2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2008 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
20 #define HAVE_STDINT_H 1
24 #include "platform-intel.h"
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
75 /* Define all supported attributes that have to be accepted by mdadm
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
79 MPB_ATTRIB_2TB_DISK | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE | \
87 /* Define attributes that are unused but not harmful */
88 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
90 #define MPB_SECTOR_CNT 2210
91 #define IMSM_RESERVED_SECTORS 4096
92 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
93 #define SECT_PER_MB_SHIFT 11
94 #define MAX_SECTOR_SIZE 4096
96 /* Disk configuration info. */
97 #define IMSM_MAX_DEVICES 255
99 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
100 __u32 total_blocks_lo
; /* 0xE8 - 0xEB total blocks lo */
101 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
102 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
103 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
104 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
105 #define JOURNAL_DISK __cpu_to_le32(0x2000000) /* Device marked as Journaling Drive */
106 __u32 status
; /* 0xF0 - 0xF3 */
107 __u32 owner_cfg_num
; /* which config 0,1,2... owns this disk */
108 __u32 total_blocks_hi
; /* 0xF4 - 0xF5 total blocks hi */
109 #define IMSM_DISK_FILLERS 3
110 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF5 - 0x107 MPB_DISK_FILLERS for future expansion */
113 /* map selector for map managment
119 /* RAID map configuration infos. */
121 __u32 pba_of_lba0_lo
; /* start address of partition */
122 __u32 blocks_per_member_lo
;/* blocks per member */
123 __u32 num_data_stripes_lo
; /* number of data stripes */
124 __u16 blocks_per_strip
;
125 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
126 #define IMSM_T_STATE_NORMAL 0
127 #define IMSM_T_STATE_UNINITIALIZED 1
128 #define IMSM_T_STATE_DEGRADED 2
129 #define IMSM_T_STATE_FAILED 3
131 #define IMSM_T_RAID0 0
132 #define IMSM_T_RAID1 1
133 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
134 __u8 num_members
; /* number of member disks */
135 __u8 num_domains
; /* number of parity domains */
136 __u8 failed_disk_num
; /* valid only when state is degraded */
138 __u32 pba_of_lba0_hi
;
139 __u32 blocks_per_member_hi
;
140 __u32 num_data_stripes_hi
;
141 __u32 filler
[4]; /* expansion area */
142 #define IMSM_ORD_REBUILD (1 << 24)
143 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
144 * top byte contains some flags
146 } __attribute__ ((packed
));
149 __u32 curr_migr_unit
;
150 __u32 checkpoint_id
; /* id to access curr_migr_unit */
151 __u8 migr_state
; /* Normal or Migrating */
153 #define MIGR_REBUILD 1
154 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
155 #define MIGR_GEN_MIGR 3
156 #define MIGR_STATE_CHANGE 4
157 #define MIGR_REPAIR 5
158 __u8 migr_type
; /* Initializing, Rebuilding, ... */
159 #define RAIDVOL_CLEAN 0
160 #define RAIDVOL_DIRTY 1
161 #define RAIDVOL_DSRECORD_VALID 2
163 __u8 fs_state
; /* fast-sync state for CnG (0xff == disabled) */
164 __u16 verify_errors
; /* number of mismatches */
165 __u16 bad_blocks
; /* number of bad blocks during verify */
167 struct imsm_map map
[1];
168 /* here comes another one if migr_state */
169 } __attribute__ ((packed
));
172 __u8 volume
[MAX_RAID_SERIAL_LEN
];
175 #define DEV_BOOTABLE __cpu_to_le32(0x01)
176 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
177 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
178 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
179 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
180 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
181 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
182 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
183 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
184 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
185 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
186 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
187 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
188 __u32 status
; /* Persistent RaidDev status */
189 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
193 __u8 cng_master_disk
;
197 __u16 my_vol_raid_dev_num
; /* Used in Unique volume Id for this RaidDev */
203 /* Unique Volume Id of the NvCache Volume associated with this volume */
204 __u32 nvc_vol_orig_family_num
;
205 __u16 nvc_vol_raid_dev_num
;
208 #define RWH_DISTRIBUTED 1
209 #define RWH_JOURNALING_DRIVE 2
210 __u8 rwh_policy
; /* Raid Write Hole Policy */
211 __u8 jd_serial
[MAX_RAID_SERIAL_LEN
]; /* Journal Drive serial number */
214 #define IMSM_DEV_FILLERS 3
215 __u32 filler
[IMSM_DEV_FILLERS
];
217 } __attribute__ ((packed
));
220 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
221 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
222 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
223 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
224 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
225 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
226 __u32 attributes
; /* 0x34 - 0x37 */
227 __u8 num_disks
; /* 0x38 Number of configured disks */
228 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
229 __u8 error_log_pos
; /* 0x3A */
230 __u8 fill
[1]; /* 0x3B */
231 __u32 cache_size
; /* 0x3c - 0x40 in mb */
232 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
233 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
234 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
235 #define IMSM_FILLERS 35
236 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
237 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
238 /* here comes imsm_dev[num_raid_devs] */
239 /* here comes BBM logs */
240 } __attribute__ ((packed
));
242 #define BBM_LOG_MAX_ENTRIES 254
243 #define BBM_LOG_MAX_LBA_ENTRY_VAL 256 /* Represents 256 LBAs */
244 #define BBM_LOG_SIGNATURE 0xabadb10c
246 struct bbm_log_block_addr
{
249 } __attribute__ ((__packed__
));
251 struct bbm_log_entry
{
252 __u8 marked_count
; /* Number of blocks marked - 1 */
253 __u8 disk_ordinal
; /* Disk entry within the imsm_super */
254 struct bbm_log_block_addr defective_block_start
;
255 } __attribute__ ((__packed__
));
258 __u32 signature
; /* 0xABADB10C */
260 struct bbm_log_entry marked_block_entries
[BBM_LOG_MAX_ENTRIES
];
261 } __attribute__ ((__packed__
));
264 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
267 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
269 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
271 #define MIGR_REC_BUF_SECTORS 1 /* size of migr_record i/o buffer in sectors */
272 #define MIGR_REC_SECTOR_POSITION 1 /* migr_record position offset on disk,
273 * MIGR_REC_BUF_SECTORS <= MIGR_REC_SECTOR_POS
276 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
277 * be recovered using srcMap */
278 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
279 * already been migrated and must
280 * be recovered from checkpoint area */
282 #define PPL_ENTRY_SPACE (128 * 1024) /* Size of the PPL, without the header */
285 __u32 rec_status
; /* Status used to determine how to restart
286 * migration in case it aborts
288 __u32 curr_migr_unit
; /* 0..numMigrUnits-1 */
289 __u32 family_num
; /* Family number of MPB
290 * containing the RaidDev
291 * that is migrating */
292 __u32 ascending_migr
; /* True if migrating in increasing
294 __u32 blocks_per_unit
; /* Num disk blocks per unit of operation */
295 __u32 dest_depth_per_unit
; /* Num member blocks each destMap
297 * advances per unit-of-operation */
298 __u32 ckpt_area_pba
; /* Pba of first block of ckpt copy area */
299 __u32 dest_1st_member_lba
; /* First member lba on first
300 * stripe of destination */
301 __u32 num_migr_units
; /* Total num migration units-of-op */
302 __u32 post_migr_vol_cap
; /* Size of volume after
303 * migration completes */
304 __u32 post_migr_vol_cap_hi
; /* Expansion space for LBA64 */
305 __u32 ckpt_read_disk_num
; /* Which member disk in destSubMap[0] the
306 * migration ckpt record was read from
307 * (for recovered migrations) */
308 } __attribute__ ((__packed__
));
313 * 2: metadata does not match
321 struct md_list
*next
;
324 #define pr_vrb(fmt, arg...) (void) (verbose && pr_err(fmt, ##arg))
326 static __u8
migr_type(struct imsm_dev
*dev
)
328 if (dev
->vol
.migr_type
== MIGR_VERIFY
&&
329 dev
->status
& DEV_VERIFY_AND_FIX
)
332 return dev
->vol
.migr_type
;
335 static void set_migr_type(struct imsm_dev
*dev
, __u8 migr_type
)
337 /* for compatibility with older oroms convert MIGR_REPAIR, into
338 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
340 if (migr_type
== MIGR_REPAIR
) {
341 dev
->vol
.migr_type
= MIGR_VERIFY
;
342 dev
->status
|= DEV_VERIFY_AND_FIX
;
344 dev
->vol
.migr_type
= migr_type
;
345 dev
->status
&= ~DEV_VERIFY_AND_FIX
;
349 static unsigned int sector_count(__u32 bytes
, unsigned int sector_size
)
351 return ROUND_UP(bytes
, sector_size
) / sector_size
;
354 static unsigned int mpb_sectors(struct imsm_super
*mpb
,
355 unsigned int sector_size
)
357 return sector_count(__le32_to_cpu(mpb
->mpb_size
), sector_size
);
361 struct imsm_dev
*dev
;
362 struct intel_dev
*next
;
367 enum sys_dev_type type
;
370 struct intel_hba
*next
;
377 /* internal representation of IMSM metadata */
380 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
381 struct imsm_super
*anchor
; /* immovable parameters */
384 void *migr_rec_buf
; /* buffer for I/O operations */
385 struct migr_record
*migr_rec
; /* migration record */
387 int clean_migration_record_by_mdmon
; /* when reshape is switched to next
388 array, it indicates that mdmon is allowed to clean migration
390 size_t len
; /* size of the 'buf' allocation */
391 size_t extra_space
; /* extra space in 'buf' that is not used yet */
392 void *next_buf
; /* for realloc'ing buf from the manager */
394 int updates_pending
; /* count of pending updates for mdmon */
395 int current_vol
; /* index of raid device undergoing creation */
396 unsigned long long create_offset
; /* common start for 'current_vol' */
397 __u32 random
; /* random data for seeding new family numbers */
398 struct intel_dev
*devlist
;
399 unsigned int sector_size
; /* sector size of used member drives */
403 __u8 serial
[MAX_RAID_SERIAL_LEN
];
406 struct imsm_disk disk
;
409 struct extent
*e
; /* for determining freespace @ create */
410 int raiddisk
; /* slot to fill in autolayout */
412 } *disks
, *current_disk
;
413 struct dl
*disk_mgmt_list
; /* list of disks to add/remove while mdmon
415 struct dl
*missing
; /* disks removed while we weren't looking */
416 struct bbm_log
*bbm_log
;
417 struct intel_hba
*hba
; /* device path of the raid controller for this metadata */
418 const struct imsm_orom
*orom
; /* platform firmware support */
419 struct intel_super
*next
; /* (temp) list for disambiguating family_num */
420 struct md_bb bb
; /* memory for get_bad_blocks call */
424 struct imsm_disk disk
;
425 #define IMSM_UNKNOWN_OWNER (-1)
427 struct intel_disk
*next
;
431 unsigned long long start
, size
;
434 /* definitions of reshape process types */
435 enum imsm_reshape_type
{
441 /* definition of messages passed to imsm_process_update */
442 enum imsm_update_type
{
443 update_activate_spare
,
447 update_add_remove_disk
,
448 update_reshape_container_disks
,
449 update_reshape_migration
,
451 update_general_migration_checkpoint
,
453 update_prealloc_badblocks_mem
,
457 struct imsm_update_activate_spare
{
458 enum imsm_update_type type
;
462 struct imsm_update_activate_spare
*next
;
468 unsigned long long size
;
475 enum takeover_direction
{
479 struct imsm_update_takeover
{
480 enum imsm_update_type type
;
482 enum takeover_direction direction
;
485 struct imsm_update_reshape
{
486 enum imsm_update_type type
;
490 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
493 struct imsm_update_reshape_migration
{
494 enum imsm_update_type type
;
497 /* fields for array migration changes
504 int new_disks
[1]; /* new_raid_disks - old_raid_disks makedev number */
507 struct imsm_update_size_change
{
508 enum imsm_update_type type
;
513 struct imsm_update_general_migration_checkpoint
{
514 enum imsm_update_type type
;
515 __u32 curr_migr_unit
;
519 __u8 serial
[MAX_RAID_SERIAL_LEN
];
522 struct imsm_update_create_array
{
523 enum imsm_update_type type
;
528 struct imsm_update_kill_array
{
529 enum imsm_update_type type
;
533 struct imsm_update_rename_array
{
534 enum imsm_update_type type
;
535 __u8 name
[MAX_RAID_SERIAL_LEN
];
539 struct imsm_update_add_remove_disk
{
540 enum imsm_update_type type
;
543 struct imsm_update_prealloc_bb_mem
{
544 enum imsm_update_type type
;
547 struct imsm_update_rwh_policy
{
548 enum imsm_update_type type
;
553 static const char *_sys_dev_type
[] = {
554 [SYS_DEV_UNKNOWN
] = "Unknown",
555 [SYS_DEV_SAS
] = "SAS",
556 [SYS_DEV_SATA
] = "SATA",
557 [SYS_DEV_NVME
] = "NVMe",
558 [SYS_DEV_VMD
] = "VMD"
561 const char *get_sys_dev_type(enum sys_dev_type type
)
563 if (type
>= SYS_DEV_MAX
)
564 type
= SYS_DEV_UNKNOWN
;
566 return _sys_dev_type
[type
];
569 static struct intel_hba
* alloc_intel_hba(struct sys_dev
*device
)
571 struct intel_hba
*result
= xmalloc(sizeof(*result
));
573 result
->type
= device
->type
;
574 result
->path
= xstrdup(device
->path
);
576 if (result
->path
&& (result
->pci_id
= strrchr(result
->path
, '/')) != NULL
)
582 static struct intel_hba
* find_intel_hba(struct intel_hba
*hba
, struct sys_dev
*device
)
584 struct intel_hba
*result
;
586 for (result
= hba
; result
; result
= result
->next
) {
587 if (result
->type
== device
->type
&& strcmp(result
->path
, device
->path
) == 0)
593 static int attach_hba_to_super(struct intel_super
*super
, struct sys_dev
*device
)
595 struct intel_hba
*hba
;
597 /* check if disk attached to Intel HBA */
598 hba
= find_intel_hba(super
->hba
, device
);
601 /* Check if HBA is already attached to super */
602 if (super
->hba
== NULL
) {
603 super
->hba
= alloc_intel_hba(device
);
608 /* Intel metadata allows for all disks attached to the same type HBA.
609 * Do not support HBA types mixing
611 if (device
->type
!= hba
->type
)
614 /* Multiple same type HBAs can be used if they share the same OROM */
615 const struct imsm_orom
*device_orom
= get_orom_by_device_id(device
->dev_id
);
617 if (device_orom
!= super
->orom
)
623 hba
->next
= alloc_intel_hba(device
);
627 static struct sys_dev
* find_disk_attached_hba(int fd
, const char *devname
)
629 struct sys_dev
*list
, *elem
;
632 if ((list
= find_intel_devices()) == NULL
)
636 disk_path
= (char *) devname
;
638 disk_path
= diskfd_to_devpath(fd
);
643 for (elem
= list
; elem
; elem
= elem
->next
)
644 if (path_attached_to_hba(disk_path
, elem
->path
))
647 if (disk_path
!= devname
)
653 static int find_intel_hba_capability(int fd
, struct intel_super
*super
,
656 static struct supertype
*match_metadata_desc_imsm(char *arg
)
658 struct supertype
*st
;
660 if (strcmp(arg
, "imsm") != 0 &&
661 strcmp(arg
, "default") != 0
665 st
= xcalloc(1, sizeof(*st
));
666 st
->ss
= &super_imsm
;
667 st
->max_devs
= IMSM_MAX_DEVICES
;
668 st
->minor_version
= 0;
674 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
676 return &mpb
->sig
[MPB_SIG_LEN
];
680 /* retrieve a disk directly from the anchor when the anchor is known to be
681 * up-to-date, currently only at load time
683 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
685 if (index
>= mpb
->num_disks
)
687 return &mpb
->disk
[index
];
690 /* retrieve the disk description based on a index of the disk
693 static struct dl
*get_imsm_dl_disk(struct intel_super
*super
, __u8 index
)
697 for (d
= super
->disks
; d
; d
= d
->next
)
698 if (d
->index
== index
)
703 /* retrieve a disk from the parsed metadata */
704 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
708 dl
= get_imsm_dl_disk(super
, index
);
715 /* generate a checksum directly from the anchor when the anchor is known to be
716 * up-to-date, currently only at load or write_super after coalescing
718 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
720 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
721 __u32
*p
= (__u32
*) mpb
;
725 sum
+= __le32_to_cpu(*p
);
729 return sum
- __le32_to_cpu(mpb
->check_sum
);
732 static size_t sizeof_imsm_map(struct imsm_map
*map
)
734 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
737 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
739 /* A device can have 2 maps if it is in the middle of a migration.
741 * MAP_0 - we return the first map
742 * MAP_1 - we return the second map if it exists, else NULL
743 * MAP_X - we return the second map if it exists, else the first
745 struct imsm_map
*map
= &dev
->vol
.map
[0];
746 struct imsm_map
*map2
= NULL
;
748 if (dev
->vol
.migr_state
)
749 map2
= (void *)map
+ sizeof_imsm_map(map
);
751 switch (second_map
) {
768 /* return the size of the device.
769 * migr_state increases the returned size if map[0] were to be duplicated
771 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
773 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
774 sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
776 /* migrating means an additional map */
777 if (dev
->vol
.migr_state
)
778 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_1
));
780 size
+= sizeof_imsm_map(get_imsm_map(dev
, MAP_0
));
786 /* retrieve disk serial number list from a metadata update */
787 static struct disk_info
*get_disk_info(struct imsm_update_create_array
*update
)
790 struct disk_info
*inf
;
792 inf
= u
+ sizeof(*update
) - sizeof(struct imsm_dev
) +
793 sizeof_imsm_dev(&update
->dev
, 0);
799 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
805 if (index
>= mpb
->num_raid_devs
)
808 /* devices start after all disks */
809 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
811 for (i
= 0; i
<= index
; i
++)
813 return _mpb
+ offset
;
815 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
820 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
822 struct intel_dev
*dv
;
824 if (index
>= super
->anchor
->num_raid_devs
)
826 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
827 if (dv
->index
== index
)
832 static inline unsigned long long __le48_to_cpu(const struct bbm_log_block_addr
835 return ((((__u64
)__le32_to_cpu(addr
->dw1
)) << 16) |
836 __le16_to_cpu(addr
->w1
));
839 static inline struct bbm_log_block_addr
__cpu_to_le48(unsigned long long sec
)
841 struct bbm_log_block_addr addr
;
843 addr
.w1
= __cpu_to_le16((__u16
)(sec
& 0xffff));
844 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];
979 #endif /* MDASSEMBLE */
981 /* allocate and load BBM log from metadata */
982 static int load_bbm_log(struct intel_super
*super
)
984 struct imsm_super
*mpb
= super
->anchor
;
985 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
987 super
->bbm_log
= xcalloc(1, sizeof(struct bbm_log
));
992 struct bbm_log
*log
= (void *)mpb
+
993 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
997 if (bbm_log_size
< sizeof(log
->signature
) +
998 sizeof(log
->entry_count
))
1001 entry_count
= __le32_to_cpu(log
->entry_count
);
1002 if ((__le32_to_cpu(log
->signature
) != BBM_LOG_SIGNATURE
) ||
1003 (entry_count
> BBM_LOG_MAX_ENTRIES
))
1007 sizeof(log
->signature
) + sizeof(log
->entry_count
) +
1008 entry_count
* sizeof(struct bbm_log_entry
))
1011 memcpy(super
->bbm_log
, log
, bbm_log_size
);
1013 super
->bbm_log
->signature
= __cpu_to_le32(BBM_LOG_SIGNATURE
);
1014 super
->bbm_log
->entry_count
= 0;
1020 /* checks if bad block is within volume boundaries */
1021 static int is_bad_block_in_volume(const struct bbm_log_entry
*entry
,
1022 const unsigned long long start_sector
,
1023 const unsigned long long size
)
1025 unsigned long long bb_start
;
1026 unsigned long long bb_end
;
1028 bb_start
= __le48_to_cpu(&entry
->defective_block_start
);
1029 bb_end
= bb_start
+ (entry
->marked_count
+ 1);
1031 if (((bb_start
>= start_sector
) && (bb_start
< start_sector
+ size
)) ||
1032 ((bb_end
>= start_sector
) && (bb_end
<= start_sector
+ size
)))
1038 /* get list of bad blocks on a drive for a volume */
1039 static void get_volume_badblocks(const struct bbm_log
*log
, const __u8 idx
,
1040 const unsigned long long start_sector
,
1041 const unsigned long long size
,
1047 for (i
= 0; i
< log
->entry_count
; i
++) {
1048 const struct bbm_log_entry
*ent
=
1049 &log
->marked_block_entries
[i
];
1050 struct md_bb_entry
*bb
;
1052 if ((ent
->disk_ordinal
== idx
) &&
1053 is_bad_block_in_volume(ent
, start_sector
, size
)) {
1055 if (!bbs
->entries
) {
1056 bbs
->entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
1062 bb
= &bbs
->entries
[count
++];
1063 bb
->sector
= __le48_to_cpu(&ent
->defective_block_start
);
1064 bb
->length
= ent
->marked_count
+ 1;
1072 * == MAP_0 get first map
1073 * == MAP_1 get second map
1074 * == MAP_X than get map according to the current migr_state
1076 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
,
1080 struct imsm_map
*map
;
1082 map
= get_imsm_map(dev
, second_map
);
1084 /* top byte identifies disk under rebuild */
1085 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1088 #define ord_to_idx(ord) (((ord) << 8) >> 8)
1089 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
, int second_map
)
1091 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, second_map
);
1093 return ord_to_idx(ord
);
1096 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
1098 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
1101 static int get_imsm_disk_slot(struct imsm_map
*map
, unsigned idx
)
1106 for (slot
= 0; slot
< map
->num_members
; slot
++) {
1107 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
1108 if (ord_to_idx(ord
) == idx
)
1115 static int get_imsm_raid_level(struct imsm_map
*map
)
1117 if (map
->raid_level
== 1) {
1118 if (map
->num_members
== 2)
1124 return map
->raid_level
;
1127 static int cmp_extent(const void *av
, const void *bv
)
1129 const struct extent
*a
= av
;
1130 const struct extent
*b
= bv
;
1131 if (a
->start
< b
->start
)
1133 if (a
->start
> b
->start
)
1138 static int count_memberships(struct dl
*dl
, struct intel_super
*super
)
1140 int memberships
= 0;
1143 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1144 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1145 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1147 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
1154 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
);
1156 static int split_ull(unsigned long long n
, __u32
*lo
, __u32
*hi
)
1158 if (lo
== 0 || hi
== 0)
1160 *lo
= __le32_to_cpu((unsigned)n
);
1161 *hi
= __le32_to_cpu((unsigned)(n
>> 32));
1165 static unsigned long long join_u32(__u32 lo
, __u32 hi
)
1167 return (unsigned long long)__le32_to_cpu(lo
) |
1168 (((unsigned long long)__le32_to_cpu(hi
)) << 32);
1171 static unsigned long long total_blocks(struct imsm_disk
*disk
)
1175 return join_u32(disk
->total_blocks_lo
, disk
->total_blocks_hi
);
1178 static unsigned long long pba_of_lba0(struct imsm_map
*map
)
1182 return join_u32(map
->pba_of_lba0_lo
, map
->pba_of_lba0_hi
);
1185 static unsigned long long blocks_per_member(struct imsm_map
*map
)
1189 return join_u32(map
->blocks_per_member_lo
, map
->blocks_per_member_hi
);
1192 static unsigned long long num_data_stripes(struct imsm_map
*map
)
1196 return join_u32(map
->num_data_stripes_lo
, map
->num_data_stripes_hi
);
1199 static void set_total_blocks(struct imsm_disk
*disk
, unsigned long long n
)
1201 split_ull(n
, &disk
->total_blocks_lo
, &disk
->total_blocks_hi
);
1204 static void set_pba_of_lba0(struct imsm_map
*map
, unsigned long long n
)
1206 split_ull(n
, &map
->pba_of_lba0_lo
, &map
->pba_of_lba0_hi
);
1209 static void set_blocks_per_member(struct imsm_map
*map
, unsigned long long n
)
1211 split_ull(n
, &map
->blocks_per_member_lo
, &map
->blocks_per_member_hi
);
1214 static void set_num_data_stripes(struct imsm_map
*map
, unsigned long long n
)
1216 split_ull(n
, &map
->num_data_stripes_lo
, &map
->num_data_stripes_hi
);
1219 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
1221 /* find a list of used extents on the given physical device */
1222 struct extent
*rv
, *e
;
1224 int memberships
= count_memberships(dl
, super
);
1227 /* trim the reserved area for spares, so they can join any array
1228 * regardless of whether the OROM has assigned sectors from the
1229 * IMSM_RESERVED_SECTORS region
1231 if (dl
->index
== -1)
1232 reservation
= imsm_min_reserved_sectors(super
);
1234 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1236 rv
= xcalloc(sizeof(struct extent
), (memberships
+ 1));
1239 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1240 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1241 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1243 if (get_imsm_disk_slot(map
, dl
->index
) >= 0) {
1244 e
->start
= pba_of_lba0(map
);
1245 e
->size
= blocks_per_member(map
);
1249 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
1251 /* determine the start of the metadata
1252 * when no raid devices are defined use the default
1253 * ...otherwise allow the metadata to truncate the value
1254 * as is the case with older versions of imsm
1257 struct extent
*last
= &rv
[memberships
- 1];
1258 unsigned long long remainder
;
1260 remainder
= total_blocks(&dl
->disk
) - (last
->start
+ last
->size
);
1261 /* round down to 1k block to satisfy precision of the kernel
1265 /* make sure remainder is still sane */
1266 if (remainder
< (unsigned)ROUND_UP(super
->len
, 512) >> 9)
1267 remainder
= ROUND_UP(super
->len
, 512) >> 9;
1268 if (reservation
> remainder
)
1269 reservation
= remainder
;
1271 e
->start
= total_blocks(&dl
->disk
) - reservation
;
1276 /* try to determine how much space is reserved for metadata from
1277 * the last get_extents() entry, otherwise fallback to the
1280 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
1286 /* for spares just return a minimal reservation which will grow
1287 * once the spare is picked up by an array
1289 if (dl
->index
== -1)
1290 return MPB_SECTOR_CNT
;
1292 e
= get_extents(super
, dl
);
1294 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1296 /* scroll to last entry */
1297 for (i
= 0; e
[i
].size
; i
++)
1300 rv
= total_blocks(&dl
->disk
) - e
[i
].start
;
1307 static int is_spare(struct imsm_disk
*disk
)
1309 return (disk
->status
& SPARE_DISK
) == SPARE_DISK
;
1312 static int is_configured(struct imsm_disk
*disk
)
1314 return (disk
->status
& CONFIGURED_DISK
) == CONFIGURED_DISK
;
1317 static int is_failed(struct imsm_disk
*disk
)
1319 return (disk
->status
& FAILED_DISK
) == FAILED_DISK
;
1322 static int is_journal(struct imsm_disk
*disk
)
1324 return (disk
->status
& JOURNAL_DISK
) == JOURNAL_DISK
;
1327 /* try to determine how much space is reserved for metadata from
1328 * the last get_extents() entry on the smallest active disk,
1329 * otherwise fallback to the default
1331 static __u32
imsm_min_reserved_sectors(struct intel_super
*super
)
1335 unsigned long long min_active
;
1337 __u32 rv
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1338 struct dl
*dl
, *dl_min
= NULL
;
1344 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1347 unsigned long long blocks
= total_blocks(&dl
->disk
);
1348 if (blocks
< min_active
|| min_active
== 0) {
1350 min_active
= blocks
;
1356 /* find last lba used by subarrays on the smallest active disk */
1357 e
= get_extents(super
, dl_min
);
1360 for (i
= 0; e
[i
].size
; i
++)
1363 remainder
= min_active
- e
[i
].start
;
1366 /* to give priority to recovery we should not require full
1367 IMSM_RESERVED_SECTORS from the spare */
1368 rv
= MPB_SECTOR_CNT
+ NUM_BLOCKS_DIRTY_STRIPE_REGION
;
1370 /* if real reservation is smaller use that value */
1371 return (remainder
< rv
) ? remainder
: rv
;
1374 /* Return minimum size of a spare that can be used in this array*/
1375 static unsigned long long min_acceptable_spare_size_imsm(struct supertype
*st
)
1377 struct intel_super
*super
= st
->sb
;
1381 unsigned long long rv
= 0;
1385 /* find first active disk in array */
1387 while (dl
&& (is_failed(&dl
->disk
) || dl
->index
== -1))
1391 /* find last lba used by subarrays */
1392 e
= get_extents(super
, dl
);
1395 for (i
= 0; e
[i
].size
; i
++)
1398 rv
= e
[i
-1].start
+ e
[i
-1].size
;
1401 /* add the amount of space needed for metadata */
1402 rv
= rv
+ imsm_min_reserved_sectors(super
);
1407 static int is_gen_migration(struct imsm_dev
*dev
);
1409 #define IMSM_4K_DIV 8
1412 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
1413 struct imsm_dev
*dev
);
1415 static void print_imsm_dev(struct intel_super
*super
,
1416 struct imsm_dev
*dev
,
1422 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1423 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
1427 printf("[%.16s]:\n", dev
->volume
);
1428 printf(" UUID : %s\n", uuid
);
1429 printf(" RAID Level : %d", get_imsm_raid_level(map
));
1431 printf(" <-- %d", get_imsm_raid_level(map2
));
1433 printf(" Members : %d", map
->num_members
);
1435 printf(" <-- %d", map2
->num_members
);
1437 printf(" Slots : [");
1438 for (i
= 0; i
< map
->num_members
; i
++) {
1439 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_0
);
1440 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1445 for (i
= 0; i
< map2
->num_members
; i
++) {
1446 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_1
);
1447 printf("%s", ord
& IMSM_ORD_REBUILD
? "_" : "U");
1452 printf(" Failed disk : ");
1453 if (map
->failed_disk_num
== 0xff)
1456 printf("%i", map
->failed_disk_num
);
1458 slot
= get_imsm_disk_slot(map
, disk_idx
);
1460 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
1461 printf(" This Slot : %d%s\n", slot
,
1462 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
1464 printf(" This Slot : ?\n");
1465 sz
= __le32_to_cpu(dev
->size_high
);
1467 sz
+= __le32_to_cpu(dev
->size_low
);
1468 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
1469 human_size(sz
* 512));
1470 sz
= blocks_per_member(map
);
1471 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
1472 human_size(sz
* 512));
1473 printf(" Sector Offset : %llu\n",
1475 printf(" Num Stripes : %llu\n",
1476 num_data_stripes(map
));
1477 printf(" Chunk Size : %u KiB",
1478 __le16_to_cpu(map
->blocks_per_strip
) / 2);
1480 printf(" <-- %u KiB",
1481 __le16_to_cpu(map2
->blocks_per_strip
) / 2);
1483 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
1484 printf(" Migrate State : ");
1485 if (dev
->vol
.migr_state
) {
1486 if (migr_type(dev
) == MIGR_INIT
)
1487 printf("initialize\n");
1488 else if (migr_type(dev
) == MIGR_REBUILD
)
1489 printf("rebuild\n");
1490 else if (migr_type(dev
) == MIGR_VERIFY
)
1492 else if (migr_type(dev
) == MIGR_GEN_MIGR
)
1493 printf("general migration\n");
1494 else if (migr_type(dev
) == MIGR_STATE_CHANGE
)
1495 printf("state change\n");
1496 else if (migr_type(dev
) == MIGR_REPAIR
)
1499 printf("<unknown:%d>\n", migr_type(dev
));
1502 printf(" Map State : %s", map_state_str
[map
->map_state
]);
1503 if (dev
->vol
.migr_state
) {
1504 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
1506 printf(" <-- %s", map_state_str
[map
->map_state
]);
1507 printf("\n Checkpoint : %u ",
1508 __le32_to_cpu(dev
->vol
.curr_migr_unit
));
1509 if (is_gen_migration(dev
) && (slot
> 1 || slot
< 0))
1512 printf("(%llu)", (unsigned long long)
1513 blocks_per_migr_unit(super
, dev
));
1516 printf(" Dirty State : %s\n", (dev
->vol
.dirty
& RAIDVOL_DIRTY
) ?
1518 printf(" RWH Policy : ");
1519 if (dev
->rwh_policy
== RWH_OFF
)
1521 else if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
1522 printf("PPL distributed\n");
1523 else if (dev
->rwh_policy
== RWH_JOURNALING_DRIVE
)
1524 printf("PPL journaling drive\n");
1526 printf("<unknown:%d>\n", dev
->rwh_policy
);
1529 static void print_imsm_disk(struct imsm_disk
*disk
,
1532 unsigned int sector_size
) {
1533 char str
[MAX_RAID_SERIAL_LEN
+ 1];
1536 if (index
< -1 || !disk
)
1540 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
1542 printf(" Disk%02d Serial : %s\n", index
, str
);
1544 printf(" Disk Serial : %s\n", str
);
1545 printf(" State :%s%s%s%s\n", is_spare(disk
) ? " spare" : "",
1546 is_configured(disk
) ? " active" : "",
1547 is_failed(disk
) ? " failed" : "",
1548 is_journal(disk
) ? " journal" : "");
1549 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
1550 sz
= total_blocks(disk
) - reserved
;
1551 printf(" Usable Size : %llu%s\n",
1552 (unsigned long long)sz
* 512 / sector_size
,
1553 human_size(sz
* 512));
1556 void convert_to_4k_imsm_migr_rec(struct intel_super
*super
)
1558 struct migr_record
*migr_rec
= super
->migr_rec
;
1560 migr_rec
->blocks_per_unit
/= IMSM_4K_DIV
;
1561 migr_rec
->ckpt_area_pba
/= IMSM_4K_DIV
;
1562 migr_rec
->dest_1st_member_lba
/= IMSM_4K_DIV
;
1563 migr_rec
->dest_depth_per_unit
/= IMSM_4K_DIV
;
1564 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1565 migr_rec
->post_migr_vol_cap_hi
) / IMSM_4K_DIV
),
1566 &migr_rec
->post_migr_vol_cap
, &migr_rec
->post_migr_vol_cap_hi
);
1569 void convert_to_4k_imsm_disk(struct imsm_disk
*disk
)
1571 set_total_blocks(disk
, (total_blocks(disk
)/IMSM_4K_DIV
));
1574 void convert_to_4k(struct intel_super
*super
)
1576 struct imsm_super
*mpb
= super
->anchor
;
1577 struct imsm_disk
*disk
;
1579 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1581 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1582 disk
= __get_imsm_disk(mpb
, i
);
1584 convert_to_4k_imsm_disk(disk
);
1586 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1587 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1588 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1590 split_ull((join_u32(dev
->size_low
, dev
->size_high
)/IMSM_4K_DIV
),
1591 &dev
->size_low
, &dev
->size_high
);
1592 dev
->vol
.curr_migr_unit
/= IMSM_4K_DIV
;
1595 set_blocks_per_member(map
, blocks_per_member(map
)/IMSM_4K_DIV
);
1596 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1597 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1599 if (dev
->vol
.migr_state
) {
1601 map
= get_imsm_map(dev
, MAP_1
);
1602 set_blocks_per_member(map
,
1603 blocks_per_member(map
)/IMSM_4K_DIV
);
1604 map
->blocks_per_strip
/= IMSM_4K_DIV
;
1605 set_pba_of_lba0(map
, pba_of_lba0(map
)/IMSM_4K_DIV
);
1609 struct bbm_log
*log
= (void *)mpb
+
1610 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1613 for (i
= 0; i
< log
->entry_count
; i
++) {
1614 struct bbm_log_entry
*entry
=
1615 &log
->marked_block_entries
[i
];
1617 __u8 count
= entry
->marked_count
+ 1;
1618 unsigned long long sector
=
1619 __le48_to_cpu(&entry
->defective_block_start
);
1621 entry
->defective_block_start
=
1622 __cpu_to_le48(sector
/IMSM_4K_DIV
);
1623 entry
->marked_count
= max(count
/IMSM_4K_DIV
, 1) - 1;
1627 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1630 void examine_migr_rec_imsm(struct intel_super
*super
)
1632 struct migr_record
*migr_rec
= super
->migr_rec
;
1633 struct imsm_super
*mpb
= super
->anchor
;
1636 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1637 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1638 struct imsm_map
*map
;
1641 if (is_gen_migration(dev
) == 0)
1644 printf("\nMigration Record Information:");
1646 /* first map under migration */
1647 map
= get_imsm_map(dev
, MAP_0
);
1649 slot
= get_imsm_disk_slot(map
, super
->disks
->index
);
1650 if (map
== NULL
|| slot
> 1 || slot
< 0) {
1651 printf(" Empty\n ");
1652 printf("Examine one of first two disks in array\n");
1655 printf("\n Status : ");
1656 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
1659 printf("Contains Data\n");
1660 printf(" Current Unit : %u\n",
1661 __le32_to_cpu(migr_rec
->curr_migr_unit
));
1662 printf(" Family : %u\n",
1663 __le32_to_cpu(migr_rec
->family_num
));
1664 printf(" Ascending : %u\n",
1665 __le32_to_cpu(migr_rec
->ascending_migr
));
1666 printf(" Blocks Per Unit : %u\n",
1667 __le32_to_cpu(migr_rec
->blocks_per_unit
));
1668 printf(" Dest. Depth Per Unit : %u\n",
1669 __le32_to_cpu(migr_rec
->dest_depth_per_unit
));
1670 printf(" Checkpoint Area pba : %u\n",
1671 __le32_to_cpu(migr_rec
->ckpt_area_pba
));
1672 printf(" First member lba : %u\n",
1673 __le32_to_cpu(migr_rec
->dest_1st_member_lba
));
1674 printf(" Total Number of Units : %u\n",
1675 __le32_to_cpu(migr_rec
->num_migr_units
));
1676 printf(" Size of volume : %u\n",
1677 __le32_to_cpu(migr_rec
->post_migr_vol_cap
));
1678 printf(" Expansion space for LBA64 : %u\n",
1679 __le32_to_cpu(migr_rec
->post_migr_vol_cap_hi
));
1680 printf(" Record was read from : %u\n",
1681 __le32_to_cpu(migr_rec
->ckpt_read_disk_num
));
1686 #endif /* MDASSEMBLE */
1688 void convert_from_4k_imsm_migr_rec(struct intel_super
*super
)
1690 struct migr_record
*migr_rec
= super
->migr_rec
;
1692 migr_rec
->blocks_per_unit
*= IMSM_4K_DIV
;
1693 migr_rec
->ckpt_area_pba
*= IMSM_4K_DIV
;
1694 migr_rec
->dest_1st_member_lba
*= IMSM_4K_DIV
;
1695 migr_rec
->dest_depth_per_unit
*= IMSM_4K_DIV
;
1696 split_ull((join_u32(migr_rec
->post_migr_vol_cap
,
1697 migr_rec
->post_migr_vol_cap_hi
) * IMSM_4K_DIV
),
1698 &migr_rec
->post_migr_vol_cap
,
1699 &migr_rec
->post_migr_vol_cap_hi
);
1702 void convert_from_4k(struct intel_super
*super
)
1704 struct imsm_super
*mpb
= super
->anchor
;
1705 struct imsm_disk
*disk
;
1707 __u32 bbm_log_size
= __le32_to_cpu(mpb
->bbm_log_size
);
1709 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1710 disk
= __get_imsm_disk(mpb
, i
);
1712 set_total_blocks(disk
, (total_blocks(disk
)*IMSM_4K_DIV
));
1715 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1716 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1717 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
1719 split_ull((join_u32(dev
->size_low
, dev
->size_high
)*IMSM_4K_DIV
),
1720 &dev
->size_low
, &dev
->size_high
);
1721 dev
->vol
.curr_migr_unit
*= IMSM_4K_DIV
;
1724 set_blocks_per_member(map
, blocks_per_member(map
)*IMSM_4K_DIV
);
1725 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1726 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1728 if (dev
->vol
.migr_state
) {
1730 map
= get_imsm_map(dev
, MAP_1
);
1731 set_blocks_per_member(map
,
1732 blocks_per_member(map
)*IMSM_4K_DIV
);
1733 map
->blocks_per_strip
*= IMSM_4K_DIV
;
1734 set_pba_of_lba0(map
, pba_of_lba0(map
)*IMSM_4K_DIV
);
1738 struct bbm_log
*log
= (void *)mpb
+
1739 __le32_to_cpu(mpb
->mpb_size
) - bbm_log_size
;
1742 for (i
= 0; i
< log
->entry_count
; i
++) {
1743 struct bbm_log_entry
*entry
=
1744 &log
->marked_block_entries
[i
];
1746 __u8 count
= entry
->marked_count
+ 1;
1747 unsigned long long sector
=
1748 __le48_to_cpu(&entry
->defective_block_start
);
1750 entry
->defective_block_start
=
1751 __cpu_to_le48(sector
*IMSM_4K_DIV
);
1752 entry
->marked_count
= count
*IMSM_4K_DIV
- 1;
1756 mpb
->check_sum
= __gen_imsm_checksum(mpb
);
1759 /*******************************************************************************
1760 * function: imsm_check_attributes
1761 * Description: Function checks if features represented by attributes flags
1762 * are supported by mdadm.
1764 * attributes - Attributes read from metadata
1766 * 0 - passed attributes contains unsupported features flags
1767 * 1 - all features are supported
1768 ******************************************************************************/
1769 static int imsm_check_attributes(__u32 attributes
)
1772 __u32 not_supported
= MPB_ATTRIB_SUPPORTED
^0xffffffff;
1774 not_supported
&= ~MPB_ATTRIB_IGNORED
;
1776 not_supported
&= attributes
;
1777 if (not_supported
) {
1778 pr_err("(IMSM): Unsupported attributes : %x\n",
1779 (unsigned)__le32_to_cpu(not_supported
));
1780 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1781 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1782 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1784 if (not_supported
& MPB_ATTRIB_2TB
) {
1785 dprintf("\t\tMPB_ATTRIB_2TB\n");
1786 not_supported
^= MPB_ATTRIB_2TB
;
1788 if (not_supported
& MPB_ATTRIB_RAID0
) {
1789 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1790 not_supported
^= MPB_ATTRIB_RAID0
;
1792 if (not_supported
& MPB_ATTRIB_RAID1
) {
1793 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1794 not_supported
^= MPB_ATTRIB_RAID1
;
1796 if (not_supported
& MPB_ATTRIB_RAID10
) {
1797 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1798 not_supported
^= MPB_ATTRIB_RAID10
;
1800 if (not_supported
& MPB_ATTRIB_RAID1E
) {
1801 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1802 not_supported
^= MPB_ATTRIB_RAID1E
;
1804 if (not_supported
& MPB_ATTRIB_RAID5
) {
1805 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1806 not_supported
^= MPB_ATTRIB_RAID5
;
1808 if (not_supported
& MPB_ATTRIB_RAIDCNG
) {
1809 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1810 not_supported
^= MPB_ATTRIB_RAIDCNG
;
1812 if (not_supported
& MPB_ATTRIB_BBM
) {
1813 dprintf("\t\tMPB_ATTRIB_BBM\n");
1814 not_supported
^= MPB_ATTRIB_BBM
;
1816 if (not_supported
& MPB_ATTRIB_CHECKSUM_VERIFY
) {
1817 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1818 not_supported
^= MPB_ATTRIB_CHECKSUM_VERIFY
;
1820 if (not_supported
& MPB_ATTRIB_EXP_STRIPE_SIZE
) {
1821 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1822 not_supported
^= MPB_ATTRIB_EXP_STRIPE_SIZE
;
1824 if (not_supported
& MPB_ATTRIB_2TB_DISK
) {
1825 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1826 not_supported
^= MPB_ATTRIB_2TB_DISK
;
1828 if (not_supported
& MPB_ATTRIB_NEVER_USE2
) {
1829 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1830 not_supported
^= MPB_ATTRIB_NEVER_USE2
;
1832 if (not_supported
& MPB_ATTRIB_NEVER_USE
) {
1833 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1834 not_supported
^= MPB_ATTRIB_NEVER_USE
;
1838 dprintf("(IMSM): Unknown attributes : %x\n", not_supported
);
1847 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1849 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
1851 struct intel_super
*super
= st
->sb
;
1852 struct imsm_super
*mpb
= super
->anchor
;
1853 char str
[MAX_SIGNATURE_LENGTH
];
1858 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
1861 strncpy(str
, (char *)mpb
->sig
, MPB_SIG_LEN
);
1862 str
[MPB_SIG_LEN
-1] = '\0';
1863 printf(" Magic : %s\n", str
);
1864 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
1865 printf(" Version : %s\n", get_imsm_version(mpb
));
1866 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb
->orig_family_num
));
1867 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
1868 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
1869 printf(" Attributes : ");
1870 if (imsm_check_attributes(mpb
->attributes
))
1871 printf("All supported\n");
1873 printf("not supported\n");
1874 getinfo_super_imsm(st
, &info
, NULL
);
1875 fname_from_uuid(st
, &info
, nbuf
, ':');
1876 printf(" UUID : %s\n", nbuf
+ 5);
1877 sum
= __le32_to_cpu(mpb
->check_sum
);
1878 printf(" Checksum : %08x %s\n", sum
,
1879 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
1880 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
, super
->sector_size
));
1881 printf(" Disks : %d\n", mpb
->num_disks
);
1882 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
1883 print_imsm_disk(__get_imsm_disk(mpb
, super
->disks
->index
),
1884 super
->disks
->index
, reserved
, super
->sector_size
);
1885 if (get_imsm_bbm_log_size(super
->bbm_log
)) {
1886 struct bbm_log
*log
= super
->bbm_log
;
1889 printf("Bad Block Management Log:\n");
1890 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
1891 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
1892 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
1894 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1896 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1898 super
->current_vol
= i
;
1899 getinfo_super_imsm(st
, &info
, NULL
);
1900 fname_from_uuid(st
, &info
, nbuf
, ':');
1901 print_imsm_dev(super
, dev
, nbuf
+ 5, super
->disks
->index
);
1903 for (i
= 0; i
< mpb
->num_disks
; i
++) {
1904 if (i
== super
->disks
->index
)
1906 print_imsm_disk(__get_imsm_disk(mpb
, i
), i
, reserved
,
1907 super
->sector_size
);
1910 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1911 if (dl
->index
== -1)
1912 print_imsm_disk(&dl
->disk
, -1, reserved
,
1913 super
->sector_size
);
1915 examine_migr_rec_imsm(super
);
1918 static void brief_examine_super_imsm(struct supertype
*st
, int verbose
)
1920 /* We just write a generic IMSM ARRAY entry */
1923 struct intel_super
*super
= st
->sb
;
1925 if (!super
->anchor
->num_raid_devs
) {
1926 printf("ARRAY metadata=imsm\n");
1930 getinfo_super_imsm(st
, &info
, NULL
);
1931 fname_from_uuid(st
, &info
, nbuf
, ':');
1932 printf("ARRAY metadata=imsm UUID=%s\n", nbuf
+ 5);
1935 static void brief_examine_subarrays_imsm(struct supertype
*st
, int verbose
)
1937 /* We just write a generic IMSM ARRAY entry */
1941 struct intel_super
*super
= st
->sb
;
1944 if (!super
->anchor
->num_raid_devs
)
1947 getinfo_super_imsm(st
, &info
, NULL
);
1948 fname_from_uuid(st
, &info
, nbuf
, ':');
1949 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1950 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
1952 super
->current_vol
= i
;
1953 getinfo_super_imsm(st
, &info
, NULL
);
1954 fname_from_uuid(st
, &info
, nbuf1
, ':');
1955 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1956 dev
->volume
, nbuf
+ 5, i
, nbuf1
+ 5);
1960 static void export_examine_super_imsm(struct supertype
*st
)
1962 struct intel_super
*super
= st
->sb
;
1963 struct imsm_super
*mpb
= super
->anchor
;
1967 getinfo_super_imsm(st
, &info
, NULL
);
1968 fname_from_uuid(st
, &info
, nbuf
, ':');
1969 printf("MD_METADATA=imsm\n");
1970 printf("MD_LEVEL=container\n");
1971 printf("MD_UUID=%s\n", nbuf
+5);
1972 printf("MD_DEVICES=%u\n", mpb
->num_disks
);
1975 static int copy_metadata_imsm(struct supertype
*st
, int from
, int to
)
1977 /* The second last sector of the device contains
1978 * the "struct imsm_super" metadata.
1979 * This contains mpb_size which is the size in bytes of the
1980 * extended metadata. This is located immediately before
1982 * We want to read all that, plus the last sector which
1983 * may contain a migration record, and write it all
1987 unsigned long long dsize
, offset
;
1989 struct imsm_super
*sb
;
1990 struct intel_super
*super
= st
->sb
;
1991 unsigned int sector_size
= super
->sector_size
;
1992 unsigned int written
= 0;
1994 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, MAX_SECTOR_SIZE
) != 0)
1997 if (!get_dev_size(from
, NULL
, &dsize
))
2000 if (lseek64(from
, dsize
-(2*sector_size
), 0) < 0)
2002 if ((unsigned int)read(from
, buf
, sector_size
) != sector_size
)
2005 if (strncmp((char*)sb
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0)
2008 sectors
= mpb_sectors(sb
, sector_size
) + 2;
2009 offset
= dsize
- sectors
* sector_size
;
2010 if (lseek64(from
, offset
, 0) < 0 ||
2011 lseek64(to
, offset
, 0) < 0)
2013 while (written
< sectors
* sector_size
) {
2014 int n
= sectors
*sector_size
- written
;
2017 if (read(from
, buf
, n
) != n
)
2019 if (write(to
, buf
, n
) != n
)
2030 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
2035 getinfo_super_imsm(st
, &info
, NULL
);
2036 fname_from_uuid(st
, &info
, nbuf
, ':');
2037 printf("\n UUID : %s\n", nbuf
+ 5);
2040 static void brief_detail_super_imsm(struct supertype
*st
)
2044 getinfo_super_imsm(st
, &info
, NULL
);
2045 fname_from_uuid(st
, &info
, nbuf
, ':');
2046 printf(" UUID=%s", nbuf
+ 5);
2049 static int imsm_read_serial(int fd
, char *devname
, __u8
*serial
);
2050 static void fd2devname(int fd
, char *name
);
2052 static int ahci_enumerate_ports(const char *hba_path
, int port_count
, int host_base
, int verbose
)
2054 /* dump an unsorted list of devices attached to AHCI Intel storage
2055 * controller, as well as non-connected ports
2057 int hba_len
= strlen(hba_path
) + 1;
2062 unsigned long port_mask
= (1 << port_count
) - 1;
2064 if (port_count
> (int)sizeof(port_mask
) * 8) {
2066 pr_err("port_count %d out of range\n", port_count
);
2070 /* scroll through /sys/dev/block looking for devices attached to
2073 dir
= opendir("/sys/dev/block");
2077 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2088 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
2090 path
= devt_to_devpath(makedev(major
, minor
));
2093 if (!path_attached_to_hba(path
, hba_path
)) {
2099 /* retrieve the scsi device type */
2100 if (asprintf(&device
, "/sys/dev/block/%d:%d/device/xxxxxxx", major
, minor
) < 0) {
2102 pr_err("failed to allocate 'device'\n");
2106 sprintf(device
, "/sys/dev/block/%d:%d/device/type", major
, minor
);
2107 if (load_sys(device
, buf
, sizeof(buf
)) != 0) {
2109 pr_err("failed to read device type for %s\n",
2115 type
= strtoul(buf
, NULL
, 10);
2117 /* if it's not a disk print the vendor and model */
2118 if (!(type
== 0 || type
== 7 || type
== 14)) {
2121 sprintf(device
, "/sys/dev/block/%d:%d/device/vendor", major
, minor
);
2122 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2123 strncpy(vendor
, buf
, sizeof(vendor
));
2124 vendor
[sizeof(vendor
) - 1] = '\0';
2125 c
= (char *) &vendor
[sizeof(vendor
) - 1];
2126 while (isspace(*c
) || *c
== '\0')
2130 sprintf(device
, "/sys/dev/block/%d:%d/device/model", major
, minor
);
2131 if (load_sys(device
, buf
, sizeof(buf
)) == 0) {
2132 strncpy(model
, buf
, sizeof(model
));
2133 model
[sizeof(model
) - 1] = '\0';
2134 c
= (char *) &model
[sizeof(model
) - 1];
2135 while (isspace(*c
) || *c
== '\0')
2139 if (vendor
[0] && model
[0])
2140 sprintf(buf
, "%.64s %.64s", vendor
, model
);
2142 switch (type
) { /* numbers from hald/linux/device.c */
2143 case 1: sprintf(buf
, "tape"); break;
2144 case 2: sprintf(buf
, "printer"); break;
2145 case 3: sprintf(buf
, "processor"); break;
2147 case 5: sprintf(buf
, "cdrom"); break;
2148 case 6: sprintf(buf
, "scanner"); break;
2149 case 8: sprintf(buf
, "media_changer"); break;
2150 case 9: sprintf(buf
, "comm"); break;
2151 case 12: sprintf(buf
, "raid"); break;
2152 default: sprintf(buf
, "unknown");
2158 /* chop device path to 'host%d' and calculate the port number */
2159 c
= strchr(&path
[hba_len
], '/');
2162 pr_err("%s - invalid path name\n", path
+ hba_len
);
2167 if ((sscanf(&path
[hba_len
], "ata%d", &port
) == 1) ||
2168 ((sscanf(&path
[hba_len
], "host%d", &port
) == 1)))
2172 *c
= '/'; /* repair the full string */
2173 pr_err("failed to determine port number for %s\n",
2180 /* mark this port as used */
2181 port_mask
&= ~(1 << port
);
2183 /* print out the device information */
2185 printf(" Port%d : - non-disk device (%s) -\n", port
, buf
);
2189 fd
= dev_open(ent
->d_name
, O_RDONLY
);
2191 printf(" Port%d : - disk info unavailable -\n", port
);
2193 fd2devname(fd
, buf
);
2194 printf(" Port%d : %s", port
, buf
);
2195 if (imsm_read_serial(fd
, NULL
, (__u8
*) buf
) == 0)
2196 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN
, buf
);
2211 for (i
= 0; i
< port_count
; i
++)
2212 if (port_mask
& (1 << i
))
2213 printf(" Port%d : - no device attached -\n", i
);
2219 static int print_vmd_attached_devs(struct sys_dev
*hba
)
2227 if (hba
->type
!= SYS_DEV_VMD
)
2230 /* scroll through /sys/dev/block looking for devices attached to
2233 dir
= opendir("/sys/bus/pci/drivers/nvme");
2237 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2240 /* is 'ent' a device? check that the 'subsystem' link exists and
2241 * that its target matches 'bus'
2243 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s/subsystem",
2245 n
= readlink(path
, link
, sizeof(link
));
2246 if (n
< 0 || n
>= (int)sizeof(link
))
2249 c
= strrchr(link
, '/');
2252 if (strncmp("pci", c
+1, strlen("pci")) != 0)
2255 sprintf(path
, "/sys/bus/pci/drivers/nvme/%s", ent
->d_name
);
2257 rp
= realpath(path
, NULL
);
2261 if (path_attached_to_hba(rp
, hba
->path
)) {
2262 printf(" NVMe under VMD : %s\n", rp
);
2271 static void print_found_intel_controllers(struct sys_dev
*elem
)
2273 for (; elem
; elem
= elem
->next
) {
2274 pr_err("found Intel(R) ");
2275 if (elem
->type
== SYS_DEV_SATA
)
2276 fprintf(stderr
, "SATA ");
2277 else if (elem
->type
== SYS_DEV_SAS
)
2278 fprintf(stderr
, "SAS ");
2279 else if (elem
->type
== SYS_DEV_NVME
)
2280 fprintf(stderr
, "NVMe ");
2282 if (elem
->type
== SYS_DEV_VMD
)
2283 fprintf(stderr
, "VMD domain");
2285 fprintf(stderr
, "RAID controller");
2288 fprintf(stderr
, " at %s", elem
->pci_id
);
2289 fprintf(stderr
, ".\n");
2294 static int ahci_get_port_count(const char *hba_path
, int *port_count
)
2301 if ((dir
= opendir(hba_path
)) == NULL
)
2304 for (ent
= readdir(dir
); ent
; ent
= readdir(dir
)) {
2307 if ((sscanf(ent
->d_name
, "ata%d", &host
) != 1) &&
2308 ((sscanf(ent
->d_name
, "host%d", &host
) != 1)))
2310 if (*port_count
== 0)
2312 else if (host
< host_base
)
2315 if (host
+ 1 > *port_count
+ host_base
)
2316 *port_count
= host
+ 1 - host_base
;
2322 static void print_imsm_capability(const struct imsm_orom
*orom
)
2324 printf(" Platform : Intel(R) ");
2325 if (orom
->capabilities
== 0 && orom
->driver_features
== 0)
2326 printf("Matrix Storage Manager\n");
2328 printf("Rapid Storage Technology%s\n",
2329 imsm_orom_is_enterprise(orom
) ? " enterprise" : "");
2330 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2331 printf(" Version : %d.%d.%d.%d\n", orom
->major_ver
,
2332 orom
->minor_ver
, orom
->hotfix_ver
, orom
->build
);
2333 printf(" RAID Levels :%s%s%s%s%s\n",
2334 imsm_orom_has_raid0(orom
) ? " raid0" : "",
2335 imsm_orom_has_raid1(orom
) ? " raid1" : "",
2336 imsm_orom_has_raid1e(orom
) ? " raid1e" : "",
2337 imsm_orom_has_raid10(orom
) ? " raid10" : "",
2338 imsm_orom_has_raid5(orom
) ? " raid5" : "");
2339 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2340 imsm_orom_has_chunk(orom
, 2) ? " 2k" : "",
2341 imsm_orom_has_chunk(orom
, 4) ? " 4k" : "",
2342 imsm_orom_has_chunk(orom
, 8) ? " 8k" : "",
2343 imsm_orom_has_chunk(orom
, 16) ? " 16k" : "",
2344 imsm_orom_has_chunk(orom
, 32) ? " 32k" : "",
2345 imsm_orom_has_chunk(orom
, 64) ? " 64k" : "",
2346 imsm_orom_has_chunk(orom
, 128) ? " 128k" : "",
2347 imsm_orom_has_chunk(orom
, 256) ? " 256k" : "",
2348 imsm_orom_has_chunk(orom
, 512) ? " 512k" : "",
2349 imsm_orom_has_chunk(orom
, 1024*1) ? " 1M" : "",
2350 imsm_orom_has_chunk(orom
, 1024*2) ? " 2M" : "",
2351 imsm_orom_has_chunk(orom
, 1024*4) ? " 4M" : "",
2352 imsm_orom_has_chunk(orom
, 1024*8) ? " 8M" : "",
2353 imsm_orom_has_chunk(orom
, 1024*16) ? " 16M" : "",
2354 imsm_orom_has_chunk(orom
, 1024*32) ? " 32M" : "",
2355 imsm_orom_has_chunk(orom
, 1024*64) ? " 64M" : "");
2356 printf(" 2TB volumes :%s supported\n",
2357 (orom
->attr
& IMSM_OROM_ATTR_2TB
)?"":" not");
2358 printf(" 2TB disks :%s supported\n",
2359 (orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
)?"":" not");
2360 printf(" Max Disks : %d\n", orom
->tds
);
2361 printf(" Max Volumes : %d per array, %d per %s\n",
2362 orom
->vpa
, orom
->vphba
,
2363 imsm_orom_is_nvme(orom
) ? "platform" : "controller");
2367 static void print_imsm_capability_export(const struct imsm_orom
*orom
)
2369 printf("MD_FIRMWARE_TYPE=imsm\n");
2370 if (orom
->major_ver
|| orom
->minor_ver
|| orom
->hotfix_ver
|| orom
->build
)
2371 printf("IMSM_VERSION=%d.%d.%d.%d\n", orom
->major_ver
, orom
->minor_ver
,
2372 orom
->hotfix_ver
, orom
->build
);
2373 printf("IMSM_SUPPORTED_RAID_LEVELS=%s%s%s%s%s\n",
2374 imsm_orom_has_raid0(orom
) ? "raid0 " : "",
2375 imsm_orom_has_raid1(orom
) ? "raid1 " : "",
2376 imsm_orom_has_raid1e(orom
) ? "raid1e " : "",
2377 imsm_orom_has_raid5(orom
) ? "raid10 " : "",
2378 imsm_orom_has_raid10(orom
) ? "raid5 " : "");
2379 printf("IMSM_SUPPORTED_CHUNK_SIZES=%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2380 imsm_orom_has_chunk(orom
, 2) ? "2k " : "",
2381 imsm_orom_has_chunk(orom
, 4) ? "4k " : "",
2382 imsm_orom_has_chunk(orom
, 8) ? "8k " : "",
2383 imsm_orom_has_chunk(orom
, 16) ? "16k " : "",
2384 imsm_orom_has_chunk(orom
, 32) ? "32k " : "",
2385 imsm_orom_has_chunk(orom
, 64) ? "64k " : "",
2386 imsm_orom_has_chunk(orom
, 128) ? "128k " : "",
2387 imsm_orom_has_chunk(orom
, 256) ? "256k " : "",
2388 imsm_orom_has_chunk(orom
, 512) ? "512k " : "",
2389 imsm_orom_has_chunk(orom
, 1024*1) ? "1M " : "",
2390 imsm_orom_has_chunk(orom
, 1024*2) ? "2M " : "",
2391 imsm_orom_has_chunk(orom
, 1024*4) ? "4M " : "",
2392 imsm_orom_has_chunk(orom
, 1024*8) ? "8M " : "",
2393 imsm_orom_has_chunk(orom
, 1024*16) ? "16M " : "",
2394 imsm_orom_has_chunk(orom
, 1024*32) ? "32M " : "",
2395 imsm_orom_has_chunk(orom
, 1024*64) ? "64M " : "");
2396 printf("IMSM_2TB_VOLUMES=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB
) ? "yes" : "no");
2397 printf("IMSM_2TB_DISKS=%s\n",(orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) ? "yes" : "no");
2398 printf("IMSM_MAX_DISKS=%d\n",orom
->tds
);
2399 printf("IMSM_MAX_VOLUMES_PER_ARRAY=%d\n",orom
->vpa
);
2400 printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom
->vphba
);
2403 static int detail_platform_imsm(int verbose
, int enumerate_only
, char *controller_path
)
2405 /* There are two components to imsm platform support, the ahci SATA
2406 * controller and the option-rom. To find the SATA controller we
2407 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
2408 * controller with the Intel vendor id is present. This approach
2409 * allows mdadm to leverage the kernel's ahci detection logic, with the
2410 * caveat that if ahci.ko is not loaded mdadm will not be able to
2411 * detect platform raid capabilities. The option-rom resides in a
2412 * platform "Adapter ROM". We scan for its signature to retrieve the
2413 * platform capabilities. If raid support is disabled in the BIOS the
2414 * option-rom capability structure will not be available.
2416 struct sys_dev
*list
, *hba
;
2421 if (enumerate_only
) {
2422 if (check_env("IMSM_NO_PLATFORM"))
2424 list
= find_intel_devices();
2427 for (hba
= list
; hba
; hba
= hba
->next
) {
2428 if (find_imsm_capability(hba
)) {
2438 list
= find_intel_devices();
2441 pr_err("no active Intel(R) RAID controller found.\n");
2443 } else if (verbose
> 0)
2444 print_found_intel_controllers(list
);
2446 for (hba
= list
; hba
; hba
= hba
->next
) {
2447 if (controller_path
&& (compare_paths(hba
->path
, controller_path
) != 0))
2449 if (!find_imsm_capability(hba
)) {
2451 pr_err("imsm capabilities not found for controller: %s (type %s)\n",
2452 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
,
2453 get_sys_dev_type(hba
->type
));
2459 if (controller_path
&& result
== 1) {
2460 pr_err("no active Intel(R) RAID controller found under %s\n",
2465 const struct orom_entry
*entry
;
2467 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2468 if (entry
->type
== SYS_DEV_VMD
) {
2469 print_imsm_capability(&entry
->orom
);
2470 printf(" 3rd party NVMe :%s supported\n",
2471 imsm_orom_has_tpv_support(&entry
->orom
)?"":" not");
2472 for (hba
= list
; hba
; hba
= hba
->next
) {
2473 if (hba
->type
== SYS_DEV_VMD
) {
2475 printf(" I/O Controller : %s (%s)\n",
2476 vmd_domain_to_controller(hba
, buf
), get_sys_dev_type(hba
->type
));
2477 if (print_vmd_attached_devs(hba
)) {
2479 pr_err("failed to get devices attached to VMD domain.\n");
2488 print_imsm_capability(&entry
->orom
);
2489 if (entry
->type
== SYS_DEV_NVME
) {
2490 for (hba
= list
; hba
; hba
= hba
->next
) {
2491 if (hba
->type
== SYS_DEV_NVME
)
2492 printf(" NVMe Device : %s\n", hba
->path
);
2498 struct devid_list
*devid
;
2499 for (devid
= entry
->devid_list
; devid
; devid
= devid
->next
) {
2500 hba
= device_by_id(devid
->devid
);
2504 printf(" I/O Controller : %s (%s)\n",
2505 hba
->path
, get_sys_dev_type(hba
->type
));
2506 if (hba
->type
== SYS_DEV_SATA
) {
2507 host_base
= ahci_get_port_count(hba
->path
, &port_count
);
2508 if (ahci_enumerate_ports(hba
->path
, port_count
, host_base
, verbose
)) {
2510 pr_err("failed to enumerate ports on SATA controller at %s.\n", hba
->pci_id
);
2521 static int export_detail_platform_imsm(int verbose
, char *controller_path
)
2523 struct sys_dev
*list
, *hba
;
2526 list
= find_intel_devices();
2529 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_INTEL_DEVICES\n");
2534 for (hba
= list
; hba
; hba
= hba
->next
) {
2535 if (controller_path
&& (compare_paths(hba
->path
,controller_path
) != 0))
2537 if (!find_imsm_capability(hba
) && verbose
> 0) {
2539 pr_err("IMSM_DETAIL_PLATFORM_ERROR=NO_IMSM_CAPABLE_DEVICE_UNDER_%s\n",
2540 hba
->type
== SYS_DEV_VMD
? vmd_domain_to_controller(hba
, buf
) : hba
->path
);
2546 const struct orom_entry
*entry
;
2548 for (entry
= orom_entries
; entry
; entry
= entry
->next
) {
2549 if (entry
->type
== SYS_DEV_VMD
) {
2550 for (hba
= list
; hba
; hba
= hba
->next
)
2551 print_imsm_capability_export(&entry
->orom
);
2554 print_imsm_capability_export(&entry
->orom
);
2562 static int match_home_imsm(struct supertype
*st
, char *homehost
)
2564 /* the imsm metadata format does not specify any host
2565 * identification information. We return -1 since we can never
2566 * confirm nor deny whether a given array is "meant" for this
2567 * host. We rely on compare_super and the 'family_num' fields to
2568 * exclude member disks that do not belong, and we rely on
2569 * mdadm.conf to specify the arrays that should be assembled.
2570 * Auto-assembly may still pick up "foreign" arrays.
2576 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
2578 /* The uuid returned here is used for:
2579 * uuid to put into bitmap file (Create, Grow)
2580 * uuid for backup header when saving critical section (Grow)
2581 * comparing uuids when re-adding a device into an array
2582 * In these cases the uuid required is that of the data-array,
2583 * not the device-set.
2584 * uuid to recognise same set when adding a missing device back
2585 * to an array. This is a uuid for the device-set.
2587 * For each of these we can make do with a truncated
2588 * or hashed uuid rather than the original, as long as
2590 * In each case the uuid required is that of the data-array,
2591 * not the device-set.
2593 /* imsm does not track uuid's so we synthesis one using sha1 on
2594 * - The signature (Which is constant for all imsm array, but no matter)
2595 * - the orig_family_num of the container
2596 * - the index number of the volume
2597 * - the 'serial' number of the volume.
2598 * Hopefully these are all constant.
2600 struct intel_super
*super
= st
->sb
;
2603 struct sha1_ctx ctx
;
2604 struct imsm_dev
*dev
= NULL
;
2607 /* some mdadm versions failed to set ->orig_family_num, in which
2608 * case fall back to ->family_num. orig_family_num will be
2609 * fixed up with the first metadata update.
2611 family_num
= super
->anchor
->orig_family_num
;
2612 if (family_num
== 0)
2613 family_num
= super
->anchor
->family_num
;
2614 sha1_init_ctx(&ctx
);
2615 sha1_process_bytes(super
->anchor
->sig
, MPB_SIG_LEN
, &ctx
);
2616 sha1_process_bytes(&family_num
, sizeof(__u32
), &ctx
);
2617 if (super
->current_vol
>= 0)
2618 dev
= get_imsm_dev(super
, super
->current_vol
);
2620 __u32 vol
= super
->current_vol
;
2621 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
2622 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
2624 sha1_finish_ctx(&ctx
, buf
);
2625 memcpy(uuid
, buf
, 4*4);
2630 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
2632 __u8
*v
= get_imsm_version(mpb
);
2633 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
2634 char major
[] = { 0, 0, 0 };
2635 char minor
[] = { 0 ,0, 0 };
2636 char patch
[] = { 0, 0, 0 };
2637 char *ver_parse
[] = { major
, minor
, patch
};
2641 while (*v
!= '\0' && v
< end
) {
2642 if (*v
!= '.' && j
< 2)
2643 ver_parse
[i
][j
++] = *v
;
2651 *m
= strtol(minor
, NULL
, 0);
2652 *p
= strtol(patch
, NULL
, 0);
2656 static __u32
migr_strip_blocks_resync(struct imsm_dev
*dev
)
2658 /* migr_strip_size when repairing or initializing parity */
2659 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2660 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2662 switch (get_imsm_raid_level(map
)) {
2667 return 128*1024 >> 9;
2671 static __u32
migr_strip_blocks_rebuild(struct imsm_dev
*dev
)
2673 /* migr_strip_size when rebuilding a degraded disk, no idea why
2674 * this is different than migr_strip_size_resync(), but it's good
2677 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2678 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2680 switch (get_imsm_raid_level(map
)) {
2683 if (map
->num_members
% map
->num_domains
== 0)
2684 return 128*1024 >> 9;
2688 return max((__u32
) 64*1024 >> 9, chunk
);
2690 return 128*1024 >> 9;
2694 static __u32
num_stripes_per_unit_resync(struct imsm_dev
*dev
)
2696 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2697 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2698 __u32 lo_chunk
= __le32_to_cpu(lo
->blocks_per_strip
);
2699 __u32 hi_chunk
= __le32_to_cpu(hi
->blocks_per_strip
);
2701 return max((__u32
) 1, hi_chunk
/ lo_chunk
);
2704 static __u32
num_stripes_per_unit_rebuild(struct imsm_dev
*dev
)
2706 struct imsm_map
*lo
= get_imsm_map(dev
, MAP_0
);
2707 int level
= get_imsm_raid_level(lo
);
2709 if (level
== 1 || level
== 10) {
2710 struct imsm_map
*hi
= get_imsm_map(dev
, MAP_1
);
2712 return hi
->num_domains
;
2714 return num_stripes_per_unit_resync(dev
);
2717 static __u8
imsm_num_data_members(struct imsm_dev
*dev
, int second_map
)
2719 /* named 'imsm_' because raid0, raid1 and raid10
2720 * counter-intuitively have the same number of data disks
2722 struct imsm_map
*map
= get_imsm_map(dev
, second_map
);
2724 switch (get_imsm_raid_level(map
)) {
2726 return map
->num_members
;
2730 return map
->num_members
/2;
2732 return map
->num_members
- 1;
2734 dprintf("unsupported raid level\n");
2739 static __u32
parity_segment_depth(struct imsm_dev
*dev
)
2741 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2742 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2744 switch(get_imsm_raid_level(map
)) {
2747 return chunk
* map
->num_domains
;
2749 return chunk
* map
->num_members
;
2755 static __u32
map_migr_block(struct imsm_dev
*dev
, __u32 block
)
2757 struct imsm_map
*map
= get_imsm_map(dev
, MAP_1
);
2758 __u32 chunk
= __le32_to_cpu(map
->blocks_per_strip
);
2759 __u32 strip
= block
/ chunk
;
2761 switch (get_imsm_raid_level(map
)) {
2764 __u32 vol_strip
= (strip
* map
->num_domains
) + 1;
2765 __u32 vol_stripe
= vol_strip
/ map
->num_members
;
2767 return vol_stripe
* chunk
+ block
% chunk
;
2769 __u32 stripe
= strip
/ (map
->num_members
- 1);
2771 return stripe
* chunk
+ block
% chunk
;
2778 static __u64
blocks_per_migr_unit(struct intel_super
*super
,
2779 struct imsm_dev
*dev
)
2781 /* calculate the conversion factor between per member 'blocks'
2782 * (md/{resync,rebuild}_start) and imsm migration units, return
2783 * 0 for the 'not migrating' and 'unsupported migration' cases
2785 if (!dev
->vol
.migr_state
)
2788 switch (migr_type(dev
)) {
2789 case MIGR_GEN_MIGR
: {
2790 struct migr_record
*migr_rec
= super
->migr_rec
;
2791 return __le32_to_cpu(migr_rec
->blocks_per_unit
);
2796 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
2797 __u32 stripes_per_unit
;
2798 __u32 blocks_per_unit
;
2807 /* yes, this is really the translation of migr_units to
2808 * per-member blocks in the 'resync' case
2810 stripes_per_unit
= num_stripes_per_unit_resync(dev
);
2811 migr_chunk
= migr_strip_blocks_resync(dev
);
2812 disks
= imsm_num_data_members(dev
, MAP_0
);
2813 blocks_per_unit
= stripes_per_unit
* migr_chunk
* disks
;
2814 stripe
= __le16_to_cpu(map
->blocks_per_strip
) * disks
;
2815 segment
= blocks_per_unit
/ stripe
;
2816 block_rel
= blocks_per_unit
- segment
* stripe
;
2817 parity_depth
= parity_segment_depth(dev
);
2818 block_map
= map_migr_block(dev
, block_rel
);
2819 return block_map
+ parity_depth
* segment
;
2821 case MIGR_REBUILD
: {
2822 __u32 stripes_per_unit
;
2825 stripes_per_unit
= num_stripes_per_unit_rebuild(dev
);
2826 migr_chunk
= migr_strip_blocks_rebuild(dev
);
2827 return migr_chunk
* stripes_per_unit
;
2829 case MIGR_STATE_CHANGE
:
2835 static int imsm_level_to_layout(int level
)
2843 return ALGORITHM_LEFT_ASYMMETRIC
;
2850 /*******************************************************************************
2851 * Function: read_imsm_migr_rec
2852 * Description: Function reads imsm migration record from last sector of disk
2854 * fd : disk descriptor
2855 * super : metadata info
2859 ******************************************************************************/
2860 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
)
2863 unsigned int sector_size
= super
->sector_size
;
2864 unsigned long long dsize
;
2866 get_dev_size(fd
, NULL
, &dsize
);
2867 if (lseek64(fd
, dsize
- (sector_size
*MIGR_REC_SECTOR_POSITION
),
2869 pr_err("Cannot seek to anchor block: %s\n",
2873 if ((unsigned int)read(fd
, super
->migr_rec_buf
,
2874 MIGR_REC_BUF_SECTORS
*sector_size
) !=
2875 MIGR_REC_BUF_SECTORS
*sector_size
) {
2876 pr_err("Cannot read migr record block: %s\n",
2881 if (sector_size
== 4096)
2882 convert_from_4k_imsm_migr_rec(super
);
2888 static struct imsm_dev
*imsm_get_device_during_migration(
2889 struct intel_super
*super
)
2892 struct intel_dev
*dv
;
2894 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
2895 if (is_gen_migration(dv
->dev
))
2901 /*******************************************************************************
2902 * Function: load_imsm_migr_rec
2903 * Description: Function reads imsm migration record (it is stored at the last
2906 * super : imsm internal array info
2907 * info : general array info
2911 * -2 : no migration in progress
2912 ******************************************************************************/
2913 static int load_imsm_migr_rec(struct intel_super
*super
, struct mdinfo
*info
)
2920 struct imsm_dev
*dev
;
2921 struct imsm_map
*map
;
2924 /* find map under migration */
2925 dev
= imsm_get_device_during_migration(super
);
2926 /* nothing to load,no migration in progress?
2932 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
2933 /* read only from one of the first two slots */
2934 if ((sd
->disk
.raid_disk
< 0) ||
2935 (sd
->disk
.raid_disk
> 1))
2938 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2939 fd
= dev_open(nm
, O_RDONLY
);
2945 map
= get_imsm_map(dev
, MAP_0
);
2946 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2947 /* skip spare and failed disks
2951 /* read only from one of the first two slots */
2953 slot
= get_imsm_disk_slot(map
, dl
->index
);
2954 if (map
== NULL
|| slot
> 1 || slot
< 0)
2956 sprintf(nm
, "%d:%d", dl
->major
, dl
->minor
);
2957 fd
= dev_open(nm
, O_RDONLY
);
2964 retval
= read_imsm_migr_rec(fd
, super
);
2973 /*******************************************************************************
2974 * function: imsm_create_metadata_checkpoint_update
2975 * Description: It creates update for checkpoint change.
2977 * super : imsm internal array info
2978 * u : pointer to prepared update
2981 * If length is equal to 0, input pointer u contains no update
2982 ******************************************************************************/
2983 static int imsm_create_metadata_checkpoint_update(
2984 struct intel_super
*super
,
2985 struct imsm_update_general_migration_checkpoint
**u
)
2988 int update_memory_size
= 0;
2990 dprintf("(enter)\n");
2996 /* size of all update data without anchor */
2997 update_memory_size
=
2998 sizeof(struct imsm_update_general_migration_checkpoint
);
3000 *u
= xcalloc(1, update_memory_size
);
3002 dprintf("error: cannot get memory\n");
3005 (*u
)->type
= update_general_migration_checkpoint
;
3006 (*u
)->curr_migr_unit
= __le32_to_cpu(super
->migr_rec
->curr_migr_unit
);
3007 dprintf("prepared for %u\n", (*u
)->curr_migr_unit
);
3009 return update_memory_size
;
3012 static void imsm_update_metadata_locally(struct supertype
*st
,
3013 void *buf
, int len
);
3015 /*******************************************************************************
3016 * Function: write_imsm_migr_rec
3017 * Description: Function writes imsm migration record
3018 * (at the last sector of disk)
3020 * super : imsm internal array info
3024 ******************************************************************************/
3025 static int write_imsm_migr_rec(struct supertype
*st
)
3027 struct intel_super
*super
= st
->sb
;
3028 unsigned int sector_size
= super
->sector_size
;
3029 unsigned long long dsize
;
3035 struct imsm_update_general_migration_checkpoint
*u
;
3036 struct imsm_dev
*dev
;
3037 struct imsm_map
*map
;
3039 /* find map under migration */
3040 dev
= imsm_get_device_during_migration(super
);
3041 /* if no migration, write buffer anyway to clear migr_record
3042 * on disk based on first available device
3045 dev
= get_imsm_dev(super
, super
->current_vol
< 0 ? 0 :
3046 super
->current_vol
);
3048 map
= get_imsm_map(dev
, MAP_0
);
3050 if (sector_size
== 4096)
3051 convert_to_4k_imsm_migr_rec(super
);
3052 for (sd
= super
->disks
; sd
; sd
= sd
->next
) {
3055 /* skip failed and spare devices */
3058 /* write to 2 first slots only */
3060 slot
= get_imsm_disk_slot(map
, sd
->index
);
3061 if (map
== NULL
|| slot
> 1 || slot
< 0)
3064 sprintf(nm
, "%d:%d", sd
->major
, sd
->minor
);
3065 fd
= dev_open(nm
, O_RDWR
);
3068 get_dev_size(fd
, NULL
, &dsize
);
3069 if (lseek64(fd
, dsize
- (MIGR_REC_SECTOR_POSITION
*sector_size
),
3071 pr_err("Cannot seek to anchor block: %s\n",
3075 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
3076 MIGR_REC_BUF_SECTORS
*sector_size
) !=
3077 MIGR_REC_BUF_SECTORS
*sector_size
) {
3078 pr_err("Cannot write migr record block: %s\n",
3085 if (sector_size
== 4096)
3086 convert_from_4k_imsm_migr_rec(super
);
3087 /* update checkpoint information in metadata */
3088 len
= imsm_create_metadata_checkpoint_update(super
, &u
);
3090 dprintf("imsm: Cannot prepare update\n");
3093 /* update metadata locally */
3094 imsm_update_metadata_locally(st
, u
, len
);
3095 /* and possibly remotely */
3096 if (st
->update_tail
) {
3097 append_metadata_update(st
, u
, len
);
3098 /* during reshape we do all work inside metadata handler
3099 * manage_reshape(), so metadata update has to be triggered
3102 flush_metadata_updates(st
);
3103 st
->update_tail
= &st
->updates
;
3113 #endif /* MDASSEMBLE */
3115 /* spare/missing disks activations are not allowe when
3116 * array/container performs reshape operation, because
3117 * all arrays in container works on the same disks set
3119 int imsm_reshape_blocks_arrays_changes(struct intel_super
*super
)
3122 struct intel_dev
*i_dev
;
3123 struct imsm_dev
*dev
;
3125 /* check whole container
3127 for (i_dev
= super
->devlist
; i_dev
; i_dev
= i_dev
->next
) {
3129 if (is_gen_migration(dev
)) {
3130 /* No repair during any migration in container
3138 static unsigned long long imsm_component_size_aligment_check(int level
,
3140 unsigned int sector_size
,
3141 unsigned long long component_size
)
3143 unsigned int component_size_alligment
;
3145 /* check component size aligment
3147 component_size_alligment
= component_size
% (chunk_size
/sector_size
);
3149 dprintf("(Level: %i, chunk_size = %i, component_size = %llu), component_size_alligment = %u\n",
3150 level
, chunk_size
, component_size
,
3151 component_size_alligment
);
3153 if (component_size_alligment
&& (level
!= 1) && (level
!= UnSet
)) {
3154 dprintf("imsm: reported component size alligned from %llu ",
3156 component_size
-= component_size_alligment
;
3157 dprintf_cont("to %llu (%i).\n",
3158 component_size
, component_size_alligment
);
3161 return component_size
;
3164 static unsigned long long get_ppl_sector(struct intel_super
*super
, int dev_idx
)
3166 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
3167 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3169 return pba_of_lba0(map
) +
3170 (num_data_stripes(map
) * map
->blocks_per_strip
);
3173 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
, char *dmap
)
3175 struct intel_super
*super
= st
->sb
;
3176 struct migr_record
*migr_rec
= super
->migr_rec
;
3177 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
3178 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
3179 struct imsm_map
*prev_map
= get_imsm_map(dev
, MAP_1
);
3180 struct imsm_map
*map_to_analyse
= map
;
3182 int map_disks
= info
->array
.raid_disks
;
3184 memset(info
, 0, sizeof(*info
));
3186 map_to_analyse
= prev_map
;
3188 dl
= super
->current_disk
;
3190 info
->container_member
= super
->current_vol
;
3191 info
->array
.raid_disks
= map
->num_members
;
3192 info
->array
.level
= get_imsm_raid_level(map_to_analyse
);
3193 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
3194 info
->array
.md_minor
= -1;
3195 info
->array
.ctime
= 0;
3196 info
->array
.utime
= 0;
3197 info
->array
.chunk_size
=
3198 __le16_to_cpu(map_to_analyse
->blocks_per_strip
) << 9;
3199 info
->array
.state
= !(dev
->vol
.dirty
& RAIDVOL_DIRTY
);
3200 info
->custom_array_size
= __le32_to_cpu(dev
->size_high
);
3201 info
->custom_array_size
<<= 32;
3202 info
->custom_array_size
|= __le32_to_cpu(dev
->size_low
);
3203 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3205 if (is_gen_migration(dev
)) {
3206 info
->reshape_active
= 1;
3207 info
->new_level
= get_imsm_raid_level(map
);
3208 info
->new_layout
= imsm_level_to_layout(info
->new_level
);
3209 info
->new_chunk
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
3210 info
->delta_disks
= map
->num_members
- prev_map
->num_members
;
3211 if (info
->delta_disks
) {
3212 /* this needs to be applied to every array
3215 info
->reshape_active
= CONTAINER_RESHAPE
;
3217 /* We shape information that we give to md might have to be
3218 * modify to cope with md's requirement for reshaping arrays.
3219 * For example, when reshaping a RAID0, md requires it to be
3220 * presented as a degraded RAID4.
3221 * Also if a RAID0 is migrating to a RAID5 we need to specify
3222 * the array as already being RAID5, but the 'before' layout
3223 * is a RAID4-like layout.
3225 switch (info
->array
.level
) {
3227 switch(info
->new_level
) {
3229 /* conversion is happening as RAID4 */
3230 info
->array
.level
= 4;
3231 info
->array
.raid_disks
+= 1;
3234 /* conversion is happening as RAID5 */
3235 info
->array
.level
= 5;
3236 info
->array
.layout
= ALGORITHM_PARITY_N
;
3237 info
->delta_disks
-= 1;
3240 /* FIXME error message */
3241 info
->array
.level
= UnSet
;
3247 info
->new_level
= UnSet
;
3248 info
->new_layout
= UnSet
;
3249 info
->new_chunk
= info
->array
.chunk_size
;
3250 info
->delta_disks
= 0;
3254 info
->disk
.major
= dl
->major
;
3255 info
->disk
.minor
= dl
->minor
;
3256 info
->disk
.number
= dl
->index
;
3257 info
->disk
.raid_disk
= get_imsm_disk_slot(map_to_analyse
,
3261 info
->data_offset
= pba_of_lba0(map_to_analyse
);
3263 if (info
->array
.level
== 5) {
3264 info
->component_size
= num_data_stripes(map_to_analyse
) *
3265 map_to_analyse
->blocks_per_strip
;
3267 info
->component_size
= blocks_per_member(map_to_analyse
);
3270 info
->component_size
= imsm_component_size_aligment_check(
3272 info
->array
.chunk_size
,
3274 info
->component_size
);
3275 info
->bb
.supported
= 1;
3277 memset(info
->uuid
, 0, sizeof(info
->uuid
));
3278 info
->recovery_start
= MaxSector
;
3280 if (info
->array
.level
== 5 && dev
->rwh_policy
== RWH_DISTRIBUTED
) {
3281 info
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
3282 info
->ppl_sector
= get_ppl_sector(super
, super
->current_vol
);
3283 info
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
) >> 9;
3284 } else if (info
->array
.level
<= 0) {
3285 info
->consistency_policy
= CONSISTENCY_POLICY_NONE
;
3287 info
->consistency_policy
= CONSISTENCY_POLICY_RESYNC
;
3290 info
->reshape_progress
= 0;
3291 info
->resync_start
= MaxSector
;
3292 if ((map_to_analyse
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
3293 !(info
->array
.state
& 1)) &&
3294 imsm_reshape_blocks_arrays_changes(super
) == 0) {
3295 info
->resync_start
= 0;
3297 if (dev
->vol
.migr_state
) {
3298 switch (migr_type(dev
)) {
3301 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3303 __u64 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
3305 info
->resync_start
= blocks_per_unit
* units
;
3308 case MIGR_GEN_MIGR
: {
3309 __u64 blocks_per_unit
= blocks_per_migr_unit(super
,
3311 __u64 units
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
3312 unsigned long long array_blocks
;
3315 if (__le32_to_cpu(migr_rec
->ascending_migr
) &&
3317 (__le32_to_cpu(migr_rec
->num_migr_units
)-1)) &&
3318 (super
->migr_rec
->rec_status
==
3319 __cpu_to_le32(UNIT_SRC_IN_CP_AREA
)))
3322 info
->reshape_progress
= blocks_per_unit
* units
;
3324 dprintf("IMSM: General Migration checkpoint : %llu (%llu) -> read reshape progress : %llu\n",
3325 (unsigned long long)units
,
3326 (unsigned long long)blocks_per_unit
,
3327 info
->reshape_progress
);
3329 used_disks
= imsm_num_data_members(dev
, MAP_1
);
3330 if (used_disks
> 0) {
3331 array_blocks
= blocks_per_member(map
) *
3333 /* round array size down to closest MB
3335 info
->custom_array_size
= (array_blocks
3336 >> SECT_PER_MB_SHIFT
)
3337 << SECT_PER_MB_SHIFT
;
3341 /* we could emulate the checkpointing of
3342 * 'sync_action=check' migrations, but for now
3343 * we just immediately complete them
3346 /* this is handled by container_content_imsm() */
3347 case MIGR_STATE_CHANGE
:
3348 /* FIXME handle other migrations */
3350 /* we are not dirty, so... */
3351 info
->resync_start
= MaxSector
;
3355 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
3356 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
3358 info
->array
.major_version
= -1;
3359 info
->array
.minor_version
= -2;
3360 sprintf(info
->text_version
, "/%s/%d", st
->container_devnm
, info
->container_member
);
3361 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
3362 uuid_from_super_imsm(st
, info
->uuid
);
3366 for (i
=0; i
<map_disks
; i
++) {
3368 if (i
< info
->array
.raid_disks
) {
3369 struct imsm_disk
*dsk
;
3370 j
= get_imsm_disk_idx(dev
, i
, MAP_X
);
3371 dsk
= get_imsm_disk(super
, j
);
3372 if (dsk
&& (dsk
->status
& CONFIGURED_DISK
))
3379 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
3380 int failed
, int look_in_map
);
3382 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
3386 static void manage_second_map(struct intel_super
*super
, struct imsm_dev
*dev
)
3388 if (is_gen_migration(dev
)) {
3391 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
3393 failed
= imsm_count_failed(super
, dev
, MAP_1
);
3394 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_1
);
3395 if (map2
->map_state
!= map_state
) {
3396 map2
->map_state
= map_state
;
3397 super
->updates_pending
++;
3403 static struct imsm_disk
*get_imsm_missing(struct intel_super
*super
, __u8 index
)
3407 for (d
= super
->missing
; d
; d
= d
->next
)
3408 if (d
->index
== index
)
3413 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
, char *map
)
3415 struct intel_super
*super
= st
->sb
;
3416 struct imsm_disk
*disk
;
3417 int map_disks
= info
->array
.raid_disks
;
3418 int max_enough
= -1;
3420 struct imsm_super
*mpb
;
3422 if (super
->current_vol
>= 0) {
3423 getinfo_super_imsm_volume(st
, info
, map
);
3426 memset(info
, 0, sizeof(*info
));
3428 /* Set raid_disks to zero so that Assemble will always pull in valid
3431 info
->array
.raid_disks
= 0;
3432 info
->array
.level
= LEVEL_CONTAINER
;
3433 info
->array
.layout
= 0;
3434 info
->array
.md_minor
= -1;
3435 info
->array
.ctime
= 0; /* N/A for imsm */
3436 info
->array
.utime
= 0;
3437 info
->array
.chunk_size
= 0;
3439 info
->disk
.major
= 0;
3440 info
->disk
.minor
= 0;
3441 info
->disk
.raid_disk
= -1;
3442 info
->reshape_active
= 0;
3443 info
->array
.major_version
= -1;
3444 info
->array
.minor_version
= -2;
3445 strcpy(info
->text_version
, "imsm");
3446 info
->safe_mode_delay
= 0;
3447 info
->disk
.number
= -1;
3448 info
->disk
.state
= 0;
3450 info
->recovery_start
= MaxSector
;
3451 info
->recovery_blocked
= imsm_reshape_blocks_arrays_changes(st
->sb
);
3452 info
->bb
.supported
= 1;
3454 /* do we have the all the insync disks that we expect? */
3455 mpb
= super
->anchor
;
3456 info
->events
= __le32_to_cpu(mpb
->generation_num
);
3458 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3459 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
3460 int failed
, enough
, j
, missing
= 0;
3461 struct imsm_map
*map
;
3464 failed
= imsm_count_failed(super
, dev
, MAP_0
);
3465 state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
3466 map
= get_imsm_map(dev
, MAP_0
);
3468 /* any newly missing disks?
3469 * (catches single-degraded vs double-degraded)
3471 for (j
= 0; j
< map
->num_members
; j
++) {
3472 __u32 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_0
);
3473 __u32 idx
= ord_to_idx(ord
);
3475 if (!(ord
& IMSM_ORD_REBUILD
) &&
3476 get_imsm_missing(super
, idx
)) {
3482 if (state
== IMSM_T_STATE_FAILED
)
3484 else if (state
== IMSM_T_STATE_DEGRADED
&&
3485 (state
!= map
->map_state
|| missing
))
3487 else /* we're normal, or already degraded */
3489 if (is_gen_migration(dev
) && missing
) {
3490 /* during general migration we need all disks
3491 * that process is running on.
3492 * No new missing disk is allowed.
3496 /* no more checks necessary
3500 /* in the missing/failed disk case check to see
3501 * if at least one array is runnable
3503 max_enough
= max(max_enough
, enough
);
3505 dprintf("enough: %d\n", max_enough
);
3506 info
->container_enough
= max_enough
;
3509 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
3511 disk
= &super
->disks
->disk
;
3512 info
->data_offset
= total_blocks(&super
->disks
->disk
) - reserved
;
3513 info
->component_size
= reserved
;
3514 info
->disk
.state
= is_configured(disk
) ? (1 << MD_DISK_ACTIVE
) : 0;
3515 /* we don't change info->disk.raid_disk here because
3516 * this state will be finalized in mdmon after we have
3517 * found the 'most fresh' version of the metadata
3519 info
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3520 info
->disk
.state
|= (is_spare(disk
) || is_journal(disk
)) ?
3521 0 : (1 << MD_DISK_SYNC
);
3524 /* only call uuid_from_super_imsm when this disk is part of a populated container,
3525 * ->compare_super may have updated the 'num_raid_devs' field for spares
3527 if (info
->disk
.state
& (1 << MD_DISK_SYNC
) || super
->anchor
->num_raid_devs
)
3528 uuid_from_super_imsm(st
, info
->uuid
);
3530 memcpy(info
->uuid
, uuid_zero
, sizeof(uuid_zero
));
3532 /* I don't know how to compute 'map' on imsm, so use safe default */
3535 for (i
= 0; i
< map_disks
; i
++)
3541 /* allocates memory and fills disk in mdinfo structure
3542 * for each disk in array */
3543 struct mdinfo
*getinfo_super_disks_imsm(struct supertype
*st
)
3545 struct mdinfo
*mddev
;
3546 struct intel_super
*super
= st
->sb
;
3547 struct imsm_disk
*disk
;
3550 if (!super
|| !super
->disks
)
3553 mddev
= xcalloc(1, sizeof(*mddev
));
3557 tmp
= xcalloc(1, sizeof(*tmp
));
3559 tmp
->next
= mddev
->devs
;
3561 tmp
->disk
.number
= count
++;
3562 tmp
->disk
.major
= dl
->major
;
3563 tmp
->disk
.minor
= dl
->minor
;
3564 tmp
->disk
.state
= is_configured(disk
) ?
3565 (1 << MD_DISK_ACTIVE
) : 0;
3566 tmp
->disk
.state
|= is_failed(disk
) ? (1 << MD_DISK_FAULTY
) : 0;
3567 tmp
->disk
.state
|= is_spare(disk
) ? 0 : (1 << MD_DISK_SYNC
);
3568 tmp
->disk
.raid_disk
= -1;
3574 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
3575 char *update
, char *devname
, int verbose
,
3576 int uuid_set
, char *homehost
)
3578 /* For 'assemble' and 'force' we need to return non-zero if any
3579 * change was made. For others, the return value is ignored.
3580 * Update options are:
3581 * force-one : This device looks a bit old but needs to be included,
3582 * update age info appropriately.
3583 * assemble: clear any 'faulty' flag to allow this device to
3585 * force-array: Array is degraded but being forced, mark it clean
3586 * if that will be needed to assemble it.
3588 * newdev: not used ????
3589 * grow: Array has gained a new device - this is currently for
3591 * resync: mark as dirty so a resync will happen.
3592 * name: update the name - preserving the homehost
3593 * uuid: Change the uuid of the array to match watch is given
3595 * Following are not relevant for this imsm:
3596 * sparc2.2 : update from old dodgey metadata
3597 * super-minor: change the preferred_minor number
3598 * summaries: update redundant counters.
3599 * homehost: update the recorded homehost
3600 * _reshape_progress: record new reshape_progress position.
3603 struct intel_super
*super
= st
->sb
;
3604 struct imsm_super
*mpb
;
3606 /* we can only update container info */
3607 if (!super
|| super
->current_vol
>= 0 || !super
->anchor
)
3610 mpb
= super
->anchor
;
3612 if (strcmp(update
, "uuid") == 0) {
3613 /* We take this to mean that the family_num should be updated.
3614 * However that is much smaller than the uuid so we cannot really
3615 * allow an explicit uuid to be given. And it is hard to reliably
3617 * So if !uuid_set we know the current uuid is random and just used
3618 * the first 'int' and copy it to the other 3 positions.
3619 * Otherwise we require the 4 'int's to be the same as would be the
3620 * case if we are using a random uuid. So an explicit uuid will be
3621 * accepted as long as all for ints are the same... which shouldn't hurt
3624 info
->uuid
[1] = info
->uuid
[2] = info
->uuid
[3] = info
->uuid
[0];
3627 if (info
->uuid
[0] != info
->uuid
[1] ||
3628 info
->uuid
[1] != info
->uuid
[2] ||
3629 info
->uuid
[2] != info
->uuid
[3])
3635 mpb
->orig_family_num
= info
->uuid
[0];
3636 } else if (strcmp(update
, "assemble") == 0)
3641 /* successful update? recompute checksum */
3643 mpb
->check_sum
= __le32_to_cpu(__gen_imsm_checksum(mpb
));
3648 static size_t disks_to_mpb_size(int disks
)
3652 size
= sizeof(struct imsm_super
);
3653 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
3654 size
+= 2 * sizeof(struct imsm_dev
);
3655 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
3656 size
+= (4 - 2) * sizeof(struct imsm_map
);
3657 /* 4 possible disk_ord_tbl's */
3658 size
+= 4 * (disks
- 1) * sizeof(__u32
);
3659 /* maximum bbm log */
3660 size
+= sizeof(struct bbm_log
);
3665 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
,
3666 unsigned long long data_offset
)
3668 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
3671 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
3674 static void free_devlist(struct intel_super
*super
)
3676 struct intel_dev
*dv
;
3678 while (super
->devlist
) {
3679 dv
= super
->devlist
->next
;
3680 free(super
->devlist
->dev
);
3681 free(super
->devlist
);
3682 super
->devlist
= dv
;
3686 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
3688 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
3691 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
3695 * 0 same, or first was empty, and second was copied
3696 * 1 second had wrong number
3698 * 3 wrong other info
3700 struct intel_super
*first
= st
->sb
;
3701 struct intel_super
*sec
= tst
->sb
;
3708 /* in platform dependent environment test if the disks
3709 * use the same Intel hba
3710 * If not on Intel hba at all, allow anything.
3712 if (!check_env("IMSM_NO_PLATFORM") && first
->hba
&& sec
->hba
) {
3713 if (first
->hba
->type
!= sec
->hba
->type
) {
3715 "HBAs of devices do not match %s != %s\n",
3716 get_sys_dev_type(first
->hba
->type
),
3717 get_sys_dev_type(sec
->hba
->type
));
3720 if (first
->orom
!= sec
->orom
) {
3722 "HBAs of devices do not match %s != %s\n",
3723 first
->hba
->pci_id
, sec
->hba
->pci_id
);
3728 /* if an anchor does not have num_raid_devs set then it is a free
3731 if (first
->anchor
->num_raid_devs
> 0 &&
3732 sec
->anchor
->num_raid_devs
> 0) {
3733 /* Determine if these disks might ever have been
3734 * related. Further disambiguation can only take place
3735 * in load_super_imsm_all
3737 __u32 first_family
= first
->anchor
->orig_family_num
;
3738 __u32 sec_family
= sec
->anchor
->orig_family_num
;
3740 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
,
3741 MAX_SIGNATURE_LENGTH
) != 0)
3744 if (first_family
== 0)
3745 first_family
= first
->anchor
->family_num
;
3746 if (sec_family
== 0)
3747 sec_family
= sec
->anchor
->family_num
;
3749 if (first_family
!= sec_family
)
3754 /* if 'first' is a spare promote it to a populated mpb with sec's
3757 if (first
->anchor
->num_raid_devs
== 0 &&
3758 sec
->anchor
->num_raid_devs
> 0) {
3760 struct intel_dev
*dv
;
3761 struct imsm_dev
*dev
;
3763 /* we need to copy raid device info from sec if an allocation
3764 * fails here we don't associate the spare
3766 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++) {
3767 dv
= xmalloc(sizeof(*dv
));
3768 dev
= xmalloc(sizeof_imsm_dev(get_imsm_dev(sec
, i
), 1));
3771 dv
->next
= first
->devlist
;
3772 first
->devlist
= dv
;
3774 if (i
< sec
->anchor
->num_raid_devs
) {
3775 /* allocation failure */
3776 free_devlist(first
);
3777 pr_err("imsm: failed to associate spare\n");
3780 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
3781 first
->anchor
->orig_family_num
= sec
->anchor
->orig_family_num
;
3782 first
->anchor
->family_num
= sec
->anchor
->family_num
;
3783 memcpy(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
);
3784 for (i
= 0; i
< sec
->anchor
->num_raid_devs
; i
++)
3785 imsm_copy_dev(get_imsm_dev(first
, i
), get_imsm_dev(sec
, i
));
3791 static void fd2devname(int fd
, char *name
)
3795 char dname
[PATH_MAX
];
3800 if (fstat(fd
, &st
) != 0)
3802 sprintf(path
, "/sys/dev/block/%d:%d",
3803 major(st
.st_rdev
), minor(st
.st_rdev
));
3805 rv
= readlink(path
, dname
, sizeof(dname
)-1);
3810 nm
= strrchr(dname
, '/');
3813 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
3817 static int nvme_get_serial(int fd
, void *buf
, size_t buf_len
)
3820 char *name
= fd2kname(fd
);
3825 if (strncmp(name
, "nvme", 4) != 0)
3828 snprintf(path
, sizeof(path
) - 1, "/sys/block/%s/device/serial", name
);
3830 return load_sys(path
, buf
, buf_len
);
3833 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
3835 static int imsm_read_serial(int fd
, char *devname
,
3836 __u8 serial
[MAX_RAID_SERIAL_LEN
])
3845 memset(buf
, 0, sizeof(buf
));
3847 rv
= nvme_get_serial(fd
, buf
, sizeof(buf
));
3850 rv
= scsi_get_serial(fd
, buf
, sizeof(buf
));
3852 if (rv
&& check_env("IMSM_DEVNAME_AS_SERIAL")) {
3853 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3854 fd2devname(fd
, (char *) serial
);
3860 pr_err("Failed to retrieve serial for %s\n",
3865 /* trim all whitespace and non-printable characters and convert
3868 for (i
= 0, dest
= buf
; i
< sizeof(buf
) && buf
[i
]; i
++) {
3871 /* ':' is reserved for use in placeholder serial
3872 * numbers for missing disks
3883 /* truncate leading characters */
3884 if (len
> MAX_RAID_SERIAL_LEN
) {
3885 dest
+= len
- MAX_RAID_SERIAL_LEN
;
3886 len
= MAX_RAID_SERIAL_LEN
;
3889 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
3890 memcpy(serial
, dest
, len
);
3895 static int serialcmp(__u8
*s1
, __u8
*s2
)
3897 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
3900 static void serialcpy(__u8
*dest
, __u8
*src
)
3902 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
3905 static struct dl
*serial_to_dl(__u8
*serial
, struct intel_super
*super
)
3909 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
3910 if (serialcmp(dl
->serial
, serial
) == 0)
3916 static struct imsm_disk
*
3917 __serial_to_disk(__u8
*serial
, struct imsm_super
*mpb
, int *idx
)
3921 for (i
= 0; i
< mpb
->num_disks
; i
++) {
3922 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
3924 if (serialcmp(disk
->serial
, serial
) == 0) {
3935 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
3937 struct imsm_disk
*disk
;
3942 __u8 serial
[MAX_RAID_SERIAL_LEN
];
3944 rv
= imsm_read_serial(fd
, devname
, serial
);
3949 dl
= xcalloc(1, sizeof(*dl
));
3952 dl
->major
= major(stb
.st_rdev
);
3953 dl
->minor
= minor(stb
.st_rdev
);
3954 dl
->next
= super
->disks
;
3955 dl
->fd
= keep_fd
? fd
: -1;
3956 assert(super
->disks
== NULL
);
3958 serialcpy(dl
->serial
, serial
);
3961 fd2devname(fd
, name
);
3963 dl
->devname
= xstrdup(devname
);
3965 dl
->devname
= xstrdup(name
);
3967 /* look up this disk's index in the current anchor */
3968 disk
= __serial_to_disk(dl
->serial
, super
->anchor
, &dl
->index
);
3971 /* only set index on disks that are a member of a
3972 * populated contianer, i.e. one with raid_devs
3974 if (is_failed(&dl
->disk
))
3976 else if (is_spare(&dl
->disk
) || is_journal(&dl
->disk
))
3984 /* When migrating map0 contains the 'destination' state while map1
3985 * contains the current state. When not migrating map0 contains the
3986 * current state. This routine assumes that map[0].map_state is set to
3987 * the current array state before being called.
3989 * Migration is indicated by one of the following states
3990 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3991 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3992 * map1state=unitialized)
3993 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3995 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3996 * map1state=degraded)
3997 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
4000 static void migrate(struct imsm_dev
*dev
, struct intel_super
*super
,
4001 __u8 to_state
, int migr_type
)
4003 struct imsm_map
*dest
;
4004 struct imsm_map
*src
= get_imsm_map(dev
, MAP_0
);
4006 dev
->vol
.migr_state
= 1;
4007 set_migr_type(dev
, migr_type
);
4008 dev
->vol
.curr_migr_unit
= 0;
4009 dest
= get_imsm_map(dev
, MAP_1
);
4011 /* duplicate and then set the target end state in map[0] */
4012 memcpy(dest
, src
, sizeof_imsm_map(src
));
4013 if (migr_type
== MIGR_REBUILD
|| migr_type
== MIGR_GEN_MIGR
) {
4017 for (i
= 0; i
< src
->num_members
; i
++) {
4018 ord
= __le32_to_cpu(src
->disk_ord_tbl
[i
]);
4019 set_imsm_ord_tbl_ent(src
, i
, ord_to_idx(ord
));
4023 if (migr_type
== MIGR_GEN_MIGR
)
4024 /* Clear migration record */
4025 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
4027 src
->map_state
= to_state
;
4030 static void end_migration(struct imsm_dev
*dev
, struct intel_super
*super
,
4033 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
4034 struct imsm_map
*prev
= get_imsm_map(dev
, dev
->vol
.migr_state
== 0 ?
4038 /* merge any IMSM_ORD_REBUILD bits that were not successfully
4039 * completed in the last migration.
4041 * FIXME add support for raid-level-migration
4043 if (map_state
!= map
->map_state
&& (is_gen_migration(dev
) == 0) &&
4044 prev
->map_state
!= IMSM_T_STATE_UNINITIALIZED
) {
4045 /* when final map state is other than expected
4046 * merge maps (not for migration)
4050 for (i
= 0; i
< prev
->num_members
; i
++)
4051 for (j
= 0; j
< map
->num_members
; j
++)
4052 /* during online capacity expansion
4053 * disks position can be changed
4054 * if takeover is used
4056 if (ord_to_idx(map
->disk_ord_tbl
[j
]) ==
4057 ord_to_idx(prev
->disk_ord_tbl
[i
])) {
4058 map
->disk_ord_tbl
[j
] |=
4059 prev
->disk_ord_tbl
[i
];
4062 failed
= imsm_count_failed(super
, dev
, MAP_0
);
4063 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
4066 dev
->vol
.migr_state
= 0;
4067 set_migr_type(dev
, 0);
4068 dev
->vol
.curr_migr_unit
= 0;
4069 map
->map_state
= map_state
;
4073 static int parse_raid_devices(struct intel_super
*super
)
4076 struct imsm_dev
*dev_new
;
4077 size_t len
, len_migr
;
4079 size_t space_needed
= 0;
4080 struct imsm_super
*mpb
= super
->anchor
;
4082 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4083 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4084 struct intel_dev
*dv
;
4086 len
= sizeof_imsm_dev(dev_iter
, 0);
4087 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
4089 space_needed
+= len_migr
- len
;
4091 dv
= xmalloc(sizeof(*dv
));
4092 if (max_len
< len_migr
)
4094 if (max_len
> len_migr
)
4095 space_needed
+= max_len
- len_migr
;
4096 dev_new
= xmalloc(max_len
);
4097 imsm_copy_dev(dev_new
, dev_iter
);
4100 dv
->next
= super
->devlist
;
4101 super
->devlist
= dv
;
4104 /* ensure that super->buf is large enough when all raid devices
4107 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
4110 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
,
4111 super
->sector_size
);
4112 if (posix_memalign(&buf
, MAX_SECTOR_SIZE
, len
) != 0)
4115 memcpy(buf
, super
->buf
, super
->len
);
4116 memset(buf
+ super
->len
, 0, len
- super
->len
);
4122 super
->extra_space
+= space_needed
;
4127 /*******************************************************************************
4128 * Function: check_mpb_migr_compatibility
4129 * Description: Function checks for unsupported migration features:
4130 * - migration optimization area (pba_of_lba0)
4131 * - descending reshape (ascending_migr)
4133 * super : imsm metadata information
4135 * 0 : migration is compatible
4136 * -1 : migration is not compatible
4137 ******************************************************************************/
4138 int check_mpb_migr_compatibility(struct intel_super
*super
)
4140 struct imsm_map
*map0
, *map1
;
4141 struct migr_record
*migr_rec
= super
->migr_rec
;
4144 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
4145 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
4148 dev_iter
->vol
.migr_state
== 1 &&
4149 dev_iter
->vol
.migr_type
== MIGR_GEN_MIGR
) {
4150 /* This device is migrating */
4151 map0
= get_imsm_map(dev_iter
, MAP_0
);
4152 map1
= get_imsm_map(dev_iter
, MAP_1
);
4153 if (pba_of_lba0(map0
) != pba_of_lba0(map1
))
4154 /* migration optimization area was used */
4156 if (migr_rec
->ascending_migr
== 0
4157 && migr_rec
->dest_depth_per_unit
> 0)
4158 /* descending reshape not supported yet */
4165 static void __free_imsm(struct intel_super
*super
, int free_disks
);
4167 /* load_imsm_mpb - read matrix metadata
4168 * allocates super->mpb to be freed by free_imsm
4170 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
4172 unsigned long long dsize
;
4173 unsigned long long sectors
;
4174 unsigned int sector_size
= super
->sector_size
;
4176 struct imsm_super
*anchor
;
4179 get_dev_size(fd
, NULL
, &dsize
);
4180 if (dsize
< 2*sector_size
) {
4182 pr_err("%s: device to small for imsm\n",
4187 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0) {
4189 pr_err("Cannot seek to anchor block on %s: %s\n",
4190 devname
, strerror(errno
));
4194 if (posix_memalign((void **)&anchor
, sector_size
, sector_size
) != 0) {
4196 pr_err("Failed to allocate imsm anchor buffer on %s\n", devname
);
4199 if ((unsigned int)read(fd
, anchor
, sector_size
) != sector_size
) {
4201 pr_err("Cannot read anchor block on %s: %s\n",
4202 devname
, strerror(errno
));
4207 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
4209 pr_err("no IMSM anchor on %s\n", devname
);
4214 __free_imsm(super
, 0);
4215 /* reload capability and hba */
4217 /* capability and hba must be updated with new super allocation */
4218 find_intel_hba_capability(fd
, super
, devname
);
4219 super
->len
= ROUND_UP(anchor
->mpb_size
, sector_size
);
4220 if (posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, super
->len
) != 0) {
4222 pr_err("unable to allocate %zu byte mpb buffer\n",
4227 memcpy(super
->buf
, anchor
, sector_size
);
4229 sectors
= mpb_sectors(anchor
, sector_size
) - 1;
4232 if (posix_memalign(&super
->migr_rec_buf
, sector_size
,
4233 MIGR_REC_BUF_SECTORS
*sector_size
) != 0) {
4234 pr_err("could not allocate migr_rec buffer\n");
4238 super
->clean_migration_record_by_mdmon
= 0;
4241 check_sum
= __gen_imsm_checksum(super
->anchor
);
4242 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4244 pr_err("IMSM checksum %x != %x on %s\n",
4246 __le32_to_cpu(super
->anchor
->check_sum
),
4254 /* read the extended mpb */
4255 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)), SEEK_SET
) < 0) {
4257 pr_err("Cannot seek to extended mpb on %s: %s\n",
4258 devname
, strerror(errno
));
4262 if ((unsigned int)read(fd
, super
->buf
+ sector_size
,
4263 super
->len
- sector_size
) != super
->len
- sector_size
) {
4265 pr_err("Cannot read extended mpb on %s: %s\n",
4266 devname
, strerror(errno
));
4270 check_sum
= __gen_imsm_checksum(super
->anchor
);
4271 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
4273 pr_err("IMSM checksum %x != %x on %s\n",
4274 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
4282 static int read_imsm_migr_rec(int fd
, struct intel_super
*super
);
4284 /* clears hi bits in metadata if MPB_ATTRIB_2TB_DISK not set */
4285 static void clear_hi(struct intel_super
*super
)
4287 struct imsm_super
*mpb
= super
->anchor
;
4289 if (mpb
->attributes
& MPB_ATTRIB_2TB_DISK
)
4291 for (i
= 0; i
< mpb
->num_disks
; ++i
) {
4292 struct imsm_disk
*disk
= &mpb
->disk
[i
];
4293 disk
->total_blocks_hi
= 0;
4295 for (i
= 0; i
< mpb
->num_raid_devs
; ++i
) {
4296 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
4299 for (n
= 0; n
< 2; ++n
) {
4300 struct imsm_map
*map
= get_imsm_map(dev
, n
);
4303 map
->pba_of_lba0_hi
= 0;
4304 map
->blocks_per_member_hi
= 0;
4305 map
->num_data_stripes_hi
= 0;
4311 load_and_parse_mpb(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
4315 err
= load_imsm_mpb(fd
, super
, devname
);
4318 if (super
->sector_size
== 4096)
4319 convert_from_4k(super
);
4320 err
= load_imsm_disk(fd
, super
, devname
, keep_fd
);
4323 err
= parse_raid_devices(super
);
4326 err
= load_bbm_log(super
);
4331 static void __free_imsm_disk(struct dl
*d
)
4343 static void free_imsm_disks(struct intel_super
*super
)
4347 while (super
->disks
) {
4349 super
->disks
= d
->next
;
4350 __free_imsm_disk(d
);
4352 while (super
->disk_mgmt_list
) {
4353 d
= super
->disk_mgmt_list
;
4354 super
->disk_mgmt_list
= d
->next
;
4355 __free_imsm_disk(d
);
4357 while (super
->missing
) {
4359 super
->missing
= d
->next
;
4360 __free_imsm_disk(d
);
4365 /* free all the pieces hanging off of a super pointer */
4366 static void __free_imsm(struct intel_super
*super
, int free_disks
)
4368 struct intel_hba
*elem
, *next
;
4374 /* unlink capability description */
4376 if (super
->migr_rec_buf
) {
4377 free(super
->migr_rec_buf
);
4378 super
->migr_rec_buf
= NULL
;
4381 free_imsm_disks(super
);
4382 free_devlist(super
);
4386 free((void *)elem
->path
);
4392 free(super
->bbm_log
);
4396 static void free_imsm(struct intel_super
*super
)
4398 __free_imsm(super
, 1);
4399 free(super
->bb
.entries
);
4403 static void free_super_imsm(struct supertype
*st
)
4405 struct intel_super
*super
= st
->sb
;
4414 static struct intel_super
*alloc_super(void)
4416 struct intel_super
*super
= xcalloc(1, sizeof(*super
));
4418 super
->current_vol
= -1;
4419 super
->create_offset
= ~((unsigned long long) 0);
4421 super
->bb
.entries
= xmalloc(BBM_LOG_MAX_ENTRIES
*
4422 sizeof(struct md_bb_entry
));
4423 if (!super
->bb
.entries
) {
4432 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
4434 static int find_intel_hba_capability(int fd
, struct intel_super
*super
, char *devname
)
4436 struct sys_dev
*hba_name
;
4439 if (fd
< 0 || check_env("IMSM_NO_PLATFORM")) {
4444 hba_name
= find_disk_attached_hba(fd
, NULL
);
4447 pr_err("%s is not attached to Intel(R) RAID controller.\n",
4451 rv
= attach_hba_to_super(super
, hba_name
);
4454 struct intel_hba
*hba
= super
->hba
;
4456 pr_err("%s is attached to Intel(R) %s %s (%s),\n"
4457 " but the container is assigned to Intel(R) %s %s (",
4459 get_sys_dev_type(hba_name
->type
),
4460 hba_name
->type
== SYS_DEV_VMD
? "domain" : "RAID controller",
4461 hba_name
->pci_id
? : "Err!",
4462 get_sys_dev_type(super
->hba
->type
),
4463 hba
->type
== SYS_DEV_VMD
? "domain" : "RAID controller");
4466 fprintf(stderr
, "%s", hba
->pci_id
? : "Err!");
4468 fprintf(stderr
, ", ");
4471 fprintf(stderr
, ").\n"
4472 " Mixing devices attached to different %s is not allowed.\n",
4473 hba_name
->type
== SYS_DEV_VMD
? "VMD domains" : "controllers");
4477 super
->orom
= find_imsm_capability(hba_name
);
4484 /* find_missing - helper routine for load_super_imsm_all that identifies
4485 * disks that have disappeared from the system. This routine relies on
4486 * the mpb being uptodate, which it is at load time.
4488 static int find_missing(struct intel_super
*super
)
4491 struct imsm_super
*mpb
= super
->anchor
;
4493 struct imsm_disk
*disk
;
4495 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4496 disk
= __get_imsm_disk(mpb
, i
);
4497 dl
= serial_to_dl(disk
->serial
, super
);
4501 dl
= xmalloc(sizeof(*dl
));
4505 dl
->devname
= xstrdup("missing");
4507 serialcpy(dl
->serial
, disk
->serial
);
4510 dl
->next
= super
->missing
;
4511 super
->missing
= dl
;
4518 static struct intel_disk
*disk_list_get(__u8
*serial
, struct intel_disk
*disk_list
)
4520 struct intel_disk
*idisk
= disk_list
;
4523 if (serialcmp(idisk
->disk
.serial
, serial
) == 0)
4525 idisk
= idisk
->next
;
4531 static int __prep_thunderdome(struct intel_super
**table
, int tbl_size
,
4532 struct intel_super
*super
,
4533 struct intel_disk
**disk_list
)
4535 struct imsm_disk
*d
= &super
->disks
->disk
;
4536 struct imsm_super
*mpb
= super
->anchor
;
4539 for (i
= 0; i
< tbl_size
; i
++) {
4540 struct imsm_super
*tbl_mpb
= table
[i
]->anchor
;
4541 struct imsm_disk
*tbl_d
= &table
[i
]->disks
->disk
;
4543 if (tbl_mpb
->family_num
== mpb
->family_num
) {
4544 if (tbl_mpb
->check_sum
== mpb
->check_sum
) {
4545 dprintf("mpb from %d:%d matches %d:%d\n",
4546 super
->disks
->major
,
4547 super
->disks
->minor
,
4548 table
[i
]->disks
->major
,
4549 table
[i
]->disks
->minor
);
4553 if (((is_configured(d
) && !is_configured(tbl_d
)) ||
4554 is_configured(d
) == is_configured(tbl_d
)) &&
4555 tbl_mpb
->generation_num
< mpb
->generation_num
) {
4556 /* current version of the mpb is a
4557 * better candidate than the one in
4558 * super_table, but copy over "cross
4559 * generational" status
4561 struct intel_disk
*idisk
;
4563 dprintf("mpb from %d:%d replaces %d:%d\n",
4564 super
->disks
->major
,
4565 super
->disks
->minor
,
4566 table
[i
]->disks
->major
,
4567 table
[i
]->disks
->minor
);
4569 idisk
= disk_list_get(tbl_d
->serial
, *disk_list
);
4570 if (idisk
&& is_failed(&idisk
->disk
))
4571 tbl_d
->status
|= FAILED_DISK
;
4574 struct intel_disk
*idisk
;
4575 struct imsm_disk
*disk
;
4577 /* tbl_mpb is more up to date, but copy
4578 * over cross generational status before
4581 disk
= __serial_to_disk(d
->serial
, mpb
, NULL
);
4582 if (disk
&& is_failed(disk
))
4583 d
->status
|= FAILED_DISK
;
4585 idisk
= disk_list_get(d
->serial
, *disk_list
);
4588 if (disk
&& is_configured(disk
))
4589 idisk
->disk
.status
|= CONFIGURED_DISK
;
4592 dprintf("mpb from %d:%d prefer %d:%d\n",
4593 super
->disks
->major
,
4594 super
->disks
->minor
,
4595 table
[i
]->disks
->major
,
4596 table
[i
]->disks
->minor
);
4604 table
[tbl_size
++] = super
;
4608 /* update/extend the merged list of imsm_disk records */
4609 for (j
= 0; j
< mpb
->num_disks
; j
++) {
4610 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, j
);
4611 struct intel_disk
*idisk
;
4613 idisk
= disk_list_get(disk
->serial
, *disk_list
);
4615 idisk
->disk
.status
|= disk
->status
;
4616 if (is_configured(&idisk
->disk
) ||
4617 is_failed(&idisk
->disk
))
4618 idisk
->disk
.status
&= ~(SPARE_DISK
);
4620 idisk
= xcalloc(1, sizeof(*idisk
));
4621 idisk
->owner
= IMSM_UNKNOWN_OWNER
;
4622 idisk
->disk
= *disk
;
4623 idisk
->next
= *disk_list
;
4627 if (serialcmp(idisk
->disk
.serial
, d
->serial
) == 0)
4634 static struct intel_super
*
4635 validate_members(struct intel_super
*super
, struct intel_disk
*disk_list
,
4638 struct imsm_super
*mpb
= super
->anchor
;
4642 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4643 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, i
);
4644 struct intel_disk
*idisk
;
4646 idisk
= disk_list_get(disk
->serial
, disk_list
);
4648 if (idisk
->owner
== owner
||
4649 idisk
->owner
== IMSM_UNKNOWN_OWNER
)
4652 dprintf("'%.16s' owner %d != %d\n",
4653 disk
->serial
, idisk
->owner
,
4656 dprintf("unknown disk %x [%d]: %.16s\n",
4657 __le32_to_cpu(mpb
->family_num
), i
,
4663 if (ok_count
== mpb
->num_disks
)
4668 static void show_conflicts(__u32 family_num
, struct intel_super
*super_list
)
4670 struct intel_super
*s
;
4672 for (s
= super_list
; s
; s
= s
->next
) {
4673 if (family_num
!= s
->anchor
->family_num
)
4675 pr_err("Conflict, offlining family %#x on '%s'\n",
4676 __le32_to_cpu(family_num
), s
->disks
->devname
);
4680 static struct intel_super
*
4681 imsm_thunderdome(struct intel_super
**super_list
, int len
)
4683 struct intel_super
*super_table
[len
];
4684 struct intel_disk
*disk_list
= NULL
;
4685 struct intel_super
*champion
, *spare
;
4686 struct intel_super
*s
, **del
;
4691 memset(super_table
, 0, sizeof(super_table
));
4692 for (s
= *super_list
; s
; s
= s
->next
)
4693 tbl_size
= __prep_thunderdome(super_table
, tbl_size
, s
, &disk_list
);
4695 for (i
= 0; i
< tbl_size
; i
++) {
4696 struct imsm_disk
*d
;
4697 struct intel_disk
*idisk
;
4698 struct imsm_super
*mpb
= super_table
[i
]->anchor
;
4701 d
= &s
->disks
->disk
;
4703 /* 'd' must appear in merged disk list for its
4704 * configuration to be valid
4706 idisk
= disk_list_get(d
->serial
, disk_list
);
4707 if (idisk
&& idisk
->owner
== i
)
4708 s
= validate_members(s
, disk_list
, i
);
4713 dprintf("marking family: %#x from %d:%d offline\n",
4715 super_table
[i
]->disks
->major
,
4716 super_table
[i
]->disks
->minor
);
4720 /* This is where the mdadm implementation differs from the Windows
4721 * driver which has no strict concept of a container. We can only
4722 * assemble one family from a container, so when returning a prodigal
4723 * array member to this system the code will not be able to disambiguate
4724 * the container contents that should be assembled ("foreign" versus
4725 * "local"). It requires user intervention to set the orig_family_num
4726 * to a new value to establish a new container. The Windows driver in
4727 * this situation fixes up the volume name in place and manages the
4728 * foreign array as an independent entity.
4733 for (i
= 0; i
< tbl_size
; i
++) {
4734 struct intel_super
*tbl_ent
= super_table
[i
];
4740 if (tbl_ent
->anchor
->num_raid_devs
== 0) {
4745 if (s
&& !is_spare
) {
4746 show_conflicts(tbl_ent
->anchor
->family_num
, *super_list
);
4748 } else if (!s
&& !is_spare
)
4761 pr_err("Chose family %#x on '%s', assemble conflicts to new container with '--update=uuid'\n",
4762 __le32_to_cpu(s
->anchor
->family_num
), s
->disks
->devname
);
4764 /* collect all dl's onto 'champion', and update them to
4765 * champion's version of the status
4767 for (s
= *super_list
; s
; s
= s
->next
) {
4768 struct imsm_super
*mpb
= champion
->anchor
;
4769 struct dl
*dl
= s
->disks
;
4774 mpb
->attributes
|= s
->anchor
->attributes
& MPB_ATTRIB_2TB_DISK
;
4776 for (i
= 0; i
< mpb
->num_disks
; i
++) {
4777 struct imsm_disk
*disk
;
4779 disk
= __serial_to_disk(dl
->serial
, mpb
, &dl
->index
);
4782 /* only set index on disks that are a member of
4783 * a populated contianer, i.e. one with
4786 if (is_failed(&dl
->disk
))
4788 else if (is_spare(&dl
->disk
))
4794 if (i
>= mpb
->num_disks
) {
4795 struct intel_disk
*idisk
;
4797 idisk
= disk_list_get(dl
->serial
, disk_list
);
4798 if (idisk
&& is_spare(&idisk
->disk
) &&
4799 !is_failed(&idisk
->disk
) && !is_configured(&idisk
->disk
))
4807 dl
->next
= champion
->disks
;
4808 champion
->disks
= dl
;
4812 /* delete 'champion' from super_list */
4813 for (del
= super_list
; *del
; ) {
4814 if (*del
== champion
) {
4815 *del
= (*del
)->next
;
4818 del
= &(*del
)->next
;
4820 champion
->next
= NULL
;
4824 struct intel_disk
*idisk
= disk_list
;
4826 disk_list
= disk_list
->next
;
4834 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
);
4835 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4836 int major
, int minor
, int keep_fd
);
4838 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4839 int *max
, int keep_fd
);
4841 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
4842 char *devname
, struct md_list
*devlist
,
4845 struct intel_super
*super_list
= NULL
;
4846 struct intel_super
*super
= NULL
;
4851 /* 'fd' is an opened container */
4852 err
= get_sra_super_block(fd
, &super_list
, devname
, &i
, keep_fd
);
4854 /* get super block from devlist devices */
4855 err
= get_devlist_super_block(devlist
, &super_list
, &i
, keep_fd
);
4858 /* all mpbs enter, maybe one leaves */
4859 super
= imsm_thunderdome(&super_list
, i
);
4865 if (find_missing(super
) != 0) {
4871 /* load migration record */
4872 err
= load_imsm_migr_rec(super
, NULL
);
4874 /* migration is in progress,
4875 * but migr_rec cannot be loaded,
4881 /* Check migration compatibility */
4882 if (err
== 0 && check_mpb_migr_compatibility(super
) != 0) {
4883 pr_err("Unsupported migration detected");
4885 fprintf(stderr
, " on %s\n", devname
);
4887 fprintf(stderr
, " (IMSM).\n");
4896 while (super_list
) {
4897 struct intel_super
*s
= super_list
;
4899 super_list
= super_list
->next
;
4908 strcpy(st
->container_devnm
, fd2devnm(fd
));
4910 st
->container_devnm
[0] = 0;
4911 if (err
== 0 && st
->ss
== NULL
) {
4912 st
->ss
= &super_imsm
;
4913 st
->minor_version
= 0;
4914 st
->max_devs
= IMSM_MAX_DEVICES
;
4920 get_devlist_super_block(struct md_list
*devlist
, struct intel_super
**super_list
,
4921 int *max
, int keep_fd
)
4923 struct md_list
*tmpdev
;
4927 for (i
= 0, tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
4928 if (tmpdev
->used
!= 1)
4930 if (tmpdev
->container
== 1) {
4932 int fd
= dev_open(tmpdev
->devname
, O_RDONLY
|O_EXCL
);
4934 pr_err("cannot open device %s: %s\n",
4935 tmpdev
->devname
, strerror(errno
));
4939 err
= get_sra_super_block(fd
, super_list
,
4940 tmpdev
->devname
, &lmax
,
4949 int major
= major(tmpdev
->st_rdev
);
4950 int minor
= minor(tmpdev
->st_rdev
);
4951 err
= get_super_block(super_list
,
4968 static int get_super_block(struct intel_super
**super_list
, char *devnm
, char *devname
,
4969 int major
, int minor
, int keep_fd
)
4971 struct intel_super
*s
;
4983 sprintf(nm
, "%d:%d", major
, minor
);
4984 dfd
= dev_open(nm
, O_RDWR
);
4990 get_dev_sector_size(dfd
, NULL
, &s
->sector_size
);
4991 find_intel_hba_capability(dfd
, s
, devname
);
4992 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
4994 /* retry the load if we might have raced against mdmon */
4995 if (err
== 3 && devnm
&& mdmon_running(devnm
))
4996 for (retry
= 0; retry
< 3; retry
++) {
4998 err
= load_and_parse_mpb(dfd
, s
, NULL
, keep_fd
);
5004 s
->next
= *super_list
;
5012 if (dfd
>= 0 && !keep_fd
)
5019 get_sra_super_block(int fd
, struct intel_super
**super_list
, char *devname
, int *max
, int keep_fd
)
5026 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
5030 if (sra
->array
.major_version
!= -1 ||
5031 sra
->array
.minor_version
!= -2 ||
5032 strcmp(sra
->text_version
, "imsm") != 0) {
5037 devnm
= fd2devnm(fd
);
5038 for (sd
= sra
->devs
, i
= 0; sd
; sd
= sd
->next
, i
++) {
5039 if (get_super_block(super_list
, devnm
, devname
,
5040 sd
->disk
.major
, sd
->disk
.minor
, keep_fd
) != 0) {
5051 static int load_container_imsm(struct supertype
*st
, int fd
, char *devname
)
5053 return load_super_imsm_all(st
, fd
, &st
->sb
, devname
, NULL
, 1);
5057 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
5059 struct intel_super
*super
;
5063 if (test_partition(fd
))
5064 /* IMSM not allowed on partitions */
5067 free_super_imsm(st
);
5069 super
= alloc_super();
5070 get_dev_sector_size(fd
, NULL
, &super
->sector_size
);
5073 /* Load hba and capabilities if they exist.
5074 * But do not preclude loading metadata in case capabilities or hba are
5075 * non-compliant and ignore_hw_compat is set.
5077 rv
= find_intel_hba_capability(fd
, super
, devname
);
5078 /* no orom/efi or non-intel hba of the disk */
5079 if (rv
!= 0 && st
->ignore_hw_compat
== 0) {
5081 pr_err("No OROM/EFI properties for %s\n", devname
);
5085 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5087 /* retry the load if we might have raced against mdmon */
5089 struct mdstat_ent
*mdstat
= NULL
;
5090 char *name
= fd2kname(fd
);
5093 mdstat
= mdstat_by_component(name
);
5095 if (mdstat
&& mdmon_running(mdstat
->devnm
) && getpid() != mdmon_pid(mdstat
->devnm
)) {
5096 for (retry
= 0; retry
< 3; retry
++) {
5098 rv
= load_and_parse_mpb(fd
, super
, devname
, 0);
5104 free_mdstat(mdstat
);
5109 pr_err("Failed to load all information sections on %s\n", devname
);
5115 if (st
->ss
== NULL
) {
5116 st
->ss
= &super_imsm
;
5117 st
->minor_version
= 0;
5118 st
->max_devs
= IMSM_MAX_DEVICES
;
5121 /* load migration record */
5122 if (load_imsm_migr_rec(super
, NULL
) == 0) {
5123 /* Check for unsupported migration features */
5124 if (check_mpb_migr_compatibility(super
) != 0) {
5125 pr_err("Unsupported migration detected");
5127 fprintf(stderr
, " on %s\n", devname
);
5129 fprintf(stderr
, " (IMSM).\n");
5137 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
5139 if (info
->level
== 1)
5141 return info
->chunk_size
>> 9;
5144 static unsigned long long info_to_blocks_per_member(mdu_array_info_t
*info
,
5145 unsigned long long size
)
5147 if (info
->level
== 1)
5150 return (size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
5153 static void imsm_update_version_info(struct intel_super
*super
)
5155 /* update the version and attributes */
5156 struct imsm_super
*mpb
= super
->anchor
;
5158 struct imsm_dev
*dev
;
5159 struct imsm_map
*map
;
5162 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5163 dev
= get_imsm_dev(super
, i
);
5164 map
= get_imsm_map(dev
, MAP_0
);
5165 if (__le32_to_cpu(dev
->size_high
) > 0)
5166 mpb
->attributes
|= MPB_ATTRIB_2TB
;
5168 /* FIXME detect when an array spans a port multiplier */
5170 mpb
->attributes
|= MPB_ATTRIB_PM
;
5173 if (mpb
->num_raid_devs
> 1 ||
5174 mpb
->attributes
!= MPB_ATTRIB_CHECKSUM_VERIFY
) {
5175 version
= MPB_VERSION_ATTRIBS
;
5176 switch (get_imsm_raid_level(map
)) {
5177 case 0: mpb
->attributes
|= MPB_ATTRIB_RAID0
; break;
5178 case 1: mpb
->attributes
|= MPB_ATTRIB_RAID1
; break;
5179 case 10: mpb
->attributes
|= MPB_ATTRIB_RAID10
; break;
5180 case 5: mpb
->attributes
|= MPB_ATTRIB_RAID5
; break;
5183 if (map
->num_members
>= 5)
5184 version
= MPB_VERSION_5OR6_DISK_ARRAY
;
5185 else if (dev
->status
== DEV_CLONE_N_GO
)
5186 version
= MPB_VERSION_CNG
;
5187 else if (get_imsm_raid_level(map
) == 5)
5188 version
= MPB_VERSION_RAID5
;
5189 else if (map
->num_members
>= 3)
5190 version
= MPB_VERSION_3OR4_DISK_ARRAY
;
5191 else if (get_imsm_raid_level(map
) == 1)
5192 version
= MPB_VERSION_RAID1
;
5194 version
= MPB_VERSION_RAID0
;
5196 strcpy(((char *) mpb
->sig
) + strlen(MPB_SIGNATURE
), version
);
5200 static int check_name(struct intel_super
*super
, char *name
, int quiet
)
5202 struct imsm_super
*mpb
= super
->anchor
;
5203 char *reason
= NULL
;
5206 if (strlen(name
) > MAX_RAID_SERIAL_LEN
)
5207 reason
= "must be 16 characters or less";
5209 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5210 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
5212 if (strncmp((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
) == 0) {
5213 reason
= "already exists";
5218 if (reason
&& !quiet
)
5219 pr_err("imsm volume name %s\n", reason
);
5224 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
5225 struct shape
*s
, char *name
,
5226 char *homehost
, int *uuid
,
5227 long long data_offset
)
5229 /* We are creating a volume inside a pre-existing container.
5230 * so st->sb is already set.
5232 struct intel_super
*super
= st
->sb
;
5233 unsigned int sector_size
= super
->sector_size
;
5234 struct imsm_super
*mpb
= super
->anchor
;
5235 struct intel_dev
*dv
;
5236 struct imsm_dev
*dev
;
5237 struct imsm_vol
*vol
;
5238 struct imsm_map
*map
;
5239 int idx
= mpb
->num_raid_devs
;
5241 unsigned long long array_blocks
;
5242 size_t size_old
, size_new
;
5243 unsigned long long num_data_stripes
;
5245 if (super
->orom
&& mpb
->num_raid_devs
>= super
->orom
->vpa
) {
5246 pr_err("This imsm-container already has the maximum of %d volumes\n", super
->orom
->vpa
);
5250 /* ensure the mpb is large enough for the new data */
5251 size_old
= __le32_to_cpu(mpb
->mpb_size
);
5252 size_new
= disks_to_mpb_size(info
->nr_disks
);
5253 if (size_new
> size_old
) {
5255 size_t size_round
= ROUND_UP(size_new
, sector_size
);
5257 if (posix_memalign(&mpb_new
, sector_size
, size_round
) != 0) {
5258 pr_err("could not allocate new mpb\n");
5261 if (posix_memalign(&super
->migr_rec_buf
, sector_size
,
5262 MIGR_REC_BUF_SECTORS
*sector_size
) != 0) {
5263 pr_err("could not allocate migr_rec buffer\n");
5269 memcpy(mpb_new
, mpb
, size_old
);
5272 super
->anchor
= mpb_new
;
5273 mpb
->mpb_size
= __cpu_to_le32(size_new
);
5274 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
5275 super
->len
= size_round
;
5277 super
->current_vol
= idx
;
5279 /* handle 'failed_disks' by either:
5280 * a) create dummy disk entries in the table if this the first
5281 * volume in the array. We add them here as this is the only
5282 * opportunity to add them. add_to_super_imsm_volume()
5283 * handles the non-failed disks and continues incrementing
5285 * b) validate that 'failed_disks' matches the current number
5286 * of missing disks if the container is populated
5288 if (super
->current_vol
== 0) {
5290 for (i
= 0; i
< info
->failed_disks
; i
++) {
5291 struct imsm_disk
*disk
;
5294 disk
= __get_imsm_disk(mpb
, i
);
5295 disk
->status
= CONFIGURED_DISK
| FAILED_DISK
;
5296 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
5297 snprintf((char *) disk
->serial
, MAX_RAID_SERIAL_LEN
,
5298 "missing:%d", (__u8
)i
);
5300 find_missing(super
);
5305 for (d
= super
->missing
; d
; d
= d
->next
)
5307 if (info
->failed_disks
> missing
) {
5308 pr_err("unable to add 'missing' disk to container\n");
5313 if (!check_name(super
, name
, 0))
5315 dv
= xmalloc(sizeof(*dv
));
5316 dev
= xcalloc(1, sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
5317 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
5318 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
5319 info
->layout
, info
->chunk_size
,
5321 /* round array size down to closest MB */
5322 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
5324 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
5325 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
5326 dev
->status
= (DEV_READ_COALESCING
| DEV_WRITE_COALESCING
);
5328 vol
->migr_state
= 0;
5329 set_migr_type(dev
, MIGR_INIT
);
5330 vol
->dirty
= !info
->state
;
5331 vol
->curr_migr_unit
= 0;
5332 map
= get_imsm_map(dev
, MAP_0
);
5333 set_pba_of_lba0(map
, super
->create_offset
);
5334 set_blocks_per_member(map
, info_to_blocks_per_member(info
, s
->size
));
5335 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
5336 map
->failed_disk_num
= ~0;
5337 if (info
->level
> 0)
5338 map
->map_state
= (info
->state
? IMSM_T_STATE_NORMAL
5339 : IMSM_T_STATE_UNINITIALIZED
);
5341 map
->map_state
= info
->failed_disks
? IMSM_T_STATE_FAILED
:
5342 IMSM_T_STATE_NORMAL
;
5345 if (info
->level
== 1 && info
->raid_disks
> 2) {
5348 pr_err("imsm does not support more than 2 disksin a raid1 volume\n");
5352 map
->raid_level
= info
->level
;
5353 if (info
->level
== 10) {
5354 map
->raid_level
= 1;
5355 map
->num_domains
= info
->raid_disks
/ 2;
5356 } else if (info
->level
== 1)
5357 map
->num_domains
= info
->raid_disks
;
5359 map
->num_domains
= 1;
5361 /* info->size is only int so use the 'size' parameter instead */
5362 num_data_stripes
= (s
->size
* 2) / info_to_blocks_per_strip(info
);
5363 num_data_stripes
/= map
->num_domains
;
5364 set_num_data_stripes(map
, num_data_stripes
);
5366 map
->num_members
= info
->raid_disks
;
5367 for (i
= 0; i
< map
->num_members
; i
++) {
5368 /* initialized in add_to_super */
5369 set_imsm_ord_tbl_ent(map
, i
, IMSM_ORD_REBUILD
);
5371 mpb
->num_raid_devs
++;
5373 if (s
->consistency_policy
== UnSet
||
5374 s
->consistency_policy
== CONSISTENCY_POLICY_RESYNC
||
5375 s
->consistency_policy
== CONSISTENCY_POLICY_NONE
) {
5376 dev
->rwh_policy
= RWH_OFF
;
5377 } else if (s
->consistency_policy
== CONSISTENCY_POLICY_PPL
) {
5378 dev
->rwh_policy
= RWH_DISTRIBUTED
;
5382 pr_err("imsm does not support consistency policy %s\n",
5383 map_num(consistency_policies
, s
->consistency_policy
));
5388 dv
->index
= super
->current_vol
;
5389 dv
->next
= super
->devlist
;
5390 super
->devlist
= dv
;
5392 imsm_update_version_info(super
);
5397 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
5398 struct shape
*s
, char *name
,
5399 char *homehost
, int *uuid
,
5400 unsigned long long data_offset
)
5402 /* This is primarily called by Create when creating a new array.
5403 * We will then get add_to_super called for each component, and then
5404 * write_init_super called to write it out to each device.
5405 * For IMSM, Create can create on fresh devices or on a pre-existing
5407 * To create on a pre-existing array a different method will be called.
5408 * This one is just for fresh drives.
5410 struct intel_super
*super
;
5411 struct imsm_super
*mpb
;
5415 if (data_offset
!= INVALID_SECTORS
) {
5416 pr_err("data-offset not supported by imsm\n");
5421 return init_super_imsm_volume(st
, info
, s
, name
, homehost
, uuid
,
5425 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
5427 mpb_size
= MAX_SECTOR_SIZE
;
5429 super
= alloc_super();
5431 posix_memalign(&super
->buf
, MAX_SECTOR_SIZE
, mpb_size
) != 0) {
5436 pr_err("could not allocate superblock\n");
5439 if (posix_memalign(&super
->migr_rec_buf
, MAX_SECTOR_SIZE
,
5440 MIGR_REC_BUF_SECTORS
*MAX_SECTOR_SIZE
) != 0) {
5441 pr_err("could not allocate migr_rec buffer\n");
5446 memset(super
->buf
, 0, mpb_size
);
5448 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5452 /* zeroing superblock */
5456 mpb
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5458 version
= (char *) mpb
->sig
;
5459 strcpy(version
, MPB_SIGNATURE
);
5460 version
+= strlen(MPB_SIGNATURE
);
5461 strcpy(version
, MPB_VERSION_RAID0
);
5467 static int add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
5468 int fd
, char *devname
)
5470 struct intel_super
*super
= st
->sb
;
5471 struct imsm_super
*mpb
= super
->anchor
;
5472 struct imsm_disk
*_disk
;
5473 struct imsm_dev
*dev
;
5474 struct imsm_map
*map
;
5478 dev
= get_imsm_dev(super
, super
->current_vol
);
5479 map
= get_imsm_map(dev
, MAP_0
);
5481 if (! (dk
->state
& (1<<MD_DISK_SYNC
))) {
5482 pr_err("%s: Cannot add spare devices to IMSM volume\n",
5488 /* we're doing autolayout so grab the pre-marked (in
5489 * validate_geometry) raid_disk
5491 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5492 if (dl
->raiddisk
== dk
->raid_disk
)
5495 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
5496 if (dl
->major
== dk
->major
&&
5497 dl
->minor
== dk
->minor
)
5502 pr_err("%s is not a member of the same container\n", devname
);
5506 /* add a pristine spare to the metadata */
5507 if (dl
->index
< 0) {
5508 dl
->index
= super
->anchor
->num_disks
;
5509 super
->anchor
->num_disks
++;
5511 /* Check the device has not already been added */
5512 slot
= get_imsm_disk_slot(map
, dl
->index
);
5514 (get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
) & IMSM_ORD_REBUILD
) == 0) {
5515 pr_err("%s has been included in this array twice\n",
5519 set_imsm_ord_tbl_ent(map
, dk
->raid_disk
, dl
->index
);
5520 dl
->disk
.status
= CONFIGURED_DISK
;
5522 /* update size of 'missing' disks to be at least as large as the
5523 * largest acitve member (we only have dummy missing disks when
5524 * creating the first volume)
5526 if (super
->current_vol
== 0) {
5527 for (df
= super
->missing
; df
; df
= df
->next
) {
5528 if (total_blocks(&dl
->disk
) > total_blocks(&df
->disk
))
5529 set_total_blocks(&df
->disk
, total_blocks(&dl
->disk
));
5530 _disk
= __get_imsm_disk(mpb
, df
->index
);
5535 /* refresh unset/failed slots to point to valid 'missing' entries */
5536 for (df
= super
->missing
; df
; df
= df
->next
)
5537 for (slot
= 0; slot
< mpb
->num_disks
; slot
++) {
5538 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
5540 if ((ord
& IMSM_ORD_REBUILD
) == 0)
5542 set_imsm_ord_tbl_ent(map
, slot
, df
->index
| IMSM_ORD_REBUILD
);
5543 if (is_gen_migration(dev
)) {
5544 struct imsm_map
*map2
= get_imsm_map(dev
,
5546 int slot2
= get_imsm_disk_slot(map2
, df
->index
);
5547 if (slot2
< map2
->num_members
&& slot2
>= 0) {
5548 __u32 ord2
= get_imsm_ord_tbl_ent(dev
,
5551 if ((unsigned)df
->index
==
5553 set_imsm_ord_tbl_ent(map2
,
5559 dprintf("set slot:%d to missing disk:%d\n", slot
, df
->index
);
5563 /* if we are creating the first raid device update the family number */
5564 if (super
->current_vol
== 0) {
5566 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
5568 _disk
= __get_imsm_disk(mpb
, dl
->index
);
5569 if (!_dev
|| !_disk
) {
5570 pr_err("BUG mpb setup error\n");
5576 sum
+= __gen_imsm_checksum(mpb
);
5577 mpb
->family_num
= __cpu_to_le32(sum
);
5578 mpb
->orig_family_num
= mpb
->family_num
;
5580 super
->current_disk
= dl
;
5585 * Function marks disk as spare and restores disk serial
5586 * in case it was previously marked as failed by takeover operation
5588 * -1 : critical error
5589 * 0 : disk is marked as spare but serial is not set
5592 int mark_spare(struct dl
*disk
)
5594 __u8 serial
[MAX_RAID_SERIAL_LEN
];
5601 if (!imsm_read_serial(disk
->fd
, NULL
, serial
)) {
5602 /* Restore disk serial number, because takeover marks disk
5603 * as failed and adds to serial ':0' before it becomes
5606 serialcpy(disk
->serial
, serial
);
5607 serialcpy(disk
->disk
.serial
, serial
);
5610 disk
->disk
.status
= SPARE_DISK
;
5616 static int add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
5617 int fd
, char *devname
,
5618 unsigned long long data_offset
)
5620 struct intel_super
*super
= st
->sb
;
5622 unsigned long long size
;
5623 unsigned int member_sector_size
;
5628 /* If we are on an RAID enabled platform check that the disk is
5629 * attached to the raid controller.
5630 * We do not need to test disks attachment for container based additions,
5631 * they shall be already tested when container was created/assembled.
5633 rv
= find_intel_hba_capability(fd
, super
, devname
);
5634 /* no orom/efi or non-intel hba of the disk */
5636 dprintf("capability: %p fd: %d ret: %d\n",
5637 super
->orom
, fd
, rv
);
5641 if (super
->current_vol
>= 0)
5642 return add_to_super_imsm_volume(st
, dk
, fd
, devname
);
5645 dd
= xcalloc(sizeof(*dd
), 1);
5646 dd
->major
= major(stb
.st_rdev
);
5647 dd
->minor
= minor(stb
.st_rdev
);
5648 dd
->devname
= devname
? xstrdup(devname
) : NULL
;
5651 dd
->action
= DISK_ADD
;
5652 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
5654 pr_err("failed to retrieve scsi serial, aborting\n");
5660 if (super
->hba
&& ((super
->hba
->type
== SYS_DEV_NVME
) ||
5661 (super
->hba
->type
== SYS_DEV_VMD
))) {
5663 char *devpath
= diskfd_to_devpath(fd
);
5664 char controller_path
[PATH_MAX
];
5667 pr_err("failed to get devpath, aborting\n");
5674 snprintf(controller_path
, PATH_MAX
-1, "%s/device", devpath
);
5677 if (devpath_to_vendor(controller_path
) == 0x8086) {
5679 * If Intel's NVMe drive has serial ended with
5680 * "-A","-B","-1" or "-2" it means that this is "x8"
5681 * device (double drive on single PCIe card).
5682 * User should be warned about potential data loss.
5684 for (i
= MAX_RAID_SERIAL_LEN
-1; i
> 0; i
--) {
5685 /* Skip empty character at the end */
5686 if (dd
->serial
[i
] == 0)
5689 if (((dd
->serial
[i
] == 'A') ||
5690 (dd
->serial
[i
] == 'B') ||
5691 (dd
->serial
[i
] == '1') ||
5692 (dd
->serial
[i
] == '2')) &&
5693 (dd
->serial
[i
-1] == '-'))
5694 pr_err("\tThe action you are about to take may put your data at risk.\n"
5695 "\tPlease note that x8 devices may consist of two separate x4 devices "
5696 "located on a single PCIe port.\n"
5697 "\tRAID 0 is the only supported configuration for this type of x8 device.\n");
5700 } else if (super
->hba
->type
== SYS_DEV_VMD
&& super
->orom
&&
5701 !imsm_orom_has_tpv_support(super
->orom
)) {
5702 pr_err("\tPlatform configuration does not support non-Intel NVMe drives.\n"
5703 "\tPlease refer to Intel(R) RSTe user guide.\n");
5710 get_dev_size(fd
, NULL
, &size
);
5711 get_dev_sector_size(fd
, NULL
, &member_sector_size
);
5713 if (super
->sector_size
== 0) {
5714 /* this a first device, so sector_size is not set yet */
5715 super
->sector_size
= member_sector_size
;
5716 } else if (member_sector_size
!= super
->sector_size
) {
5717 pr_err("Mixing between different sector size is forbidden, aborting...\n");
5724 /* clear migr_rec when adding disk to container */
5725 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*super
->sector_size
);
5726 if (lseek64(fd
, size
- MIGR_REC_SECTOR_POSITION
*super
->sector_size
,
5728 if ((unsigned int)write(fd
, super
->migr_rec_buf
,
5729 MIGR_REC_BUF_SECTORS
*super
->sector_size
) !=
5730 MIGR_REC_BUF_SECTORS
*super
->sector_size
)
5731 perror("Write migr_rec failed");
5735 serialcpy(dd
->disk
.serial
, dd
->serial
);
5736 set_total_blocks(&dd
->disk
, size
);
5737 if (__le32_to_cpu(dd
->disk
.total_blocks_hi
) > 0) {
5738 struct imsm_super
*mpb
= super
->anchor
;
5739 mpb
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5742 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
5743 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
5745 dd
->disk
.scsi_id
= __cpu_to_le32(0);
5747 if (st
->update_tail
) {
5748 dd
->next
= super
->disk_mgmt_list
;
5749 super
->disk_mgmt_list
= dd
;
5751 dd
->next
= super
->disks
;
5753 super
->updates_pending
++;
5759 static int remove_from_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
)
5761 struct intel_super
*super
= st
->sb
;
5764 /* remove from super works only in mdmon - for communication
5765 * manager - monitor. Check if communication memory buffer
5768 if (!st
->update_tail
) {
5769 pr_err("shall be used in mdmon context only\n");
5772 dd
= xcalloc(1, sizeof(*dd
));
5773 dd
->major
= dk
->major
;
5774 dd
->minor
= dk
->minor
;
5777 dd
->action
= DISK_REMOVE
;
5779 dd
->next
= super
->disk_mgmt_list
;
5780 super
->disk_mgmt_list
= dd
;
5785 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
);
5788 char buf
[MAX_SECTOR_SIZE
];
5789 struct imsm_super anchor
;
5790 } spare_record
__attribute__ ((aligned(MAX_SECTOR_SIZE
)));
5792 /* spare records have their own family number and do not have any defined raid
5795 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
5797 struct imsm_super
*mpb
= super
->anchor
;
5798 struct imsm_super
*spare
= &spare_record
.anchor
;
5802 spare
->mpb_size
= __cpu_to_le32(sizeof(struct imsm_super
));
5803 spare
->generation_num
= __cpu_to_le32(1UL);
5804 spare
->attributes
= MPB_ATTRIB_CHECKSUM_VERIFY
;
5805 spare
->num_disks
= 1;
5806 spare
->num_raid_devs
= 0;
5807 spare
->cache_size
= mpb
->cache_size
;
5808 spare
->pwr_cycle_count
= __cpu_to_le32(1);
5810 snprintf((char *) spare
->sig
, MAX_SIGNATURE_LENGTH
,
5811 MPB_SIGNATURE MPB_VERSION_RAID0
);
5813 for (d
= super
->disks
; d
; d
= d
->next
) {
5817 spare
->disk
[0] = d
->disk
;
5818 if (__le32_to_cpu(d
->disk
.total_blocks_hi
) > 0)
5819 spare
->attributes
|= MPB_ATTRIB_2TB_DISK
;
5821 if (super
->sector_size
== 4096)
5822 convert_to_4k_imsm_disk(&spare
->disk
[0]);
5824 sum
= __gen_imsm_checksum(spare
);
5825 spare
->family_num
= __cpu_to_le32(sum
);
5826 spare
->orig_family_num
= 0;
5827 sum
= __gen_imsm_checksum(spare
);
5828 spare
->check_sum
= __cpu_to_le32(sum
);
5830 if (store_imsm_mpb(d
->fd
, spare
)) {
5831 pr_err("failed for device %d:%d %s\n",
5832 d
->major
, d
->minor
, strerror(errno
));
5844 static int write_super_imsm(struct supertype
*st
, int doclose
)
5846 struct intel_super
*super
= st
->sb
;
5847 unsigned int sector_size
= super
->sector_size
;
5848 struct imsm_super
*mpb
= super
->anchor
;
5854 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
5856 int clear_migration_record
= 1;
5859 /* 'generation' is incremented everytime the metadata is written */
5860 generation
= __le32_to_cpu(mpb
->generation_num
);
5862 mpb
->generation_num
= __cpu_to_le32(generation
);
5864 /* fix up cases where previous mdadm releases failed to set
5867 if (mpb
->orig_family_num
== 0)
5868 mpb
->orig_family_num
= mpb
->family_num
;
5870 for (d
= super
->disks
; d
; d
= d
->next
) {
5874 mpb
->disk
[d
->index
] = d
->disk
;
5878 for (d
= super
->missing
; d
; d
= d
->next
) {
5879 mpb
->disk
[d
->index
] = d
->disk
;
5882 mpb
->num_disks
= num_disks
;
5883 mpb_size
+= sizeof(struct imsm_disk
) * mpb
->num_disks
;
5885 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
5886 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
5887 struct imsm_dev
*dev2
= get_imsm_dev(super
, i
);
5889 imsm_copy_dev(dev
, dev2
);
5890 mpb_size
+= sizeof_imsm_dev(dev
, 0);
5892 if (is_gen_migration(dev2
))
5893 clear_migration_record
= 0;
5896 bbm_log_size
= get_imsm_bbm_log_size(super
->bbm_log
);
5899 memcpy((void *)mpb
+ mpb_size
, super
->bbm_log
, bbm_log_size
);
5900 mpb
->attributes
|= MPB_ATTRIB_BBM
;
5902 mpb
->attributes
&= ~MPB_ATTRIB_BBM
;
5904 super
->anchor
->bbm_log_size
= __cpu_to_le32(bbm_log_size
);
5905 mpb_size
+= bbm_log_size
;
5906 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
5909 assert(super
->len
== 0 || mpb_size
<= super
->len
);
5912 /* recalculate checksum */
5913 sum
= __gen_imsm_checksum(mpb
);
5914 mpb
->check_sum
= __cpu_to_le32(sum
);
5916 if (super
->clean_migration_record_by_mdmon
) {
5917 clear_migration_record
= 1;
5918 super
->clean_migration_record_by_mdmon
= 0;
5920 if (clear_migration_record
)
5921 memset(super
->migr_rec_buf
, 0,
5922 MIGR_REC_BUF_SECTORS
*sector_size
);
5924 if (sector_size
== 4096)
5925 convert_to_4k(super
);
5927 /* write the mpb for disks that compose raid devices */
5928 for (d
= super
->disks
; d
; d
= d
->next
) {
5929 if (d
->index
< 0 || is_failed(&d
->disk
))
5932 if (clear_migration_record
) {
5933 unsigned long long dsize
;
5935 get_dev_size(d
->fd
, NULL
, &dsize
);
5936 if (lseek64(d
->fd
, dsize
- sector_size
,
5938 if ((unsigned int)write(d
->fd
,
5939 super
->migr_rec_buf
,
5940 MIGR_REC_BUF_SECTORS
*sector_size
) !=
5941 MIGR_REC_BUF_SECTORS
*sector_size
)
5942 perror("Write migr_rec failed");
5946 if (store_imsm_mpb(d
->fd
, mpb
))
5948 "failed for device %d:%d (fd: %d)%s\n",
5950 d
->fd
, strerror(errno
));
5959 return write_super_imsm_spares(super
, doclose
);
5964 static int create_array(struct supertype
*st
, int dev_idx
)
5967 struct imsm_update_create_array
*u
;
5968 struct intel_super
*super
= st
->sb
;
5969 struct imsm_dev
*dev
= get_imsm_dev(super
, dev_idx
);
5970 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
5971 struct disk_info
*inf
;
5972 struct imsm_disk
*disk
;
5975 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0) +
5976 sizeof(*inf
) * map
->num_members
;
5978 u
->type
= update_create_array
;
5979 u
->dev_idx
= dev_idx
;
5980 imsm_copy_dev(&u
->dev
, dev
);
5981 inf
= get_disk_info(u
);
5982 for (i
= 0; i
< map
->num_members
; i
++) {
5983 int idx
= get_imsm_disk_idx(dev
, i
, MAP_X
);
5985 disk
= get_imsm_disk(super
, idx
);
5987 disk
= get_imsm_missing(super
, idx
);
5988 serialcpy(inf
[i
].serial
, disk
->serial
);
5990 append_metadata_update(st
, u
, len
);
5995 static int mgmt_disk(struct supertype
*st
)
5997 struct intel_super
*super
= st
->sb
;
5999 struct imsm_update_add_remove_disk
*u
;
6001 if (!super
->disk_mgmt_list
)
6006 u
->type
= update_add_remove_disk
;
6007 append_metadata_update(st
, u
, len
);
6013 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
6015 static int write_init_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
, int fd
)
6017 struct intel_super
*super
= st
->sb
;
6019 struct ppl_header
*ppl_hdr
;
6022 ret
= posix_memalign(&buf
, 4096, PPL_HEADER_SIZE
);
6024 pr_err("Failed to allocate PPL header buffer\n");
6028 memset(buf
, 0, PPL_HEADER_SIZE
);
6030 memset(ppl_hdr
->reserved
, 0xff, PPL_HDR_RESERVED
);
6031 ppl_hdr
->signature
= __cpu_to_le32(super
->anchor
->orig_family_num
);
6032 ppl_hdr
->checksum
= __cpu_to_le32(~crc32c_le(~0, buf
, PPL_HEADER_SIZE
));
6034 if (lseek64(fd
, info
->ppl_sector
* 512, SEEK_SET
) < 0) {
6036 perror("Failed to seek to PPL header location");
6039 if (!ret
&& write(fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6041 perror("Write PPL header failed");
6051 static int validate_ppl_imsm(struct supertype
*st
, struct mdinfo
*info
,
6052 struct mdinfo
*disk
)
6054 struct intel_super
*super
= st
->sb
;
6058 struct ppl_header
*ppl_hdr
;
6060 struct imsm_dev
*dev
;
6061 struct imsm_map
*map
;
6064 if (disk
->disk
.raid_disk
< 0)
6067 if (posix_memalign(&buf
, 4096, PPL_HEADER_SIZE
)) {
6068 pr_err("Failed to allocate PPL header buffer\n");
6072 dev
= get_imsm_dev(super
, info
->container_member
);
6073 map
= get_imsm_map(dev
, MAP_X
);
6074 idx
= get_imsm_disk_idx(dev
, disk
->disk
.raid_disk
, MAP_X
);
6075 d
= get_imsm_dl_disk(super
, idx
);
6077 if (!d
|| d
->index
< 0 || is_failed(&d
->disk
))
6080 if (lseek64(d
->fd
, info
->ppl_sector
* 512, SEEK_SET
) < 0) {
6081 perror("Failed to seek to PPL header location");
6086 if (read(d
->fd
, buf
, PPL_HEADER_SIZE
) != PPL_HEADER_SIZE
) {
6087 perror("Read PPL header failed");
6094 crc
= __le32_to_cpu(ppl_hdr
->checksum
);
6095 ppl_hdr
->checksum
= 0;
6097 if (crc
!= ~crc32c_le(~0, buf
, PPL_HEADER_SIZE
)) {
6098 dprintf("Wrong PPL header checksum on %s\n",
6103 if (!ret
&& (__le32_to_cpu(ppl_hdr
->signature
) !=
6104 super
->anchor
->orig_family_num
)) {
6105 dprintf("Wrong PPL header signature on %s\n",
6113 if (ret
== 1 && map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
6114 return st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6121 static int write_init_ppl_imsm_all(struct supertype
*st
, struct mdinfo
*info
)
6123 struct intel_super
*super
= st
->sb
;
6127 if (info
->consistency_policy
!= CONSISTENCY_POLICY_PPL
||
6128 info
->array
.level
!= 5)
6131 for (d
= super
->disks
; d
; d
= d
->next
) {
6132 if (d
->index
< 0 || is_failed(&d
->disk
))
6135 ret
= st
->ss
->write_init_ppl(st
, info
, d
->fd
);
6143 static int write_init_super_imsm(struct supertype
*st
)
6145 struct intel_super
*super
= st
->sb
;
6146 int current_vol
= super
->current_vol
;
6150 getinfo_super_imsm(st
, &info
, NULL
);
6152 /* we are done with current_vol reset it to point st at the container */
6153 super
->current_vol
= -1;
6155 if (st
->update_tail
) {
6156 /* queue the recently created array / added disk
6157 * as a metadata update */
6159 /* determine if we are creating a volume or adding a disk */
6160 if (current_vol
< 0) {
6161 /* in the mgmt (add/remove) disk case we are running
6162 * in mdmon context, so don't close fd's
6166 rv
= write_init_ppl_imsm_all(st
, &info
);
6168 rv
= create_array(st
, current_vol
);
6172 for (d
= super
->disks
; d
; d
= d
->next
)
6173 Kill(d
->devname
, NULL
, 0, -1, 1);
6174 if (current_vol
>= 0)
6175 rv
= write_init_ppl_imsm_all(st
, &info
);
6177 rv
= write_super_imsm(st
, 1);
6184 static int store_super_imsm(struct supertype
*st
, int fd
)
6186 struct intel_super
*super
= st
->sb
;
6187 struct imsm_super
*mpb
= super
? super
->anchor
: NULL
;
6193 if (super
->sector_size
== 4096)
6194 convert_to_4k(super
);
6195 return store_imsm_mpb(fd
, mpb
);
6202 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
6203 int layout
, int raiddisks
, int chunk
,
6204 unsigned long long size
,
6205 unsigned long long data_offset
,
6207 unsigned long long *freesize
,
6211 unsigned long long ldsize
;
6212 struct intel_super
*super
;
6215 if (level
!= LEVEL_CONTAINER
)
6220 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
6223 pr_err("imsm: Cannot open %s: %s\n",
6224 dev
, strerror(errno
));
6227 if (!get_dev_size(fd
, dev
, &ldsize
)) {
6232 /* capabilities retrieve could be possible
6233 * note that there is no fd for the disks in array.
6235 super
= alloc_super();
6240 if (!get_dev_sector_size(fd
, NULL
, &super
->sector_size
)) {
6246 rv
= find_intel_hba_capability(fd
, super
, verbose
> 0 ? dev
: NULL
);
6250 fd2devname(fd
, str
);
6251 dprintf("fd: %d %s orom: %p rv: %d raiddisk: %d\n",
6252 fd
, str
, super
->orom
, rv
, raiddisks
);
6254 /* no orom/efi or non-intel hba of the disk */
6261 if (raiddisks
> super
->orom
->tds
) {
6263 pr_err("%d exceeds maximum number of platform supported disks: %d\n",
6264 raiddisks
, super
->orom
->tds
);
6268 if ((super
->orom
->attr
& IMSM_OROM_ATTR_2TB_DISK
) == 0 &&
6269 (ldsize
>> 9) >> 32 > 0) {
6271 pr_err("%s exceeds maximum platform supported size\n", dev
);
6277 *freesize
= avail_size_imsm(st
, ldsize
>> 9, data_offset
);
6283 static unsigned long long find_size(struct extent
*e
, int *idx
, int num_extents
)
6285 const unsigned long long base_start
= e
[*idx
].start
;
6286 unsigned long long end
= base_start
+ e
[*idx
].size
;
6289 if (base_start
== end
)
6293 for (i
= *idx
; i
< num_extents
; i
++) {
6294 /* extend overlapping extents */
6295 if (e
[i
].start
>= base_start
&&
6296 e
[i
].start
<= end
) {
6299 if (e
[i
].start
+ e
[i
].size
> end
)
6300 end
= e
[i
].start
+ e
[i
].size
;
6301 } else if (e
[i
].start
> end
) {
6307 return end
- base_start
;
6310 static unsigned long long merge_extents(struct intel_super
*super
, int sum_extents
)
6312 /* build a composite disk with all known extents and generate a new
6313 * 'maxsize' given the "all disks in an array must share a common start
6314 * offset" constraint
6316 struct extent
*e
= xcalloc(sum_extents
, sizeof(*e
));
6320 unsigned long long pos
;
6321 unsigned long long start
= 0;
6322 unsigned long long maxsize
;
6323 unsigned long reserve
;
6325 /* coalesce and sort all extents. also, check to see if we need to
6326 * reserve space between member arrays
6329 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6332 for (i
= 0; i
< dl
->extent_cnt
; i
++)
6335 qsort(e
, sum_extents
, sizeof(*e
), cmp_extent
);
6340 while (i
< sum_extents
) {
6341 e
[j
].start
= e
[i
].start
;
6342 e
[j
].size
= find_size(e
, &i
, sum_extents
);
6344 if (e
[j
-1].size
== 0)
6353 unsigned long long esize
;
6355 esize
= e
[i
].start
- pos
;
6356 if (esize
>= maxsize
) {
6361 pos
= e
[i
].start
+ e
[i
].size
;
6363 } while (e
[i
-1].size
);
6369 /* FIXME assumes volume at offset 0 is the first volume in a
6372 if (start_extent
> 0)
6373 reserve
= IMSM_RESERVED_SECTORS
; /* gap between raid regions */
6377 if (maxsize
< reserve
)
6380 super
->create_offset
= ~((unsigned long long) 0);
6381 if (start
+ reserve
> super
->create_offset
)
6382 return 0; /* start overflows create_offset */
6383 super
->create_offset
= start
+ reserve
;
6385 return maxsize
- reserve
;
6388 static int is_raid_level_supported(const struct imsm_orom
*orom
, int level
, int raiddisks
)
6390 if (level
< 0 || level
== 6 || level
== 4)
6393 /* if we have an orom prevent invalid raid levels */
6396 case 0: return imsm_orom_has_raid0(orom
);
6399 return imsm_orom_has_raid1e(orom
);
6400 return imsm_orom_has_raid1(orom
) && raiddisks
== 2;
6401 case 10: return imsm_orom_has_raid10(orom
) && raiddisks
== 4;
6402 case 5: return imsm_orom_has_raid5(orom
) && raiddisks
> 2;
6405 return 1; /* not on an Intel RAID platform so anything goes */
6411 active_arrays_by_format(char *name
, char* hba
, struct md_list
**devlist
,
6412 int dpa
, int verbose
)
6414 struct mdstat_ent
*mdstat
= mdstat_read(0, 0);
6415 struct mdstat_ent
*memb
;
6421 for (memb
= mdstat
; memb
; memb
= memb
->next
) {
6422 if (memb
->metadata_version
&&
6423 (strncmp(memb
->metadata_version
, "external:", 9) == 0) &&
6424 (strcmp(&memb
->metadata_version
[9], name
) == 0) &&
6425 !is_subarray(memb
->metadata_version
+9) &&
6427 struct dev_member
*dev
= memb
->members
;
6429 while(dev
&& (fd
< 0)) {
6430 char *path
= xmalloc(strlen(dev
->name
) + strlen("/dev/") + 1);
6431 num
= sprintf(path
, "%s%s", "/dev/", dev
->name
);
6433 fd
= open(path
, O_RDONLY
, 0);
6434 if (num
<= 0 || fd
< 0) {
6435 pr_vrb("Cannot open %s: %s\n",
6436 dev
->name
, strerror(errno
));
6442 if (fd
>= 0 && disk_attached_to_hba(fd
, hba
)) {
6443 struct mdstat_ent
*vol
;
6444 for (vol
= mdstat
; vol
; vol
= vol
->next
) {
6445 if (vol
->active
> 0 &&
6446 vol
->metadata_version
&&
6447 is_container_member(vol
, memb
->devnm
)) {
6452 if (*devlist
&& (found
< dpa
)) {
6453 dv
= xcalloc(1, sizeof(*dv
));
6454 dv
->devname
= xmalloc(strlen(memb
->devnm
) + strlen("/dev/") + 1);
6455 sprintf(dv
->devname
, "%s%s", "/dev/", memb
->devnm
);
6458 dv
->next
= *devlist
;
6466 free_mdstat(mdstat
);
6471 static struct md_list
*
6472 get_loop_devices(void)
6475 struct md_list
*devlist
= NULL
;
6478 for(i
= 0; i
< 12; i
++) {
6479 dv
= xcalloc(1, sizeof(*dv
));
6480 dv
->devname
= xmalloc(40);
6481 sprintf(dv
->devname
, "/dev/loop%d", i
);
6489 static struct md_list
*
6490 get_devices(const char *hba_path
)
6492 struct md_list
*devlist
= NULL
;
6499 devlist
= get_loop_devices();
6502 /* scroll through /sys/dev/block looking for devices attached to
6505 dir
= opendir("/sys/dev/block");
6506 for (ent
= dir
? readdir(dir
) : NULL
; ent
; ent
= readdir(dir
)) {
6511 if (sscanf(ent
->d_name
, "%d:%d", &major
, &minor
) != 2)
6513 path
= devt_to_devpath(makedev(major
, minor
));
6516 if (!path_attached_to_hba(path
, hba_path
)) {
6523 fd
= dev_open(ent
->d_name
, O_RDONLY
);
6525 fd2devname(fd
, buf
);
6528 pr_err("cannot open device: %s\n",
6533 dv
= xcalloc(1, sizeof(*dv
));
6534 dv
->devname
= xstrdup(buf
);
6541 devlist
= devlist
->next
;
6551 count_volumes_list(struct md_list
*devlist
, char *homehost
,
6552 int verbose
, int *found
)
6554 struct md_list
*tmpdev
;
6556 struct supertype
*st
;
6558 /* first walk the list of devices to find a consistent set
6559 * that match the criterea, if that is possible.
6560 * We flag the ones we like with 'used'.
6563 st
= match_metadata_desc_imsm("imsm");
6565 pr_vrb("cannot allocate memory for imsm supertype\n");
6569 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6570 char *devname
= tmpdev
->devname
;
6572 struct supertype
*tst
;
6574 if (tmpdev
->used
> 1)
6576 tst
= dup_super(st
);
6578 pr_vrb("cannot allocate memory for imsm supertype\n");
6581 tmpdev
->container
= 0;
6582 dfd
= dev_open(devname
, O_RDONLY
|O_EXCL
);
6584 dprintf("cannot open device %s: %s\n",
6585 devname
, strerror(errno
));
6587 } else if (fstat(dfd
, &stb
)< 0) {
6589 dprintf("fstat failed for %s: %s\n",
6590 devname
, strerror(errno
));
6592 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
6593 dprintf("%s is not a block device.\n",
6596 } else if (must_be_container(dfd
)) {
6597 struct supertype
*cst
;
6598 cst
= super_by_fd(dfd
, NULL
);
6600 dprintf("cannot recognize container type %s\n",
6603 } else if (tst
->ss
!= st
->ss
) {
6604 dprintf("non-imsm container - ignore it: %s\n",
6607 } else if (!tst
->ss
->load_container
||
6608 tst
->ss
->load_container(tst
, dfd
, NULL
))
6611 tmpdev
->container
= 1;
6614 cst
->ss
->free_super(cst
);
6616 tmpdev
->st_rdev
= stb
.st_rdev
;
6617 if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
6618 dprintf("no RAID superblock on %s\n",
6621 } else if (tst
->ss
->compare_super
== NULL
) {
6622 dprintf("Cannot assemble %s metadata on %s\n",
6623 tst
->ss
->name
, devname
);
6629 if (tmpdev
->used
== 2 || tmpdev
->used
== 4) {
6630 /* Ignore unrecognised devices during auto-assembly */
6635 tst
->ss
->getinfo_super(tst
, &info
, NULL
);
6637 if (st
->minor_version
== -1)
6638 st
->minor_version
= tst
->minor_version
;
6640 if (memcmp(info
.uuid
, uuid_zero
,
6641 sizeof(int[4])) == 0) {
6642 /* this is a floating spare. It cannot define
6643 * an array unless there are no more arrays of
6644 * this type to be found. It can be included
6645 * in an array of this type though.
6651 if (st
->ss
!= tst
->ss
||
6652 st
->minor_version
!= tst
->minor_version
||
6653 st
->ss
->compare_super(st
, tst
) != 0) {
6654 /* Some mismatch. If exactly one array matches this host,
6655 * we can resolve on that one.
6656 * Or, if we are auto assembling, we just ignore the second
6659 dprintf("superblock on %s doesn't match others - assembly aborted\n",
6665 dprintf("found: devname: %s\n", devname
);
6669 tst
->ss
->free_super(tst
);
6673 if ((err
= load_super_imsm_all(st
, -1, &st
->sb
, NULL
, devlist
, 0)) == 0) {
6674 struct mdinfo
*iter
, *head
= st
->ss
->container_content(st
, NULL
);
6675 for (iter
= head
; iter
; iter
= iter
->next
) {
6676 dprintf("content->text_version: %s vol\n",
6677 iter
->text_version
);
6678 if (iter
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
6679 /* do not assemble arrays with unsupported
6681 dprintf("Cannot activate member %s.\n",
6682 iter
->text_version
);
6689 dprintf("No valid super block on device list: err: %d %p\n",
6693 dprintf("no more devices to examine\n");
6696 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
6697 if (tmpdev
->used
== 1 && tmpdev
->found
) {
6699 if (count
< tmpdev
->found
)
6702 count
-= tmpdev
->found
;
6705 if (tmpdev
->used
== 1)
6710 st
->ss
->free_super(st
);
6714 static int __count_volumes(char *hba_path
, int dpa
, int verbose
,
6717 struct sys_dev
*idev
, *intel_devices
= find_intel_devices();
6719 const struct orom_entry
*entry
;
6720 struct devid_list
*dv
, *devid_list
;
6725 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
6726 if (strstr(idev
->path
, hba_path
))
6730 if (!idev
|| !idev
->dev_id
)
6733 entry
= get_orom_entry_by_device_id(idev
->dev_id
);
6735 if (!entry
|| !entry
->devid_list
)
6738 devid_list
= entry
->devid_list
;
6739 for (dv
= devid_list
; dv
; dv
= dv
->next
) {
6740 struct md_list
*devlist
;
6741 struct sys_dev
*device
= NULL
;
6746 device
= device_by_id_and_path(dv
->devid
, hba_path
);
6748 device
= device_by_id(dv
->devid
);
6751 hpath
= device
->path
;
6755 devlist
= get_devices(hpath
);
6756 /* if no intel devices return zero volumes */
6757 if (devlist
== NULL
)
6760 count
+= active_arrays_by_format("imsm", hpath
, &devlist
, dpa
,
6762 dprintf("path: %s active arrays: %d\n", hpath
, count
);
6763 if (devlist
== NULL
)
6767 count
+= count_volumes_list(devlist
,
6771 dprintf("found %d count: %d\n", found
, count
);
6774 dprintf("path: %s total number of volumes: %d\n", hpath
, count
);
6777 struct md_list
*dv
= devlist
;
6778 devlist
= devlist
->next
;
6786 static int count_volumes(struct intel_hba
*hba
, int dpa
, int verbose
)
6790 if (hba
->type
== SYS_DEV_VMD
) {
6791 struct sys_dev
*dev
;
6794 for (dev
= find_intel_devices(); dev
; dev
= dev
->next
) {
6795 if (dev
->type
== SYS_DEV_VMD
)
6796 count
+= __count_volumes(dev
->path
, dpa
,
6801 return __count_volumes(hba
->path
, dpa
, verbose
, 0);
6804 static int imsm_default_chunk(const struct imsm_orom
*orom
)
6806 /* up to 512 if the plaform supports it, otherwise the platform max.
6807 * 128 if no platform detected
6809 int fs
= max(7, orom
? fls(orom
->sss
) : 0);
6811 return min(512, (1 << fs
));
6815 validate_geometry_imsm_orom(struct intel_super
*super
, int level
, int layout
,
6816 int raiddisks
, int *chunk
, unsigned long long size
, int verbose
)
6818 /* check/set platform and metadata limits/defaults */
6819 if (super
->orom
&& raiddisks
> super
->orom
->dpa
) {
6820 pr_vrb("platform supports a maximum of %d disks per array\n",
6825 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
6826 if (!is_raid_level_supported(super
->orom
, level
, raiddisks
)) {
6827 pr_vrb("platform does not support raid%d with %d disk%s\n",
6828 level
, raiddisks
, raiddisks
> 1 ? "s" : "");
6832 if (*chunk
== 0 || *chunk
== UnSet
)
6833 *chunk
= imsm_default_chunk(super
->orom
);
6835 if (super
->orom
&& !imsm_orom_has_chunk(super
->orom
, *chunk
)) {
6836 pr_vrb("platform does not support a chunk size of: %d\n", *chunk
);
6840 if (layout
!= imsm_level_to_layout(level
)) {
6842 pr_vrb("imsm raid 5 only supports the left-asymmetric layout\n");
6843 else if (level
== 10)
6844 pr_vrb("imsm raid 10 only supports the n2 layout\n");
6846 pr_vrb("imsm unknown layout %#x for this raid level %d\n",
6851 if (super
->orom
&& (super
->orom
->attr
& IMSM_OROM_ATTR_2TB
) == 0 &&
6852 (calc_array_size(level
, raiddisks
, layout
, *chunk
, size
) >> 32) > 0) {
6853 pr_vrb("platform does not support a volume size over 2TB\n");
6860 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
6861 * FIX ME add ahci details
6863 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
6864 int layout
, int raiddisks
, int *chunk
,
6865 unsigned long long size
,
6866 unsigned long long data_offset
,
6868 unsigned long long *freesize
,
6872 struct intel_super
*super
= st
->sb
;
6873 struct imsm_super
*mpb
;
6875 unsigned long long pos
= 0;
6876 unsigned long long maxsize
;
6880 /* We must have the container info already read in. */
6884 mpb
= super
->anchor
;
6886 if (!validate_geometry_imsm_orom(super
, level
, layout
, raiddisks
, chunk
, size
, verbose
)) {
6887 pr_err("RAID gemetry validation failed. Cannot proceed with the action(s).\n");
6891 /* General test: make sure there is space for
6892 * 'raiddisks' device extents of size 'size' at a given
6895 unsigned long long minsize
= size
;
6896 unsigned long long start_offset
= MaxSector
;
6899 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
6900 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6905 e
= get_extents(super
, dl
);
6908 unsigned long long esize
;
6909 esize
= e
[i
].start
- pos
;
6910 if (esize
>= minsize
)
6912 if (found
&& start_offset
== MaxSector
) {
6915 } else if (found
&& pos
!= start_offset
) {
6919 pos
= e
[i
].start
+ e
[i
].size
;
6921 } while (e
[i
-1].size
);
6926 if (dcnt
< raiddisks
) {
6928 pr_err("imsm: Not enough devices with space for this array (%d < %d)\n",
6935 /* This device must be a member of the set */
6936 if (stat(dev
, &stb
) < 0)
6938 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
6940 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
6941 if (dl
->major
== (int)major(stb
.st_rdev
) &&
6942 dl
->minor
== (int)minor(stb
.st_rdev
))
6947 pr_err("%s is not in the same imsm set\n", dev
);
6949 } else if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
) {
6950 /* If a volume is present then the current creation attempt
6951 * cannot incorporate new spares because the orom may not
6952 * understand this configuration (all member disks must be
6953 * members of each array in the container).
6955 pr_err("%s is a spare and a volume is already defined for this container\n", dev
);
6956 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6958 } else if (super
->orom
&& mpb
->num_raid_devs
> 0 &&
6959 mpb
->num_disks
!= raiddisks
) {
6960 pr_err("The option-rom requires all member disks to be a member of all volumes\n");
6964 /* retrieve the largest free space block */
6965 e
= get_extents(super
, dl
);
6970 unsigned long long esize
;
6972 esize
= e
[i
].start
- pos
;
6973 if (esize
>= maxsize
)
6975 pos
= e
[i
].start
+ e
[i
].size
;
6977 } while (e
[i
-1].size
);
6982 pr_err("unable to determine free space for: %s\n",
6986 if (maxsize
< size
) {
6988 pr_err("%s not enough space (%llu < %llu)\n",
6989 dev
, maxsize
, size
);
6993 /* count total number of extents for merge */
6995 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
6997 i
+= dl
->extent_cnt
;
6999 maxsize
= merge_extents(super
, i
);
7001 if (!check_env("IMSM_NO_PLATFORM") &&
7002 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7003 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7007 if (maxsize
< size
|| maxsize
== 0) {
7010 pr_err("no free space left on device. Aborting...\n");
7012 pr_err("not enough space to create volume of given size (%llu < %llu). Aborting...\n",
7018 *freesize
= maxsize
;
7021 int count
= count_volumes(super
->hba
,
7022 super
->orom
->dpa
, verbose
);
7023 if (super
->orom
->vphba
<= count
) {
7024 pr_vrb("platform does not support more than %d raid volumes.\n",
7025 super
->orom
->vphba
);
7032 static int imsm_get_free_size(struct supertype
*st
, int raiddisks
,
7033 unsigned long long size
, int chunk
,
7034 unsigned long long *freesize
)
7036 struct intel_super
*super
= st
->sb
;
7037 struct imsm_super
*mpb
= super
->anchor
;
7042 unsigned long long maxsize
;
7043 unsigned long long minsize
;
7047 /* find the largest common start free region of the possible disks */
7051 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
7057 /* don't activate new spares if we are orom constrained
7058 * and there is already a volume active in the container
7060 if (super
->orom
&& dl
->index
< 0 && mpb
->num_raid_devs
)
7063 e
= get_extents(super
, dl
);
7066 for (i
= 1; e
[i
-1].size
; i
++)
7074 maxsize
= merge_extents(super
, extent_cnt
);
7078 minsize
= chunk
* 2;
7080 if (cnt
< raiddisks
||
7081 (super
->orom
&& used
&& used
!= raiddisks
) ||
7082 maxsize
< minsize
||
7084 pr_err("not enough devices with space to create array.\n");
7085 return 0; /* No enough free spaces large enough */
7096 if (!check_env("IMSM_NO_PLATFORM") &&
7097 mpb
->num_raid_devs
> 0 && size
&& size
!= maxsize
) {
7098 pr_err("attempting to create a second volume with size less then remaining space. Aborting...\n");
7102 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7104 dl
->raiddisk
= cnt
++;
7108 dprintf("imsm: imsm_get_free_size() returns : %llu\n", size
);
7113 static int reserve_space(struct supertype
*st
, int raiddisks
,
7114 unsigned long long size
, int chunk
,
7115 unsigned long long *freesize
)
7117 struct intel_super
*super
= st
->sb
;
7122 rv
= imsm_get_free_size(st
, raiddisks
, size
, chunk
, freesize
);
7125 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
7127 dl
->raiddisk
= cnt
++;
7134 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
7135 int raiddisks
, int *chunk
, unsigned long long size
,
7136 unsigned long long data_offset
,
7137 char *dev
, unsigned long long *freesize
,
7138 int consistency_policy
, int verbose
)
7145 * if given unused devices create a container
7146 * if given given devices in a container create a member volume
7148 if (level
== LEVEL_CONTAINER
) {
7149 /* Must be a fresh device to add to a container */
7150 return validate_geometry_imsm_container(st
, level
, layout
,
7160 struct intel_super
*super
= st
->sb
;
7161 if (!validate_geometry_imsm_orom(st
->sb
, level
, layout
,
7162 raiddisks
, chunk
, size
,
7165 /* we are being asked to automatically layout a
7166 * new volume based on the current contents of
7167 * the container. If the the parameters can be
7168 * satisfied reserve_space will record the disks,
7169 * start offset, and size of the volume to be
7170 * created. add_to_super and getinfo_super
7171 * detect when autolayout is in progress.
7173 /* assuming that freesize is always given when array is
7175 if (super
->orom
&& freesize
) {
7177 count
= count_volumes(super
->hba
,
7178 super
->orom
->dpa
, verbose
);
7179 if (super
->orom
->vphba
<= count
) {
7180 pr_vrb("platform does not support more than %d raid volumes.\n",
7181 super
->orom
->vphba
);
7186 return reserve_space(st
, raiddisks
, size
,
7192 /* creating in a given container */
7193 return validate_geometry_imsm_volume(st
, level
, layout
,
7194 raiddisks
, chunk
, size
,
7196 dev
, freesize
, verbose
);
7199 /* This device needs to be a device in an 'imsm' container */
7200 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
7203 pr_err("Cannot create this array on device %s\n",
7208 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
7210 pr_err("Cannot open %s: %s\n",
7211 dev
, strerror(errno
));
7214 /* Well, it is in use by someone, maybe an 'imsm' container. */
7215 cfd
= open_container(fd
);
7219 pr_err("Cannot use %s: It is busy\n",
7223 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
7224 if (sra
&& sra
->array
.major_version
== -1 &&
7225 strcmp(sra
->text_version
, "imsm") == 0)
7229 /* This is a member of a imsm container. Load the container
7230 * and try to create a volume
7232 struct intel_super
*super
;
7234 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, NULL
, 1) == 0) {
7236 strcpy(st
->container_devnm
, fd2devnm(cfd
));
7238 return validate_geometry_imsm_volume(st
, level
, layout
,
7240 size
, data_offset
, dev
,
7247 pr_err("failed container membership check\n");
7253 static void default_geometry_imsm(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
7255 struct intel_super
*super
= st
->sb
;
7257 if (level
&& *level
== UnSet
)
7258 *level
= LEVEL_CONTAINER
;
7260 if (level
&& layout
&& *layout
== UnSet
)
7261 *layout
= imsm_level_to_layout(*level
);
7263 if (chunk
&& (*chunk
== UnSet
|| *chunk
== 0))
7264 *chunk
= imsm_default_chunk(super
->orom
);
7267 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
);
7269 static int kill_subarray_imsm(struct supertype
*st
)
7271 /* remove the subarray currently referenced by ->current_vol */
7273 struct intel_dev
**dp
;
7274 struct intel_super
*super
= st
->sb
;
7275 __u8 current_vol
= super
->current_vol
;
7276 struct imsm_super
*mpb
= super
->anchor
;
7278 if (super
->current_vol
< 0)
7280 super
->current_vol
= -1; /* invalidate subarray cursor */
7282 /* block deletions that would change the uuid of active subarrays
7284 * FIXME when immutable ids are available, but note that we'll
7285 * also need to fixup the invalidated/active subarray indexes in
7288 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7291 if (i
< current_vol
)
7293 sprintf(subarray
, "%u", i
);
7294 if (is_subarray_active(subarray
, st
->devnm
)) {
7295 pr_err("deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
7302 if (st
->update_tail
) {
7303 struct imsm_update_kill_array
*u
= xmalloc(sizeof(*u
));
7305 u
->type
= update_kill_array
;
7306 u
->dev_idx
= current_vol
;
7307 append_metadata_update(st
, u
, sizeof(*u
));
7312 for (dp
= &super
->devlist
; *dp
;)
7313 if ((*dp
)->index
== current_vol
) {
7316 handle_missing(super
, (*dp
)->dev
);
7317 if ((*dp
)->index
> current_vol
)
7322 /* no more raid devices, all active components are now spares,
7323 * but of course failed are still failed
7325 if (--mpb
->num_raid_devs
== 0) {
7328 for (d
= super
->disks
; d
; d
= d
->next
)
7333 super
->updates_pending
++;
7338 static int update_subarray_imsm(struct supertype
*st
, char *subarray
,
7339 char *update
, struct mddev_ident
*ident
)
7341 /* update the subarray currently referenced by ->current_vol */
7342 struct intel_super
*super
= st
->sb
;
7343 struct imsm_super
*mpb
= super
->anchor
;
7345 if (strcmp(update
, "name") == 0) {
7346 char *name
= ident
->name
;
7350 if (is_subarray_active(subarray
, st
->devnm
)) {
7351 pr_err("Unable to update name of active subarray\n");
7355 if (!check_name(super
, name
, 0))
7358 vol
= strtoul(subarray
, &ep
, 10);
7359 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7362 if (st
->update_tail
) {
7363 struct imsm_update_rename_array
*u
= xmalloc(sizeof(*u
));
7365 u
->type
= update_rename_array
;
7367 strncpy((char *) u
->name
, name
, MAX_RAID_SERIAL_LEN
);
7368 u
->name
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7369 append_metadata_update(st
, u
, sizeof(*u
));
7371 struct imsm_dev
*dev
;
7374 dev
= get_imsm_dev(super
, vol
);
7375 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
7376 dev
->volume
[MAX_RAID_SERIAL_LEN
-1] = '\0';
7377 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7378 dev
= get_imsm_dev(super
, i
);
7379 handle_missing(super
, dev
);
7381 super
->updates_pending
++;
7383 } else if (strcmp(update
, "ppl") == 0 ||
7384 strcmp(update
, "no-ppl") == 0) {
7387 int vol
= strtoul(subarray
, &ep
, 10);
7389 if (*ep
!= '\0' || vol
>= super
->anchor
->num_raid_devs
)
7392 if (strcmp(update
, "ppl") == 0)
7393 new_policy
= RWH_DISTRIBUTED
;
7395 new_policy
= RWH_OFF
;
7397 if (st
->update_tail
) {
7398 struct imsm_update_rwh_policy
*u
= xmalloc(sizeof(*u
));
7400 u
->type
= update_rwh_policy
;
7402 u
->new_policy
= new_policy
;
7403 append_metadata_update(st
, u
, sizeof(*u
));
7405 struct imsm_dev
*dev
;
7407 dev
= get_imsm_dev(super
, vol
);
7408 dev
->rwh_policy
= new_policy
;
7409 super
->updates_pending
++;
7416 #endif /* MDASSEMBLE */
7418 static int is_gen_migration(struct imsm_dev
*dev
)
7423 if (!dev
->vol
.migr_state
)
7426 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7432 static int is_rebuilding(struct imsm_dev
*dev
)
7434 struct imsm_map
*migr_map
;
7436 if (!dev
->vol
.migr_state
)
7439 if (migr_type(dev
) != MIGR_REBUILD
)
7442 migr_map
= get_imsm_map(dev
, MAP_1
);
7444 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
7451 static int is_initializing(struct imsm_dev
*dev
)
7453 struct imsm_map
*migr_map
;
7455 if (!dev
->vol
.migr_state
)
7458 if (migr_type(dev
) != MIGR_INIT
)
7461 migr_map
= get_imsm_map(dev
, MAP_1
);
7463 if (migr_map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
7470 static void update_recovery_start(struct intel_super
*super
,
7471 struct imsm_dev
*dev
,
7472 struct mdinfo
*array
)
7474 struct mdinfo
*rebuild
= NULL
;
7478 if (!is_rebuilding(dev
))
7481 /* Find the rebuild target, but punt on the dual rebuild case */
7482 for (d
= array
->devs
; d
; d
= d
->next
)
7483 if (d
->recovery_start
== 0) {
7490 /* (?) none of the disks are marked with
7491 * IMSM_ORD_REBUILD, so assume they are missing and the
7492 * disk_ord_tbl was not correctly updated
7494 dprintf("failed to locate out-of-sync disk\n");
7498 units
= __le32_to_cpu(dev
->vol
.curr_migr_unit
);
7499 rebuild
->recovery_start
= units
* blocks_per_migr_unit(super
, dev
);
7503 static int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
);
7506 static struct mdinfo
*container_content_imsm(struct supertype
*st
, char *subarray
)
7508 /* Given a container loaded by load_super_imsm_all,
7509 * extract information about all the arrays into
7511 * If 'subarray' is given, just extract info about that array.
7513 * For each imsm_dev create an mdinfo, fill it in,
7514 * then look for matching devices in super->disks
7515 * and create appropriate device mdinfo.
7517 struct intel_super
*super
= st
->sb
;
7518 struct imsm_super
*mpb
= super
->anchor
;
7519 struct mdinfo
*rest
= NULL
;
7523 int spare_disks
= 0;
7525 /* do not assemble arrays when not all attributes are supported */
7526 if (imsm_check_attributes(mpb
->attributes
) == 0) {
7528 pr_err("Unsupported attributes in IMSM metadata.Arrays activation is blocked.\n");
7531 /* count spare devices, not used in maps
7533 for (d
= super
->disks
; d
; d
= d
->next
)
7537 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
7538 struct imsm_dev
*dev
;
7539 struct imsm_map
*map
;
7540 struct imsm_map
*map2
;
7541 struct mdinfo
*this;
7549 (i
!= strtoul(subarray
, &ep
, 10) || *ep
!= '\0'))
7552 dev
= get_imsm_dev(super
, i
);
7553 map
= get_imsm_map(dev
, MAP_0
);
7554 map2
= get_imsm_map(dev
, MAP_1
);
7556 /* do not publish arrays that are in the middle of an
7557 * unsupported migration
7559 if (dev
->vol
.migr_state
&&
7560 (migr_type(dev
) == MIGR_STATE_CHANGE
)) {
7561 pr_err("cannot assemble volume '%.16s': unsupported migration in progress\n",
7565 /* do not publish arrays that are not support by controller's
7569 this = xmalloc(sizeof(*this));
7571 super
->current_vol
= i
;
7572 getinfo_super_imsm_volume(st
, this, NULL
);
7575 chunk
= __le16_to_cpu(map
->blocks_per_strip
) >> 1;
7576 /* mdadm does not support all metadata features- set the bit in all arrays state */
7577 if (!validate_geometry_imsm_orom(super
,
7578 get_imsm_raid_level(map
), /* RAID level */
7579 imsm_level_to_layout(get_imsm_raid_level(map
)),
7580 map
->num_members
, /* raid disks */
7581 &chunk
, join_u32(dev
->size_low
, dev
->size_high
),
7583 pr_err("IMSM RAID geometry validation failed. Array %s activation is blocked.\n",
7585 this->array
.state
|=
7586 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7587 (1<<MD_SB_BLOCK_VOLUME
);
7591 /* if array has bad blocks, set suitable bit in all arrays state */
7593 this->array
.state
|=
7594 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
) |
7595 (1<<MD_SB_BLOCK_VOLUME
);
7597 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
7598 unsigned long long recovery_start
;
7599 struct mdinfo
*info_d
;
7606 idx
= get_imsm_disk_idx(dev
, slot
, MAP_0
);
7607 ord
= get_imsm_ord_tbl_ent(dev
, slot
, MAP_X
);
7608 for (d
= super
->disks
; d
; d
= d
->next
)
7609 if (d
->index
== idx
)
7612 recovery_start
= MaxSector
;
7615 if (d
&& is_failed(&d
->disk
))
7617 if (ord
& IMSM_ORD_REBUILD
)
7621 * if we skip some disks the array will be assmebled degraded;
7622 * reset resync start to avoid a dirty-degraded
7623 * situation when performing the intial sync
7625 * FIXME handle dirty degraded
7627 if ((skip
|| recovery_start
== 0) &&
7628 !(dev
->vol
.dirty
& RAIDVOL_DIRTY
))
7629 this->resync_start
= MaxSector
;
7633 info_d
= xcalloc(1, sizeof(*info_d
));
7634 info_d
->next
= this->devs
;
7635 this->devs
= info_d
;
7637 info_d
->disk
.number
= d
->index
;
7638 info_d
->disk
.major
= d
->major
;
7639 info_d
->disk
.minor
= d
->minor
;
7640 info_d
->disk
.raid_disk
= slot
;
7641 info_d
->recovery_start
= recovery_start
;
7643 if (slot
< map2
->num_members
)
7644 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7646 this->array
.spare_disks
++;
7648 if (slot
< map
->num_members
)
7649 info_d
->disk
.state
= (1 << MD_DISK_ACTIVE
);
7651 this->array
.spare_disks
++;
7653 if (info_d
->recovery_start
== MaxSector
)
7654 this->array
.working_disks
++;
7656 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
7657 info_d
->data_offset
= pba_of_lba0(map
);
7659 if (map
->raid_level
== 5) {
7660 info_d
->component_size
=
7661 num_data_stripes(map
) *
7662 map
->blocks_per_strip
;
7663 info_d
->ppl_sector
= this->ppl_sector
;
7664 info_d
->ppl_size
= this->ppl_size
;
7666 info_d
->component_size
= blocks_per_member(map
);
7668 info_d
->consistency_policy
= this->consistency_policy
;
7670 info_d
->bb
.supported
= 1;
7671 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
),
7672 info_d
->data_offset
,
7673 info_d
->component_size
,
7676 /* now that the disk list is up-to-date fixup recovery_start */
7677 update_recovery_start(super
, dev
, this);
7678 this->array
.spare_disks
+= spare_disks
;
7681 /* check for reshape */
7682 if (this->reshape_active
== 1)
7683 recover_backup_imsm(st
, this);
7691 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
,
7692 int failed
, int look_in_map
)
7694 struct imsm_map
*map
;
7696 map
= get_imsm_map(dev
, look_in_map
);
7699 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
7700 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
7702 switch (get_imsm_raid_level(map
)) {
7704 return IMSM_T_STATE_FAILED
;
7707 if (failed
< map
->num_members
)
7708 return IMSM_T_STATE_DEGRADED
;
7710 return IMSM_T_STATE_FAILED
;
7715 * check to see if any mirrors have failed, otherwise we
7716 * are degraded. Even numbered slots are mirrored on
7720 /* gcc -Os complains that this is unused */
7721 int insync
= insync
;
7723 for (i
= 0; i
< map
->num_members
; i
++) {
7724 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
, MAP_X
);
7725 int idx
= ord_to_idx(ord
);
7726 struct imsm_disk
*disk
;
7728 /* reset the potential in-sync count on even-numbered
7729 * slots. num_copies is always 2 for imsm raid10
7734 disk
= get_imsm_disk(super
, idx
);
7735 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7738 /* no in-sync disks left in this mirror the
7742 return IMSM_T_STATE_FAILED
;
7745 return IMSM_T_STATE_DEGRADED
;
7749 return IMSM_T_STATE_DEGRADED
;
7751 return IMSM_T_STATE_FAILED
;
7757 return map
->map_state
;
7760 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
,
7765 struct imsm_disk
*disk
;
7766 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
7767 struct imsm_map
*prev
= get_imsm_map(dev
, MAP_1
);
7768 struct imsm_map
*map_for_loop
;
7773 /* at the beginning of migration we set IMSM_ORD_REBUILD on
7774 * disks that are being rebuilt. New failures are recorded to
7775 * map[0]. So we look through all the disks we started with and
7776 * see if any failures are still present, or if any new ones
7780 if (prev
&& (map
->num_members
< prev
->num_members
))
7781 map_for_loop
= prev
;
7783 for (i
= 0; i
< map_for_loop
->num_members
; i
++) {
7785 /* when MAP_X is passed both maps failures are counted
7788 (look_in_map
== MAP_1
|| look_in_map
== MAP_X
) &&
7789 i
< prev
->num_members
) {
7790 ord
= __le32_to_cpu(prev
->disk_ord_tbl
[i
]);
7791 idx_1
= ord_to_idx(ord
);
7793 disk
= get_imsm_disk(super
, idx_1
);
7794 if (!disk
|| is_failed(disk
) || ord
& IMSM_ORD_REBUILD
)
7797 if ((look_in_map
== MAP_0
|| look_in_map
== MAP_X
) &&
7798 i
< map
->num_members
) {
7799 ord
= __le32_to_cpu(map
->disk_ord_tbl
[i
]);
7800 idx
= ord_to_idx(ord
);
7803 disk
= get_imsm_disk(super
, idx
);
7804 if (!disk
|| is_failed(disk
) ||
7805 ord
& IMSM_ORD_REBUILD
)
7815 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
7818 struct intel_super
*super
= c
->sb
;
7819 struct imsm_super
*mpb
= super
->anchor
;
7820 struct imsm_update_prealloc_bb_mem u
;
7822 if (atoi(inst
) >= mpb
->num_raid_devs
) {
7823 pr_err("subarry index %d, out of range\n", atoi(inst
));
7827 dprintf("imsm: open_new %s\n", inst
);
7828 a
->info
.container_member
= atoi(inst
);
7830 u
.type
= update_prealloc_badblocks_mem
;
7831 imsm_update_metadata_locally(c
, &u
, sizeof(u
));
7836 static int is_resyncing(struct imsm_dev
*dev
)
7838 struct imsm_map
*migr_map
;
7840 if (!dev
->vol
.migr_state
)
7843 if (migr_type(dev
) == MIGR_INIT
||
7844 migr_type(dev
) == MIGR_REPAIR
)
7847 if (migr_type(dev
) == MIGR_GEN_MIGR
)
7850 migr_map
= get_imsm_map(dev
, MAP_1
);
7852 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
&&
7853 dev
->vol
.migr_type
!= MIGR_GEN_MIGR
)
7859 /* return true if we recorded new information */
7860 static int mark_failure(struct intel_super
*super
,
7861 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7865 struct imsm_map
*map
;
7866 char buf
[MAX_RAID_SERIAL_LEN
+3];
7867 unsigned int len
, shift
= 0;
7869 /* new failures are always set in map[0] */
7870 map
= get_imsm_map(dev
, MAP_0
);
7872 slot
= get_imsm_disk_slot(map
, idx
);
7876 ord
= __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
7877 if (is_failed(disk
) && (ord
& IMSM_ORD_REBUILD
))
7880 memcpy(buf
, disk
->serial
, MAX_RAID_SERIAL_LEN
);
7881 buf
[MAX_RAID_SERIAL_LEN
] = '\000';
7883 if ((len
= strlen(buf
)) >= MAX_RAID_SERIAL_LEN
)
7884 shift
= len
- MAX_RAID_SERIAL_LEN
+ 1;
7885 strncpy((char *)disk
->serial
, &buf
[shift
], MAX_RAID_SERIAL_LEN
);
7887 disk
->status
|= FAILED_DISK
;
7888 set_imsm_ord_tbl_ent(map
, slot
, idx
| IMSM_ORD_REBUILD
);
7889 /* mark failures in second map if second map exists and this disk
7891 * This is valid for migration, initialization and rebuild
7893 if (dev
->vol
.migr_state
) {
7894 struct imsm_map
*map2
= get_imsm_map(dev
, MAP_1
);
7895 int slot2
= get_imsm_disk_slot(map2
, idx
);
7897 if (slot2
< map2
->num_members
&& slot2
>= 0)
7898 set_imsm_ord_tbl_ent(map2
, slot2
,
7899 idx
| IMSM_ORD_REBUILD
);
7901 if (map
->failed_disk_num
== 0xff)
7902 map
->failed_disk_num
= slot
;
7904 clear_disk_badblocks(super
->bbm_log
, ord_to_idx(ord
));
7909 static void mark_missing(struct intel_super
*super
,
7910 struct imsm_dev
*dev
, struct imsm_disk
*disk
, int idx
)
7912 mark_failure(super
, dev
, disk
, idx
);
7914 if (disk
->scsi_id
== __cpu_to_le32(~(__u32
)0))
7917 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
7918 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
7921 static void handle_missing(struct intel_super
*super
, struct imsm_dev
*dev
)
7925 if (!super
->missing
)
7928 /* When orom adds replacement for missing disk it does
7929 * not remove entry of missing disk, but just updates map with
7930 * new added disk. So it is not enough just to test if there is
7931 * any missing disk, we have to look if there are any failed disks
7932 * in map to stop migration */
7934 dprintf("imsm: mark missing\n");
7935 /* end process for initialization and rebuild only
7937 if (is_gen_migration(dev
) == 0) {
7941 failed
= imsm_count_failed(super
, dev
, MAP_0
);
7942 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
7945 end_migration(dev
, super
, map_state
);
7947 for (dl
= super
->missing
; dl
; dl
= dl
->next
)
7948 mark_missing(super
, dev
, &dl
->disk
, dl
->index
);
7949 super
->updates_pending
++;
7952 static unsigned long long imsm_set_array_size(struct imsm_dev
*dev
,
7955 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
7956 unsigned long long array_blocks
;
7957 struct imsm_map
*map
;
7959 if (used_disks
== 0) {
7960 /* when problems occures
7961 * return current array_blocks value
7963 array_blocks
= __le32_to_cpu(dev
->size_high
);
7964 array_blocks
= array_blocks
<< 32;
7965 array_blocks
+= __le32_to_cpu(dev
->size_low
);
7967 return array_blocks
;
7970 /* set array size in metadata
7972 if (new_size
<= 0) {
7973 /* OLCE size change is caused by added disks
7975 map
= get_imsm_map(dev
, MAP_0
);
7976 array_blocks
= blocks_per_member(map
) * used_disks
;
7978 /* Online Volume Size Change
7979 * Using available free space
7981 array_blocks
= new_size
;
7984 /* round array size down to closest MB
7986 array_blocks
= (array_blocks
>> SECT_PER_MB_SHIFT
) << SECT_PER_MB_SHIFT
;
7987 dev
->size_low
= __cpu_to_le32((__u32
)array_blocks
);
7988 dev
->size_high
= __cpu_to_le32((__u32
)(array_blocks
>> 32));
7990 return array_blocks
;
7993 static void imsm_set_disk(struct active_array
*a
, int n
, int state
);
7995 static void imsm_progress_container_reshape(struct intel_super
*super
)
7997 /* if no device has a migr_state, but some device has a
7998 * different number of members than the previous device, start
7999 * changing the number of devices in this device to match
8002 struct imsm_super
*mpb
= super
->anchor
;
8003 int prev_disks
= -1;
8007 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8008 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
8009 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8010 struct imsm_map
*map2
;
8011 int prev_num_members
;
8013 if (dev
->vol
.migr_state
)
8016 if (prev_disks
== -1)
8017 prev_disks
= map
->num_members
;
8018 if (prev_disks
== map
->num_members
)
8021 /* OK, this array needs to enter reshape mode.
8022 * i.e it needs a migr_state
8025 copy_map_size
= sizeof_imsm_map(map
);
8026 prev_num_members
= map
->num_members
;
8027 map
->num_members
= prev_disks
;
8028 dev
->vol
.migr_state
= 1;
8029 dev
->vol
.curr_migr_unit
= 0;
8030 set_migr_type(dev
, MIGR_GEN_MIGR
);
8031 for (i
= prev_num_members
;
8032 i
< map
->num_members
; i
++)
8033 set_imsm_ord_tbl_ent(map
, i
, i
);
8034 map2
= get_imsm_map(dev
, MAP_1
);
8035 /* Copy the current map */
8036 memcpy(map2
, map
, copy_map_size
);
8037 map2
->num_members
= prev_num_members
;
8039 imsm_set_array_size(dev
, -1);
8040 super
->clean_migration_record_by_mdmon
= 1;
8041 super
->updates_pending
++;
8045 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
8046 * states are handled in imsm_set_disk() with one exception, when a
8047 * resync is stopped due to a new failure this routine will set the
8048 * 'degraded' state for the array.
8050 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
8052 int inst
= a
->info
.container_member
;
8053 struct intel_super
*super
= a
->container
->sb
;
8054 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8055 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8056 int failed
= imsm_count_failed(super
, dev
, MAP_0
);
8057 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8058 __u32 blocks_per_unit
;
8060 if (dev
->vol
.migr_state
&&
8061 dev
->vol
.migr_type
== MIGR_GEN_MIGR
) {
8062 /* array state change is blocked due to reshape action
8064 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
8065 * - finish the reshape (if last_checkpoint is big and action != reshape)
8066 * - update curr_migr_unit
8068 if (a
->curr_action
== reshape
) {
8069 /* still reshaping, maybe update curr_migr_unit */
8070 goto mark_checkpoint
;
8072 if (a
->last_checkpoint
== 0 && a
->prev_action
== reshape
) {
8073 /* for some reason we aborted the reshape.
8075 * disable automatic metadata rollback
8076 * user action is required to recover process
8079 struct imsm_map
*map2
=
8080 get_imsm_map(dev
, MAP_1
);
8081 dev
->vol
.migr_state
= 0;
8082 set_migr_type(dev
, 0);
8083 dev
->vol
.curr_migr_unit
= 0;
8085 sizeof_imsm_map(map2
));
8086 super
->updates_pending
++;
8089 if (a
->last_checkpoint
>= a
->info
.component_size
) {
8090 unsigned long long array_blocks
;
8094 used_disks
= imsm_num_data_members(dev
, MAP_0
);
8095 if (used_disks
> 0) {
8097 blocks_per_member(map
) *
8099 /* round array size down to closest MB
8101 array_blocks
= (array_blocks
8102 >> SECT_PER_MB_SHIFT
)
8103 << SECT_PER_MB_SHIFT
;
8104 a
->info
.custom_array_size
= array_blocks
;
8105 /* encourage manager to update array
8109 a
->check_reshape
= 1;
8111 /* finalize online capacity expansion/reshape */
8112 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8114 mdi
->disk
.raid_disk
,
8117 imsm_progress_container_reshape(super
);
8122 /* before we activate this array handle any missing disks */
8123 if (consistent
== 2)
8124 handle_missing(super
, dev
);
8126 if (consistent
== 2 &&
8127 (!is_resync_complete(&a
->info
) ||
8128 map_state
!= IMSM_T_STATE_NORMAL
||
8129 dev
->vol
.migr_state
))
8132 if (is_resync_complete(&a
->info
)) {
8133 /* complete intialization / resync,
8134 * recovery and interrupted recovery is completed in
8137 if (is_resyncing(dev
)) {
8138 dprintf("imsm: mark resync done\n");
8139 end_migration(dev
, super
, map_state
);
8140 super
->updates_pending
++;
8141 a
->last_checkpoint
= 0;
8143 } else if ((!is_resyncing(dev
) && !failed
) &&
8144 (imsm_reshape_blocks_arrays_changes(super
) == 0)) {
8145 /* mark the start of the init process if nothing is failed */
8146 dprintf("imsm: mark resync start\n");
8147 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
)
8148 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_INIT
);
8150 migrate(dev
, super
, IMSM_T_STATE_NORMAL
, MIGR_REPAIR
);
8151 super
->updates_pending
++;
8155 /* skip checkpointing for general migration,
8156 * it is controlled in mdadm
8158 if (is_gen_migration(dev
))
8159 goto skip_mark_checkpoint
;
8161 /* check if we can update curr_migr_unit from resync_start, recovery_start */
8162 blocks_per_unit
= blocks_per_migr_unit(super
, dev
);
8163 if (blocks_per_unit
) {
8167 units
= a
->last_checkpoint
/ blocks_per_unit
;
8170 /* check that we did not overflow 32-bits, and that
8171 * curr_migr_unit needs updating
8173 if (units32
== units
&&
8175 __le32_to_cpu(dev
->vol
.curr_migr_unit
) != units32
) {
8176 dprintf("imsm: mark checkpoint (%u)\n", units32
);
8177 dev
->vol
.curr_migr_unit
= __cpu_to_le32(units32
);
8178 super
->updates_pending
++;
8182 skip_mark_checkpoint
:
8183 /* mark dirty / clean */
8184 if (((dev
->vol
.dirty
& RAIDVOL_DIRTY
) && consistent
) ||
8185 (!(dev
->vol
.dirty
& RAIDVOL_DIRTY
) && !consistent
)) {
8186 dprintf("imsm: mark '%s'\n", consistent
? "clean" : "dirty");
8188 dev
->vol
.dirty
= RAIDVOL_CLEAN
;
8190 dev
->vol
.dirty
= RAIDVOL_DIRTY
;
8191 if (dev
->rwh_policy
== RWH_DISTRIBUTED
)
8192 dev
->vol
.dirty
|= RAIDVOL_DSRECORD_VALID
;
8194 super
->updates_pending
++;
8200 static int imsm_disk_slot_to_ord(struct active_array
*a
, int slot
)
8202 int inst
= a
->info
.container_member
;
8203 struct intel_super
*super
= a
->container
->sb
;
8204 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8205 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8207 if (slot
> map
->num_members
) {
8208 pr_err("imsm: imsm_disk_slot_to_ord %d out of range 0..%d\n",
8209 slot
, map
->num_members
- 1);
8216 return get_imsm_ord_tbl_ent(dev
, slot
, MAP_0
);
8219 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
8221 int inst
= a
->info
.container_member
;
8222 struct intel_super
*super
= a
->container
->sb
;
8223 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8224 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8225 struct imsm_disk
*disk
;
8227 int recovery_not_finished
= 0;
8232 ord
= imsm_disk_slot_to_ord(a
, n
);
8236 dprintf("imsm: set_disk %d:%x\n", n
, state
);
8237 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
8239 /* check for new failures */
8240 if (state
& DS_FAULTY
) {
8241 if (mark_failure(super
, dev
, disk
, ord_to_idx(ord
)))
8242 super
->updates_pending
++;
8245 /* check if in_sync */
8246 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
&& is_rebuilding(dev
)) {
8247 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8249 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
8250 super
->updates_pending
++;
8253 failed
= imsm_count_failed(super
, dev
, MAP_0
);
8254 map_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
8256 /* check if recovery complete, newly degraded, or failed */
8257 dprintf("imsm: Detected transition to state ");
8258 switch (map_state
) {
8259 case IMSM_T_STATE_NORMAL
: /* transition to normal state */
8260 dprintf("normal: ");
8261 if (is_rebuilding(dev
)) {
8262 dprintf_cont("while rebuilding");
8263 /* check if recovery is really finished */
8264 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
8265 if (mdi
->recovery_start
!= MaxSector
) {
8266 recovery_not_finished
= 1;
8269 if (recovery_not_finished
) {
8271 dprintf("Rebuild has not finished yet, state not changed");
8272 if (a
->last_checkpoint
< mdi
->recovery_start
) {
8273 a
->last_checkpoint
= mdi
->recovery_start
;
8274 super
->updates_pending
++;
8278 end_migration(dev
, super
, map_state
);
8279 map
= get_imsm_map(dev
, MAP_0
);
8280 map
->failed_disk_num
= ~0;
8281 super
->updates_pending
++;
8282 a
->last_checkpoint
= 0;
8285 if (is_gen_migration(dev
)) {
8286 dprintf_cont("while general migration");
8287 if (a
->last_checkpoint
>= a
->info
.component_size
)
8288 end_migration(dev
, super
, map_state
);
8290 map
->map_state
= map_state
;
8291 map
= get_imsm_map(dev
, MAP_0
);
8292 map
->failed_disk_num
= ~0;
8293 super
->updates_pending
++;
8297 case IMSM_T_STATE_DEGRADED
: /* transition to degraded state */
8298 dprintf_cont("degraded: ");
8299 if (map
->map_state
!= map_state
&& !dev
->vol
.migr_state
) {
8300 dprintf_cont("mark degraded");
8301 map
->map_state
= map_state
;
8302 super
->updates_pending
++;
8303 a
->last_checkpoint
= 0;
8306 if (is_rebuilding(dev
)) {
8307 dprintf_cont("while rebuilding.");
8308 if (map
->map_state
!= map_state
) {
8309 dprintf_cont(" Map state change");
8310 end_migration(dev
, super
, map_state
);
8311 super
->updates_pending
++;
8315 if (is_gen_migration(dev
)) {
8316 dprintf_cont("while general migration");
8317 if (a
->last_checkpoint
>= a
->info
.component_size
)
8318 end_migration(dev
, super
, map_state
);
8320 map
->map_state
= map_state
;
8321 manage_second_map(super
, dev
);
8323 super
->updates_pending
++;
8326 if (is_initializing(dev
)) {
8327 dprintf_cont("while initialization.");
8328 map
->map_state
= map_state
;
8329 super
->updates_pending
++;
8333 case IMSM_T_STATE_FAILED
: /* transition to failed state */
8334 dprintf_cont("failed: ");
8335 if (is_gen_migration(dev
)) {
8336 dprintf_cont("while general migration");
8337 map
->map_state
= map_state
;
8338 super
->updates_pending
++;
8341 if (map
->map_state
!= map_state
) {
8342 dprintf_cont("mark failed");
8343 end_migration(dev
, super
, map_state
);
8344 super
->updates_pending
++;
8345 a
->last_checkpoint
= 0;
8350 dprintf_cont("state %i\n", map_state
);
8355 static int store_imsm_mpb(int fd
, struct imsm_super
*mpb
)
8358 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
8359 unsigned long long dsize
;
8360 unsigned long long sectors
;
8361 unsigned int sector_size
;
8363 get_dev_sector_size(fd
, NULL
, §or_size
);
8364 get_dev_size(fd
, NULL
, &dsize
);
8366 if (mpb_size
> sector_size
) {
8367 /* -1 to account for anchor */
8368 sectors
= mpb_sectors(mpb
, sector_size
) - 1;
8370 /* write the extended mpb to the sectors preceeding the anchor */
8371 if (lseek64(fd
, dsize
- (sector_size
* (2 + sectors
)),
8375 if ((unsigned long long)write(fd
, buf
+ sector_size
,
8376 sector_size
* sectors
) != sector_size
* sectors
)
8380 /* first block is stored on second to last sector of the disk */
8381 if (lseek64(fd
, dsize
- (sector_size
* 2), SEEK_SET
) < 0)
8384 if ((unsigned int)write(fd
, buf
, sector_size
) != sector_size
)
8390 static void imsm_sync_metadata(struct supertype
*container
)
8392 struct intel_super
*super
= container
->sb
;
8394 dprintf("sync metadata: %d\n", super
->updates_pending
);
8395 if (!super
->updates_pending
)
8398 write_super_imsm(container
, 0);
8400 super
->updates_pending
= 0;
8403 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
8405 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8406 int i
= get_imsm_disk_idx(dev
, idx
, MAP_X
);
8409 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8413 if (dl
&& is_failed(&dl
->disk
))
8417 dprintf("found %x:%x\n", dl
->major
, dl
->minor
);
8422 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
,
8423 struct active_array
*a
, int activate_new
,
8424 struct mdinfo
*additional_test_list
)
8426 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
8427 int idx
= get_imsm_disk_idx(dev
, slot
, MAP_X
);
8428 struct imsm_super
*mpb
= super
->anchor
;
8429 struct imsm_map
*map
;
8430 unsigned long long pos
;
8435 __u32 array_start
= 0;
8436 __u32 array_end
= 0;
8438 struct mdinfo
*test_list
;
8440 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8441 /* If in this array, skip */
8442 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8443 if (d
->state_fd
>= 0 &&
8444 d
->disk
.major
== dl
->major
&&
8445 d
->disk
.minor
== dl
->minor
) {
8446 dprintf("%x:%x already in array\n",
8447 dl
->major
, dl
->minor
);
8452 test_list
= additional_test_list
;
8454 if (test_list
->disk
.major
== dl
->major
&&
8455 test_list
->disk
.minor
== dl
->minor
) {
8456 dprintf("%x:%x already in additional test list\n",
8457 dl
->major
, dl
->minor
);
8460 test_list
= test_list
->next
;
8465 /* skip in use or failed drives */
8466 if (is_failed(&dl
->disk
) || idx
== dl
->index
||
8468 dprintf("%x:%x status (failed: %d index: %d)\n",
8469 dl
->major
, dl
->minor
, is_failed(&dl
->disk
), idx
);
8473 /* skip pure spares when we are looking for partially
8474 * assimilated drives
8476 if (dl
->index
== -1 && !activate_new
)
8479 /* Does this unused device have the requisite free space?
8480 * It needs to be able to cover all member volumes
8482 ex
= get_extents(super
, dl
);
8484 dprintf("cannot get extents\n");
8487 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
8488 dev
= get_imsm_dev(super
, i
);
8489 map
= get_imsm_map(dev
, MAP_0
);
8491 /* check if this disk is already a member of
8494 if (get_imsm_disk_slot(map
, dl
->index
) >= 0)
8500 array_start
= pba_of_lba0(map
);
8501 array_end
= array_start
+
8502 blocks_per_member(map
) - 1;
8505 /* check that we can start at pba_of_lba0 with
8506 * blocks_per_member of space
8508 if (array_start
>= pos
&& array_end
< ex
[j
].start
) {
8512 pos
= ex
[j
].start
+ ex
[j
].size
;
8514 } while (ex
[j
-1].size
);
8521 if (i
< mpb
->num_raid_devs
) {
8522 dprintf("%x:%x does not have %u to %u available\n",
8523 dl
->major
, dl
->minor
, array_start
, array_end
);
8533 static int imsm_rebuild_allowed(struct supertype
*cont
, int dev_idx
, int failed
)
8535 struct imsm_dev
*dev2
;
8536 struct imsm_map
*map
;
8542 dev2
= get_imsm_dev(cont
->sb
, dev_idx
);
8544 state
= imsm_check_degraded(cont
->sb
, dev2
, failed
, MAP_0
);
8545 if (state
== IMSM_T_STATE_FAILED
) {
8546 map
= get_imsm_map(dev2
, MAP_0
);
8549 for (slot
= 0; slot
< map
->num_members
; slot
++) {
8551 * Check if failed disks are deleted from intel
8552 * disk list or are marked to be deleted
8554 idx
= get_imsm_disk_idx(dev2
, slot
, MAP_X
);
8555 idisk
= get_imsm_dl_disk(cont
->sb
, idx
);
8557 * Do not rebuild the array if failed disks
8558 * from failed sub-array are not removed from
8562 is_failed(&idisk
->disk
) &&
8563 (idisk
->action
!= DISK_REMOVE
))
8571 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
8572 struct metadata_update
**updates
)
8575 * Find a device with unused free space and use it to replace a
8576 * failed/vacant region in an array. We replace failed regions one a
8577 * array at a time. The result is that a new spare disk will be added
8578 * to the first failed array and after the monitor has finished
8579 * propagating failures the remainder will be consumed.
8581 * FIXME add a capability for mdmon to request spares from another
8585 struct intel_super
*super
= a
->container
->sb
;
8586 int inst
= a
->info
.container_member
;
8587 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
8588 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8589 int failed
= a
->info
.array
.raid_disks
;
8590 struct mdinfo
*rv
= NULL
;
8593 struct metadata_update
*mu
;
8595 struct imsm_update_activate_spare
*u
;
8600 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
8601 if ((d
->curr_state
& DS_FAULTY
) &&
8603 /* wait for Removal to happen */
8605 if (d
->state_fd
>= 0)
8609 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
8610 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
8612 if (imsm_reshape_blocks_arrays_changes(super
))
8615 /* Cannot activate another spare if rebuild is in progress already
8617 if (is_rebuilding(dev
)) {
8618 dprintf("imsm: No spare activation allowed. Rebuild in progress already.\n");
8622 if (a
->info
.array
.level
== 4)
8623 /* No repair for takeovered array
8624 * imsm doesn't support raid4
8628 if (imsm_check_degraded(super
, dev
, failed
, MAP_0
) !=
8629 IMSM_T_STATE_DEGRADED
)
8632 if (get_imsm_map(dev
, MAP_0
)->map_state
== IMSM_T_STATE_UNINITIALIZED
) {
8633 dprintf("imsm: No spare activation allowed. Volume is not initialized.\n");
8638 * If there are any failed disks check state of the other volume.
8639 * Block rebuild if the another one is failed until failed disks
8640 * are removed from container.
8643 dprintf("found failed disks in %.*s, check if there anotherfailed sub-array.\n",
8644 MAX_RAID_SERIAL_LEN
, dev
->volume
);
8645 /* check if states of the other volumes allow for rebuild */
8646 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
8648 allowed
= imsm_rebuild_allowed(a
->container
,
8656 /* For each slot, if it is not working, find a spare */
8657 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
8658 for (d
= a
->info
.devs
; d
; d
= d
->next
)
8659 if (d
->disk
.raid_disk
== i
)
8661 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
8662 if (d
&& (d
->state_fd
>= 0))
8666 * OK, this device needs recovery. Try to re-add the
8667 * previous occupant of this slot, if this fails see if
8668 * we can continue the assimilation of a spare that was
8669 * partially assimilated, finally try to activate a new
8672 dl
= imsm_readd(super
, i
, a
);
8674 dl
= imsm_add_spare(super
, i
, a
, 0, rv
);
8676 dl
= imsm_add_spare(super
, i
, a
, 1, rv
);
8680 /* found a usable disk with enough space */
8681 di
= xcalloc(1, sizeof(*di
));
8683 /* dl->index will be -1 in the case we are activating a
8684 * pristine spare. imsm_process_update() will create a
8685 * new index in this case. Once a disk is found to be
8686 * failed in all member arrays it is kicked from the
8689 di
->disk
.number
= dl
->index
;
8691 /* (ab)use di->devs to store a pointer to the device
8694 di
->devs
= (struct mdinfo
*) dl
;
8696 di
->disk
.raid_disk
= i
;
8697 di
->disk
.major
= dl
->major
;
8698 di
->disk
.minor
= dl
->minor
;
8700 di
->recovery_start
= 0;
8701 di
->data_offset
= pba_of_lba0(map
);
8702 di
->component_size
= a
->info
.component_size
;
8703 di
->container_member
= inst
;
8704 di
->bb
.supported
= 1;
8705 if (dev
->rwh_policy
== RWH_DISTRIBUTED
) {
8706 di
->consistency_policy
= CONSISTENCY_POLICY_PPL
;
8707 di
->ppl_sector
= get_ppl_sector(super
, inst
);
8708 di
->ppl_size
= (PPL_HEADER_SIZE
+ PPL_ENTRY_SPACE
) >> 9;
8710 super
->random
= random32();
8714 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
8715 i
, di
->data_offset
);
8719 /* No spares found */
8721 /* Now 'rv' has a list of devices to return.
8722 * Create a metadata_update record to update the
8723 * disk_ord_tbl for the array
8725 mu
= xmalloc(sizeof(*mu
));
8726 mu
->buf
= xcalloc(num_spares
,
8727 sizeof(struct imsm_update_activate_spare
));
8729 mu
->space_list
= NULL
;
8730 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
8731 mu
->next
= *updates
;
8732 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
8734 for (di
= rv
; di
; di
= di
->next
) {
8735 u
->type
= update_activate_spare
;
8736 u
->dl
= (struct dl
*) di
->devs
;
8738 u
->slot
= di
->disk
.raid_disk
;
8749 static int disks_overlap(struct intel_super
*super
, int idx
, struct imsm_update_create_array
*u
)
8751 struct imsm_dev
*dev
= get_imsm_dev(super
, idx
);
8752 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8753 struct imsm_map
*new_map
= get_imsm_map(&u
->dev
, MAP_0
);
8754 struct disk_info
*inf
= get_disk_info(u
);
8755 struct imsm_disk
*disk
;
8759 for (i
= 0; i
< map
->num_members
; i
++) {
8760 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
, MAP_X
));
8761 for (j
= 0; j
< new_map
->num_members
; j
++)
8762 if (serialcmp(disk
->serial
, inf
[j
].serial
) == 0)
8769 static struct dl
*get_disk_super(struct intel_super
*super
, int major
, int minor
)
8773 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
8774 if (dl
->major
== major
&& dl
->minor
== minor
)
8779 static int remove_disk_super(struct intel_super
*super
, int major
, int minor
)
8785 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
8786 if (dl
->major
== major
&& dl
->minor
== minor
) {
8789 prev
->next
= dl
->next
;
8791 super
->disks
= dl
->next
;
8793 __free_imsm_disk(dl
);
8794 dprintf("removed %x:%x\n", major
, minor
);
8802 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
);
8804 static int add_remove_disk_update(struct intel_super
*super
)
8806 int check_degraded
= 0;
8809 /* add/remove some spares to/from the metadata/contrainer */
8810 while (super
->disk_mgmt_list
) {
8811 struct dl
*disk_cfg
;
8813 disk_cfg
= super
->disk_mgmt_list
;
8814 super
->disk_mgmt_list
= disk_cfg
->next
;
8815 disk_cfg
->next
= NULL
;
8817 if (disk_cfg
->action
== DISK_ADD
) {
8818 disk_cfg
->next
= super
->disks
;
8819 super
->disks
= disk_cfg
;
8821 dprintf("added %x:%x\n",
8822 disk_cfg
->major
, disk_cfg
->minor
);
8823 } else if (disk_cfg
->action
== DISK_REMOVE
) {
8824 dprintf("Disk remove action processed: %x.%x\n",
8825 disk_cfg
->major
, disk_cfg
->minor
);
8826 disk
= get_disk_super(super
,
8830 /* store action status */
8831 disk
->action
= DISK_REMOVE
;
8832 /* remove spare disks only */
8833 if (disk
->index
== -1) {
8834 remove_disk_super(super
,
8839 /* release allocate disk structure */
8840 __free_imsm_disk(disk_cfg
);
8843 return check_degraded
;
8846 static int apply_reshape_migration_update(struct imsm_update_reshape_migration
*u
,
8847 struct intel_super
*super
,
8850 struct intel_dev
*id
;
8851 void **tofree
= NULL
;
8854 dprintf("(enter)\n");
8855 if (u
->subdev
< 0 || u
->subdev
> 1) {
8856 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8859 if (space_list
== NULL
|| *space_list
== NULL
) {
8860 dprintf("imsm: Error: Memory is not allocated\n");
8864 for (id
= super
->devlist
; id
; id
= id
->next
) {
8865 if (id
->index
== (unsigned)u
->subdev
) {
8866 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8867 struct imsm_map
*map
;
8868 struct imsm_dev
*new_dev
=
8869 (struct imsm_dev
*)*space_list
;
8870 struct imsm_map
*migr_map
= get_imsm_map(dev
, MAP_1
);
8872 struct dl
*new_disk
;
8874 if (new_dev
== NULL
)
8876 *space_list
= **space_list
;
8877 memcpy(new_dev
, dev
, sizeof_imsm_dev(dev
, 0));
8878 map
= get_imsm_map(new_dev
, MAP_0
);
8880 dprintf("imsm: Error: migration in progress");
8884 to_state
= map
->map_state
;
8885 if ((u
->new_level
== 5) && (map
->raid_level
== 0)) {
8887 /* this should not happen */
8888 if (u
->new_disks
[0] < 0) {
8889 map
->failed_disk_num
=
8890 map
->num_members
- 1;
8891 to_state
= IMSM_T_STATE_DEGRADED
;
8893 to_state
= IMSM_T_STATE_NORMAL
;
8895 migrate(new_dev
, super
, to_state
, MIGR_GEN_MIGR
);
8896 if (u
->new_level
> -1)
8897 map
->raid_level
= u
->new_level
;
8898 migr_map
= get_imsm_map(new_dev
, MAP_1
);
8899 if ((u
->new_level
== 5) &&
8900 (migr_map
->raid_level
== 0)) {
8901 int ord
= map
->num_members
- 1;
8902 migr_map
->num_members
--;
8903 if (u
->new_disks
[0] < 0)
8904 ord
|= IMSM_ORD_REBUILD
;
8905 set_imsm_ord_tbl_ent(map
,
8906 map
->num_members
- 1,
8910 tofree
= (void **)dev
;
8912 /* update chunk size
8914 if (u
->new_chunksize
> 0) {
8915 unsigned long long num_data_stripes
;
8917 imsm_num_data_members(dev
, MAP_0
);
8919 if (used_disks
== 0)
8922 map
->blocks_per_strip
=
8923 __cpu_to_le16(u
->new_chunksize
* 2);
8925 (join_u32(dev
->size_low
, dev
->size_high
)
8927 num_data_stripes
/= map
->blocks_per_strip
;
8928 num_data_stripes
/= map
->num_domains
;
8929 set_num_data_stripes(map
, num_data_stripes
);
8934 if (u
->new_level
!= 5 || migr_map
->raid_level
!= 0 ||
8935 migr_map
->raid_level
== map
->raid_level
)
8938 if (u
->new_disks
[0] >= 0) {
8941 new_disk
= get_disk_super(super
,
8942 major(u
->new_disks
[0]),
8943 minor(u
->new_disks
[0]));
8944 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
8945 major(u
->new_disks
[0]),
8946 minor(u
->new_disks
[0]),
8947 new_disk
, new_disk
->index
);
8948 if (new_disk
== NULL
)
8949 goto error_disk_add
;
8951 new_disk
->index
= map
->num_members
- 1;
8952 /* slot to fill in autolayout
8954 new_disk
->raiddisk
= new_disk
->index
;
8955 new_disk
->disk
.status
|= CONFIGURED_DISK
;
8956 new_disk
->disk
.status
&= ~SPARE_DISK
;
8958 goto error_disk_add
;
8961 *tofree
= *space_list
;
8962 /* calculate new size
8964 imsm_set_array_size(new_dev
, -1);
8971 *space_list
= tofree
;
8975 dprintf("Error: imsm: Cannot find disk.\n");
8979 static int apply_size_change_update(struct imsm_update_size_change
*u
,
8980 struct intel_super
*super
)
8982 struct intel_dev
*id
;
8985 dprintf("(enter)\n");
8986 if (u
->subdev
< 0 || u
->subdev
> 1) {
8987 dprintf("imsm: Error: Wrong subdev: %i\n", u
->subdev
);
8991 for (id
= super
->devlist
; id
; id
= id
->next
) {
8992 if (id
->index
== (unsigned)u
->subdev
) {
8993 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subdev
);
8994 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
8995 int used_disks
= imsm_num_data_members(dev
, MAP_0
);
8996 unsigned long long blocks_per_member
;
8997 unsigned long long num_data_stripes
;
8999 /* calculate new size
9001 blocks_per_member
= u
->new_size
/ used_disks
;
9002 num_data_stripes
= blocks_per_member
/
9003 map
->blocks_per_strip
;
9004 num_data_stripes
/= map
->num_domains
;
9005 dprintf("(size: %llu, blocks per member: %llu, num_data_stipes: %llu)\n",
9006 u
->new_size
, blocks_per_member
,
9008 set_blocks_per_member(map
, blocks_per_member
);
9009 set_num_data_stripes(map
, num_data_stripes
);
9010 imsm_set_array_size(dev
, u
->new_size
);
9020 static int apply_update_activate_spare(struct imsm_update_activate_spare
*u
,
9021 struct intel_super
*super
,
9022 struct active_array
*active_array
)
9024 struct imsm_super
*mpb
= super
->anchor
;
9025 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
9026 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9027 struct imsm_map
*migr_map
;
9028 struct active_array
*a
;
9029 struct imsm_disk
*disk
;
9036 int second_map_created
= 0;
9038 for (; u
; u
= u
->next
) {
9039 victim
= get_imsm_disk_idx(dev
, u
->slot
, MAP_X
);
9044 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
9049 pr_err("error: imsm_activate_spare passed an unknown disk (index: %d)\n",
9054 /* count failures (excluding rebuilds and the victim)
9055 * to determine map[0] state
9058 for (i
= 0; i
< map
->num_members
; i
++) {
9061 disk
= get_imsm_disk(super
,
9062 get_imsm_disk_idx(dev
, i
, MAP_X
));
9063 if (!disk
|| is_failed(disk
))
9067 /* adding a pristine spare, assign a new index */
9068 if (dl
->index
< 0) {
9069 dl
->index
= super
->anchor
->num_disks
;
9070 super
->anchor
->num_disks
++;
9073 disk
->status
|= CONFIGURED_DISK
;
9074 disk
->status
&= ~SPARE_DISK
;
9077 to_state
= imsm_check_degraded(super
, dev
, failed
, MAP_0
);
9078 if (!second_map_created
) {
9079 second_map_created
= 1;
9080 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9081 migrate(dev
, super
, to_state
, MIGR_REBUILD
);
9083 map
->map_state
= to_state
;
9084 migr_map
= get_imsm_map(dev
, MAP_1
);
9085 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
9086 set_imsm_ord_tbl_ent(migr_map
, u
->slot
,
9087 dl
->index
| IMSM_ORD_REBUILD
);
9089 /* update the family_num to mark a new container
9090 * generation, being careful to record the existing
9091 * family_num in orig_family_num to clean up after
9092 * earlier mdadm versions that neglected to set it.
9094 if (mpb
->orig_family_num
== 0)
9095 mpb
->orig_family_num
= mpb
->family_num
;
9096 mpb
->family_num
+= super
->random
;
9098 /* count arrays using the victim in the metadata */
9100 for (a
= active_array
; a
; a
= a
->next
) {
9101 dev
= get_imsm_dev(super
, a
->info
.container_member
);
9102 map
= get_imsm_map(dev
, MAP_0
);
9104 if (get_imsm_disk_slot(map
, victim
) >= 0)
9108 /* delete the victim if it is no longer being
9114 /* We know that 'manager' isn't touching anything,
9115 * so it is safe to delete
9117 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
9118 if ((*dlp
)->index
== victim
)
9121 /* victim may be on the missing list */
9123 for (dlp
= &super
->missing
; *dlp
;
9124 dlp
= &(*dlp
)->next
)
9125 if ((*dlp
)->index
== victim
)
9127 imsm_delete(super
, dlp
, victim
);
9134 static int apply_reshape_container_disks_update(struct imsm_update_reshape
*u
,
9135 struct intel_super
*super
,
9138 struct dl
*new_disk
;
9139 struct intel_dev
*id
;
9141 int delta_disks
= u
->new_raid_disks
- u
->old_raid_disks
;
9142 int disk_count
= u
->old_raid_disks
;
9143 void **tofree
= NULL
;
9144 int devices_to_reshape
= 1;
9145 struct imsm_super
*mpb
= super
->anchor
;
9147 unsigned int dev_id
;
9149 dprintf("(enter)\n");
9151 /* enable spares to use in array */
9152 for (i
= 0; i
< delta_disks
; i
++) {
9153 new_disk
= get_disk_super(super
,
9154 major(u
->new_disks
[i
]),
9155 minor(u
->new_disks
[i
]));
9156 dprintf("imsm: new disk for reshape is: %i:%i (%p, index = %i)\n",
9157 major(u
->new_disks
[i
]), minor(u
->new_disks
[i
]),
9158 new_disk
, new_disk
->index
);
9159 if (new_disk
== NULL
||
9160 (new_disk
->index
>= 0 &&
9161 new_disk
->index
< u
->old_raid_disks
))
9162 goto update_reshape_exit
;
9163 new_disk
->index
= disk_count
++;
9164 /* slot to fill in autolayout
9166 new_disk
->raiddisk
= new_disk
->index
;
9167 new_disk
->disk
.status
|=
9169 new_disk
->disk
.status
&= ~SPARE_DISK
;
9172 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
9173 mpb
->num_raid_devs
);
9174 /* manage changes in volume
9176 for (dev_id
= 0; dev_id
< mpb
->num_raid_devs
; dev_id
++) {
9177 void **sp
= *space_list
;
9178 struct imsm_dev
*newdev
;
9179 struct imsm_map
*newmap
, *oldmap
;
9181 for (id
= super
->devlist
; id
; id
= id
->next
) {
9182 if (id
->index
== dev_id
)
9191 /* Copy the dev, but not (all of) the map */
9192 memcpy(newdev
, id
->dev
, sizeof(*newdev
));
9193 oldmap
= get_imsm_map(id
->dev
, MAP_0
);
9194 newmap
= get_imsm_map(newdev
, MAP_0
);
9195 /* Copy the current map */
9196 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9197 /* update one device only
9199 if (devices_to_reshape
) {
9200 dprintf("imsm: modifying subdev: %i\n",
9202 devices_to_reshape
--;
9203 newdev
->vol
.migr_state
= 1;
9204 newdev
->vol
.curr_migr_unit
= 0;
9205 set_migr_type(newdev
, MIGR_GEN_MIGR
);
9206 newmap
->num_members
= u
->new_raid_disks
;
9207 for (i
= 0; i
< delta_disks
; i
++) {
9208 set_imsm_ord_tbl_ent(newmap
,
9209 u
->old_raid_disks
+ i
,
9210 u
->old_raid_disks
+ i
);
9212 /* New map is correct, now need to save old map
9214 newmap
= get_imsm_map(newdev
, MAP_1
);
9215 memcpy(newmap
, oldmap
, sizeof_imsm_map(oldmap
));
9217 imsm_set_array_size(newdev
, -1);
9220 sp
= (void **)id
->dev
;
9225 /* Clear migration record */
9226 memset(super
->migr_rec
, 0, sizeof(struct migr_record
));
9229 *space_list
= tofree
;
9232 update_reshape_exit
:
9237 static int apply_takeover_update(struct imsm_update_takeover
*u
,
9238 struct intel_super
*super
,
9241 struct imsm_dev
*dev
= NULL
;
9242 struct intel_dev
*dv
;
9243 struct imsm_dev
*dev_new
;
9244 struct imsm_map
*map
;
9248 for (dv
= super
->devlist
; dv
; dv
= dv
->next
)
9249 if (dv
->index
== (unsigned int)u
->subarray
) {
9257 map
= get_imsm_map(dev
, MAP_0
);
9259 if (u
->direction
== R10_TO_R0
) {
9260 unsigned long long num_data_stripes
;
9262 map
->num_domains
= 1;
9263 num_data_stripes
= blocks_per_member(map
);
9264 num_data_stripes
/= map
->blocks_per_strip
;
9265 num_data_stripes
/= map
->num_domains
;
9266 set_num_data_stripes(map
, num_data_stripes
);
9268 /* Number of failed disks must be half of initial disk number */
9269 if (imsm_count_failed(super
, dev
, MAP_0
) !=
9270 (map
->num_members
/ 2))
9273 /* iterate through devices to mark removed disks as spare */
9274 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9275 if (dm
->disk
.status
& FAILED_DISK
) {
9276 int idx
= dm
->index
;
9277 /* update indexes on the disk list */
9278 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
9279 the index values will end up being correct.... NB */
9280 for (du
= super
->disks
; du
; du
= du
->next
)
9281 if (du
->index
> idx
)
9283 /* mark as spare disk */
9288 map
->num_members
= map
->num_members
/ 2;
9289 map
->map_state
= IMSM_T_STATE_NORMAL
;
9290 map
->num_domains
= 1;
9291 map
->raid_level
= 0;
9292 map
->failed_disk_num
= -1;
9295 if (u
->direction
== R0_TO_R10
) {
9297 /* update slots in current disk list */
9298 for (dm
= super
->disks
; dm
; dm
= dm
->next
) {
9302 /* create new *missing* disks */
9303 for (i
= 0; i
< map
->num_members
; i
++) {
9304 space
= *space_list
;
9307 *space_list
= *space
;
9309 memcpy(du
, super
->disks
, sizeof(*du
));
9313 du
->index
= (i
* 2) + 1;
9314 sprintf((char *)du
->disk
.serial
,
9315 " MISSING_%d", du
->index
);
9316 sprintf((char *)du
->serial
,
9317 "MISSING_%d", du
->index
);
9318 du
->next
= super
->missing
;
9319 super
->missing
= du
;
9321 /* create new dev and map */
9322 space
= *space_list
;
9325 *space_list
= *space
;
9326 dev_new
= (void *)space
;
9327 memcpy(dev_new
, dev
, sizeof(*dev
));
9328 /* update new map */
9329 map
= get_imsm_map(dev_new
, MAP_0
);
9330 map
->num_members
= map
->num_members
* 2;
9331 map
->map_state
= IMSM_T_STATE_DEGRADED
;
9332 map
->num_domains
= 2;
9333 map
->raid_level
= 1;
9334 /* replace dev<->dev_new */
9337 /* update disk order table */
9338 for (du
= super
->disks
; du
; du
= du
->next
)
9340 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9341 for (du
= super
->missing
; du
; du
= du
->next
)
9342 if (du
->index
>= 0) {
9343 set_imsm_ord_tbl_ent(map
, du
->index
, du
->index
);
9344 mark_missing(super
, dv
->dev
, &du
->disk
, du
->index
);
9350 static void imsm_process_update(struct supertype
*st
,
9351 struct metadata_update
*update
)
9354 * crack open the metadata_update envelope to find the update record
9355 * update can be one of:
9356 * update_reshape_container_disks - all the arrays in the container
9357 * are being reshaped to have more devices. We need to mark
9358 * the arrays for general migration and convert selected spares
9359 * into active devices.
9360 * update_activate_spare - a spare device has replaced a failed
9361 * device in an array, update the disk_ord_tbl. If this disk is
9362 * present in all member arrays then also clear the SPARE_DISK
9364 * update_create_array
9366 * update_rename_array
9367 * update_add_remove_disk
9369 struct intel_super
*super
= st
->sb
;
9370 struct imsm_super
*mpb
;
9371 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
9373 /* update requires a larger buf but the allocation failed */
9374 if (super
->next_len
&& !super
->next_buf
) {
9375 super
->next_len
= 0;
9379 if (super
->next_buf
) {
9380 memcpy(super
->next_buf
, super
->buf
, super
->len
);
9382 super
->len
= super
->next_len
;
9383 super
->buf
= super
->next_buf
;
9385 super
->next_len
= 0;
9386 super
->next_buf
= NULL
;
9389 mpb
= super
->anchor
;
9392 case update_general_migration_checkpoint
: {
9393 struct intel_dev
*id
;
9394 struct imsm_update_general_migration_checkpoint
*u
=
9395 (void *)update
->buf
;
9397 dprintf("called for update_general_migration_checkpoint\n");
9399 /* find device under general migration */
9400 for (id
= super
->devlist
; id
; id
= id
->next
) {
9401 if (is_gen_migration(id
->dev
)) {
9402 id
->dev
->vol
.curr_migr_unit
=
9403 __cpu_to_le32(u
->curr_migr_unit
);
9404 super
->updates_pending
++;
9409 case update_takeover
: {
9410 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9411 if (apply_takeover_update(u
, super
, &update
->space_list
)) {
9412 imsm_update_version_info(super
);
9413 super
->updates_pending
++;
9418 case update_reshape_container_disks
: {
9419 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9420 if (apply_reshape_container_disks_update(
9421 u
, super
, &update
->space_list
))
9422 super
->updates_pending
++;
9425 case update_reshape_migration
: {
9426 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9427 if (apply_reshape_migration_update(
9428 u
, super
, &update
->space_list
))
9429 super
->updates_pending
++;
9432 case update_size_change
: {
9433 struct imsm_update_size_change
*u
= (void *)update
->buf
;
9434 if (apply_size_change_update(u
, super
))
9435 super
->updates_pending
++;
9438 case update_activate_spare
: {
9439 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
9440 if (apply_update_activate_spare(u
, super
, st
->arrays
))
9441 super
->updates_pending
++;
9444 case update_create_array
: {
9445 /* someone wants to create a new array, we need to be aware of
9446 * a few races/collisions:
9447 * 1/ 'Create' called by two separate instances of mdadm
9448 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
9449 * devices that have since been assimilated via
9451 * In the event this update can not be carried out mdadm will
9452 * (FIX ME) notice that its update did not take hold.
9454 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9455 struct intel_dev
*dv
;
9456 struct imsm_dev
*dev
;
9457 struct imsm_map
*map
, *new_map
;
9458 unsigned long long start
, end
;
9459 unsigned long long new_start
, new_end
;
9461 struct disk_info
*inf
;
9464 /* handle racing creates: first come first serve */
9465 if (u
->dev_idx
< mpb
->num_raid_devs
) {
9466 dprintf("subarray %d already defined\n", u
->dev_idx
);
9470 /* check update is next in sequence */
9471 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
9472 dprintf("can not create array %d expected index %d\n",
9473 u
->dev_idx
, mpb
->num_raid_devs
);
9477 new_map
= get_imsm_map(&u
->dev
, MAP_0
);
9478 new_start
= pba_of_lba0(new_map
);
9479 new_end
= new_start
+ blocks_per_member(new_map
);
9480 inf
= get_disk_info(u
);
9482 /* handle activate_spare versus create race:
9483 * check to make sure that overlapping arrays do not include
9486 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9487 dev
= get_imsm_dev(super
, i
);
9488 map
= get_imsm_map(dev
, MAP_0
);
9489 start
= pba_of_lba0(map
);
9490 end
= start
+ blocks_per_member(map
);
9491 if ((new_start
>= start
&& new_start
<= end
) ||
9492 (start
>= new_start
&& start
<= new_end
))
9497 if (disks_overlap(super
, i
, u
)) {
9498 dprintf("arrays overlap\n");
9503 /* check that prepare update was successful */
9504 if (!update
->space
) {
9505 dprintf("prepare update failed\n");
9509 /* check that all disks are still active before committing
9510 * changes. FIXME: could we instead handle this by creating a
9511 * degraded array? That's probably not what the user expects,
9512 * so better to drop this update on the floor.
9514 for (i
= 0; i
< new_map
->num_members
; i
++) {
9515 dl
= serial_to_dl(inf
[i
].serial
, super
);
9517 dprintf("disk disappeared\n");
9522 super
->updates_pending
++;
9524 /* convert spares to members and fixup ord_tbl */
9525 for (i
= 0; i
< new_map
->num_members
; i
++) {
9526 dl
= serial_to_dl(inf
[i
].serial
, super
);
9527 if (dl
->index
== -1) {
9528 dl
->index
= mpb
->num_disks
;
9530 dl
->disk
.status
|= CONFIGURED_DISK
;
9531 dl
->disk
.status
&= ~SPARE_DISK
;
9533 set_imsm_ord_tbl_ent(new_map
, i
, dl
->index
);
9538 update
->space
= NULL
;
9539 imsm_copy_dev(dev
, &u
->dev
);
9540 dv
->index
= u
->dev_idx
;
9541 dv
->next
= super
->devlist
;
9542 super
->devlist
= dv
;
9543 mpb
->num_raid_devs
++;
9545 imsm_update_version_info(super
);
9548 /* mdmon knows how to release update->space, but not
9549 * ((struct intel_dev *) update->space)->dev
9551 if (update
->space
) {
9557 case update_kill_array
: {
9558 struct imsm_update_kill_array
*u
= (void *) update
->buf
;
9559 int victim
= u
->dev_idx
;
9560 struct active_array
*a
;
9561 struct intel_dev
**dp
;
9562 struct imsm_dev
*dev
;
9564 /* sanity check that we are not affecting the uuid of
9565 * active arrays, or deleting an active array
9567 * FIXME when immutable ids are available, but note that
9568 * we'll also need to fixup the invalidated/active
9569 * subarray indexes in mdstat
9571 for (a
= st
->arrays
; a
; a
= a
->next
)
9572 if (a
->info
.container_member
>= victim
)
9574 /* by definition if mdmon is running at least one array
9575 * is active in the container, so checking
9576 * mpb->num_raid_devs is just extra paranoia
9578 dev
= get_imsm_dev(super
, victim
);
9579 if (a
|| !dev
|| mpb
->num_raid_devs
== 1) {
9580 dprintf("failed to delete subarray-%d\n", victim
);
9584 for (dp
= &super
->devlist
; *dp
;)
9585 if ((*dp
)->index
== (unsigned)super
->current_vol
) {
9588 if ((*dp
)->index
> (unsigned)victim
)
9592 mpb
->num_raid_devs
--;
9593 super
->updates_pending
++;
9596 case update_rename_array
: {
9597 struct imsm_update_rename_array
*u
= (void *) update
->buf
;
9598 char name
[MAX_RAID_SERIAL_LEN
+1];
9599 int target
= u
->dev_idx
;
9600 struct active_array
*a
;
9601 struct imsm_dev
*dev
;
9603 /* sanity check that we are not affecting the uuid of
9606 snprintf(name
, MAX_RAID_SERIAL_LEN
, "%s", (char *) u
->name
);
9607 name
[MAX_RAID_SERIAL_LEN
] = '\0';
9608 for (a
= st
->arrays
; a
; a
= a
->next
)
9609 if (a
->info
.container_member
== target
)
9611 dev
= get_imsm_dev(super
, u
->dev_idx
);
9612 if (a
|| !dev
|| !check_name(super
, name
, 1)) {
9613 dprintf("failed to rename subarray-%d\n", target
);
9617 snprintf((char *) dev
->volume
, MAX_RAID_SERIAL_LEN
, "%s", name
);
9618 super
->updates_pending
++;
9621 case update_add_remove_disk
: {
9622 /* we may be able to repair some arrays if disks are
9623 * being added, check the status of add_remove_disk
9624 * if discs has been added.
9626 if (add_remove_disk_update(super
)) {
9627 struct active_array
*a
;
9629 super
->updates_pending
++;
9630 for (a
= st
->arrays
; a
; a
= a
->next
)
9631 a
->check_degraded
= 1;
9635 case update_prealloc_badblocks_mem
:
9637 case update_rwh_policy
: {
9638 struct imsm_update_rwh_policy
*u
= (void *)update
->buf
;
9639 int target
= u
->dev_idx
;
9640 struct imsm_dev
*dev
= get_imsm_dev(super
, target
);
9642 dprintf("could not find subarray-%d\n", target
);
9646 if (dev
->rwh_policy
!= u
->new_policy
) {
9647 dev
->rwh_policy
= u
->new_policy
;
9648 super
->updates_pending
++;
9653 pr_err("error: unsuported process update type:(type: %d)\n", type
);
9657 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
);
9659 static int imsm_prepare_update(struct supertype
*st
,
9660 struct metadata_update
*update
)
9663 * Allocate space to hold new disk entries, raid-device entries or a new
9664 * mpb if necessary. The manager synchronously waits for updates to
9665 * complete in the monitor, so new mpb buffers allocated here can be
9666 * integrated by the monitor thread without worrying about live pointers
9667 * in the manager thread.
9669 enum imsm_update_type type
;
9670 struct intel_super
*super
= st
->sb
;
9671 unsigned int sector_size
= super
->sector_size
;
9672 struct imsm_super
*mpb
= super
->anchor
;
9676 if (update
->len
< (int)sizeof(type
))
9679 type
= *(enum imsm_update_type
*) update
->buf
;
9682 case update_general_migration_checkpoint
:
9683 if (update
->len
< (int)sizeof(struct imsm_update_general_migration_checkpoint
))
9685 dprintf("called for update_general_migration_checkpoint\n");
9687 case update_takeover
: {
9688 struct imsm_update_takeover
*u
= (void *)update
->buf
;
9689 if (update
->len
< (int)sizeof(*u
))
9691 if (u
->direction
== R0_TO_R10
) {
9692 void **tail
= (void **)&update
->space_list
;
9693 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->subarray
);
9694 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9695 int num_members
= map
->num_members
;
9698 /* allocate memory for added disks */
9699 for (i
= 0; i
< num_members
; i
++) {
9700 size
= sizeof(struct dl
);
9701 space
= xmalloc(size
);
9706 /* allocate memory for new device */
9707 size
= sizeof_imsm_dev(super
->devlist
->dev
, 0) +
9708 (num_members
* sizeof(__u32
));
9709 space
= xmalloc(size
);
9713 len
= disks_to_mpb_size(num_members
* 2);
9718 case update_reshape_container_disks
: {
9719 /* Every raid device in the container is about to
9720 * gain some more devices, and we will enter a
9722 * So each 'imsm_map' will be bigger, and the imsm_vol
9723 * will now hold 2 of them.
9724 * Thus we need new 'struct imsm_dev' allocations sized
9725 * as sizeof_imsm_dev but with more devices in both maps.
9727 struct imsm_update_reshape
*u
= (void *)update
->buf
;
9728 struct intel_dev
*dl
;
9729 void **space_tail
= (void**)&update
->space_list
;
9731 if (update
->len
< (int)sizeof(*u
))
9734 dprintf("for update_reshape\n");
9736 for (dl
= super
->devlist
; dl
; dl
= dl
->next
) {
9737 int size
= sizeof_imsm_dev(dl
->dev
, 1);
9739 if (u
->new_raid_disks
> u
->old_raid_disks
)
9740 size
+= sizeof(__u32
)*2*
9741 (u
->new_raid_disks
- u
->old_raid_disks
);
9748 len
= disks_to_mpb_size(u
->new_raid_disks
);
9749 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9752 case update_reshape_migration
: {
9753 /* for migration level 0->5 we need to add disks
9754 * so the same as for container operation we will copy
9755 * device to the bigger location.
9756 * in memory prepared device and new disk area are prepared
9757 * for usage in process update
9759 struct imsm_update_reshape_migration
*u
= (void *)update
->buf
;
9760 struct intel_dev
*id
;
9761 void **space_tail
= (void **)&update
->space_list
;
9764 int current_level
= -1;
9766 if (update
->len
< (int)sizeof(*u
))
9769 dprintf("for update_reshape\n");
9771 /* add space for bigger array in update
9773 for (id
= super
->devlist
; id
; id
= id
->next
) {
9774 if (id
->index
== (unsigned)u
->subdev
) {
9775 size
= sizeof_imsm_dev(id
->dev
, 1);
9776 if (u
->new_raid_disks
> u
->old_raid_disks
)
9777 size
+= sizeof(__u32
)*2*
9778 (u
->new_raid_disks
- u
->old_raid_disks
);
9786 if (update
->space_list
== NULL
)
9789 /* add space for disk in update
9791 size
= sizeof(struct dl
);
9797 /* add spare device to update
9799 for (id
= super
->devlist
; id
; id
= id
->next
)
9800 if (id
->index
== (unsigned)u
->subdev
) {
9801 struct imsm_dev
*dev
;
9802 struct imsm_map
*map
;
9804 dev
= get_imsm_dev(super
, u
->subdev
);
9805 map
= get_imsm_map(dev
, MAP_0
);
9806 current_level
= map
->raid_level
;
9809 if (u
->new_level
== 5 && u
->new_level
!= current_level
) {
9810 struct mdinfo
*spares
;
9812 spares
= get_spares_for_grow(st
);
9820 makedev(dev
->disk
.major
,
9822 dl
= get_disk_super(super
,
9825 dl
->index
= u
->old_raid_disks
;
9831 len
= disks_to_mpb_size(u
->new_raid_disks
);
9832 dprintf("New anchor length is %llu\n", (unsigned long long)len
);
9835 case update_size_change
: {
9836 if (update
->len
< (int)sizeof(struct imsm_update_size_change
))
9840 case update_activate_spare
: {
9841 if (update
->len
< (int)sizeof(struct imsm_update_activate_spare
))
9845 case update_create_array
: {
9846 struct imsm_update_create_array
*u
= (void *) update
->buf
;
9847 struct intel_dev
*dv
;
9848 struct imsm_dev
*dev
= &u
->dev
;
9849 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
9851 struct disk_info
*inf
;
9855 if (update
->len
< (int)sizeof(*u
))
9858 inf
= get_disk_info(u
);
9859 len
= sizeof_imsm_dev(dev
, 1);
9860 /* allocate a new super->devlist entry */
9861 dv
= xmalloc(sizeof(*dv
));
9862 dv
->dev
= xmalloc(len
);
9865 /* count how many spares will be converted to members */
9866 for (i
= 0; i
< map
->num_members
; i
++) {
9867 dl
= serial_to_dl(inf
[i
].serial
, super
);
9869 /* hmm maybe it failed?, nothing we can do about
9874 if (count_memberships(dl
, super
) == 0)
9877 len
+= activate
* sizeof(struct imsm_disk
);
9880 case update_kill_array
: {
9881 if (update
->len
< (int)sizeof(struct imsm_update_kill_array
))
9885 case update_rename_array
: {
9886 if (update
->len
< (int)sizeof(struct imsm_update_rename_array
))
9890 case update_add_remove_disk
:
9891 /* no update->len needed */
9893 case update_prealloc_badblocks_mem
:
9894 super
->extra_space
+= sizeof(struct bbm_log
) -
9895 get_imsm_bbm_log_size(super
->bbm_log
);
9897 case update_rwh_policy
: {
9898 if (update
->len
< (int)sizeof(struct imsm_update_rwh_policy
))
9906 /* check if we need a larger metadata buffer */
9907 if (super
->next_buf
)
9908 buf_len
= super
->next_len
;
9910 buf_len
= super
->len
;
9912 if (__le32_to_cpu(mpb
->mpb_size
) + super
->extra_space
+ len
> buf_len
) {
9913 /* ok we need a larger buf than what is currently allocated
9914 * if this allocation fails process_update will notice that
9915 * ->next_len is set and ->next_buf is NULL
9917 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) +
9918 super
->extra_space
+ len
, sector_size
);
9919 if (super
->next_buf
)
9920 free(super
->next_buf
);
9922 super
->next_len
= buf_len
;
9923 if (posix_memalign(&super
->next_buf
, sector_size
, buf_len
) == 0)
9924 memset(super
->next_buf
, 0, buf_len
);
9926 super
->next_buf
= NULL
;
9931 /* must be called while manager is quiesced */
9932 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, unsigned index
)
9934 struct imsm_super
*mpb
= super
->anchor
;
9936 struct imsm_dev
*dev
;
9937 struct imsm_map
*map
;
9938 unsigned int i
, j
, num_members
;
9940 struct bbm_log
*log
= super
->bbm_log
;
9942 dprintf("deleting device[%d] from imsm_super\n", index
);
9944 /* shift all indexes down one */
9945 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
9946 if (iter
->index
> (int)index
)
9948 for (iter
= super
->missing
; iter
; iter
= iter
->next
)
9949 if (iter
->index
> (int)index
)
9952 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
9953 dev
= get_imsm_dev(super
, i
);
9954 map
= get_imsm_map(dev
, MAP_0
);
9955 num_members
= map
->num_members
;
9956 for (j
= 0; j
< num_members
; j
++) {
9957 /* update ord entries being careful not to propagate
9958 * ord-flags to the first map
9960 ord
= get_imsm_ord_tbl_ent(dev
, j
, MAP_X
);
9962 if (ord_to_idx(ord
) <= index
)
9965 map
= get_imsm_map(dev
, MAP_0
);
9966 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
9967 map
= get_imsm_map(dev
, MAP_1
);
9969 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
9973 for (i
= 0; i
< log
->entry_count
; i
++) {
9974 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[i
];
9976 if (entry
->disk_ordinal
<= index
)
9978 entry
->disk_ordinal
--;
9982 super
->updates_pending
++;
9984 struct dl
*dl
= *dlp
;
9986 *dlp
= (*dlp
)->next
;
9987 __free_imsm_disk(dl
);
9990 #endif /* MDASSEMBLE */
9992 static void close_targets(int *targets
, int new_disks
)
9999 for (i
= 0; i
< new_disks
; i
++) {
10000 if (targets
[i
] >= 0) {
10007 static int imsm_get_allowed_degradation(int level
, int raid_disks
,
10008 struct intel_super
*super
,
10009 struct imsm_dev
*dev
)
10015 struct imsm_map
*map
;
10018 ret_val
= raid_disks
/2;
10019 /* check map if all disks pairs not failed
10022 map
= get_imsm_map(dev
, MAP_0
);
10023 for (i
= 0; i
< ret_val
; i
++) {
10024 int degradation
= 0;
10025 if (get_imsm_disk(super
, i
) == NULL
)
10027 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10029 if (degradation
== 2)
10032 map
= get_imsm_map(dev
, MAP_1
);
10033 /* if there is no second map
10034 * result can be returned
10038 /* check degradation in second map
10040 for (i
= 0; i
< ret_val
; i
++) {
10041 int degradation
= 0;
10042 if (get_imsm_disk(super
, i
) == NULL
)
10044 if (get_imsm_disk(super
, i
+ 1) == NULL
)
10046 if (degradation
== 2)
10060 /*******************************************************************************
10061 * Function: open_backup_targets
10062 * Description: Function opens file descriptors for all devices given in
10065 * info : general array info
10066 * raid_disks : number of disks
10067 * raid_fds : table of device's file descriptors
10068 * super : intel super for raid10 degradation check
10069 * dev : intel device for raid10 degradation check
10073 ******************************************************************************/
10074 int open_backup_targets(struct mdinfo
*info
, int raid_disks
, int *raid_fds
,
10075 struct intel_super
*super
, struct imsm_dev
*dev
)
10081 for (i
= 0; i
< raid_disks
; i
++)
10084 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10087 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
10088 dprintf("disk is faulty!!\n");
10092 if (sd
->disk
.raid_disk
>= raid_disks
|| sd
->disk
.raid_disk
< 0)
10095 dn
= map_dev(sd
->disk
.major
,
10096 sd
->disk
.minor
, 1);
10097 raid_fds
[sd
->disk
.raid_disk
] = dev_open(dn
, O_RDWR
);
10098 if (raid_fds
[sd
->disk
.raid_disk
] < 0) {
10099 pr_err("cannot open component\n");
10104 /* check if maximum array degradation level is not exceeded
10106 if ((raid_disks
- opened
) >
10107 imsm_get_allowed_degradation(info
->new_level
, raid_disks
,
10109 pr_err("Not enough disks can be opened.\n");
10110 close_targets(raid_fds
, raid_disks
);
10116 /*******************************************************************************
10117 * Function: validate_container_imsm
10118 * Description: This routine validates container after assemble,
10119 * eg. if devices in container are under the same controller.
10122 * info : linked list with info about devices used in array
10126 ******************************************************************************/
10127 int validate_container_imsm(struct mdinfo
*info
)
10129 if (check_env("IMSM_NO_PLATFORM"))
10132 struct sys_dev
*idev
;
10133 struct sys_dev
*hba
= NULL
;
10134 struct sys_dev
*intel_devices
= find_intel_devices();
10135 char *dev_path
= devt_to_devpath(makedev(info
->disk
.major
,
10136 info
->disk
.minor
));
10138 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10139 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10148 pr_err("WARNING - Cannot detect HBA for device %s!\n",
10149 devid2kname(makedev(info
->disk
.major
, info
->disk
.minor
)));
10153 const struct imsm_orom
*orom
= get_orom_by_device_id(hba
->dev_id
);
10154 struct mdinfo
*dev
;
10156 for (dev
= info
->next
; dev
; dev
= dev
->next
) {
10157 dev_path
= devt_to_devpath(makedev(dev
->disk
.major
, dev
->disk
.minor
));
10159 struct sys_dev
*hba2
= NULL
;
10160 for (idev
= intel_devices
; idev
; idev
= idev
->next
) {
10161 if (dev_path
&& strstr(dev_path
, idev
->path
)) {
10169 const struct imsm_orom
*orom2
= hba2
== NULL
? NULL
:
10170 get_orom_by_device_id(hba2
->dev_id
);
10172 if (hba2
&& hba
->type
!= hba2
->type
) {
10173 pr_err("WARNING - HBAs of devices do not match %s != %s\n",
10174 get_sys_dev_type(hba
->type
), get_sys_dev_type(hba2
->type
));
10178 if (orom
!= orom2
) {
10179 pr_err("WARNING - IMSM container assembled with disks under different HBAs!\n"
10180 " This operation is not supported and can lead to data loss.\n");
10185 pr_err("WARNING - IMSM container assembled with disks under HBAs without IMSM platform support!\n"
10186 " This operation is not supported and can lead to data loss.\n");
10194 /*******************************************************************************
10195 * Function: imsm_record_badblock
10196 * Description: This routine stores new bad block record in BBM log
10199 * a : array containing a bad block
10200 * slot : disk number containing a bad block
10201 * sector : bad block sector
10202 * length : bad block sectors range
10206 ******************************************************************************/
10207 static int imsm_record_badblock(struct active_array
*a
, int slot
,
10208 unsigned long long sector
, int length
)
10210 struct intel_super
*super
= a
->container
->sb
;
10214 ord
= imsm_disk_slot_to_ord(a
, slot
);
10218 ret
= record_new_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
,
10221 super
->updates_pending
++;
10225 /*******************************************************************************
10226 * Function: imsm_clear_badblock
10227 * Description: This routine clears bad block record from BBM log
10230 * a : array containing a bad block
10231 * slot : disk number containing a bad block
10232 * sector : bad block sector
10233 * length : bad block sectors range
10237 ******************************************************************************/
10238 static int imsm_clear_badblock(struct active_array
*a
, int slot
,
10239 unsigned long long sector
, int length
)
10241 struct intel_super
*super
= a
->container
->sb
;
10245 ord
= imsm_disk_slot_to_ord(a
, slot
);
10249 ret
= clear_badblock(super
->bbm_log
, ord_to_idx(ord
), sector
, length
);
10251 super
->updates_pending
++;
10255 /*******************************************************************************
10256 * Function: imsm_get_badblocks
10257 * Description: This routine get list of bad blocks for an array
10261 * slot : disk number
10263 * bb : structure containing bad blocks
10265 ******************************************************************************/
10266 static struct md_bb
*imsm_get_badblocks(struct active_array
*a
, int slot
)
10268 int inst
= a
->info
.container_member
;
10269 struct intel_super
*super
= a
->container
->sb
;
10270 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
10271 struct imsm_map
*map
= get_imsm_map(dev
, MAP_0
);
10274 ord
= imsm_disk_slot_to_ord(a
, slot
);
10278 get_volume_badblocks(super
->bbm_log
, ord_to_idx(ord
), pba_of_lba0(map
),
10279 blocks_per_member(map
), &super
->bb
);
10283 /*******************************************************************************
10284 * Function: examine_badblocks_imsm
10285 * Description: Prints list of bad blocks on a disk to the standard output
10288 * st : metadata handler
10289 * fd : open file descriptor for device
10290 * devname : device name
10294 ******************************************************************************/
10295 static int examine_badblocks_imsm(struct supertype
*st
, int fd
, char *devname
)
10297 struct intel_super
*super
= st
->sb
;
10298 struct bbm_log
*log
= super
->bbm_log
;
10299 struct dl
*d
= NULL
;
10302 for (d
= super
->disks
; d
; d
= d
->next
) {
10303 if (strcmp(d
->devname
, devname
) == 0)
10307 if ((d
== NULL
) || (d
->index
< 0)) { /* serial mismatch probably */
10308 pr_err("%s doesn't appear to be part of a raid array\n",
10315 struct bbm_log_entry
*entry
= &log
->marked_block_entries
[0];
10317 for (i
= 0; i
< log
->entry_count
; i
++) {
10318 if (entry
[i
].disk_ordinal
== d
->index
) {
10319 unsigned long long sector
= __le48_to_cpu(
10320 &entry
[i
].defective_block_start
);
10321 int cnt
= entry
[i
].marked_count
+ 1;
10324 printf("Bad-blocks on %s:\n", devname
);
10328 printf("%20llu for %d sectors\n", sector
, cnt
);
10334 printf("No bad-blocks list configured on %s\n", devname
);
10338 /*******************************************************************************
10339 * Function: init_migr_record_imsm
10340 * Description: Function inits imsm migration record
10342 * super : imsm internal array info
10343 * dev : device under migration
10344 * info : general array info to find the smallest device
10347 ******************************************************************************/
10348 void init_migr_record_imsm(struct supertype
*st
, struct imsm_dev
*dev
,
10349 struct mdinfo
*info
)
10351 struct intel_super
*super
= st
->sb
;
10352 struct migr_record
*migr_rec
= super
->migr_rec
;
10353 int new_data_disks
;
10354 unsigned long long dsize
, dev_sectors
;
10355 long long unsigned min_dev_sectors
= -1LLU;
10359 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10360 struct imsm_map
*map_src
= get_imsm_map(dev
, MAP_1
);
10361 unsigned long long num_migr_units
;
10362 unsigned long long array_blocks
;
10364 memset(migr_rec
, 0, sizeof(struct migr_record
));
10365 migr_rec
->family_num
= __cpu_to_le32(super
->anchor
->family_num
);
10367 /* only ascending reshape supported now */
10368 migr_rec
->ascending_migr
= __cpu_to_le32(1);
10370 migr_rec
->dest_depth_per_unit
= GEN_MIGR_AREA_SIZE
/
10371 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10372 migr_rec
->dest_depth_per_unit
*=
10373 max(map_dest
->blocks_per_strip
, map_src
->blocks_per_strip
);
10374 new_data_disks
= imsm_num_data_members(dev
, MAP_0
);
10375 migr_rec
->blocks_per_unit
=
10376 __cpu_to_le32(migr_rec
->dest_depth_per_unit
* new_data_disks
);
10377 migr_rec
->dest_depth_per_unit
=
10378 __cpu_to_le32(migr_rec
->dest_depth_per_unit
);
10379 array_blocks
= info
->component_size
* new_data_disks
;
10381 array_blocks
/ __le32_to_cpu(migr_rec
->blocks_per_unit
);
10383 if (array_blocks
% __le32_to_cpu(migr_rec
->blocks_per_unit
))
10385 migr_rec
->num_migr_units
= __cpu_to_le32(num_migr_units
);
10387 migr_rec
->post_migr_vol_cap
= dev
->size_low
;
10388 migr_rec
->post_migr_vol_cap_hi
= dev
->size_high
;
10390 /* Find the smallest dev */
10391 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
10392 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
10393 fd
= dev_open(nm
, O_RDONLY
);
10396 get_dev_size(fd
, NULL
, &dsize
);
10397 dev_sectors
= dsize
/ 512;
10398 if (dev_sectors
< min_dev_sectors
)
10399 min_dev_sectors
= dev_sectors
;
10402 migr_rec
->ckpt_area_pba
= __cpu_to_le32(min_dev_sectors
-
10403 RAID_DISK_RESERVED_BLOCKS_IMSM_HI
);
10405 write_imsm_migr_rec(st
);
10410 /*******************************************************************************
10411 * Function: save_backup_imsm
10412 * Description: Function saves critical data stripes to Migration Copy Area
10413 * and updates the current migration unit status.
10414 * Use restore_stripes() to form a destination stripe,
10415 * and to write it to the Copy Area.
10417 * st : supertype information
10418 * dev : imsm device that backup is saved for
10419 * info : general array info
10420 * buf : input buffer
10421 * length : length of data to backup (blocks_per_unit)
10425 ******************************************************************************/
10426 int save_backup_imsm(struct supertype
*st
,
10427 struct imsm_dev
*dev
,
10428 struct mdinfo
*info
,
10433 struct intel_super
*super
= st
->sb
;
10434 unsigned long long *target_offsets
;
10437 struct imsm_map
*map_dest
= get_imsm_map(dev
, MAP_0
);
10438 int new_disks
= map_dest
->num_members
;
10439 int dest_layout
= 0;
10441 unsigned long long start
;
10442 int data_disks
= imsm_num_data_members(dev
, MAP_0
);
10444 targets
= xmalloc(new_disks
* sizeof(int));
10446 for (i
= 0; i
< new_disks
; i
++)
10449 target_offsets
= xcalloc(new_disks
, sizeof(unsigned long long));
10451 start
= info
->reshape_progress
* 512;
10452 for (i
= 0; i
< new_disks
; i
++) {
10453 target_offsets
[i
] = (unsigned long long)
10454 __le32_to_cpu(super
->migr_rec
->ckpt_area_pba
) * 512;
10455 /* move back copy area adderss, it will be moved forward
10456 * in restore_stripes() using start input variable
10458 target_offsets
[i
] -= start
/data_disks
;
10461 if (open_backup_targets(info
, new_disks
, targets
,
10465 dest_layout
= imsm_level_to_layout(map_dest
->raid_level
);
10466 dest_chunk
= __le16_to_cpu(map_dest
->blocks_per_strip
) * 512;
10468 if (restore_stripes(targets
, /* list of dest devices */
10469 target_offsets
, /* migration record offsets */
10472 map_dest
->raid_level
,
10474 -1, /* source backup file descriptor */
10475 0, /* input buf offset
10476 * always 0 buf is already offseted */
10480 pr_err("Error restoring stripes\n");
10488 close_targets(targets
, new_disks
);
10491 free(target_offsets
);
10496 /*******************************************************************************
10497 * Function: save_checkpoint_imsm
10498 * Description: Function called for current unit status update
10499 * in the migration record. It writes it to disk.
10501 * super : imsm internal array info
10502 * info : general array info
10506 * 2: failure, means no valid migration record
10507 * / no general migration in progress /
10508 ******************************************************************************/
10509 int save_checkpoint_imsm(struct supertype
*st
, struct mdinfo
*info
, int state
)
10511 struct intel_super
*super
= st
->sb
;
10512 unsigned long long blocks_per_unit
;
10513 unsigned long long curr_migr_unit
;
10515 if (load_imsm_migr_rec(super
, info
) != 0) {
10516 dprintf("imsm: ERROR: Cannot read migration record for checkpoint save.\n");
10520 blocks_per_unit
= __le32_to_cpu(super
->migr_rec
->blocks_per_unit
);
10521 if (blocks_per_unit
== 0) {
10522 dprintf("imsm: no migration in progress.\n");
10525 curr_migr_unit
= info
->reshape_progress
/ blocks_per_unit
;
10526 /* check if array is alligned to copy area
10527 * if it is not alligned, add one to current migration unit value
10528 * this can happend on array reshape finish only
10530 if (info
->reshape_progress
% blocks_per_unit
)
10533 super
->migr_rec
->curr_migr_unit
=
10534 __cpu_to_le32(curr_migr_unit
);
10535 super
->migr_rec
->rec_status
= __cpu_to_le32(state
);
10536 super
->migr_rec
->dest_1st_member_lba
=
10537 __cpu_to_le32(curr_migr_unit
*
10538 __le32_to_cpu(super
->migr_rec
->dest_depth_per_unit
));
10539 if (write_imsm_migr_rec(st
) < 0) {
10540 dprintf("imsm: Cannot write migration record outside backup area\n");
10547 /*******************************************************************************
10548 * Function: recover_backup_imsm
10549 * Description: Function recovers critical data from the Migration Copy Area
10550 * while assembling an array.
10552 * super : imsm internal array info
10553 * info : general array info
10555 * 0 : success (or there is no data to recover)
10557 ******************************************************************************/
10558 int recover_backup_imsm(struct supertype
*st
, struct mdinfo
*info
)
10560 struct intel_super
*super
= st
->sb
;
10561 struct migr_record
*migr_rec
= super
->migr_rec
;
10562 struct imsm_map
*map_dest
;
10563 struct intel_dev
*id
= NULL
;
10564 unsigned long long read_offset
;
10565 unsigned long long write_offset
;
10567 int *targets
= NULL
;
10568 int new_disks
, i
, err
;
10571 unsigned int sector_size
= super
->sector_size
;
10572 unsigned long curr_migr_unit
= __le32_to_cpu(migr_rec
->curr_migr_unit
);
10573 unsigned long num_migr_units
= __le32_to_cpu(migr_rec
->num_migr_units
);
10575 int skipped_disks
= 0;
10577 err
= sysfs_get_str(info
, NULL
, "array_state", (char *)buffer
, 20);
10581 /* recover data only during assemblation */
10582 if (strncmp(buffer
, "inactive", 8) != 0)
10584 /* no data to recover */
10585 if (__le32_to_cpu(migr_rec
->rec_status
) == UNIT_SRC_NORMAL
)
10587 if (curr_migr_unit
>= num_migr_units
)
10590 /* find device during reshape */
10591 for (id
= super
->devlist
; id
; id
= id
->next
)
10592 if (is_gen_migration(id
->dev
))
10597 map_dest
= get_imsm_map(id
->dev
, MAP_0
);
10598 new_disks
= map_dest
->num_members
;
10600 read_offset
= (unsigned long long)
10601 __le32_to_cpu(migr_rec
->ckpt_area_pba
) * 512;
10603 write_offset
= ((unsigned long long)
10604 __le32_to_cpu(migr_rec
->dest_1st_member_lba
) +
10605 pba_of_lba0(map_dest
)) * 512;
10607 unit_len
= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
10608 if (posix_memalign((void **)&buf
, sector_size
, unit_len
) != 0)
10610 targets
= xcalloc(new_disks
, sizeof(int));
10612 if (open_backup_targets(info
, new_disks
, targets
, super
, id
->dev
)) {
10613 pr_err("Cannot open some devices belonging to array.\n");
10617 for (i
= 0; i
< new_disks
; i
++) {
10618 if (targets
[i
] < 0) {
10622 if (lseek64(targets
[i
], read_offset
, SEEK_SET
) < 0) {
10623 pr_err("Cannot seek to block: %s\n",
10628 if ((unsigned)read(targets
[i
], buf
, unit_len
) != unit_len
) {
10629 pr_err("Cannot read copy area block: %s\n",
10634 if (lseek64(targets
[i
], write_offset
, SEEK_SET
) < 0) {
10635 pr_err("Cannot seek to block: %s\n",
10640 if ((unsigned)write(targets
[i
], buf
, unit_len
) != unit_len
) {
10641 pr_err("Cannot restore block: %s\n",
10648 if (skipped_disks
> imsm_get_allowed_degradation(info
->new_level
,
10652 pr_err("Cannot restore data from backup. Too many failed disks\n");
10656 if (save_checkpoint_imsm(st
, info
, UNIT_SRC_NORMAL
)) {
10657 /* ignore error == 2, this can mean end of reshape here
10659 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL) during restart\n");
10665 for (i
= 0; i
< new_disks
; i
++)
10674 static char disk_by_path
[] = "/dev/disk/by-path/";
10676 static const char *imsm_get_disk_controller_domain(const char *path
)
10678 char disk_path
[PATH_MAX
];
10682 strcpy(disk_path
, disk_by_path
);
10683 strncat(disk_path
, path
, PATH_MAX
- strlen(disk_path
) - 1);
10684 if (stat(disk_path
, &st
) == 0) {
10685 struct sys_dev
* hba
;
10688 path
= devt_to_devpath(st
.st_rdev
);
10691 hba
= find_disk_attached_hba(-1, path
);
10692 if (hba
&& hba
->type
== SYS_DEV_SAS
)
10694 else if (hba
&& hba
->type
== SYS_DEV_SATA
)
10698 dprintf("path: %s hba: %s attached: %s\n",
10699 path
, (hba
) ? hba
->path
: "NULL", drv
);
10705 static char *imsm_find_array_devnm_by_subdev(int subdev
, char *container
)
10707 static char devnm
[32];
10708 char subdev_name
[20];
10709 struct mdstat_ent
*mdstat
;
10711 sprintf(subdev_name
, "%d", subdev
);
10712 mdstat
= mdstat_by_subdev(subdev_name
, container
);
10716 strcpy(devnm
, mdstat
->devnm
);
10717 free_mdstat(mdstat
);
10721 static int imsm_reshape_is_allowed_on_container(struct supertype
*st
,
10722 struct geo_params
*geo
,
10723 int *old_raid_disks
,
10726 /* currently we only support increasing the number of devices
10727 * for a container. This increases the number of device for each
10728 * member array. They must all be RAID0 or RAID5.
10731 struct mdinfo
*info
, *member
;
10732 int devices_that_can_grow
= 0;
10734 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): st->devnm = (%s)\n", st
->devnm
);
10736 if (geo
->size
> 0 ||
10737 geo
->level
!= UnSet
||
10738 geo
->layout
!= UnSet
||
10739 geo
->chunksize
!= 0 ||
10740 geo
->raid_disks
== UnSet
) {
10741 dprintf("imsm: Container operation is allowed for raid disks number change only.\n");
10745 if (direction
== ROLLBACK_METADATA_CHANGES
) {
10746 dprintf("imsm: Metadata changes rollback is not supported for container operation.\n");
10750 info
= container_content_imsm(st
, NULL
);
10751 for (member
= info
; member
; member
= member
->next
) {
10754 dprintf("imsm: checking device_num: %i\n",
10755 member
->container_member
);
10757 if (geo
->raid_disks
<= member
->array
.raid_disks
) {
10758 /* we work on container for Online Capacity Expansion
10759 * only so raid_disks has to grow
10761 dprintf("imsm: for container operation raid disks increase is required\n");
10765 if (info
->array
.level
!= 0 && info
->array
.level
!= 5) {
10766 /* we cannot use this container with other raid level
10768 dprintf("imsm: for container operation wrong raid level (%i) detected\n",
10769 info
->array
.level
);
10772 /* check for platform support
10773 * for this raid level configuration
10775 struct intel_super
*super
= st
->sb
;
10776 if (!is_raid_level_supported(super
->orom
,
10777 member
->array
.level
,
10778 geo
->raid_disks
)) {
10779 dprintf("platform does not support raid%d with %d disk%s\n",
10782 geo
->raid_disks
> 1 ? "s" : "");
10785 /* check if component size is aligned to chunk size
10787 if (info
->component_size
%
10788 (info
->array
.chunk_size
/512)) {
10789 dprintf("Component size is not aligned to chunk size\n");
10794 if (*old_raid_disks
&&
10795 info
->array
.raid_disks
!= *old_raid_disks
)
10797 *old_raid_disks
= info
->array
.raid_disks
;
10799 /* All raid5 and raid0 volumes in container
10800 * have to be ready for Online Capacity Expansion
10801 * so they need to be assembled. We have already
10802 * checked that no recovery etc is happening.
10804 result
= imsm_find_array_devnm_by_subdev(member
->container_member
,
10805 st
->container_devnm
);
10806 if (result
== NULL
) {
10807 dprintf("imsm: cannot find array\n");
10810 devices_that_can_grow
++;
10813 if (!member
&& devices_that_can_grow
)
10817 dprintf("Container operation allowed\n");
10819 dprintf("Error: %i\n", ret_val
);
10824 /* Function: get_spares_for_grow
10825 * Description: Allocates memory and creates list of spare devices
10826 * avaliable in container. Checks if spare drive size is acceptable.
10827 * Parameters: Pointer to the supertype structure
10828 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
10831 static struct mdinfo
*get_spares_for_grow(struct supertype
*st
)
10833 unsigned long long min_size
= min_acceptable_spare_size_imsm(st
);
10834 return container_choose_spares(st
, min_size
, NULL
, NULL
, NULL
, 0);
10837 /******************************************************************************
10838 * function: imsm_create_metadata_update_for_reshape
10839 * Function creates update for whole IMSM container.
10841 ******************************************************************************/
10842 static int imsm_create_metadata_update_for_reshape(
10843 struct supertype
*st
,
10844 struct geo_params
*geo
,
10845 int old_raid_disks
,
10846 struct imsm_update_reshape
**updatep
)
10848 struct intel_super
*super
= st
->sb
;
10849 struct imsm_super
*mpb
= super
->anchor
;
10850 int update_memory_size
;
10851 struct imsm_update_reshape
*u
;
10852 struct mdinfo
*spares
;
10855 struct mdinfo
*dev
;
10857 dprintf("(enter) raid_disks = %i\n", geo
->raid_disks
);
10859 delta_disks
= geo
->raid_disks
- old_raid_disks
;
10861 /* size of all update data without anchor */
10862 update_memory_size
= sizeof(struct imsm_update_reshape
);
10864 /* now add space for spare disks that we need to add. */
10865 update_memory_size
+= sizeof(u
->new_disks
[0]) * (delta_disks
- 1);
10867 u
= xcalloc(1, update_memory_size
);
10868 u
->type
= update_reshape_container_disks
;
10869 u
->old_raid_disks
= old_raid_disks
;
10870 u
->new_raid_disks
= geo
->raid_disks
;
10872 /* now get spare disks list
10874 spares
= get_spares_for_grow(st
);
10877 || delta_disks
> spares
->array
.spare_disks
) {
10878 pr_err("imsm: ERROR: Cannot get spare devices for %s.\n", geo
->dev_name
);
10883 /* we have got spares
10884 * update disk list in imsm_disk list table in anchor
10886 dprintf("imsm: %i spares are available.\n\n",
10887 spares
->array
.spare_disks
);
10889 dev
= spares
->devs
;
10890 for (i
= 0; i
< delta_disks
; i
++) {
10895 u
->new_disks
[i
] = makedev(dev
->disk
.major
,
10897 dl
= get_disk_super(super
, dev
->disk
.major
, dev
->disk
.minor
);
10898 dl
->index
= mpb
->num_disks
;
10906 sysfs_free(spares
);
10908 dprintf("imsm: reshape update preparation :");
10909 if (i
== delta_disks
) {
10910 dprintf_cont(" OK\n");
10912 return update_memory_size
;
10915 dprintf_cont(" Error\n");
10920 /******************************************************************************
10921 * function: imsm_create_metadata_update_for_size_change()
10922 * Creates update for IMSM array for array size change.
10924 ******************************************************************************/
10925 static int imsm_create_metadata_update_for_size_change(
10926 struct supertype
*st
,
10927 struct geo_params
*geo
,
10928 struct imsm_update_size_change
**updatep
)
10930 struct intel_super
*super
= st
->sb
;
10931 int update_memory_size
;
10932 struct imsm_update_size_change
*u
;
10934 dprintf("(enter) New size = %llu\n", geo
->size
);
10936 /* size of all update data without anchor */
10937 update_memory_size
= sizeof(struct imsm_update_size_change
);
10939 u
= xcalloc(1, update_memory_size
);
10940 u
->type
= update_size_change
;
10941 u
->subdev
= super
->current_vol
;
10942 u
->new_size
= geo
->size
;
10944 dprintf("imsm: reshape update preparation : OK\n");
10947 return update_memory_size
;
10950 /******************************************************************************
10951 * function: imsm_create_metadata_update_for_migration()
10952 * Creates update for IMSM array.
10954 ******************************************************************************/
10955 static int imsm_create_metadata_update_for_migration(
10956 struct supertype
*st
,
10957 struct geo_params
*geo
,
10958 struct imsm_update_reshape_migration
**updatep
)
10960 struct intel_super
*super
= st
->sb
;
10961 int update_memory_size
;
10962 struct imsm_update_reshape_migration
*u
;
10963 struct imsm_dev
*dev
;
10964 int previous_level
= -1;
10966 dprintf("(enter) New Level = %i\n", geo
->level
);
10968 /* size of all update data without anchor */
10969 update_memory_size
= sizeof(struct imsm_update_reshape_migration
);
10971 u
= xcalloc(1, update_memory_size
);
10972 u
->type
= update_reshape_migration
;
10973 u
->subdev
= super
->current_vol
;
10974 u
->new_level
= geo
->level
;
10975 u
->new_layout
= geo
->layout
;
10976 u
->new_raid_disks
= u
->old_raid_disks
= geo
->raid_disks
;
10977 u
->new_disks
[0] = -1;
10978 u
->new_chunksize
= -1;
10980 dev
= get_imsm_dev(super
, u
->subdev
);
10982 struct imsm_map
*map
;
10984 map
= get_imsm_map(dev
, MAP_0
);
10986 int current_chunk_size
=
10987 __le16_to_cpu(map
->blocks_per_strip
) / 2;
10989 if (geo
->chunksize
!= current_chunk_size
) {
10990 u
->new_chunksize
= geo
->chunksize
/ 1024;
10991 dprintf("imsm: chunk size change from %i to %i\n",
10992 current_chunk_size
, u
->new_chunksize
);
10994 previous_level
= map
->raid_level
;
10997 if (geo
->level
== 5 && previous_level
== 0) {
10998 struct mdinfo
*spares
= NULL
;
11000 u
->new_raid_disks
++;
11001 spares
= get_spares_for_grow(st
);
11002 if (spares
== NULL
|| spares
->array
.spare_disks
< 1) {
11004 sysfs_free(spares
);
11005 update_memory_size
= 0;
11006 pr_err("cannot get spare device for requested migration\n");
11009 sysfs_free(spares
);
11011 dprintf("imsm: reshape update preparation : OK\n");
11014 return update_memory_size
;
11017 static void imsm_update_metadata_locally(struct supertype
*st
,
11018 void *buf
, int len
)
11020 struct metadata_update mu
;
11025 mu
.space_list
= NULL
;
11027 if (imsm_prepare_update(st
, &mu
))
11028 imsm_process_update(st
, &mu
);
11030 while (mu
.space_list
) {
11031 void **space
= mu
.space_list
;
11032 mu
.space_list
= *space
;
11037 /***************************************************************************
11038 * Function: imsm_analyze_change
11039 * Description: Function analyze change for single volume
11040 * and validate if transition is supported
11041 * Parameters: Geometry parameters, supertype structure,
11042 * metadata change direction (apply/rollback)
11043 * Returns: Operation type code on success, -1 if fail
11044 ****************************************************************************/
11045 enum imsm_reshape_type
imsm_analyze_change(struct supertype
*st
,
11046 struct geo_params
*geo
,
11049 struct mdinfo info
;
11051 int check_devs
= 0;
11053 /* number of added/removed disks in operation result */
11054 int devNumChange
= 0;
11055 /* imsm compatible layout value for array geometry verification */
11056 int imsm_layout
= -1;
11058 struct imsm_dev
*dev
;
11059 struct intel_super
*super
;
11060 unsigned long long current_size
;
11061 unsigned long long free_size
;
11062 unsigned long long max_size
;
11065 getinfo_super_imsm_volume(st
, &info
, NULL
);
11066 if (geo
->level
!= info
.array
.level
&& geo
->level
>= 0 &&
11067 geo
->level
!= UnSet
) {
11068 switch (info
.array
.level
) {
11070 if (geo
->level
== 5) {
11071 change
= CH_MIGRATION
;
11072 if (geo
->layout
!= ALGORITHM_LEFT_ASYMMETRIC
) {
11073 pr_err("Error. Requested Layout not supported (left-asymmetric layout is supported only)!\n");
11075 goto analyse_change_exit
;
11077 imsm_layout
= geo
->layout
;
11079 devNumChange
= 1; /* parity disk added */
11080 } else if (geo
->level
== 10) {
11081 change
= CH_TAKEOVER
;
11083 devNumChange
= 2; /* two mirrors added */
11084 imsm_layout
= 0x102; /* imsm supported layout */
11089 if (geo
->level
== 0) {
11090 change
= CH_TAKEOVER
;
11092 devNumChange
= -(geo
->raid_disks
/2);
11093 imsm_layout
= 0; /* imsm raid0 layout */
11097 if (change
== -1) {
11098 pr_err("Error. Level Migration from %d to %d not supported!\n",
11099 info
.array
.level
, geo
->level
);
11100 goto analyse_change_exit
;
11103 geo
->level
= info
.array
.level
;
11105 if (geo
->layout
!= info
.array
.layout
&&
11106 (geo
->layout
!= UnSet
&& geo
->layout
!= -1)) {
11107 change
= CH_MIGRATION
;
11108 if (info
.array
.layout
== 0 && info
.array
.level
== 5 &&
11109 geo
->layout
== 5) {
11110 /* reshape 5 -> 4 */
11111 } else if (info
.array
.layout
== 5 && info
.array
.level
== 5 &&
11112 geo
->layout
== 0) {
11113 /* reshape 4 -> 5 */
11117 pr_err("Error. Layout Migration from %d to %d not supported!\n",
11118 info
.array
.layout
, geo
->layout
);
11120 goto analyse_change_exit
;
11123 geo
->layout
= info
.array
.layout
;
11124 if (imsm_layout
== -1)
11125 imsm_layout
= info
.array
.layout
;
11128 if (geo
->chunksize
> 0 && geo
->chunksize
!= UnSet
&&
11129 geo
->chunksize
!= info
.array
.chunk_size
) {
11130 if (info
.array
.level
== 10) {
11131 pr_err("Error. Chunk size change for RAID 10 is not supported.\n");
11133 goto analyse_change_exit
;
11134 } else if (info
.component_size
% (geo
->chunksize
/512)) {
11135 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk). Aborting...\n",
11136 geo
->chunksize
/1024, info
.component_size
/2);
11138 goto analyse_change_exit
;
11140 change
= CH_MIGRATION
;
11142 geo
->chunksize
= info
.array
.chunk_size
;
11145 chunk
= geo
->chunksize
/ 1024;
11148 dev
= get_imsm_dev(super
, super
->current_vol
);
11149 data_disks
= imsm_num_data_members(dev
, MAP_0
);
11150 /* compute current size per disk member
11152 current_size
= info
.custom_array_size
/ data_disks
;
11154 if (geo
->size
> 0 && geo
->size
!= MAX_SIZE
) {
11155 /* align component size
11157 geo
->size
= imsm_component_size_aligment_check(
11158 get_imsm_raid_level(dev
->vol
.map
),
11159 chunk
* 1024, super
->sector_size
,
11161 if (geo
->size
== 0) {
11162 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is 0).\n",
11164 goto analyse_change_exit
;
11168 if (current_size
!= geo
->size
&& geo
->size
> 0) {
11169 if (change
!= -1) {
11170 pr_err("Error. Size change should be the only one at a time.\n");
11172 goto analyse_change_exit
;
11174 if ((super
->current_vol
+ 1) != super
->anchor
->num_raid_devs
) {
11175 pr_err("Error. The last volume in container can be expanded only (%i/%s).\n",
11176 super
->current_vol
, st
->devnm
);
11177 goto analyse_change_exit
;
11179 /* check the maximum available size
11181 rv
= imsm_get_free_size(st
, dev
->vol
.map
->num_members
,
11182 0, chunk
, &free_size
);
11184 /* Cannot find maximum available space
11188 max_size
= free_size
+ current_size
;
11189 /* align component size
11191 max_size
= imsm_component_size_aligment_check(
11192 get_imsm_raid_level(dev
->vol
.map
),
11193 chunk
* 1024, super
->sector_size
,
11196 if (geo
->size
== MAX_SIZE
) {
11197 /* requested size change to the maximum available size
11199 if (max_size
== 0) {
11200 pr_err("Error. Cannot find maximum available space.\n");
11202 goto analyse_change_exit
;
11204 geo
->size
= max_size
;
11207 if (direction
== ROLLBACK_METADATA_CHANGES
) {
11208 /* accept size for rollback only
11211 /* round size due to metadata compatibility
11213 geo
->size
= (geo
->size
>> SECT_PER_MB_SHIFT
)
11214 << SECT_PER_MB_SHIFT
;
11215 dprintf("Prepare update for size change to %llu\n",
11217 if (current_size
>= geo
->size
) {
11218 pr_err("Error. Size expansion is supported only (current size is %llu, requested size /rounded/ is %llu).\n",
11219 current_size
, geo
->size
);
11220 goto analyse_change_exit
;
11222 if (max_size
&& geo
->size
> max_size
) {
11223 pr_err("Error. Requested size is larger than maximum available size (maximum available size is %llu, requested size /rounded/ is %llu).\n",
11224 max_size
, geo
->size
);
11225 goto analyse_change_exit
;
11228 geo
->size
*= data_disks
;
11229 geo
->raid_disks
= dev
->vol
.map
->num_members
;
11230 change
= CH_ARRAY_SIZE
;
11232 if (!validate_geometry_imsm(st
,
11235 geo
->raid_disks
+ devNumChange
,
11237 geo
->size
, INVALID_SECTORS
,
11238 0, 0, info
.consistency_policy
, 1))
11242 struct intel_super
*super
= st
->sb
;
11243 struct imsm_super
*mpb
= super
->anchor
;
11245 if (mpb
->num_raid_devs
> 1) {
11246 pr_err("Error. Cannot perform operation on %s- for this operation it MUST be single array in container\n",
11252 analyse_change_exit
:
11253 if (direction
== ROLLBACK_METADATA_CHANGES
&&
11254 (change
== CH_MIGRATION
|| change
== CH_TAKEOVER
)) {
11255 dprintf("imsm: Metadata changes rollback is not supported for migration and takeover operations.\n");
11261 int imsm_takeover(struct supertype
*st
, struct geo_params
*geo
)
11263 struct intel_super
*super
= st
->sb
;
11264 struct imsm_update_takeover
*u
;
11266 u
= xmalloc(sizeof(struct imsm_update_takeover
));
11268 u
->type
= update_takeover
;
11269 u
->subarray
= super
->current_vol
;
11271 /* 10->0 transition */
11272 if (geo
->level
== 0)
11273 u
->direction
= R10_TO_R0
;
11275 /* 0->10 transition */
11276 if (geo
->level
== 10)
11277 u
->direction
= R0_TO_R10
;
11279 /* update metadata locally */
11280 imsm_update_metadata_locally(st
, u
,
11281 sizeof(struct imsm_update_takeover
));
11282 /* and possibly remotely */
11283 if (st
->update_tail
)
11284 append_metadata_update(st
, u
,
11285 sizeof(struct imsm_update_takeover
));
11292 static int imsm_reshape_super(struct supertype
*st
, unsigned long long size
,
11294 int layout
, int chunksize
, int raid_disks
,
11295 int delta_disks
, char *backup
, char *dev
,
11296 int direction
, int verbose
)
11299 struct geo_params geo
;
11301 dprintf("(enter)\n");
11303 memset(&geo
, 0, sizeof(struct geo_params
));
11305 geo
.dev_name
= dev
;
11306 strcpy(geo
.devnm
, st
->devnm
);
11309 geo
.layout
= layout
;
11310 geo
.chunksize
= chunksize
;
11311 geo
.raid_disks
= raid_disks
;
11312 if (delta_disks
!= UnSet
)
11313 geo
.raid_disks
+= delta_disks
;
11315 dprintf("for level : %i\n", geo
.level
);
11316 dprintf("for raid_disks : %i\n", geo
.raid_disks
);
11318 if (experimental() == 0)
11321 if (strcmp(st
->container_devnm
, st
->devnm
) == 0) {
11322 /* On container level we can only increase number of devices. */
11323 dprintf("imsm: info: Container operation\n");
11324 int old_raid_disks
= 0;
11326 if (imsm_reshape_is_allowed_on_container(
11327 st
, &geo
, &old_raid_disks
, direction
)) {
11328 struct imsm_update_reshape
*u
= NULL
;
11331 len
= imsm_create_metadata_update_for_reshape(
11332 st
, &geo
, old_raid_disks
, &u
);
11335 dprintf("imsm: Cannot prepare update\n");
11336 goto exit_imsm_reshape_super
;
11340 /* update metadata locally */
11341 imsm_update_metadata_locally(st
, u
, len
);
11342 /* and possibly remotely */
11343 if (st
->update_tail
)
11344 append_metadata_update(st
, u
, len
);
11349 pr_err("(imsm) Operation is not allowed on this container\n");
11352 /* On volume level we support following operations
11353 * - takeover: raid10 -> raid0; raid0 -> raid10
11354 * - chunk size migration
11355 * - migration: raid5 -> raid0; raid0 -> raid5
11357 struct intel_super
*super
= st
->sb
;
11358 struct intel_dev
*dev
= super
->devlist
;
11360 dprintf("imsm: info: Volume operation\n");
11361 /* find requested device */
11364 imsm_find_array_devnm_by_subdev(
11365 dev
->index
, st
->container_devnm
);
11366 if (devnm
&& strcmp(devnm
, geo
.devnm
) == 0)
11371 pr_err("Cannot find %s (%s) subarray\n",
11372 geo
.dev_name
, geo
.devnm
);
11373 goto exit_imsm_reshape_super
;
11375 super
->current_vol
= dev
->index
;
11376 change
= imsm_analyze_change(st
, &geo
, direction
);
11379 ret_val
= imsm_takeover(st
, &geo
);
11381 case CH_MIGRATION
: {
11382 struct imsm_update_reshape_migration
*u
= NULL
;
11384 imsm_create_metadata_update_for_migration(
11387 dprintf("imsm: Cannot prepare update\n");
11391 /* update metadata locally */
11392 imsm_update_metadata_locally(st
, u
, len
);
11393 /* and possibly remotely */
11394 if (st
->update_tail
)
11395 append_metadata_update(st
, u
, len
);
11400 case CH_ARRAY_SIZE
: {
11401 struct imsm_update_size_change
*u
= NULL
;
11403 imsm_create_metadata_update_for_size_change(
11406 dprintf("imsm: Cannot prepare update\n");
11410 /* update metadata locally */
11411 imsm_update_metadata_locally(st
, u
, len
);
11412 /* and possibly remotely */
11413 if (st
->update_tail
)
11414 append_metadata_update(st
, u
, len
);
11424 exit_imsm_reshape_super
:
11425 dprintf("imsm: reshape_super Exit code = %i\n", ret_val
);
11429 #define COMPLETED_OK 0
11430 #define COMPLETED_NONE 1
11431 #define COMPLETED_DELAYED 2
11433 static int read_completed(int fd
, unsigned long long *val
)
11438 ret
= sysfs_fd_get_str(fd
, buf
, 50);
11442 ret
= COMPLETED_OK
;
11443 if (strncmp(buf
, "none", 4) == 0) {
11444 ret
= COMPLETED_NONE
;
11445 } else if (strncmp(buf
, "delayed", 7) == 0) {
11446 ret
= COMPLETED_DELAYED
;
11449 *val
= strtoull(buf
, &ep
, 0);
11450 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
11456 /*******************************************************************************
11457 * Function: wait_for_reshape_imsm
11458 * Description: Function writes new sync_max value and waits until
11459 * reshape process reach new position
11461 * sra : general array info
11462 * ndata : number of disks in new array's layout
11465 * 1 : there is no reshape in progress,
11467 ******************************************************************************/
11468 int wait_for_reshape_imsm(struct mdinfo
*sra
, int ndata
)
11470 int fd
= sysfs_get_fd(sra
, NULL
, "sync_completed");
11472 unsigned long long completed
;
11473 /* to_complete : new sync_max position */
11474 unsigned long long to_complete
= sra
->reshape_progress
;
11475 unsigned long long position_to_set
= to_complete
/ ndata
;
11478 dprintf("cannot open reshape_position\n");
11483 if (sysfs_fd_get_ll(fd
, &completed
) < 0) {
11485 dprintf("cannot read reshape_position (no reshape in progres)\n");
11494 if (completed
> position_to_set
) {
11495 dprintf("wrong next position to set %llu (%llu)\n",
11496 to_complete
, position_to_set
);
11500 dprintf("Position set: %llu\n", position_to_set
);
11501 if (sysfs_set_num(sra
, NULL
, "sync_max",
11502 position_to_set
) != 0) {
11503 dprintf("cannot set reshape position to %llu\n",
11512 int timeout
= 3000;
11514 sysfs_wait(fd
, &timeout
);
11515 if (sysfs_get_str(sra
, NULL
, "sync_action",
11517 strncmp(action
, "reshape", 7) != 0) {
11518 if (strncmp(action
, "idle", 4) == 0)
11524 rc
= read_completed(fd
, &completed
);
11526 dprintf("cannot read reshape_position (in loop)\n");
11529 } else if (rc
== COMPLETED_NONE
)
11531 } while (completed
< position_to_set
);
11537 /*******************************************************************************
11538 * Function: check_degradation_change
11539 * Description: Check that array hasn't become failed.
11541 * info : for sysfs access
11542 * sources : source disks descriptors
11543 * degraded: previous degradation level
11545 * degradation level
11546 ******************************************************************************/
11547 int check_degradation_change(struct mdinfo
*info
,
11551 unsigned long long new_degraded
;
11554 rv
= sysfs_get_ll(info
, NULL
, "degraded", &new_degraded
);
11555 if (rv
== -1 || (new_degraded
!= (unsigned long long)degraded
)) {
11556 /* check each device to ensure it is still working */
11559 for (sd
= info
->devs
; sd
; sd
= sd
->next
) {
11560 if (sd
->disk
.state
& (1<<MD_DISK_FAULTY
))
11562 if (sd
->disk
.state
& (1<<MD_DISK_SYNC
)) {
11565 if (sysfs_get_str(info
,
11566 sd
, "state", sbuf
, sizeof(sbuf
)) < 0 ||
11567 strstr(sbuf
, "faulty") ||
11568 strstr(sbuf
, "in_sync") == NULL
) {
11569 /* this device is dead */
11570 sd
->disk
.state
= (1<<MD_DISK_FAULTY
);
11571 if (sd
->disk
.raid_disk
>= 0 &&
11572 sources
[sd
->disk
.raid_disk
] >= 0) {
11574 sd
->disk
.raid_disk
]);
11575 sources
[sd
->disk
.raid_disk
] =
11584 return new_degraded
;
11587 /*******************************************************************************
11588 * Function: imsm_manage_reshape
11589 * Description: Function finds array under reshape and it manages reshape
11590 * process. It creates stripes backups (if required) and sets
11593 * afd : Backup handle (nattive) - not used
11594 * sra : general array info
11595 * reshape : reshape parameters - not used
11596 * st : supertype structure
11597 * blocks : size of critical section [blocks]
11598 * fds : table of source device descriptor
11599 * offsets : start of array (offest per devices)
11601 * destfd : table of destination device descriptor
11602 * destoffsets : table of destination offsets (per device)
11604 * 1 : success, reshape is done
11606 ******************************************************************************/
11607 static int imsm_manage_reshape(
11608 int afd
, struct mdinfo
*sra
, struct reshape
*reshape
,
11609 struct supertype
*st
, unsigned long backup_blocks
,
11610 int *fds
, unsigned long long *offsets
,
11611 int dests
, int *destfd
, unsigned long long *destoffsets
)
11614 struct intel_super
*super
= st
->sb
;
11615 struct intel_dev
*dv
;
11616 unsigned int sector_size
= super
->sector_size
;
11617 struct imsm_dev
*dev
= NULL
;
11618 struct imsm_map
*map_src
;
11619 int migr_vol_qan
= 0;
11620 int ndata
, odata
; /* [bytes] */
11621 int chunk
; /* [bytes] */
11622 struct migr_record
*migr_rec
;
11624 unsigned int buf_size
; /* [bytes] */
11625 unsigned long long max_position
; /* array size [bytes] */
11626 unsigned long long next_step
; /* [blocks]/[bytes] */
11627 unsigned long long old_data_stripe_length
;
11628 unsigned long long start_src
; /* [bytes] */
11629 unsigned long long start
; /* [bytes] */
11630 unsigned long long start_buf_shift
; /* [bytes] */
11632 int source_layout
= 0;
11637 if (!fds
|| !offsets
)
11640 /* Find volume during the reshape */
11641 for (dv
= super
->devlist
; dv
; dv
= dv
->next
) {
11642 if (dv
->dev
->vol
.migr_type
== MIGR_GEN_MIGR
11643 && dv
->dev
->vol
.migr_state
== 1) {
11648 /* Only one volume can migrate at the same time */
11649 if (migr_vol_qan
!= 1) {
11650 pr_err("%s", migr_vol_qan
?
11651 "Number of migrating volumes greater than 1\n" :
11652 "There is no volume during migrationg\n");
11656 map_src
= get_imsm_map(dev
, MAP_1
);
11657 if (map_src
== NULL
)
11660 ndata
= imsm_num_data_members(dev
, MAP_0
);
11661 odata
= imsm_num_data_members(dev
, MAP_1
);
11663 chunk
= __le16_to_cpu(map_src
->blocks_per_strip
) * 512;
11664 old_data_stripe_length
= odata
* chunk
;
11666 migr_rec
= super
->migr_rec
;
11668 /* initialize migration record for start condition */
11669 if (sra
->reshape_progress
== 0)
11670 init_migr_record_imsm(st
, dev
, sra
);
11672 if (__le32_to_cpu(migr_rec
->rec_status
) != UNIT_SRC_NORMAL
) {
11673 dprintf("imsm: cannot restart migration when data are present in copy area.\n");
11676 /* Save checkpoint to update migration record for current
11677 * reshape position (in md). It can be farther than current
11678 * reshape position in metadata.
11680 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11681 /* ignore error == 2, this can mean end of reshape here
11683 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL, initial save)\n");
11688 /* size for data */
11689 buf_size
= __le32_to_cpu(migr_rec
->blocks_per_unit
) * 512;
11690 /* extend buffer size for parity disk */
11691 buf_size
+= __le32_to_cpu(migr_rec
->dest_depth_per_unit
) * 512;
11692 /* add space for stripe aligment */
11693 buf_size
+= old_data_stripe_length
;
11694 if (posix_memalign((void **)&buf
, MAX_SECTOR_SIZE
, buf_size
)) {
11695 dprintf("imsm: Cannot allocate checkpoint buffer\n");
11699 max_position
= sra
->component_size
* ndata
;
11700 source_layout
= imsm_level_to_layout(map_src
->raid_level
);
11702 while (__le32_to_cpu(migr_rec
->curr_migr_unit
) <
11703 __le32_to_cpu(migr_rec
->num_migr_units
)) {
11704 /* current reshape position [blocks] */
11705 unsigned long long current_position
=
11706 __le32_to_cpu(migr_rec
->blocks_per_unit
)
11707 * __le32_to_cpu(migr_rec
->curr_migr_unit
);
11708 unsigned long long border
;
11710 /* Check that array hasn't become failed.
11712 degraded
= check_degradation_change(sra
, fds
, degraded
);
11713 if (degraded
> 1) {
11714 dprintf("imsm: Abort reshape due to degradation level (%i)\n", degraded
);
11718 next_step
= __le32_to_cpu(migr_rec
->blocks_per_unit
);
11720 if ((current_position
+ next_step
) > max_position
)
11721 next_step
= max_position
- current_position
;
11723 start
= current_position
* 512;
11725 /* align reading start to old geometry */
11726 start_buf_shift
= start
% old_data_stripe_length
;
11727 start_src
= start
- start_buf_shift
;
11729 border
= (start_src
/ odata
) - (start
/ ndata
);
11731 if (border
<= __le32_to_cpu(migr_rec
->dest_depth_per_unit
)) {
11732 /* save critical stripes to buf
11733 * start - start address of current unit
11734 * to backup [bytes]
11735 * start_src - start address of current unit
11736 * to backup alligned to source array
11739 unsigned long long next_step_filler
;
11740 unsigned long long copy_length
= next_step
* 512;
11742 /* allign copy area length to stripe in old geometry */
11743 next_step_filler
= ((copy_length
+ start_buf_shift
)
11744 % old_data_stripe_length
);
11745 if (next_step_filler
)
11746 next_step_filler
= (old_data_stripe_length
11747 - next_step_filler
);
11748 dprintf("save_stripes() parameters: start = %llu,\tstart_src = %llu,\tnext_step*512 = %llu,\tstart_in_buf_shift = %llu,\tnext_step_filler = %llu\n",
11749 start
, start_src
, copy_length
,
11750 start_buf_shift
, next_step_filler
);
11752 if (save_stripes(fds
, offsets
, map_src
->num_members
,
11753 chunk
, map_src
->raid_level
,
11754 source_layout
, 0, NULL
, start_src
,
11756 next_step_filler
+ start_buf_shift
,
11758 dprintf("imsm: Cannot save stripes to buffer\n");
11761 /* Convert data to destination format and store it
11762 * in backup general migration area
11764 if (save_backup_imsm(st
, dev
, sra
,
11765 buf
+ start_buf_shift
, copy_length
)) {
11766 dprintf("imsm: Cannot save stripes to target devices\n");
11769 if (save_checkpoint_imsm(st
, sra
,
11770 UNIT_SRC_IN_CP_AREA
)) {
11771 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_IN_CP_AREA)\n");
11775 /* set next step to use whole border area */
11776 border
/= next_step
;
11778 next_step
*= border
;
11780 /* When data backed up, checkpoint stored,
11781 * kick the kernel to reshape unit of data
11783 next_step
= next_step
+ sra
->reshape_progress
;
11784 /* limit next step to array max position */
11785 if (next_step
> max_position
)
11786 next_step
= max_position
;
11787 sysfs_set_num(sra
, NULL
, "suspend_lo", sra
->reshape_progress
);
11788 sysfs_set_num(sra
, NULL
, "suspend_hi", next_step
);
11789 sra
->reshape_progress
= next_step
;
11791 /* wait until reshape finish */
11792 if (wait_for_reshape_imsm(sra
, ndata
)) {
11793 dprintf("wait_for_reshape_imsm returned error!\n");
11799 if (save_checkpoint_imsm(st
, sra
, UNIT_SRC_NORMAL
) == 1) {
11800 /* ignore error == 2, this can mean end of reshape here
11802 dprintf("imsm: Cannot write checkpoint to migration record (UNIT_SRC_NORMAL)\n");
11808 /* clear migr_rec on disks after successful migration */
11811 memset(super
->migr_rec_buf
, 0, MIGR_REC_BUF_SECTORS
*sector_size
);
11812 for (d
= super
->disks
; d
; d
= d
->next
) {
11813 if (d
->index
< 0 || is_failed(&d
->disk
))
11815 unsigned long long dsize
;
11817 get_dev_size(d
->fd
, NULL
, &dsize
);
11818 if (lseek64(d
->fd
, dsize
- MIGR_REC_SECTOR_POSITION
*sector_size
,
11820 if ((unsigned int)write(d
->fd
, super
->migr_rec_buf
,
11821 MIGR_REC_BUF_SECTORS
*sector_size
) !=
11822 MIGR_REC_BUF_SECTORS
*sector_size
)
11823 perror("Write migr_rec failed");
11827 /* return '1' if done */
11831 /* See Grow.c: abort_reshape() for further explanation */
11832 sysfs_set_num(sra
, NULL
, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL
);
11833 sysfs_set_num(sra
, NULL
, "suspend_hi", 0);
11834 sysfs_set_num(sra
, NULL
, "suspend_lo", 0);
11839 #endif /* MDASSEMBLE */
11841 struct superswitch super_imsm
= {
11843 .examine_super
= examine_super_imsm
,
11844 .brief_examine_super
= brief_examine_super_imsm
,
11845 .brief_examine_subarrays
= brief_examine_subarrays_imsm
,
11846 .export_examine_super
= export_examine_super_imsm
,
11847 .detail_super
= detail_super_imsm
,
11848 .brief_detail_super
= brief_detail_super_imsm
,
11849 .write_init_super
= write_init_super_imsm
,
11850 .validate_geometry
= validate_geometry_imsm
,
11851 .add_to_super
= add_to_super_imsm
,
11852 .remove_from_super
= remove_from_super_imsm
,
11853 .detail_platform
= detail_platform_imsm
,
11854 .export_detail_platform
= export_detail_platform_imsm
,
11855 .kill_subarray
= kill_subarray_imsm
,
11856 .update_subarray
= update_subarray_imsm
,
11857 .load_container
= load_container_imsm
,
11858 .default_geometry
= default_geometry_imsm
,
11859 .get_disk_controller_domain
= imsm_get_disk_controller_domain
,
11860 .reshape_super
= imsm_reshape_super
,
11861 .manage_reshape
= imsm_manage_reshape
,
11862 .recover_backup
= recover_backup_imsm
,
11863 .copy_metadata
= copy_metadata_imsm
,
11864 .examine_badblocks
= examine_badblocks_imsm
,
11866 .match_home
= match_home_imsm
,
11867 .uuid_from_super
= uuid_from_super_imsm
,
11868 .getinfo_super
= getinfo_super_imsm
,
11869 .getinfo_super_disks
= getinfo_super_disks_imsm
,
11870 .update_super
= update_super_imsm
,
11872 .avail_size
= avail_size_imsm
,
11873 .min_acceptable_spare_size
= min_acceptable_spare_size_imsm
,
11875 .compare_super
= compare_super_imsm
,
11877 .load_super
= load_super_imsm
,
11878 .init_super
= init_super_imsm
,
11879 .store_super
= store_super_imsm
,
11880 .free_super
= free_super_imsm
,
11881 .match_metadata_desc
= match_metadata_desc_imsm
,
11882 .container_content
= container_content_imsm
,
11883 .validate_container
= validate_container_imsm
,
11885 .write_init_ppl
= write_init_ppl_imsm
,
11886 .validate_ppl
= validate_ppl_imsm
,
11893 .open_new
= imsm_open_new
,
11894 .set_array_state
= imsm_set_array_state
,
11895 .set_disk
= imsm_set_disk
,
11896 .sync_metadata
= imsm_sync_metadata
,
11897 .activate_spare
= imsm_activate_spare
,
11898 .process_update
= imsm_process_update
,
11899 .prepare_update
= imsm_prepare_update
,
11900 .record_bad_block
= imsm_record_badblock
,
11901 .clear_bad_block
= imsm_clear_badblock
,
11902 .get_bad_blocks
= imsm_get_badblocks
,
11903 #endif /* MDASSEMBLE */