE
);
5474 version
+= strlen(MPB_SIGNATURE
);
5475 strcpy(version
, MPB_VERSION_RAID0
);
5480 static int drive_validate_sector_size(struct intel_super
*super
, struct dl
*dl
)
5482 unsigned int member_sector_size
;
5485 pr_err("Invalid file descriptor for %s\n", dl
->devname
);
5489 if (!get_dev_sector_size(dl
->fd
, dl
->devname
, &member_sector_size
))
5491 if (member_sector_size
!= super
->sector_size
)
5496 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5497 int fd
, char *devname
)
5499 struct intel_super
*super
= st
->sb
;
5500 struct imsm_super
*mpb
= super
->anchor
;
5501 struct imsm_disk
*_disk
;
5502 struct imsm_dev
*dev
;
5503 struct imsm_map
*map
;
5507 dev
= get_imsm_dev(super
, super
->current_vol
);
5508 map
= get_imsm_map(dev
, MAP_0
);
5510 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5511 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5517 /* we're doing autolayout so grab the pre-marked (in
5518 * validate_geometry) raid_disk
5520 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5521 if (dl
->raiddisk
== dk
->raid_disk
)
5524 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5525 if (dl
->major
== dk
->major
&&
5526 dl
->minor
== dk
->minor
)
5531 pr_err("%s is not a member of the same container\n", devname
);
5535 if (!drive_validate_sector_size(super
, dl
)) {
5536 pr_err("Combining drives of different sector size in one volume is not allowed\n");
5540 /* add a pristine spare to the metadata */
5541 if (dl
->index
< 0) {
5542 dl
->index
= super
->anchor
->num_disks
;
5543 super
->anchor
->num_disks
++;
5545 /* Check the device has not already been added */
5546 slot
= get_imsm_disk_slot(map
, dl
->index
);
5548 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5549 pr_err("%s has been included in this array twice\n",
5553 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5554 dl
->disk
.status
= CONFIGURED_DISK
;
5556 /* update size of 'missing' disks to be at least as large as the
5557 * largest acitve member (we only have dummy missing disks when
5558 * creating the first volume)
5560 if (super
->current_vol
== 0) {
5561 for (df
= super
->missing
; df
; df
= df
->next
) {
5562 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5563 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5564 _disk
= __get_imsm_disk(mpb
, df
->index
);
5569 /* refresh unset/failed slots to point to valid 'missing' entries */
5570 for (df
= super
->missing
; df
; df
= df
->next
)
5571 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5572 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5574 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5576 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5577 if (is_gen_migration(dev
)) {
5578 struct imsm_map
*map2
= get_imsm_map(dev
,
5580 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5581 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5582 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5585 if ((unsigned)df
->index
==
5587 set_imsm_ord_tbl_ent(map2
,
5593 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5597 /* if we are creating the first raid device update the family number */
5598 if (super
->current_vol
== 0) {
5600 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5602 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5603 if (!_dev
|| !_disk
) {
5604 pr_err("BUG mpb setup error\n");
5610 sum
+= __gen_imsm_checksum(mpb
);
5611 mpb
->family_num
= __cpu_to_le32(sum
);
5612 mpb
->orig_family_num
= mpb
->family_num
;
5614 super
->current_disk
= dl
;
5619 * Function marks disk as spare and restores disk serial
5620 * in case it was previously marked as failed by takeover operation
5622 * -1 : critical error
5623 * 0 : disk is marked as spare but serial is not set
5626 int mark_spare(struct dl
*disk
)
5628 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5635 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5636 /* Restore disk serial number, because takeover marks disk
5637 * as failed and adds to serial ':0' before it becomes
5640 serialcpy(disk
->serial
, serial
);
5641 serialcpy(disk
->disk
.serial
, serial
);
5644 disk
->disk
.status
= SPARE_DISK
;
5650 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5651 int fd
, char *devname
,
5652 unsigned long long data_offset
)
5654 struct intel_super
*super
= st
->sb
;
5656 unsigned long long size
;
5657 unsigned int member_sector_size
;
5662 /* If we are on an RAID enabled platform check that the disk is
5663 * attached to the raid controller.
5664 * We do not need to test disks attachment for container based additions,
5665 * they shall be already tested when container was created/assembled.
5667 rv
= find_intel_hba_capability(fd
, super
, devname
);
5668 /* no orom/efi or non-intel hba of the disk */
5670 dprintf("capability: %p fd: %d ret: %d\n",
5671 super
->orom
, fd
, rv
);
5675 if (super
->current_vol
>= 0)
5676 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5679 dd
= xcalloc(sizeof(*dd
), 1);
5680 dd
->major
= major(stb
.st_rdev
);
5681 dd
->minor
= minor(stb
.st_rdev
);
5682 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5685 dd
->action
= DISK_ADD
;
5686 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5688 pr_err("failed to retrieve scsi serial, aborting\n");
5694 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5695 (super
->hba
->type
== SYS_DEV_VMD
))) {
5697 char *devpath
= diskfd_to_devpath(fd
);
5698 char controller_path
[PATH_MAX
];
5701 pr_err("failed to get devpath, aborting\n");
5708 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5711 if (devpath_to_vendor(controller_path
) == 0x8086) {
5713 * If Intel's NVMe drive has serial ended with
5714 * "-A","-B","-1" or "-2" it means that this is "x8"
5715 * device (double drive on single PCIe card).
5716 * User should be warned about potential data loss.
5718 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5719 /* Skip empty character at the end */
5720 if (dd
->serial
[i
] == 0)
5723 if (((dd
->serial
[i
] == 'A') ||
5724 (dd
->serial
[i
] == 'B') ||
5725 (dd
->serial
[i
] == '1') ||
5726 (dd
->serial
[i
] == '2')) &&
5727 (dd
->serial
[i
-1] == '-'))
5728 pr_err("\tThe action you are about to take may put your data at risk.\n"
5729 "\tPlease note that x8 devices may consist of two separate x4 devices "
5730 "located on a single PCIe port.\n"
5731 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5734 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5735 !imsm_orom_has_tpv_support(super
->orom
)) {
5736 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5737 "\tPlease refer to Intel(R) RSTe user guide.\n");
5744 get_dev_size(fd
, NULL
, &size
);
5745 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5747 if (super
->sector_size
== 0) {
5748 /* this a first device, so sector_size is not set yet */
5749 super
->sector_size
= member_sector_size
;
5752 /* clear migr_rec when adding disk to container */
5753 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5754 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*member_sector_size
,
5756 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5757 MIGR_REC_BUF_SECTORS
*member_sector_size
) !=
5758 MIGR_REC_BUF_SECTORS
*member_sector_size
)
5759 perror("Write migr_rec failed");
5763 serialcpy(dd
->disk
.serial
, dd
->serial
);
5764 set_total_blocks(&dd
->disk
, size
);
5765 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5766 struct imsm_super
*mpb
= super
->anchor
;
5767 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5770 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5771 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5773 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5775 if (st
->update_tail
) {
5776 dd
->next
= super
->disk_mgmt_list
;
5777 super
->disk_mgmt_list
= dd
;
5779 dd
->next
= super
->disks
;
5781 super
->updates_pending
++;
5787 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5789 struct intel_super
*super
= st
->sb
;
5792 /* remove from super works only in mdmon - for communication
5793 * manager - monitor. Check if communication memory buffer
5796 if (!st
->update_tail
) {
5797 pr_err("shall be used in mdmon context only\n");
5800 dd
= xcalloc(1, sizeof(*dd
));
5801 dd
->major
= dk
->major
;
5802 dd
->minor
= dk
->minor
;
5805 dd
->action
= DISK_REMOVE
;
5807 dd
->next
= super
->disk_mgmt_list
;
5808 super
->disk_mgmt_list
= dd
;
5813 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5816 char buf
[MAX_SECTOR_SIZE
];
5817 struct imsm_super anchor
;
5818 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5820 /* spare records have their own family number and do not have any defined raid
5823 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5825 struct imsm_super
*mpb
= super
->anchor
;
5826 struct imsm_super
*spare
= &spare_record
.anchor
;
5830 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5831 spare
->generation_num
= __cpu_to_le32(1UL);
5832 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5833 spare
->num_disks
= 1;
5834 spare
->num_raid_devs
= 0;
5835 spare
->cache_size
= mpb
->cache_size
;
5836 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5838 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5839 MPB_SIGNATURE MPB_VERSION_RAID0
);
5841 for (d
= super
->disks
; d
; d
= d
->next
) {
5845 spare
->disk
[0] = d
->disk
;
5846 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5847 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5849 if (super
->sector_size
== 4096)
5850 convert_to_4k_imsm_disk(&spare
->disk
[0]);
5852 sum
= __gen_imsm_checksum(spare
);
5853 spare
->family_num
= __cpu_to_le32(sum
);
5854 spare
->orig_family_num
= 0;
5855 sum
= __gen_imsm_checksum(spare
);
5856 spare
->check_sum
= __cpu_to_le32(sum
);
5858 if (store_imsm_mpb(d
->fd
, spare
)) {
5859 pr_err("failed for device %d:%d %s\n",
5860 d
->major
, d
->minor
, strerror(errno
));
5872 static int write_super_imsm(struct supertype
*st
, int doclose
)
5874 struct intel_super
*super
= st
->sb
;
5875 unsigned int sector_size
= super
->sector_size
;
5876 struct imsm_super
*mpb
= super
->anchor
;
5882 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5884 int clear_migration_record
= 1;
5887 /* 'generation' is incremented everytime the metadata is written */
5888 generation
= __le32_to_cpu(mpb
->generation_num
);
5890 mpb
->generation_num
= __cpu_to_le32(generation
);
5892 /* fix up cases where previous mdadm releases failed to set
5895 if (mpb
->orig_family_num
== 0)
5896 mpb
->orig_family_num
= mpb
->family_num
;
5898 for (d
= super
->disks
; d
; d
= d
->next
) {
5902 mpb
->disk
[d
->index
] = d
->disk
;
5906 for (d
= super
->missing
; d
; d
= d
->next
) {
5907 mpb
->disk
[d
->index
] = d
->disk
;
5910 mpb
->num_disks
= num_disks
;
5911 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5913 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5914 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5915 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5917 imsm_copy_dev(dev
, dev2
);
5918 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5920 if (is_gen_migration(dev2
))
5921 clear_migration_record
= 0;
5924 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
5927 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
5928 mpb
->attributes
|= MPB_ATTRIB_BBM
;
5930 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
5932 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
5933 mpb_size
+= bbm_log_size
;
5934 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5937 assert(super
->len
== 0 || mpb_size
<= super
->len
);
5940 /* recalculate checksum */
5941 sum
= __gen_imsm_checksum(mpb
);
5942 mpb
->check_sum
= __cpu_to_le32(sum
);
5944 if (super
->clean_migration_record_by_mdmon
) {
5945 clear_migration_record
= 1;
5946 super
->clean_migration_record_by_mdmon
= 0;
5948 if (clear_migration_record
)
5949 memset(super
->migr_rec_buf
, 0,
5950 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
5952 if (sector_size
== 4096)
5953 convert_to_4k(super
);
5955 /* write the mpb for disks that compose raid devices */
5956 for (d
= super
->disks
; d
; d
= d
->next
) {
5957 if (d
->index
< 0 || is_failed(&d
->disk
))
5960 if (clear_migration_record
) {
5961 unsigned long long dsize
;
5963 get_dev_size(d
->fd
, NULL
, &dsize
);
5964 if (lseek64(d
->fd
, dsize
- sector_size
,
5966 if ((unsigned int)write(d
->fd
,
5967 super
->migr_rec_buf
,
5968 MIGR_REC_BUF_SECTORS
*sector_size
) !=
5969 MIGR_REC_BUF_SECTORS
*sector_size
)
5970 perror("Write migr_rec failed");
5974 if (store_imsm_mpb(d
->fd
, mpb
))
5976 "failed for device %d:%d (fd: %d)%s\n",
5978 d
->fd
, strerror(errno
));
5987 return write_super_imsm_spares(super
, doclose
);
5992 static int create_array(struct supertype
*st
, int dev_idx
)
5995 struct imsm_update_create_array
*u
;
5996 struct intel_super
*super
= st
->sb
;
5997 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5998 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5999 struct disk_info
*inf
;
6000 struct imsm_disk
*disk
;
6003 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
6004 sizeof(*inf
) * map
->num_members
;
6006 u
->type
= update_create_array
;
6007 u
->dev_idx
= dev_idx
;
6008 imsm_copy_dev(&u
->dev
, dev
);
6009 inf
= get_disk_info(u
);
6010 for (i
= 0; i
< map
->num_members
; i
++) {
6011 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
6013 disk
= get_imsm_disk(super
, idx
);
6015 disk
= get_imsm_missing(super
, idx
);
6016 serialcpy(inf
[i
].serial
, disk
->serial
);
6018 append_metadata_update(st
, u
, len
);
6023 static int mgmt_disk(struct supertype
*st
)
6025 struct intel_super
*super
= st
->sb
;
6027 struct imsm_update_add_remove_disk
*u
;
6029 if (!super
->disk_mgmt_list
)
6034 u
->type
= update_add_remove_disk
;
6035 append_metadata_update(st
, u
, len
);
6040 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6042 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6044 struct intel_super
*super
= st
->sb
;
6046 struct ppl_header
*ppl_hdr
;
6049 ret
= posix_memalign(&buf
, 4096, PPL_HEADER_SIZE
);
6051 pr_err("Failed to allocate PPL header buffer\n");
6055 memset(buf
, 0, PPL_HEADER_SIZE
);
6057 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6058 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6059 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6061 if (lseek64(fd
, info
->ppl_sector
* 512, SEEK_SET
) < 0) {
6063 perror("Failed to seek to PPL header location");
6066 if (!ret
&& write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6068 perror("Write PPL header failed");
6078 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6079 struct mdinfo
*disk
)
6081 struct intel_super
*super
= st
->sb
;
6085 struct ppl_header
*ppl_hdr
;
6087 struct imsm_dev
*dev
;
6088 struct imsm_map
*map
;
6091 if (disk
->disk
.raid_disk
< 0)
6094 if (posix_memalign(&buf
, 4096, PPL_HEADER_SIZE
)) {
6095 pr_err("Failed to allocate PPL header buffer\n");
6099 dev
= get_imsm_dev(super
, info
->container_member
);
6100 map
= get_imsm_map(dev
, MAP_X
);
6101 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_X
);
6102 d
= get_imsm_dl_disk(super
, idx
);
6104 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6107 if (lseek64(d
->fd
, info
->ppl_sector
* 512, SEEK_SET
) < 0) {
6108 perror("Failed to seek to PPL header location");
6113 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6114 perror("Read PPL header failed");
6121 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6122 ppl_hdr
->checksum
= 0;
6124 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6125 dprintf("Wrong PPL header checksum on %s\n",
6130 if (!ret
&& (__le32_to_cpu(ppl_hdr
->signature
) !=
6131 super
->anchor
->orig_family_num
)) {
6132 dprintf("Wrong PPL header signature on %s\n",
6140 if (ret
== 1 && map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6141 return st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6146 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6148 struct intel_super
*super
= st
->sb
;
6152 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6153 info
->array
.level
!= 5)
6156 for (d
= super
->disks
; d
; d
= d
->next
) {
6157 if (d
->index
< 0 || is_failed(&d
->disk
))
6160 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6168 static int write_init_super_imsm(struct supertype
*st
)
6170 struct intel_super
*super
= st
->sb
;
6171 int current_vol
= super
->current_vol
;
6175 getinfo_super_imsm(st
, &info
, NULL
);
6177 /* we are done with current_vol reset it to point st at the container */
6178 super
->current_vol
= -1;
6180 if (st
->update_tail
) {
6181 /* queue the recently created array / added disk
6182 * as a metadata update */
6184 /* determine if we are creating a volume or adding a disk */
6185 if (current_vol
< 0) {
6186 /* in the mgmt (add/remove) disk case we are running
6187 * in mdmon context, so don't close fd's
6191 rv
= write_init_ppl_imsm_all(st
, &info
);
6193 rv
= create_array(st
, current_vol
);
6197 for (d
= super
->disks
; d
; d
= d
->next
)
6198 Kill(d
->devname
, NULL
, 0, -1, 1);
6199 if (current_vol
>= 0)
6200 rv
= write_init_ppl_imsm_all(st
, &info
);
6202 rv
= write_super_imsm(st
, 1);
6208 static int store_super_imsm(struct supertype
*st
, int fd
)
6210 struct intel_super
*super
= st
->sb
;
6211 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6216 if (super
->sector_size
== 4096)
6217 convert_to_4k(super
);
6218 return store_imsm_mpb(fd
, mpb
);
6221 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6222 int layout
, int raiddisks
, int chunk
,
6223 unsigned long long size
,
6224 unsigned long long data_offset
,
6226 unsigned long long *freesize
,
6230 unsigned long long ldsize
;
6231 struct intel_super
*super
;
6234 if (level
!= LEVEL_CONTAINER
)
6239 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6242 pr_err("imsm: Cannot open %s: %s\n",
6243 dev
, strerror(errno
));
6246 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6251 /* capabilities retrieve could be possible
6252 * note that there is no fd for the disks in array.
6254 super
= alloc_super();
6259 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6265 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6269 fd2devname(fd
, str
);
6270 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6271 fd
, str
, super
->orom
, rv
, raiddisks
);
6273 /* no orom/efi or non-intel hba of the disk */
6280 if (raiddisks
> super
->orom
->tds
) {
6282 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6283 raiddisks
, super
->orom
->tds
);
6287 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6288 (ldsize
>> 9) >> 32 > 0) {
6290 pr_err("%s exceeds maximum platform supported size\n", dev
);
6296 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6302 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6304 const unsigned long long base_start
= e
[*idx
].start
;
6305 unsigned long long end
= base_start
+ e
[*idx
].size
;
6308 if (base_start
== end
)
6312 for (i
= *idx
; i
< num_extents
; i
++) {
6313 /* extend overlapping extents */
6314 if (e
[i
].start
>= base_start
&&
6315 e
[i
].start
<= end
) {
6318 if (e
[i
].start
+ e
[i
].size
> end
)
6319 end
= e
[i
].start
+ e
[i
].size
;
6320 } else if (e
[i
].start
> end
) {
6326 return end
- base_start
;
6329 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6331 /* build a composite disk with all known extents and generate a new
6332 * 'maxsize' given the "all disks in an array must share a common start
6333 * offset" constraint
6335 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6339 unsigned long long pos
;
6340 unsigned long long start
= 0;
6341 unsigned long long maxsize
;
6342 unsigned long reserve
;
6344 /* coalesce and sort all extents. also, check to see if we need to
6345 * reserve space between member arrays
6348 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6351 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6354 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6359 while (i
< sum_extents
) {
6360 e
[j
].start
= e
[i
].start
;
6361 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6363 if (e
[j
-1].size
== 0)
6372 unsigned long long esize
;
6374 esize
= e
[i
].start
- pos
;
6375 if (esize
>= maxsize
) {
6380 pos
= e
[i
].start
+ e
[i
].size
;
6382 } while (e
[i
-1].size
);
6388 /* FIXME assumes volume at offset 0 is the first volume in a
6391 if (start_extent
> 0)
6392 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6396 if (maxsize
< reserve
)
6399 super
->create_offset
= ~((unsigned long long) 0);
6400 if (start
+ reserve
> super
->create_offset
)
6401 return 0; /* start overflows create_offset */
6402 super
->create_offset
= start
+ reserve
;
6404 return maxsize
- reserve
;
6407 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6409 if (level
< 0 || level
== 6 || level
== 4)
6412 /* if we have an orom prevent invalid raid levels */
6415 case 0: return imsm_orom_has_raid0(orom
);
6418 return imsm_orom_has_raid1e(orom
);
6419 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6420 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6421 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6424 return 1; /* not on an Intel RAID platform so anything goes */
6430 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6431 int dpa
, int verbose
)
6433 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6434 struct mdstat_ent
*memb
;
6440 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6441 if (memb
->metadata_version
&&
6442 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6443 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6444 !is_subarray(memb
->metadata_version
+9) &&
6446 struct dev_member
*dev
= memb
->members
;
6448 while(dev
&& (fd
< 0)) {
6449 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6450 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6452 fd
= open(path
, O_RDONLY
, 0);
6453 if (num
<= 0 || fd
< 0) {
6454 pr_vrb("Cannot open %s: %s\n",
6455 dev
->name
, strerror(errno
));
6461 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6462 struct mdstat_ent
*vol
;
6463 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6464 if (vol
->active
> 0 &&
6465 vol
->metadata_version
&&
6466 is_container_member(vol
, memb
->devnm
)) {
6471 if (*devlist
&& (found
< dpa
)) {
6472 dv
= xcalloc(1, sizeof(*dv
));
6473 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6474 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6477 dv
->next
= *devlist
;
6485 free_mdstat(mdstat
);
6490 static struct md_list
*
6491 get_loop_devices(void)
6494 struct md_list
*devlist
= NULL
;
6497 for(i
= 0; i
< 12; i
++) {
6498 dv
= xcalloc(1, sizeof(*dv
));
6499 dv
->devname
= xmalloc(40);
6500 sprintf(dv
->devname
, "/dev/loop%d", i
);
6508 static struct md_list
*
6509 get_devices(const char *hba_path
)
6511 struct md_list
*devlist
= NULL
;
6518 devlist
= get_loop_devices();
6521 /* scroll through /sys/dev/block looking for devices attached to
6524 dir
= opendir("/sys/dev/block");
6525 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6530 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6532 path
= devt_to_devpath(makedev(major
, minor
));
6535 if (!path_attached_to_hba(path
, hba_path
)) {
6542 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6544 fd2devname(fd
, buf
);
6547 pr_err("cannot open device: %s\n",
6552 dv
= xcalloc(1, sizeof(*dv
));
6553 dv
->devname
= xstrdup(buf
);
6560 devlist
= devlist
->next
;
6570 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6571 int verbose
, int *found
)
6573 struct md_list
*tmpdev
;
6575 struct supertype
*st
;
6577 /* first walk the list of devices to find a consistent set
6578 * that match the criterea, if that is possible.
6579 * We flag the ones we like with 'used'.
6582 st
= match_metadata_desc_imsm("imsm");
6584 pr_vrb("cannot allocate memory for imsm supertype\n");
6588 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6589 char *devname
= tmpdev
->devname
;
6591 struct supertype
*tst
;
6593 if (tmpdev
->used
> 1)
6595 tst
= dup_super(st
);
6597 pr_vrb("cannot allocate memory for imsm supertype\n");
6600 tmpdev
->container
= 0;
6601 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6603 dprintf("cannot open device %s: %s\n",
6604 devname
, strerror(errno
));
6606 } else if (!fstat_is_blkdev(dfd
, devname
, &rdev
)) {
6608 } else if (must_be_container(dfd
)) {
6609 struct supertype
*cst
;
6610 cst
= super_by_fd(dfd
, NULL
);
6612 dprintf("cannot recognize container type %s\n",
6615 } else if (tst
->ss
!= st
->ss
) {
6616 dprintf("non-imsm container - ignore it: %s\n",
6619 } else if (!tst
->ss
->load_container
||
6620 tst
->ss
->load_container(tst
, dfd
, NULL
))
6623 tmpdev
->container
= 1;
6626 cst
->ss
->free_super(cst
);
6628 tmpdev
->st_rdev
= rdev
;
6629 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6630 dprintf("no RAID superblock on %s\n",
6633 } else if (tst
->ss
->compare_super
== NULL
) {
6634 dprintf("Cannot assemble %s metadata on %s\n",
6635 tst
->ss
->name
, devname
);
6641 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6642 /* Ignore unrecognised devices during auto-assembly */
6647 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6649 if (st
->minor_version
== -1)
6650 st
->minor_version
= tst
->minor_version
;
6652 if (memcmp(info
.uuid
, uuid_zero
,
6653 sizeof(int[4])) == 0) {
6654 /* this is a floating spare. It cannot define
6655 * an array unless there are no more arrays of
6656 * this type to be found. It can be included
6657 * in an array of this type though.
6663 if (st
->ss
!= tst
->ss
||
6664 st
->minor_version
!= tst
->minor_version
||
6665 st
->ss
->compare_super(st
, tst
) != 0) {
6666 /* Some mismatch. If exactly one array matches this host,
6667 * we can resolve on that one.
6668 * Or, if we are auto assembling, we just ignore the second
6671 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6677 dprintf("found: devname: %s\n", devname
);
6681 tst
->ss
->free_super(tst
);
6685 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6686 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6687 for (iter
= head
; iter
; iter
= iter
->next
) {
6688 dprintf("content->text_version: %s vol\n",
6689 iter
->text_version
);
6690 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6691 /* do not assemble arrays with unsupported
6693 dprintf("Cannot activate member %s.\n",
6694 iter
->text_version
);
6701 dprintf("No valid super block on device list: err: %d %p\n",
6705 dprintf("no more devices to examine\n");
6708 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6709 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6711 if (count
< tmpdev
->found
)
6714 count
-= tmpdev
->found
;
6717 if (tmpdev
->used
== 1)
6722 st
->ss
->free_super(st
);
6726 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6729 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6731 const struct orom_entry
*entry
;
6732 struct devid_list
*dv
, *devid_list
;
6737 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6738 if (strstr(idev
->path
, hba_path
))
6742 if (!idev
|| !idev
->dev_id
)
6745 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6747 if (!entry
|| !entry
->devid_list
)
6750 devid_list
= entry
->devid_list
;
6751 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6752 struct md_list
*devlist
;
6753 struct sys_dev
*device
= NULL
;
6758 device
= device_by_id_and_path(dv
->devid
, hba_path
);
6760 device
= device_by_id(dv
->devid
);
6763 hpath
= device
->path
;
6767 devlist
= get_devices(hpath
);
6768 /* if no intel devices return zero volumes */
6769 if (devlist
== NULL
)
6772 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
6774 dprintf("path: %s active arrays: %d\n", hpath
, count
);
6775 if (devlist
== NULL
)
6779 count
+= count_volumes_list(devlist
,
6783 dprintf("found %d count: %d\n", found
, count
);
6786 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
6789 struct md_list
*dv
= devlist
;
6790 devlist
= devlist
->next
;
6798 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
6802 if (hba
->type
== SYS_DEV_VMD
) {
6803 struct sys_dev
*dev
;
6806 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
6807 if (dev
->type
== SYS_DEV_VMD
)
6808 count
+= __count_volumes(dev
->path
, dpa
,
6813 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
6816 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6818 /* up to 512 if the plaform supports it, otherwise the platform max.
6819 * 128 if no platform detected
6821 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6823 return min(512, (1 << fs
));
6827 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6828 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6830 /* check/set platform and metadata limits/defaults */
6831 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6832 pr_vrb("platform supports a maximum of %d disks per array\n",
6837 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6838 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6839 pr_vrb("platform does not support raid%d with %d disk%s\n",
6840 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6844 if (*chunk
== 0 || *chunk
== UnSet
)
6845 *chunk
= imsm_default_chunk(super
->orom
);
6847 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6848 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
6852 if (layout
!= imsm_level_to_layout(level
)) {
6854 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
6855 else if (level
== 10)
6856 pr_vrb("imsm raid 10 only supports the n2 layout\n");
6858 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
6863 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6864 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6865 pr_vrb("platform does not support a volume size over 2TB\n");
6872 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6873 * FIX ME add ahci details
6875 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6876 int layout
, int raiddisks
, int *chunk
,
6877 unsigned long long size
,
6878 unsigned long long data_offset
,
6880 unsigned long long *freesize
,
6884 struct intel_super
*super
= st
->sb
;
6885 struct imsm_super
*mpb
;
6887 unsigned long long pos
= 0;
6888 unsigned long long maxsize
;
6892 /* We must have the container info already read in. */
6896 mpb
= super
->anchor
;
6898 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6899 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6903 /* General test: make sure there is space for
6904 * 'raiddisks' device extents of size 'size' at a given
6907 unsigned long long minsize
= size
;
6908 unsigned long long start_offset
= MaxSector
;
6911 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6912 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6917 e
= get_extents(super
, dl
);
6920 unsigned long long esize
;
6921 esize
= e
[i
].start
- pos
;
6922 if (esize
>= minsize
)
6924 if (found
&& start_offset
== MaxSector
) {
6927 } else if (found
&& pos
!= start_offset
) {
6931 pos
= e
[i
].start
+ e
[i
].size
;
6933 } while (e
[i
-1].size
);
6938 if (dcnt
< raiddisks
) {
6940 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6947 /* This device must be a member of the set */
6948 if (!stat_is_blkdev(dev
, &rdev
))
6950 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6951 if (dl
->major
== (int)major(rdev
) &&
6952 dl
->minor
== (int)minor(rdev
))
6957 pr_err("%s is not in the same imsm set\n", dev
);
6959 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6960 /* If a volume is present then the current creation attempt
6961 * cannot incorporate new spares because the orom may not
6962 * understand this configuration (all member disks must be
6963 * members of each array in the container).
6965 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6966 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6968 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6969 mpb
->num_disks
!= raiddisks
) {
6970 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6974 /* retrieve the largest free space block */
6975 e
= get_extents(super
, dl
);
6980 unsigned long long esize
;
6982 esize
= e
[i
].start
- pos
;
6983 if (esize
>= maxsize
)
6985 pos
= e
[i
].start
+ e
[i
].size
;
6987 } while (e
[i
-1].size
);
6992 pr_err("unable to determine free space for: %s\n",
6996 if (maxsize
< size
) {
6998 pr_err("%s not enough space (%llu < %llu)\n",
6999 dev
, maxsize
, size
);
7003 /* count total number of extents for merge */
7005 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7007 i
+= dl
->extent_cnt
;
7009 maxsize
= merge_extents(super
, i
);
7011 if (!check_env("IMSM_NO_PLATFORM") &&
7012 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7013 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7017 if (maxsize
< size
|| maxsize
== 0) {
7020 pr_err("no free space left on device. Aborting...\n");
7022 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7028 *freesize
= maxsize
;
7031 int count
= count_volumes(super
->hba
,
7032 super
->orom
->dpa
, verbose
);
7033 if (super
->orom
->vphba
<= count
) {
7034 pr_vrb("platform does not support more than %d raid volumes.\n",
7035 super
->orom
->vphba
);
7042 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7043 unsigned long long size
, int chunk
,
7044 unsigned long long *freesize
)
7046 struct intel_super
*super
= st
->sb
;
7047 struct imsm_super
*mpb
= super
->anchor
;
7052 unsigned long long maxsize
;
7053 unsigned long long minsize
;
7057 /* find the largest common start free region of the possible disks */
7061 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7067 /* don't activate new spares if we are orom constrained
7068 * and there is already a volume active in the container
7070 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7073 e
= get_extents(super
, dl
);
7076 for (i
= 1; e
[i
-1].size
; i
++)
7084 maxsize
= merge_extents(super
, extent_cnt
);
7088 minsize
= chunk
* 2;
7090 if (cnt
< raiddisks
||
7091 (super
->orom
&& used
&& used
!= raiddisks
) ||
7092 maxsize
< minsize
||
7094 pr_err("not enough devices with space to create array.\n");
7095 return 0; /* No enough free spaces large enough */
7106 if (!check_env("IMSM_NO_PLATFORM") &&
7107 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7108 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7112 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7114 dl
->raiddisk
= cnt
++;
7118 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7123 static int reserve_space(struct supertype
*st
, int raiddisks
,
7124 unsigned long long size
, int chunk
,
7125 unsigned long long *freesize
)
7127 struct intel_super
*super
= st
->sb
;
7132 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7135 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7137 dl
->raiddisk
= cnt
++;
7144 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7145 int raiddisks
, int *chunk
, unsigned long long size
,
7146 unsigned long long data_offset
,
7147 char *dev
, unsigned long long *freesize
,
7148 int consistency_policy
, int verbose
)
7155 * if given unused devices create a container
7156 * if given given devices in a container create a member volume
7158 if (level
== LEVEL_CONTAINER
) {
7159 /* Must be a fresh device to add to a container */
7160 return validate_geometry_imsm_container(st
, level
, layout
,
7170 struct intel_super
*super
= st
->sb
;
7171 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7172 raiddisks
, chunk
, size
,
7175 /* we are being asked to automatically layout a
7176 * new volume based on the current contents of
7177 * the container. If the the parameters can be
7178 * satisfied reserve_space will record the disks,
7179 * start offset, and size of the volume to be
7180 * created. add_to_super and getinfo_super
7181 * detect when autolayout is in progress.
7183 /* assuming that freesize is always given when array is
7185 if (super
->orom
&& freesize
) {
7187 count
= count_volumes(super
->hba
,
7188 super
->orom
->dpa
, verbose
);
7189 if (super
->orom
->vphba
<= count
) {
7190 pr_vrb("platform does not support more than %d raid volumes.\n",
7191 super
->orom
->vphba
);
7196 return reserve_space(st
, raiddisks
, size
,
7202 /* creating in a given container */
7203 return validate_geometry_imsm_volume(st
, level
, layout
,
7204 raiddisks
, chunk
, size
,
7206 dev
, freesize
, verbose
);
7209 /* This device needs to be a device in an 'imsm' container */
7210 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7213 pr_err("Cannot create this array on device %s\n",
7218 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7220 pr_err("Cannot open %s: %s\n",
7221 dev
, strerror(errno
));
7224 /* Well, it is in use by someone, maybe an 'imsm' container. */
7225 cfd
= open_container(fd
);
7229 pr_err("Cannot use %s: It is busy\n",
7233 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7234 if (sra
&& sra
->array
.major_version
== -1 &&
7235 strcmp(sra
->text_version
, "imsm") == 0)
7239 /* This is a member of a imsm container. Load the container
7240 * and try to create a volume
7242 struct intel_super
*super
;
7244 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7246 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7248 return validate_geometry_imsm_volume(st
, level
, layout
,
7250 size
, data_offset
, dev
,
7257 pr_err("failed container membership check\n");
7263 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7265 struct intel_super
*super
= st
->sb
;
7267 if (level
&& *level
== UnSet
)
7268 *level
= LEVEL_CONTAINER
;
7270 if (level
&& layout
&& *layout
== UnSet
)
7271 *layout
= imsm_level_to_layout(*level
);
7273 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7274 *chunk
= imsm_default_chunk(super
->orom
);
7277 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7279 static int kill_subarray_imsm(struct supertype
*st
)
7281 /* remove the subarray currently referenced by ->current_vol */
7283 struct intel_dev
**dp
;
7284 struct intel_super
*super
= st
->sb
;
7285 __u8 current_vol
= super
->current_vol
;
7286 struct imsm_super
*mpb
= super
->anchor
;
7288 if (super
->current_vol
< 0)
7290 super
->current_vol
= -1; /* invalidate subarray cursor */
7292 /* block deletions that would change the uuid of active subarrays
7294 * FIXME when immutable ids are available, but note that we'll
7295 * also need to fixup the invalidated/active subarray indexes in
7298 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7301 if (i
< current_vol
)
7303 sprintf(subarray
, "%u", i
);
7304 if (is_subarray_active(subarray
, st
->devnm
)) {
7305 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7312 if (st
->update_tail
) {
7313 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7315 u
->type
= update_kill_array
;
7316 u
->dev_idx
= current_vol
;
7317 append_metadata_update(st
, u
, sizeof(*u
));
7322 for (dp
= &super
->devlist
; *dp
;)
7323 if ((*dp
)->index
== current_vol
) {
7326 handle_missing(super
, (*dp
)->dev
);
7327 if ((*dp
)->index
> current_vol
)
7332 /* no more raid devices, all active components are now spares,
7333 * but of course failed are still failed
7335 if (--mpb
->num_raid_devs
== 0) {
7338 for (d
= super
->disks
; d
; d
= d
->next
)
7343 super
->updates_pending
++;
7348 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7349 char *update
, struct mddev_ident
*ident
)
7351 /* update the subarray currently referenced by ->current_vol */
7352 struct intel_super
*super
= st
->sb
;
7353 struct imsm_super
*mpb
= super
->anchor
;
7355 if (strcmp(update
, "name") == 0) {
7356 char *name
= ident
->name
;
7360 if (is_subarray_active(subarray
, st
->devnm
)) {
7361 pr_err("Unable to update name of active subarray\n");
7365 if (!check_name(super
, name
, 0))
7368 vol
= strtoul(subarray
, &ep
, 10);
7369 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7372 if (st
->update_tail
) {
7373 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7375 u
->type
= update_rename_array
;
7377 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7378 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7379 append_metadata_update(st
, u
, sizeof(*u
));
7381 struct imsm_dev
*dev
;
7384 dev
= get_imsm_dev(super
, vol
);
7385 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
7386 dev
->volume
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7387 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7388 dev
= get_imsm_dev(super
, i
);
7389 handle_missing(super
, dev
);
7391 super
->updates_pending
++;
7393 } else if (strcmp(update
, "ppl") == 0 ||
7394 strcmp(update
, "no-ppl") == 0) {
7397 int vol
= strtoul(subarray
, &ep
, 10);
7399 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7402 if (strcmp(update
, "ppl") == 0)
7403 new_policy
= RWH_DISTRIBUTED
;
7405 new_policy
= RWH_OFF
;
7407 if (st
->update_tail
) {
7408 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7410 u
->type
= update_rwh_policy
;
7412 u
->new_policy
= new_policy
;
7413 append_metadata_update(st
, u
, sizeof(*u
));
7415 struct imsm_dev
*dev
;
7417 dev
= get_imsm_dev(super
, vol
);
7418 dev
->rwh_policy
= new_policy
;
7419 super
->updates_pending
++;
7427 static int is_gen_migration(struct imsm_dev
*dev
)
7432 if (!dev
->vol
.migr_state
)
7435 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7441 static int is_rebuilding(struct imsm_dev
*dev
)
7443 struct imsm_map
*migr_map
;
7445 if (!dev
->vol
.migr_state
)
7448 if (migr_type(dev
) != MIGR_REBUILD
)
7451 migr_map
= get_imsm_map(dev
, MAP_1
);
7453 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7459 static int is_initializing(struct imsm_dev
*dev
)
7461 struct imsm_map
*migr_map
;
7463 if (!dev
->vol
.migr_state
)
7466 if (migr_type(dev
) != MIGR_INIT
)
7469 migr_map
= get_imsm_map(dev
, MAP_1
);
7471 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7477 static void update_recovery_start(struct intel_super
*super
,
7478 struct imsm_dev
*dev
,
7479 struct mdinfo
*array
)
7481 struct mdinfo
*rebuild
= NULL
;
7485 if (!is_rebuilding(dev
))
7488 /* Find the rebuild target, but punt on the dual rebuild case */
7489 for (d
= array
->devs
; d
; d
= d
->next
)
7490 if (d
->recovery_start
== 0) {
7497 /* (?) none of the disks are marked with
7498 * IMSM_ORD_REBUILD, so assume they are missing and the
7499 * disk_ord_tbl was not correctly updated
7501 dprintf("failed to locate out-of-sync disk\n");
7505 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7506 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7509 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7511 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7513 /* Given a container loaded by load_super_imsm_all,
7514 * extract information about all the arrays into
7516 * If 'subarray' is given, just extract info about that array.
7518 * For each imsm_dev create an mdinfo, fill it in,
7519 * then look for matching devices in super->disks
7520 * and create appropriate device mdinfo.
7522 struct intel_super
*super
= st
->sb
;
7523 struct imsm_super
*mpb
= super
->anchor
;
7524 struct mdinfo
*rest
= NULL
;
7528 int spare_disks
= 0;
7530 /* do not assemble arrays when not all attributes are supported */
7531 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7533 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7536 /* count spare devices, not used in maps
7538 for (d
= super
->disks
; d
; d
= d
->next
)
7542 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7543 struct imsm_dev
*dev
;
7544 struct imsm_map
*map
;
7545 struct imsm_map
*map2
;
7546 struct mdinfo
*this;
7552 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7555 dev
= get_imsm_dev(super
, i
);
7556 map
= get_imsm_map(dev
, MAP_0
);
7557 map2
= get_imsm_map(dev
, MAP_1
);
7559 /* do not publish arrays that are in the middle of an
7560 * unsupported migration
7562 if (dev
->vol
.migr_state
&&
7563 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7564 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7568 /* do not publish arrays that are not support by controller's
7572 this = xmalloc(sizeof(*this));
7574 super
->current_vol
= i
;
7575 getinfo_super_imsm_volume(st
, this, NULL
);
7577 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7578 /* mdadm does not support all metadata features- set the bit in all arrays state */
7579 if (!validate_geometry_imsm_orom(super
,
7580 get_imsm_raid_level(map
), /* RAID level */
7581 imsm_level_to_layout(get_imsm_raid_level(map
)),
7582 map
->num_members
, /* raid disks */
7583 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
7585 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7587 this->array
.state
|=
7588 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7589 (1<<MD_SB_BLOCK_VOLUME
);
7592 /* if array has bad blocks, set suitable bit in all arrays state */
7594 this->array
.state
|=
7595 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7596 (1<<MD_SB_BLOCK_VOLUME
);
7598 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7599 unsigned long long recovery_start
;
7600 struct mdinfo
*info_d
;
7607 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7608 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7609 for (d
= super
->disks
; d
; d
= d
->next
)
7610 if (d
->index
== idx
)
7613 recovery_start
= MaxSector
;
7616 if (d
&& is_failed(&d
->disk
))
7618 if (ord
& IMSM_ORD_REBUILD
)
7622 * if we skip some disks the array will be assmebled degraded;
7623 * reset resync start to avoid a dirty-degraded
7624 * situation when performing the intial sync
7626 * FIXME handle dirty degraded
7628 if ((skip
|| recovery_start
== 0) &&
7629 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
))
7630 this->resync_start
= MaxSector
;
7634 info_d
= xcalloc(1, sizeof(*info_d
));
7635 info_d
->next
= this->devs
;
7636 this->devs
= info_d
;
7638 info_d
->disk
.number
= d
->index
;
7639 info_d
->disk
.major
= d
->major
;
7640 info_d
->disk
.minor
= d
->minor
;
7641 info_d
->disk
.raid_disk
= slot
;
7642 info_d
->recovery_start
= recovery_start
;
7644 if (slot
< map2
->num_members
)
7645 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7647 this->array
.spare_disks
++;
7649 if (slot
< map
->num_members
)
7650 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7652 this->array
.spare_disks
++;
7654 if (info_d
->recovery_start
== MaxSector
)
7655 this->array
.working_disks
++;
7657 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7658 info_d
->data_offset
= pba_of_lba0(map
);
7660 if (map
->raid_level
== 5) {
7661 info_d
->component_size
=
7662 num_data_stripes(map
) *
7663 map
->blocks_per_strip
;
7664 info_d
->ppl_sector
= this->ppl_sector
;
7665 info_d
->ppl_size
= this->ppl_size
;
7667 info_d
->component_size
= blocks_per_member(map
);
7669 info_d
->consistency_policy
= this->consistency_policy
;
7671 info_d
->bb
.supported
= 1;
7672 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7673 info_d
->data_offset
,
7674 info_d
->component_size
,
7677 /* now that the disk list is up-to-date fixup recovery_start */
7678 update_recovery_start(super
, dev
, this);
7679 this->array
.spare_disks
+= spare_disks
;
7681 /* check for reshape */
7682 if (this->reshape_active
== 1)
7683 recover_backup_imsm(st
, this);
7690 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7691 int failed
, int look_in_map
)
7693 struct imsm_map
*map
;
7695 map
= get_imsm_map(dev
, look_in_map
);
7698 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7699 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7701 switch (get_imsm_raid_level(map
)) {
7703 return IMSM_T_STATE_FAILED
;
7706 if (failed
< map
->num_members
)
7707 return IMSM_T_STATE_DEGRADED
;
7709 return IMSM_T_STATE_FAILED
;
7714 * check to see if any mirrors have failed, otherwise we
7715 * are degraded. Even numbered slots are mirrored on
7719 /* gcc -Os complains that this is unused */
7720 int insync
= insync
;
7722 for (i
= 0; i
< map
->num_members
; i
++) {
7723 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
7724 int idx
= ord_to_idx(ord
);
7725 struct imsm_disk
*disk
;
7727 /* reset the potential in-sync count on even-numbered
7728 * slots. num_copies is always 2 for imsm raid10
7733 disk
= get_imsm_disk(super
, idx
);
7734 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7737 /* no in-sync disks left in this mirror the
7741 return IMSM_T_STATE_FAILED
;
7744 return IMSM_T_STATE_DEGRADED
;
7748 return IMSM_T_STATE_DEGRADED
;
7750 return IMSM_T_STATE_FAILED
;
7756 return map
->map_state
;
7759 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7764 struct imsm_disk
*disk
;
7765 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7766 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7767 struct imsm_map
*map_for_loop
;
7772 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7773 * disks that are being rebuilt. New failures are recorded to
7774 * map[0]. So we look through all the disks we started with and
7775 * see if any failures are still present, or if any new ones
7779 if (prev
&& (map
->num_members
< prev
->num_members
))
7780 map_for_loop
= prev
;
7782 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
7784 /* when MAP_X is passed both maps failures are counted
7787 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
7788 i
< prev
->num_members
) {
7789 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
7790 idx_1
= ord_to_idx(ord
);
7792 disk
= get_imsm_disk(super
, idx_1
);
7793 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7796 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
7797 i
< map
->num_members
) {
7798 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
7799 idx
= ord_to_idx(ord
);
7802 disk
= get_imsm_disk(super
, idx
);
7803 if (!disk
|| is_failed(disk
) ||
7804 ord
& IMSM_ORD_REBUILD
)
7813 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7816 struct intel_super
*super
= c
->sb
;
7817 struct imsm_super
*mpb
= super
->anchor
;
7818 struct imsm_update_prealloc_bb_mem u
;
7820 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7821 pr_err("subarry index %d, out of range\n", atoi(inst
));
7825 dprintf("imsm: open_new %s\n", inst
);
7826 a
->info
.container_member
= atoi(inst
);
7828 u
.type
= update_prealloc_badblocks_mem
;
7829 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
7834 static int is_resyncing(struct imsm_dev
*dev
)
7836 struct imsm_map
*migr_map
;
7838 if (!dev
->vol
.migr_state
)
7841 if (migr_type(dev
) == MIGR_INIT
||
7842 migr_type(dev
) == MIGR_REPAIR
)
7845 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7848 migr_map
= get_imsm_map(dev
, MAP_1
);
7850 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
7851 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
7857 /* return true if we recorded new information */
7858 static int mark_failure(struct intel_super
*super
,
7859 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7863 struct imsm_map
*map
;
7864 char buf
[MAX_RAID_SERIAL_LEN
+3];
7865 unsigned int len
, shift
= 0;
7867 /* new failures are always set in map[0] */
7868 map
= get_imsm_map(dev
, MAP_0
);
7870 slot
= get_imsm_disk_slot(map
, idx
);
7874 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7875 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7878 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7879 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7881 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7882 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7883 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7885 disk
->status
|= FAILED_DISK
;
7886 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7887 /* mark failures in second map if second map exists and this disk
7889 * This is valid for migration, initialization and rebuild
7891 if (dev
->vol
.migr_state
) {
7892 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7893 int slot2
= get_imsm_disk_slot(map2
, idx
);
7895 if (slot2
< map2
->num_members
&& slot2
>= 0)
7896 set_imsm_ord_tbl_ent(map2
, slot2
,
7897 idx
| IMSM_ORD_REBUILD
);
7899 if (map
->failed_disk_num
== 0xff)
7900 map
->failed_disk_num
= slot
;
7902 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
7907 static void mark_missing(struct intel_super
*super
,
7908 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7910 mark_failure(super
, dev
, disk
, idx
);
7912 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7915 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7916 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7919 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7923 if (!super
->missing
)
7926 /* When orom adds replacement for missing disk it does
7927 * not remove entry of missing disk, but just updates map with
7928 * new added disk. So it is not enough just to test if there is
7929 * any missing disk, we have to look if there are any failed disks
7930 * in map to stop migration */
7932 dprintf("imsm: mark missing\n");
7933 /* end process for initialization and rebuild only
7935 if (is_gen_migration(dev
) == 0) {
7939 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7940 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7943 end_migration(dev
, super
, map_state
);
7945 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7946 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
7947 super
->updates_pending
++;
7950 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7953 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7954 unsigned long long array_blocks
;
7955 struct imsm_map
*map
;
7957 if (used_disks
== 0) {
7958 /* when problems occures
7959 * return current array_blocks value
7961 array_blocks
= __le32_to_cpu(dev
->size_high
);
7962 array_blocks
= array_blocks
<< 32;
7963 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7965 return array_blocks
;
7968 /* set array size in metadata
7970 if (new_size
<= 0) {
7971 /* OLCE size change is caused by added disks
7973 map
= get_imsm_map(dev
, MAP_0
);
7974 array_blocks
= blocks_per_member(map
) * used_disks
;
7976 /* Online Volume Size Change
7977 * Using available free space
7979 array_blocks
= new_size
;
7982 array_blocks
= round_size_to_mb(array_blocks
, used_disks
);
7983 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7984 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7986 return array_blocks
;
7989 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7991 static void imsm_progress_container_reshape(struct intel_super
*super
)
7993 /* if no device has a migr_state, but some device has a
7994 * different number of members than the previous device, start
7995 * changing the number of devices in this device to match
7998 struct imsm_super
*mpb
= super
->anchor
;
7999 int prev_disks
= -1;
8003 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8004 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8005 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8006 struct imsm_map
*map2
;
8007 int prev_num_members
;
8009 if (dev
->vol
.migr_state
)
8012 if (prev_disks
== -1)
8013 prev_disks
= map
->num_members
;
8014 if (prev_disks
== map
->num_members
)
8017 /* OK, this array needs to enter reshape mode.
8018 * i.e it needs a migr_state
8021 copy_map_size
= sizeof_imsm_map(map
);
8022 prev_num_members
= map
->num_members
;
8023 map
->num_members
= prev_disks
;
8024 dev
->vol
.migr_state
= 1;
8025 dev
->vol
.curr_migr_unit
= 0;
8026 set_migr_type(dev
, MIGR_GEN_MIGR
);
8027 for (i
= prev_num_members
;
8028 i
< map
->num_members
; i
++)
8029 set_imsm_ord_tbl_ent(map
, i
, i
);
8030 map2
= get_imsm_map(dev
, MAP_1
);
8031 /* Copy the current map */
8032 memcpy(map2
, map
, copy_map_size
);
8033 map2
->num_members
= prev_num_members
;
8035 imsm_set_array_size(dev
, -1);
8036 super
->clean_migration_record_by_mdmon
= 1;
8037 super
->updates_pending
++;
8041 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8042 * states are handled in imsm_set_disk() with one exception, when a
8043 * resync is stopped due to a new failure this routine will set the
8044 * 'degraded' state for the array.
8046 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8048 int inst
= a
->info
.container_member
;
8049 struct intel_super
*super
= a
->container
->sb
;
8050 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8051 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8052 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8053 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8054 __u32 blocks_per_unit
;
8056 if (dev
->vol
.migr_state
&&
8057 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8058 /* array state change is blocked due to reshape action
8060 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8061 * - finish the reshape (if last_checkpoint is big and action != reshape)
8062 * - update curr_migr_unit
8064 if (a
->curr_action
== reshape
) {
8065 /* still reshaping, maybe update curr_migr_unit */
8066 goto mark_checkpoint
;
8068 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8069 /* for some reason we aborted the reshape.
8071 * disable automatic metadata rollback
8072 * user action is required to recover process
8075 struct imsm_map
*map2
=
8076 get_imsm_map(dev
, MAP_1
);
8077 dev
->vol
.migr_state
= 0;
8078 set_migr_type(dev
, 0);
8079 dev
->vol
.curr_migr_unit
= 0;
8081 sizeof_imsm_map(map2
));
8082 super
->updates_pending
++;
8085 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8086 unsigned long long array_blocks
;
8090 used_disks
= imsm_num_data_members(dev
, MAP_0
);
8091 if (used_disks
> 0) {
8093 blocks_per_member(map
) *
8096 round_size_to_mb(array_blocks
,
8098 a
->info
.custom_array_size
= array_blocks
;
8099 /* encourage manager to update array
8103 a
->check_reshape
= 1;
8105 /* finalize online capacity expansion/reshape */
8106 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8108 mdi
->disk
.raid_disk
,
8111 imsm_progress_container_reshape(super
);
8116 /* before we activate this array handle any missing disks */
8117 if (consistent
== 2)
8118 handle_missing(super
, dev
);
8120 if (consistent
== 2 &&
8121 (!is_resync_complete(&a
->info
) ||
8122 map_state
!= IMSM_T_STATE_NORMAL
||
8123 dev
->vol
.migr_state
))
8126 if (is_resync_complete(&a
->info
)) {
8127 /* complete intialization / resync,
8128 * recovery and interrupted recovery is completed in
8131 if (is_resyncing(dev
)) {
8132 dprintf("imsm: mark resync done\n");
8133 end_migration(dev
, super
, map_state
);
8134 super
->updates_pending
++;
8135 a
->last_checkpoint
= 0;
8137 } else if ((!is_resyncing(dev
) && !failed
) &&
8138 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8139 /* mark the start of the init process if nothing is failed */
8140 dprintf("imsm: mark resync start\n");
8141 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8142 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8144 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8145 super
->updates_pending
++;
8149 /* skip checkpointing for general migration,
8150 * it is controlled in mdadm
8152 if (is_gen_migration(dev
))
8153 goto skip_mark_checkpoint
;
8155 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8156 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8157 if (blocks_per_unit
) {
8161 units
= a
->last_checkpoint
/ blocks_per_unit
;
8164 /* check that we did not overflow 32-bits, and that
8165 * curr_migr_unit needs updating
8167 if (units32
== units
&&
8169 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8170 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8171 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8172 super
->updates_pending
++;
8176 skip_mark_checkpoint
:
8177 /* mark dirty / clean */
8178 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8179 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8180 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8182 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8184 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8185 if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
8186 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8188 super
->updates_pending
++;
8194 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8196 int inst
= a
->info
.container_member
;
8197 struct intel_super
*super
= a
->container
->sb
;
8198 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8199 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8201 if (slot
> map
->num_members
) {
8202 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8203 slot
, map
->num_members
- 1);
8210 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8213 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8215 int inst
= a
->info
.container_member
;
8216 struct intel_super
*super
= a
->container
->sb
;
8217 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8218 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8219 struct imsm_disk
*disk
;
8221 int recovery_not_finished
= 0;
8226 ord
= imsm_disk_slot_to_ord(a
, n
);
8230 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8231 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8233 /* check for new failures */
8234 if (state
& DS_FAULTY
) {
8235 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8236 super
->updates_pending
++;
8239 /* check if in_sync */
8240 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8241 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8243 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8244 super
->updates_pending
++;
8247 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8248 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8250 /* check if recovery complete, newly degraded, or failed */
8251 dprintf("imsm: Detected transition to state ");
8252 switch (map_state
) {
8253 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8254 dprintf("normal: ");
8255 if (is_rebuilding(dev
)) {
8256 dprintf_cont("while rebuilding");
8257 /* check if recovery is really finished */
8258 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8259 if (mdi
->recovery_start
!= MaxSector
) {
8260 recovery_not_finished
= 1;
8263 if (recovery_not_finished
) {
8265 dprintf("Rebuild has not finished yet, state not changed");
8266 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8267 a
->last_checkpoint
= mdi
->recovery_start
;
8268 super
->updates_pending
++;
8272 end_migration(dev
, super
, map_state
);
8273 map
= get_imsm_map(dev
, MAP_0
);
8274 map
->failed_disk_num
= ~0;
8275 super
->updates_pending
++;
8276 a
->last_checkpoint
= 0;
8279 if (is_gen_migration(dev
)) {
8280 dprintf_cont("while general migration");
8281 if (a
->last_checkpoint
>= a
->info
.component_size
)
8282 end_migration(dev
, super
, map_state
);
8284 map
->map_state
= map_state
;
8285 map
= get_imsm_map(dev
, MAP_0
);
8286 map
->failed_disk_num
= ~0;
8287 super
->updates_pending
++;
8291 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8292 dprintf_cont("degraded: ");
8293 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8294 dprintf_cont("mark degraded");
8295 map
->map_state
= map_state
;
8296 super
->updates_pending
++;
8297 a
->last_checkpoint
= 0;
8300 if (is_rebuilding(dev
)) {
8301 dprintf_cont("while rebuilding.");
8302 if (map
->map_state
!= map_state
) {
8303 dprintf_cont(" Map state change");
8304 end_migration(dev
, super
, map_state
);
8305 super
->updates_pending
++;
8309 if (is_gen_migration(dev
)) {
8310 dprintf_cont("while general migration");
8311 if (a
->last_checkpoint
>= a
->info
.component_size
)
8312 end_migration(dev
, super
, map_state
);
8314 map
->map_state
= map_state
;
8315 manage_second_map(super
, dev
);
8317 super
->updates_pending
++;
8320 if (is_initializing(dev
)) {
8321 dprintf_cont("while initialization.");
8322 map
->map_state
= map_state
;
8323 super
->updates_pending
++;
8327 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8328 dprintf_cont("failed: ");
8329 if (is_gen_migration(dev
)) {
8330 dprintf_cont("while general migration");
8331 map
->map_state
= map_state
;
8332 super
->updates_pending
++;
8335 if (map
->map_state
!= map_state
) {
8336 dprintf_cont("mark failed");
8337 end_migration(dev
, super
, map_state
);
8338 super
->updates_pending
++;
8339 a
->last_checkpoint
= 0;
8344 dprintf_cont("state %i\n", map_state
);
8349 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8352 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8353 unsigned long long dsize
;
8354 unsigned long long sectors
;
8355 unsigned int sector_size
;
8357 get_dev_sector_size(fd
, NULL
, §or_size
);
8358 get_dev_size(fd
, NULL
, &dsize
);
8360 if (mpb_size
> sector_size
) {
8361 /* -1 to account for anchor */
8362 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8364 /* write the extended mpb to the sectors preceeding the anchor */
8365 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8369 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8370 sector_size
* sectors
) != sector_size
* sectors
)
8374 /* first block is stored on second to last sector of the disk */
8375 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8378 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8384 static void imsm_sync_metadata(struct supertype
*container
)
8386 struct intel_super
*super
= container
->sb
;
8388 dprintf("sync metadata: %d\n", super
->updates_pending
);
8389 if (!super
->updates_pending
)
8392 write_super_imsm(container
, 0);
8394 super
->updates_pending
= 0;
8397 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8399 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8400 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8403 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8407 if (dl
&& is_failed(&dl
->disk
))
8411 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8416 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8417 struct active_array
*a
, int activate_new
,
8418 struct mdinfo
*additional_test_list
)
8420 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8421 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8422 struct imsm_super
*mpb
= super
->anchor
;
8423 struct imsm_map
*map
;
8424 unsigned long long pos
;
8429 __u32 array_start
= 0;
8430 __u32 array_end
= 0;
8432 struct mdinfo
*test_list
;
8434 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8435 /* If in this array, skip */
8436 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8437 if (d
->state_fd
>= 0 &&
8438 d
->disk
.major
== dl
->major
&&
8439 d
->disk
.minor
== dl
->minor
) {
8440 dprintf("%x:%x already in array\n",
8441 dl
->major
, dl
->minor
);
8446 test_list
= additional_test_list
;
8448 if (test_list
->disk
.major
== dl
->major
&&
8449 test_list
->disk
.minor
== dl
->minor
) {
8450 dprintf("%x:%x already in additional test list\n",
8451 dl
->major
, dl
->minor
);
8454 test_list
= test_list
->next
;
8459 /* skip in use or failed drives */
8460 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8462 dprintf("%x:%x status (failed: %d index: %d)\n",
8463 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8467 /* skip pure spares when we are looking for partially
8468 * assimilated drives
8470 if (dl
->index
== -1 && !activate_new
)
8473 if (!drive_validate_sector_size(super
, dl
))
8476 /* Does this unused device have the requisite free space?
8477 * It needs to be able to cover all member volumes
8479 ex
= get_extents(super
, dl
);
8481 dprintf("cannot get extents\n");
8484 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8485 dev
= get_imsm_dev(super
, i
);
8486 map
= get_imsm_map(dev
, MAP_0
);
8488 /* check if this disk is already a member of
8491 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8497 array_start
= pba_of_lba0(map
);
8498 array_end
= array_start
+
8499 blocks_per_member(map
) - 1;
8502 /* check that we can start at pba_of_lba0 with
8503 * blocks_per_member of space
8505 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8509 pos
= ex
[j
].start
+ ex
[j
].size
;
8511 } while (ex
[j
-1].size
);
8518 if (i
< mpb
->num_raid_devs
) {
8519 dprintf("%x:%x does not have %u to %u available\n",
8520 dl
->major
, dl
->minor
, array_start
, array_end
);
8530 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8532 struct imsm_dev
*dev2
;
8533 struct imsm_map
*map
;
8539 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8541 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8542 if (state
== IMSM_T_STATE_FAILED
) {
8543 map
= get_imsm_map(dev2
, MAP_0
);
8546 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8548 * Check if failed disks are deleted from intel
8549 * disk list or are marked to be deleted
8551 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8552 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8554 * Do not rebuild the array if failed disks
8555 * from failed sub-array are not removed from
8559 is_failed(&idisk
->disk
) &&
8560 (idisk
->action
!= DISK_REMOVE
))
8568 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8569 struct metadata_update
**updates
)
8572 * Find a device with unused free space and use it to replace a
8573 * failed/vacant region in an array. We replace failed regions one a
8574 * array at a time. The result is that a new spare disk will be added
8575 * to the first failed array and after the monitor has finished
8576 * propagating failures the remainder will be consumed.
8578 * FIXME add a capability for mdmon to request spares from another
8582 struct intel_super
*super
= a
->container
->sb
;
8583 int inst
= a
->info
.container_member
;
8584 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8585 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8586 int failed
= a
->info
.array
.raid_disks
;
8587 struct mdinfo
*rv
= NULL
;
8590 struct metadata_update
*mu
;
8592 struct imsm_update_activate_spare
*u
;
8597 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8598 if ((d
->curr_state
& DS_FAULTY
) &&
8600 /* wait for Removal to happen */
8602 if (d
->state_fd
>= 0)
8606 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8607 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8609 if (imsm_reshape_blocks_arrays_changes(super
))
8612 /* Cannot activate another spare if rebuild is in progress already
8614 if (is_rebuilding(dev
)) {
8615 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8619 if (a
->info
.array
.level
== 4)
8620 /* No repair for takeovered array
8621 * imsm doesn't support raid4
8625 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8626 IMSM_T_STATE_DEGRADED
)
8629 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8630 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8635 * If there are any failed disks check state of the other volume.
8636 * Block rebuild if the another one is failed until failed disks
8637 * are removed from container.
8640 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8641 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8642 /* check if states of the other volumes allow for rebuild */
8643 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8645 allowed
= imsm_rebuild_allowed(a
->container
,
8653 /* For each slot, if it is not working, find a spare */
8654 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8655 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8656 if (d
->disk
.raid_disk
== i
)
8658 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8659 if (d
&& (d
->state_fd
>= 0))
8663 * OK, this device needs recovery. Try to re-add the
8664 * previous occupant of this slot, if this fails see if
8665 * we can continue the assimilation of a spare that was
8666 * partially assimilated, finally try to activate a new
8669 dl
= imsm_readd(super
, i
, a
);
8671 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
8673 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
8677 /* found a usable disk with enough space */
8678 di
= xcalloc(1, sizeof(*di
));
8680 /* dl->index will be -1 in the case we are activating a
8681 * pristine spare. imsm_process_update() will create a
8682 * new index in this case. Once a disk is found to be
8683 * failed in all member arrays it is kicked from the
8686 di
->disk
.number
= dl
->index
;
8688 /* (ab)use di->devs to store a pointer to the device
8691 di
->devs
= (struct mdinfo
*) dl
;
8693 di
->disk
.raid_disk
= i
;
8694 di
->disk
.major
= dl
->major
;
8695 di
->disk
.minor
= dl
->minor
;
8697 di
->recovery_start
= 0;
8698 di
->data_offset
= pba_of_lba0(map
);
8699 di
->component_size
= a
->info
.component_size
;
8700 di
->container_member
= inst
;
8701 di
->bb
.supported
= 1;
8702 if (dev
->rwh_policy
== RWH_DISTRIBUTED
) {
8703 di
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
8704 di
->ppl_sector
= get_ppl_sector(super
, inst
);
8705 di
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
) >> 9;
8707 super
->random
= random32();
8711 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
8712 i
, di
->data_offset
);
8716 /* No spares found */
8718 /* Now 'rv' has a list of devices to return.
8719 * Create a metadata_update record to update the
8720 * disk_ord_tbl for the array
8722 mu
= xmalloc(sizeof(*mu
));
8723 mu
->buf
= xcalloc(num_spares
,
8724 sizeof(struct imsm_update_activate_spare
));
8726 mu
->space_list
= NULL
;
8727 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
8728 mu
->next
= *updates
;
8729 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
8731 for (di
= rv
; di
; di
= di
->next
) {
8732 u
->type
= update_activate_spare
;
8733 u
->dl
= (struct dl
*) di
->devs
;
8735 u
->slot
= di
->disk
.raid_disk
;
8746 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
8748 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
8749 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8750 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8751 struct disk_info
*inf
= get_disk_info(u
);
8752 struct imsm_disk
*disk
;
8756 for (i
= 0; i
< map
->num_members
; i
++) {
8757 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
8758 for (j
= 0; j
< new_map
->num_members
; j
++)
8759 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
8766 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
8770 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8771 if (dl
->major
== major
&& dl
->minor
== minor
)
8776 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
8782 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8783 if (dl
->major
== major
&& dl
->minor
== minor
) {
8786 prev
->next
= dl
->next
;
8788 super
->disks
= dl
->next
;
8790 __free_imsm_disk(dl
);
8791 dprintf("removed %x:%x\n", major
, minor
);
8799 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
8801 static int add_remove_disk_update(struct intel_super
*super
)
8803 int check_degraded
= 0;
8806 /* add/remove some spares to/from the metadata/contrainer */
8807 while (super
->disk_mgmt_list
) {
8808 struct dl
*disk_cfg
;
8810 disk_cfg
= super
->disk_mgmt_list
;
8811 super
->disk_mgmt_list
= disk_cfg
->next
;
8812 disk_cfg
->next
= NULL
;
8814 if (disk_cfg
->action
== DISK_ADD
) {
8815 disk_cfg
->next
= super
->disks
;
8816 super
->disks
= disk_cfg
;
8818 dprintf("added %x:%x\n",
8819 disk_cfg
->major
, disk_cfg
->minor
);
8820 } else if (disk_cfg
->action
== DISK_REMOVE
) {
8821 dprintf("Disk remove action processed: %x.%x\n",
8822 disk_cfg
->major
, disk_cfg
->minor
);
8823 disk
= get_disk_super(super
,
8827 /* store action status */
8828 disk
->action
= DISK_REMOVE
;
8829 /* remove spare disks only */
8830 if (disk
->index
== -1) {
8831 remove_disk_super(super
,
8836 /* release allocate disk structure */
8837 __free_imsm_disk(disk_cfg
);
8840 return check_degraded
;
8843 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
8844 struct intel_super
*super
,
8847 struct intel_dev
*id
;
8848 void **tofree
= NULL
;
8851 dprintf("(enter)\n");
8852 if (u
->subdev
< 0 || u
->subdev
> 1) {
8853 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8856 if (space_list
== NULL
|| *space_list
== NULL
) {
8857 dprintf("imsm: Error: Memory is not allocated\n");
8861 for (id
= super
->devlist
; id
; id
= id
->next
) {
8862 if (id
->index
== (unsigned)u
->subdev
) {
8863 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8864 struct imsm_map
*map
;
8865 struct imsm_dev
*new_dev
=
8866 (struct imsm_dev
*)*space_list
;
8867 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8869 struct dl
*new_disk
;
8871 if (new_dev
== NULL
)
8873 *space_list
= **space_list
;
8874 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8875 map
= get_imsm_map(new_dev
, MAP_0
);
8877 dprintf("imsm: Error: migration in progress");
8881 to_state
= map
->map_state
;
8882 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8884 /* this should not happen */
8885 if (u
->new_disks
[0] < 0) {
8886 map
->failed_disk_num
=
8887 map
->num_members
- 1;
8888 to_state
= IMSM_T_STATE_DEGRADED
;
8890 to_state
= IMSM_T_STATE_NORMAL
;
8892 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8893 if (u
->new_level
> -1)
8894 map
->raid_level
= u
->new_level
;
8895 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8896 if ((u
->new_level
== 5) &&
8897 (migr_map
->raid_level
== 0)) {
8898 int ord
= map
->num_members
- 1;
8899 migr_map
->num_members
--;
8900 if (u
->new_disks
[0] < 0)
8901 ord
|= IMSM_ORD_REBUILD
;
8902 set_imsm_ord_tbl_ent(map
,
8903 map
->num_members
- 1,
8907 tofree
= (void **)dev
;
8909 /* update chunk size
8911 if (u
->new_chunksize
> 0) {
8912 unsigned long long num_data_stripes
;
8914 imsm_num_data_members(dev
, MAP_0
);
8916 if (used_disks
== 0)
8919 map
->blocks_per_strip
=
8920 __cpu_to_le16(u
->new_chunksize
* 2);
8922 (join_u32(dev
->size_low
, dev
->size_high
)
8924 num_data_stripes
/= map
->blocks_per_strip
;
8925 num_data_stripes
/= map
->num_domains
;
8926 set_num_data_stripes(map
, num_data_stripes
);
8931 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
8932 migr_map
->raid_level
== map
->raid_level
)
8935 if (u
->new_disks
[0] >= 0) {
8938 new_disk
= get_disk_super(super
,
8939 major(u
->new_disks
[0]),
8940 minor(u
->new_disks
[0]));
8941 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8942 major(u
->new_disks
[0]),
8943 minor(u
->new_disks
[0]),
8944 new_disk
, new_disk
->index
);
8945 if (new_disk
== NULL
)
8946 goto error_disk_add
;
8948 new_disk
->index
= map
->num_members
- 1;
8949 /* slot to fill in autolayout
8951 new_disk
->raiddisk
= new_disk
->index
;
8952 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8953 new_disk
->disk
.status
&= ~SPARE_DISK
;
8955 goto error_disk_add
;
8958 *tofree
= *space_list
;
8959 /* calculate new size
8961 imsm_set_array_size(new_dev
, -1);
8968 *space_list
= tofree
;
8972 dprintf("Error: imsm: Cannot find disk.\n");
8976 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8977 struct intel_super
*super
)
8979 struct intel_dev
*id
;
8982 dprintf("(enter)\n");
8983 if (u
->subdev
< 0 || u
->subdev
> 1) {
8984 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8988 for (id
= super
->devlist
; id
; id
= id
->next
) {
8989 if (id
->index
== (unsigned)u
->subdev
) {
8990 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8991 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8992 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8993 unsigned long long blocks_per_member
;
8994 unsigned long long num_data_stripes
;
8996 /* calculate new size
8998 blocks_per_member
= u
->new_size
/ used_disks
;
8999 num_data_stripes
= blocks_per_member
/
9000 map
->blocks_per_strip
;
9001 num_data_stripes
/= map
->num_domains
;
9002 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9003 u
->new_size
, blocks_per_member
,
9005 set_blocks_per_member(map
, blocks_per_member
);
9006 set_num_data_stripes(map
, num_data_stripes
);
9007 imsm_set_array_size(dev
, u
->new_size
);
9017 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9018 struct intel_super
*super
,
9019 struct active_array
*active_array
)
9021 struct imsm_super
*mpb
= super
->anchor
;
9022 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9023 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9024 struct imsm_map
*migr_map
;
9025 struct active_array
*a
;
9026 struct imsm_disk
*disk
;
9033 int second_map_created
= 0;
9035 for (; u
; u
= u
->next
) {
9036 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9041 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9046 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9051 /* count failures (excluding rebuilds and the victim)
9052 * to determine map[0] state
9055 for (i
= 0; i
< map
->num_members
; i
++) {
9058 disk
= get_imsm_disk(super
,
9059 get_imsm_disk_idx(dev
, i
, MAP_X
));
9060 if (!disk
|| is_failed(disk
))
9064 /* adding a pristine spare, assign a new index */
9065 if (dl
->index
< 0) {
9066 dl
->index
= super
->anchor
->num_disks
;
9067 super
->anchor
->num_disks
++;
9070 disk
->status
|= CONFIGURED_DISK
;
9071 disk
->status
&= ~SPARE_DISK
;
9074 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9075 if (!second_map_created
) {
9076 second_map_created
= 1;
9077 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9078 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9080 map
->map_state
= to_state
;
9081 migr_map
= get_imsm_map(dev
, MAP_1
);
9082 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9083 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9084 dl
->index
| IMSM_ORD_REBUILD
);
9086 /* update the family_num to mark a new container
9087 * generation, being careful to record the existing
9088 * family_num in orig_family_num to clean up after
9089 * earlier mdadm versions that neglected to set it.
9091 if (mpb
->orig_family_num
== 0)
9092 mpb
->orig_family_num
= mpb
->family_num
;
9093 mpb
->family_num
+= super
->random
;
9095 /* count arrays using the victim in the metadata */
9097 for (a
= active_array
; a
; a
= a
->next
) {
9098 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9099 map
= get_imsm_map(dev
, MAP_0
);
9101 if (get_imsm_disk_slot(map
, victim
) >= 0)
9105 /* delete the victim if it is no longer being
9111 /* We know that 'manager' isn't touching anything,
9112 * so it is safe to delete
9114 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9115 if ((*dlp
)->index
== victim
)
9118 /* victim may be on the missing list */
9120 for (dlp
= &super
->missing
; *dlp
;
9121 dlp
= &(*dlp
)->next
)
9122 if ((*dlp
)->index
== victim
)
9124 imsm_delete(super
, dlp
, victim
);
9131 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9132 struct intel_super
*super
,
9135 struct dl
*new_disk
;
9136 struct intel_dev
*id
;
9138 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9139 int disk_count
= u
->old_raid_disks
;
9140 void **tofree
= NULL
;
9141 int devices_to_reshape
= 1;
9142 struct imsm_super
*mpb
= super
->anchor
;
9144 unsigned int dev_id
;
9146 dprintf("(enter)\n");
9148 /* enable spares to use in array */
9149 for (i
= 0; i
< delta_disks
; i
++) {
9150 new_disk
= get_disk_super(super
,
9151 major(u
->new_disks
[i
]),
9152 minor(u
->new_disks
[i
]));
9153 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9154 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9155 new_disk
, new_disk
->index
);
9156 if (new_disk
== NULL
||
9157 (new_disk
->index
>= 0 &&
9158 new_disk
->index
< u
->old_raid_disks
))
9159 goto update_reshape_exit
;
9160 new_disk
->index
= disk_count
++;
9161 /* slot to fill in autolayout
9163 new_disk
->raiddisk
= new_disk
->index
;
9164 new_disk
->disk
.status
|=
9166 new_disk
->disk
.status
&= ~SPARE_DISK
;
9169 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9170 mpb
->num_raid_devs
);
9171 /* manage changes in volume
9173 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9174 void **sp
= *space_list
;
9175 struct imsm_dev
*newdev
;
9176 struct imsm_map
*newmap
, *oldmap
;
9178 for (id
= super
->devlist
; id
; id
= id
->next
) {
9179 if (id
->index
== dev_id
)
9188 /* Copy the dev, but not (all of) the map */
9189 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9190 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9191 newmap
= get_imsm_map(newdev
, MAP_0
);
9192 /* Copy the current map */
9193 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9194 /* update one device only
9196 if (devices_to_reshape
) {
9197 dprintf("imsm: modifying subdev: %i\n",
9199 devices_to_reshape
--;
9200 newdev
->vol
.migr_state
= 1;
9201 newdev
->vol
.curr_migr_unit
= 0;
9202 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9203 newmap
->num_members
= u
->new_raid_disks
;
9204 for (i
= 0; i
< delta_disks
; i
++) {
9205 set_imsm_ord_tbl_ent(newmap
,
9206 u
->old_raid_disks
+ i
,
9207 u
->old_raid_disks
+ i
);
9209 /* New map is correct, now need to save old map
9211 newmap
= get_imsm_map(newdev
, MAP_1
);
9212 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9214 imsm_set_array_size(newdev
, -1);
9217 sp
= (void **)id
->dev
;
9222 /* Clear migration record */
9223 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9226 *space_list
= tofree
;
9229 update_reshape_exit
:
9234 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9235 struct intel_super
*super
,
9238 struct imsm_dev
*dev
= NULL
;
9239 struct intel_dev
*dv
;
9240 struct imsm_dev
*dev_new
;
9241 struct imsm_map
*map
;
9245 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9246 if (dv
->index
== (unsigned int)u
->subarray
) {
9254 map
= get_imsm_map(dev
, MAP_0
);
9256 if (u
->direction
== R10_TO_R0
) {
9257 unsigned long long num_data_stripes
;
9259 map
->num_domains
= 1;
9260 num_data_stripes
= blocks_per_member(map
);
9261 num_data_stripes
/= map
->blocks_per_strip
;
9262 num_data_stripes
/= map
->num_domains
;
9263 set_num_data_stripes(map
, num_data_stripes
);
9265 /* Number of failed disks must be half of initial disk number */
9266 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9267 (map
->num_members
/ 2))
9270 /* iterate through devices to mark removed disks as spare */
9271 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9272 if (dm
->disk
.status
& FAILED_DISK
) {
9273 int idx
= dm
->index
;
9274 /* update indexes on the disk list */
9275 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9276 the index values will end up being correct.... NB */
9277 for (du
= super
->disks
; du
; du
= du
->next
)
9278 if (du
->index
> idx
)
9280 /* mark as spare disk */
9285 map
->num_members
= map
->num_members
/ 2;
9286 map
->map_state
= IMSM_T_STATE_NORMAL
;
9287 map
->num_domains
= 1;
9288 map
->raid_level
= 0;
9289 map
->failed_disk_num
= -1;
9292 if (u
->direction
== R0_TO_R10
) {
9294 /* update slots in current disk list */
9295 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9299 /* create new *missing* disks */
9300 for (i
= 0; i
< map
->num_members
; i
++) {
9301 space
= *space_list
;
9304 *space_list
= *space
;
9306 memcpy(du
, super
->disks
, sizeof(*du
));
9310 du
->index
= (i
* 2) + 1;
9311 sprintf((char *)du
->disk
.serial
,
9312 " MISSING_%d", du
->index
);
9313 sprintf((char *)du
->serial
,
9314 "MISSING_%d", du
->index
);
9315 du
->next
= super
->missing
;
9316 super
->missing
= du
;
9318 /* create new dev and map */
9319 space
= *space_list
;
9322 *space_list
= *space
;
9323 dev_new
= (void *)space
;
9324 memcpy(dev_new
, dev
, sizeof(*dev
));
9325 /* update new map */
9326 map
= get_imsm_map(dev_new
, MAP_0
);
9327 map
->num_members
= map
->num_members
* 2;
9328 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9329 map
->num_domains
= 2;
9330 map
->raid_level
= 1;
9331 /* replace dev<->dev_new */
9334 /* update disk order table */
9335 for (du
= super
->disks
; du
; du
= du
->next
)
9337 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9338 for (du
= super
->missing
; du
; du
= du
->next
)
9339 if (du
->index
>= 0) {
9340 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9341 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9347 static void imsm_process_update(struct supertype
*st
,
9348 struct metadata_update
*update
)
9351 * crack open the metadata_update envelope to find the update record
9352 * update can be one of:
9353 * update_reshape_container_disks - all the arrays in the container
9354 * are being reshaped to have more devices. We need to mark
9355 * the arrays for general migration and convert selected spares
9356 * into active devices.
9357 * update_activate_spare - a spare device has replaced a failed
9358 * device in an array, update the disk_ord_tbl. If this disk is
9359 * present in all member arrays then also clear the SPARE_DISK
9361 * update_create_array
9363 * update_rename_array
9364 * update_add_remove_disk
9366 struct intel_super
*super
= st
->sb
;
9367 struct imsm_super
*mpb
;
9368 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9370 /* update requires a larger buf but the allocation failed */
9371 if (super
->next_len
&& !super
->next_buf
) {
9372 super
->next_len
= 0;
9376 if (super
->next_buf
) {
9377 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9379 super
->len
= super
->next_len
;
9380 super
->buf
= super
->next_buf
;
9382 super
->next_len
= 0;
9383 super
->next_buf
= NULL
;
9386 mpb
= super
->anchor
;
9389 case update_general_migration_checkpoint
: {
9390 struct intel_dev
*id
;
9391 struct imsm_update_general_migration_checkpoint
*u
=
9392 (void *)update
->buf
;
9394 dprintf("called for update_general_migration_checkpoint\n");
9396 /* find device under general migration */
9397 for (id
= super
->devlist
; id
; id
= id
->next
) {
9398 if (is_gen_migration(id
->dev
)) {
9399 id
->dev
->vol
.curr_migr_unit
=
9400 __cpu_to_le32(u
->curr_migr_unit
);
9401 super
->updates_pending
++;
9406 case update_takeover
: {
9407 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9408 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9409 imsm_update_version_info(super
);
9410 super
->updates_pending
++;
9415 case update_reshape_container_disks
: {
9416 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9417 if (apply_reshape_container_disks_update(
9418 u
, super
, &update
->space_list
))
9419 super
->updates_pending
++;
9422 case update_reshape_migration
: {
9423 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9424 if (apply_reshape_migration_update(
9425 u
, super
, &update
->space_list
))
9426 super
->updates_pending
++;
9429 case update_size_change
: {
9430 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9431 if (apply_size_change_update(u
, super
))
9432 super
->updates_pending
++;
9435 case update_activate_spare
: {
9436 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9437 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9438 super
->updates_pending
++;
9441 case update_create_array
: {
9442 /* someone wants to create a new array, we need to be aware of
9443 * a few races/collisions:
9444 * 1/ 'Create' called by two separate instances of mdadm
9445 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9446 * devices that have since been assimilated via
9448 * In the event this update can not be carried out mdadm will
9449 * (FIX ME) notice that its update did not take hold.
9451 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9452 struct intel_dev
*dv
;
9453 struct imsm_dev
*dev
;
9454 struct imsm_map
*map
, *new_map
;
9455 unsigned long long start
, end
;
9456 unsigned long long new_start
, new_end
;
9458 struct disk_info
*inf
;
9461 /* handle racing creates: first come first serve */
9462 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9463 dprintf("subarray %d already defined\n", u
->dev_idx
);
9467 /* check update is next in sequence */
9468 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9469 dprintf("can not create array %d expected index %d\n",
9470 u
->dev_idx
, mpb
->num_raid_devs
);
9474 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9475 new_start
= pba_of_lba0(new_map
);
9476 new_end
= new_start
+ blocks_per_member(new_map
);
9477 inf
= get_disk_info(u
);
9479 /* handle activate_spare versus create race:
9480 * check to make sure that overlapping arrays do not include
9483 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9484 dev
= get_imsm_dev(super
, i
);
9485 map
= get_imsm_map(dev
, MAP_0
);
9486 start
= pba_of_lba0(map
);
9487 end
= start
+ blocks_per_member(map
);
9488 if ((new_start
>= start
&& new_start
<= end
) ||
9489 (start
>= new_start
&& start
<= new_end
))
9494 if (disks_overlap(super
, i
, u
)) {
9495 dprintf("arrays overlap\n");
9500 /* check that prepare update was successful */
9501 if (!update
->space
) {
9502 dprintf("prepare update failed\n");
9506 /* check that all disks are still active before committing
9507 * changes. FIXME: could we instead handle this by creating a
9508 * degraded array? That's probably not what the user expects,
9509 * so better to drop this update on the floor.
9511 for (i
= 0; i
< new_map
->num_members
; i
++) {
9512 dl
= serial_to_dl(inf
[i
].serial
, super
);
9514 dprintf("disk disappeared\n");
9519 super
->updates_pending
++;
9521 /* convert spares to members and fixup ord_tbl */
9522 for (i
= 0; i
< new_map
->num_members
; i
++) {
9523 dl
= serial_to_dl(inf
[i
].serial
, super
);
9524 if (dl
->index
== -1) {
9525 dl
->index
= mpb
->num_disks
;
9527 dl
->disk
.status
|= CONFIGURED_DISK
;
9528 dl
->disk
.status
&= ~SPARE_DISK
;
9530 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9535 update
->space
= NULL
;
9536 imsm_copy_dev(dev
, &u
->dev
);
9537 dv
->index
= u
->dev_idx
;
9538 dv
->next
= super
->devlist
;
9539 super
->devlist
= dv
;
9540 mpb
->num_raid_devs
++;
9542 imsm_update_version_info(super
);
9545 /* mdmon knows how to release update->space, but not
9546 * ((struct intel_dev *) update->space)->dev
9548 if (update
->space
) {
9554 case update_kill_array
: {
9555 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9556 int victim
= u
->dev_idx
;
9557 struct active_array
*a
;
9558 struct intel_dev
**dp
;
9559 struct imsm_dev
*dev
;
9561 /* sanity check that we are not affecting the uuid of
9562 * active arrays, or deleting an active array
9564 * FIXME when immutable ids are available, but note that
9565 * we'll also need to fixup the invalidated/active
9566 * subarray indexes in mdstat
9568 for (a
= st
->arrays
; a
; a
= a
->next
)
9569 if (a
->info
.container_member
>= victim
)
9571 /* by definition if mdmon is running at least one array
9572 * is active in the container, so checking
9573 * mpb->num_raid_devs is just extra paranoia
9575 dev
= get_imsm_dev(super
, victim
);
9576 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9577 dprintf("failed to delete subarray-%d\n", victim
);
9581 for (dp
= &super
->devlist
; *dp
;)
9582 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9585 if ((*dp
)->index
> (unsigned)victim
)
9589 mpb
->num_raid_devs
--;
9590 super
->updates_pending
++;
9593 case update_rename_array
: {
9594 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9595 char name
[MAX_RAID_SERIAL_LEN
+1];
9596 int target
= u
->dev_idx
;
9597 struct active_array
*a
;
9598 struct imsm_dev
*dev
;
9600 /* sanity check that we are not affecting the uuid of
9603 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9604 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9605 for (a
= st
->arrays
; a
; a
= a
->next
)
9606 if (a
->info
.container_member
== target
)
9608 dev
= get_imsm_dev(super
, u
->dev_idx
);
9609 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9610 dprintf("failed to rename subarray-%d\n", target
);
9614 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
9615 super
->updates_pending
++;
9618 case update_add_remove_disk
: {
9619 /* we may be able to repair some arrays if disks are
9620 * being added, check the status of add_remove_disk
9621 * if discs has been added.
9623 if (add_remove_disk_update(super
)) {
9624 struct active_array
*a
;
9626 super
->updates_pending
++;
9627 for (a
= st
->arrays
; a
; a
= a
->next
)
9628 a
->check_degraded
= 1;
9632 case update_prealloc_badblocks_mem
:
9634 case update_rwh_policy
: {
9635 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
9636 int target
= u
->dev_idx
;
9637 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
9639 dprintf("could not find subarray-%d\n", target
);
9643 if (dev
->rwh_policy
!= u
->new_policy
) {
9644 dev
->rwh_policy
= u
->new_policy
;
9645 super
->updates_pending
++;
9650 pr_err("error: unsuported process update type:(type: %d)\n", type
);
9654 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
9656 static int imsm_prepare_update(struct supertype
*st
,
9657 struct metadata_update
*update
)
9660 * Allocate space to hold new disk entries, raid-device entries or a new
9661 * mpb if necessary. The manager synchronously waits for updates to
9662 * complete in the monitor, so new mpb buffers allocated here can be
9663 * integrated by the monitor thread without worrying about live pointers
9664 * in the manager thread.
9666 enum imsm_update_type type
;
9667 struct intel_super
*super
= st
->sb
;
9668 unsigned int sector_size
= super
->sector_size
;
9669 struct imsm_super
*mpb
= super
->anchor
;
9673 if (update
->len
< (int)sizeof(type
))
9676 type
= *(enum imsm_update_type
*) update
->buf
;
9679 case update_general_migration_checkpoint
:
9680 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
9682 dprintf("called for update_general_migration_checkpoint\n");
9684 case update_takeover
: {
9685 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9686 if (update
->len
< (int)sizeof(*u
))
9688 if (u
->direction
== R0_TO_R10
) {
9689 void **tail
= (void **)&update
->space_list
;
9690 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
9691 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9692 int num_members
= map
->num_members
;
9695 /* allocate memory for added disks */
9696 for (i
= 0; i
< num_members
; i
++) {
9697 size
= sizeof(struct dl
);
9698 space
= xmalloc(size
);
9703 /* allocate memory for new device */
9704 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
9705 (num_members
* sizeof(__u32
));
9706 space
= xmalloc(size
);
9710 len
= disks_to_mpb_size(num_members
* 2);
9715 case update_reshape_container_disks
: {
9716 /* Every raid device in the container is about to
9717 * gain some more devices, and we will enter a
9719 * So each 'imsm_map' will be bigger, and the imsm_vol
9720 * will now hold 2 of them.
9721 * Thus we need new 'struct imsm_dev' allocations sized
9722 * as sizeof_imsm_dev but with more devices in both maps.
9724 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9725 struct intel_dev
*dl
;
9726 void **space_tail
= (void**)&update
->space_list
;
9728 if (update
->len
< (int)sizeof(*u
))
9731 dprintf("for update_reshape\n");
9733 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
9734 int size
= sizeof_imsm_dev(dl
->dev
, 1);
9736 if (u
->new_raid_disks
> u
->old_raid_disks
)
9737 size
+= sizeof(__u32
)*2*
9738 (u
->new_raid_disks
- u
->old_raid_disks
);
9745 len
= disks_to_mpb_size(u
->new_raid_disks
);
9746 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9749 case update_reshape_migration
: {
9750 /* for migration level 0->5 we need to add disks
9751 * so the same as for container operation we will copy
9752 * device to the bigger location.
9753 * in memory prepared device and new disk area are prepared
9754 * for usage in process update
9756 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9757 struct intel_dev
*id
;
9758 void **space_tail
= (void **)&update
->space_list
;
9761 int current_level
= -1;
9763 if (update
->len
< (int)sizeof(*u
))
9766 dprintf("for update_reshape\n");
9768 /* add space for bigger array in update
9770 for (id
= super
->devlist
; id
; id
= id
->next
) {
9771 if (id
->index
== (unsigned)u
->subdev
) {
9772 size
= sizeof_imsm_dev(id
->dev
, 1);
9773 if (u
->new_raid_disks
> u
->old_raid_disks
)
9774 size
+= sizeof(__u32
)*2*
9775 (u
->new_raid_disks
- u
->old_raid_disks
);
9783 if (update
->space_list
== NULL
)
9786 /* add space for disk in update
9788 size
= sizeof(struct dl
);
9794 /* add spare device to update
9796 for (id
= super
->devlist
; id
; id
= id
->next
)
9797 if (id
->index
== (unsigned)u
->subdev
) {
9798 struct imsm_dev
*dev
;
9799 struct imsm_map
*map
;
9801 dev
= get_imsm_dev(super
, u
->subdev
);
9802 map
= get_imsm_map(dev
, MAP_0
);
9803 current_level
= map
->raid_level
;
9806 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
9807 struct mdinfo
*spares
;
9809 spares
= get_spares_for_grow(st
);
9817 makedev(dev
->disk
.major
,
9819 dl
= get_disk_super(super
,
9822 dl
->index
= u
->old_raid_disks
;
9828 len
= disks_to_mpb_size(u
->new_raid_disks
);
9829 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9832 case update_size_change
: {
9833 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
9837 case update_activate_spare
: {
9838 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
9842 case update_create_array
: {
9843 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9844 struct intel_dev
*dv
;
9845 struct imsm_dev
*dev
= &u
->dev
;
9846 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9848 struct disk_info
*inf
;
9852 if (update
->len
< (int)sizeof(*u
))
9855 inf
= get_disk_info(u
);
9856 len
= sizeof_imsm_dev(dev
, 1);
9857 /* allocate a new super->devlist entry */
9858 dv
= xmalloc(sizeof(*dv
));
9859 dv
->dev
= xmalloc(len
);
9862 /* count how many spares will be converted to members */
9863 for (i
= 0; i
< map
->num_members
; i
++) {
9864 dl
= serial_to_dl(inf
[i
].serial
, super
);
9866 /* hmm maybe it failed?, nothing we can do about
9871 if (count_memberships(dl
, super
) == 0)
9874 len
+= activate
* sizeof(struct imsm_disk
);
9877 case update_kill_array
: {
9878 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
9882 case update_rename_array
: {
9883 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
9887 case update_add_remove_disk
:
9888 /* no update->len needed */
9890 case update_prealloc_badblocks_mem
:
9891 super
->extra_space
+= sizeof(struct bbm_log
) -
9892 get_imsm_bbm_log_size(super
->bbm_log
);
9894 case update_rwh_policy
: {
9895 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
9903 /* check if we need a larger metadata buffer */
9904 if (super
->next_buf
)
9905 buf_len
= super
->next_len
;
9907 buf_len
= super
->len
;
9909 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
9910 /* ok we need a larger buf than what is currently allocated
9911 * if this allocation fails process_update will notice that
9912 * ->next_len is set and ->next_buf is NULL
9914 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
9915 super
->extra_space
+ len
, sector_size
);
9916 if (super
->next_buf
)
9917 free(super
->next_buf
);
9919 super
->next_len
= buf_len
;
9920 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
9921 memset(super
->next_buf
, 0, buf_len
);
9923 super
->next_buf
= NULL
;
9928 /* must be called while manager is quiesced */
9929 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9931 struct imsm_super
*mpb
= super
->anchor
;
9933 struct imsm_dev
*dev
;
9934 struct imsm_map
*map
;
9935 unsigned int i
, j
, num_members
;
9937 struct bbm_log
*log
= super
->bbm_log
;
9939 dprintf("deleting device[%d] from imsm_super\n", index
);
9941 /* shift all indexes down one */
9942 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
9943 if (iter
->index
> (int)index
)
9945 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
9946 if (iter
->index
> (int)index
)
9949 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9950 dev
= get_imsm_dev(super
, i
);
9951 map
= get_imsm_map(dev
, MAP_0
);
9952 num_members
= map
->num_members
;
9953 for (j
= 0; j
< num_members
; j
++) {
9954 /* update ord entries being careful not to propagate
9955 * ord-flags to the first map
9957 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
9959 if (ord_to_idx(ord
) <= index
)
9962 map
= get_imsm_map(dev
, MAP_0
);
9963 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
9964 map
= get_imsm_map(dev
, MAP_1
);
9966 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
9970 for (i
= 0; i
< log
->entry_count
; i
++) {
9971 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
9973 if (entry
->disk_ordinal
<= index
)
9975 entry
->disk_ordinal
--;
9979 super
->updates_pending
++;
9981 struct dl
*dl
= *dlp
;
9983 *dlp
= (*dlp
)->next
;
9984 __free_imsm_disk(dl
);
9988 static void close_targets(int *targets
, int new_disks
)
9995 for (i
= 0; i
< new_disks
; i
++) {
9996 if (targets
[i
] >= 0) {
10003 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10004 struct intel_super
*super
,
10005 struct imsm_dev
*dev
)
10011 struct imsm_map
*map
;
10014 ret_val
= raid_disks
/2;
10015 /* check map if all disks pairs not failed
10018 map
= get_imsm_map(dev
, MAP_0
);
10019 for (i
= 0; i
< ret_val
; i
++) {
10020 int degradation
= 0;
10021 if (get_imsm_disk(super
, i
) == NULL
)
10023 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10025 if (degradation
== 2)
10028 map
= get_imsm_map(dev
, MAP_1
);
10029 /* if there is no second map
10030 * result can be returned
10034 /* check degradation in second map
10036 for (i
= 0; i
< ret_val
; i
++) {
10037 int degradation
= 0;
10038 if (get_imsm_disk(super
, i
) == NULL
)
10040 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10042 if (degradation
== 2)
10056 /*******************************************************************************
10057 * Function: open_backup_targets
10058 * Description: Function opens file descriptors for all devices given in
10061 * info : general array info
10062 * raid_disks : number of disks
10063 * raid_fds : table of device's file descriptors
10064 * super : intel super for raid10 degradation check
10065 * dev : intel device for raid10 degradation check
10069 ******************************************************************************/
10070 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10071 struct intel_super
*super
, struct imsm_dev
*dev
)
10077 for (i
= 0; i
< raid_disks
; i
++)
10080 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10083 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10084 dprintf("disk is faulty!!\n");
10088 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10091 dn
= map_dev(sd
->disk
.major
,
10092 sd
->disk
.minor
, 1);
10093 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10094 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10095 pr_err("cannot open component\n");
10100 /* check if maximum array degradation level is not exceeded
10102 if ((raid_disks
- opened
) >
10103 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10105 pr_err("Not enough disks can be opened.\n");
10106 close_targets(raid_fds
, raid_disks
);
10112 /*******************************************************************************
10113 * Function: validate_container_imsm
10114 * Description: This routine validates container after assemble,
10115 * eg. if devices in container are under the same controller.
10118 * info : linked list with info about devices used in array
10122 ******************************************************************************/
10123 int validate_container_imsm(struct mdinfo
*info
)
10125 if (check_env("IMSM_NO_PLATFORM"))
10128 struct sys_dev
*idev
;
10129 struct sys_dev
*hba
= NULL
;
10130 struct sys_dev
*intel_devices
= find_intel_devices();
10131 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10132 info
->disk
.minor
));
10134 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10135 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10144 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10145 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10149 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10150 struct mdinfo
*dev
;
10152 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10153 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10155 struct sys_dev
*hba2
= NULL
;
10156 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10157 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10165 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10166 get_orom_by_device_id(hba2
->dev_id
);
10168 if (hba2
&& hba
->type
!= hba2
->type
) {
10169 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10170 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10174 if (orom
!= orom2
) {
10175 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10176 " This operation is not supported and can lead to data loss.\n");
10181 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10182 " This operation is not supported and can lead to data loss.\n");
10190 /*******************************************************************************
10191 * Function: imsm_record_badblock
10192 * Description: This routine stores new bad block record in BBM log
10195 * a : array containing a bad block
10196 * slot : disk number containing a bad block
10197 * sector : bad block sector
10198 * length : bad block sectors range
10202 ******************************************************************************/
10203 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10204 unsigned long long sector
, int length
)
10206 struct intel_super
*super
= a
->container
->sb
;
10210 ord
= imsm_disk_slot_to_ord(a
, slot
);
10214 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10217 super
->updates_pending
++;
10221 /*******************************************************************************
10222 * Function: imsm_clear_badblock
10223 * Description: This routine clears bad block record from BBM log
10226 * a : array containing a bad block
10227 * slot : disk number containing a bad block
10228 * sector : bad block sector
10229 * length : bad block sectors range
10233 ******************************************************************************/
10234 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10235 unsigned long long sector
, int length
)
10237 struct intel_super
*super
= a
->container
->sb
;
10241 ord
= imsm_disk_slot_to_ord(a
, slot
);
10245 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10247 super
->updates_pending
++;
10251 /*******************************************************************************
10252 * Function: imsm_get_badblocks
10253 * Description: This routine get list of bad blocks for an array
10257 * slot : disk number
10259 * bb : structure containing bad blocks
10261 ******************************************************************************/
10262 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10264 int inst
= a
->info
.container_member
;
10265 struct intel_super
*super
= a
->container
->sb
;
10266 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10267 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10270 ord
= imsm_disk_slot_to_ord(a
, slot
);
10274 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10275 blocks_per_member(map
), &super
->bb
);
10279 /*******************************************************************************
10280 * Function: examine_badblocks_imsm
10281 * Description: Prints list of bad blocks on a disk to the standard output
10284 * st : metadata handler
10285 * fd : open file descriptor for device
10286 * devname : device name
10290 ******************************************************************************/
10291 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10293 struct intel_super
*super
= st
->sb
;
10294 struct bbm_log
*log
= super
->bbm_log
;
10295 struct dl
*d
= NULL
;
10298 for (d
= super
->disks
; d
; d
= d
->next
) {
10299 if (strcmp(d
->devname
, devname
) == 0)
10303 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10304 pr_err("%s doesn't appear to be part of a raid array\n",
10311 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10313 for (i
= 0; i
< log
->entry_count
; i
++) {
10314 if (entry
[i
].disk_ordinal
== d
->index
) {
10315 unsigned long long sector
= __le48_to_cpu(
10316 &entry
[i
].defective_block_start
);
10317 int cnt
= entry
[i
].marked_count
+ 1;
10320 printf("Bad-blocks on %s:\n", devname
);
10324 printf("%20llu for %d sectors\n", sector
, cnt
);
10330 printf("No bad-blocks list configured on %s\n", devname
);
10334 /*******************************************************************************
10335 * Function: init_migr_record_imsm
10336 * Description: Function inits imsm migration record
10338 * super : imsm internal array info
10339 * dev : device under migration
10340 * info : general array info to find the smallest device
10343 ******************************************************************************/
10344 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10345 struct mdinfo
*info
)
10347 struct intel_super
*super
= st
->sb
;
10348 struct migr_record
*migr_rec
= super
->migr_rec
;
10349 int new_data_disks
;
10350 unsigned long long dsize
, dev_sectors
;
10351 long long unsigned min_dev_sectors
= -1LLU;
10355 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10356 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10357 unsigned long long num_migr_units
;
10358 unsigned long long array_blocks
;
10360 memset(migr_rec
, 0, sizeof(struct migr_record
));
10361 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10363 /* only ascending reshape supported now */
10364 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10366 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10367 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10368 migr_rec
->dest_depth_per_unit
*=
10369 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10370 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
10371 migr_rec
->blocks_per_unit
=
10372 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10373 migr_rec
->dest_depth_per_unit
=
10374 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10375 array_blocks
= info
->component_size
* new_data_disks
;
10377 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10379 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10381 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
10383 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10384 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10386 /* Find the smallest dev */
10387 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10388 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10389 fd
= dev_open(nm
, O_RDONLY
);
10392 get_dev_size(fd
, NULL
, &dsize
);
10393 dev_sectors
= dsize
/ 512;
10394 if (dev_sectors
< min_dev_sectors
)
10395 min_dev_sectors
= dev_sectors
;
10398 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
10399 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10401 write_imsm_migr_rec(st
);
10406 /*******************************************************************************
10407 * Function: save_backup_imsm
10408 * Description: Function saves critical data stripes to Migration Copy Area
10409 * and updates the current migration unit status.
10410 * Use restore_stripes() to form a destination stripe,
10411 * and to write it to the Copy Area.
10413 * st : supertype information
10414 * dev : imsm device that backup is saved for
10415 * info : general array info
10416 * buf : input buffer
10417 * length : length of data to backup (blocks_per_unit)
10421 ******************************************************************************/
10422 int save_backup_imsm(struct supertype
*st
,
10423 struct imsm_dev
*dev
,
10424 struct mdinfo
*info
,
10429 struct intel_super
*super
= st
->sb
;
10430 unsigned long long *target_offsets
;
10433 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10434 int new_disks
= map_dest
->num_members
;
10435 int dest_layout
= 0;
10437 unsigned long long start
;
10438 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
10440 targets
= xmalloc(new_disks
* sizeof(int));
10442 for (i
= 0; i
< new_disks
; i
++)
10445 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10447 start
= info
->reshape_progress
* 512;
10448 for (i
= 0; i
< new_disks
; i
++) {
10449 target_offsets
[i
] = (unsigned long long)
10450 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
10451 /* move back copy area adderss, it will be moved forward
10452 * in restore_stripes() using start input variable
10454 target_offsets
[i
] -= start
/data_disks
;
10457 if (open_backup_targets(info
, new_disks
, targets
,
10461 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10462 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10464 if (restore_stripes(targets
, /* list of dest devices */
10465 target_offsets
, /* migration record offsets */
10468 map_dest
->raid_level
,
10470 -1, /* source backup file descriptor */
10471 0, /* input buf offset
10472 * always 0 buf is already offseted */
10476 pr_err("Error restoring stripes\n");
10484 close_targets(targets
, new_disks
);
10487 free(target_offsets
);
10492 /*******************************************************************************
10493 * Function: save_checkpoint_imsm
10494 * Description: Function called for current unit status update
10495 * in the migration record. It writes it to disk.
10497 * super : imsm internal array info
10498 * info : general array info
10502 * 2: failure, means no valid migration record
10503 * / no general migration in progress /
10504 ******************************************************************************/
10505 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10507 struct intel_super
*super
= st
->sb
;
10508 unsigned long long blocks_per_unit
;
10509 unsigned long long curr_migr_unit
;
10511 if (load_imsm_migr_rec(super
, info
) != 0) {
10512 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10516 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10517 if (blocks_per_unit
== 0) {
10518 dprintf("imsm: no migration in progress.\n");
10521 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10522 /* check if array is alligned to copy area
10523 * if it is not alligned, add one to current migration unit value
10524 * this can happend on array reshape finish only
10526 if (info
->reshape_progress
% blocks_per_unit
)
10529 super
->migr_rec
->curr_migr_unit
=
10530 __cpu_to_le32(curr_migr_unit
);
10531 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10532 super
->migr_rec
->dest_1st_member_lba
=
10533 __cpu_to_le32(curr_migr_unit
*
10534 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
10535 if (write_imsm_migr_rec(st
) < 0) {
10536 dprintf("imsm: Cannot write migration record outside backup area\n");
10543 /*******************************************************************************
10544 * Function: recover_backup_imsm
10545 * Description: Function recovers critical data from the Migration Copy Area
10546 * while assembling an array.
10548 * super : imsm internal array info
10549 * info : general array info
10551 * 0 : success (or there is no data to recover)
10553 ******************************************************************************/
10554 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10556 struct intel_super
*super
= st
->sb
;
10557 struct migr_record
*migr_rec
= super
->migr_rec
;
10558 struct imsm_map
*map_dest
;
10559 struct intel_dev
*id
= NULL
;
10560 unsigned long long read_offset
;
10561 unsigned long long write_offset
;
10563 int *targets
= NULL
;
10564 int new_disks
, i
, err
;
10567 unsigned int sector_size
= super
->sector_size
;
10568 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
10569 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
10571 int skipped_disks
= 0;
10573 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10577 /* recover data only during assemblation */
10578 if (strncmp(buffer
, "inactive", 8) != 0)
10580 /* no data to recover */
10581 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10583 if (curr_migr_unit
>= num_migr_units
)
10586 /* find device during reshape */
10587 for (id
= super
->devlist
; id
; id
= id
->next
)
10588 if (is_gen_migration(id
->dev
))
10593 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10594 new_disks
= map_dest
->num_members
;
10596 read_offset
= (unsigned long long)
10597 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
10599 write_offset
= ((unsigned long long)
10600 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
10601 pba_of_lba0(map_dest
)) * 512;
10603 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10604 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10606 targets
= xcalloc(new_disks
, sizeof(int));
10608 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10609 pr_err("Cannot open some devices belonging to array.\n");
10613 for (i
= 0; i
< new_disks
; i
++) {
10614 if (targets
[i
] < 0) {
10618 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10619 pr_err("Cannot seek to block: %s\n",
10624 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10625 pr_err("Cannot read copy area block: %s\n",
10630 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
10631 pr_err("Cannot seek to block: %s\n",
10636 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
10637 pr_err("Cannot restore block: %s\n",
10644 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
10648 pr_err("Cannot restore data from backup. Too many failed disks\n");
10652 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
10653 /* ignore error == 2, this can mean end of reshape here
10655 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
10661 for (i
= 0; i
< new_disks
; i
++)
10670 static char disk_by_path
[] = "/dev/disk/by-path/";
10672 static const char *imsm_get_disk_controller_domain(const char *path
)
10674 char disk_path
[PATH_MAX
];
10678 strcpy(disk_path
, disk_by_path
);
10679 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
10680 if (stat(disk_path
, &st
) == 0) {
10681 struct sys_dev
* hba
;
10684 path
= devt_to_devpath(st
.st_rdev
);
10687 hba
= find_disk_attached_hba(-1, path
);
10688 if (hba
&& hba
->type
== SYS_DEV_SAS
)
10690 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
10694 dprintf("path: %s hba: %s attached: %s\n",
10695 path
, (hba
) ? hba
->path
: "NULL", drv
);
10701 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
10703 static char devnm
[32];
10704 char subdev_name
[20];
10705 struct mdstat_ent
*mdstat
;
10707 sprintf(subdev_name
, "%d", subdev
);
10708 mdstat
= mdstat_by_subdev(subdev_name
, container
);
10712 strcpy(devnm
, mdstat
->devnm
);
10713 free_mdstat(mdstat
);
10717 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
10718 struct geo_params
*geo
,
10719 int *old_raid_disks
,
10722 /* currently we only support increasing the number of devices
10723 * for a container. This increases the number of device for each
10724 * member array. They must all be RAID0 or RAID5.
10727 struct mdinfo
*info
, *member
;
10728 int devices_that_can_grow
= 0;
10730 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
10732 if (geo
->size
> 0 ||
10733 geo
->level
!= UnSet
||
10734 geo
->layout
!= UnSet
||
10735 geo
->chunksize
!= 0 ||
10736 geo
->raid_disks
== UnSet
) {
10737 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
10741 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10742 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
10746 info
= container_content_imsm(st
, NULL
);
10747 for (member
= info
; member
; member
= member
->next
) {
10750 dprintf("imsm: checking device_num: %i\n",
10751 member
->container_member
);
10753 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
10754 /* we work on container for Online Capacity Expansion
10755 * only so raid_disks has to grow
10757 dprintf("imsm: for container operation raid disks increase is required\n");
10761 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
10762 /* we cannot use this container with other raid level
10764 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
10765 info
->array
.level
);
10768 /* check for platform support
10769 * for this raid level configuration
10771 struct intel_super
*super
= st
->sb
;
10772 if (!is_raid_level_supported(super
->orom
,
10773 member
->array
.level
,
10774 geo
->raid_disks
)) {
10775 dprintf("platform does not support raid%d with %d disk%s\n",
10778 geo
->raid_disks
> 1 ? "s" : "");
10781 /* check if component size is aligned to chunk size
10783 if (info
->component_size
%
10784 (info
->array
.chunk_size
/512)) {
10785 dprintf("Component size is not aligned to chunk size\n");
10790 if (*old_raid_disks
&&
10791 info
->array
.raid_disks
!= *old_raid_disks
)
10793 *old_raid_disks
= info
->array
.raid_disks
;
10795 /* All raid5 and raid0 volumes in container
10796 * have to be ready for Online Capacity Expansion
10797 * so they need to be assembled. We have already
10798 * checked that no recovery etc is happening.
10800 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
10801 st
->container_devnm
);
10802 if (result
== NULL
) {
10803 dprintf("imsm: cannot find array\n");
10806 devices_that_can_grow
++;
10809 if (!member
&& devices_that_can_grow
)
10813 dprintf("Container operation allowed\n");
10815 dprintf("Error: %i\n", ret_val
);
10820 /* Function: get_spares_for_grow
10821 * Description: Allocates memory and creates list of spare devices
10822 * avaliable in container. Checks if spare drive size is acceptable.
10823 * Parameters: Pointer to the supertype structure
10824 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
10827 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
10829 struct spare_criteria sc
;
10831 get_spare_criteria_imsm(st
, &sc
);
10832 return container_choose_spares(st
, &sc
, NULL
, NULL
, NULL
, 0);
10835 /******************************************************************************
10836 * function: imsm_create_metadata_update_for_reshape
10837 * Function creates update for whole IMSM container.
10839 ******************************************************************************/
10840 static int imsm_create_metadata_update_for_reshape(
10841 struct supertype
*st
,
10842 struct geo_params
*geo
,
10843 int old_raid_disks
,
10844 struct imsm_update_reshape
**updatep
)
10846 struct intel_super
*super
= st
->sb
;
10847 struct imsm_super
*mpb
= super
->anchor
;
10848 int update_memory_size
;
10849 struct imsm_update_reshape
*u
;
10850 struct mdinfo
*spares
;
10853 struct mdinfo
*dev
;
10855 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
10857 delta_disks
= geo
->raid_disks
- old_raid_disks
;
10859 /* size of all update data without anchor */
10860 update_memory_size
= sizeof(struct imsm_update_reshape
);
10862 /* now add space for spare disks that we need to add. */
10863 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
10865 u
= xcalloc(1, update_memory_size
);
10866 u
->type
= update_reshape_container_disks
;
10867 u
->old_raid_disks
= old_raid_disks
;
10868 u
->new_raid_disks
= geo
->raid_disks
;
10870 /* now get spare disks list
10872 spares
= get_spares_for_grow(st
);
10875 || delta_disks
> spares
->array
.spare_disks
) {
10876 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
10881 /* we have got spares
10882 * update disk list in imsm_disk list table in anchor
10884 dprintf("imsm: %i spares are available.\n\n",
10885 spares
->array
.spare_disks
);
10887 dev
= spares
->devs
;
10888 for (i
= 0; i
< delta_disks
; i
++) {
10893 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
10895 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
10896 dl
->index
= mpb
->num_disks
;
10904 sysfs_free(spares
);
10906 dprintf("imsm: reshape update preparation :");
10907 if (i
== delta_disks
) {
10908 dprintf_cont(" OK\n");
10910 return update_memory_size
;
10913 dprintf_cont(" Error\n");
10918 /******************************************************************************
10919 * function: imsm_create_metadata_update_for_size_change()
10920 * Creates update for IMSM array for array size change.
10922 ******************************************************************************/
10923 static int imsm_create_metadata_update_for_size_change(
10924 struct supertype
*st
,
10925 struct geo_params
*geo
,
10926 struct imsm_update_size_change
**updatep
)
10928 struct intel_super
*super
= st
->sb
;
10929 int update_memory_size
;
10930 struct imsm_update_size_change
*u
;
10932 dprintf("(enter) New size = %llu\n", geo
->size
);
10934 /* size of all update data without anchor */
10935 update_memory_size
= sizeof(struct imsm_update_size_change
);
10937 u
= xcalloc(1, update_memory_size
);
10938 u
->type
= update_size_change
;
10939 u
->subdev
= super
->current_vol
;
10940 u
->new_size
= geo
->size
;
10942 dprintf("imsm: reshape update preparation : OK\n");
10945 return update_memory_size
;
10948 /******************************************************************************
10949 * function: imsm_create_metadata_update_for_migration()
10950 * Creates update for IMSM array.
10952 ******************************************************************************/
10953 static int imsm_create_metadata_update_for_migration(
10954 struct supertype
*st
,
10955 struct geo_params
*geo
,
10956 struct imsm_update_reshape_migration
**updatep
)
10958 struct intel_super
*super
= st
->sb
;
10959 int update_memory_size
;
10960 struct imsm_update_reshape_migration
*u
;
10961 struct imsm_dev
*dev
;
10962 int previous_level
= -1;
10964 dprintf("(enter) New Level = %i\n", geo
->level
);
10966 /* size of all update data without anchor */
10967 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
10969 u
= xcalloc(1, update_memory_size
);
10970 u
->type
= update_reshape_migration
;
10971 u
->subdev
= super
->current_vol
;
10972 u
->new_level
= geo
->level
;
10973 u
->new_layout
= geo
->layout
;
10974 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
10975 u
->new_disks
[0] = -1;
10976 u
->new_chunksize
= -1;
10978 dev
= get_imsm_dev(super
, u
->subdev
);
10980 struct imsm_map
*map
;
10982 map
= get_imsm_map(dev
, MAP_0
);
10984 int current_chunk_size
=
10985 __le16_to_cpu(map
->blocks_per_strip
) / 2;
10987 if (geo
->chunksize
!= current_chunk_size
) {
10988 u
->new_chunksize
= geo
->chunksize
/ 1024;
10989 dprintf("imsm: chunk size change from %i to %i\n",
10990 current_chunk_size
, u
->new_chunksize
);
10992 previous_level
= map
->raid_level
;
10995 if (geo
->level
== 5 && previous_level
== 0) {
10996 struct mdinfo
*spares
= NULL
;
10998 u
->new_raid_disks
++;
10999 spares
= get_spares_for_grow(st
);
11000 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11002 sysfs_free(spares
);
11003 update_memory_size
= 0;
11004 pr_err("cannot get spare device for requested migration\n");
11007 sysfs_free(spares
);
11009 dprintf("imsm: reshape update preparation : OK\n");
11012 return update_memory_size
;
11015 static void imsm_update_metadata_locally(struct supertype
*st
,
11016 void *buf
, int len
)
11018 struct metadata_update mu
;
11023 mu
.space_list
= NULL
;
11025 if (imsm_prepare_update(st
, &mu
))
11026 imsm_process_update(st
, &mu
);
11028 while (mu
.space_list
) {
11029 void **space
= mu
.space_list
;
11030 mu
.space_list
= *space
;
11035 /***************************************************************************
11036 * Function: imsm_analyze_change
11037 * Description: Function analyze change for single volume
11038 * and validate if transition is supported
11039 * Parameters: Geometry parameters, supertype structure,
11040 * metadata change direction (apply/rollback)
11041 * Returns: Operation type code on success, -1 if fail
11042 ****************************************************************************/
11043 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11044 struct geo_params
*geo
,
11047 struct mdinfo info
;
11049 int check_devs
= 0;
11051 /* number of added/removed disks in operation result */
11052 int devNumChange
= 0;
11053 /* imsm compatible layout value for array geometry verification */
11054 int imsm_layout
= -1;
11056 struct imsm_dev
*dev
;
11057 struct intel_super
*super
;
11058 unsigned long long current_size
;
11059 unsigned long long free_size
;
11060 unsigned long long max_size
;
11063 getinfo_super_imsm_volume(st
, &info
, NULL
);
11064 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11065 geo
->level
!= UnSet
) {
11066 switch (info
.array
.level
) {
11068 if (geo
->level
== 5) {
11069 change
= CH_MIGRATION
;
11070 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11071 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11073 goto analyse_change_exit
;
11075 imsm_layout
= geo
->layout
;
11077 devNumChange
= 1; /* parity disk added */
11078 } else if (geo
->level
== 10) {
11079 change
= CH_TAKEOVER
;
11081 devNumChange
= 2; /* two mirrors added */
11082 imsm_layout
= 0x102; /* imsm supported layout */
11087 if (geo
->level
== 0) {
11088 change
= CH_TAKEOVER
;
11090 devNumChange
= -(geo
->raid_disks
/2);
11091 imsm_layout
= 0; /* imsm raid0 layout */
11095 if (change
== -1) {
11096 pr_err("Error. Level Migration from %d to %d not supported!\n",
11097 info
.array
.level
, geo
->level
);
11098 goto analyse_change_exit
;
11101 geo
->level
= info
.array
.level
;
11103 if (geo
->layout
!= info
.array
.layout
&&
11104 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11105 change
= CH_MIGRATION
;
11106 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11107 geo
->layout
== 5) {
11108 /* reshape 5 -> 4 */
11109 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11110 geo
->layout
== 0) {
11111 /* reshape 4 -> 5 */
11115 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11116 info
.array
.layout
, geo
->layout
);
11118 goto analyse_change_exit
;
11121 geo
->layout
= info
.array
.layout
;
11122 if (imsm_layout
== -1)
11123 imsm_layout
= info
.array
.layout
;
11126 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11127 geo
->chunksize
!= info
.array
.chunk_size
) {
11128 if (info
.array
.level
== 10) {
11129 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11131 goto analyse_change_exit
;
11132 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11133 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11134 geo
->chunksize
/1024, info
.component_size
/2);
11136 goto analyse_change_exit
;
11138 change
= CH_MIGRATION
;
11140 geo
->chunksize
= info
.array
.chunk_size
;
11143 chunk
= geo
->chunksize
/ 1024;
11146 dev
= get_imsm_dev(super
, super
->current_vol
);
11147 data_disks
= imsm_num_data_members(dev
, MAP_0
);
11148 /* compute current size per disk member
11150 current_size
= info
.custom_array_size
/ data_disks
;
11152 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11153 /* align component size
11155 geo
->size
= imsm_component_size_aligment_check(
11156 get_imsm_raid_level(dev
->vol
.map
),
11157 chunk
* 1024, super
->sector_size
,
11159 if (geo
->size
== 0) {
11160 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11162 goto analyse_change_exit
;
11166 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11167 if (change
!= -1) {
11168 pr_err("Error. Size change should be the only one at a time.\n");
11170 goto analyse_change_exit
;
11172 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11173 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11174 super
->current_vol
, st
->devnm
);
11175 goto analyse_change_exit
;
11177 /* check the maximum available size
11179 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11180 0, chunk
, &free_size
);
11182 /* Cannot find maximum available space
11186 max_size
= free_size
+ current_size
;
11187 /* align component size
11189 max_size
= imsm_component_size_aligment_check(
11190 get_imsm_raid_level(dev
->vol
.map
),
11191 chunk
* 1024, super
->sector_size
,
11194 if (geo
->size
== MAX_SIZE
) {
11195 /* requested size change to the maximum available size
11197 if (max_size
== 0) {
11198 pr_err("Error. Cannot find maximum available space.\n");
11200 goto analyse_change_exit
;
11202 geo
->size
= max_size
;
11205 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11206 /* accept size for rollback only
11209 /* round size due to metadata compatibility
11211 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11212 << SECT_PER_MB_SHIFT
;
11213 dprintf("Prepare update for size change to %llu\n",
11215 if (current_size
>= geo
->size
) {
11216 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11217 current_size
, geo
->size
);
11218 goto analyse_change_exit
;
11220 if (max_size
&& geo
->size
> max_size
) {
11221 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11222 max_size
, geo
->size
);
11223 goto analyse_change_exit
;
11226 geo
->size
*= data_disks
;
11227 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11228 change
= CH_ARRAY_SIZE
;
11230 if (!validate_geometry_imsm(st
,
11233 geo
->raid_disks
+ devNumChange
,
11235 geo
->size
, INVALID_SECTORS
,
11236 0, 0, info
.consistency_policy
, 1))
11240 struct intel_super
*super
= st
->sb
;
11241 struct imsm_super
*mpb
= super
->anchor
;
11243 if (mpb
->num_raid_devs
> 1) {
11244 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11250 analyse_change_exit
:
11251 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11252 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11253 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11259 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11261 struct intel_super
*super
= st
->sb
;
11262 struct imsm_update_takeover
*u
;
11264 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11266 u
->type
= update_takeover
;
11267 u
->subarray
= super
->current_vol
;
11269 /* 10->0 transition */
11270 if (geo
->level
== 0)
11271 u
->direction
= R10_TO_R0
;
11273 /* 0->10 transition */
11274 if (geo
->level
== 10)
11275 u
->direction
= R0_TO_R10
;
11277 /* update metadata locally */
11278 imsm_update_metadata_locally(st
, u
,
11279 sizeof(struct imsm_update_takeover
));
11280 /* and possibly remotely */
11281 if (st
->update_tail
)
11282 append_metadata_update(st
, u
,
11283 sizeof(struct imsm_update_takeover
));
11290 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11292 int layout
, int chunksize
, int raid_disks
,
11293 int delta_disks
, char *backup
, char *dev
,
11294 int direction
, int verbose
)
11297 struct geo_params geo
;
11299 dprintf("(enter)\n");
11301 memset(&geo
, 0, sizeof(struct geo_params
));
11303 geo
.dev_name
= dev
;
11304 strcpy(geo
.devnm
, st
->devnm
);
11307 geo
.layout
= layout
;
11308 geo
.chunksize
= chunksize
;
11309 geo
.raid_disks
= raid_disks
;
11310 if (delta_disks
!= UnSet
)
11311 geo
.raid_disks
+= delta_disks
;
11313 dprintf("for level : %i\n", geo
.level
);
11314 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11316 if (experimental() == 0)
11319 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11320 /* On container level we can only increase number of devices. */
11321 dprintf("imsm: info: Container operation\n");
11322 int old_raid_disks
= 0;
11324 if (imsm_reshape_is_allowed_on_container(
11325 st
, &geo
, &old_raid_disks
, direction
)) {
11326 struct imsm_update_reshape
*u
= NULL
;
11329 len
= imsm_create_metadata_update_for_reshape(
11330 st
, &geo
, old_raid_disks
, &u
);
11333 dprintf("imsm: Cannot prepare update\n");
11334 goto exit_imsm_reshape_super
;
11338 /* update metadata locally */
11339 imsm_update_metadata_locally(st
, u
, len
);
11340 /* and possibly remotely */
11341 if (st
->update_tail
)
11342 append_metadata_update(st
, u
, len
);
11347 pr_err("(imsm) Operation is not allowed on this container\n");
11350 /* On volume level we support following operations
11351 * - takeover: raid10 -> raid0; raid0 -> raid10
11352 * - chunk size migration
11353 * - migration: raid5 -> raid0; raid0 -> raid5
11355 struct intel_super
*super
= st
->sb
;
11356 struct intel_dev
*dev
= super
->devlist
;
11358 dprintf("imsm: info: Volume operation\n");
11359 /* find requested device */
11362 imsm_find_array_devnm_by_subdev(
11363 dev
->index
, st
->container_devnm
);
11364 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11369 pr_err("Cannot find %s (%s) subarray\n",
11370 geo
.dev_name
, geo
.devnm
);
11371 goto exit_imsm_reshape_super
;
11373 super
->current_vol
= dev
->index
;
11374 change
= imsm_analyze_change(st
, &geo
, direction
);
11377 ret_val
= imsm_takeover(st
, &geo
);
11379 case CH_MIGRATION
: {
11380 struct imsm_update_reshape_migration
*u
= NULL
;
11382 imsm_create_metadata_update_for_migration(
11385 dprintf("imsm: Cannot prepare update\n");
11389 /* update metadata locally */
11390 imsm_update_metadata_locally(st
, u
, len
);
11391 /* and possibly remotely */
11392 if (st
->update_tail
)
11393 append_metadata_update(st
, u
, len
);
11398 case CH_ARRAY_SIZE
: {
11399 struct imsm_update_size_change
*u
= NULL
;
11401 imsm_create_metadata_update_for_size_change(
11404 dprintf("imsm: Cannot prepare update\n");
11408 /* update metadata locally */
11409 imsm_update_metadata_locally(st
, u
, len
);
11410 /* and possibly remotely */
11411 if (st
->update_tail
)
11412 append_metadata_update(st
, u
, len
);
11422 exit_imsm_reshape_super
:
11423 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11427 #define COMPLETED_OK 0
11428 #define COMPLETED_NONE 1
11429 #define COMPLETED_DELAYED 2
11431 static int read_completed(int fd
, unsigned long long *val
)
11436 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11440 ret
= COMPLETED_OK
;
11441 if (strncmp(buf
, "none", 4) == 0) {
11442 ret
= COMPLETED_NONE
;
11443 } else if (strncmp(buf
, "delayed", 7) == 0) {
11444 ret
= COMPLETED_DELAYED
;
11447 *val
= strtoull(buf
, &ep
, 0);
11448 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11454 /*******************************************************************************
11455 * Function: wait_for_reshape_imsm
11456 * Description: Function writes new sync_max value and waits until
11457 * reshape process reach new position
11459 * sra : general array info
11460 * ndata : number of disks in new array's layout
11463 * 1 : there is no reshape in progress,
11465 ******************************************************************************/
11466 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11468 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11470 unsigned long long completed
;
11471 /* to_complete : new sync_max position */
11472 unsigned long long to_complete
= sra
->reshape_progress
;
11473 unsigned long long position_to_set
= to_complete
/ ndata
;
11476 dprintf("cannot open reshape_position\n");
11481 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11483 dprintf("cannot read reshape_position (no reshape in progres)\n");
11492 if (completed
> position_to_set
) {
11493 dprintf("wrong next position to set %llu (%llu)\n",
11494 to_complete
, position_to_set
);
11498 dprintf("Position set: %llu\n", position_to_set
);
11499 if (sysfs_set_num(sra
, NULL
, "sync_max",
11500 position_to_set
) != 0) {
11501 dprintf("cannot set reshape position to %llu\n",
11510 int timeout
= 3000;
11512 sysfs_wait(fd
, &timeout
);
11513 if (sysfs_get_str(sra
, NULL
, "sync_action",
11515 strncmp(action
, "reshape", 7) != 0) {
11516 if (strncmp(action
, "idle", 4) == 0)
11522 rc
= read_completed(fd
, &completed
);
11524 dprintf("cannot read reshape_position (in loop)\n");
11527 } else if (rc
== COMPLETED_NONE
)
11529 } while (completed
< position_to_set
);
11535 /*******************************************************************************
11536 * Function: check_degradation_change
11537 * Description: Check that array hasn't become failed.
11539 * info : for sysfs access
11540 * sources : source disks descriptors
11541 * degraded: previous degradation level
11543 * degradation level
11544 ******************************************************************************/
11545 int check_degradation_change(struct mdinfo
*info
,
11549 unsigned long long new_degraded
;
11552 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11553 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11554 /* check each device to ensure it is still working */
11557 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11558 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11560 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11563 if (sysfs_get_str(info
,
11564 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11565 strstr(sbuf
, "faulty") ||
11566 strstr(sbuf
, "in_sync") == NULL
) {
11567 /* this device is dead */
11568 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11569 if (sd
->disk
.raid_disk
>= 0 &&
11570 sources
[sd
->disk
.raid_disk
] >= 0) {
11572 sd
->disk
.raid_disk
]);
11573 sources
[sd
->disk
.raid_disk
] =
11582 return new_degraded
;
11585 /*******************************************************************************
11586 * Function: imsm_manage_reshape
11587 * Description: Function finds array under reshape and it manages reshape
11588 * process. It creates stripes backups (if required) and sets
11591 * afd : Backup handle (nattive) - not used
11592 * sra : general array info
11593 * reshape : reshape parameters - not used
11594 * st : supertype structure
11595 * blocks : size of critical section [blocks]
11596 * fds : table of source device descriptor
11597 * offsets : start of array (offest per devices)
11599 * destfd : table of destination device descriptor
11600 * destoffsets : table of destination offsets (per device)
11602 * 1 : success, reshape is done
11604 ******************************************************************************/
11605 static int imsm_manage_reshape(
11606 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11607 struct supertype
*st
, unsigned long backup_blocks
,
11608 int *fds
, unsigned long long *offsets
,
11609 int dests
, int *destfd
, unsigned long long *destoffsets
)
11612 struct intel_super
*super
= st
->sb
;
11613 struct intel_dev
*dv
;
11614 unsigned int sector_size
= super
->sector_size
;
11615 struct imsm_dev
*dev
= NULL
;
11616 struct imsm_map
*map_src
;
11617 int migr_vol_qan
= 0;
11618 int ndata
, odata
; /* [bytes] */
11619 int chunk
; /* [bytes] */
11620 struct migr_record
*migr_rec
;
11622 unsigned int buf_size
; /* [bytes] */
11623 unsigned long long max_position
; /* array size [bytes] */
11624 unsigned long long next_step
; /* [blocks]/[bytes] */
11625 unsigned long long old_data_stripe_length
;
11626 unsigned long long start_src
; /* [bytes] */
11627 unsigned long long start
; /* [bytes] */
11628 unsigned long long start_buf_shift
; /* [bytes] */
11630 int source_layout
= 0;
11635 if (!fds
|| !offsets
)
11638 /* Find volume during the reshape */
11639 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
11640 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
11641 && dv
->dev
->vol
.migr_state
== 1) {
11646 /* Only one volume can migrate at the same time */
11647 if (migr_vol_qan
!= 1) {
11648 pr_err("%s", migr_vol_qan
?
11649 "Number of migrating volumes greater than 1\n" :
11650 "There is no volume during migrationg\n");
11654 map_src
= get_imsm_map(dev
, MAP_1
);
11655 if (map_src
== NULL
)
11658 ndata
= imsm_num_data_members(dev
, MAP_0
);
11659 odata
= imsm_num_data_members(dev
, MAP_1
);
11661 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
11662 old_data_stripe_length
= odata
* chunk
;
11664 migr_rec
= super
->migr_rec
;
11666 /* initialize migration record for start condition */
11667 if (sra
->reshape_progress
== 0)
11668 init_migr_record_imsm(st
, dev
, sra
);
11670 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
11671 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
11674 /* Save checkpoint to update migration record for current
11675 * reshape position (in md). It can be farther than current
11676 * reshape position in metadata.
11678 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11679 /* ignore error == 2, this can mean end of reshape here
11681 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
11686 /* size for data */
11687 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
11688 /* extend buffer size for parity disk */
11689 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11690 /* add space for stripe aligment */
11691 buf_size
+= old_data_stripe_length
;
11692 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
11693 dprintf("imsm: Cannot allocate checkpoint buffer\n");
11697 max_position
= sra
->component_size
* ndata
;
11698 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
11700 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
11701 __le32_to_cpu(migr_rec
->num_migr_units
)) {
11702 /* current reshape position [blocks] */
11703 unsigned long long current_position
=
11704 __le32_to_cpu(migr_rec
->blocks_per_unit
)
11705 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
11706 unsigned long long border
;
11708 /* Check that array hasn't become failed.
11710 degraded
= check_degradation_change(sra
, fds
, degraded
);
11711 if (degraded
> 1) {
11712 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
11716 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
11718 if ((current_position
+ next_step
) > max_position
)
11719 next_step
= max_position
- current_position
;
11721 start
= current_position
* 512;
11723 /* align reading start to old geometry */
11724 start_buf_shift
= start
% old_data_stripe_length
;
11725 start_src
= start
- start_buf_shift
;
11727 border
= (start_src
/ odata
) - (start
/ ndata
);
11729 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
11730 /* save critical stripes to buf
11731 * start - start address of current unit
11732 * to backup [bytes]
11733 * start_src - start address of current unit
11734 * to backup alligned to source array
11737 unsigned long long next_step_filler
;
11738 unsigned long long copy_length
= next_step
* 512;
11740 /* allign copy area length to stripe in old geometry */
11741 next_step_filler
= ((copy_length
+ start_buf_shift
)
11742 % old_data_stripe_length
);
11743 if (next_step_filler
)
11744 next_step_filler
= (old_data_stripe_length
11745 - next_step_filler
);
11746 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
11747 start
, start_src
, copy_length
,
11748 start_buf_shift
, next_step_filler
);
11750 if (save_stripes(fds
, offsets
, map_src
->num_members
,
11751 chunk
, map_src
->raid_level
,
11752 source_layout
, 0, NULL
, start_src
,
11754 next_step_filler
+ start_buf_shift
,
11756 dprintf("imsm: Cannot save stripes to buffer\n");
11759 /* Convert data to destination format and store it
11760 * in backup general migration area
11762 if (save_backup_imsm(st
, dev
, sra
,
11763 buf
+ start_buf_shift
, copy_length
)) {
11764 dprintf("imsm: Cannot save stripes to target devices\n");
11767 if (save_checkpoint_imsm(st
, sra
,
11768 UNIT_SRC_IN_CP_AREA
)) {
11769 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
11773 /* set next step to use whole border area */
11774 border
/= next_step
;
11776 next_step
*= border
;
11778 /* When data backed up, checkpoint stored,
11779 * kick the kernel to reshape unit of data
11781 next_step
= next_step
+ sra
->reshape_progress
;
11782 /* limit next step to array max position */
11783 if (next_step
> max_position
)
11784 next_step
= max_position
;
11785 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
11786 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
11787 sra
->reshape_progress
= next_step
;
11789 /* wait until reshape finish */
11790 if (wait_for_reshape_imsm(sra
, ndata
)) {
11791 dprintf("wait_for_reshape_imsm returned error!\n");
11797 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11798 /* ignore error == 2, this can mean end of reshape here
11800 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
11806 /* clear migr_rec on disks after successful migration */
11809 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
);
11810 for (d
= super
->disks
; d
; d
= d
->next
) {
11811 if (d
->index
< 0 || is_failed(&d
->disk
))
11813 unsigned long long dsize
;
11815 get_dev_size(d
->fd
, NULL
, &dsize
);
11816 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
11818 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
11819 MIGR_REC_BUF_SECTORS
*sector_size
) !=
11820 MIGR_REC_BUF_SECTORS
*sector_size
)
11821 perror("Write migr_rec failed");
11825 /* return '1' if done */
11829 /* See Grow.c: abort_reshape() for further explanation */
11830 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
11831 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
11832 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
11837 struct superswitch super_imsm
= {
11838 .examine_super
= examine_super_imsm
,
11839 .brief_examine_super
= brief_examine_super_imsm
,
11840 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
11841 .export_examine_super
= export_examine_super_imsm
,
11842 .detail_super
= detail_super_imsm
,
11843 .brief_detail_super
= brief_detail_super_imsm
,
11844 .write_init_super
= write_init_super_imsm
,
11845 .validate_geometry
= validate_geometry_imsm
,
11846 .add_to_super
= add_to_super_imsm
,
11847 .remove_from_super
= remove_from_super_imsm
,
11848 .detail_platform
= detail_platform_imsm
,
11849 .export_detail_platform
= export_detail_platform_imsm
,
11850 .kill_subarray
= kill_subarray_imsm
,
11851 .update_subarray
= update_subarray_imsm
,
11852 .load_container
= load_container_imsm
,
11853 .default_geometry
= default_geometry_imsm
,
11854 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
11855 .reshape_super
= imsm_reshape_super
,
11856 .manage_reshape
= imsm_manage_reshape
,
11857 .recover_backup
= recover_backup_imsm
,
11858 .copy_metadata
= copy_metadata_imsm
,
11859 .examine_badblocks
= examine_badblocks_imsm
,
11860 .match_home
= match_home_imsm
,
11861 .uuid_from_super
= uuid_from_super_imsm
,
11862 .getinfo_super
= getinfo_super_imsm
,
11863 .getinfo_super_disks
= getinfo_super_disks_imsm
,
11864 .update_super
= update_super_imsm
,
11866 .avail_size
= avail_size_imsm
,
11867 .get_spare_criteria
= get_spare_criteria_imsm
,
11869 .compare_super
= compare_super_imsm
,
11871 .load_super
= load_super_imsm
,
11872 .init_super
= init_super_imsm
,
11873 .store_super
= store_super_imsm
,
11874 .free_super
= free_super_imsm
,
11875 .match_metadata_desc
= match_metadata_desc_imsm
,
11876 .container_content
= container_content_imsm
,
11877 .validate_container
= validate_container_imsm
,
11879 .write_init_ppl
= write_init_ppl_imsm
,
11880 .validate_ppl
= validate_ppl_imsm
,
11886 .open_new
= imsm_open_new
,
11887 .set_array_state
= imsm_set_array_state
,
11888 .set_disk
= imsm_set_disk
,
11889 .sync_metadata
= imsm_sync_metadata
,
11890 .activate_spare
= imsm_activate_spare
,
11891 .process_update
= imsm_process_update
,
11892 .prepare_update
= imsm_prepare_update
,
11893 .record_bad_block
= imsm_record_badblock
,
11894 .clear_bad_block
= imsm_clear_badblock
,
11895 .get_bad_blocks
= imsm_get_badblocks
,