]> git.ipfire.org Git - thirdparty/mdadm.git/blame_incremental - super-intel.c
imsm: FIX: Correct ords merging in end_migration()
[thirdparty/mdadm.git] / super-intel.c
... / ...
CommitLineData
1/*
2 * mdadm - Intel(R) Matrix Storage Manager Support
3 *
4 * Copyright (C) 2002-2008 Intel Corporation
5 *
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.
9 *
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
13 * more details.
14 *
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.
18 */
19
20#define HAVE_STDINT_H 1
21#include "mdadm.h"
22#include "mdmon.h"
23#include "sha1.h"
24#include "platform-intel.h"
25#include <values.h>
26#include <scsi/sg.h>
27#include <ctype.h>
28#include <dirent.h>
29
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
43
44/* supports RAID0 */
45#define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
46/* supports RAID1 */
47#define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
48/* supports RAID10 */
49#define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50/* supports RAID1E */
51#define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
52/* supports RAID5 */
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)
58
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)
65
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)
74
75/* Define all supported attributes that have to be accepted by mdadm
76 */
77#define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
78 MPB_ATTRIB_2TB | \
79 MPB_ATTRIB_2TB_DISK | \
80 MPB_ATTRIB_RAID0 | \
81 MPB_ATTRIB_RAID1 | \
82 MPB_ATTRIB_RAID10 | \
83 MPB_ATTRIB_RAID5 | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE)
85
86/* Define attributes that are unused but not harmful */
87#define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
88
89#define MPB_SECTOR_CNT 2210
90#define IMSM_RESERVED_SECTORS 4096
91#define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
92#define SECT_PER_MB_SHIFT 11
93
94/* Disk configuration info. */
95#define IMSM_MAX_DEVICES 255
96struct imsm_disk {
97 __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
98 __u32 total_blocks; /* 0xE8 - 0xEB total blocks */
99 __u32 scsi_id; /* 0xEC - 0xEF scsi ID */
100#define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
101#define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
102#define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
103 __u32 status; /* 0xF0 - 0xF3 */
104 __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
105#define IMSM_DISK_FILLERS 4
106 __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
107};
108
109/* map selector for map managment
110 */
111#define MAP_0 2
112#define MAP_1 4
113
114/* RAID map configuration infos. */
115struct imsm_map {
116 __u32 pba_of_lba0; /* start address of partition */
117 __u32 blocks_per_member;/* blocks per member */
118 __u32 num_data_stripes; /* number of data stripes */
119 __u16 blocks_per_strip;
120 __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */
121#define IMSM_T_STATE_NORMAL 0
122#define IMSM_T_STATE_UNINITIALIZED 1
123#define IMSM_T_STATE_DEGRADED 2
124#define IMSM_T_STATE_FAILED 3
125 __u8 raid_level;
126#define IMSM_T_RAID0 0
127#define IMSM_T_RAID1 1
128#define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
129 __u8 num_members; /* number of member disks */
130 __u8 num_domains; /* number of parity domains */
131 __u8 failed_disk_num; /* valid only when state is degraded */
132 __u8 ddf;
133 __u32 filler[7]; /* expansion area */
134#define IMSM_ORD_REBUILD (1 << 24)
135 __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
136 * top byte contains some flags
137 */
138} __attribute__ ((packed));
139
140struct imsm_vol {
141 __u32 curr_migr_unit;
142 __u32 checkpoint_id; /* id to access curr_migr_unit */
143 __u8 migr_state; /* Normal or Migrating */
144#define MIGR_INIT 0
145#define MIGR_REBUILD 1
146#define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
147#define MIGR_GEN_MIGR 3
148#define MIGR_STATE_CHANGE 4
149#define MIGR_REPAIR 5
150 __u8 migr_type; /* Initializing, Rebuilding, ... */
151 __u8 dirty;
152 __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */
153 __u16 verify_errors; /* number of mismatches */
154 __u16 bad_blocks; /* number of bad blocks during verify */
155 __u32 filler[4];
156 struct imsm_map map[1];
157 /* here comes another one if migr_state */
158} __attribute__ ((packed));
159
160struct imsm_dev {
161 __u8 volume[MAX_RAID_SERIAL_LEN];
162 __u32 size_low;
163 __u32 size_high;
164#define DEV_BOOTABLE __cpu_to_le32(0x01)
165#define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
166#define DEV_READ_COALESCING __cpu_to_le32(0x04)
167#define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
168#define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
169#define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
170#define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
171#define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
172#define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
173#define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
174#define DEV_CLONE_N_GO __cpu_to_le32(0x400)
175#define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
176#define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
177 __u32 status; /* Persistent RaidDev status */
178 __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
179 __u8 migr_priority;
180 __u8 num_sub_vols;
181 __u8 tid;
182 __u8 cng_master_disk;
183 __u16 cache_policy;
184 __u8 cng_state;
185 __u8 cng_sub_state;
186#define IMSM_DEV_FILLERS 10
187 __u32 filler[IMSM_DEV_FILLERS];
188 struct imsm_vol vol;
189} __attribute__ ((packed));
190
191struct imsm_super {
192 __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
193 __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */
194 __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */
195 __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */
196 __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
197 __u32 error_log_size; /* 0x30 - 0x33 in bytes */
198 __u32 attributes; /* 0x34 - 0x37 */
199 __u8 num_disks; /* 0x38 Number of configured disks */
200 __u8 num_raid_devs; /* 0x39 Number of configured volumes */
201 __u8 error_log_pos; /* 0x3A */
202 __u8 fill[1]; /* 0x3B */
203 __u32 cache_size; /* 0x3c - 0x40 in mb */
204 __u32 orig_family_num; /* 0x40 - 0x43 original family num */
205 __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */
206 __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
207#define IMSM_FILLERS 35
208 __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
209 struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */
210 /* here comes imsm_dev[num_raid_devs] */
211 /* here comes BBM logs */
212} __attribute__ ((packed));
213
214#define BBM_LOG_MAX_ENTRIES 254
215
216struct bbm_log_entry {
217 __u64 defective_block_start;
218#define UNREADABLE 0xFFFFFFFF
219 __u32 spare_block_offset;
220 __u16 remapped_marked_count;
221 __u16 disk_ordinal;
222} __attribute__ ((__packed__));
223
224struct bbm_log {
225 __u32 signature; /* 0xABADB10C */
226 __u32 entry_count;
227 __u32 reserved_spare_block_count; /* 0 */
228 __u32 reserved; /* 0xFFFF */
229 __u64 first_spare_lba;
230 struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
231} __attribute__ ((__packed__));
232
233
234#ifndef MDASSEMBLE
235static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
236#endif
237
238#define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
239
240#define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
241
242#define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
243 * be recovered using srcMap */
244#define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
245 * already been migrated and must
246 * be recovered from checkpoint area */
247struct migr_record {
248 __u32 rec_status; /* Status used to determine how to restart
249 * migration in case it aborts
250 * in some fashion */
251 __u32 curr_migr_unit; /* 0..numMigrUnits-1 */
252 __u32 family_num; /* Family number of MPB
253 * containing the RaidDev
254 * that is migrating */
255 __u32 ascending_migr; /* True if migrating in increasing
256 * order of lbas */
257 __u32 blocks_per_unit; /* Num disk blocks per unit of operation */
258 __u32 dest_depth_per_unit; /* Num member blocks each destMap
259 * member disk
260 * advances per unit-of-operation */
261 __u32 ckpt_area_pba; /* Pba of first block of ckpt copy area */
262 __u32 dest_1st_member_lba; /* First member lba on first
263 * stripe of destination */
264 __u32 num_migr_units; /* Total num migration units-of-op */
265 __u32 post_migr_vol_cap; /* Size of volume after
266 * migration completes */
267 __u32 post_migr_vol_cap_hi; /* Expansion space for LBA64 */
268 __u32 ckpt_read_disk_num; /* Which member disk in destSubMap[0] the
269 * migration ckpt record was read from
270 * (for recovered migrations) */
271} __attribute__ ((__packed__));
272
273static __u8 migr_type(struct imsm_dev *dev)
274{
275 if (dev->vol.migr_type == MIGR_VERIFY &&
276 dev->status & DEV_VERIFY_AND_FIX)
277 return MIGR_REPAIR;
278 else
279 return dev->vol.migr_type;
280}
281
282static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
283{
284 /* for compatibility with older oroms convert MIGR_REPAIR, into
285 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
286 */
287 if (migr_type == MIGR_REPAIR) {
288 dev->vol.migr_type = MIGR_VERIFY;
289 dev->status |= DEV_VERIFY_AND_FIX;
290 } else {
291 dev->vol.migr_type = migr_type;
292 dev->status &= ~DEV_VERIFY_AND_FIX;
293 }
294}
295
296static unsigned int sector_count(__u32 bytes)
297{
298 return ((bytes + (512-1)) & (~(512-1))) / 512;
299}
300
301static unsigned int mpb_sectors(struct imsm_super *mpb)
302{
303 return sector_count(__le32_to_cpu(mpb->mpb_size));
304}
305
306struct intel_dev {
307 struct imsm_dev *dev;
308 struct intel_dev *next;
309 unsigned index;
310};
311
312struct intel_hba {
313 enum sys_dev_type type;
314 char *path;
315 char *pci_id;
316 struct intel_hba *next;
317};
318
319enum action {
320 DISK_REMOVE = 1,
321 DISK_ADD
322};
323/* internal representation of IMSM metadata */
324struct intel_super {
325 union {
326 void *buf; /* O_DIRECT buffer for reading/writing metadata */
327 struct imsm_super *anchor; /* immovable parameters */
328 };
329 union {
330 void *migr_rec_buf; /* buffer for I/O operations */
331 struct migr_record *migr_rec; /* migration record */
332 };
333 size_t len; /* size of the 'buf' allocation */
334 void *next_buf; /* for realloc'ing buf from the manager */
335 size_t next_len;
336 int updates_pending; /* count of pending updates for mdmon */
337 int current_vol; /* index of raid device undergoing creation */
338 __u32 create_offset; /* common start for 'current_vol' */
339 __u32 random; /* random data for seeding new family numbers */
340 struct intel_dev *devlist;
341 struct dl {
342 struct dl *next;
343 int index;
344 __u8 serial[MAX_RAID_SERIAL_LEN];
345 int major, minor;
346 char *devname;
347 struct imsm_disk disk;
348 int fd;
349 int extent_cnt;
350 struct extent *e; /* for determining freespace @ create */
351 int raiddisk; /* slot to fill in autolayout */
352 enum action action;
353 } *disks, *current_disk;
354 struct dl *disk_mgmt_list; /* list of disks to add/remove while mdmon
355 active */
356 struct dl *missing; /* disks removed while we weren't looking */
357 struct bbm_log *bbm_log;
358 struct intel_hba *hba; /* device path of the raid controller for this metadata */
359 const struct imsm_orom *orom; /* platform firmware support */
360 struct intel_super *next; /* (temp) list for disambiguating family_num */
361};
362
363struct intel_disk {
364 struct imsm_disk disk;
365 #define IMSM_UNKNOWN_OWNER (-1)
366 int owner;
367 struct intel_disk *next;
368};
369
370struct extent {
371 unsigned long long start, size;
372};
373
374/* definitions of reshape process types */
375enum imsm_reshape_type {
376 CH_TAKEOVER,
377 CH_MIGRATION,
378};
379
380/* definition of messages passed to imsm_process_update */
381enum imsm_update_type {
382 update_activate_spare,
383 update_create_array,
384 update_kill_array,
385 update_rename_array,
386 update_add_remove_disk,
387 update_reshape_container_disks,
388 update_reshape_migration,
389 update_takeover,
390 update_general_migration_checkpoint,
391};
392
393struct imsm_update_activate_spare {
394 enum imsm_update_type type;
395 struct dl *dl;
396 int slot;
397 int array;
398 struct imsm_update_activate_spare *next;
399};
400
401struct geo_params {
402 int dev_id;
403 char *dev_name;
404 long long size;
405 int level;
406 int layout;
407 int chunksize;
408 int raid_disks;
409};
410
411enum takeover_direction {
412 R10_TO_R0,
413 R0_TO_R10
414};
415struct imsm_update_takeover {
416 enum imsm_update_type type;
417 int subarray;
418 enum takeover_direction direction;
419};
420
421struct imsm_update_reshape {
422 enum imsm_update_type type;
423 int old_raid_disks;
424 int new_raid_disks;
425
426 int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */
427};
428
429struct imsm_update_reshape_migration {
430 enum imsm_update_type type;
431 int old_raid_disks;
432 int new_raid_disks;
433 /* fields for array migration changes
434 */
435 int subdev;
436 int new_level;
437 int new_layout;
438 int new_chunksize;
439
440 int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */
441};
442
443struct imsm_update_general_migration_checkpoint {
444 enum imsm_update_type type;
445 __u32 curr_migr_unit;
446};
447
448struct disk_info {
449 __u8 serial[MAX_RAID_SERIAL_LEN];
450};
451
452struct imsm_update_create_array {
453 enum imsm_update_type type;
454 int dev_idx;
455 struct imsm_dev dev;
456};
457
458struct imsm_update_kill_array {
459 enum imsm_update_type type;
460 int dev_idx;
461};
462
463struct imsm_update_rename_array {
464 enum imsm_update_type type;
465 __u8 name[MAX_RAID_SERIAL_LEN];
466 int dev_idx;
467};
468
469struct imsm_update_add_remove_disk {
470 enum imsm_update_type type;
471};
472
473
474static const char *_sys_dev_type[] = {
475 [SYS_DEV_UNKNOWN] = "Unknown",
476 [SYS_DEV_SAS] = "SAS",
477 [SYS_DEV_SATA] = "SATA"
478};
479
480const char *get_sys_dev_type(enum sys_dev_type type)
481{
482 if (type >= SYS_DEV_MAX)
483 type = SYS_DEV_UNKNOWN;
484
485 return _sys_dev_type[type];
486}
487
488static struct intel_hba * alloc_intel_hba(struct sys_dev *device)
489{
490 struct intel_hba *result = malloc(sizeof(*result));
491 if (result) {
492 result->type = device->type;
493 result->path = strdup(device->path);
494 result->next = NULL;
495 if (result->path && (result->pci_id = strrchr(result->path, '/')) != NULL)
496 result->pci_id++;
497 }
498 return result;
499}
500
501static struct intel_hba * find_intel_hba(struct intel_hba *hba, struct sys_dev *device)
502{
503 struct intel_hba *result=NULL;
504 for (result = hba; result; result = result->next) {
505 if (result->type == device->type && strcmp(result->path, device->path) == 0)
506 break;
507 }
508 return result;
509}
510
511static int attach_hba_to_super(struct intel_super *super, struct sys_dev *device)
512{
513 struct intel_hba *hba;
514
515 /* check if disk attached to Intel HBA */
516 hba = find_intel_hba(super->hba, device);
517 if (hba != NULL)
518 return 1;
519 /* Check if HBA is already attached to super */
520 if (super->hba == NULL) {
521 super->hba = alloc_intel_hba(device);
522 return 1;
523 }
524
525 hba = super->hba;
526 /* Intel metadata allows for all disks attached to the same type HBA.
527 * Do not sypport odf HBA types mixing
528 */
529 if (device->type != hba->type)
530 return 2;
531
532 while (hba->next)
533 hba = hba->next;
534
535 hba->next = alloc_intel_hba(device);
536 return 1;
537}
538
539static struct sys_dev* find_disk_attached_hba(int fd, const char *devname)
540{
541 struct sys_dev *list, *elem, *prev;
542 char *disk_path;
543
544 if ((list = find_intel_devices()) == NULL)
545 return 0;
546
547 if (fd < 0)
548 disk_path = (char *) devname;
549 else
550 disk_path = diskfd_to_devpath(fd);
551
552 if (!disk_path) {
553 free_sys_dev(&list);
554 return 0;
555 }
556
557 for (prev = NULL, elem = list; elem; prev = elem, elem = elem->next) {
558 if (path_attached_to_hba(disk_path, elem->path)) {
559 if (prev == NULL)
560 list = list->next;
561 else
562 prev->next = elem->next;
563 elem->next = NULL;
564 if (disk_path != devname)
565 free(disk_path);
566 free_sys_dev(&list);
567 return elem;
568 }
569 }
570 if (disk_path != devname)
571 free(disk_path);
572 free_sys_dev(&list);
573
574 return NULL;
575}
576
577
578static int find_intel_hba_capability(int fd, struct intel_super *super,
579 char *devname);
580
581static struct supertype *match_metadata_desc_imsm(char *arg)
582{
583 struct supertype *st;
584
585 if (strcmp(arg, "imsm") != 0 &&
586 strcmp(arg, "default") != 0
587 )
588 return NULL;
589
590 st = malloc(sizeof(*st));
591 if (!st)
592 return NULL;
593 memset(st, 0, sizeof(*st));
594 st->container_dev = NoMdDev;
595 st->ss = &super_imsm;
596 st->max_devs = IMSM_MAX_DEVICES;
597 st->minor_version = 0;
598 st->sb = NULL;
599 return st;
600}
601
602#ifndef MDASSEMBLE
603static __u8 *get_imsm_version(struct imsm_super *mpb)
604{
605 return &mpb->sig[MPB_SIG_LEN];
606}
607#endif
608
609/* retrieve a disk directly from the anchor when the anchor is known to be
610 * up-to-date, currently only at load time
611 */
612static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
613{
614 if (index >= mpb->num_disks)
615 return NULL;
616 return &mpb->disk[index];
617}
618
619/* retrieve the disk description based on a index of the disk
620 * in the sub-array
621 */
622static struct dl *get_imsm_dl_disk(struct intel_super *super, __u8 index)
623{
624 struct dl *d;
625
626 for (d = super->disks; d; d = d->next)
627 if (d->index == index)
628 return d;
629
630 return NULL;
631}
632/* retrieve a disk from the parsed metadata */
633static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
634{
635 struct dl *dl;
636
637 dl = get_imsm_dl_disk(super, index);
638 if (dl)
639 return &dl->disk;
640
641 return NULL;
642}
643
644/* generate a checksum directly from the anchor when the anchor is known to be
645 * up-to-date, currently only at load or write_super after coalescing
646 */
647static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
648{
649 __u32 end = mpb->mpb_size / sizeof(end);
650 __u32 *p = (__u32 *) mpb;
651 __u32 sum = 0;
652
653 while (end--) {
654 sum += __le32_to_cpu(*p);
655 p++;
656 }
657
658 return sum - __le32_to_cpu(mpb->check_sum);
659}
660
661static size_t sizeof_imsm_map(struct imsm_map *map)
662{
663 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
664}
665
666struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
667{
668 /* A device can have 2 maps if it is in the middle of a migration.
669 * If second_map is:
670 * MAP_0 or 0 - we return the first map
671 * MAP_1 or 1 - we return the second map if it exists, else NULL
672 * -1 - we return the second map if it exists, else the first
673 */
674 struct imsm_map *map = &dev->vol.map[0];
675 struct imsm_map *map2 = NULL;
676
677 if (dev->vol.migr_state)
678 map2 = (void *)map + sizeof_imsm_map(map);
679
680 switch (second_map) {
681 case MAP_0:
682 case 0:
683 break;
684 case MAP_1:
685 case 1:
686 map = map2;
687 break;
688 case -1:
689 if (map2)
690 map = map2;
691 break;
692 default:
693 map = NULL;
694 }
695 return map;
696
697}
698
699/* return the size of the device.
700 * migr_state increases the returned size if map[0] were to be duplicated
701 */
702static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
703{
704 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
705 sizeof_imsm_map(get_imsm_map(dev, 0));
706
707 /* migrating means an additional map */
708 if (dev->vol.migr_state)
709 size += sizeof_imsm_map(get_imsm_map(dev, 1));
710 else if (migr_state)
711 size += sizeof_imsm_map(get_imsm_map(dev, 0));
712
713 return size;
714}
715
716#ifndef MDASSEMBLE
717/* retrieve disk serial number list from a metadata update */
718static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
719{
720 void *u = update;
721 struct disk_info *inf;
722
723 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
724 sizeof_imsm_dev(&update->dev, 0);
725
726 return inf;
727}
728#endif
729
730static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
731{
732 int offset;
733 int i;
734 void *_mpb = mpb;
735
736 if (index >= mpb->num_raid_devs)
737 return NULL;
738
739 /* devices start after all disks */
740 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
741
742 for (i = 0; i <= index; i++)
743 if (i == index)
744 return _mpb + offset;
745 else
746 offset += sizeof_imsm_dev(_mpb + offset, 0);
747
748 return NULL;
749}
750
751static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
752{
753 struct intel_dev *dv;
754
755 if (index >= super->anchor->num_raid_devs)
756 return NULL;
757 for (dv = super->devlist; dv; dv = dv->next)
758 if (dv->index == index)
759 return dv->dev;
760 return NULL;
761}
762
763/*
764 * for second_map:
765 * == 0 get first map
766 * == 1 get second map
767 * == -1 than get map according to the current migr_state
768 */
769static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev,
770 int slot,
771 int second_map)
772{
773 struct imsm_map *map;
774
775 map = get_imsm_map(dev, second_map);
776
777 /* top byte identifies disk under rebuild */
778 return __le32_to_cpu(map->disk_ord_tbl[slot]);
779}
780
781#define ord_to_idx(ord) (((ord) << 8) >> 8)
782static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot, int second_map)
783{
784 __u32 ord = get_imsm_ord_tbl_ent(dev, slot, second_map);
785
786 return ord_to_idx(ord);
787}
788
789static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
790{
791 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
792}
793
794static int get_imsm_disk_slot(struct imsm_map *map, unsigned idx)
795{
796 int slot;
797 __u32 ord;
798
799 for (slot = 0; slot < map->num_members; slot++) {
800 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
801 if (ord_to_idx(ord) == idx)
802 return slot;
803 }
804
805 return -1;
806}
807
808static int get_imsm_raid_level(struct imsm_map *map)
809{
810 if (map->raid_level == 1) {
811 if (map->num_members == 2)
812 return 1;
813 else
814 return 10;
815 }
816
817 return map->raid_level;
818}
819
820static int cmp_extent(const void *av, const void *bv)
821{
822 const struct extent *a = av;
823 const struct extent *b = bv;
824 if (a->start < b->start)
825 return -1;
826 if (a->start > b->start)
827 return 1;
828 return 0;
829}
830
831static int count_memberships(struct dl *dl, struct intel_super *super)
832{
833 int memberships = 0;
834 int i;
835
836 for (i = 0; i < super->anchor->num_raid_devs; i++) {
837 struct imsm_dev *dev = get_imsm_dev(super, i);
838 struct imsm_map *map = get_imsm_map(dev, 0);
839
840 if (get_imsm_disk_slot(map, dl->index) >= 0)
841 memberships++;
842 }
843
844 return memberships;
845}
846
847static __u32 imsm_min_reserved_sectors(struct intel_super *super);
848
849static struct extent *get_extents(struct intel_super *super, struct dl *dl)
850{
851 /* find a list of used extents on the given physical device */
852 struct extent *rv, *e;
853 int i;
854 int memberships = count_memberships(dl, super);
855 __u32 reservation;
856
857 /* trim the reserved area for spares, so they can join any array
858 * regardless of whether the OROM has assigned sectors from the
859 * IMSM_RESERVED_SECTORS region
860 */
861 if (dl->index == -1)
862 reservation = imsm_min_reserved_sectors(super);
863 else
864 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
865
866 rv = malloc(sizeof(struct extent) * (memberships + 1));
867 if (!rv)
868 return NULL;
869 e = rv;
870
871 for (i = 0; i < super->anchor->num_raid_devs; i++) {
872 struct imsm_dev *dev = get_imsm_dev(super, i);
873 struct imsm_map *map = get_imsm_map(dev, 0);
874
875 if (get_imsm_disk_slot(map, dl->index) >= 0) {
876 e->start = __le32_to_cpu(map->pba_of_lba0);
877 e->size = __le32_to_cpu(map->blocks_per_member);
878 e++;
879 }
880 }
881 qsort(rv, memberships, sizeof(*rv), cmp_extent);
882
883 /* determine the start of the metadata
884 * when no raid devices are defined use the default
885 * ...otherwise allow the metadata to truncate the value
886 * as is the case with older versions of imsm
887 */
888 if (memberships) {
889 struct extent *last = &rv[memberships - 1];
890 __u32 remainder;
891
892 remainder = __le32_to_cpu(dl->disk.total_blocks) -
893 (last->start + last->size);
894 /* round down to 1k block to satisfy precision of the kernel
895 * 'size' interface
896 */
897 remainder &= ~1UL;
898 /* make sure remainder is still sane */
899 if (remainder < (unsigned)ROUND_UP(super->len, 512) >> 9)
900 remainder = ROUND_UP(super->len, 512) >> 9;
901 if (reservation > remainder)
902 reservation = remainder;
903 }
904 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
905 e->size = 0;
906 return rv;
907}
908
909/* try to determine how much space is reserved for metadata from
910 * the last get_extents() entry, otherwise fallback to the
911 * default
912 */
913static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
914{
915 struct extent *e;
916 int i;
917 __u32 rv;
918
919 /* for spares just return a minimal reservation which will grow
920 * once the spare is picked up by an array
921 */
922 if (dl->index == -1)
923 return MPB_SECTOR_CNT;
924
925 e = get_extents(super, dl);
926 if (!e)
927 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
928
929 /* scroll to last entry */
930 for (i = 0; e[i].size; i++)
931 continue;
932
933 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
934
935 free(e);
936
937 return rv;
938}
939
940static int is_spare(struct imsm_disk *disk)
941{
942 return (disk->status & SPARE_DISK) == SPARE_DISK;
943}
944
945static int is_configured(struct imsm_disk *disk)
946{
947 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
948}
949
950static int is_failed(struct imsm_disk *disk)
951{
952 return (disk->status & FAILED_DISK) == FAILED_DISK;
953}
954
955/* try to determine how much space is reserved for metadata from
956 * the last get_extents() entry on the smallest active disk,
957 * otherwise fallback to the default
958 */
959static __u32 imsm_min_reserved_sectors(struct intel_super *super)
960{
961 struct extent *e;
962 int i;
963 __u32 min_active, remainder;
964 __u32 rv = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
965 struct dl *dl, *dl_min = NULL;
966
967 if (!super)
968 return rv;
969
970 min_active = 0;
971 for (dl = super->disks; dl; dl = dl->next) {
972 if (dl->index < 0)
973 continue;
974 if (dl->disk.total_blocks < min_active || min_active == 0) {
975 dl_min = dl;
976 min_active = dl->disk.total_blocks;
977 }
978 }
979 if (!dl_min)
980 return rv;
981
982 /* find last lba used by subarrays on the smallest active disk */
983 e = get_extents(super, dl_min);
984 if (!e)
985 return rv;
986 for (i = 0; e[i].size; i++)
987 continue;
988
989 remainder = min_active - e[i].start;
990 free(e);
991
992 /* to give priority to recovery we should not require full
993 IMSM_RESERVED_SECTORS from the spare */
994 rv = MPB_SECTOR_CNT + NUM_BLOCKS_DIRTY_STRIPE_REGION;
995
996 /* if real reservation is smaller use that value */
997 return (remainder < rv) ? remainder : rv;
998}
999
1000/* Return minimum size of a spare that can be used in this array*/
1001static unsigned long long min_acceptable_spare_size_imsm(struct supertype *st)
1002{
1003 struct intel_super *super = st->sb;
1004 struct dl *dl;
1005 struct extent *e;
1006 int i;
1007 unsigned long long rv = 0;
1008
1009 if (!super)
1010 return rv;
1011 /* find first active disk in array */
1012 dl = super->disks;
1013 while (dl && (is_failed(&dl->disk) || dl->index == -1))
1014 dl = dl->next;
1015 if (!dl)
1016 return rv;
1017 /* find last lba used by subarrays */
1018 e = get_extents(super, dl);
1019 if (!e)
1020 return rv;
1021 for (i = 0; e[i].size; i++)
1022 continue;
1023 if (i > 0)
1024 rv = e[i-1].start + e[i-1].size;
1025 free(e);
1026
1027 /* add the amount of space needed for metadata */
1028 rv = rv + imsm_min_reserved_sectors(super);
1029
1030 return rv * 512;
1031}
1032
1033static int is_gen_migration(struct imsm_dev *dev);
1034
1035#ifndef MDASSEMBLE
1036static __u64 blocks_per_migr_unit(struct intel_super *super,
1037 struct imsm_dev *dev);
1038
1039static void print_imsm_dev(struct intel_super *super,
1040 struct imsm_dev *dev,
1041 char *uuid,
1042 int disk_idx)
1043{
1044 __u64 sz;
1045 int slot, i;
1046 struct imsm_map *map = get_imsm_map(dev, 0);
1047 struct imsm_map *map2 = get_imsm_map(dev, 1);
1048 __u32 ord;
1049
1050 printf("\n");
1051 printf("[%.16s]:\n", dev->volume);
1052 printf(" UUID : %s\n", uuid);
1053 printf(" RAID Level : %d", get_imsm_raid_level(map));
1054 if (map2)
1055 printf(" <-- %d", get_imsm_raid_level(map2));
1056 printf("\n");
1057 printf(" Members : %d", map->num_members);
1058 if (map2)
1059 printf(" <-- %d", map2->num_members);
1060 printf("\n");
1061 printf(" Slots : [");
1062 for (i = 0; i < map->num_members; i++) {
1063 ord = get_imsm_ord_tbl_ent(dev, i, 0);
1064 printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
1065 }
1066 printf("]");
1067 if (map2) {
1068 printf(" <-- [");
1069 for (i = 0; i < map2->num_members; i++) {
1070 ord = get_imsm_ord_tbl_ent(dev, i, 1);
1071 printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
1072 }
1073 printf("]");
1074 }
1075 printf("\n");
1076 printf(" Failed disk : ");
1077 if (map->failed_disk_num == 0xff)
1078 printf("none");
1079 else
1080 printf("%i", map->failed_disk_num);
1081 printf("\n");
1082 slot = get_imsm_disk_slot(map, disk_idx);
1083 if (slot >= 0) {
1084 ord = get_imsm_ord_tbl_ent(dev, slot, -1);
1085 printf(" This Slot : %d%s\n", slot,
1086 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
1087 } else
1088 printf(" This Slot : ?\n");
1089 sz = __le32_to_cpu(dev->size_high);
1090 sz <<= 32;
1091 sz += __le32_to_cpu(dev->size_low);
1092 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
1093 human_size(sz * 512));
1094 sz = __le32_to_cpu(map->blocks_per_member);
1095 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
1096 human_size(sz * 512));
1097 printf(" Sector Offset : %u\n",
1098 __le32_to_cpu(map->pba_of_lba0));
1099 printf(" Num Stripes : %u\n",
1100 __le32_to_cpu(map->num_data_stripes));
1101 printf(" Chunk Size : %u KiB",
1102 __le16_to_cpu(map->blocks_per_strip) / 2);
1103 if (map2)
1104 printf(" <-- %u KiB",
1105 __le16_to_cpu(map2->blocks_per_strip) / 2);
1106 printf("\n");
1107 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
1108 printf(" Migrate State : ");
1109 if (dev->vol.migr_state) {
1110 if (migr_type(dev) == MIGR_INIT)
1111 printf("initialize\n");
1112 else if (migr_type(dev) == MIGR_REBUILD)
1113 printf("rebuild\n");
1114 else if (migr_type(dev) == MIGR_VERIFY)
1115 printf("check\n");
1116 else if (migr_type(dev) == MIGR_GEN_MIGR)
1117 printf("general migration\n");
1118 else if (migr_type(dev) == MIGR_STATE_CHANGE)
1119 printf("state change\n");
1120 else if (migr_type(dev) == MIGR_REPAIR)
1121 printf("repair\n");
1122 else
1123 printf("<unknown:%d>\n", migr_type(dev));
1124 } else
1125 printf("idle\n");
1126 printf(" Map State : %s", map_state_str[map->map_state]);
1127 if (dev->vol.migr_state) {
1128 struct imsm_map *map = get_imsm_map(dev, 1);
1129
1130 printf(" <-- %s", map_state_str[map->map_state]);
1131 printf("\n Checkpoint : %u ",
1132 __le32_to_cpu(dev->vol.curr_migr_unit));
1133 if ((is_gen_migration(dev)) && (super->disks->index > 1))
1134 printf("(N/A)");
1135 else
1136 printf("(%llu)", (unsigned long long)
1137 blocks_per_migr_unit(super, dev));
1138 }
1139 printf("\n");
1140 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
1141}
1142
1143static void print_imsm_disk(struct imsm_disk *disk, int index, __u32 reserved)
1144{
1145 char str[MAX_RAID_SERIAL_LEN + 1];
1146 __u64 sz;
1147
1148 if (index < -1 || !disk)
1149 return;
1150
1151 printf("\n");
1152 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
1153 if (index >= 0)
1154 printf(" Disk%02d Serial : %s\n", index, str);
1155 else
1156 printf(" Disk Serial : %s\n", str);
1157 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
1158 is_configured(disk) ? " active" : "",
1159 is_failed(disk) ? " failed" : "");
1160 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
1161 sz = __le32_to_cpu(disk->total_blocks) - reserved;
1162 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
1163 human_size(sz * 512));
1164}
1165
1166void examine_migr_rec_imsm(struct intel_super *super)
1167{
1168 struct migr_record *migr_rec = super->migr_rec;
1169 struct imsm_super *mpb = super->anchor;
1170 int i;
1171
1172 for (i = 0; i < mpb->num_raid_devs; i++) {
1173 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
1174 if (is_gen_migration(dev) == 0)
1175 continue;
1176
1177 printf("\nMigration Record Information:");
1178 if (super->disks->index > 1) {
1179 printf(" Empty\n ");
1180 printf("Examine one of first two disks in array\n");
1181 break;
1182 }
1183 printf("\n Status : ");
1184 if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL)
1185 printf("Normal\n");
1186 else
1187 printf("Contains Data\n");
1188 printf(" Current Unit : %u\n",
1189 __le32_to_cpu(migr_rec->curr_migr_unit));
1190 printf(" Family : %u\n",
1191 __le32_to_cpu(migr_rec->family_num));
1192 printf(" Ascending : %u\n",
1193 __le32_to_cpu(migr_rec->ascending_migr));
1194 printf(" Blocks Per Unit : %u\n",
1195 __le32_to_cpu(migr_rec->blocks_per_unit));
1196 printf(" Dest. Depth Per Unit : %u\n",
1197 __le32_to_cpu(migr_rec->dest_depth_per_unit));
1198 printf(" Checkpoint Area pba : %u\n",
1199 __le32_to_cpu(migr_rec->ckpt_area_pba));
1200 printf(" First member lba : %u\n",
1201 __le32_to_cpu(migr_rec->dest_1st_member_lba));
1202 printf(" Total Number of Units : %u\n",
1203 __le32_to_cpu(migr_rec->num_migr_units));
1204 printf(" Size of volume : %u\n",
1205 __le32_to_cpu(migr_rec->post_migr_vol_cap));
1206 printf(" Expansion space for LBA64 : %u\n",
1207 __le32_to_cpu(migr_rec->post_migr_vol_cap_hi));
1208 printf(" Record was read from : %u\n",
1209 __le32_to_cpu(migr_rec->ckpt_read_disk_num));
1210
1211 break;
1212 }
1213}
1214#endif /* MDASSEMBLE */
1215/*******************************************************************************
1216 * function: imsm_check_attributes
1217 * Description: Function checks if features represented by attributes flags
1218 * are supported by mdadm.
1219 * Parameters:
1220 * attributes - Attributes read from metadata
1221 * Returns:
1222 * 0 - passed attributes contains unsupported features flags
1223 * 1 - all features are supported
1224 ******************************************************************************/
1225static int imsm_check_attributes(__u32 attributes)
1226{
1227 int ret_val = 1;
1228 __u32 not_supported = MPB_ATTRIB_SUPPORTED^0xffffffff;
1229
1230 not_supported &= ~MPB_ATTRIB_IGNORED;
1231
1232 not_supported &= attributes;
1233 if (not_supported) {
1234 fprintf(stderr, Name "(IMSM): Unsupported attributes : %x\n",
1235 (unsigned)__le32_to_cpu(not_supported));
1236 if (not_supported & MPB_ATTRIB_CHECKSUM_VERIFY) {
1237 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1238 not_supported ^= MPB_ATTRIB_CHECKSUM_VERIFY;
1239 }
1240 if (not_supported & MPB_ATTRIB_2TB) {
1241 dprintf("\t\tMPB_ATTRIB_2TB\n");
1242 not_supported ^= MPB_ATTRIB_2TB;
1243 }
1244 if (not_supported & MPB_ATTRIB_RAID0) {
1245 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1246 not_supported ^= MPB_ATTRIB_RAID0;
1247 }
1248 if (not_supported & MPB_ATTRIB_RAID1) {
1249 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1250 not_supported ^= MPB_ATTRIB_RAID1;
1251 }
1252 if (not_supported & MPB_ATTRIB_RAID10) {
1253 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1254 not_supported ^= MPB_ATTRIB_RAID10;
1255 }
1256 if (not_supported & MPB_ATTRIB_RAID1E) {
1257 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1258 not_supported ^= MPB_ATTRIB_RAID1E;
1259 }
1260 if (not_supported & MPB_ATTRIB_RAID5) {
1261 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1262 not_supported ^= MPB_ATTRIB_RAID5;
1263 }
1264 if (not_supported & MPB_ATTRIB_RAIDCNG) {
1265 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1266 not_supported ^= MPB_ATTRIB_RAIDCNG;
1267 }
1268 if (not_supported & MPB_ATTRIB_BBM) {
1269 dprintf("\t\tMPB_ATTRIB_BBM\n");
1270 not_supported ^= MPB_ATTRIB_BBM;
1271 }
1272 if (not_supported & MPB_ATTRIB_CHECKSUM_VERIFY) {
1273 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1274 not_supported ^= MPB_ATTRIB_CHECKSUM_VERIFY;
1275 }
1276 if (not_supported & MPB_ATTRIB_EXP_STRIPE_SIZE) {
1277 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1278 not_supported ^= MPB_ATTRIB_EXP_STRIPE_SIZE;
1279 }
1280 if (not_supported & MPB_ATTRIB_2TB_DISK) {
1281 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1282 not_supported ^= MPB_ATTRIB_2TB_DISK;
1283 }
1284 if (not_supported & MPB_ATTRIB_NEVER_USE2) {
1285 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1286 not_supported ^= MPB_ATTRIB_NEVER_USE2;
1287 }
1288 if (not_supported & MPB_ATTRIB_NEVER_USE) {
1289 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1290 not_supported ^= MPB_ATTRIB_NEVER_USE;
1291 }
1292
1293 if (not_supported)
1294 dprintf(Name "(IMSM): Unknown attributes : %x\n", not_supported);
1295
1296 ret_val = 0;
1297 }
1298
1299 return ret_val;
1300}
1301
1302#ifndef MDASSEMBLE
1303static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map);
1304
1305static void examine_super_imsm(struct supertype *st, char *homehost)
1306{
1307 struct intel_super *super = st->sb;
1308 struct imsm_super *mpb = super->anchor;
1309 char str[MAX_SIGNATURE_LENGTH];
1310 int i;
1311 struct mdinfo info;
1312 char nbuf[64];
1313 __u32 sum;
1314 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1315 struct dl *dl;
1316
1317 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
1318 printf(" Magic : %s\n", str);
1319 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
1320 printf(" Version : %s\n", get_imsm_version(mpb));
1321 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
1322 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
1323 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
1324 printf(" Attributes : ");
1325 if (imsm_check_attributes(mpb->attributes))
1326 printf("All supported\n");
1327 else
1328 printf("not supported\n");
1329 getinfo_super_imsm(st, &info, NULL);
1330 fname_from_uuid(st, &info, nbuf, ':');
1331 printf(" UUID : %s\n", nbuf + 5);
1332 sum = __le32_to_cpu(mpb->check_sum);
1333 printf(" Checksum : %08x %s\n", sum,
1334 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
1335 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
1336 printf(" Disks : %d\n", mpb->num_disks);
1337 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
1338 print_imsm_disk(__get_imsm_disk(mpb, super->disks->index), super->disks->index, reserved);
1339 if (super->bbm_log) {
1340 struct bbm_log *log = super->bbm_log;
1341
1342 printf("\n");
1343 printf("Bad Block Management Log:\n");
1344 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
1345 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
1346 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
1347 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
1348 printf(" First Spare : %llx\n",
1349 (unsigned long long) __le64_to_cpu(log->first_spare_lba));
1350 }
1351 for (i = 0; i < mpb->num_raid_devs; i++) {
1352 struct mdinfo info;
1353 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
1354
1355 super->current_vol = i;
1356 getinfo_super_imsm(st, &info, NULL);
1357 fname_from_uuid(st, &info, nbuf, ':');
1358 print_imsm_dev(super, dev, nbuf + 5, super->disks->index);
1359 }
1360 for (i = 0; i < mpb->num_disks; i++) {
1361 if (i == super->disks->index)
1362 continue;
1363 print_imsm_disk(__get_imsm_disk(mpb, i), i, reserved);
1364 }
1365
1366 for (dl = super->disks; dl; dl = dl->next)
1367 if (dl->index == -1)
1368 print_imsm_disk(&dl->disk, -1, reserved);
1369
1370 examine_migr_rec_imsm(super);
1371}
1372
1373static void brief_examine_super_imsm(struct supertype *st, int verbose)
1374{
1375 /* We just write a generic IMSM ARRAY entry */
1376 struct mdinfo info;
1377 char nbuf[64];
1378 struct intel_super *super = st->sb;
1379
1380 if (!super->anchor->num_raid_devs) {
1381 printf("ARRAY metadata=imsm\n");
1382 return;
1383 }
1384
1385 getinfo_super_imsm(st, &info, NULL);
1386 fname_from_uuid(st, &info, nbuf, ':');
1387 printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
1388}
1389
1390static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
1391{
1392 /* We just write a generic IMSM ARRAY entry */
1393 struct mdinfo info;
1394 char nbuf[64];
1395 char nbuf1[64];
1396 struct intel_super *super = st->sb;
1397 int i;
1398
1399 if (!super->anchor->num_raid_devs)
1400 return;
1401
1402 getinfo_super_imsm(st, &info, NULL);
1403 fname_from_uuid(st, &info, nbuf, ':');
1404 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1405 struct imsm_dev *dev = get_imsm_dev(super, i);
1406
1407 super->current_vol = i;
1408 getinfo_super_imsm(st, &info, NULL);
1409 fname_from_uuid(st, &info, nbuf1, ':');
1410 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1411 dev->volume, nbuf + 5, i, nbuf1 + 5);
1412 }
1413}
1414
1415static void export_examine_super_imsm(struct supertype *st)
1416{
1417 struct intel_super *super = st->sb;
1418 struct imsm_super *mpb = super->anchor;
1419 struct mdinfo info;
1420 char nbuf[64];
1421
1422 getinfo_super_imsm(st, &info, NULL);
1423 fname_from_uuid(st, &info, nbuf, ':');
1424 printf("MD_METADATA=imsm\n");
1425 printf("MD_LEVEL=container\n");
1426 printf("MD_UUID=%s\n", nbuf+5);
1427 printf("MD_DEVICES=%u\n", mpb->num_disks);
1428}
1429
1430static void detail_super_imsm(struct supertype *st, char *homehost)
1431{
1432 struct mdinfo info;
1433 char nbuf[64];
1434
1435 getinfo_super_imsm(st, &info, NULL);
1436 fname_from_uuid(st, &info, nbuf, ':');
1437 printf("\n UUID : %s\n", nbuf + 5);
1438}
1439
1440static void brief_detail_super_imsm(struct supertype *st)
1441{
1442 struct mdinfo info;
1443 char nbuf[64];
1444 getinfo_super_imsm(st, &info, NULL);
1445 fname_from_uuid(st, &info, nbuf, ':');
1446 printf(" UUID=%s", nbuf + 5);
1447}
1448
1449static int imsm_read_serial(int fd, char *devname, __u8 *serial);
1450static void fd2devname(int fd, char *name);
1451
1452static int ahci_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
1453{
1454 /* dump an unsorted list of devices attached to AHCI Intel storage
1455 * controller, as well as non-connected ports
1456 */
1457 int hba_len = strlen(hba_path) + 1;
1458 struct dirent *ent;
1459 DIR *dir;
1460 char *path = NULL;
1461 int err = 0;
1462 unsigned long port_mask = (1 << port_count) - 1;
1463
1464 if (port_count > (int)sizeof(port_mask) * 8) {
1465 if (verbose)
1466 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
1467 return 2;
1468 }
1469
1470 /* scroll through /sys/dev/block looking for devices attached to
1471 * this hba
1472 */
1473 dir = opendir("/sys/dev/block");
1474 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1475 int fd;
1476 char model[64];
1477 char vendor[64];
1478 char buf[1024];
1479 int major, minor;
1480 char *device;
1481 char *c;
1482 int port;
1483 int type;
1484
1485 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
1486 continue;
1487 path = devt_to_devpath(makedev(major, minor));
1488 if (!path)
1489 continue;
1490 if (!path_attached_to_hba(path, hba_path)) {
1491 free(path);
1492 path = NULL;
1493 continue;
1494 }
1495
1496 /* retrieve the scsi device type */
1497 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
1498 if (verbose)
1499 fprintf(stderr, Name ": failed to allocate 'device'\n");
1500 err = 2;
1501 break;
1502 }
1503 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
1504 if (load_sys(device, buf) != 0) {
1505 if (verbose)
1506 fprintf(stderr, Name ": failed to read device type for %s\n",
1507 path);
1508 err = 2;
1509 free(device);
1510 break;
1511 }
1512 type = strtoul(buf, NULL, 10);
1513
1514 /* if it's not a disk print the vendor and model */
1515 if (!(type == 0 || type == 7 || type == 14)) {
1516 vendor[0] = '\0';
1517 model[0] = '\0';
1518 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
1519 if (load_sys(device, buf) == 0) {
1520 strncpy(vendor, buf, sizeof(vendor));
1521 vendor[sizeof(vendor) - 1] = '\0';
1522 c = (char *) &vendor[sizeof(vendor) - 1];
1523 while (isspace(*c) || *c == '\0')
1524 *c-- = '\0';
1525
1526 }
1527 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
1528 if (load_sys(device, buf) == 0) {
1529 strncpy(model, buf, sizeof(model));
1530 model[sizeof(model) - 1] = '\0';
1531 c = (char *) &model[sizeof(model) - 1];
1532 while (isspace(*c) || *c == '\0')
1533 *c-- = '\0';
1534 }
1535
1536 if (vendor[0] && model[0])
1537 sprintf(buf, "%.64s %.64s", vendor, model);
1538 else
1539 switch (type) { /* numbers from hald/linux/device.c */
1540 case 1: sprintf(buf, "tape"); break;
1541 case 2: sprintf(buf, "printer"); break;
1542 case 3: sprintf(buf, "processor"); break;
1543 case 4:
1544 case 5: sprintf(buf, "cdrom"); break;
1545 case 6: sprintf(buf, "scanner"); break;
1546 case 8: sprintf(buf, "media_changer"); break;
1547 case 9: sprintf(buf, "comm"); break;
1548 case 12: sprintf(buf, "raid"); break;
1549 default: sprintf(buf, "unknown");
1550 }
1551 } else
1552 buf[0] = '\0';
1553 free(device);
1554
1555 /* chop device path to 'host%d' and calculate the port number */
1556 c = strchr(&path[hba_len], '/');
1557 if (!c) {
1558 if (verbose)
1559 fprintf(stderr, Name ": %s - invalid path name\n", path + hba_len);
1560 err = 2;
1561 break;
1562 }
1563 *c = '\0';
1564 if (sscanf(&path[hba_len], "host%d", &port) == 1)
1565 port -= host_base;
1566 else {
1567 if (verbose) {
1568 *c = '/'; /* repair the full string */
1569 fprintf(stderr, Name ": failed to determine port number for %s\n",
1570 path);
1571 }
1572 err = 2;
1573 break;
1574 }
1575
1576 /* mark this port as used */
1577 port_mask &= ~(1 << port);
1578
1579 /* print out the device information */
1580 if (buf[0]) {
1581 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
1582 continue;
1583 }
1584
1585 fd = dev_open(ent->d_name, O_RDONLY);
1586 if (fd < 0)
1587 printf(" Port%d : - disk info unavailable -\n", port);
1588 else {
1589 fd2devname(fd, buf);
1590 printf(" Port%d : %s", port, buf);
1591 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
1592 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN, buf);
1593 else
1594 printf(" ()\n");
1595 close(fd);
1596 }
1597 free(path);
1598 path = NULL;
1599 }
1600 if (path)
1601 free(path);
1602 if (dir)
1603 closedir(dir);
1604 if (err == 0) {
1605 int i;
1606
1607 for (i = 0; i < port_count; i++)
1608 if (port_mask & (1 << i))
1609 printf(" Port%d : - no device attached -\n", i);
1610 }
1611
1612 return err;
1613}
1614
1615static void print_found_intel_controllers(struct sys_dev *elem)
1616{
1617 for (; elem; elem = elem->next) {
1618 fprintf(stderr, Name ": found Intel(R) ");
1619 if (elem->type == SYS_DEV_SATA)
1620 fprintf(stderr, "SATA ");
1621 else if (elem->type == SYS_DEV_SAS)
1622 fprintf(stderr, "SAS ");
1623 fprintf(stderr, "RAID controller");
1624 if (elem->pci_id)
1625 fprintf(stderr, " at %s", elem->pci_id);
1626 fprintf(stderr, ".\n");
1627 }
1628 fflush(stderr);
1629}
1630
1631static int ahci_get_port_count(const char *hba_path, int *port_count)
1632{
1633 struct dirent *ent;
1634 DIR *dir;
1635 int host_base = -1;
1636
1637 *port_count = 0;
1638 if ((dir = opendir(hba_path)) == NULL)
1639 return -1;
1640
1641 for (ent = readdir(dir); ent; ent = readdir(dir)) {
1642 int host;
1643
1644 if (sscanf(ent->d_name, "host%d", &host) != 1)
1645 continue;
1646 if (*port_count == 0)
1647 host_base = host;
1648 else if (host < host_base)
1649 host_base = host;
1650
1651 if (host + 1 > *port_count + host_base)
1652 *port_count = host + 1 - host_base;
1653 }
1654 closedir(dir);
1655 return host_base;
1656}
1657
1658static void print_imsm_capability(const struct imsm_orom *orom)
1659{
1660 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1661 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
1662 orom->hotfix_ver, orom->build);
1663 printf(" RAID Levels :%s%s%s%s%s\n",
1664 imsm_orom_has_raid0(orom) ? " raid0" : "",
1665 imsm_orom_has_raid1(orom) ? " raid1" : "",
1666 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
1667 imsm_orom_has_raid10(orom) ? " raid10" : "",
1668 imsm_orom_has_raid5(orom) ? " raid5" : "");
1669 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1670 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
1671 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
1672 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
1673 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
1674 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
1675 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
1676 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
1677 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
1678 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
1679 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
1680 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
1681 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
1682 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
1683 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
1684 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
1685 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1686 printf(" Max Disks : %d\n", orom->tds);
1687 printf(" Max Volumes : %d\n", orom->vpa);
1688 return;
1689}
1690
1691static int detail_platform_imsm(int verbose, int enumerate_only)
1692{
1693 /* There are two components to imsm platform support, the ahci SATA
1694 * controller and the option-rom. To find the SATA controller we
1695 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1696 * controller with the Intel vendor id is present. This approach
1697 * allows mdadm to leverage the kernel's ahci detection logic, with the
1698 * caveat that if ahci.ko is not loaded mdadm will not be able to
1699 * detect platform raid capabilities. The option-rom resides in a
1700 * platform "Adapter ROM". We scan for its signature to retrieve the
1701 * platform capabilities. If raid support is disabled in the BIOS the
1702 * option-rom capability structure will not be available.
1703 */
1704 const struct imsm_orom *orom;
1705 struct sys_dev *list, *hba;
1706 int host_base = 0;
1707 int port_count = 0;
1708 int result=0;
1709
1710 if (enumerate_only) {
1711 if (check_env("IMSM_NO_PLATFORM"))
1712 return 0;
1713 list = find_intel_devices();
1714 if (!list)
1715 return 2;
1716 for (hba = list; hba; hba = hba->next) {
1717 orom = find_imsm_capability(hba->type);
1718 if (!orom) {
1719 result = 2;
1720 break;
1721 }
1722 }
1723 free_sys_dev(&list);
1724 return result;
1725 }
1726
1727 list = find_intel_devices();
1728 if (!list) {
1729 if (verbose)
1730 fprintf(stderr, Name ": no active Intel(R) RAID "
1731 "controller found.\n");
1732 free_sys_dev(&list);
1733 return 2;
1734 } else if (verbose)
1735 print_found_intel_controllers(list);
1736
1737 for (hba = list; hba; hba = hba->next) {
1738 orom = find_imsm_capability(hba->type);
1739 if (!orom)
1740 fprintf(stderr, Name ": imsm capabilities not found for controller: %s (type %s)\n",
1741 hba->path, get_sys_dev_type(hba->type));
1742 else
1743 print_imsm_capability(orom);
1744 }
1745
1746 for (hba = list; hba; hba = hba->next) {
1747 printf(" I/O Controller : %s (%s)\n",
1748 hba->path, get_sys_dev_type(hba->type));
1749
1750 if (hba->type == SYS_DEV_SATA) {
1751 host_base = ahci_get_port_count(hba->path, &port_count);
1752 if (ahci_enumerate_ports(hba->path, port_count, host_base, verbose)) {
1753 if (verbose)
1754 fprintf(stderr, Name ": failed to enumerate "
1755 "ports on SATA controller at %s.", hba->pci_id);
1756 result |= 2;
1757 }
1758 }
1759 }
1760
1761 free_sys_dev(&list);
1762 return result;
1763}
1764#endif
1765
1766static int match_home_imsm(struct supertype *st, char *homehost)
1767{
1768 /* the imsm metadata format does not specify any host
1769 * identification information. We return -1 since we can never
1770 * confirm nor deny whether a given array is "meant" for this
1771 * host. We rely on compare_super and the 'family_num' fields to
1772 * exclude member disks that do not belong, and we rely on
1773 * mdadm.conf to specify the arrays that should be assembled.
1774 * Auto-assembly may still pick up "foreign" arrays.
1775 */
1776
1777 return -1;
1778}
1779
1780static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1781{
1782 /* The uuid returned here is used for:
1783 * uuid to put into bitmap file (Create, Grow)
1784 * uuid for backup header when saving critical section (Grow)
1785 * comparing uuids when re-adding a device into an array
1786 * In these cases the uuid required is that of the data-array,
1787 * not the device-set.
1788 * uuid to recognise same set when adding a missing device back
1789 * to an array. This is a uuid for the device-set.
1790 *
1791 * For each of these we can make do with a truncated
1792 * or hashed uuid rather than the original, as long as
1793 * everyone agrees.
1794 * In each case the uuid required is that of the data-array,
1795 * not the device-set.
1796 */
1797 /* imsm does not track uuid's so we synthesis one using sha1 on
1798 * - The signature (Which is constant for all imsm array, but no matter)
1799 * - the orig_family_num of the container
1800 * - the index number of the volume
1801 * - the 'serial' number of the volume.
1802 * Hopefully these are all constant.
1803 */
1804 struct intel_super *super = st->sb;
1805
1806 char buf[20];
1807 struct sha1_ctx ctx;
1808 struct imsm_dev *dev = NULL;
1809 __u32 family_num;
1810
1811 /* some mdadm versions failed to set ->orig_family_num, in which
1812 * case fall back to ->family_num. orig_family_num will be
1813 * fixed up with the first metadata update.
1814 */
1815 family_num = super->anchor->orig_family_num;
1816 if (family_num == 0)
1817 family_num = super->anchor->family_num;
1818 sha1_init_ctx(&ctx);
1819 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1820 sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
1821 if (super->current_vol >= 0)
1822 dev = get_imsm_dev(super, super->current_vol);
1823 if (dev) {
1824 __u32 vol = super->current_vol;
1825 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1826 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1827 }
1828 sha1_finish_ctx(&ctx, buf);
1829 memcpy(uuid, buf, 4*4);
1830}
1831
1832#if 0
1833static void
1834get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1835{
1836 __u8 *v = get_imsm_version(mpb);
1837 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1838 char major[] = { 0, 0, 0 };
1839 char minor[] = { 0 ,0, 0 };
1840 char patch[] = { 0, 0, 0 };
1841 char *ver_parse[] = { major, minor, patch };
1842 int i, j;
1843
1844 i = j = 0;
1845 while (*v != '\0' && v < end) {
1846 if (*v != '.' && j < 2)
1847 ver_parse[i][j++] = *v;
1848 else {
1849 i++;
1850 j = 0;
1851 }
1852 v++;
1853 }
1854
1855 *m = strtol(minor, NULL, 0);
1856 *p = strtol(patch, NULL, 0);
1857}
1858#endif
1859
1860static __u32 migr_strip_blocks_resync(struct imsm_dev *dev)
1861{
1862 /* migr_strip_size when repairing or initializing parity */
1863 struct imsm_map *map = get_imsm_map(dev, 0);
1864 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1865
1866 switch (get_imsm_raid_level(map)) {
1867 case 5:
1868 case 10:
1869 return chunk;
1870 default:
1871 return 128*1024 >> 9;
1872 }
1873}
1874
1875static __u32 migr_strip_blocks_rebuild(struct imsm_dev *dev)
1876{
1877 /* migr_strip_size when rebuilding a degraded disk, no idea why
1878 * this is different than migr_strip_size_resync(), but it's good
1879 * to be compatible
1880 */
1881 struct imsm_map *map = get_imsm_map(dev, 1);
1882 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1883
1884 switch (get_imsm_raid_level(map)) {
1885 case 1:
1886 case 10:
1887 if (map->num_members % map->num_domains == 0)
1888 return 128*1024 >> 9;
1889 else
1890 return chunk;
1891 case 5:
1892 return max((__u32) 64*1024 >> 9, chunk);
1893 default:
1894 return 128*1024 >> 9;
1895 }
1896}
1897
1898static __u32 num_stripes_per_unit_resync(struct imsm_dev *dev)
1899{
1900 struct imsm_map *lo = get_imsm_map(dev, 0);
1901 struct imsm_map *hi = get_imsm_map(dev, 1);
1902 __u32 lo_chunk = __le32_to_cpu(lo->blocks_per_strip);
1903 __u32 hi_chunk = __le32_to_cpu(hi->blocks_per_strip);
1904
1905 return max((__u32) 1, hi_chunk / lo_chunk);
1906}
1907
1908static __u32 num_stripes_per_unit_rebuild(struct imsm_dev *dev)
1909{
1910 struct imsm_map *lo = get_imsm_map(dev, 0);
1911 int level = get_imsm_raid_level(lo);
1912
1913 if (level == 1 || level == 10) {
1914 struct imsm_map *hi = get_imsm_map(dev, 1);
1915
1916 return hi->num_domains;
1917 } else
1918 return num_stripes_per_unit_resync(dev);
1919}
1920
1921static __u8 imsm_num_data_members(struct imsm_dev *dev, int second_map)
1922{
1923 /* named 'imsm_' because raid0, raid1 and raid10
1924 * counter-intuitively have the same number of data disks
1925 */
1926 struct imsm_map *map = get_imsm_map(dev, second_map);
1927
1928 switch (get_imsm_raid_level(map)) {
1929 case 0:
1930 case 1:
1931 case 10:
1932 return map->num_members;
1933 case 5:
1934 return map->num_members - 1;
1935 default:
1936 dprintf("%s: unsupported raid level\n", __func__);
1937 return 0;
1938 }
1939}
1940
1941static __u32 parity_segment_depth(struct imsm_dev *dev)
1942{
1943 struct imsm_map *map = get_imsm_map(dev, 0);
1944 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1945
1946 switch(get_imsm_raid_level(map)) {
1947 case 1:
1948 case 10:
1949 return chunk * map->num_domains;
1950 case 5:
1951 return chunk * map->num_members;
1952 default:
1953 return chunk;
1954 }
1955}
1956
1957static __u32 map_migr_block(struct imsm_dev *dev, __u32 block)
1958{
1959 struct imsm_map *map = get_imsm_map(dev, 1);
1960 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1961 __u32 strip = block / chunk;
1962
1963 switch (get_imsm_raid_level(map)) {
1964 case 1:
1965 case 10: {
1966 __u32 vol_strip = (strip * map->num_domains) + 1;
1967 __u32 vol_stripe = vol_strip / map->num_members;
1968
1969 return vol_stripe * chunk + block % chunk;
1970 } case 5: {
1971 __u32 stripe = strip / (map->num_members - 1);
1972
1973 return stripe * chunk + block % chunk;
1974 }
1975 default:
1976 return 0;
1977 }
1978}
1979
1980static __u64 blocks_per_migr_unit(struct intel_super *super,
1981 struct imsm_dev *dev)
1982{
1983 /* calculate the conversion factor between per member 'blocks'
1984 * (md/{resync,rebuild}_start) and imsm migration units, return
1985 * 0 for the 'not migrating' and 'unsupported migration' cases
1986 */
1987 if (!dev->vol.migr_state)
1988 return 0;
1989
1990 switch (migr_type(dev)) {
1991 case MIGR_GEN_MIGR: {
1992 struct migr_record *migr_rec = super->migr_rec;
1993 return __le32_to_cpu(migr_rec->blocks_per_unit);
1994 }
1995 case MIGR_VERIFY:
1996 case MIGR_REPAIR:
1997 case MIGR_INIT: {
1998 struct imsm_map *map = get_imsm_map(dev, 0);
1999 __u32 stripes_per_unit;
2000 __u32 blocks_per_unit;
2001 __u32 parity_depth;
2002 __u32 migr_chunk;
2003 __u32 block_map;
2004 __u32 block_rel;
2005 __u32 segment;
2006 __u32 stripe;
2007 __u8 disks;
2008
2009 /* yes, this is really the translation of migr_units to
2010 * per-member blocks in the 'resync' case
2011 */
2012 stripes_per_unit = num_stripes_per_unit_resync(dev);
2013 migr_chunk = migr_strip_blocks_resync(dev);
2014 disks = imsm_num_data_members(dev, 0);
2015 blocks_per_unit = stripes_per_unit * migr_chunk * disks;
2016 stripe = __le16_to_cpu(map->blocks_per_strip) * disks;
2017 segment = blocks_per_unit / stripe;
2018 block_rel = blocks_per_unit - segment * stripe;
2019 parity_depth = parity_segment_depth(dev);
2020 block_map = map_migr_block(dev, block_rel);
2021 return block_map + parity_depth * segment;
2022 }
2023 case MIGR_REBUILD: {
2024 __u32 stripes_per_unit;
2025 __u32 migr_chunk;
2026
2027 stripes_per_unit = num_stripes_per_unit_rebuild(dev);
2028 migr_chunk = migr_strip_blocks_rebuild(dev);
2029 return migr_chunk * stripes_per_unit;
2030 }
2031 case MIGR_STATE_CHANGE:
2032 default:
2033 return 0;
2034 }
2035}
2036
2037static int imsm_level_to_layout(int level)
2038{
2039 switch (level) {
2040 case 0:
2041 case 1:
2042 return 0;
2043 case 5:
2044 case 6:
2045 return ALGORITHM_LEFT_ASYMMETRIC;
2046 case 10:
2047 return 0x102;
2048 }
2049 return UnSet;
2050}
2051
2052/*******************************************************************************
2053 * Function: read_imsm_migr_rec
2054 * Description: Function reads imsm migration record from last sector of disk
2055 * Parameters:
2056 * fd : disk descriptor
2057 * super : metadata info
2058 * Returns:
2059 * 0 : success,
2060 * -1 : fail
2061 ******************************************************************************/
2062static int read_imsm_migr_rec(int fd, struct intel_super *super)
2063{
2064 int ret_val = -1;
2065 unsigned long long dsize;
2066
2067 get_dev_size(fd, NULL, &dsize);
2068 if (lseek64(fd, dsize - 512, SEEK_SET) < 0) {
2069 fprintf(stderr,
2070 Name ": Cannot seek to anchor block: %s\n",
2071 strerror(errno));
2072 goto out;
2073 }
2074 if (read(fd, super->migr_rec_buf, 512) != 512) {
2075 fprintf(stderr,
2076 Name ": Cannot read migr record block: %s\n",
2077 strerror(errno));
2078 goto out;
2079 }
2080 ret_val = 0;
2081
2082out:
2083 return ret_val;
2084}
2085
2086/*******************************************************************************
2087 * Function: load_imsm_migr_rec
2088 * Description: Function reads imsm migration record (it is stored at the last
2089 * sector of disk)
2090 * Parameters:
2091 * super : imsm internal array info
2092 * info : general array info
2093 * Returns:
2094 * 0 : success
2095 * -1 : fail
2096 ******************************************************************************/
2097static int load_imsm_migr_rec(struct intel_super *super, struct mdinfo *info)
2098{
2099 struct mdinfo *sd;
2100 struct dl *dl = NULL;
2101 char nm[30];
2102 int retval = -1;
2103 int fd = -1;
2104
2105 if (info) {
2106 for (sd = info->devs ; sd ; sd = sd->next) {
2107 /* read only from one of the first two slots */
2108 if ((sd->disk.raid_disk > 1) ||
2109 (sd->disk.raid_disk < 0))
2110 continue;
2111 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2112 fd = dev_open(nm, O_RDONLY);
2113 if (fd >= 0)
2114 break;
2115 }
2116 }
2117 if (fd < 0) {
2118 for (dl = super->disks; dl; dl = dl->next) {
2119 /* read only from one of the first two slots */
2120 if (dl->index > 1)
2121 continue;
2122 sprintf(nm, "%d:%d", dl->major, dl->minor);
2123 fd = dev_open(nm, O_RDONLY);
2124 if (fd >= 0)
2125 break;
2126 }
2127 }
2128 if (fd < 0)
2129 goto out;
2130 retval = read_imsm_migr_rec(fd, super);
2131
2132out:
2133 if (fd >= 0)
2134 close(fd);
2135 return retval;
2136}
2137
2138#ifndef MDASSEMBLE
2139/*******************************************************************************
2140 * function: imsm_create_metadata_checkpoint_update
2141 * Description: It creates update for checkpoint change.
2142 * Parameters:
2143 * super : imsm internal array info
2144 * u : pointer to prepared update
2145 * Returns:
2146 * Uptate length.
2147 * If length is equal to 0, input pointer u contains no update
2148 ******************************************************************************/
2149static int imsm_create_metadata_checkpoint_update(
2150 struct intel_super *super,
2151 struct imsm_update_general_migration_checkpoint **u)
2152{
2153
2154 int update_memory_size = 0;
2155
2156 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2157
2158 if (u == NULL)
2159 return 0;
2160 *u = NULL;
2161
2162 /* size of all update data without anchor */
2163 update_memory_size =
2164 sizeof(struct imsm_update_general_migration_checkpoint);
2165
2166 *u = calloc(1, update_memory_size);
2167 if (*u == NULL) {
2168 dprintf("error: cannot get memory for "
2169 "imsm_create_metadata_checkpoint_update update\n");
2170 return 0;
2171 }
2172 (*u)->type = update_general_migration_checkpoint;
2173 (*u)->curr_migr_unit = __le32_to_cpu(super->migr_rec->curr_migr_unit);
2174 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2175 (*u)->curr_migr_unit);
2176
2177 return update_memory_size;
2178}
2179
2180
2181static void imsm_update_metadata_locally(struct supertype *st,
2182 void *buf, int len);
2183
2184/*******************************************************************************
2185 * Function: write_imsm_migr_rec
2186 * Description: Function writes imsm migration record
2187 * (at the last sector of disk)
2188 * Parameters:
2189 * super : imsm internal array info
2190 * Returns:
2191 * 0 : success
2192 * -1 : if fail
2193 ******************************************************************************/
2194static int write_imsm_migr_rec(struct supertype *st)
2195{
2196 struct intel_super *super = st->sb;
2197 unsigned long long dsize;
2198 char nm[30];
2199 int fd = -1;
2200 int retval = -1;
2201 struct dl *sd;
2202 int len;
2203 struct imsm_update_general_migration_checkpoint *u;
2204
2205 for (sd = super->disks ; sd ; sd = sd->next) {
2206 /* write to 2 first slots only */
2207 if ((sd->index < 0) || (sd->index > 1))
2208 continue;
2209 sprintf(nm, "%d:%d", sd->major, sd->minor);
2210 fd = dev_open(nm, O_RDWR);
2211 if (fd < 0)
2212 continue;
2213 get_dev_size(fd, NULL, &dsize);
2214 if (lseek64(fd, dsize - 512, SEEK_SET) < 0) {
2215 fprintf(stderr,
2216 Name ": Cannot seek to anchor block: %s\n",
2217 strerror(errno));
2218 goto out;
2219 }
2220 if (write(fd, super->migr_rec_buf, 512) != 512) {
2221 fprintf(stderr,
2222 Name ": Cannot write migr record block: %s\n",
2223 strerror(errno));
2224 goto out;
2225 }
2226 close(fd);
2227 fd = -1;
2228 }
2229 /* update checkpoint information in metadata */
2230 len = imsm_create_metadata_checkpoint_update(super, &u);
2231
2232 if (len <= 0) {
2233 dprintf("imsm: Cannot prepare update\n");
2234 goto out;
2235 }
2236 /* update metadata locally */
2237 imsm_update_metadata_locally(st, u, len);
2238 /* and possibly remotely */
2239 if (st->update_tail) {
2240 append_metadata_update(st, u, len);
2241 /* during reshape we do all work inside metadata handler
2242 * manage_reshape(), so metadata update has to be triggered
2243 * insida it
2244 */
2245 flush_metadata_updates(st);
2246 st->update_tail = &st->updates;
2247 } else
2248 free(u);
2249
2250 retval = 0;
2251 out:
2252 if (fd >= 0)
2253 close(fd);
2254 return retval;
2255}
2256#endif /* MDASSEMBLE */
2257
2258/* spare/missing disks activations are not allowe when
2259 * array/container performs reshape operation, because
2260 * all arrays in container works on the same disks set
2261 */
2262int imsm_reshape_blocks_arrays_changes(struct intel_super *super)
2263{
2264 int rv = 0;
2265 struct intel_dev *i_dev;
2266 struct imsm_dev *dev;
2267
2268 /* check whole container
2269 */
2270 for (i_dev = super->devlist; i_dev; i_dev = i_dev->next) {
2271 dev = i_dev->dev;
2272 if (is_gen_migration(dev)) {
2273 /* No repair during any migration in container
2274 */
2275 rv = 1;
2276 break;
2277 }
2278 }
2279 return rv;
2280}
2281
2282static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info, char *dmap)
2283{
2284 struct intel_super *super = st->sb;
2285 struct migr_record *migr_rec = super->migr_rec;
2286 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
2287 struct imsm_map *map = get_imsm_map(dev, 0);
2288 struct imsm_map *prev_map = get_imsm_map(dev, 1);
2289 struct imsm_map *map_to_analyse = map;
2290 struct dl *dl;
2291 char *devname;
2292 unsigned int component_size_alligment;
2293 int map_disks = info->array.raid_disks;
2294
2295 memset(info, 0, sizeof(*info));
2296 if (prev_map)
2297 map_to_analyse = prev_map;
2298
2299 dl = super->current_disk;
2300
2301 info->container_member = super->current_vol;
2302 info->array.raid_disks = map->num_members;
2303 info->array.level = get_imsm_raid_level(map_to_analyse);
2304 info->array.layout = imsm_level_to_layout(info->array.level);
2305 info->array.md_minor = -1;
2306 info->array.ctime = 0;
2307 info->array.utime = 0;
2308 info->array.chunk_size =
2309 __le16_to_cpu(map_to_analyse->blocks_per_strip) << 9;
2310 info->array.state = !dev->vol.dirty;
2311 info->custom_array_size = __le32_to_cpu(dev->size_high);
2312 info->custom_array_size <<= 32;
2313 info->custom_array_size |= __le32_to_cpu(dev->size_low);
2314 info->recovery_blocked = imsm_reshape_blocks_arrays_changes(st->sb);
2315
2316 if (is_gen_migration(dev)) {
2317 info->reshape_active = 1;
2318 info->new_level = get_imsm_raid_level(map);
2319 info->new_layout = imsm_level_to_layout(info->new_level);
2320 info->new_chunk = __le16_to_cpu(map->blocks_per_strip) << 9;
2321 info->delta_disks = map->num_members - prev_map->num_members;
2322 if (info->delta_disks) {
2323 /* this needs to be applied to every array
2324 * in the container.
2325 */
2326 info->reshape_active = CONTAINER_RESHAPE;
2327 }
2328 /* We shape information that we give to md might have to be
2329 * modify to cope with md's requirement for reshaping arrays.
2330 * For example, when reshaping a RAID0, md requires it to be
2331 * presented as a degraded RAID4.
2332 * Also if a RAID0 is migrating to a RAID5 we need to specify
2333 * the array as already being RAID5, but the 'before' layout
2334 * is a RAID4-like layout.
2335 */
2336 switch (info->array.level) {
2337 case 0:
2338 switch(info->new_level) {
2339 case 0:
2340 /* conversion is happening as RAID4 */
2341 info->array.level = 4;
2342 info->array.raid_disks += 1;
2343 break;
2344 case 5:
2345 /* conversion is happening as RAID5 */
2346 info->array.level = 5;
2347 info->array.layout = ALGORITHM_PARITY_N;
2348 info->delta_disks -= 1;
2349 break;
2350 default:
2351 /* FIXME error message */
2352 info->array.level = UnSet;
2353 break;
2354 }
2355 break;
2356 }
2357 } else {
2358 info->new_level = UnSet;
2359 info->new_layout = UnSet;
2360 info->new_chunk = info->array.chunk_size;
2361 info->delta_disks = 0;
2362 }
2363
2364 if (dl) {
2365 info->disk.major = dl->major;
2366 info->disk.minor = dl->minor;
2367 info->disk.number = dl->index;
2368 info->disk.raid_disk = get_imsm_disk_slot(map_to_analyse,
2369 dl->index);
2370 }
2371
2372 info->data_offset = __le32_to_cpu(map_to_analyse->pba_of_lba0);
2373 info->component_size =
2374 __le32_to_cpu(map_to_analyse->blocks_per_member);
2375
2376 /* check component size aligment
2377 */
2378 component_size_alligment =
2379 info->component_size % (info->array.chunk_size/512);
2380
2381 if (component_size_alligment &&
2382 (info->array.level != 1) && (info->array.level != UnSet)) {
2383 dprintf("imsm: reported component size alligned from %llu ",
2384 info->component_size);
2385 info->component_size -= component_size_alligment;
2386 dprintf("to %llu (%i).\n",
2387 info->component_size, component_size_alligment);
2388 }
2389
2390 memset(info->uuid, 0, sizeof(info->uuid));
2391 info->recovery_start = MaxSector;
2392
2393 info->reshape_progress = 0;
2394 info->resync_start = MaxSector;
2395 if ((map_to_analyse->map_state == IMSM_T_STATE_UNINITIALIZED ||
2396 dev->vol.dirty) &&
2397 imsm_reshape_blocks_arrays_changes(super) == 0) {
2398 info->resync_start = 0;
2399 }
2400 if (dev->vol.migr_state) {
2401 switch (migr_type(dev)) {
2402 case MIGR_REPAIR:
2403 case MIGR_INIT: {
2404 __u64 blocks_per_unit = blocks_per_migr_unit(super,
2405 dev);
2406 __u64 units = __le32_to_cpu(dev->vol.curr_migr_unit);
2407
2408 info->resync_start = blocks_per_unit * units;
2409 break;
2410 }
2411 case MIGR_GEN_MIGR: {
2412 __u64 blocks_per_unit = blocks_per_migr_unit(super,
2413 dev);
2414 __u64 units = __le32_to_cpu(migr_rec->curr_migr_unit);
2415 unsigned long long array_blocks;
2416 int used_disks;
2417
2418 if (__le32_to_cpu(migr_rec->ascending_migr) &&
2419 (units <
2420 (__le32_to_cpu(migr_rec->num_migr_units)-1)) &&
2421 (super->migr_rec->rec_status ==
2422 __cpu_to_le32(UNIT_SRC_IN_CP_AREA)))
2423 units++;
2424
2425 info->reshape_progress = blocks_per_unit * units;
2426
2427 dprintf("IMSM: General Migration checkpoint : %llu "
2428 "(%llu) -> read reshape progress : %llu\n",
2429 (unsigned long long)units,
2430 (unsigned long long)blocks_per_unit,
2431 info->reshape_progress);
2432
2433 used_disks = imsm_num_data_members(dev, 1);
2434 if (used_disks > 0) {
2435 array_blocks = map->blocks_per_member *
2436 used_disks;
2437 /* round array size down to closest MB
2438 */
2439 info->custom_array_size = (array_blocks
2440 >> SECT_PER_MB_SHIFT)
2441 << SECT_PER_MB_SHIFT;
2442 }
2443 }
2444 case MIGR_VERIFY:
2445 /* we could emulate the checkpointing of
2446 * 'sync_action=check' migrations, but for now
2447 * we just immediately complete them
2448 */
2449 case MIGR_REBUILD:
2450 /* this is handled by container_content_imsm() */
2451 case MIGR_STATE_CHANGE:
2452 /* FIXME handle other migrations */
2453 default:
2454 /* we are not dirty, so... */
2455 info->resync_start = MaxSector;
2456 }
2457 }
2458
2459 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
2460 info->name[MAX_RAID_SERIAL_LEN] = 0;
2461
2462 info->array.major_version = -1;
2463 info->array.minor_version = -2;
2464 devname = devnum2devname(st->container_dev);
2465 *info->text_version = '\0';
2466 if (devname)
2467 sprintf(info->text_version, "/%s/%d", devname, info->container_member);
2468 free(devname);
2469 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
2470 uuid_from_super_imsm(st, info->uuid);
2471
2472 if (dmap) {
2473 int i, j;
2474 for (i=0; i<map_disks; i++) {
2475 dmap[i] = 0;
2476 if (i < info->array.raid_disks) {
2477 struct imsm_disk *dsk;
2478 j = get_imsm_disk_idx(dev, i, -1);
2479 dsk = get_imsm_disk(super, j);
2480 if (dsk && (dsk->status & CONFIGURED_DISK))
2481 dmap[i] = 1;
2482 }
2483 }
2484 }
2485}
2486
2487static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev,
2488 int failed, int look_in_map);
2489
2490static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev,
2491 int look_in_map);
2492
2493static void manage_second_map(struct intel_super *super, struct imsm_dev *dev)
2494{
2495 if (is_gen_migration(dev)) {
2496 int failed;
2497 __u8 map_state;
2498 struct imsm_map *map2 = get_imsm_map(dev, MAP_1);
2499
2500 failed = imsm_count_failed(super, dev, MAP_1);
2501 map_state = imsm_check_degraded(super, dev, failed,
2502 MAP_1);
2503 if (map2->map_state != map_state) {
2504 map2->map_state = map_state;
2505 super->updates_pending++;
2506 }
2507 }
2508}
2509
2510static struct imsm_disk *get_imsm_missing(struct intel_super *super, __u8 index)
2511{
2512 struct dl *d;
2513
2514 for (d = super->missing; d; d = d->next)
2515 if (d->index == index)
2516 return &d->disk;
2517 return NULL;
2518}
2519
2520static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map)
2521{
2522 struct intel_super *super = st->sb;
2523 struct imsm_disk *disk;
2524 int map_disks = info->array.raid_disks;
2525 int max_enough = -1;
2526 int i;
2527 struct imsm_super *mpb;
2528
2529 if (super->current_vol >= 0) {
2530 getinfo_super_imsm_volume(st, info, map);
2531 return;
2532 }
2533 memset(info, 0, sizeof(*info));
2534
2535 /* Set raid_disks to zero so that Assemble will always pull in valid
2536 * spares
2537 */
2538 info->array.raid_disks = 0;
2539 info->array.level = LEVEL_CONTAINER;
2540 info->array.layout = 0;
2541 info->array.md_minor = -1;
2542 info->array.ctime = 0; /* N/A for imsm */
2543 info->array.utime = 0;
2544 info->array.chunk_size = 0;
2545
2546 info->disk.major = 0;
2547 info->disk.minor = 0;
2548 info->disk.raid_disk = -1;
2549 info->reshape_active = 0;
2550 info->array.major_version = -1;
2551 info->array.minor_version = -2;
2552 strcpy(info->text_version, "imsm");
2553 info->safe_mode_delay = 0;
2554 info->disk.number = -1;
2555 info->disk.state = 0;
2556 info->name[0] = 0;
2557 info->recovery_start = MaxSector;
2558 info->recovery_blocked = imsm_reshape_blocks_arrays_changes(st->sb);
2559
2560 /* do we have the all the insync disks that we expect? */
2561 mpb = super->anchor;
2562
2563 for (i = 0; i < mpb->num_raid_devs; i++) {
2564 struct imsm_dev *dev = get_imsm_dev(super, i);
2565 int failed, enough, j, missing = 0;
2566 struct imsm_map *map;
2567 __u8 state;
2568
2569 failed = imsm_count_failed(super, dev, MAP_0);
2570 state = imsm_check_degraded(super, dev, failed, MAP_0);
2571 map = get_imsm_map(dev, 0);
2572
2573 /* any newly missing disks?
2574 * (catches single-degraded vs double-degraded)
2575 */
2576 for (j = 0; j < map->num_members; j++) {
2577 __u32 ord = get_imsm_ord_tbl_ent(dev, j, 0);
2578 __u32 idx = ord_to_idx(ord);
2579
2580 if (!(ord & IMSM_ORD_REBUILD) &&
2581 get_imsm_missing(super, idx)) {
2582 missing = 1;
2583 break;
2584 }
2585 }
2586
2587 if (state == IMSM_T_STATE_FAILED)
2588 enough = -1;
2589 else if (state == IMSM_T_STATE_DEGRADED &&
2590 (state != map->map_state || missing))
2591 enough = 0;
2592 else /* we're normal, or already degraded */
2593 enough = 1;
2594
2595 /* in the missing/failed disk case check to see
2596 * if at least one array is runnable
2597 */
2598 max_enough = max(max_enough, enough);
2599 }
2600 dprintf("%s: enough: %d\n", __func__, max_enough);
2601 info->container_enough = max_enough;
2602
2603 if (super->disks) {
2604 __u32 reserved = imsm_reserved_sectors(super, super->disks);
2605
2606 disk = &super->disks->disk;
2607 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
2608 info->component_size = reserved;
2609 info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
2610 /* we don't change info->disk.raid_disk here because
2611 * this state will be finalized in mdmon after we have
2612 * found the 'most fresh' version of the metadata
2613 */
2614 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
2615 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
2616 }
2617
2618 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2619 * ->compare_super may have updated the 'num_raid_devs' field for spares
2620 */
2621 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
2622 uuid_from_super_imsm(st, info->uuid);
2623 else
2624 memcpy(info->uuid, uuid_zero, sizeof(uuid_zero));
2625
2626 /* I don't know how to compute 'map' on imsm, so use safe default */
2627 if (map) {
2628 int i;
2629 for (i = 0; i < map_disks; i++)
2630 map[i] = 1;
2631 }
2632
2633}
2634
2635/* allocates memory and fills disk in mdinfo structure
2636 * for each disk in array */
2637struct mdinfo *getinfo_super_disks_imsm(struct supertype *st)
2638{
2639 struct mdinfo *mddev = NULL;
2640 struct intel_super *super = st->sb;
2641 struct imsm_disk *disk;
2642 int count = 0;
2643 struct dl *dl;
2644 if (!super || !super->disks)
2645 return NULL;
2646 dl = super->disks;
2647 mddev = malloc(sizeof(*mddev));
2648 if (!mddev) {
2649 fprintf(stderr, Name ": Failed to allocate memory.\n");
2650 return NULL;
2651 }
2652 memset(mddev, 0, sizeof(*mddev));
2653 while (dl) {
2654 struct mdinfo *tmp;
2655 disk = &dl->disk;
2656 tmp = malloc(sizeof(*tmp));
2657 if (!tmp) {
2658 fprintf(stderr, Name ": Failed to allocate memory.\n");
2659 if (mddev)
2660 sysfs_free(mddev);
2661 return NULL;
2662 }
2663 memset(tmp, 0, sizeof(*tmp));
2664 if (mddev->devs)
2665 tmp->next = mddev->devs;
2666 mddev->devs = tmp;
2667 tmp->disk.number = count++;
2668 tmp->disk.major = dl->major;
2669 tmp->disk.minor = dl->minor;
2670 tmp->disk.state = is_configured(disk) ?
2671 (1 << MD_DISK_ACTIVE) : 0;
2672 tmp->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
2673 tmp->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
2674 tmp->disk.raid_disk = -1;
2675 dl = dl->next;
2676 }
2677 return mddev;
2678}
2679
2680static int update_super_imsm(struct supertype *st, struct mdinfo *info,
2681 char *update, char *devname, int verbose,
2682 int uuid_set, char *homehost)
2683{
2684 /* For 'assemble' and 'force' we need to return non-zero if any
2685 * change was made. For others, the return value is ignored.
2686 * Update options are:
2687 * force-one : This device looks a bit old but needs to be included,
2688 * update age info appropriately.
2689 * assemble: clear any 'faulty' flag to allow this device to
2690 * be assembled.
2691 * force-array: Array is degraded but being forced, mark it clean
2692 * if that will be needed to assemble it.
2693 *
2694 * newdev: not used ????
2695 * grow: Array has gained a new device - this is currently for
2696 * linear only
2697 * resync: mark as dirty so a resync will happen.
2698 * name: update the name - preserving the homehost
2699 * uuid: Change the uuid of the array to match watch is given
2700 *
2701 * Following are not relevant for this imsm:
2702 * sparc2.2 : update from old dodgey metadata
2703 * super-minor: change the preferred_minor number
2704 * summaries: update redundant counters.
2705 * homehost: update the recorded homehost
2706 * _reshape_progress: record new reshape_progress position.
2707 */
2708 int rv = 1;
2709 struct intel_super *super = st->sb;
2710 struct imsm_super *mpb;
2711
2712 /* we can only update container info */
2713 if (!super || super->current_vol >= 0 || !super->anchor)
2714 return 1;
2715
2716 mpb = super->anchor;
2717
2718 if (strcmp(update, "uuid") == 0 && uuid_set && !info->update_private)
2719 rv = -1;
2720 else if (strcmp(update, "uuid") == 0 && uuid_set && info->update_private) {
2721 mpb->orig_family_num = *((__u32 *) info->update_private);
2722 rv = 0;
2723 } else if (strcmp(update, "uuid") == 0) {
2724 __u32 *new_family = malloc(sizeof(*new_family));
2725
2726 /* update orig_family_number with the incoming random
2727 * data, report the new effective uuid, and store the
2728 * new orig_family_num for future updates.
2729 */
2730 if (new_family) {
2731 memcpy(&mpb->orig_family_num, info->uuid, sizeof(__u32));
2732 uuid_from_super_imsm(st, info->uuid);
2733 *new_family = mpb->orig_family_num;
2734 info->update_private = new_family;
2735 rv = 0;
2736 }
2737 } else if (strcmp(update, "assemble") == 0)
2738 rv = 0;
2739 else
2740 rv = -1;
2741
2742 /* successful update? recompute checksum */
2743 if (rv == 0)
2744 mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb));
2745
2746 return rv;
2747}
2748
2749static size_t disks_to_mpb_size(int disks)
2750{
2751 size_t size;
2752
2753 size = sizeof(struct imsm_super);
2754 size += (disks - 1) * sizeof(struct imsm_disk);
2755 size += 2 * sizeof(struct imsm_dev);
2756 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2757 size += (4 - 2) * sizeof(struct imsm_map);
2758 /* 4 possible disk_ord_tbl's */
2759 size += 4 * (disks - 1) * sizeof(__u32);
2760
2761 return size;
2762}
2763
2764static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
2765{
2766 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
2767 return 0;
2768
2769 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
2770}
2771
2772static void free_devlist(struct intel_super *super)
2773{
2774 struct intel_dev *dv;
2775
2776 while (super->devlist) {
2777 dv = super->devlist->next;
2778 free(super->devlist->dev);
2779 free(super->devlist);
2780 super->devlist = dv;
2781 }
2782}
2783
2784static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
2785{
2786 memcpy(dest, src, sizeof_imsm_dev(src, 0));
2787}
2788
2789static int compare_super_imsm(struct supertype *st, struct supertype *tst)
2790{
2791 /*
2792 * return:
2793 * 0 same, or first was empty, and second was copied
2794 * 1 second had wrong number
2795 * 2 wrong uuid
2796 * 3 wrong other info
2797 */
2798 struct intel_super *first = st->sb;
2799 struct intel_super *sec = tst->sb;
2800
2801 if (!first) {
2802 st->sb = tst->sb;
2803 tst->sb = NULL;
2804 return 0;
2805 }
2806 /* in platform dependent environment test if the disks
2807 * use the same Intel hba
2808 */
2809 if (!check_env("IMSM_NO_PLATFORM")) {
2810 if (!first->hba || !sec->hba ||
2811 (first->hba->type != sec->hba->type)) {
2812 fprintf(stderr,
2813 "HBAs of devices does not match %s != %s\n",
2814 first->hba ? get_sys_dev_type(first->hba->type) : NULL,
2815 sec->hba ? get_sys_dev_type(sec->hba->type) : NULL);
2816 return 3;
2817 }
2818 }
2819
2820 /* if an anchor does not have num_raid_devs set then it is a free
2821 * floating spare
2822 */
2823 if (first->anchor->num_raid_devs > 0 &&
2824 sec->anchor->num_raid_devs > 0) {
2825 /* Determine if these disks might ever have been
2826 * related. Further disambiguation can only take place
2827 * in load_super_imsm_all
2828 */
2829 __u32 first_family = first->anchor->orig_family_num;
2830 __u32 sec_family = sec->anchor->orig_family_num;
2831
2832 if (memcmp(first->anchor->sig, sec->anchor->sig,
2833 MAX_SIGNATURE_LENGTH) != 0)
2834 return 3;
2835
2836 if (first_family == 0)
2837 first_family = first->anchor->family_num;
2838 if (sec_family == 0)
2839 sec_family = sec->anchor->family_num;
2840
2841 if (first_family != sec_family)
2842 return 3;
2843
2844 }
2845
2846
2847 /* if 'first' is a spare promote it to a populated mpb with sec's
2848 * family number
2849 */
2850 if (first->anchor->num_raid_devs == 0 &&
2851 sec->anchor->num_raid_devs > 0) {
2852 int i;
2853 struct intel_dev *dv;
2854 struct imsm_dev *dev;
2855
2856 /* we need to copy raid device info from sec if an allocation
2857 * fails here we don't associate the spare
2858 */
2859 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
2860 dv = malloc(sizeof(*dv));
2861 if (!dv)
2862 break;
2863 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
2864 if (!dev) {
2865 free(dv);
2866 break;
2867 }
2868 dv->dev = dev;
2869 dv->index = i;
2870 dv->next = first->devlist;
2871 first->devlist = dv;
2872 }
2873 if (i < sec->anchor->num_raid_devs) {
2874 /* allocation failure */
2875 free_devlist(first);
2876 fprintf(stderr, "imsm: failed to associate spare\n");
2877 return 3;
2878 }
2879 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
2880 first->anchor->orig_family_num = sec->anchor->orig_family_num;
2881 first->anchor->family_num = sec->anchor->family_num;
2882 memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH);
2883 for (i = 0; i < sec->anchor->num_raid_devs; i++)
2884 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
2885 }
2886
2887 return 0;
2888}
2889
2890static void fd2devname(int fd, char *name)
2891{
2892 struct stat st;
2893 char path[256];
2894 char dname[PATH_MAX];
2895 char *nm;
2896 int rv;
2897
2898 name[0] = '\0';
2899 if (fstat(fd, &st) != 0)
2900 return;
2901 sprintf(path, "/sys/dev/block/%d:%d",
2902 major(st.st_rdev), minor(st.st_rdev));
2903
2904 rv = readlink(path, dname, sizeof(dname)-1);
2905 if (rv <= 0)
2906 return;
2907
2908 dname[rv] = '\0';
2909 nm = strrchr(dname, '/');
2910 if (nm) {
2911 nm++;
2912 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
2913 }
2914}
2915
2916extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
2917
2918static int imsm_read_serial(int fd, char *devname,
2919 __u8 serial[MAX_RAID_SERIAL_LEN])
2920{
2921 unsigned char scsi_serial[255];
2922 int rv;
2923 int rsp_len;
2924 int len;
2925 char *dest;
2926 char *src;
2927 char *rsp_buf;
2928 int i;
2929
2930 memset(scsi_serial, 0, sizeof(scsi_serial));
2931
2932 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
2933
2934 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
2935 memset(serial, 0, MAX_RAID_SERIAL_LEN);
2936 fd2devname(fd, (char *) serial);
2937 return 0;
2938 }
2939
2940 if (rv != 0) {
2941 if (devname)
2942 fprintf(stderr,
2943 Name ": Failed to retrieve serial for %s\n",
2944 devname);
2945 return rv;
2946 }
2947
2948 rsp_len = scsi_serial[3];
2949 if (!rsp_len) {
2950 if (devname)
2951 fprintf(stderr,
2952 Name ": Failed to retrieve serial for %s\n",
2953 devname);
2954 return 2;
2955 }
2956 rsp_buf = (char *) &scsi_serial[4];
2957
2958 /* trim all whitespace and non-printable characters and convert
2959 * ':' to ';'
2960 */
2961 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
2962 src = &rsp_buf[i];
2963 if (*src > 0x20) {
2964 /* ':' is reserved for use in placeholder serial
2965 * numbers for missing disks
2966 */
2967 if (*src == ':')
2968 *dest++ = ';';
2969 else
2970 *dest++ = *src;
2971 }
2972 }
2973 len = dest - rsp_buf;
2974 dest = rsp_buf;
2975
2976 /* truncate leading characters */
2977 if (len > MAX_RAID_SERIAL_LEN) {
2978 dest += len - MAX_RAID_SERIAL_LEN;
2979 len = MAX_RAID_SERIAL_LEN;
2980 }
2981
2982 memset(serial, 0, MAX_RAID_SERIAL_LEN);
2983 memcpy(serial, dest, len);
2984
2985 return 0;
2986}
2987
2988static int serialcmp(__u8 *s1, __u8 *s2)
2989{
2990 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
2991}
2992
2993static void serialcpy(__u8 *dest, __u8 *src)
2994{
2995 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
2996}
2997
2998static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
2999{
3000 struct dl *dl;
3001
3002 for (dl = super->disks; dl; dl = dl->next)
3003 if (serialcmp(dl->serial, serial) == 0)
3004 break;
3005
3006 return dl;
3007}
3008
3009static struct imsm_disk *
3010__serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
3011{
3012 int i;
3013
3014 for (i = 0; i < mpb->num_disks; i++) {
3015 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
3016
3017 if (serialcmp(disk->serial, serial) == 0) {
3018 if (idx)
3019 *idx = i;
3020 return disk;
3021 }
3022 }
3023
3024 return NULL;
3025}
3026
3027static int
3028load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
3029{
3030 struct imsm_disk *disk;
3031 struct dl *dl;
3032 struct stat stb;
3033 int rv;
3034 char name[40];
3035 __u8 serial[MAX_RAID_SERIAL_LEN];
3036
3037 rv = imsm_read_serial(fd, devname, serial);
3038
3039 if (rv != 0)
3040 return 2;
3041
3042 dl = calloc(1, sizeof(*dl));
3043 if (!dl) {
3044 if (devname)
3045 fprintf(stderr,
3046 Name ": failed to allocate disk buffer for %s\n",
3047 devname);
3048 return 2;
3049 }
3050
3051 fstat(fd, &stb);
3052 dl->major = major(stb.st_rdev);
3053 dl->minor = minor(stb.st_rdev);
3054 dl->next = super->disks;
3055 dl->fd = keep_fd ? fd : -1;
3056 assert(super->disks == NULL);
3057 super->disks = dl;
3058 serialcpy(dl->serial, serial);
3059 dl->index = -2;
3060 dl->e = NULL;
3061 fd2devname(fd, name);
3062 if (devname)
3063 dl->devname = strdup(devname);
3064 else
3065 dl->devname = strdup(name);
3066
3067 /* look up this disk's index in the current anchor */
3068 disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
3069 if (disk) {
3070 dl->disk = *disk;
3071 /* only set index on disks that are a member of a
3072 * populated contianer, i.e. one with raid_devs
3073 */
3074 if (is_failed(&dl->disk))
3075 dl->index = -2;
3076 else if (is_spare(&dl->disk))
3077 dl->index = -1;
3078 }
3079
3080 return 0;
3081}
3082
3083#ifndef MDASSEMBLE
3084/* When migrating map0 contains the 'destination' state while map1
3085 * contains the current state. When not migrating map0 contains the
3086 * current state. This routine assumes that map[0].map_state is set to
3087 * the current array state before being called.
3088 *
3089 * Migration is indicated by one of the following states
3090 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3091 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3092 * map1state=unitialized)
3093 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3094 * map1state=normal)
3095 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3096 * map1state=degraded)
3097 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3098 * map1state=normal)
3099 */
3100static void migrate(struct imsm_dev *dev, struct intel_super *super,
3101 __u8 to_state, int migr_type)
3102{
3103 struct imsm_map *dest;
3104 struct imsm_map *src = get_imsm_map(dev, 0);
3105
3106 dev->vol.migr_state = 1;
3107 set_migr_type(dev, migr_type);
3108 dev->vol.curr_migr_unit = 0;
3109 dest = get_imsm_map(dev, 1);
3110
3111 /* duplicate and then set the target end state in map[0] */
3112 memcpy(dest, src, sizeof_imsm_map(src));
3113 if ((migr_type == MIGR_REBUILD) ||
3114 (migr_type == MIGR_GEN_MIGR)) {
3115 __u32 ord;
3116 int i;
3117
3118 for (i = 0; i < src->num_members; i++) {
3119 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
3120 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
3121 }
3122 }
3123
3124 if (migr_type == MIGR_GEN_MIGR)
3125 /* Clear migration record */
3126 memset(super->migr_rec, 0, sizeof(struct migr_record));
3127
3128 src->map_state = to_state;
3129}
3130
3131static void end_migration(struct imsm_dev *dev, struct intel_super *super,
3132 __u8 map_state)
3133{
3134 struct imsm_map *map = get_imsm_map(dev, 0);
3135 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3136 int i, j;
3137
3138 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3139 * completed in the last migration.
3140 *
3141 * FIXME add support for raid-level-migration
3142 */
3143 if ((map_state != map->map_state) && (is_gen_migration(dev) == 0) &&
3144 (prev->map_state != IMSM_T_STATE_UNINITIALIZED)) {
3145 /* when final map state is other than expected
3146 * merge maps (not for migration)
3147 */
3148 int failed;
3149
3150 for (i = 0; i < prev->num_members; i++)
3151 for (j = 0; j < map->num_members; j++)
3152 /* during online capacity expansion
3153 * disks position can be changed
3154 * if takeover is used
3155 */
3156 if (ord_to_idx(map->disk_ord_tbl[j]) ==
3157 ord_to_idx(prev->disk_ord_tbl[i])) {
3158 map->disk_ord_tbl[j] |=
3159 prev->disk_ord_tbl[i];
3160 break;
3161 }
3162 failed = imsm_count_failed(super, dev, MAP_0);
3163 map_state = imsm_check_degraded(super, dev, failed, MAP_0);
3164 }
3165
3166 dev->vol.migr_state = 0;
3167 set_migr_type(dev, 0);
3168 dev->vol.curr_migr_unit = 0;
3169 map->map_state = map_state;
3170}
3171#endif
3172
3173static int parse_raid_devices(struct intel_super *super)
3174{
3175 int i;
3176 struct imsm_dev *dev_new;
3177 size_t len, len_migr;
3178 size_t max_len = 0;
3179 size_t space_needed = 0;
3180 struct imsm_super *mpb = super->anchor;
3181
3182 for (i = 0; i < super->anchor->num_raid_devs; i++) {
3183 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
3184 struct intel_dev *dv;
3185
3186 len = sizeof_imsm_dev(dev_iter, 0);
3187 len_migr = sizeof_imsm_dev(dev_iter, 1);
3188 if (len_migr > len)
3189 space_needed += len_migr - len;
3190
3191 dv = malloc(sizeof(*dv));
3192 if (!dv)
3193 return 1;
3194 if (max_len < len_migr)
3195 max_len = len_migr;
3196 if (max_len > len_migr)
3197 space_needed += max_len - len_migr;
3198 dev_new = malloc(max_len);
3199 if (!dev_new) {
3200 free(dv);
3201 return 1;
3202 }
3203 imsm_copy_dev(dev_new, dev_iter);
3204 dv->dev = dev_new;
3205 dv->index = i;
3206 dv->next = super->devlist;
3207 super->devlist = dv;
3208 }
3209
3210 /* ensure that super->buf is large enough when all raid devices
3211 * are migrating
3212 */
3213 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
3214 void *buf;
3215
3216 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
3217 if (posix_memalign(&buf, 512, len) != 0)
3218 return 1;
3219
3220 memcpy(buf, super->buf, super->len);
3221 memset(buf + super->len, 0, len - super->len);
3222 free(super->buf);
3223 super->buf = buf;
3224 super->len = len;
3225 }
3226
3227 return 0;
3228}
3229
3230/* retrieve a pointer to the bbm log which starts after all raid devices */
3231struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
3232{
3233 void *ptr = NULL;
3234
3235 if (__le32_to_cpu(mpb->bbm_log_size)) {
3236 ptr = mpb;
3237 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
3238 }
3239
3240 return ptr;
3241}
3242
3243/*******************************************************************************
3244 * Function: check_mpb_migr_compatibility
3245 * Description: Function checks for unsupported migration features:
3246 * - migration optimization area (pba_of_lba0)
3247 * - descending reshape (ascending_migr)
3248 * Parameters:
3249 * super : imsm metadata information
3250 * Returns:
3251 * 0 : migration is compatible
3252 * -1 : migration is not compatible
3253 ******************************************************************************/
3254int check_mpb_migr_compatibility(struct intel_super *super)
3255{
3256 struct imsm_map *map0, *map1;
3257 struct migr_record *migr_rec = super->migr_rec;
3258 int i;
3259
3260 for (i = 0; i < super->anchor->num_raid_devs; i++) {
3261 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
3262
3263 if (dev_iter &&
3264 dev_iter->vol.migr_state == 1 &&
3265 dev_iter->vol.migr_type == MIGR_GEN_MIGR) {
3266 /* This device is migrating */
3267 map0 = get_imsm_map(dev_iter, 0);
3268 map1 = get_imsm_map(dev_iter, 1);
3269 if (map0->pba_of_lba0 != map1->pba_of_lba0)
3270 /* migration optimization area was used */
3271 return -1;
3272 if (migr_rec->ascending_migr == 0
3273 && migr_rec->dest_depth_per_unit > 0)
3274 /* descending reshape not supported yet */
3275 return -1;
3276 }
3277 }
3278 return 0;
3279}
3280
3281static void __free_imsm(struct intel_super *super, int free_disks);
3282
3283/* load_imsm_mpb - read matrix metadata
3284 * allocates super->mpb to be freed by free_imsm
3285 */
3286static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
3287{
3288 unsigned long long dsize;
3289 unsigned long long sectors;
3290 struct stat;
3291 struct imsm_super *anchor;
3292 __u32 check_sum;
3293
3294 get_dev_size(fd, NULL, &dsize);
3295 if (dsize < 1024) {
3296 if (devname)
3297 fprintf(stderr,
3298 Name ": %s: device to small for imsm\n",
3299 devname);
3300 return 1;
3301 }
3302
3303 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
3304 if (devname)
3305 fprintf(stderr, Name
3306 ": Cannot seek to anchor block on %s: %s\n",
3307 devname, strerror(errno));
3308 return 1;
3309 }
3310
3311 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
3312 if (devname)
3313 fprintf(stderr,
3314 Name ": Failed to allocate imsm anchor buffer"
3315 " on %s\n", devname);
3316 return 1;
3317 }
3318 if (read(fd, anchor, 512) != 512) {
3319 if (devname)
3320 fprintf(stderr,
3321 Name ": Cannot read anchor block on %s: %s\n",
3322 devname, strerror(errno));
3323 free(anchor);
3324 return 1;
3325 }
3326
3327 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
3328 if (devname)
3329 fprintf(stderr,
3330 Name ": no IMSM anchor on %s\n", devname);
3331 free(anchor);
3332 return 2;
3333 }
3334
3335 __free_imsm(super, 0);
3336 /* reload capability and hba */
3337
3338 /* capability and hba must be updated with new super allocation */
3339 find_intel_hba_capability(fd, super, devname);
3340 super->len = ROUND_UP(anchor->mpb_size, 512);
3341 if (posix_memalign(&super->buf, 512, super->len) != 0) {
3342 if (devname)
3343 fprintf(stderr,
3344 Name ": unable to allocate %zu byte mpb buffer\n",
3345 super->len);
3346 free(anchor);
3347 return 2;
3348 }
3349 memcpy(super->buf, anchor, 512);
3350
3351 sectors = mpb_sectors(anchor) - 1;
3352 free(anchor);
3353
3354 if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) {
3355 fprintf(stderr, Name
3356 ": %s could not allocate migr_rec buffer\n", __func__);
3357 free(super->buf);
3358 return 2;
3359 }
3360
3361 if (!sectors) {
3362 check_sum = __gen_imsm_checksum(super->anchor);
3363 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
3364 if (devname)
3365 fprintf(stderr,
3366 Name ": IMSM checksum %x != %x on %s\n",
3367 check_sum,
3368 __le32_to_cpu(super->anchor->check_sum),
3369 devname);
3370 return 2;
3371 }
3372
3373 return 0;
3374 }
3375
3376 /* read the extended mpb */
3377 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
3378 if (devname)
3379 fprintf(stderr,
3380 Name ": Cannot seek to extended mpb on %s: %s\n",
3381 devname, strerror(errno));
3382 return 1;
3383 }
3384
3385 if ((unsigned)read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
3386 if (devname)
3387 fprintf(stderr,
3388 Name ": Cannot read extended mpb on %s: %s\n",
3389 devname, strerror(errno));
3390 return 2;
3391 }
3392
3393 check_sum = __gen_imsm_checksum(super->anchor);
3394 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
3395 if (devname)
3396 fprintf(stderr,
3397 Name ": IMSM checksum %x != %x on %s\n",
3398 check_sum, __le32_to_cpu(super->anchor->check_sum),
3399 devname);
3400 return 3;
3401 }
3402
3403 /* FIXME the BBM log is disk specific so we cannot use this global
3404 * buffer for all disks. Ok for now since we only look at the global
3405 * bbm_log_size parameter to gate assembly
3406 */
3407 super->bbm_log = __get_imsm_bbm_log(super->anchor);
3408
3409 return 0;
3410}
3411
3412static int read_imsm_migr_rec(int fd, struct intel_super *super);
3413
3414static int
3415load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
3416{
3417 int err;
3418
3419 err = load_imsm_mpb(fd, super, devname);
3420 if (err)
3421 return err;
3422 err = load_imsm_disk(fd, super, devname, keep_fd);
3423 if (err)
3424 return err;
3425 err = parse_raid_devices(super);
3426
3427 return err;
3428}
3429
3430static void __free_imsm_disk(struct dl *d)
3431{
3432 if (d->fd >= 0)
3433 close(d->fd);
3434 if (d->devname)
3435 free(d->devname);
3436 if (d->e)
3437 free(d->e);
3438 free(d);
3439
3440}
3441
3442static void free_imsm_disks(struct intel_super *super)
3443{
3444 struct dl *d;
3445
3446 while (super->disks) {
3447 d = super->disks;
3448 super->disks = d->next;
3449 __free_imsm_disk(d);
3450 }
3451 while (super->disk_mgmt_list) {
3452 d = super->disk_mgmt_list;
3453 super->disk_mgmt_list = d->next;
3454 __free_imsm_disk(d);
3455 }
3456 while (super->missing) {
3457 d = super->missing;
3458 super->missing = d->next;
3459 __free_imsm_disk(d);
3460 }
3461
3462}
3463
3464/* free all the pieces hanging off of a super pointer */
3465static void __free_imsm(struct intel_super *super, int free_disks)
3466{
3467 struct intel_hba *elem, *next;
3468
3469 if (super->buf) {
3470 free(super->buf);
3471 super->buf = NULL;
3472 }
3473 /* unlink capability description */
3474 super->orom = NULL;
3475 if (super->migr_rec_buf) {
3476 free(super->migr_rec_buf);
3477 super->migr_rec_buf = NULL;
3478 }
3479 if (free_disks)
3480 free_imsm_disks(super);
3481 free_devlist(super);
3482 elem = super->hba;
3483 while (elem) {
3484 if (elem->path)
3485 free((void *)elem->path);
3486 next = elem->next;
3487 free(elem);
3488 elem = next;
3489 }
3490 super->hba = NULL;
3491}
3492
3493static void free_imsm(struct intel_super *super)
3494{
3495 __free_imsm(super, 1);
3496 free(super);
3497}
3498
3499static void free_super_imsm(struct supertype *st)
3500{
3501 struct intel_super *super = st->sb;
3502
3503 if (!super)
3504 return;
3505
3506 free_imsm(super);
3507 st->sb = NULL;
3508}
3509
3510static struct intel_super *alloc_super(void)
3511{
3512 struct intel_super *super = malloc(sizeof(*super));
3513
3514 if (super) {
3515 memset(super, 0, sizeof(*super));
3516 super->current_vol = -1;
3517 super->create_offset = ~((__u32 ) 0);
3518 }
3519 return super;
3520}
3521
3522/*
3523 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3524 */
3525static int find_intel_hba_capability(int fd, struct intel_super *super, char *devname)
3526{
3527 struct sys_dev *hba_name;
3528 int rv = 0;
3529
3530 if ((fd < 0) || check_env("IMSM_NO_PLATFORM")) {
3531 super->orom = NULL;
3532 super->hba = NULL;
3533 return 0;
3534 }
3535 hba_name = find_disk_attached_hba(fd, NULL);
3536 if (!hba_name) {
3537 if (devname)
3538 fprintf(stderr,
3539 Name ": %s is not attached to Intel(R) RAID controller.\n",
3540 devname);
3541 return 1;
3542 }
3543 rv = attach_hba_to_super(super, hba_name);
3544 if (rv == 2) {
3545 if (devname) {
3546 struct intel_hba *hba = super->hba;
3547
3548 fprintf(stderr, Name ": %s is attached to Intel(R) %s RAID "
3549 "controller (%s),\n"
3550 " but the container is assigned to Intel(R) "
3551 "%s RAID controller (",
3552 devname,
3553 hba_name->path,
3554 hba_name->pci_id ? : "Err!",
3555 get_sys_dev_type(hba_name->type));
3556
3557 while (hba) {
3558 fprintf(stderr, "%s", hba->pci_id ? : "Err!");
3559 if (hba->next)
3560 fprintf(stderr, ", ");
3561 hba = hba->next;
3562 }
3563
3564 fprintf(stderr, ").\n"
3565 " Mixing devices attached to different controllers "
3566 "is not allowed.\n");
3567 }
3568 free_sys_dev(&hba_name);
3569 return 2;
3570 }
3571 super->orom = find_imsm_capability(hba_name->type);
3572 free_sys_dev(&hba_name);
3573 if (!super->orom)
3574 return 3;
3575 return 0;
3576}
3577
3578/* find_missing - helper routine for load_super_imsm_all that identifies
3579 * disks that have disappeared from the system. This routine relies on
3580 * the mpb being uptodate, which it is at load time.
3581 */
3582static int find_missing(struct intel_super *super)
3583{
3584 int i;
3585 struct imsm_super *mpb = super->anchor;
3586 struct dl *dl;
3587 struct imsm_disk *disk;
3588
3589 for (i = 0; i < mpb->num_disks; i++) {
3590 disk = __get_imsm_disk(mpb, i);
3591 dl = serial_to_dl(disk->serial, super);
3592 if (dl)
3593 continue;
3594
3595 dl = malloc(sizeof(*dl));
3596 if (!dl)
3597 return 1;
3598 dl->major = 0;
3599 dl->minor = 0;
3600 dl->fd = -1;
3601 dl->devname = strdup("missing");
3602 dl->index = i;
3603 serialcpy(dl->serial, disk->serial);
3604 dl->disk = *disk;
3605 dl->e = NULL;
3606 dl->next = super->missing;
3607 super->missing = dl;
3608 }
3609
3610 return 0;
3611}
3612
3613#ifndef MDASSEMBLE
3614static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
3615{
3616 struct intel_disk *idisk = disk_list;
3617
3618 while (idisk) {
3619 if (serialcmp(idisk->disk.serial, serial) == 0)
3620 break;
3621 idisk = idisk->next;
3622 }
3623
3624 return idisk;
3625}
3626
3627static int __prep_thunderdome(struct intel_super **table, int tbl_size,
3628 struct intel_super *super,
3629 struct intel_disk **disk_list)
3630{
3631 struct imsm_disk *d = &super->disks->disk;
3632 struct imsm_super *mpb = super->anchor;
3633 int i, j;
3634
3635 for (i = 0; i < tbl_size; i++) {
3636 struct imsm_super *tbl_mpb = table[i]->anchor;
3637 struct imsm_disk *tbl_d = &table[i]->disks->disk;
3638
3639 if (tbl_mpb->family_num == mpb->family_num) {
3640 if (tbl_mpb->check_sum == mpb->check_sum) {
3641 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3642 __func__, super->disks->major,
3643 super->disks->minor,
3644 table[i]->disks->major,
3645 table[i]->disks->minor);
3646 break;
3647 }
3648
3649 if (((is_configured(d) && !is_configured(tbl_d)) ||
3650 is_configured(d) == is_configured(tbl_d)) &&
3651 tbl_mpb->generation_num < mpb->generation_num) {
3652 /* current version of the mpb is a
3653 * better candidate than the one in
3654 * super_table, but copy over "cross
3655 * generational" status
3656 */
3657 struct intel_disk *idisk;
3658
3659 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3660 __func__, super->disks->major,
3661 super->disks->minor,
3662 table[i]->disks->major,
3663 table[i]->disks->minor);
3664
3665 idisk = disk_list_get(tbl_d->serial, *disk_list);
3666 if (idisk && is_failed(&idisk->disk))
3667 tbl_d->status |= FAILED_DISK;
3668 break;
3669 } else {
3670 struct intel_disk *idisk;
3671 struct imsm_disk *disk;
3672
3673 /* tbl_mpb is more up to date, but copy
3674 * over cross generational status before
3675 * returning
3676 */
3677 disk = __serial_to_disk(d->serial, mpb, NULL);
3678 if (disk && is_failed(disk))
3679 d->status |= FAILED_DISK;
3680
3681 idisk = disk_list_get(d->serial, *disk_list);
3682 if (idisk) {
3683 idisk->owner = i;
3684 if (disk && is_configured(disk))
3685 idisk->disk.status |= CONFIGURED_DISK;
3686 }
3687
3688 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3689 __func__, super->disks->major,
3690 super->disks->minor,
3691 table[i]->disks->major,
3692 table[i]->disks->minor);
3693
3694 return tbl_size;
3695 }
3696 }
3697 }
3698
3699 if (i >= tbl_size)
3700 table[tbl_size++] = super;
3701 else
3702 table[i] = super;
3703
3704 /* update/extend the merged list of imsm_disk records */
3705 for (j = 0; j < mpb->num_disks; j++) {
3706 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
3707 struct intel_disk *idisk;
3708
3709 idisk = disk_list_get(disk->serial, *disk_list);
3710 if (idisk) {
3711 idisk->disk.status |= disk->status;
3712 if (is_configured(&idisk->disk) ||
3713 is_failed(&idisk->disk))
3714 idisk->disk.status &= ~(SPARE_DISK);
3715 } else {
3716 idisk = calloc(1, sizeof(*idisk));
3717 if (!idisk)
3718 return -1;
3719 idisk->owner = IMSM_UNKNOWN_OWNER;
3720 idisk->disk = *disk;
3721 idisk->next = *disk_list;
3722 *disk_list = idisk;
3723 }
3724
3725 if (serialcmp(idisk->disk.serial, d->serial) == 0)
3726 idisk->owner = i;
3727 }
3728
3729 return tbl_size;
3730}
3731
3732static struct intel_super *
3733validate_members(struct intel_super *super, struct intel_disk *disk_list,
3734 const int owner)
3735{
3736 struct imsm_super *mpb = super->anchor;
3737 int ok_count = 0;
3738 int i;
3739
3740 for (i = 0; i < mpb->num_disks; i++) {
3741 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
3742 struct intel_disk *idisk;
3743
3744 idisk = disk_list_get(disk->serial, disk_list);
3745 if (idisk) {
3746 if (idisk->owner == owner ||
3747 idisk->owner == IMSM_UNKNOWN_OWNER)
3748 ok_count++;
3749 else
3750 dprintf("%s: '%.16s' owner %d != %d\n",
3751 __func__, disk->serial, idisk->owner,
3752 owner);
3753 } else {
3754 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3755 __func__, __le32_to_cpu(mpb->family_num), i,
3756 disk->serial);
3757 break;
3758 }
3759 }
3760
3761 if (ok_count == mpb->num_disks)
3762 return super;
3763 return NULL;
3764}
3765
3766static void show_conflicts(__u32 family_num, struct intel_super *super_list)
3767{
3768 struct intel_super *s;
3769
3770 for (s = super_list; s; s = s->next) {
3771 if (family_num != s->anchor->family_num)
3772 continue;
3773 fprintf(stderr, "Conflict, offlining family %#x on '%s'\n",
3774 __le32_to_cpu(family_num), s->disks->devname);
3775 }
3776}
3777
3778static struct intel_super *
3779imsm_thunderdome(struct intel_super **super_list, int len)
3780{
3781 struct intel_super *super_table[len];
3782 struct intel_disk *disk_list = NULL;
3783 struct intel_super *champion, *spare;
3784 struct intel_super *s, **del;
3785 int tbl_size = 0;
3786 int conflict;
3787 int i;
3788
3789 memset(super_table, 0, sizeof(super_table));
3790 for (s = *super_list; s; s = s->next)
3791 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
3792
3793 for (i = 0; i < tbl_size; i++) {
3794 struct imsm_disk *d;
3795 struct intel_disk *idisk;
3796 struct imsm_super *mpb = super_table[i]->anchor;
3797
3798 s = super_table[i];
3799 d = &s->disks->disk;
3800
3801 /* 'd' must appear in merged disk list for its
3802 * configuration to be valid
3803 */
3804 idisk = disk_list_get(d->serial, disk_list);
3805 if (idisk && idisk->owner == i)
3806 s = validate_members(s, disk_list, i);
3807 else
3808 s = NULL;
3809
3810 if (!s)
3811 dprintf("%s: marking family: %#x from %d:%d offline\n",
3812 __func__, mpb->family_num,
3813 super_table[i]->disks->major,
3814 super_table[i]->disks->minor);
3815 super_table[i] = s;
3816 }
3817
3818 /* This is where the mdadm implementation differs from the Windows
3819 * driver which has no strict concept of a container. We can only
3820 * assemble one family from a container, so when returning a prodigal
3821 * array member to this system the code will not be able to disambiguate
3822 * the container contents that should be assembled ("foreign" versus
3823 * "local"). It requires user intervention to set the orig_family_num
3824 * to a new value to establish a new container. The Windows driver in
3825 * this situation fixes up the volume name in place and manages the
3826 * foreign array as an independent entity.
3827 */
3828 s = NULL;
3829 spare = NULL;
3830 conflict = 0;
3831 for (i = 0; i < tbl_size; i++) {
3832 struct intel_super *tbl_ent = super_table[i];
3833 int is_spare = 0;
3834
3835 if (!tbl_ent)
3836 continue;
3837
3838 if (tbl_ent->anchor->num_raid_devs == 0) {
3839 spare = tbl_ent;
3840 is_spare = 1;
3841 }
3842
3843 if (s && !is_spare) {
3844 show_conflicts(tbl_ent->anchor->family_num, *super_list);
3845 conflict++;
3846 } else if (!s && !is_spare)
3847 s = tbl_ent;
3848 }
3849
3850 if (!s)
3851 s = spare;
3852 if (!s) {
3853 champion = NULL;
3854 goto out;
3855 }
3856 champion = s;
3857
3858 if (conflict)
3859 fprintf(stderr, "Chose family %#x on '%s', "
3860 "assemble conflicts to new container with '--update=uuid'\n",
3861 __le32_to_cpu(s->anchor->family_num), s->disks->devname);
3862
3863 /* collect all dl's onto 'champion', and update them to
3864 * champion's version of the status
3865 */
3866 for (s = *super_list; s; s = s->next) {
3867 struct imsm_super *mpb = champion->anchor;
3868 struct dl *dl = s->disks;
3869
3870 if (s == champion)
3871 continue;
3872
3873 for (i = 0; i < mpb->num_disks; i++) {
3874 struct imsm_disk *disk;
3875
3876 disk = __serial_to_disk(dl->serial, mpb, &dl->index);
3877 if (disk) {
3878 dl->disk = *disk;
3879 /* only set index on disks that are a member of
3880 * a populated contianer, i.e. one with
3881 * raid_devs
3882 */
3883 if (is_failed(&dl->disk))
3884 dl->index = -2;
3885 else if (is_spare(&dl->disk))
3886 dl->index = -1;
3887 break;
3888 }
3889 }
3890
3891 if (i >= mpb->num_disks) {
3892 struct intel_disk *idisk;
3893
3894 idisk = disk_list_get(dl->serial, disk_list);
3895 if (idisk && is_spare(&idisk->disk) &&
3896 !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
3897 dl->index = -1;
3898 else {
3899 dl->index = -2;
3900 continue;
3901 }
3902 }
3903
3904 dl->next = champion->disks;
3905 champion->disks = dl;
3906 s->disks = NULL;
3907 }
3908
3909 /* delete 'champion' from super_list */
3910 for (del = super_list; *del; ) {
3911 if (*del == champion) {
3912 *del = (*del)->next;
3913 break;
3914 } else
3915 del = &(*del)->next;
3916 }
3917 champion->next = NULL;
3918
3919 out:
3920 while (disk_list) {
3921 struct intel_disk *idisk = disk_list;
3922
3923 disk_list = disk_list->next;
3924 free(idisk);
3925 }
3926
3927 return champion;
3928}
3929
3930static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
3931 char *devname)
3932{
3933 struct mdinfo *sra;
3934 struct intel_super *super_list = NULL;
3935 struct intel_super *super = NULL;
3936 int devnum = fd2devnum(fd);
3937 struct mdinfo *sd;
3938 int retry;
3939 int err = 0;
3940 int i;
3941
3942 /* check if 'fd' an opened container */
3943 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
3944 if (!sra)
3945 return 1;
3946
3947 if (sra->array.major_version != -1 ||
3948 sra->array.minor_version != -2 ||
3949 strcmp(sra->text_version, "imsm") != 0) {
3950 err = 1;
3951 goto error;
3952 }
3953 /* load all mpbs */
3954 for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
3955 struct intel_super *s = alloc_super();
3956 char nm[32];
3957 int dfd;
3958 int rv;
3959
3960 err = 1;
3961 if (!s)
3962 goto error;
3963 s->next = super_list;
3964 super_list = s;
3965
3966 err = 2;
3967 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3968 dfd = dev_open(nm, O_RDWR);
3969 if (dfd < 0)
3970 goto error;
3971
3972 rv = find_intel_hba_capability(dfd, s, devname);
3973 /* no orom/efi or non-intel hba of the disk */
3974 if (rv != 0)
3975 goto error;
3976
3977 err = load_and_parse_mpb(dfd, s, NULL, 1);
3978
3979 /* retry the load if we might have raced against mdmon */
3980 if (err == 3 && mdmon_running(devnum))
3981 for (retry = 0; retry < 3; retry++) {
3982 usleep(3000);
3983 err = load_and_parse_mpb(dfd, s, NULL, 1);
3984 if (err != 3)
3985 break;
3986 }
3987 if (err)
3988 goto error;
3989 }
3990
3991 /* all mpbs enter, maybe one leaves */
3992 super = imsm_thunderdome(&super_list, i);
3993 if (!super) {
3994 err = 1;
3995 goto error;
3996 }
3997
3998 if (find_missing(super) != 0) {
3999 free_imsm(super);
4000 err = 2;
4001 goto error;
4002 }
4003
4004 /* load migration record */
4005 err = load_imsm_migr_rec(super, NULL);
4006 if (err) {
4007 err = 4;
4008 goto error;
4009 }
4010
4011 /* Check migration compatibility */
4012 if (check_mpb_migr_compatibility(super) != 0) {
4013 fprintf(stderr, Name ": Unsupported migration detected");
4014 if (devname)
4015 fprintf(stderr, " on %s\n", devname);
4016 else
4017 fprintf(stderr, " (IMSM).\n");
4018
4019 err = 5;
4020 goto error;
4021 }
4022
4023 err = 0;
4024
4025 error:
4026 while (super_list) {
4027 struct intel_super *s = super_list;
4028
4029 super_list = super_list->next;
4030 free_imsm(s);
4031 }
4032 sysfs_free(sra);
4033
4034 if (err)
4035 return err;
4036
4037 *sbp = super;
4038 st->container_dev = devnum;
4039 if (err == 0 && st->ss == NULL) {
4040 st->ss = &super_imsm;
4041 st->minor_version = 0;
4042 st->max_devs = IMSM_MAX_DEVICES;
4043 }
4044 return 0;
4045}
4046
4047static int load_container_imsm(struct supertype *st, int fd, char *devname)
4048{
4049 return load_super_imsm_all(st, fd, &st->sb, devname);
4050}
4051#endif
4052
4053static int load_super_imsm(struct supertype *st, int fd, char *devname)
4054{
4055 struct intel_super *super;
4056 int rv;
4057
4058 if (test_partition(fd))
4059 /* IMSM not allowed on partitions */
4060 return 1;
4061
4062 free_super_imsm(st);
4063
4064 super = alloc_super();
4065 if (!super) {
4066 fprintf(stderr,
4067 Name ": malloc of %zu failed.\n",
4068 sizeof(*super));
4069 return 1;
4070 }
4071 /* Load hba and capabilities if they exist.
4072 * But do not preclude loading metadata in case capabilities or hba are
4073 * non-compliant and ignore_hw_compat is set.
4074 */
4075 rv = find_intel_hba_capability(fd, super, devname);
4076 /* no orom/efi or non-intel hba of the disk */
4077 if ((rv != 0) && (st->ignore_hw_compat == 0)) {
4078 if (devname)
4079 fprintf(stderr,
4080 Name ": No OROM/EFI properties for %s\n", devname);
4081 free_imsm(super);
4082 return 2;
4083 }
4084 rv = load_and_parse_mpb(fd, super, devname, 0);
4085
4086 if (rv) {
4087 if (devname)
4088 fprintf(stderr,
4089 Name ": Failed to load all information "
4090 "sections on %s\n", devname);
4091 free_imsm(super);
4092 return rv;
4093 }
4094
4095 st->sb = super;
4096 if (st->ss == NULL) {
4097 st->ss = &super_imsm;
4098 st->minor_version = 0;
4099 st->max_devs = IMSM_MAX_DEVICES;
4100 }
4101
4102 /* load migration record */
4103 if (load_imsm_migr_rec(super, NULL) == 0) {
4104 /* Check for unsupported migration features */
4105 if (check_mpb_migr_compatibility(super) != 0) {
4106 fprintf(stderr,
4107 Name ": Unsupported migration detected");
4108 if (devname)
4109 fprintf(stderr, " on %s\n", devname);
4110 else
4111 fprintf(stderr, " (IMSM).\n");
4112 return 3;
4113 }
4114 }
4115
4116 return 0;
4117}
4118
4119static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
4120{
4121 if (info->level == 1)
4122 return 128;
4123 return info->chunk_size >> 9;
4124}
4125
4126static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
4127{
4128 __u32 num_stripes;
4129
4130 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
4131 num_stripes /= num_domains;
4132
4133 return num_stripes;
4134}
4135
4136static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
4137{
4138 if (info->level == 1)
4139 return info->size * 2;
4140 else
4141 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
4142}
4143
4144static void imsm_update_version_info(struct intel_super *super)
4145{
4146 /* update the version and attributes */
4147 struct imsm_super *mpb = super->anchor;
4148 char *version;
4149 struct imsm_dev *dev;
4150 struct imsm_map *map;
4151 int i;
4152
4153 for (i = 0; i < mpb->num_raid_devs; i++) {
4154 dev = get_imsm_dev(super, i);
4155 map = get_imsm_map(dev, 0);
4156 if (__le32_to_cpu(dev->size_high) > 0)
4157 mpb->attributes |= MPB_ATTRIB_2TB;
4158
4159 /* FIXME detect when an array spans a port multiplier */
4160 #if 0
4161 mpb->attributes |= MPB_ATTRIB_PM;
4162 #endif
4163
4164 if (mpb->num_raid_devs > 1 ||
4165 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
4166 version = MPB_VERSION_ATTRIBS;
4167 switch (get_imsm_raid_level(map)) {
4168 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
4169 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
4170 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
4171 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
4172 }
4173 } else {
4174 if (map->num_members >= 5)
4175 version = MPB_VERSION_5OR6_DISK_ARRAY;
4176 else if (dev->status == DEV_CLONE_N_GO)
4177 version = MPB_VERSION_CNG;
4178 else if (get_imsm_raid_level(map) == 5)
4179 version = MPB_VERSION_RAID5;
4180 else if (map->num_members >= 3)
4181 version = MPB_VERSION_3OR4_DISK_ARRAY;
4182 else if (get_imsm_raid_level(map) == 1)
4183 version = MPB_VERSION_RAID1;
4184 else
4185 version = MPB_VERSION_RAID0;
4186 }
4187 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
4188 }
4189}
4190
4191static int check_name(struct intel_super *super, char *name, int quiet)
4192{
4193 struct imsm_super *mpb = super->anchor;
4194 char *reason = NULL;
4195 int i;
4196
4197 if (strlen(name) > MAX_RAID_SERIAL_LEN)
4198 reason = "must be 16 characters or less";
4199
4200 for (i = 0; i < mpb->num_raid_devs; i++) {
4201 struct imsm_dev *dev = get_imsm_dev(super, i);
4202
4203 if (strncmp((char *) dev->volume, name, MAX_RAID_SERIAL_LEN) == 0) {
4204 reason = "already exists";
4205 break;
4206 }
4207 }
4208
4209 if (reason && !quiet)
4210 fprintf(stderr, Name ": imsm volume name %s\n", reason);
4211
4212 return !reason;
4213}
4214
4215static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
4216 unsigned long long size, char *name,
4217 char *homehost, int *uuid)
4218{
4219 /* We are creating a volume inside a pre-existing container.
4220 * so st->sb is already set.
4221 */
4222 struct intel_super *super = st->sb;
4223 struct imsm_super *mpb = super->anchor;
4224 struct intel_dev *dv;
4225 struct imsm_dev *dev;
4226 struct imsm_vol *vol;
4227 struct imsm_map *map;
4228 int idx = mpb->num_raid_devs;
4229 int i;
4230 unsigned long long array_blocks;
4231 size_t size_old, size_new;
4232 __u32 num_data_stripes;
4233
4234 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
4235 fprintf(stderr, Name": This imsm-container already has the "
4236 "maximum of %d volumes\n", super->orom->vpa);
4237 return 0;
4238 }
4239
4240 /* ensure the mpb is large enough for the new data */
4241 size_old = __le32_to_cpu(mpb->mpb_size);
4242 size_new = disks_to_mpb_size(info->nr_disks);
4243 if (size_new > size_old) {
4244 void *mpb_new;
4245 size_t size_round = ROUND_UP(size_new, 512);
4246
4247 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
4248 fprintf(stderr, Name": could not allocate new mpb\n");
4249 return 0;
4250 }
4251 if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) {
4252 fprintf(stderr, Name
4253 ": %s could not allocate migr_rec buffer\n",
4254 __func__);
4255 free(super->buf);
4256 free(super);
4257 free(mpb_new);
4258 return 0;
4259 }
4260 memcpy(mpb_new, mpb, size_old);
4261 free(mpb);
4262 mpb = mpb_new;
4263 super->anchor = mpb_new;
4264 mpb->mpb_size = __cpu_to_le32(size_new);
4265 memset(mpb_new + size_old, 0, size_round - size_old);
4266 }
4267 super->current_vol = idx;
4268
4269 /* handle 'failed_disks' by either:
4270 * a) create dummy disk entries in the table if this the first
4271 * volume in the array. We add them here as this is the only
4272 * opportunity to add them. add_to_super_imsm_volume()
4273 * handles the non-failed disks and continues incrementing
4274 * mpb->num_disks.
4275 * b) validate that 'failed_disks' matches the current number
4276 * of missing disks if the container is populated
4277 */
4278 if (super->current_vol == 0) {
4279 mpb->num_disks = 0;
4280 for (i = 0; i < info->failed_disks; i++) {
4281 struct imsm_disk *disk;
4282
4283 mpb->num_disks++;
4284 disk = __get_imsm_disk(mpb, i);
4285 disk->status = CONFIGURED_DISK | FAILED_DISK;
4286 disk->scsi_id = __cpu_to_le32(~(__u32)0);
4287 snprintf((char *) disk->serial, MAX_RAID_SERIAL_LEN,
4288 "missing:%d", i);
4289 }
4290 find_missing(super);
4291 } else {
4292 int missing = 0;
4293 struct dl *d;
4294
4295 for (d = super->missing; d; d = d->next)
4296 missing++;
4297 if (info->failed_disks > missing) {
4298 fprintf(stderr, Name": unable to add 'missing' disk to container\n");
4299 return 0;
4300 }
4301 }
4302
4303 if (!check_name(super, name, 0))
4304 return 0;
4305 dv = malloc(sizeof(*dv));
4306 if (!dv) {
4307 fprintf(stderr, Name ": failed to allocate device list entry\n");
4308 return 0;
4309 }
4310 dev = calloc(1, sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
4311 if (!dev) {
4312 free(dv);
4313 fprintf(stderr, Name": could not allocate raid device\n");
4314 return 0;
4315 }
4316
4317 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
4318 if (info->level == 1)
4319 array_blocks = info_to_blocks_per_member(info);
4320 else
4321 array_blocks = calc_array_size(info->level, info->raid_disks,
4322 info->layout, info->chunk_size,
4323 info->size*2);
4324 /* round array size down to closest MB */
4325 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
4326
4327 dev->size_low = __cpu_to_le32((__u32) array_blocks);
4328 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
4329 dev->status = (DEV_READ_COALESCING | DEV_WRITE_COALESCING);
4330 vol = &dev->vol;
4331 vol->migr_state = 0;
4332 set_migr_type(dev, MIGR_INIT);
4333 vol->dirty = !info->state;
4334 vol->curr_migr_unit = 0;
4335 map = get_imsm_map(dev, 0);
4336 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
4337 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
4338 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
4339 map->failed_disk_num = ~0;
4340 if (info->level > 0)
4341 map->map_state = IMSM_T_STATE_UNINITIALIZED;
4342 else
4343 map->map_state = info->failed_disks ? IMSM_T_STATE_FAILED :
4344 IMSM_T_STATE_NORMAL;
4345 map->ddf = 1;
4346
4347 if (info->level == 1 && info->raid_disks > 2) {
4348 free(dev);
4349 free(dv);
4350 fprintf(stderr, Name": imsm does not support more than 2 disks"
4351 "in a raid1 volume\n");
4352 return 0;
4353 }
4354
4355 map->raid_level = info->level;
4356 if (info->level == 10) {
4357 map->raid_level = 1;
4358 map->num_domains = info->raid_disks / 2;
4359 } else if (info->level == 1)
4360 map->num_domains = info->raid_disks;
4361 else
4362 map->num_domains = 1;
4363
4364 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
4365 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
4366
4367 map->num_members = info->raid_disks;
4368 for (i = 0; i < map->num_members; i++) {
4369 /* initialized in add_to_super */
4370 set_imsm_ord_tbl_ent(map, i, IMSM_ORD_REBUILD);
4371 }
4372 mpb->num_raid_devs++;
4373
4374 dv->dev = dev;
4375 dv->index = super->current_vol;
4376 dv->next = super->devlist;
4377 super->devlist = dv;
4378
4379 imsm_update_version_info(super);
4380
4381 return 1;
4382}
4383
4384static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
4385 unsigned long long size, char *name,
4386 char *homehost, int *uuid)
4387{
4388 /* This is primarily called by Create when creating a new array.
4389 * We will then get add_to_super called for each component, and then
4390 * write_init_super called to write it out to each device.
4391 * For IMSM, Create can create on fresh devices or on a pre-existing
4392 * array.
4393 * To create on a pre-existing array a different method will be called.
4394 * This one is just for fresh drives.
4395 */
4396 struct intel_super *super;
4397 struct imsm_super *mpb;
4398 size_t mpb_size;
4399 char *version;
4400
4401 if (st->sb)
4402 return init_super_imsm_volume(st, info, size, name, homehost, uuid);
4403
4404 if (info)
4405 mpb_size = disks_to_mpb_size(info->nr_disks);
4406 else
4407 mpb_size = 512;
4408
4409 super = alloc_super();
4410 if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) {
4411 free(super);
4412 super = NULL;
4413 }
4414 if (!super) {
4415 fprintf(stderr, Name
4416 ": %s could not allocate superblock\n", __func__);
4417 return 0;
4418 }
4419 if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) {
4420 fprintf(stderr, Name
4421 ": %s could not allocate migr_rec buffer\n", __func__);
4422 free(super->buf);
4423 free(super);
4424 return 0;
4425 }
4426 memset(super->buf, 0, mpb_size);
4427 mpb = super->buf;
4428 mpb->mpb_size = __cpu_to_le32(mpb_size);
4429 st->sb = super;
4430
4431 if (info == NULL) {
4432 /* zeroing superblock */
4433 return 0;
4434 }
4435
4436 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
4437
4438 version = (char *) mpb->sig;
4439 strcpy(version, MPB_SIGNATURE);
4440 version += strlen(MPB_SIGNATURE);
4441 strcpy(version, MPB_VERSION_RAID0);
4442
4443 return 1;
4444}
4445
4446#ifndef MDASSEMBLE
4447static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
4448 int fd, char *devname)
4449{
4450 struct intel_super *super = st->sb;
4451 struct imsm_super *mpb = super->anchor;
4452 struct imsm_disk *_disk;
4453 struct imsm_dev *dev;
4454 struct imsm_map *map;
4455 struct dl *dl, *df;
4456 int slot;
4457
4458 dev = get_imsm_dev(super, super->current_vol);
4459 map = get_imsm_map(dev, 0);
4460
4461 if (! (dk->state & (1<<MD_DISK_SYNC))) {
4462 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
4463 devname);
4464 return 1;
4465 }
4466
4467 if (fd == -1) {
4468 /* we're doing autolayout so grab the pre-marked (in
4469 * validate_geometry) raid_disk
4470 */
4471 for (dl = super->disks; dl; dl = dl->next)
4472 if (dl->raiddisk == dk->raid_disk)
4473 break;
4474 } else {
4475 for (dl = super->disks; dl ; dl = dl->next)
4476 if (dl->major == dk->major &&
4477 dl->minor == dk->minor)
4478 break;
4479 }
4480
4481 if (!dl) {
4482 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
4483 return 1;
4484 }
4485
4486 /* add a pristine spare to the metadata */
4487 if (dl->index < 0) {
4488 dl->index = super->anchor->num_disks;
4489 super->anchor->num_disks++;
4490 }
4491 /* Check the device has not already been added */
4492 slot = get_imsm_disk_slot(map, dl->index);
4493 if (slot >= 0 &&
4494 (get_imsm_ord_tbl_ent(dev, slot, -1) & IMSM_ORD_REBUILD) == 0) {
4495 fprintf(stderr, Name ": %s has been included in this array twice\n",
4496 devname);
4497 return 1;
4498 }
4499 set_imsm_ord_tbl_ent(map, dk->raid_disk, dl->index);
4500 dl->disk.status = CONFIGURED_DISK;
4501
4502 /* update size of 'missing' disks to be at least as large as the
4503 * largest acitve member (we only have dummy missing disks when
4504 * creating the first volume)
4505 */
4506 if (super->current_vol == 0) {
4507 for (df = super->missing; df; df = df->next) {
4508 if (dl->disk.total_blocks > df->disk.total_blocks)
4509 df->disk.total_blocks = dl->disk.total_blocks;
4510 _disk = __get_imsm_disk(mpb, df->index);
4511 *_disk = df->disk;
4512 }
4513 }
4514
4515 /* refresh unset/failed slots to point to valid 'missing' entries */
4516 for (df = super->missing; df; df = df->next)
4517 for (slot = 0; slot < mpb->num_disks; slot++) {
4518 __u32 ord = get_imsm_ord_tbl_ent(dev, slot, -1);
4519
4520 if ((ord & IMSM_ORD_REBUILD) == 0)
4521 continue;
4522 set_imsm_ord_tbl_ent(map, slot, df->index | IMSM_ORD_REBUILD);
4523 if (is_gen_migration(dev)) {
4524 struct imsm_map *map2 = get_imsm_map(dev, 1);
4525 if (slot < map2->num_members) {
4526 __u32 ord2 = get_imsm_ord_tbl_ent(dev,
4527 slot,
4528 1);
4529 if ((unsigned)df->index ==
4530 ord_to_idx(ord2))
4531 set_imsm_ord_tbl_ent(map2,
4532 slot,
4533 df->index |
4534 IMSM_ORD_REBUILD);
4535 }
4536 }
4537 dprintf("set slot:%d to missing disk:%d\n", slot, df->index);
4538 break;
4539 }
4540
4541 /* if we are creating the first raid device update the family number */
4542 if (super->current_vol == 0) {
4543 __u32 sum;
4544 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
4545
4546 _disk = __get_imsm_disk(mpb, dl->index);
4547 if (!_dev || !_disk) {
4548 fprintf(stderr, Name ": BUG mpb setup error\n");
4549 return 1;
4550 }
4551 *_dev = *dev;
4552 *_disk = dl->disk;
4553 sum = random32();
4554 sum += __gen_imsm_checksum(mpb);
4555 mpb->family_num = __cpu_to_le32(sum);
4556 mpb->orig_family_num = mpb->family_num;
4557 }
4558 super->current_disk = dl;
4559 return 0;
4560}
4561
4562/* mark_spare()
4563 * Function marks disk as spare and restores disk serial
4564 * in case it was previously marked as failed by takeover operation
4565 * reruns:
4566 * -1 : critical error
4567 * 0 : disk is marked as spare but serial is not set
4568 * 1 : success
4569 */
4570int mark_spare(struct dl *disk)
4571{
4572 __u8 serial[MAX_RAID_SERIAL_LEN];
4573 int ret_val = -1;
4574
4575 if (!disk)
4576 return ret_val;
4577
4578 ret_val = 0;
4579 if (!imsm_read_serial(disk->fd, NULL, serial)) {
4580 /* Restore disk serial number, because takeover marks disk
4581 * as failed and adds to serial ':0' before it becomes
4582 * a spare disk.
4583 */
4584 serialcpy(disk->serial, serial);
4585 serialcpy(disk->disk.serial, serial);
4586 ret_val = 1;
4587 }
4588 disk->disk.status = SPARE_DISK;
4589 disk->index = -1;
4590
4591 return ret_val;
4592}
4593
4594static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
4595 int fd, char *devname)
4596{
4597 struct intel_super *super = st->sb;
4598 struct dl *dd;
4599 unsigned long long size;
4600 __u32 id;
4601 int rv;
4602 struct stat stb;
4603
4604 /* If we are on an RAID enabled platform check that the disk is
4605 * attached to the raid controller.
4606 * We do not need to test disks attachment for container based additions,
4607 * they shall be already tested when container was created/assembled.
4608 */
4609 rv = find_intel_hba_capability(fd, super, devname);
4610 /* no orom/efi or non-intel hba of the disk */
4611 if (rv != 0) {
4612 dprintf("capability: %p fd: %d ret: %d\n",
4613 super->orom, fd, rv);
4614 return 1;
4615 }
4616
4617 if (super->current_vol >= 0)
4618 return add_to_super_imsm_volume(st, dk, fd, devname);
4619
4620 fstat(fd, &stb);
4621 dd = malloc(sizeof(*dd));
4622 if (!dd) {
4623 fprintf(stderr,
4624 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
4625 return 1;
4626 }
4627 memset(dd, 0, sizeof(*dd));
4628 dd->major = major(stb.st_rdev);
4629 dd->minor = minor(stb.st_rdev);
4630 dd->devname = devname ? strdup(devname) : NULL;
4631 dd->fd = fd;
4632 dd->e = NULL;
4633 dd->action = DISK_ADD;
4634 rv = imsm_read_serial(fd, devname, dd->serial);
4635 if (rv) {
4636 fprintf(stderr,
4637 Name ": failed to retrieve scsi serial, aborting\n");
4638 free(dd);
4639 abort();
4640 }
4641
4642 get_dev_size(fd, NULL, &size);
4643 size /= 512;
4644 serialcpy(dd->disk.serial, dd->serial);
4645 dd->disk.total_blocks = __cpu_to_le32(size);
4646 mark_spare(dd);
4647 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
4648 dd->disk.scsi_id = __cpu_to_le32(id);
4649 else
4650 dd->disk.scsi_id = __cpu_to_le32(0);
4651
4652 if (st->update_tail) {
4653 dd->next = super->disk_mgmt_list;
4654 super->disk_mgmt_list = dd;
4655 } else {
4656 dd->next = super->disks;
4657 super->disks = dd;
4658 super->updates_pending++;
4659 }
4660
4661 return 0;
4662}
4663
4664
4665static int remove_from_super_imsm(struct supertype *st, mdu_disk_info_t *dk)
4666{
4667 struct intel_super *super = st->sb;
4668 struct dl *dd;
4669
4670 /* remove from super works only in mdmon - for communication
4671 * manager - monitor. Check if communication memory buffer
4672 * is prepared.
4673 */
4674 if (!st->update_tail) {
4675 fprintf(stderr,
4676 Name ": %s shall be used in mdmon context only"
4677 "(line %d).\n", __func__, __LINE__);
4678 return 1;
4679 }
4680 dd = malloc(sizeof(*dd));
4681 if (!dd) {
4682 fprintf(stderr,
4683 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
4684 return 1;
4685 }
4686 memset(dd, 0, sizeof(*dd));
4687 dd->major = dk->major;
4688 dd->minor = dk->minor;
4689 dd->fd = -1;
4690 mark_spare(dd);
4691 dd->action = DISK_REMOVE;
4692
4693 dd->next = super->disk_mgmt_list;
4694 super->disk_mgmt_list = dd;
4695
4696
4697 return 0;
4698}
4699
4700static int store_imsm_mpb(int fd, struct imsm_super *mpb);
4701
4702static union {
4703 char buf[512];
4704 struct imsm_super anchor;
4705} spare_record __attribute__ ((aligned(512)));
4706
4707/* spare records have their own family number and do not have any defined raid
4708 * devices
4709 */
4710static int write_super_imsm_spares(struct intel_super *super, int doclose)
4711{
4712 struct imsm_super *mpb = super->anchor;
4713 struct imsm_super *spare = &spare_record.anchor;
4714 __u32 sum;
4715 struct dl *d;
4716
4717 spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)),
4718 spare->generation_num = __cpu_to_le32(1UL),
4719 spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
4720 spare->num_disks = 1,
4721 spare->num_raid_devs = 0,
4722 spare->cache_size = mpb->cache_size,
4723 spare->pwr_cycle_count = __cpu_to_le32(1),
4724
4725 snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
4726 MPB_SIGNATURE MPB_VERSION_RAID0);
4727
4728 for (d = super->disks; d; d = d->next) {
4729 if (d->index != -1)
4730 continue;
4731
4732 spare->disk[0] = d->disk;
4733 sum = __gen_imsm_checksum(spare);
4734 spare->family_num = __cpu_to_le32(sum);
4735 spare->orig_family_num = 0;
4736 sum = __gen_imsm_checksum(spare);
4737 spare->check_sum = __cpu_to_le32(sum);
4738
4739 if (store_imsm_mpb(d->fd, spare)) {
4740 fprintf(stderr, "%s: failed for device %d:%d %s\n",
4741 __func__, d->major, d->minor, strerror(errno));
4742 return 1;
4743 }
4744 if (doclose) {
4745 close(d->fd);
4746 d->fd = -1;
4747 }
4748 }
4749
4750 return 0;
4751}
4752
4753static int write_super_imsm(struct supertype *st, int doclose)
4754{
4755 struct intel_super *super = st->sb;
4756 struct imsm_super *mpb = super->anchor;
4757 struct dl *d;
4758 __u32 generation;
4759 __u32 sum;
4760 int spares = 0;
4761 int i;
4762 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
4763 int num_disks = 0;
4764 int clear_migration_record = 1;
4765
4766 /* 'generation' is incremented everytime the metadata is written */
4767 generation = __le32_to_cpu(mpb->generation_num);
4768 generation++;
4769 mpb->generation_num = __cpu_to_le32(generation);
4770
4771 /* fix up cases where previous mdadm releases failed to set
4772 * orig_family_num
4773 */
4774 if (mpb->orig_family_num == 0)
4775 mpb->orig_family_num = mpb->family_num;
4776
4777 for (d = super->disks; d; d = d->next) {
4778 if (d->index == -1)
4779 spares++;
4780 else {
4781 mpb->disk[d->index] = d->disk;
4782 num_disks++;
4783 }
4784 }
4785 for (d = super->missing; d; d = d->next) {
4786 mpb->disk[d->index] = d->disk;
4787 num_disks++;
4788 }
4789 mpb->num_disks = num_disks;
4790 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
4791
4792 for (i = 0; i < mpb->num_raid_devs; i++) {
4793 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
4794 struct imsm_dev *dev2 = get_imsm_dev(super, i);
4795 if (dev && dev2) {
4796 imsm_copy_dev(dev, dev2);
4797 mpb_size += sizeof_imsm_dev(dev, 0);
4798 }
4799 if (is_gen_migration(dev2))
4800 clear_migration_record = 0;
4801 }
4802 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
4803 mpb->mpb_size = __cpu_to_le32(mpb_size);
4804
4805 /* recalculate checksum */
4806 sum = __gen_imsm_checksum(mpb);
4807 mpb->check_sum = __cpu_to_le32(sum);
4808
4809 if (clear_migration_record)
4810 memset(super->migr_rec_buf, 0, 512);
4811
4812 /* write the mpb for disks that compose raid devices */
4813 for (d = super->disks; d ; d = d->next) {
4814 if (d->index < 0 || is_failed(&d->disk))
4815 continue;
4816 if (store_imsm_mpb(d->fd, mpb))
4817 fprintf(stderr, "%s: failed for device %d:%d %s\n",
4818 __func__, d->major, d->minor, strerror(errno));
4819 if (clear_migration_record) {
4820 unsigned long long dsize;
4821
4822 get_dev_size(d->fd, NULL, &dsize);
4823 if (lseek64(d->fd, dsize - 512, SEEK_SET) >= 0) {
4824 if (write(d->fd, super->migr_rec_buf, 512) != 512)
4825 perror("Write migr_rec failed");
4826 }
4827 }
4828 if (doclose) {
4829 close(d->fd);
4830 d->fd = -1;
4831 }
4832 }
4833
4834 if (spares)
4835 return write_super_imsm_spares(super, doclose);
4836
4837 return 0;
4838}
4839
4840
4841static int create_array(struct supertype *st, int dev_idx)
4842{
4843 size_t len;
4844 struct imsm_update_create_array *u;
4845 struct intel_super *super = st->sb;
4846 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
4847 struct imsm_map *map = get_imsm_map(dev, 0);
4848 struct disk_info *inf;
4849 struct imsm_disk *disk;
4850 int i;
4851
4852 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
4853 sizeof(*inf) * map->num_members;
4854 u = malloc(len);
4855 if (!u) {
4856 fprintf(stderr, "%s: failed to allocate update buffer\n",
4857 __func__);
4858 return 1;
4859 }
4860
4861 u->type = update_create_array;
4862 u->dev_idx = dev_idx;
4863 imsm_copy_dev(&u->dev, dev);
4864 inf = get_disk_info(u);
4865 for (i = 0; i < map->num_members; i++) {
4866 int idx = get_imsm_disk_idx(dev, i, -1);
4867
4868 disk = get_imsm_disk(super, idx);
4869 serialcpy(inf[i].serial, disk->serial);
4870 }
4871 append_metadata_update(st, u, len);
4872
4873 return 0;
4874}
4875
4876static int mgmt_disk(struct supertype *st)
4877{
4878 struct intel_super *super = st->sb;
4879 size_t len;
4880 struct imsm_update_add_remove_disk *u;
4881
4882 if (!super->disk_mgmt_list)
4883 return 0;
4884
4885 len = sizeof(*u);
4886 u = malloc(len);
4887 if (!u) {
4888 fprintf(stderr, "%s: failed to allocate update buffer\n",
4889 __func__);
4890 return 1;
4891 }
4892
4893 u->type = update_add_remove_disk;
4894 append_metadata_update(st, u, len);
4895
4896 return 0;
4897}
4898
4899static int write_init_super_imsm(struct supertype *st)
4900{
4901 struct intel_super *super = st->sb;
4902 int current_vol = super->current_vol;
4903
4904 /* we are done with current_vol reset it to point st at the container */
4905 super->current_vol = -1;
4906
4907 if (st->update_tail) {
4908 /* queue the recently created array / added disk
4909 * as a metadata update */
4910 int rv;
4911
4912 /* determine if we are creating a volume or adding a disk */
4913 if (current_vol < 0) {
4914 /* in the mgmt (add/remove) disk case we are running
4915 * in mdmon context, so don't close fd's
4916 */
4917 return mgmt_disk(st);
4918 } else
4919 rv = create_array(st, current_vol);
4920
4921 return rv;
4922 } else {
4923 struct dl *d;
4924 for (d = super->disks; d; d = d->next)
4925 Kill(d->devname, NULL, 0, 1, 1);
4926 return write_super_imsm(st, 1);
4927 }
4928}
4929#endif
4930
4931static int store_super_imsm(struct supertype *st, int fd)
4932{
4933 struct intel_super *super = st->sb;
4934 struct imsm_super *mpb = super ? super->anchor : NULL;
4935
4936 if (!mpb)
4937 return 1;
4938
4939#ifndef MDASSEMBLE
4940 return store_imsm_mpb(fd, mpb);
4941#else
4942 return 1;
4943#endif
4944}
4945
4946static int imsm_bbm_log_size(struct imsm_super *mpb)
4947{
4948 return __le32_to_cpu(mpb->bbm_log_size);
4949}
4950
4951#ifndef MDASSEMBLE
4952static int validate_geometry_imsm_container(struct supertype *st, int level,
4953 int layout, int raiddisks, int chunk,
4954 unsigned long long size, char *dev,
4955 unsigned long long *freesize,
4956 int verbose)
4957{
4958 int fd;
4959 unsigned long long ldsize;
4960 struct intel_super *super=NULL;
4961 int rv = 0;
4962
4963 if (level != LEVEL_CONTAINER)
4964 return 0;
4965 if (!dev)
4966 return 1;
4967
4968 fd = open(dev, O_RDONLY|O_EXCL, 0);
4969 if (fd < 0) {
4970 if (verbose)
4971 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
4972 dev, strerror(errno));
4973 return 0;
4974 }
4975 if (!get_dev_size(fd, dev, &ldsize)) {
4976 close(fd);
4977 return 0;
4978 }
4979
4980 /* capabilities retrieve could be possible
4981 * note that there is no fd for the disks in array.
4982 */
4983 super = alloc_super();
4984 if (!super) {
4985 fprintf(stderr,
4986 Name ": malloc of %zu failed.\n",
4987 sizeof(*super));
4988 close(fd);
4989 return 0;
4990 }
4991
4992 rv = find_intel_hba_capability(fd, super, verbose ? dev : NULL);
4993 if (rv != 0) {
4994#if DEBUG
4995 char str[256];
4996 fd2devname(fd, str);
4997 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4998 fd, str, super->orom, rv, raiddisks);
4999#endif
5000 /* no orom/efi or non-intel hba of the disk */
5001 close(fd);
5002 free_imsm(super);
5003 return 0;
5004 }
5005 close(fd);
5006 if (super->orom && raiddisks > super->orom->tds) {
5007 if (verbose)
5008 fprintf(stderr, Name ": %d exceeds maximum number of"
5009 " platform supported disks: %d\n",
5010 raiddisks, super->orom->tds);
5011
5012 free_imsm(super);
5013 return 0;
5014 }
5015
5016 *freesize = avail_size_imsm(st, ldsize >> 9);
5017 free_imsm(super);
5018
5019 return 1;
5020}
5021
5022static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
5023{
5024 const unsigned long long base_start = e[*idx].start;
5025 unsigned long long end = base_start + e[*idx].size;
5026 int i;
5027
5028 if (base_start == end)
5029 return 0;
5030
5031 *idx = *idx + 1;
5032 for (i = *idx; i < num_extents; i++) {
5033 /* extend overlapping extents */
5034 if (e[i].start >= base_start &&
5035 e[i].start <= end) {
5036 if (e[i].size == 0)
5037 return 0;
5038 if (e[i].start + e[i].size > end)
5039 end = e[i].start + e[i].size;
5040 } else if (e[i].start > end) {
5041 *idx = i;
5042 break;
5043 }
5044 }
5045
5046 return end - base_start;
5047}
5048
5049static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
5050{
5051 /* build a composite disk with all known extents and generate a new
5052 * 'maxsize' given the "all disks in an array must share a common start
5053 * offset" constraint
5054 */
5055 struct extent *e = calloc(sum_extents, sizeof(*e));
5056 struct dl *dl;
5057 int i, j;
5058 int start_extent;
5059 unsigned long long pos;
5060 unsigned long long start = 0;
5061 unsigned long long maxsize;
5062 unsigned long reserve;
5063
5064 if (!e)
5065 return 0;
5066
5067 /* coalesce and sort all extents. also, check to see if we need to
5068 * reserve space between member arrays
5069 */
5070 j = 0;
5071 for (dl = super->disks; dl; dl = dl->next) {
5072 if (!dl->e)
5073 continue;
5074 for (i = 0; i < dl->extent_cnt; i++)
5075 e[j++] = dl->e[i];
5076 }
5077 qsort(e, sum_extents, sizeof(*e), cmp_extent);
5078
5079 /* merge extents */
5080 i = 0;
5081 j = 0;
5082 while (i < sum_extents) {
5083 e[j].start = e[i].start;
5084 e[j].size = find_size(e, &i, sum_extents);
5085 j++;
5086 if (e[j-1].size == 0)
5087 break;
5088 }
5089
5090 pos = 0;
5091 maxsize = 0;
5092 start_extent = 0;
5093 i = 0;
5094 do {
5095 unsigned long long esize;
5096
5097 esize = e[i].start - pos;
5098 if (esize >= maxsize) {
5099 maxsize = esize;
5100 start = pos;
5101 start_extent = i;
5102 }
5103 pos = e[i].start + e[i].size;
5104 i++;
5105 } while (e[i-1].size);
5106 free(e);
5107
5108 if (maxsize == 0)
5109 return 0;
5110
5111 /* FIXME assumes volume at offset 0 is the first volume in a
5112 * container
5113 */
5114 if (start_extent > 0)
5115 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
5116 else
5117 reserve = 0;
5118
5119 if (maxsize < reserve)
5120 return 0;
5121
5122 super->create_offset = ~((__u32) 0);
5123 if (start + reserve > super->create_offset)
5124 return 0; /* start overflows create_offset */
5125 super->create_offset = start + reserve;
5126
5127 return maxsize - reserve;
5128}
5129
5130static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
5131{
5132 if (level < 0 || level == 6 || level == 4)
5133 return 0;
5134
5135 /* if we have an orom prevent invalid raid levels */
5136 if (orom)
5137 switch (level) {
5138 case 0: return imsm_orom_has_raid0(orom);
5139 case 1:
5140 if (raiddisks > 2)
5141 return imsm_orom_has_raid1e(orom);
5142 return imsm_orom_has_raid1(orom) && raiddisks == 2;
5143 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
5144 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
5145 }
5146 else
5147 return 1; /* not on an Intel RAID platform so anything goes */
5148
5149 return 0;
5150}
5151
5152static int imsm_default_chunk(const struct imsm_orom *orom)
5153{
5154 /* up to 512 if the plaform supports it, otherwise the platform max.
5155 * 128 if no platform detected
5156 */
5157 int fs = max(7, orom ? fls(orom->sss) : 0);
5158
5159 return min(512, (1 << fs));
5160}
5161
5162#define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5163static int
5164validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
5165 int raiddisks, int *chunk, int verbose)
5166{
5167 /* check/set platform and metadata limits/defaults */
5168 if (super->orom && raiddisks > super->orom->dpa) {
5169 pr_vrb(": platform supports a maximum of %d disks per array\n",
5170 super->orom->dpa);
5171 return 0;
5172 }
5173
5174 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5175 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
5176 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5177 level, raiddisks, raiddisks > 1 ? "s" : "");
5178 return 0;
5179 }
5180
5181 if (chunk && (*chunk == 0 || *chunk == UnSet))
5182 *chunk = imsm_default_chunk(super->orom);
5183
5184 if (super->orom && chunk && !imsm_orom_has_chunk(super->orom, *chunk)) {
5185 pr_vrb(": platform does not support a chunk size of: "
5186 "%d\n", *chunk);
5187 return 0;
5188 }
5189
5190 if (layout != imsm_level_to_layout(level)) {
5191 if (level == 5)
5192 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5193 else if (level == 10)
5194 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5195 else
5196 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5197 layout, level);
5198 return 0;
5199 }
5200 return 1;
5201}
5202
5203/* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5204 * FIX ME add ahci details
5205 */
5206static int validate_geometry_imsm_volume(struct supertype *st, int level,
5207 int layout, int raiddisks, int *chunk,
5208 unsigned long long size, char *dev,
5209 unsigned long long *freesize,
5210 int verbose)
5211{
5212 struct stat stb;
5213 struct intel_super *super = st->sb;
5214 struct imsm_super *mpb;
5215 struct dl *dl;
5216 unsigned long long pos = 0;
5217 unsigned long long maxsize;
5218 struct extent *e;
5219 int i;
5220
5221 /* We must have the container info already read in. */
5222 if (!super)
5223 return 0;
5224
5225 mpb = super->anchor;
5226
5227 if (mpb->num_raid_devs > 0 && mpb->num_disks != raiddisks) {
5228 fprintf(stderr, Name ": the option-rom requires all "
5229 "member disks to be a member of all volumes.\n");
5230 return 0;
5231 }
5232
5233 if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose)) {
5234 fprintf(stderr, Name ": RAID gemetry validation failed. "
5235 "Cannot proceed with the action(s).\n");
5236 return 0;
5237 }
5238 if (!dev) {
5239 /* General test: make sure there is space for
5240 * 'raiddisks' device extents of size 'size' at a given
5241 * offset
5242 */
5243 unsigned long long minsize = size;
5244 unsigned long long start_offset = MaxSector;
5245 int dcnt = 0;
5246 if (minsize == 0)
5247 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
5248 for (dl = super->disks; dl ; dl = dl->next) {
5249 int found = 0;
5250
5251 pos = 0;
5252 i = 0;
5253 e = get_extents(super, dl);
5254 if (!e) continue;
5255 do {
5256 unsigned long long esize;
5257 esize = e[i].start - pos;
5258 if (esize >= minsize)
5259 found = 1;
5260 if (found && start_offset == MaxSector) {
5261 start_offset = pos;
5262 break;
5263 } else if (found && pos != start_offset) {
5264 found = 0;
5265 break;
5266 }
5267 pos = e[i].start + e[i].size;
5268 i++;
5269 } while (e[i-1].size);
5270 if (found)
5271 dcnt++;
5272 free(e);
5273 }
5274 if (dcnt < raiddisks) {
5275 if (verbose)
5276 fprintf(stderr, Name ": imsm: Not enough "
5277 "devices with space for this array "
5278 "(%d < %d)\n",
5279 dcnt, raiddisks);
5280 return 0;
5281 }
5282 return 1;
5283 }
5284
5285 /* This device must be a member of the set */
5286 if (stat(dev, &stb) < 0)
5287 return 0;
5288 if ((S_IFMT & stb.st_mode) != S_IFBLK)
5289 return 0;
5290 for (dl = super->disks ; dl ; dl = dl->next) {
5291 if (dl->major == (int)major(stb.st_rdev) &&
5292 dl->minor == (int)minor(stb.st_rdev))
5293 break;
5294 }
5295 if (!dl) {
5296 if (verbose)
5297 fprintf(stderr, Name ": %s is not in the "
5298 "same imsm set\n", dev);
5299 return 0;
5300 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
5301 /* If a volume is present then the current creation attempt
5302 * cannot incorporate new spares because the orom may not
5303 * understand this configuration (all member disks must be
5304 * members of each array in the container).
5305 */
5306 fprintf(stderr, Name ": %s is a spare and a volume"
5307 " is already defined for this container\n", dev);
5308 fprintf(stderr, Name ": The option-rom requires all member"
5309 " disks to be a member of all volumes\n");
5310 return 0;
5311 }
5312
5313 /* retrieve the largest free space block */
5314 e = get_extents(super, dl);
5315 maxsize = 0;
5316 i = 0;
5317 if (e) {
5318 do {
5319 unsigned long long esize;
5320
5321 esize = e[i].start - pos;
5322 if (esize >= maxsize)
5323 maxsize = esize;
5324 pos = e[i].start + e[i].size;
5325 i++;
5326 } while (e[i-1].size);
5327 dl->e = e;
5328 dl->extent_cnt = i;
5329 } else {
5330 if (verbose)
5331 fprintf(stderr, Name ": unable to determine free space for: %s\n",
5332 dev);
5333 return 0;
5334 }
5335 if (maxsize < size) {
5336 if (verbose)
5337 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
5338 dev, maxsize, size);
5339 return 0;
5340 }
5341
5342 /* count total number of extents for merge */
5343 i = 0;
5344 for (dl = super->disks; dl; dl = dl->next)
5345 if (dl->e)
5346 i += dl->extent_cnt;
5347
5348 maxsize = merge_extents(super, i);
5349
5350 if (!check_env("IMSM_NO_PLATFORM") &&
5351 mpb->num_raid_devs > 0 && size && size != maxsize) {
5352 fprintf(stderr, Name ": attempting to create a second "
5353 "volume with size less then remaining space. "
5354 "Aborting...\n");
5355 return 0;
5356 }
5357
5358 if (maxsize < size || maxsize == 0) {
5359 if (verbose)
5360 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
5361 maxsize, size);
5362 return 0;
5363 }
5364
5365 *freesize = maxsize;
5366
5367 return 1;
5368}
5369
5370static int reserve_space(struct supertype *st, int raiddisks,
5371 unsigned long long size, int chunk,
5372 unsigned long long *freesize)
5373{
5374 struct intel_super *super = st->sb;
5375 struct imsm_super *mpb = super->anchor;
5376 struct dl *dl;
5377 int i;
5378 int extent_cnt;
5379 struct extent *e;
5380 unsigned long long maxsize;
5381 unsigned long long minsize;
5382 int cnt;
5383 int used;
5384
5385 /* find the largest common start free region of the possible disks */
5386 used = 0;
5387 extent_cnt = 0;
5388 cnt = 0;
5389 for (dl = super->disks; dl; dl = dl->next) {
5390 dl->raiddisk = -1;
5391
5392 if (dl->index >= 0)
5393 used++;
5394
5395 /* don't activate new spares if we are orom constrained
5396 * and there is already a volume active in the container
5397 */
5398 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
5399 continue;
5400
5401 e = get_extents(super, dl);
5402 if (!e)
5403 continue;
5404 for (i = 1; e[i-1].size; i++)
5405 ;
5406 dl->e = e;
5407 dl->extent_cnt = i;
5408 extent_cnt += i;
5409 cnt++;
5410 }
5411
5412 maxsize = merge_extents(super, extent_cnt);
5413 minsize = size;
5414 if (size == 0)
5415 /* chunk is in K */
5416 minsize = chunk * 2;
5417
5418 if (cnt < raiddisks ||
5419 (super->orom && used && used != raiddisks) ||
5420 maxsize < minsize ||
5421 maxsize == 0) {
5422 fprintf(stderr, Name ": not enough devices with space to create array.\n");
5423 return 0; /* No enough free spaces large enough */
5424 }
5425
5426 if (size == 0) {
5427 size = maxsize;
5428 if (chunk) {
5429 size /= 2 * chunk;
5430 size *= 2 * chunk;
5431 }
5432 }
5433
5434 cnt = 0;
5435 for (dl = super->disks; dl; dl = dl->next)
5436 if (dl->e)
5437 dl->raiddisk = cnt++;
5438
5439 *freesize = size;
5440
5441 return 1;
5442}
5443
5444static int validate_geometry_imsm(struct supertype *st, int level, int layout,
5445 int raiddisks, int *chunk, unsigned long long size,
5446 char *dev, unsigned long long *freesize,
5447 int verbose)
5448{
5449 int fd, cfd;
5450 struct mdinfo *sra;
5451 int is_member = 0;
5452
5453 /* load capability
5454 * if given unused devices create a container
5455 * if given given devices in a container create a member volume
5456 */
5457 if (level == LEVEL_CONTAINER) {
5458 /* Must be a fresh device to add to a container */
5459 return validate_geometry_imsm_container(st, level, layout,
5460 raiddisks,
5461 chunk?*chunk:0, size,
5462 dev, freesize,
5463 verbose);
5464 }
5465
5466 if (!dev) {
5467 if (st->sb) {
5468 if (!validate_geometry_imsm_orom(st->sb, level, layout,
5469 raiddisks, chunk,
5470 verbose))
5471 return 0;
5472 /* we are being asked to automatically layout a
5473 * new volume based on the current contents of
5474 * the container. If the the parameters can be
5475 * satisfied reserve_space will record the disks,
5476 * start offset, and size of the volume to be
5477 * created. add_to_super and getinfo_super
5478 * detect when autolayout is in progress.
5479 */
5480 if (freesize)
5481 return reserve_space(st, raiddisks, size,
5482 chunk?*chunk:0, freesize);
5483 }
5484 return 1;
5485 }
5486 if (st->sb) {
5487 /* creating in a given container */
5488 return validate_geometry_imsm_volume(st, level, layout,
5489 raiddisks, chunk, size,
5490 dev, freesize, verbose);
5491 }
5492
5493 /* This device needs to be a device in an 'imsm' container */
5494 fd = open(dev, O_RDONLY|O_EXCL, 0);
5495 if (fd >= 0) {
5496 if (verbose)
5497 fprintf(stderr,
5498 Name ": Cannot create this array on device %s\n",
5499 dev);
5500 close(fd);
5501 return 0;
5502 }
5503 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
5504 if (verbose)
5505 fprintf(stderr, Name ": Cannot open %s: %s\n",
5506 dev, strerror(errno));
5507 return 0;
5508 }
5509 /* Well, it is in use by someone, maybe an 'imsm' container. */
5510 cfd = open_container(fd);
5511 close(fd);
5512 if (cfd < 0) {
5513 if (verbose)
5514 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
5515 dev);
5516 return 0;
5517 }
5518 sra = sysfs_read(cfd, 0, GET_VERSION);
5519 if (sra && sra->array.major_version == -1 &&
5520 strcmp(sra->text_version, "imsm") == 0)
5521 is_member = 1;
5522 sysfs_free(sra);
5523 if (is_member) {
5524 /* This is a member of a imsm container. Load the container
5525 * and try to create a volume
5526 */
5527 struct intel_super *super;
5528
5529 if (load_super_imsm_all(st, cfd, (void **) &super, NULL) == 0) {
5530 st->sb = super;
5531 st->container_dev = fd2devnum(cfd);
5532 close(cfd);
5533 return validate_geometry_imsm_volume(st, level, layout,
5534 raiddisks, chunk,
5535 size, dev,
5536 freesize, 1)
5537 ? 1 : -1;
5538 }
5539 }
5540
5541 if (verbose)
5542 fprintf(stderr, Name ": failed container membership check\n");
5543
5544 close(cfd);
5545 return 0;
5546}
5547
5548static void default_geometry_imsm(struct supertype *st, int *level, int *layout, int *chunk)
5549{
5550 struct intel_super *super = st->sb;
5551
5552 if (level && *level == UnSet)
5553 *level = LEVEL_CONTAINER;
5554
5555 if (level && layout && *layout == UnSet)
5556 *layout = imsm_level_to_layout(*level);
5557
5558 if (chunk && (*chunk == UnSet || *chunk == 0))
5559 *chunk = imsm_default_chunk(super->orom);
5560}
5561
5562static void handle_missing(struct intel_super *super, struct imsm_dev *dev);
5563
5564static int kill_subarray_imsm(struct supertype *st)
5565{
5566 /* remove the subarray currently referenced by ->current_vol */
5567 __u8 i;
5568 struct intel_dev **dp;
5569 struct intel_super *super = st->sb;
5570 __u8 current_vol = super->current_vol;
5571 struct imsm_super *mpb = super->anchor;
5572
5573 if (super->current_vol < 0)
5574 return 2;
5575 super->current_vol = -1; /* invalidate subarray cursor */
5576
5577 /* block deletions that would change the uuid of active subarrays
5578 *
5579 * FIXME when immutable ids are available, but note that we'll
5580 * also need to fixup the invalidated/active subarray indexes in
5581 * mdstat
5582 */
5583 for (i = 0; i < mpb->num_raid_devs; i++) {
5584 char subarray[4];
5585
5586 if (i < current_vol)
5587 continue;
5588 sprintf(subarray, "%u", i);
5589 if (is_subarray_active(subarray, st->devname)) {
5590 fprintf(stderr,
5591 Name ": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5592 current_vol, i);
5593
5594 return 2;
5595 }
5596 }
5597
5598 if (st->update_tail) {
5599 struct imsm_update_kill_array *u = malloc(sizeof(*u));
5600
5601 if (!u)
5602 return 2;
5603 u->type = update_kill_array;
5604 u->dev_idx = current_vol;
5605 append_metadata_update(st, u, sizeof(*u));
5606
5607 return 0;
5608 }
5609
5610 for (dp = &super->devlist; *dp;)
5611 if ((*dp)->index == current_vol) {
5612 *dp = (*dp)->next;
5613 } else {
5614 handle_missing(super, (*dp)->dev);
5615 if ((*dp)->index > current_vol)
5616 (*dp)->index--;
5617 dp = &(*dp)->next;
5618 }
5619
5620 /* no more raid devices, all active components are now spares,
5621 * but of course failed are still failed
5622 */
5623 if (--mpb->num_raid_devs == 0) {
5624 struct dl *d;
5625
5626 for (d = super->disks; d; d = d->next)
5627 if (d->index > -2)
5628 mark_spare(d);
5629 }
5630
5631 super->updates_pending++;
5632
5633 return 0;
5634}
5635
5636static int update_subarray_imsm(struct supertype *st, char *subarray,
5637 char *update, struct mddev_ident *ident)
5638{
5639 /* update the subarray currently referenced by ->current_vol */
5640 struct intel_super *super = st->sb;
5641 struct imsm_super *mpb = super->anchor;
5642
5643 if (strcmp(update, "name") == 0) {
5644 char *name = ident->name;
5645 char *ep;
5646 int vol;
5647
5648 if (is_subarray_active(subarray, st->devname)) {
5649 fprintf(stderr,
5650 Name ": Unable to update name of active subarray\n");
5651 return 2;
5652 }
5653
5654 if (!check_name(super, name, 0))
5655 return 2;
5656
5657 vol = strtoul(subarray, &ep, 10);
5658 if (*ep != '\0' || vol >= super->anchor->num_raid_devs)
5659 return 2;
5660
5661 if (st->update_tail) {
5662 struct imsm_update_rename_array *u = malloc(sizeof(*u));
5663
5664 if (!u)
5665 return 2;
5666 u->type = update_rename_array;
5667 u->dev_idx = vol;
5668 snprintf((char *) u->name, MAX_RAID_SERIAL_LEN, "%s", name);
5669 append_metadata_update(st, u, sizeof(*u));
5670 } else {
5671 struct imsm_dev *dev;
5672 int i;
5673
5674 dev = get_imsm_dev(super, vol);
5675 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
5676 for (i = 0; i < mpb->num_raid_devs; i++) {
5677 dev = get_imsm_dev(super, i);
5678 handle_missing(super, dev);
5679 }
5680 super->updates_pending++;
5681 }
5682 } else
5683 return 2;
5684
5685 return 0;
5686}
5687#endif /* MDASSEMBLE */
5688
5689static int is_gen_migration(struct imsm_dev *dev)
5690{
5691 if (dev == NULL)
5692 return 0;
5693
5694 if (!dev->vol.migr_state)
5695 return 0;
5696
5697 if (migr_type(dev) == MIGR_GEN_MIGR)
5698 return 1;
5699
5700 return 0;
5701}
5702
5703static int is_rebuilding(struct imsm_dev *dev)
5704{
5705 struct imsm_map *migr_map;
5706
5707 if (!dev->vol.migr_state)
5708 return 0;
5709
5710 if (migr_type(dev) != MIGR_REBUILD)
5711 return 0;
5712
5713 migr_map = get_imsm_map(dev, 1);
5714
5715 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
5716 return 1;
5717 else
5718 return 0;
5719}
5720
5721static void update_recovery_start(struct intel_super *super,
5722 struct imsm_dev *dev,
5723 struct mdinfo *array)
5724{
5725 struct mdinfo *rebuild = NULL;
5726 struct mdinfo *d;
5727 __u32 units;
5728
5729 if (!is_rebuilding(dev))
5730 return;
5731
5732 /* Find the rebuild target, but punt on the dual rebuild case */
5733 for (d = array->devs; d; d = d->next)
5734 if (d->recovery_start == 0) {
5735 if (rebuild)
5736 return;
5737 rebuild = d;
5738 }
5739
5740 if (!rebuild) {
5741 /* (?) none of the disks are marked with
5742 * IMSM_ORD_REBUILD, so assume they are missing and the
5743 * disk_ord_tbl was not correctly updated
5744 */
5745 dprintf("%s: failed to locate out-of-sync disk\n", __func__);
5746 return;
5747 }
5748
5749 units = __le32_to_cpu(dev->vol.curr_migr_unit);
5750 rebuild->recovery_start = units * blocks_per_migr_unit(super, dev);
5751}
5752
5753#ifndef MDASSEMBLE
5754static int recover_backup_imsm(struct supertype *st, struct mdinfo *info);
5755#endif
5756
5757static struct mdinfo *container_content_imsm(struct supertype *st, char *subarray)
5758{
5759 /* Given a container loaded by load_super_imsm_all,
5760 * extract information about all the arrays into
5761 * an mdinfo tree.
5762 * If 'subarray' is given, just extract info about that array.
5763 *
5764 * For each imsm_dev create an mdinfo, fill it in,
5765 * then look for matching devices in super->disks
5766 * and create appropriate device mdinfo.
5767 */
5768 struct intel_super *super = st->sb;
5769 struct imsm_super *mpb = super->anchor;
5770 struct mdinfo *rest = NULL;
5771 unsigned int i;
5772 int sb_errors = 0;
5773 struct dl *d;
5774 int spare_disks = 0;
5775
5776 /* do not assemble arrays when not all attributes are supported */
5777 if (imsm_check_attributes(mpb->attributes) == 0) {
5778 sb_errors = 1;
5779 fprintf(stderr, Name ": Unsupported attributes in IMSM metadata."
5780 "Arrays activation is blocked.\n");
5781 }
5782
5783 /* check for bad blocks */
5784 if (imsm_bbm_log_size(super->anchor)) {
5785 fprintf(stderr, Name ": BBM log found in IMSM metadata."
5786 "Arrays activation is blocked.\n");
5787 sb_errors = 1;
5788 }
5789
5790
5791 /* count spare devices, not used in maps
5792 */
5793 for (d = super->disks; d; d = d->next)
5794 if (d->index == -1)
5795 spare_disks++;
5796
5797 for (i = 0; i < mpb->num_raid_devs; i++) {
5798 struct imsm_dev *dev;
5799 struct imsm_map *map;
5800 struct imsm_map *map2;
5801 struct mdinfo *this;
5802 int slot, chunk;
5803 char *ep;
5804
5805 if (subarray &&
5806 (i != strtoul(subarray, &ep, 10) || *ep != '\0'))
5807 continue;
5808
5809 dev = get_imsm_dev(super, i);
5810 map = get_imsm_map(dev, 0);
5811 map2 = get_imsm_map(dev, 1);
5812
5813 /* do not publish arrays that are in the middle of an
5814 * unsupported migration
5815 */
5816 if (dev->vol.migr_state &&
5817 (migr_type(dev) == MIGR_STATE_CHANGE)) {
5818 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
5819 " unsupported migration in progress\n",
5820 dev->volume);
5821 continue;
5822 }
5823 /* do not publish arrays that are not support by controller's
5824 * OROM/EFI
5825 */
5826
5827 chunk = __le16_to_cpu(map->blocks_per_strip) >> 1;
5828 this = malloc(sizeof(*this));
5829 if (!this) {
5830 fprintf(stderr, Name ": failed to allocate %zu bytes\n",
5831 sizeof(*this));
5832 break;
5833 }
5834
5835 super->current_vol = i;
5836 getinfo_super_imsm_volume(st, this, NULL);
5837 this->next = rest;
5838#ifndef MDASSEMBLE
5839 /* mdadm does not support all metadata features- set the bit in all arrays state */
5840 if (!validate_geometry_imsm_orom(super,
5841 get_imsm_raid_level(map), /* RAID level */
5842 imsm_level_to_layout(get_imsm_raid_level(map)),
5843 map->num_members, /* raid disks */
5844 &chunk,
5845 1 /* verbose */)) {
5846 fprintf(stderr, Name ": IMSM RAID geometry validation"
5847 " failed. Array %s activation is blocked.\n",
5848 dev->volume);
5849 this->array.state |=
5850 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE) |
5851 (1<<MD_SB_BLOCK_VOLUME);
5852 }
5853#endif
5854
5855 /* if array has bad blocks, set suitable bit in all arrays state */
5856 if (sb_errors)
5857 this->array.state |=
5858 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE) |
5859 (1<<MD_SB_BLOCK_VOLUME);
5860
5861 for (slot = 0 ; slot < map->num_members; slot++) {
5862 unsigned long long recovery_start;
5863 struct mdinfo *info_d;
5864 struct dl *d;
5865 int idx;
5866 int skip;
5867 __u32 ord;
5868
5869 skip = 0;
5870 idx = get_imsm_disk_idx(dev, slot, 0);
5871 ord = get_imsm_ord_tbl_ent(dev, slot, -1);
5872 for (d = super->disks; d ; d = d->next)
5873 if (d->index == idx)
5874 break;
5875
5876 recovery_start = MaxSector;
5877 if (d == NULL)
5878 skip = 1;
5879 if (d && is_failed(&d->disk))
5880 skip = 1;
5881 if (ord & IMSM_ORD_REBUILD)
5882 recovery_start = 0;
5883
5884 /*
5885 * if we skip some disks the array will be assmebled degraded;
5886 * reset resync start to avoid a dirty-degraded
5887 * situation when performing the intial sync
5888 *
5889 * FIXME handle dirty degraded
5890 */
5891 if ((skip || recovery_start == 0) && !dev->vol.dirty)
5892 this->resync_start = MaxSector;
5893 if (skip)
5894 continue;
5895
5896 info_d = calloc(1, sizeof(*info_d));
5897 if (!info_d) {
5898 fprintf(stderr, Name ": failed to allocate disk"
5899 " for volume %.16s\n", dev->volume);
5900 info_d = this->devs;
5901 while (info_d) {
5902 struct mdinfo *d = info_d->next;
5903
5904 free(info_d);
5905 info_d = d;
5906 }
5907 free(this);
5908 this = rest;
5909 break;
5910 }
5911 info_d->next = this->devs;
5912 this->devs = info_d;
5913
5914 info_d->disk.number = d->index;
5915 info_d->disk.major = d->major;
5916 info_d->disk.minor = d->minor;
5917 info_d->disk.raid_disk = slot;
5918 info_d->recovery_start = recovery_start;
5919 if (map2) {
5920 if (slot < map2->num_members)
5921 info_d->disk.state = (1 << MD_DISK_ACTIVE);
5922 else
5923 this->array.spare_disks++;
5924 } else {
5925 if (slot < map->num_members)
5926 info_d->disk.state = (1 << MD_DISK_ACTIVE);
5927 else
5928 this->array.spare_disks++;
5929 }
5930 if (info_d->recovery_start == MaxSector)
5931 this->array.working_disks++;
5932
5933 info_d->events = __le32_to_cpu(mpb->generation_num);
5934 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
5935 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
5936 }
5937 /* now that the disk list is up-to-date fixup recovery_start */
5938 update_recovery_start(super, dev, this);
5939 this->array.spare_disks += spare_disks;
5940
5941#ifndef MDASSEMBLE
5942 /* check for reshape */
5943 if (this->reshape_active == 1)
5944 recover_backup_imsm(st, this);
5945#endif
5946 rest = this;
5947 }
5948
5949 return rest;
5950}
5951
5952
5953static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev,
5954 int failed, int look_in_map)
5955{
5956 struct imsm_map *map;
5957
5958 map = get_imsm_map(dev, look_in_map);
5959
5960 if (!failed)
5961 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
5962 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
5963
5964 switch (get_imsm_raid_level(map)) {
5965 case 0:
5966 return IMSM_T_STATE_FAILED;
5967 break;
5968 case 1:
5969 if (failed < map->num_members)
5970 return IMSM_T_STATE_DEGRADED;
5971 else
5972 return IMSM_T_STATE_FAILED;
5973 break;
5974 case 10:
5975 {
5976 /**
5977 * check to see if any mirrors have failed, otherwise we
5978 * are degraded. Even numbered slots are mirrored on
5979 * slot+1
5980 */
5981 int i;
5982 /* gcc -Os complains that this is unused */
5983 int insync = insync;
5984
5985 for (i = 0; i < map->num_members; i++) {
5986 __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1);
5987 int idx = ord_to_idx(ord);
5988 struct imsm_disk *disk;
5989
5990 /* reset the potential in-sync count on even-numbered
5991 * slots. num_copies is always 2 for imsm raid10
5992 */
5993 if ((i & 1) == 0)
5994 insync = 2;
5995
5996 disk = get_imsm_disk(super, idx);
5997 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
5998 insync--;
5999
6000 /* no in-sync disks left in this mirror the
6001 * array has failed
6002 */
6003 if (insync == 0)
6004 return IMSM_T_STATE_FAILED;
6005 }
6006
6007 return IMSM_T_STATE_DEGRADED;
6008 }
6009 case 5:
6010 if (failed < 2)
6011 return IMSM_T_STATE_DEGRADED;
6012 else
6013 return IMSM_T_STATE_FAILED;
6014 break;
6015 default:
6016 break;
6017 }
6018
6019 return map->map_state;
6020}
6021
6022static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev,
6023 int look_in_map)
6024{
6025 int i;
6026 int failed = 0;
6027 struct imsm_disk *disk;
6028 struct imsm_map *map = get_imsm_map(dev, MAP_0);
6029 struct imsm_map *prev = get_imsm_map(dev, MAP_1);
6030 struct imsm_map *map_for_loop;
6031 __u32 ord;
6032 int idx;
6033 int idx_1;
6034
6035 /* at the beginning of migration we set IMSM_ORD_REBUILD on
6036 * disks that are being rebuilt. New failures are recorded to
6037 * map[0]. So we look through all the disks we started with and
6038 * see if any failures are still present, or if any new ones
6039 * have arrived
6040 */
6041 map_for_loop = map;
6042 if (prev && (map->num_members < prev->num_members))
6043 map_for_loop = prev;
6044
6045 for (i = 0; i < map_for_loop->num_members; i++) {
6046 idx_1 = -255;
6047 if (prev &&
6048 (look_in_map & MAP_1) && (i < prev->num_members)) {
6049 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
6050 idx_1 = ord_to_idx(ord);
6051
6052 disk = get_imsm_disk(super, idx_1);
6053 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
6054 failed++;
6055 }
6056 if ((look_in_map & MAP_0) && (i < map->num_members)) {
6057 ord = __le32_to_cpu(map->disk_ord_tbl[i]);
6058 idx = ord_to_idx(ord);
6059
6060 if (idx != idx_1) {
6061 disk = get_imsm_disk(super, idx);
6062 if (!disk || is_failed(disk) ||
6063 ord & IMSM_ORD_REBUILD)
6064 failed++;
6065 }
6066 }
6067 }
6068
6069 return failed;
6070}
6071
6072#ifndef MDASSEMBLE
6073static int imsm_open_new(struct supertype *c, struct active_array *a,
6074 char *inst)
6075{
6076 struct intel_super *super = c->sb;
6077 struct imsm_super *mpb = super->anchor;
6078
6079 if (atoi(inst) >= mpb->num_raid_devs) {
6080 fprintf(stderr, "%s: subarry index %d, out of range\n",
6081 __func__, atoi(inst));
6082 return -ENODEV;
6083 }
6084
6085 dprintf("imsm: open_new %s\n", inst);
6086 a->info.container_member = atoi(inst);
6087 return 0;
6088}
6089
6090static int is_resyncing(struct imsm_dev *dev)
6091{
6092 struct imsm_map *migr_map;
6093
6094 if (!dev->vol.migr_state)
6095 return 0;
6096
6097 if (migr_type(dev) == MIGR_INIT ||
6098 migr_type(dev) == MIGR_REPAIR)
6099 return 1;
6100
6101 if (migr_type(dev) == MIGR_GEN_MIGR)
6102 return 0;
6103
6104 migr_map = get_imsm_map(dev, 1);
6105
6106 if ((migr_map->map_state == IMSM_T_STATE_NORMAL) &&
6107 (dev->vol.migr_type != MIGR_GEN_MIGR))
6108 return 1;
6109 else
6110 return 0;
6111}
6112
6113/* return true if we recorded new information */
6114static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
6115{
6116 __u32 ord;
6117 int slot;
6118 struct imsm_map *map;
6119 char buf[MAX_RAID_SERIAL_LEN+3];
6120 unsigned int len, shift = 0;
6121
6122 /* new failures are always set in map[0] */
6123 map = get_imsm_map(dev, 0);
6124
6125 slot = get_imsm_disk_slot(map, idx);
6126 if (slot < 0)
6127 return 0;
6128
6129 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
6130 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
6131 return 0;
6132
6133 memcpy(buf, disk->serial, MAX_RAID_SERIAL_LEN);
6134 buf[MAX_RAID_SERIAL_LEN] = '\000';
6135 strcat(buf, ":0");
6136 if ((len = strlen(buf)) >= MAX_RAID_SERIAL_LEN)
6137 shift = len - MAX_RAID_SERIAL_LEN + 1;
6138 strncpy((char *)disk->serial, &buf[shift], MAX_RAID_SERIAL_LEN);
6139
6140 disk->status |= FAILED_DISK;
6141 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
6142 if (is_gen_migration(dev)) {
6143 struct imsm_map *map2 = get_imsm_map(dev, 1);
6144 if (slot < map2->num_members)
6145 set_imsm_ord_tbl_ent(map2, slot,
6146 idx | IMSM_ORD_REBUILD);
6147 }
6148 if (map->failed_disk_num == 0xff)
6149 map->failed_disk_num = slot;
6150 return 1;
6151}
6152
6153static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
6154{
6155 mark_failure(dev, disk, idx);
6156
6157 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
6158 return;
6159
6160 disk->scsi_id = __cpu_to_le32(~(__u32)0);
6161 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
6162}
6163
6164static void handle_missing(struct intel_super *super, struct imsm_dev *dev)
6165{
6166 struct dl *dl;
6167
6168 if (!super->missing)
6169 return;
6170
6171 dprintf("imsm: mark missing\n");
6172 for (dl = super->missing; dl; dl = dl->next)
6173 mark_missing(dev, &dl->disk, dl->index);
6174 super->updates_pending++;
6175}
6176
6177static unsigned long long imsm_set_array_size(struct imsm_dev *dev)
6178{
6179 int used_disks = imsm_num_data_members(dev, 0);
6180 unsigned long long array_blocks;
6181 struct imsm_map *map;
6182
6183 if (used_disks == 0) {
6184 /* when problems occures
6185 * return current array_blocks value
6186 */
6187 array_blocks = __le32_to_cpu(dev->size_high);
6188 array_blocks = array_blocks << 32;
6189 array_blocks += __le32_to_cpu(dev->size_low);
6190
6191 return array_blocks;
6192 }
6193
6194 /* set array size in metadata
6195 */
6196 map = get_imsm_map(dev, 0);
6197 array_blocks = map->blocks_per_member * used_disks;
6198
6199 /* round array size down to closest MB
6200 */
6201 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
6202 dev->size_low = __cpu_to_le32((__u32)array_blocks);
6203 dev->size_high = __cpu_to_le32((__u32)(array_blocks >> 32));
6204
6205 return array_blocks;
6206}
6207
6208static void imsm_set_disk(struct active_array *a, int n, int state);
6209
6210static void imsm_progress_container_reshape(struct intel_super *super)
6211{
6212 /* if no device has a migr_state, but some device has a
6213 * different number of members than the previous device, start
6214 * changing the number of devices in this device to match
6215 * previous.
6216 */
6217 struct imsm_super *mpb = super->anchor;
6218 int prev_disks = -1;
6219 int i;
6220 int copy_map_size;
6221
6222 for (i = 0; i < mpb->num_raid_devs; i++) {
6223 struct imsm_dev *dev = get_imsm_dev(super, i);
6224 struct imsm_map *map = get_imsm_map(dev, 0);
6225 struct imsm_map *map2;
6226 int prev_num_members;
6227
6228 if (dev->vol.migr_state)
6229 return;
6230
6231 if (prev_disks == -1)
6232 prev_disks = map->num_members;
6233 if (prev_disks == map->num_members)
6234 continue;
6235
6236 /* OK, this array needs to enter reshape mode.
6237 * i.e it needs a migr_state
6238 */
6239
6240 copy_map_size = sizeof_imsm_map(map);
6241 prev_num_members = map->num_members;
6242 map->num_members = prev_disks;
6243 dev->vol.migr_state = 1;
6244 dev->vol.curr_migr_unit = 0;
6245 set_migr_type(dev, MIGR_GEN_MIGR);
6246 for (i = prev_num_members;
6247 i < map->num_members; i++)
6248 set_imsm_ord_tbl_ent(map, i, i);
6249 map2 = get_imsm_map(dev, 1);
6250 /* Copy the current map */
6251 memcpy(map2, map, copy_map_size);
6252 map2->num_members = prev_num_members;
6253
6254 imsm_set_array_size(dev);
6255 super->updates_pending++;
6256 }
6257}
6258
6259/* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6260 * states are handled in imsm_set_disk() with one exception, when a
6261 * resync is stopped due to a new failure this routine will set the
6262 * 'degraded' state for the array.
6263 */
6264static int imsm_set_array_state(struct active_array *a, int consistent)
6265{
6266 int inst = a->info.container_member;
6267 struct intel_super *super = a->container->sb;
6268 struct imsm_dev *dev = get_imsm_dev(super, inst);
6269 struct imsm_map *map = get_imsm_map(dev, 0);
6270 int failed = imsm_count_failed(super, dev, MAP_0);
6271 __u8 map_state = imsm_check_degraded(super, dev, failed, MAP_0);
6272 __u32 blocks_per_unit;
6273
6274 if (dev->vol.migr_state &&
6275 dev->vol.migr_type == MIGR_GEN_MIGR) {
6276 /* array state change is blocked due to reshape action
6277 * We might need to
6278 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6279 * - finish the reshape (if last_checkpoint is big and action != reshape)
6280 * - update curr_migr_unit
6281 */
6282 if (a->curr_action == reshape) {
6283 /* still reshaping, maybe update curr_migr_unit */
6284 goto mark_checkpoint;
6285 } else {
6286 if (a->last_checkpoint == 0 && a->prev_action == reshape) {
6287 /* for some reason we aborted the reshape.
6288 *
6289 * disable automatic metadata rollback
6290 * user action is required to recover process
6291 */
6292 if (0) {
6293 struct imsm_map *map2 = get_imsm_map(dev, 1);
6294 dev->vol.migr_state = 0;
6295 set_migr_type(dev, 0);
6296 dev->vol.curr_migr_unit = 0;
6297 memcpy(map, map2, sizeof_imsm_map(map2));
6298 super->updates_pending++;
6299 }
6300 }
6301 if (a->last_checkpoint >= a->info.component_size) {
6302 unsigned long long array_blocks;
6303 int used_disks;
6304 struct mdinfo *mdi;
6305
6306 used_disks = imsm_num_data_members(dev, 0);
6307 if (used_disks > 0) {
6308 array_blocks =
6309 map->blocks_per_member *
6310 used_disks;
6311 /* round array size down to closest MB
6312 */
6313 array_blocks = (array_blocks
6314 >> SECT_PER_MB_SHIFT)
6315 << SECT_PER_MB_SHIFT;
6316 a->info.custom_array_size = array_blocks;
6317 /* encourage manager to update array
6318 * size
6319 */
6320
6321 a->check_reshape = 1;
6322 }
6323 /* finalize online capacity expansion/reshape */
6324 for (mdi = a->info.devs; mdi; mdi = mdi->next)
6325 imsm_set_disk(a,
6326 mdi->disk.raid_disk,
6327 mdi->curr_state);
6328
6329 imsm_progress_container_reshape(super);
6330 }
6331 }
6332 }
6333
6334 /* before we activate this array handle any missing disks */
6335 if (consistent == 2)
6336 handle_missing(super, dev);
6337
6338 if (consistent == 2 &&
6339 (!is_resync_complete(&a->info) ||
6340 map_state != IMSM_T_STATE_NORMAL ||
6341 dev->vol.migr_state))
6342 consistent = 0;
6343
6344 if (is_resync_complete(&a->info)) {
6345 /* complete intialization / resync,
6346 * recovery and interrupted recovery is completed in
6347 * ->set_disk
6348 */
6349 if (is_resyncing(dev)) {
6350 dprintf("imsm: mark resync done\n");
6351 end_migration(dev, super, map_state);
6352 super->updates_pending++;
6353 a->last_checkpoint = 0;
6354 }
6355 } else if ((!is_resyncing(dev) && !failed) &&
6356 (imsm_reshape_blocks_arrays_changes(super) == 0)) {
6357 /* mark the start of the init process if nothing is failed */
6358 dprintf("imsm: mark resync start\n");
6359 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
6360 migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_INIT);
6361 else
6362 migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
6363 super->updates_pending++;
6364 }
6365
6366mark_checkpoint:
6367 /* skip checkpointing for general migration,
6368 * it is controlled in mdadm
6369 */
6370 if (is_gen_migration(dev))
6371 goto skip_mark_checkpoint;
6372
6373 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6374 blocks_per_unit = blocks_per_migr_unit(super, dev);
6375 if (blocks_per_unit) {
6376 __u32 units32;
6377 __u64 units;
6378
6379 units = a->last_checkpoint / blocks_per_unit;
6380 units32 = units;
6381
6382 /* check that we did not overflow 32-bits, and that
6383 * curr_migr_unit needs updating
6384 */
6385 if (units32 == units &&
6386 units32 != 0 &&
6387 __le32_to_cpu(dev->vol.curr_migr_unit) != units32) {
6388 dprintf("imsm: mark checkpoint (%u)\n", units32);
6389 dev->vol.curr_migr_unit = __cpu_to_le32(units32);
6390 super->updates_pending++;
6391 }
6392 }
6393
6394skip_mark_checkpoint:
6395 /* mark dirty / clean */
6396 if (dev->vol.dirty != !consistent) {
6397 dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty");
6398 if (consistent)
6399 dev->vol.dirty = 0;
6400 else
6401 dev->vol.dirty = 1;
6402 super->updates_pending++;
6403 }
6404
6405 return consistent;
6406}
6407
6408static void imsm_set_disk(struct active_array *a, int n, int state)
6409{
6410 int inst = a->info.container_member;
6411 struct intel_super *super = a->container->sb;
6412 struct imsm_dev *dev = get_imsm_dev(super, inst);
6413 struct imsm_map *map = get_imsm_map(dev, 0);
6414 struct imsm_disk *disk;
6415 int failed;
6416 __u32 ord;
6417 __u8 map_state;
6418
6419 if (n > map->num_members)
6420 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
6421 n, map->num_members - 1);
6422
6423 if (n < 0)
6424 return;
6425
6426 dprintf("imsm: set_disk %d:%x\n", n, state);
6427
6428 ord = get_imsm_ord_tbl_ent(dev, n, 0);
6429 disk = get_imsm_disk(super, ord_to_idx(ord));
6430
6431 /* check for new failures */
6432 if (state & DS_FAULTY) {
6433 if (mark_failure(dev, disk, ord_to_idx(ord)))
6434 super->updates_pending++;
6435 }
6436
6437 /* check if in_sync */
6438 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
6439 struct imsm_map *migr_map = get_imsm_map(dev, 1);
6440
6441 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
6442 super->updates_pending++;
6443 }
6444
6445 failed = imsm_count_failed(super, dev, MAP_0);
6446 map_state = imsm_check_degraded(super, dev, failed, MAP_0);
6447
6448 /* check if recovery complete, newly degraded, or failed */
6449 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
6450 end_migration(dev, super, map_state);
6451 map = get_imsm_map(dev, 0);
6452 map->failed_disk_num = ~0;
6453 super->updates_pending++;
6454 a->last_checkpoint = 0;
6455 } else if (map_state == IMSM_T_STATE_DEGRADED &&
6456 map->map_state != map_state &&
6457 !dev->vol.migr_state) {
6458 dprintf("imsm: mark degraded\n");
6459 map->map_state = map_state;
6460 super->updates_pending++;
6461 a->last_checkpoint = 0;
6462 } else if (map_state == IMSM_T_STATE_FAILED &&
6463 map->map_state != map_state) {
6464 dprintf("imsm: mark failed\n");
6465 end_migration(dev, super, map_state);
6466 super->updates_pending++;
6467 a->last_checkpoint = 0;
6468 } else if (is_gen_migration(dev)) {
6469 dprintf("imsm: Detected General Migration in state: ");
6470
6471 switch (map_state) {
6472 case IMSM_T_STATE_NORMAL:
6473 dprintf("normal\n");
6474 if (a->last_checkpoint >= a->info.component_size)
6475 end_migration(dev, super, map_state);
6476 map = get_imsm_map(dev, 0);
6477 map->failed_disk_num = ~0;
6478 break;
6479 case IMSM_T_STATE_DEGRADED:
6480 dprintf("degraded\n");
6481 if (a->last_checkpoint >= a->info.component_size)
6482 end_migration(dev, super, map_state);
6483 else
6484 manage_second_map(super, dev);
6485 break;
6486 default:
6487 dprintf("failed\n");
6488 }
6489 map->map_state = map_state;
6490 super->updates_pending++;
6491 }
6492}
6493
6494static int store_imsm_mpb(int fd, struct imsm_super *mpb)
6495{
6496 void *buf = mpb;
6497 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
6498 unsigned long long dsize;
6499 unsigned long long sectors;
6500
6501 get_dev_size(fd, NULL, &dsize);
6502
6503 if (mpb_size > 512) {
6504 /* -1 to account for anchor */
6505 sectors = mpb_sectors(mpb) - 1;
6506
6507 /* write the extended mpb to the sectors preceeding the anchor */
6508 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
6509 return 1;
6510
6511 if ((unsigned long long)write(fd, buf + 512, 512 * sectors)
6512 != 512 * sectors)
6513 return 1;
6514 }
6515
6516 /* first block is stored on second to last sector of the disk */
6517 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
6518 return 1;
6519
6520 if (write(fd, buf, 512) != 512)
6521 return 1;
6522
6523 return 0;
6524}
6525
6526static void imsm_sync_metadata(struct supertype *container)
6527{
6528 struct intel_super *super = container->sb;
6529
6530 dprintf("sync metadata: %d\n", super->updates_pending);
6531 if (!super->updates_pending)
6532 return;
6533
6534 write_super_imsm(container, 0);
6535
6536 super->updates_pending = 0;
6537}
6538
6539static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
6540{
6541 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
6542 int i = get_imsm_disk_idx(dev, idx, -1);
6543 struct dl *dl;
6544
6545 for (dl = super->disks; dl; dl = dl->next)
6546 if (dl->index == i)
6547 break;
6548
6549 if (dl && is_failed(&dl->disk))
6550 dl = NULL;
6551
6552 if (dl)
6553 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
6554
6555 return dl;
6556}
6557
6558static struct dl *imsm_add_spare(struct intel_super *super, int slot,
6559 struct active_array *a, int activate_new,
6560 struct mdinfo *additional_test_list)
6561{
6562 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
6563 int idx = get_imsm_disk_idx(dev, slot, -1);
6564 struct imsm_super *mpb = super->anchor;
6565 struct imsm_map *map;
6566 unsigned long long pos;
6567 struct mdinfo *d;
6568 struct extent *ex;
6569 int i, j;
6570 int found;
6571 __u32 array_start = 0;
6572 __u32 array_end = 0;
6573 struct dl *dl;
6574 struct mdinfo *test_list;
6575
6576 for (dl = super->disks; dl; dl = dl->next) {
6577 /* If in this array, skip */
6578 for (d = a->info.devs ; d ; d = d->next)
6579 if (d->state_fd >= 0 &&
6580 d->disk.major == dl->major &&
6581 d->disk.minor == dl->minor) {
6582 dprintf("%x:%x already in array\n",
6583 dl->major, dl->minor);
6584 break;
6585 }
6586 if (d)
6587 continue;
6588 test_list = additional_test_list;
6589 while (test_list) {
6590 if (test_list->disk.major == dl->major &&
6591 test_list->disk.minor == dl->minor) {
6592 dprintf("%x:%x already in additional test list\n",
6593 dl->major, dl->minor);
6594 break;
6595 }
6596 test_list = test_list->next;
6597 }
6598 if (test_list)
6599 continue;
6600
6601 /* skip in use or failed drives */
6602 if (is_failed(&dl->disk) || idx == dl->index ||
6603 dl->index == -2) {
6604 dprintf("%x:%x status (failed: %d index: %d)\n",
6605 dl->major, dl->minor, is_failed(&dl->disk), idx);
6606 continue;
6607 }
6608
6609 /* skip pure spares when we are looking for partially
6610 * assimilated drives
6611 */
6612 if (dl->index == -1 && !activate_new)
6613 continue;
6614
6615 /* Does this unused device have the requisite free space?
6616 * It needs to be able to cover all member volumes
6617 */
6618 ex = get_extents(super, dl);
6619 if (!ex) {
6620 dprintf("cannot get extents\n");
6621 continue;
6622 }
6623 for (i = 0; i < mpb->num_raid_devs; i++) {
6624 dev = get_imsm_dev(super, i);
6625 map = get_imsm_map(dev, 0);
6626
6627 /* check if this disk is already a member of
6628 * this array
6629 */
6630 if (get_imsm_disk_slot(map, dl->index) >= 0)
6631 continue;
6632
6633 found = 0;
6634 j = 0;
6635 pos = 0;
6636 array_start = __le32_to_cpu(map->pba_of_lba0);
6637 array_end = array_start +
6638 __le32_to_cpu(map->blocks_per_member) - 1;
6639
6640 do {
6641 /* check that we can start at pba_of_lba0 with
6642 * blocks_per_member of space
6643 */
6644 if (array_start >= pos && array_end < ex[j].start) {
6645 found = 1;
6646 break;
6647 }
6648 pos = ex[j].start + ex[j].size;
6649 j++;
6650 } while (ex[j-1].size);
6651
6652 if (!found)
6653 break;
6654 }
6655
6656 free(ex);
6657 if (i < mpb->num_raid_devs) {
6658 dprintf("%x:%x does not have %u to %u available\n",
6659 dl->major, dl->minor, array_start, array_end);
6660 /* No room */
6661 continue;
6662 }
6663 return dl;
6664 }
6665
6666 return dl;
6667}
6668
6669
6670static int imsm_rebuild_allowed(struct supertype *cont, int dev_idx, int failed)
6671{
6672 struct imsm_dev *dev2;
6673 struct imsm_map *map;
6674 struct dl *idisk;
6675 int slot;
6676 int idx;
6677 __u8 state;
6678
6679 dev2 = get_imsm_dev(cont->sb, dev_idx);
6680 if (dev2) {
6681 state = imsm_check_degraded(cont->sb, dev2, failed,
6682 MAP_0);
6683 if (state == IMSM_T_STATE_FAILED) {
6684 map = get_imsm_map(dev2, 0);
6685 if (!map)
6686 return 1;
6687 for (slot = 0; slot < map->num_members; slot++) {
6688 /*
6689 * Check if failed disks are deleted from intel
6690 * disk list or are marked to be deleted
6691 */
6692 idx = get_imsm_disk_idx(dev2, slot, -1);
6693 idisk = get_imsm_dl_disk(cont->sb, idx);
6694 /*
6695 * Do not rebuild the array if failed disks
6696 * from failed sub-array are not removed from
6697 * container.
6698 */
6699 if (idisk &&
6700 is_failed(&idisk->disk) &&
6701 (idisk->action != DISK_REMOVE))
6702 return 0;
6703 }
6704 }
6705 }
6706 return 1;
6707}
6708
6709static struct mdinfo *imsm_activate_spare(struct active_array *a,
6710 struct metadata_update **updates)
6711{
6712 /**
6713 * Find a device with unused free space and use it to replace a
6714 * failed/vacant region in an array. We replace failed regions one a
6715 * array at a time. The result is that a new spare disk will be added
6716 * to the first failed array and after the monitor has finished
6717 * propagating failures the remainder will be consumed.
6718 *
6719 * FIXME add a capability for mdmon to request spares from another
6720 * container.
6721 */
6722
6723 struct intel_super *super = a->container->sb;
6724 int inst = a->info.container_member;
6725 struct imsm_dev *dev = get_imsm_dev(super, inst);
6726 struct imsm_map *map = get_imsm_map(dev, 0);
6727 int failed = a->info.array.raid_disks;
6728 struct mdinfo *rv = NULL;
6729 struct mdinfo *d;
6730 struct mdinfo *di;
6731 struct metadata_update *mu;
6732 struct dl *dl;
6733 struct imsm_update_activate_spare *u;
6734 int num_spares = 0;
6735 int i;
6736 int allowed;
6737
6738 for (d = a->info.devs ; d ; d = d->next) {
6739 if ((d->curr_state & DS_FAULTY) &&
6740 d->state_fd >= 0)
6741 /* wait for Removal to happen */
6742 return NULL;
6743 if (d->state_fd >= 0)
6744 failed--;
6745 }
6746
6747 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6748 inst, failed, a->info.array.raid_disks, a->info.array.level);
6749
6750 if (imsm_reshape_blocks_arrays_changes(super))
6751 return NULL;
6752
6753 if (a->info.array.level == 4)
6754 /* No repair for takeovered array
6755 * imsm doesn't support raid4
6756 */
6757 return NULL;
6758
6759 if (imsm_check_degraded(super, dev, failed, MAP_0) !=
6760 IMSM_T_STATE_DEGRADED)
6761 return NULL;
6762
6763 /*
6764 * If there are any failed disks check state of the other volume.
6765 * Block rebuild if the another one is failed until failed disks
6766 * are removed from container.
6767 */
6768 if (failed) {
6769 dprintf("found failed disks in %.*s, check if there another"
6770 "failed sub-array.\n",
6771 MAX_RAID_SERIAL_LEN, dev->volume);
6772 /* check if states of the other volumes allow for rebuild */
6773 for (i = 0; i < super->anchor->num_raid_devs; i++) {
6774 if (i != inst) {
6775 allowed = imsm_rebuild_allowed(a->container,
6776 i, failed);
6777 if (!allowed)
6778 return NULL;
6779 }
6780 }
6781 }
6782
6783 /* For each slot, if it is not working, find a spare */
6784 for (i = 0; i < a->info.array.raid_disks; i++) {
6785 for (d = a->info.devs ; d ; d = d->next)
6786 if (d->disk.raid_disk == i)
6787 break;
6788 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
6789 if (d && (d->state_fd >= 0))
6790 continue;
6791
6792 /*
6793 * OK, this device needs recovery. Try to re-add the
6794 * previous occupant of this slot, if this fails see if
6795 * we can continue the assimilation of a spare that was
6796 * partially assimilated, finally try to activate a new
6797 * spare.
6798 */
6799 dl = imsm_readd(super, i, a);
6800 if (!dl)
6801 dl = imsm_add_spare(super, i, a, 0, rv);
6802 if (!dl)
6803 dl = imsm_add_spare(super, i, a, 1, rv);
6804 if (!dl)
6805 continue;
6806
6807 /* found a usable disk with enough space */
6808 di = malloc(sizeof(*di));
6809 if (!di)
6810 continue;
6811 memset(di, 0, sizeof(*di));
6812
6813 /* dl->index will be -1 in the case we are activating a
6814 * pristine spare. imsm_process_update() will create a
6815 * new index in this case. Once a disk is found to be
6816 * failed in all member arrays it is kicked from the
6817 * metadata
6818 */
6819 di->disk.number = dl->index;
6820
6821 /* (ab)use di->devs to store a pointer to the device
6822 * we chose
6823 */
6824 di->devs = (struct mdinfo *) dl;
6825
6826 di->disk.raid_disk = i;
6827 di->disk.major = dl->major;
6828 di->disk.minor = dl->minor;
6829 di->disk.state = 0;
6830 di->recovery_start = 0;
6831 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
6832 di->component_size = a->info.component_size;
6833 di->container_member = inst;
6834 super->random = random32();
6835 di->next = rv;
6836 rv = di;
6837 num_spares++;
6838 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
6839 i, di->data_offset);
6840 }
6841
6842 if (!rv)
6843 /* No spares found */
6844 return rv;
6845 /* Now 'rv' has a list of devices to return.
6846 * Create a metadata_update record to update the
6847 * disk_ord_tbl for the array
6848 */
6849 mu = malloc(sizeof(*mu));
6850 if (mu) {
6851 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
6852 if (mu->buf == NULL) {
6853 free(mu);
6854 mu = NULL;
6855 }
6856 }
6857 if (!mu) {
6858 while (rv) {
6859 struct mdinfo *n = rv->next;
6860
6861 free(rv);
6862 rv = n;
6863 }
6864 return NULL;
6865 }
6866
6867 mu->space = NULL;
6868 mu->space_list = NULL;
6869 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
6870 mu->next = *updates;
6871 u = (struct imsm_update_activate_spare *) mu->buf;
6872
6873 for (di = rv ; di ; di = di->next) {
6874 u->type = update_activate_spare;
6875 u->dl = (struct dl *) di->devs;
6876 di->devs = NULL;
6877 u->slot = di->disk.raid_disk;
6878 u->array = inst;
6879 u->next = u + 1;
6880 u++;
6881 }
6882 (u-1)->next = NULL;
6883 *updates = mu;
6884
6885 return rv;
6886}
6887
6888static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
6889{
6890 struct imsm_dev *dev = get_imsm_dev(super, idx);
6891 struct imsm_map *map = get_imsm_map(dev, 0);
6892 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
6893 struct disk_info *inf = get_disk_info(u);
6894 struct imsm_disk *disk;
6895 int i;
6896 int j;
6897
6898 for (i = 0; i < map->num_members; i++) {
6899 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i, -1));
6900 for (j = 0; j < new_map->num_members; j++)
6901 if (serialcmp(disk->serial, inf[j].serial) == 0)
6902 return 1;
6903 }
6904
6905 return 0;
6906}
6907
6908
6909static struct dl *get_disk_super(struct intel_super *super, int major, int minor)
6910{
6911 struct dl *dl = NULL;
6912 for (dl = super->disks; dl; dl = dl->next)
6913 if ((dl->major == major) && (dl->minor == minor))
6914 return dl;
6915 return NULL;
6916}
6917
6918static int remove_disk_super(struct intel_super *super, int major, int minor)
6919{
6920 struct dl *prev = NULL;
6921 struct dl *dl;
6922
6923 prev = NULL;
6924 for (dl = super->disks; dl; dl = dl->next) {
6925 if ((dl->major == major) && (dl->minor == minor)) {
6926 /* remove */
6927 if (prev)
6928 prev->next = dl->next;
6929 else
6930 super->disks = dl->next;
6931 dl->next = NULL;
6932 __free_imsm_disk(dl);
6933 dprintf("%s: removed %x:%x\n",
6934 __func__, major, minor);
6935 break;
6936 }
6937 prev = dl;
6938 }
6939 return 0;
6940}
6941
6942static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index);
6943
6944static int add_remove_disk_update(struct intel_super *super)
6945{
6946 int check_degraded = 0;
6947 struct dl *disk = NULL;
6948 /* add/remove some spares to/from the metadata/contrainer */
6949 while (super->disk_mgmt_list) {
6950 struct dl *disk_cfg;
6951
6952 disk_cfg = super->disk_mgmt_list;
6953 super->disk_mgmt_list = disk_cfg->next;
6954 disk_cfg->next = NULL;
6955
6956 if (disk_cfg->action == DISK_ADD) {
6957 disk_cfg->next = super->disks;
6958 super->disks = disk_cfg;
6959 check_degraded = 1;
6960 dprintf("%s: added %x:%x\n",
6961 __func__, disk_cfg->major,
6962 disk_cfg->minor);
6963 } else if (disk_cfg->action == DISK_REMOVE) {
6964 dprintf("Disk remove action processed: %x.%x\n",
6965 disk_cfg->major, disk_cfg->minor);
6966 disk = get_disk_super(super,
6967 disk_cfg->major,
6968 disk_cfg->minor);
6969 if (disk) {
6970 /* store action status */
6971 disk->action = DISK_REMOVE;
6972 /* remove spare disks only */
6973 if (disk->index == -1) {
6974 remove_disk_super(super,
6975 disk_cfg->major,
6976 disk_cfg->minor);
6977 }
6978 }
6979 /* release allocate disk structure */
6980 __free_imsm_disk(disk_cfg);
6981 }
6982 }
6983 return check_degraded;
6984}
6985
6986
6987static int apply_reshape_migration_update(struct imsm_update_reshape_migration *u,
6988 struct intel_super *super,
6989 void ***space_list)
6990{
6991 struct intel_dev *id;
6992 void **tofree = NULL;
6993 int ret_val = 0;
6994
6995 dprintf("apply_reshape_migration_update()\n");
6996 if ((u->subdev < 0) ||
6997 (u->subdev > 1)) {
6998 dprintf("imsm: Error: Wrong subdev: %i\n", u->subdev);
6999 return ret_val;
7000 }
7001 if ((space_list == NULL) || (*space_list == NULL)) {
7002 dprintf("imsm: Error: Memory is not allocated\n");
7003 return ret_val;
7004 }
7005
7006 for (id = super->devlist ; id; id = id->next) {
7007 if (id->index == (unsigned)u->subdev) {
7008 struct imsm_dev *dev = get_imsm_dev(super, u->subdev);
7009 struct imsm_map *map;
7010 struct imsm_dev *new_dev =
7011 (struct imsm_dev *)*space_list;
7012 struct imsm_map *migr_map = get_imsm_map(dev, 1);
7013 int to_state;
7014 struct dl *new_disk;
7015
7016 if (new_dev == NULL)
7017 return ret_val;
7018 *space_list = **space_list;
7019 memcpy(new_dev, dev, sizeof_imsm_dev(dev, 0));
7020 map = get_imsm_map(new_dev, 0);
7021 if (migr_map) {
7022 dprintf("imsm: Error: migration in progress");
7023 return ret_val;
7024 }
7025
7026 to_state = map->map_state;
7027 if ((u->new_level == 5) && (map->raid_level == 0)) {
7028 map->num_members++;
7029 /* this should not happen */
7030 if (u->new_disks[0] < 0) {
7031 map->failed_disk_num =
7032 map->num_members - 1;
7033 to_state = IMSM_T_STATE_DEGRADED;
7034 } else
7035 to_state = IMSM_T_STATE_NORMAL;
7036 }
7037 migrate(new_dev, super, to_state, MIGR_GEN_MIGR);
7038 if (u->new_level > -1)
7039 map->raid_level = u->new_level;
7040 migr_map = get_imsm_map(new_dev, 1);
7041 if ((u->new_level == 5) &&
7042 (migr_map->raid_level == 0)) {
7043 int ord = map->num_members - 1;
7044 migr_map->num_members--;
7045 if (u->new_disks[0] < 0)
7046 ord |= IMSM_ORD_REBUILD;
7047 set_imsm_ord_tbl_ent(map,
7048 map->num_members - 1,
7049 ord);
7050 }
7051 id->dev = new_dev;
7052 tofree = (void **)dev;
7053
7054 /* update chunk size
7055 */
7056 if (u->new_chunksize > 0)
7057 map->blocks_per_strip =
7058 __cpu_to_le16(u->new_chunksize * 2);
7059
7060 /* add disk
7061 */
7062 if ((u->new_level != 5) ||
7063 (migr_map->raid_level != 0) ||
7064 (migr_map->raid_level == map->raid_level))
7065 goto skip_disk_add;
7066
7067 if (u->new_disks[0] >= 0) {
7068 /* use passes spare
7069 */
7070 new_disk = get_disk_super(super,
7071 major(u->new_disks[0]),
7072 minor(u->new_disks[0]));
7073 dprintf("imsm: new disk for reshape is: %i:%i "
7074 "(%p, index = %i)\n",
7075 major(u->new_disks[0]),
7076 minor(u->new_disks[0]),
7077 new_disk, new_disk->index);
7078 if (new_disk == NULL)
7079 goto error_disk_add;
7080
7081 new_disk->index = map->num_members - 1;
7082 /* slot to fill in autolayout
7083 */
7084 new_disk->raiddisk = new_disk->index;
7085 new_disk->disk.status |= CONFIGURED_DISK;
7086 new_disk->disk.status &= ~SPARE_DISK;
7087 } else
7088 goto error_disk_add;
7089
7090skip_disk_add:
7091 *tofree = *space_list;
7092 /* calculate new size
7093 */
7094 imsm_set_array_size(new_dev);
7095
7096 ret_val = 1;
7097 }
7098 }
7099
7100 if (tofree)
7101 *space_list = tofree;
7102 return ret_val;
7103
7104error_disk_add:
7105 dprintf("Error: imsm: Cannot find disk.\n");
7106 return ret_val;
7107}
7108
7109static int apply_update_activate_spare(struct imsm_update_activate_spare *u,
7110 struct intel_super *super,
7111 struct active_array *active_array)
7112{
7113 struct imsm_super *mpb = super->anchor;
7114 struct imsm_dev *dev = get_imsm_dev(super, u->array);
7115 struct imsm_map *map = get_imsm_map(dev, 0);
7116 struct imsm_map *migr_map;
7117 struct active_array *a;
7118 struct imsm_disk *disk;
7119 __u8 to_state;
7120 struct dl *dl;
7121 unsigned int found;
7122 int failed;
7123 int victim;
7124 int i;
7125 int second_map_created = 0;
7126
7127 for (; u; u = u->next) {
7128 victim = get_imsm_disk_idx(dev, u->slot, -1);
7129
7130 if (victim < 0)
7131 return 0;
7132
7133 for (dl = super->disks; dl; dl = dl->next)
7134 if (dl == u->dl)
7135 break;
7136
7137 if (!dl) {
7138 fprintf(stderr, "error: imsm_activate_spare passed "
7139 "an unknown disk (index: %d)\n",
7140 u->dl->index);
7141 return 0;
7142 }
7143
7144 /* count failures (excluding rebuilds and the victim)
7145 * to determine map[0] state
7146 */
7147 failed = 0;
7148 for (i = 0; i < map->num_members; i++) {
7149 if (i == u->slot)
7150 continue;
7151 disk = get_imsm_disk(super,
7152 get_imsm_disk_idx(dev, i, -1));
7153 if (!disk || is_failed(disk))
7154 failed++;
7155 }
7156
7157 /* adding a pristine spare, assign a new index */
7158 if (dl->index < 0) {
7159 dl->index = super->anchor->num_disks;
7160 super->anchor->num_disks++;
7161 }
7162 disk = &dl->disk;
7163 disk->status |= CONFIGURED_DISK;
7164 disk->status &= ~SPARE_DISK;
7165
7166 /* mark rebuild */
7167 to_state = imsm_check_degraded(super, dev, failed,
7168 MAP_0);
7169 if (!second_map_created) {
7170 second_map_created = 1;
7171 map->map_state = IMSM_T_STATE_DEGRADED;
7172 migrate(dev, super, to_state, MIGR_REBUILD);
7173 } else
7174 map->map_state = to_state;
7175 migr_map = get_imsm_map(dev, 1);
7176 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
7177 set_imsm_ord_tbl_ent(migr_map, u->slot,
7178 dl->index | IMSM_ORD_REBUILD);
7179
7180 /* update the family_num to mark a new container
7181 * generation, being careful to record the existing
7182 * family_num in orig_family_num to clean up after
7183 * earlier mdadm versions that neglected to set it.
7184 */
7185 if (mpb->orig_family_num == 0)
7186 mpb->orig_family_num = mpb->family_num;
7187 mpb->family_num += super->random;
7188
7189 /* count arrays using the victim in the metadata */
7190 found = 0;
7191 for (a = active_array; a ; a = a->next) {
7192 dev = get_imsm_dev(super, a->info.container_member);
7193 map = get_imsm_map(dev, 0);
7194
7195 if (get_imsm_disk_slot(map, victim) >= 0)
7196 found++;
7197 }
7198
7199 /* delete the victim if it is no longer being
7200 * utilized anywhere
7201 */
7202 if (!found) {
7203 struct dl **dlp;
7204
7205 /* We know that 'manager' isn't touching anything,
7206 * so it is safe to delete
7207 */
7208 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
7209 if ((*dlp)->index == victim)
7210 break;
7211
7212 /* victim may be on the missing list */
7213 if (!*dlp)
7214 for (dlp = &super->missing; *dlp;
7215 dlp = &(*dlp)->next)
7216 if ((*dlp)->index == victim)
7217 break;
7218 imsm_delete(super, dlp, victim);
7219 }
7220 }
7221
7222 return 1;
7223}
7224
7225static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
7226 struct intel_super *super,
7227 void ***space_list)
7228{
7229 struct dl *new_disk;
7230 struct intel_dev *id;
7231 int i;
7232 int delta_disks = u->new_raid_disks - u->old_raid_disks;
7233 int disk_count = u->old_raid_disks;
7234 void **tofree = NULL;
7235 int devices_to_reshape = 1;
7236 struct imsm_super *mpb = super->anchor;
7237 int ret_val = 0;
7238 unsigned int dev_id;
7239
7240 dprintf("imsm: apply_reshape_container_disks_update()\n");
7241
7242 /* enable spares to use in array */
7243 for (i = 0; i < delta_disks; i++) {
7244 new_disk = get_disk_super(super,
7245 major(u->new_disks[i]),
7246 minor(u->new_disks[i]));
7247 dprintf("imsm: new disk for reshape is: %i:%i "
7248 "(%p, index = %i)\n",
7249 major(u->new_disks[i]), minor(u->new_disks[i]),
7250 new_disk, new_disk->index);
7251 if ((new_disk == NULL) ||
7252 ((new_disk->index >= 0) &&
7253 (new_disk->index < u->old_raid_disks)))
7254 goto update_reshape_exit;
7255 new_disk->index = disk_count++;
7256 /* slot to fill in autolayout
7257 */
7258 new_disk->raiddisk = new_disk->index;
7259 new_disk->disk.status |=
7260 CONFIGURED_DISK;
7261 new_disk->disk.status &= ~SPARE_DISK;
7262 }
7263
7264 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7265 mpb->num_raid_devs);
7266 /* manage changes in volume
7267 */
7268 for (dev_id = 0; dev_id < mpb->num_raid_devs; dev_id++) {
7269 void **sp = *space_list;
7270 struct imsm_dev *newdev;
7271 struct imsm_map *newmap, *oldmap;
7272
7273 for (id = super->devlist ; id; id = id->next) {
7274 if (id->index == dev_id)
7275 break;
7276 }
7277 if (id == NULL)
7278 break;
7279 if (!sp)
7280 continue;
7281 *space_list = *sp;
7282 newdev = (void*)sp;
7283 /* Copy the dev, but not (all of) the map */
7284 memcpy(newdev, id->dev, sizeof(*newdev));
7285 oldmap = get_imsm_map(id->dev, 0);
7286 newmap = get_imsm_map(newdev, 0);
7287 /* Copy the current map */
7288 memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
7289 /* update one device only
7290 */
7291 if (devices_to_reshape) {
7292 dprintf("imsm: modifying subdev: %i\n",
7293 id->index);
7294 devices_to_reshape--;
7295 newdev->vol.migr_state = 1;
7296 newdev->vol.curr_migr_unit = 0;
7297 set_migr_type(newdev, MIGR_GEN_MIGR);
7298 newmap->num_members = u->new_raid_disks;
7299 for (i = 0; i < delta_disks; i++) {
7300 set_imsm_ord_tbl_ent(newmap,
7301 u->old_raid_disks + i,
7302 u->old_raid_disks + i);
7303 }
7304 /* New map is correct, now need to save old map
7305 */
7306 newmap = get_imsm_map(newdev, 1);
7307 memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
7308
7309 imsm_set_array_size(newdev);
7310 }
7311
7312 sp = (void **)id->dev;
7313 id->dev = newdev;
7314 *sp = tofree;
7315 tofree = sp;
7316
7317 /* Clear migration record */
7318 memset(super->migr_rec, 0, sizeof(struct migr_record));
7319 }
7320 if (tofree)
7321 *space_list = tofree;
7322 ret_val = 1;
7323
7324update_reshape_exit:
7325
7326 return ret_val;
7327}
7328
7329static int apply_takeover_update(struct imsm_update_takeover *u,
7330 struct intel_super *super,
7331 void ***space_list)
7332{
7333 struct imsm_dev *dev = NULL;
7334 struct intel_dev *dv;
7335 struct imsm_dev *dev_new;
7336 struct imsm_map *map;
7337 struct dl *dm, *du;
7338 int i;
7339
7340 for (dv = super->devlist; dv; dv = dv->next)
7341 if (dv->index == (unsigned int)u->subarray) {
7342 dev = dv->dev;
7343 break;
7344 }
7345
7346 if (dev == NULL)
7347 return 0;
7348
7349 map = get_imsm_map(dev, 0);
7350
7351 if (u->direction == R10_TO_R0) {
7352 /* Number of failed disks must be half of initial disk number */
7353 if (imsm_count_failed(super, dev, MAP_0) !=
7354 (map->num_members / 2))
7355 return 0;
7356
7357 /* iterate through devices to mark removed disks as spare */
7358 for (dm = super->disks; dm; dm = dm->next) {
7359 if (dm->disk.status & FAILED_DISK) {
7360 int idx = dm->index;
7361 /* update indexes on the disk list */
7362/* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7363 the index values will end up being correct.... NB */
7364 for (du = super->disks; du; du = du->next)
7365 if (du->index > idx)
7366 du->index--;
7367 /* mark as spare disk */
7368 mark_spare(dm);
7369 }
7370 }
7371 /* update map */
7372 map->num_members = map->num_members / 2;
7373 map->map_state = IMSM_T_STATE_NORMAL;
7374 map->num_domains = 1;
7375 map->raid_level = 0;
7376 map->failed_disk_num = -1;
7377 }
7378
7379 if (u->direction == R0_TO_R10) {
7380 void **space;
7381 /* update slots in current disk list */
7382 for (dm = super->disks; dm; dm = dm->next) {
7383 if (dm->index >= 0)
7384 dm->index *= 2;
7385 }
7386 /* create new *missing* disks */
7387 for (i = 0; i < map->num_members; i++) {
7388 space = *space_list;
7389 if (!space)
7390 continue;
7391 *space_list = *space;
7392 du = (void *)space;
7393 memcpy(du, super->disks, sizeof(*du));
7394 du->fd = -1;
7395 du->minor = 0;
7396 du->major = 0;
7397 du->index = (i * 2) + 1;
7398 sprintf((char *)du->disk.serial,
7399 " MISSING_%d", du->index);
7400 sprintf((char *)du->serial,
7401 "MISSING_%d", du->index);
7402 du->next = super->missing;
7403 super->missing = du;
7404 }
7405 /* create new dev and map */
7406 space = *space_list;
7407 if (!space)
7408 return 0;
7409 *space_list = *space;
7410 dev_new = (void *)space;
7411 memcpy(dev_new, dev, sizeof(*dev));
7412 /* update new map */
7413 map = get_imsm_map(dev_new, 0);
7414 map->num_members = map->num_members * 2;
7415 map->map_state = IMSM_T_STATE_DEGRADED;
7416 map->num_domains = 2;
7417 map->raid_level = 1;
7418 /* replace dev<->dev_new */
7419 dv->dev = dev_new;
7420 }
7421 /* update disk order table */
7422 for (du = super->disks; du; du = du->next)
7423 if (du->index >= 0)
7424 set_imsm_ord_tbl_ent(map, du->index, du->index);
7425 for (du = super->missing; du; du = du->next)
7426 if (du->index >= 0) {
7427 set_imsm_ord_tbl_ent(map, du->index, du->index);
7428 mark_missing(dv->dev, &du->disk, du->index);
7429 }
7430
7431 return 1;
7432}
7433
7434static void imsm_process_update(struct supertype *st,
7435 struct metadata_update *update)
7436{
7437 /**
7438 * crack open the metadata_update envelope to find the update record
7439 * update can be one of:
7440 * update_reshape_container_disks - all the arrays in the container
7441 * are being reshaped to have more devices. We need to mark
7442 * the arrays for general migration and convert selected spares
7443 * into active devices.
7444 * update_activate_spare - a spare device has replaced a failed
7445 * device in an array, update the disk_ord_tbl. If this disk is
7446 * present in all member arrays then also clear the SPARE_DISK
7447 * flag
7448 * update_create_array
7449 * update_kill_array
7450 * update_rename_array
7451 * update_add_remove_disk
7452 */
7453 struct intel_super *super = st->sb;
7454 struct imsm_super *mpb;
7455 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
7456
7457 /* update requires a larger buf but the allocation failed */
7458 if (super->next_len && !super->next_buf) {
7459 super->next_len = 0;
7460 return;
7461 }
7462
7463 if (super->next_buf) {
7464 memcpy(super->next_buf, super->buf, super->len);
7465 free(super->buf);
7466 super->len = super->next_len;
7467 super->buf = super->next_buf;
7468
7469 super->next_len = 0;
7470 super->next_buf = NULL;
7471 }
7472
7473 mpb = super->anchor;
7474
7475 switch (type) {
7476 case update_general_migration_checkpoint: {
7477 struct intel_dev *id;
7478 struct imsm_update_general_migration_checkpoint *u =
7479 (void *)update->buf;
7480
7481 dprintf("imsm: process_update() "
7482 "for update_general_migration_checkpoint called\n");
7483
7484 /* find device under general migration */
7485 for (id = super->devlist ; id; id = id->next) {
7486 if (is_gen_migration(id->dev)) {
7487 id->dev->vol.curr_migr_unit =
7488 __cpu_to_le32(u->curr_migr_unit);
7489 super->updates_pending++;
7490 }
7491 }
7492 break;
7493 }
7494 case update_takeover: {
7495 struct imsm_update_takeover *u = (void *)update->buf;
7496 if (apply_takeover_update(u, super, &update->space_list)) {
7497 imsm_update_version_info(super);
7498 super->updates_pending++;
7499 }
7500 break;
7501 }
7502
7503 case update_reshape_container_disks: {
7504 struct imsm_update_reshape *u = (void *)update->buf;
7505 if (apply_reshape_container_disks_update(
7506 u, super, &update->space_list))
7507 super->updates_pending++;
7508 break;
7509 }
7510 case update_reshape_migration: {
7511 struct imsm_update_reshape_migration *u = (void *)update->buf;
7512 if (apply_reshape_migration_update(
7513 u, super, &update->space_list))
7514 super->updates_pending++;
7515 break;
7516 }
7517 case update_activate_spare: {
7518 struct imsm_update_activate_spare *u = (void *) update->buf;
7519 if (apply_update_activate_spare(u, super, st->arrays))
7520 super->updates_pending++;
7521 break;
7522 }
7523 case update_create_array: {
7524 /* someone wants to create a new array, we need to be aware of
7525 * a few races/collisions:
7526 * 1/ 'Create' called by two separate instances of mdadm
7527 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7528 * devices that have since been assimilated via
7529 * activate_spare.
7530 * In the event this update can not be carried out mdadm will
7531 * (FIX ME) notice that its update did not take hold.
7532 */
7533 struct imsm_update_create_array *u = (void *) update->buf;
7534 struct intel_dev *dv;
7535 struct imsm_dev *dev;
7536 struct imsm_map *map, *new_map;
7537 unsigned long long start, end;
7538 unsigned long long new_start, new_end;
7539 int i;
7540 struct disk_info *inf;
7541 struct dl *dl;
7542
7543 /* handle racing creates: first come first serve */
7544 if (u->dev_idx < mpb->num_raid_devs) {
7545 dprintf("%s: subarray %d already defined\n",
7546 __func__, u->dev_idx);
7547 goto create_error;
7548 }
7549
7550 /* check update is next in sequence */
7551 if (u->dev_idx != mpb->num_raid_devs) {
7552 dprintf("%s: can not create array %d expected index %d\n",
7553 __func__, u->dev_idx, mpb->num_raid_devs);
7554 goto create_error;
7555 }
7556
7557 new_map = get_imsm_map(&u->dev, 0);
7558 new_start = __le32_to_cpu(new_map->pba_of_lba0);
7559 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
7560 inf = get_disk_info(u);
7561
7562 /* handle activate_spare versus create race:
7563 * check to make sure that overlapping arrays do not include
7564 * overalpping disks
7565 */
7566 for (i = 0; i < mpb->num_raid_devs; i++) {
7567 dev = get_imsm_dev(super, i);
7568 map = get_imsm_map(dev, 0);
7569 start = __le32_to_cpu(map->pba_of_lba0);
7570 end = start + __le32_to_cpu(map->blocks_per_member);
7571 if ((new_start >= start && new_start <= end) ||
7572 (start >= new_start && start <= new_end))
7573 /* overlap */;
7574 else
7575 continue;
7576
7577 if (disks_overlap(super, i, u)) {
7578 dprintf("%s: arrays overlap\n", __func__);
7579 goto create_error;
7580 }
7581 }
7582
7583 /* check that prepare update was successful */
7584 if (!update->space) {
7585 dprintf("%s: prepare update failed\n", __func__);
7586 goto create_error;
7587 }
7588
7589 /* check that all disks are still active before committing
7590 * changes. FIXME: could we instead handle this by creating a
7591 * degraded array? That's probably not what the user expects,
7592 * so better to drop this update on the floor.
7593 */
7594 for (i = 0; i < new_map->num_members; i++) {
7595 dl = serial_to_dl(inf[i].serial, super);
7596 if (!dl) {
7597 dprintf("%s: disk disappeared\n", __func__);
7598 goto create_error;
7599 }
7600 }
7601
7602 super->updates_pending++;
7603
7604 /* convert spares to members and fixup ord_tbl */
7605 for (i = 0; i < new_map->num_members; i++) {
7606 dl = serial_to_dl(inf[i].serial, super);
7607 if (dl->index == -1) {
7608 dl->index = mpb->num_disks;
7609 mpb->num_disks++;
7610 dl->disk.status |= CONFIGURED_DISK;
7611 dl->disk.status &= ~SPARE_DISK;
7612 }
7613 set_imsm_ord_tbl_ent(new_map, i, dl->index);
7614 }
7615
7616 dv = update->space;
7617 dev = dv->dev;
7618 update->space = NULL;
7619 imsm_copy_dev(dev, &u->dev);
7620 dv->index = u->dev_idx;
7621 dv->next = super->devlist;
7622 super->devlist = dv;
7623 mpb->num_raid_devs++;
7624
7625 imsm_update_version_info(super);
7626 break;
7627 create_error:
7628 /* mdmon knows how to release update->space, but not
7629 * ((struct intel_dev *) update->space)->dev
7630 */
7631 if (update->space) {
7632 dv = update->space;
7633 free(dv->dev);
7634 }
7635 break;
7636 }
7637 case update_kill_array: {
7638 struct imsm_update_kill_array *u = (void *) update->buf;
7639 int victim = u->dev_idx;
7640 struct active_array *a;
7641 struct intel_dev **dp;
7642 struct imsm_dev *dev;
7643
7644 /* sanity check that we are not affecting the uuid of
7645 * active arrays, or deleting an active array
7646 *
7647 * FIXME when immutable ids are available, but note that
7648 * we'll also need to fixup the invalidated/active
7649 * subarray indexes in mdstat
7650 */
7651 for (a = st->arrays; a; a = a->next)
7652 if (a->info.container_member >= victim)
7653 break;
7654 /* by definition if mdmon is running at least one array
7655 * is active in the container, so checking
7656 * mpb->num_raid_devs is just extra paranoia
7657 */
7658 dev = get_imsm_dev(super, victim);
7659 if (a || !dev || mpb->num_raid_devs == 1) {
7660 dprintf("failed to delete subarray-%d\n", victim);
7661 break;
7662 }
7663
7664 for (dp = &super->devlist; *dp;)
7665 if ((*dp)->index == (unsigned)super->current_vol) {
7666 *dp = (*dp)->next;
7667 } else {
7668 if ((*dp)->index > (unsigned)victim)
7669 (*dp)->index--;
7670 dp = &(*dp)->next;
7671 }
7672 mpb->num_raid_devs--;
7673 super->updates_pending++;
7674 break;
7675 }
7676 case update_rename_array: {
7677 struct imsm_update_rename_array *u = (void *) update->buf;
7678 char name[MAX_RAID_SERIAL_LEN+1];
7679 int target = u->dev_idx;
7680 struct active_array *a;
7681 struct imsm_dev *dev;
7682
7683 /* sanity check that we are not affecting the uuid of
7684 * an active array
7685 */
7686 snprintf(name, MAX_RAID_SERIAL_LEN, "%s", (char *) u->name);
7687 name[MAX_RAID_SERIAL_LEN] = '\0';
7688 for (a = st->arrays; a; a = a->next)
7689 if (a->info.container_member == target)
7690 break;
7691 dev = get_imsm_dev(super, u->dev_idx);
7692 if (a || !dev || !check_name(super, name, 1)) {
7693 dprintf("failed to rename subarray-%d\n", target);
7694 break;
7695 }
7696
7697 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
7698 super->updates_pending++;
7699 break;
7700 }
7701 case update_add_remove_disk: {
7702 /* we may be able to repair some arrays if disks are
7703 * being added, check teh status of add_remove_disk
7704 * if discs has been added.
7705 */
7706 if (add_remove_disk_update(super)) {
7707 struct active_array *a;
7708
7709 super->updates_pending++;
7710 for (a = st->arrays; a; a = a->next)
7711 a->check_degraded = 1;
7712 }
7713 break;
7714 }
7715 default:
7716 fprintf(stderr, "error: unsuported process update type:"
7717 "(type: %d)\n", type);
7718 }
7719}
7720
7721static struct mdinfo *get_spares_for_grow(struct supertype *st);
7722
7723static void imsm_prepare_update(struct supertype *st,
7724 struct metadata_update *update)
7725{
7726 /**
7727 * Allocate space to hold new disk entries, raid-device entries or a new
7728 * mpb if necessary. The manager synchronously waits for updates to
7729 * complete in the monitor, so new mpb buffers allocated here can be
7730 * integrated by the monitor thread without worrying about live pointers
7731 * in the manager thread.
7732 */
7733 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
7734 struct intel_super *super = st->sb;
7735 struct imsm_super *mpb = super->anchor;
7736 size_t buf_len;
7737 size_t len = 0;
7738
7739 switch (type) {
7740 case update_general_migration_checkpoint:
7741 dprintf("imsm: prepare_update() "
7742 "for update_general_migration_checkpoint called\n");
7743 break;
7744 case update_takeover: {
7745 struct imsm_update_takeover *u = (void *)update->buf;
7746 if (u->direction == R0_TO_R10) {
7747 void **tail = (void **)&update->space_list;
7748 struct imsm_dev *dev = get_imsm_dev(super, u->subarray);
7749 struct imsm_map *map = get_imsm_map(dev, 0);
7750 int num_members = map->num_members;
7751 void *space;
7752 int size, i;
7753 int err = 0;
7754 /* allocate memory for added disks */
7755 for (i = 0; i < num_members; i++) {
7756 size = sizeof(struct dl);
7757 space = malloc(size);
7758 if (!space) {
7759 err++;
7760 break;
7761 }
7762 *tail = space;
7763 tail = space;
7764 *tail = NULL;
7765 }
7766 /* allocate memory for new device */
7767 size = sizeof_imsm_dev(super->devlist->dev, 0) +
7768 (num_members * sizeof(__u32));
7769 space = malloc(size);
7770 if (!space)
7771 err++;
7772 else {
7773 *tail = space;
7774 tail = space;
7775 *tail = NULL;
7776 }
7777 if (!err) {
7778 len = disks_to_mpb_size(num_members * 2);
7779 } else {
7780 /* if allocation didn't success, free buffer */
7781 while (update->space_list) {
7782 void **sp = update->space_list;
7783 update->space_list = *sp;
7784 free(sp);
7785 }
7786 }
7787 }
7788
7789 break;
7790 }
7791 case update_reshape_container_disks: {
7792 /* Every raid device in the container is about to
7793 * gain some more devices, and we will enter a
7794 * reconfiguration.
7795 * So each 'imsm_map' will be bigger, and the imsm_vol
7796 * will now hold 2 of them.
7797 * Thus we need new 'struct imsm_dev' allocations sized
7798 * as sizeof_imsm_dev but with more devices in both maps.
7799 */
7800 struct imsm_update_reshape *u = (void *)update->buf;
7801 struct intel_dev *dl;
7802 void **space_tail = (void**)&update->space_list;
7803
7804 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7805
7806 for (dl = super->devlist; dl; dl = dl->next) {
7807 int size = sizeof_imsm_dev(dl->dev, 1);
7808 void *s;
7809 if (u->new_raid_disks > u->old_raid_disks)
7810 size += sizeof(__u32)*2*
7811 (u->new_raid_disks - u->old_raid_disks);
7812 s = malloc(size);
7813 if (!s)
7814 break;
7815 *space_tail = s;
7816 space_tail = s;
7817 *space_tail = NULL;
7818 }
7819
7820 len = disks_to_mpb_size(u->new_raid_disks);
7821 dprintf("New anchor length is %llu\n", (unsigned long long)len);
7822 break;
7823 }
7824 case update_reshape_migration: {
7825 /* for migration level 0->5 we need to add disks
7826 * so the same as for container operation we will copy
7827 * device to the bigger location.
7828 * in memory prepared device and new disk area are prepared
7829 * for usage in process update
7830 */
7831 struct imsm_update_reshape_migration *u = (void *)update->buf;
7832 struct intel_dev *id;
7833 void **space_tail = (void **)&update->space_list;
7834 int size;
7835 void *s;
7836 int current_level = -1;
7837
7838 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7839
7840 /* add space for bigger array in update
7841 */
7842 for (id = super->devlist; id; id = id->next) {
7843 if (id->index == (unsigned)u->subdev) {
7844 size = sizeof_imsm_dev(id->dev, 1);
7845 if (u->new_raid_disks > u->old_raid_disks)
7846 size += sizeof(__u32)*2*
7847 (u->new_raid_disks - u->old_raid_disks);
7848 s = malloc(size);
7849 if (!s)
7850 break;
7851 *space_tail = s;
7852 space_tail = s;
7853 *space_tail = NULL;
7854 break;
7855 }
7856 }
7857 if (update->space_list == NULL)
7858 break;
7859
7860 /* add space for disk in update
7861 */
7862 size = sizeof(struct dl);
7863 s = malloc(size);
7864 if (!s) {
7865 free(update->space_list);
7866 update->space_list = NULL;
7867 break;
7868 }
7869 *space_tail = s;
7870 space_tail = s;
7871 *space_tail = NULL;
7872
7873 /* add spare device to update
7874 */
7875 for (id = super->devlist ; id; id = id->next)
7876 if (id->index == (unsigned)u->subdev) {
7877 struct imsm_dev *dev;
7878 struct imsm_map *map;
7879
7880 dev = get_imsm_dev(super, u->subdev);
7881 map = get_imsm_map(dev, 0);
7882 current_level = map->raid_level;
7883 break;
7884 }
7885 if ((u->new_level == 5) && (u->new_level != current_level)) {
7886 struct mdinfo *spares;
7887
7888 spares = get_spares_for_grow(st);
7889 if (spares) {
7890 struct dl *dl;
7891 struct mdinfo *dev;
7892
7893 dev = spares->devs;
7894 if (dev) {
7895 u->new_disks[0] =
7896 makedev(dev->disk.major,
7897 dev->disk.minor);
7898 dl = get_disk_super(super,
7899 dev->disk.major,
7900 dev->disk.minor);
7901 dl->index = u->old_raid_disks;
7902 dev = dev->next;
7903 }
7904 sysfs_free(spares);
7905 }
7906 }
7907 len = disks_to_mpb_size(u->new_raid_disks);
7908 dprintf("New anchor length is %llu\n", (unsigned long long)len);
7909 break;
7910 }
7911 case update_create_array: {
7912 struct imsm_update_create_array *u = (void *) update->buf;
7913 struct intel_dev *dv;
7914 struct imsm_dev *dev = &u->dev;
7915 struct imsm_map *map = get_imsm_map(dev, 0);
7916 struct dl *dl;
7917 struct disk_info *inf;
7918 int i;
7919 int activate = 0;
7920
7921 inf = get_disk_info(u);
7922 len = sizeof_imsm_dev(dev, 1);
7923 /* allocate a new super->devlist entry */
7924 dv = malloc(sizeof(*dv));
7925 if (dv) {
7926 dv->dev = malloc(len);
7927 if (dv->dev)
7928 update->space = dv;
7929 else {
7930 free(dv);
7931 update->space = NULL;
7932 }
7933 }
7934
7935 /* count how many spares will be converted to members */
7936 for (i = 0; i < map->num_members; i++) {
7937 dl = serial_to_dl(inf[i].serial, super);
7938 if (!dl) {
7939 /* hmm maybe it failed?, nothing we can do about
7940 * it here
7941 */
7942 continue;
7943 }
7944 if (count_memberships(dl, super) == 0)
7945 activate++;
7946 }
7947 len += activate * sizeof(struct imsm_disk);
7948 break;
7949 default:
7950 break;
7951 }
7952 }
7953
7954 /* check if we need a larger metadata buffer */
7955 if (super->next_buf)
7956 buf_len = super->next_len;
7957 else
7958 buf_len = super->len;
7959
7960 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
7961 /* ok we need a larger buf than what is currently allocated
7962 * if this allocation fails process_update will notice that
7963 * ->next_len is set and ->next_buf is NULL
7964 */
7965 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
7966 if (super->next_buf)
7967 free(super->next_buf);
7968
7969 super->next_len = buf_len;
7970 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
7971 memset(super->next_buf, 0, buf_len);
7972 else
7973 super->next_buf = NULL;
7974 }
7975}
7976
7977/* must be called while manager is quiesced */
7978static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index)
7979{
7980 struct imsm_super *mpb = super->anchor;
7981 struct dl *iter;
7982 struct imsm_dev *dev;
7983 struct imsm_map *map;
7984 int i, j, num_members;
7985 __u32 ord;
7986
7987 dprintf("%s: deleting device[%d] from imsm_super\n",
7988 __func__, index);
7989
7990 /* shift all indexes down one */
7991 for (iter = super->disks; iter; iter = iter->next)
7992 if (iter->index > (int)index)
7993 iter->index--;
7994 for (iter = super->missing; iter; iter = iter->next)
7995 if (iter->index > (int)index)
7996 iter->index--;
7997
7998 for (i = 0; i < mpb->num_raid_devs; i++) {
7999 dev = get_imsm_dev(super, i);
8000 map = get_imsm_map(dev, 0);
8001 num_members = map->num_members;
8002 for (j = 0; j < num_members; j++) {
8003 /* update ord entries being careful not to propagate
8004 * ord-flags to the first map
8005 */
8006 ord = get_imsm_ord_tbl_ent(dev, j, -1);
8007
8008 if (ord_to_idx(ord) <= index)
8009 continue;
8010
8011 map = get_imsm_map(dev, 0);
8012 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
8013 map = get_imsm_map(dev, 1);
8014 if (map)
8015 set_imsm_ord_tbl_ent(map, j, ord - 1);
8016 }
8017 }
8018
8019 mpb->num_disks--;
8020 super->updates_pending++;
8021 if (*dlp) {
8022 struct dl *dl = *dlp;
8023
8024 *dlp = (*dlp)->next;
8025 __free_imsm_disk(dl);
8026 }
8027}
8028#endif /* MDASSEMBLE */
8029
8030static void close_targets(int *targets, int new_disks)
8031{
8032 int i;
8033
8034 if (!targets)
8035 return;
8036
8037 for (i = 0; i < new_disks; i++) {
8038 if (targets[i] >= 0) {
8039 close(targets[i]);
8040 targets[i] = -1;
8041 }
8042 }
8043}
8044
8045static int imsm_get_allowed_degradation(int level, int raid_disks,
8046 struct intel_super *super,
8047 struct imsm_dev *dev)
8048{
8049 switch (level) {
8050 case 10:{
8051 int ret_val = 0;
8052 struct imsm_map *map;
8053 int i;
8054
8055 ret_val = raid_disks/2;
8056 /* check map if all disks pairs not failed
8057 * in both maps
8058 */
8059 map = get_imsm_map(dev, 0);
8060 for (i = 0; i < ret_val; i++) {
8061 int degradation = 0;
8062 if (get_imsm_disk(super, i) == NULL)
8063 degradation++;
8064 if (get_imsm_disk(super, i + 1) == NULL)
8065 degradation++;
8066 if (degradation == 2)
8067 return 0;
8068 }
8069 map = get_imsm_map(dev, 1);
8070 /* if there is no second map
8071 * result can be returned
8072 */
8073 if (map == NULL)
8074 return ret_val;
8075 /* check degradation in second map
8076 */
8077 for (i = 0; i < ret_val; i++) {
8078 int degradation = 0;
8079 if (get_imsm_disk(super, i) == NULL)
8080 degradation++;
8081 if (get_imsm_disk(super, i + 1) == NULL)
8082 degradation++;
8083 if (degradation == 2)
8084 return 0;
8085 }
8086 return ret_val;
8087 }
8088 case 5:
8089 return 1;
8090 case 6:
8091 return 2;
8092 default:
8093 return 0;
8094 }
8095}
8096
8097
8098/*******************************************************************************
8099 * Function: open_backup_targets
8100 * Description: Function opens file descriptors for all devices given in
8101 * info->devs
8102 * Parameters:
8103 * info : general array info
8104 * raid_disks : number of disks
8105 * raid_fds : table of device's file descriptors
8106 * super : intel super for raid10 degradation check
8107 * dev : intel device for raid10 degradation check
8108 * Returns:
8109 * 0 : success
8110 * -1 : fail
8111 ******************************************************************************/
8112int open_backup_targets(struct mdinfo *info, int raid_disks, int *raid_fds,
8113 struct intel_super *super, struct imsm_dev *dev)
8114{
8115 struct mdinfo *sd;
8116 int i;
8117 int opened = 0;
8118
8119 for (i = 0; i < raid_disks; i++)
8120 raid_fds[i] = -1;
8121
8122 for (sd = info->devs ; sd ; sd = sd->next) {
8123 char *dn;
8124
8125 if (sd->disk.state & (1<<MD_DISK_FAULTY)) {
8126 dprintf("disk is faulty!!\n");
8127 continue;
8128 }
8129
8130 if ((sd->disk.raid_disk >= raid_disks) ||
8131 (sd->disk.raid_disk < 0))
8132 continue;
8133
8134 dn = map_dev(sd->disk.major,
8135 sd->disk.minor, 1);
8136 raid_fds[sd->disk.raid_disk] = dev_open(dn, O_RDWR);
8137 if (raid_fds[sd->disk.raid_disk] < 0) {
8138 fprintf(stderr, "cannot open component\n");
8139 continue;
8140 }
8141 opened++;
8142 }
8143 /* check if maximum array degradation level is not exceeded
8144 */
8145 if ((raid_disks - opened) >
8146 imsm_get_allowed_degradation(info->new_level,
8147 raid_disks,
8148 super, dev)) {
8149 fprintf(stderr, "Not enough disks can be opened.\n");
8150 close_targets(raid_fds, raid_disks);
8151 return -2;
8152 }
8153 return 0;
8154}
8155
8156#ifndef MDASSEMBLE
8157/*******************************************************************************
8158 * Function: init_migr_record_imsm
8159 * Description: Function inits imsm migration record
8160 * Parameters:
8161 * super : imsm internal array info
8162 * dev : device under migration
8163 * info : general array info to find the smallest device
8164 * Returns:
8165 * none
8166 ******************************************************************************/
8167void init_migr_record_imsm(struct supertype *st, struct imsm_dev *dev,
8168 struct mdinfo *info)
8169{
8170 struct intel_super *super = st->sb;
8171 struct migr_record *migr_rec = super->migr_rec;
8172 int new_data_disks;
8173 unsigned long long dsize, dev_sectors;
8174 long long unsigned min_dev_sectors = -1LLU;
8175 struct mdinfo *sd;
8176 char nm[30];
8177 int fd;
8178 struct imsm_map *map_dest = get_imsm_map(dev, 0);
8179 struct imsm_map *map_src = get_imsm_map(dev, 1);
8180 unsigned long long num_migr_units;
8181 unsigned long long array_blocks;
8182
8183 memset(migr_rec, 0, sizeof(struct migr_record));
8184 migr_rec->family_num = __cpu_to_le32(super->anchor->family_num);
8185
8186 /* only ascending reshape supported now */
8187 migr_rec->ascending_migr = __cpu_to_le32(1);
8188
8189 migr_rec->dest_depth_per_unit = GEN_MIGR_AREA_SIZE /
8190 max(map_dest->blocks_per_strip, map_src->blocks_per_strip);
8191 migr_rec->dest_depth_per_unit *= map_dest->blocks_per_strip;
8192 new_data_disks = imsm_num_data_members(dev, 0);
8193 migr_rec->blocks_per_unit =
8194 __cpu_to_le32(migr_rec->dest_depth_per_unit * new_data_disks);
8195 migr_rec->dest_depth_per_unit =
8196 __cpu_to_le32(migr_rec->dest_depth_per_unit);
8197 array_blocks = info->component_size * new_data_disks;
8198 num_migr_units =
8199 array_blocks / __le32_to_cpu(migr_rec->blocks_per_unit);
8200
8201 if (array_blocks % __le32_to_cpu(migr_rec->blocks_per_unit))
8202 num_migr_units++;
8203 migr_rec->num_migr_units = __cpu_to_le32(num_migr_units);
8204
8205 migr_rec->post_migr_vol_cap = dev->size_low;
8206 migr_rec->post_migr_vol_cap_hi = dev->size_high;
8207
8208
8209 /* Find the smallest dev */
8210 for (sd = info->devs ; sd ; sd = sd->next) {
8211 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
8212 fd = dev_open(nm, O_RDONLY);
8213 if (fd < 0)
8214 continue;
8215 get_dev_size(fd, NULL, &dsize);
8216 dev_sectors = dsize / 512;
8217 if (dev_sectors < min_dev_sectors)
8218 min_dev_sectors = dev_sectors;
8219 close(fd);
8220 }
8221 migr_rec->ckpt_area_pba = __cpu_to_le32(min_dev_sectors -
8222 RAID_DISK_RESERVED_BLOCKS_IMSM_HI);
8223
8224 write_imsm_migr_rec(st);
8225
8226 return;
8227}
8228
8229/*******************************************************************************
8230 * Function: save_backup_imsm
8231 * Description: Function saves critical data stripes to Migration Copy Area
8232 * and updates the current migration unit status.
8233 * Use restore_stripes() to form a destination stripe,
8234 * and to write it to the Copy Area.
8235 * Parameters:
8236 * st : supertype information
8237 * dev : imsm device that backup is saved for
8238 * info : general array info
8239 * buf : input buffer
8240 * length : length of data to backup (blocks_per_unit)
8241 * Returns:
8242 * 0 : success
8243 *, -1 : fail
8244 ******************************************************************************/
8245int save_backup_imsm(struct supertype *st,
8246 struct imsm_dev *dev,
8247 struct mdinfo *info,
8248 void *buf,
8249 int length)
8250{
8251 int rv = -1;
8252 struct intel_super *super = st->sb;
8253 unsigned long long *target_offsets = NULL;
8254 int *targets = NULL;
8255 int i;
8256 struct imsm_map *map_dest = get_imsm_map(dev, 0);
8257 int new_disks = map_dest->num_members;
8258 int dest_layout = 0;
8259 int dest_chunk;
8260 unsigned long long start;
8261 int data_disks = imsm_num_data_members(dev, 0);
8262
8263 targets = malloc(new_disks * sizeof(int));
8264 if (!targets)
8265 goto abort;
8266
8267 for (i = 0; i < new_disks; i++)
8268 targets[i] = -1;
8269
8270 target_offsets = malloc(new_disks * sizeof(unsigned long long));
8271 if (!target_offsets)
8272 goto abort;
8273
8274 start = info->reshape_progress * 512;
8275 for (i = 0; i < new_disks; i++) {
8276 target_offsets[i] = (unsigned long long)
8277 __le32_to_cpu(super->migr_rec->ckpt_area_pba) * 512;
8278 /* move back copy area adderss, it will be moved forward
8279 * in restore_stripes() using start input variable
8280 */
8281 target_offsets[i] -= start/data_disks;
8282 }
8283
8284 if (open_backup_targets(info, new_disks, targets,
8285 super, dev))
8286 goto abort;
8287
8288 dest_layout = imsm_level_to_layout(map_dest->raid_level);
8289 dest_chunk = __le16_to_cpu(map_dest->blocks_per_strip) * 512;
8290
8291 if (restore_stripes(targets, /* list of dest devices */
8292 target_offsets, /* migration record offsets */
8293 new_disks,
8294 dest_chunk,
8295 map_dest->raid_level,
8296 dest_layout,
8297 -1, /* source backup file descriptor */
8298 0, /* input buf offset
8299 * always 0 buf is already offseted */
8300 start,
8301 length,
8302 buf) != 0) {
8303 fprintf(stderr, Name ": Error restoring stripes\n");
8304 goto abort;
8305 }
8306
8307 rv = 0;
8308
8309abort:
8310 if (targets) {
8311 close_targets(targets, new_disks);
8312 free(targets);
8313 }
8314 free(target_offsets);
8315
8316 return rv;
8317}
8318
8319/*******************************************************************************
8320 * Function: save_checkpoint_imsm
8321 * Description: Function called for current unit status update
8322 * in the migration record. It writes it to disk.
8323 * Parameters:
8324 * super : imsm internal array info
8325 * info : general array info
8326 * Returns:
8327 * 0: success
8328 * 1: failure
8329 * 2: failure, means no valid migration record
8330 * / no general migration in progress /
8331 ******************************************************************************/
8332int save_checkpoint_imsm(struct supertype *st, struct mdinfo *info, int state)
8333{
8334 struct intel_super *super = st->sb;
8335 unsigned long long blocks_per_unit;
8336 unsigned long long curr_migr_unit;
8337
8338 if (load_imsm_migr_rec(super, info) != 0) {
8339 dprintf("imsm: ERROR: Cannot read migration record "
8340 "for checkpoint save.\n");
8341 return 1;
8342 }
8343
8344 blocks_per_unit = __le32_to_cpu(super->migr_rec->blocks_per_unit);
8345 if (blocks_per_unit == 0) {
8346 dprintf("imsm: no migration in progress.\n");
8347 return 2;
8348 }
8349 curr_migr_unit = info->reshape_progress / blocks_per_unit;
8350 /* check if array is alligned to copy area
8351 * if it is not alligned, add one to current migration unit value
8352 * this can happend on array reshape finish only
8353 */
8354 if (info->reshape_progress % blocks_per_unit)
8355 curr_migr_unit++;
8356
8357 super->migr_rec->curr_migr_unit =
8358 __cpu_to_le32(curr_migr_unit);
8359 super->migr_rec->rec_status = __cpu_to_le32(state);
8360 super->migr_rec->dest_1st_member_lba =
8361 __cpu_to_le32(curr_migr_unit *
8362 __le32_to_cpu(super->migr_rec->dest_depth_per_unit));
8363 if (write_imsm_migr_rec(st) < 0) {
8364 dprintf("imsm: Cannot write migration record "
8365 "outside backup area\n");
8366 return 1;
8367 }
8368
8369 return 0;
8370}
8371
8372/*******************************************************************************
8373 * Function: recover_backup_imsm
8374 * Description: Function recovers critical data from the Migration Copy Area
8375 * while assembling an array.
8376 * Parameters:
8377 * super : imsm internal array info
8378 * info : general array info
8379 * Returns:
8380 * 0 : success (or there is no data to recover)
8381 * 1 : fail
8382 ******************************************************************************/
8383int recover_backup_imsm(struct supertype *st, struct mdinfo *info)
8384{
8385 struct intel_super *super = st->sb;
8386 struct migr_record *migr_rec = super->migr_rec;
8387 struct imsm_map *map_dest = NULL;
8388 struct intel_dev *id = NULL;
8389 unsigned long long read_offset;
8390 unsigned long long write_offset;
8391 unsigned unit_len;
8392 int *targets = NULL;
8393 int new_disks, i, err;
8394 char *buf = NULL;
8395 int retval = 1;
8396 unsigned long curr_migr_unit = __le32_to_cpu(migr_rec->curr_migr_unit);
8397 unsigned long num_migr_units = __le32_to_cpu(migr_rec->num_migr_units);
8398 char buffer[20];
8399 int skipped_disks = 0;
8400
8401 err = sysfs_get_str(info, NULL, "array_state", (char *)buffer, 20);
8402 if (err < 1)
8403 return 1;
8404
8405 /* recover data only during assemblation */
8406 if (strncmp(buffer, "inactive", 8) != 0)
8407 return 0;
8408 /* no data to recover */
8409 if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL)
8410 return 0;
8411 if (curr_migr_unit >= num_migr_units)
8412 return 1;
8413
8414 /* find device during reshape */
8415 for (id = super->devlist; id; id = id->next)
8416 if (is_gen_migration(id->dev))
8417 break;
8418 if (id == NULL)
8419 return 1;
8420
8421 map_dest = get_imsm_map(id->dev, 0);
8422 new_disks = map_dest->num_members;
8423
8424 read_offset = (unsigned long long)
8425 __le32_to_cpu(migr_rec->ckpt_area_pba) * 512;
8426
8427 write_offset = ((unsigned long long)
8428 __le32_to_cpu(migr_rec->dest_1st_member_lba) +
8429 __le32_to_cpu(map_dest->pba_of_lba0)) * 512;
8430
8431 unit_len = __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512;
8432 if (posix_memalign((void **)&buf, 512, unit_len) != 0)
8433 goto abort;
8434 targets = malloc(new_disks * sizeof(int));
8435 if (!targets)
8436 goto abort;
8437
8438 if (open_backup_targets(info, new_disks, targets, super, id->dev)) {
8439 fprintf(stderr,
8440 Name ": Cannot open some devices belonging to array.\n");
8441 goto abort;
8442 }
8443
8444 for (i = 0; i < new_disks; i++) {
8445 if (targets[i] < 0) {
8446 skipped_disks++;
8447 continue;
8448 }
8449 if (lseek64(targets[i], read_offset, SEEK_SET) < 0) {
8450 fprintf(stderr,
8451 Name ": Cannot seek to block: %s\n",
8452 strerror(errno));
8453 skipped_disks++;
8454 continue;
8455 }
8456 if ((unsigned)read(targets[i], buf, unit_len) != unit_len) {
8457 fprintf(stderr,
8458 Name ": Cannot read copy area block: %s\n",
8459 strerror(errno));
8460 skipped_disks++;
8461 continue;
8462 }
8463 if (lseek64(targets[i], write_offset, SEEK_SET) < 0) {
8464 fprintf(stderr,
8465 Name ": Cannot seek to block: %s\n",
8466 strerror(errno));
8467 skipped_disks++;
8468 continue;
8469 }
8470 if ((unsigned)write(targets[i], buf, unit_len) != unit_len) {
8471 fprintf(stderr,
8472 Name ": Cannot restore block: %s\n",
8473 strerror(errno));
8474 skipped_disks++;
8475 continue;
8476 }
8477 }
8478
8479 if (skipped_disks > imsm_get_allowed_degradation(info->new_level,
8480 new_disks,
8481 super,
8482 id->dev)) {
8483 fprintf(stderr,
8484 Name ": Cannot restore data from backup."
8485 " Too many failed disks\n");
8486 goto abort;
8487 }
8488
8489 if (save_checkpoint_imsm(st, info, UNIT_SRC_NORMAL)) {
8490 /* ignore error == 2, this can mean end of reshape here
8491 */
8492 dprintf("imsm: Cannot write checkpoint to "
8493 "migration record (UNIT_SRC_NORMAL) during restart\n");
8494 } else
8495 retval = 0;
8496
8497abort:
8498 if (targets) {
8499 for (i = 0; i < new_disks; i++)
8500 if (targets[i])
8501 close(targets[i]);
8502 free(targets);
8503 }
8504 free(buf);
8505 return retval;
8506}
8507
8508static char disk_by_path[] = "/dev/disk/by-path/";
8509
8510static const char *imsm_get_disk_controller_domain(const char *path)
8511{
8512 char disk_path[PATH_MAX];
8513 char *drv=NULL;
8514 struct stat st;
8515
8516 strncpy(disk_path, disk_by_path, PATH_MAX - 1);
8517 strncat(disk_path, path, PATH_MAX - strlen(disk_path) - 1);
8518 if (stat(disk_path, &st) == 0) {
8519 struct sys_dev* hba;
8520 char *path=NULL;
8521
8522 path = devt_to_devpath(st.st_rdev);
8523 if (path == NULL)
8524 return "unknown";
8525 hba = find_disk_attached_hba(-1, path);
8526 if (hba && hba->type == SYS_DEV_SAS)
8527 drv = "isci";
8528 else if (hba && hba->type == SYS_DEV_SATA)
8529 drv = "ahci";
8530 else
8531 drv = "unknown";
8532 dprintf("path: %s hba: %s attached: %s\n",
8533 path, (hba) ? hba->path : "NULL", drv);
8534 free(path);
8535 if (hba)
8536 free_sys_dev(&hba);
8537 }
8538 return drv;
8539}
8540
8541static int imsm_find_array_minor_by_subdev(int subdev, int container, int *minor)
8542{
8543 char subdev_name[20];
8544 struct mdstat_ent *mdstat;
8545
8546 sprintf(subdev_name, "%d", subdev);
8547 mdstat = mdstat_by_subdev(subdev_name, container);
8548 if (!mdstat)
8549 return -1;
8550
8551 *minor = mdstat->devnum;
8552 free_mdstat(mdstat);
8553 return 0;
8554}
8555
8556static int imsm_reshape_is_allowed_on_container(struct supertype *st,
8557 struct geo_params *geo,
8558 int *old_raid_disks)
8559{
8560 /* currently we only support increasing the number of devices
8561 * for a container. This increases the number of device for each
8562 * member array. They must all be RAID0 or RAID5.
8563 */
8564 int ret_val = 0;
8565 struct mdinfo *info, *member;
8566 int devices_that_can_grow = 0;
8567
8568 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8569 "st->devnum = (%i)\n",
8570 st->devnum);
8571
8572 if (geo->size != -1 ||
8573 geo->level != UnSet ||
8574 geo->layout != UnSet ||
8575 geo->chunksize != 0 ||
8576 geo->raid_disks == UnSet) {
8577 dprintf("imsm: Container operation is allowed for "
8578 "raid disks number change only.\n");
8579 return ret_val;
8580 }
8581
8582 info = container_content_imsm(st, NULL);
8583 for (member = info; member; member = member->next) {
8584 int result;
8585 int minor;
8586
8587 dprintf("imsm: checking device_num: %i\n",
8588 member->container_member);
8589
8590 if (geo->raid_disks <= member->array.raid_disks) {
8591 /* we work on container for Online Capacity Expansion
8592 * only so raid_disks has to grow
8593 */
8594 dprintf("imsm: for container operation raid disks "
8595 "increase is required\n");
8596 break;
8597 }
8598
8599 if ((info->array.level != 0) &&
8600 (info->array.level != 5)) {
8601 /* we cannot use this container with other raid level
8602 */
8603 dprintf("imsm: for container operation wrong"
8604 " raid level (%i) detected\n",
8605 info->array.level);
8606 break;
8607 } else {
8608 /* check for platform support
8609 * for this raid level configuration
8610 */
8611 struct intel_super *super = st->sb;
8612 if (!is_raid_level_supported(super->orom,
8613 member->array.level,
8614 geo->raid_disks)) {
8615 dprintf("platform does not support raid%d with"
8616 " %d disk%s\n",
8617 info->array.level,
8618 geo->raid_disks,
8619 geo->raid_disks > 1 ? "s" : "");
8620 break;
8621 }
8622 /* check if component size is aligned to chunk size
8623 */
8624 if (info->component_size %
8625 (info->array.chunk_size/512)) {
8626 dprintf("Component size is not aligned to "
8627 "chunk size\n");
8628 break;
8629 }
8630 }
8631
8632 if (*old_raid_disks &&
8633 info->array.raid_disks != *old_raid_disks)
8634 break;
8635 *old_raid_disks = info->array.raid_disks;
8636
8637 /* All raid5 and raid0 volumes in container
8638 * have to be ready for Online Capacity Expansion
8639 * so they need to be assembled. We have already
8640 * checked that no recovery etc is happening.
8641 */
8642 result = imsm_find_array_minor_by_subdev(member->container_member,
8643 st->container_dev,
8644 &minor);
8645 if (result < 0) {
8646 dprintf("imsm: cannot find array\n");
8647 break;
8648 }
8649 devices_that_can_grow++;
8650 }
8651 sysfs_free(info);
8652 if (!member && devices_that_can_grow)
8653 ret_val = 1;
8654
8655 if (ret_val)
8656 dprintf("\tContainer operation allowed\n");
8657 else
8658 dprintf("\tError: %i\n", ret_val);
8659
8660 return ret_val;
8661}
8662
8663/* Function: get_spares_for_grow
8664 * Description: Allocates memory and creates list of spare devices
8665 * avaliable in container. Checks if spare drive size is acceptable.
8666 * Parameters: Pointer to the supertype structure
8667 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8668 * NULL if fail
8669 */
8670static struct mdinfo *get_spares_for_grow(struct supertype *st)
8671{
8672 unsigned long long min_size = min_acceptable_spare_size_imsm(st);
8673 return container_choose_spares(st, min_size, NULL, NULL, NULL, 0);
8674}
8675
8676/******************************************************************************
8677 * function: imsm_create_metadata_update_for_reshape
8678 * Function creates update for whole IMSM container.
8679 *
8680 ******************************************************************************/
8681static int imsm_create_metadata_update_for_reshape(
8682 struct supertype *st,
8683 struct geo_params *geo,
8684 int old_raid_disks,
8685 struct imsm_update_reshape **updatep)
8686{
8687 struct intel_super *super = st->sb;
8688 struct imsm_super *mpb = super->anchor;
8689 int update_memory_size = 0;
8690 struct imsm_update_reshape *u = NULL;
8691 struct mdinfo *spares = NULL;
8692 int i;
8693 int delta_disks = 0;
8694 struct mdinfo *dev;
8695
8696 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8697 geo->raid_disks);
8698
8699 delta_disks = geo->raid_disks - old_raid_disks;
8700
8701 /* size of all update data without anchor */
8702 update_memory_size = sizeof(struct imsm_update_reshape);
8703
8704 /* now add space for spare disks that we need to add. */
8705 update_memory_size += sizeof(u->new_disks[0]) * (delta_disks - 1);
8706
8707 u = calloc(1, update_memory_size);
8708 if (u == NULL) {
8709 dprintf("error: "
8710 "cannot get memory for imsm_update_reshape update\n");
8711 return 0;
8712 }
8713 u->type = update_reshape_container_disks;
8714 u->old_raid_disks = old_raid_disks;
8715 u->new_raid_disks = geo->raid_disks;
8716
8717 /* now get spare disks list
8718 */
8719 spares = get_spares_for_grow(st);
8720
8721 if (spares == NULL
8722 || delta_disks > spares->array.spare_disks) {
8723 fprintf(stderr, Name ": imsm: ERROR: Cannot get spare devices "
8724 "for %s.\n", geo->dev_name);
8725 i = -1;
8726 goto abort;
8727 }
8728
8729 /* we have got spares
8730 * update disk list in imsm_disk list table in anchor
8731 */
8732 dprintf("imsm: %i spares are available.\n\n",
8733 spares->array.spare_disks);
8734
8735 dev = spares->devs;
8736 for (i = 0; i < delta_disks; i++) {
8737 struct dl *dl;
8738
8739 if (dev == NULL)
8740 break;
8741 u->new_disks[i] = makedev(dev->disk.major,
8742 dev->disk.minor);
8743 dl = get_disk_super(super, dev->disk.major, dev->disk.minor);
8744 dl->index = mpb->num_disks;
8745 mpb->num_disks++;
8746 dev = dev->next;
8747 }
8748
8749abort:
8750 /* free spares
8751 */
8752 sysfs_free(spares);
8753
8754 dprintf("imsm: reshape update preparation :");
8755 if (i == delta_disks) {
8756 dprintf(" OK\n");
8757 *updatep = u;
8758 return update_memory_size;
8759 }
8760 free(u);
8761 dprintf(" Error\n");
8762
8763 return 0;
8764}
8765
8766/******************************************************************************
8767 * function: imsm_create_metadata_update_for_migration()
8768 * Creates update for IMSM array.
8769 *
8770 ******************************************************************************/
8771static int imsm_create_metadata_update_for_migration(
8772 struct supertype *st,
8773 struct geo_params *geo,
8774 struct imsm_update_reshape_migration **updatep)
8775{
8776 struct intel_super *super = st->sb;
8777 int update_memory_size = 0;
8778 struct imsm_update_reshape_migration *u = NULL;
8779 struct imsm_dev *dev;
8780 int previous_level = -1;
8781
8782 dprintf("imsm_create_metadata_update_for_migration(enter)"
8783 " New Level = %i\n", geo->level);
8784
8785 /* size of all update data without anchor */
8786 update_memory_size = sizeof(struct imsm_update_reshape_migration);
8787
8788 u = calloc(1, update_memory_size);
8789 if (u == NULL) {
8790 dprintf("error: cannot get memory for "
8791 "imsm_create_metadata_update_for_migration\n");
8792 return 0;
8793 }
8794 u->type = update_reshape_migration;
8795 u->subdev = super->current_vol;
8796 u->new_level = geo->level;
8797 u->new_layout = geo->layout;
8798 u->new_raid_disks = u->old_raid_disks = geo->raid_disks;
8799 u->new_disks[0] = -1;
8800 u->new_chunksize = -1;
8801
8802 dev = get_imsm_dev(super, u->subdev);
8803 if (dev) {
8804 struct imsm_map *map;
8805
8806 map = get_imsm_map(dev, 0);
8807 if (map) {
8808 int current_chunk_size =
8809 __le16_to_cpu(map->blocks_per_strip) / 2;
8810
8811 if (geo->chunksize != current_chunk_size) {
8812 u->new_chunksize = geo->chunksize / 1024;
8813 dprintf("imsm: "
8814 "chunk size change from %i to %i\n",
8815 current_chunk_size, u->new_chunksize);
8816 }
8817 previous_level = map->raid_level;
8818 }
8819 }
8820 if ((geo->level == 5) && (previous_level == 0)) {
8821 struct mdinfo *spares = NULL;
8822
8823 u->new_raid_disks++;
8824 spares = get_spares_for_grow(st);
8825 if ((spares == NULL) || (spares->array.spare_disks < 1)) {
8826 free(u);
8827 sysfs_free(spares);
8828 update_memory_size = 0;
8829 dprintf("error: cannot get spare device "
8830 "for requested migration");
8831 return 0;
8832 }
8833 sysfs_free(spares);
8834 }
8835 dprintf("imsm: reshape update preparation : OK\n");
8836 *updatep = u;
8837
8838 return update_memory_size;
8839}
8840
8841static void imsm_update_metadata_locally(struct supertype *st,
8842 void *buf, int len)
8843{
8844 struct metadata_update mu;
8845
8846 mu.buf = buf;
8847 mu.len = len;
8848 mu.space = NULL;
8849 mu.space_list = NULL;
8850 mu.next = NULL;
8851 imsm_prepare_update(st, &mu);
8852 imsm_process_update(st, &mu);
8853
8854 while (mu.space_list) {
8855 void **space = mu.space_list;
8856 mu.space_list = *space;
8857 free(space);
8858 }
8859}
8860
8861/***************************************************************************
8862* Function: imsm_analyze_change
8863* Description: Function analyze change for single volume
8864* and validate if transition is supported
8865* Parameters: Geometry parameters, supertype structure
8866* Returns: Operation type code on success, -1 if fail
8867****************************************************************************/
8868enum imsm_reshape_type imsm_analyze_change(struct supertype *st,
8869 struct geo_params *geo)
8870{
8871 struct mdinfo info;
8872 int change = -1;
8873 int check_devs = 0;
8874 int chunk;
8875 int devNumChange=0;
8876 int layout = -1;
8877
8878 getinfo_super_imsm_volume(st, &info, NULL);
8879 if ((geo->level != info.array.level) &&
8880 (geo->level >= 0) &&
8881 (geo->level != UnSet)) {
8882 switch (info.array.level) {
8883 case 0:
8884 if (geo->level == 5) {
8885 change = CH_MIGRATION;
8886 if (geo->layout != ALGORITHM_LEFT_ASYMMETRIC) {
8887 fprintf(stderr,
8888 Name " Error. Requested Layout "
8889 "not supported (left-asymmetric layout "
8890 "is supported only)!\n");
8891 change = -1;
8892 goto analyse_change_exit;
8893 }
8894 layout = geo->layout;
8895 check_devs = 1;
8896 devNumChange = 1; /* parity disk added */
8897 } else if (geo->level == 10) {
8898 change = CH_TAKEOVER;
8899 check_devs = 1;
8900 devNumChange = 2; /* two mirrors added */
8901 layout = 0x102; /* imsm supported layout */
8902 }
8903 break;
8904 case 1:
8905 case 10:
8906 if (geo->level == 0) {
8907 change = CH_TAKEOVER;
8908 check_devs = 1;
8909 devNumChange = -(geo->raid_disks/2);
8910 layout = 0; /* imsm raid0 layout */
8911 }
8912 break;
8913 }
8914 if (change == -1) {
8915 fprintf(stderr,
8916 Name " Error. Level Migration from %d to %d "
8917 "not supported!\n",
8918 info.array.level, geo->level);
8919 goto analyse_change_exit;
8920 }
8921 } else
8922 geo->level = info.array.level;
8923
8924 if ((geo->layout != info.array.layout)
8925 && ((geo->layout != UnSet) && (geo->layout != -1))) {
8926 change = CH_MIGRATION;
8927 if ((info.array.layout == 0)
8928 && (info.array.level == 5)
8929 && (geo->layout == 5)) {
8930 /* reshape 5 -> 4 */
8931 } else if ((info.array.layout == 5)
8932 && (info.array.level == 5)
8933 && (geo->layout == 0)) {
8934 /* reshape 4 -> 5 */
8935 geo->layout = 0;
8936 geo->level = 5;
8937 } else {
8938 fprintf(stderr,
8939 Name " Error. Layout Migration from %d to %d "
8940 "not supported!\n",
8941 info.array.layout, geo->layout);
8942 change = -1;
8943 goto analyse_change_exit;
8944 }
8945 } else
8946 geo->layout = info.array.layout;
8947
8948 if ((geo->chunksize > 0) && (geo->chunksize != UnSet)
8949 && (geo->chunksize != info.array.chunk_size))
8950 change = CH_MIGRATION;
8951 else
8952 geo->chunksize = info.array.chunk_size;
8953
8954 chunk = geo->chunksize / 1024;
8955 if (!validate_geometry_imsm(st,
8956 geo->level,
8957 layout,
8958 geo->raid_disks + devNumChange,
8959 &chunk,
8960 geo->size,
8961 0, 0, 1))
8962 change = -1;
8963
8964 if (check_devs) {
8965 struct intel_super *super = st->sb;
8966 struct imsm_super *mpb = super->anchor;
8967
8968 if (mpb->num_raid_devs > 1) {
8969 fprintf(stderr,
8970 Name " Error. Cannot perform operation on %s"
8971 "- for this operation it MUST be single "
8972 "array in container\n",
8973 geo->dev_name);
8974 change = -1;
8975 }
8976 }
8977
8978analyse_change_exit:
8979
8980 return change;
8981}
8982
8983int imsm_takeover(struct supertype *st, struct geo_params *geo)
8984{
8985 struct intel_super *super = st->sb;
8986 struct imsm_update_takeover *u;
8987
8988 u = malloc(sizeof(struct imsm_update_takeover));
8989 if (u == NULL)
8990 return 1;
8991
8992 u->type = update_takeover;
8993 u->subarray = super->current_vol;
8994
8995 /* 10->0 transition */
8996 if (geo->level == 0)
8997 u->direction = R10_TO_R0;
8998
8999 /* 0->10 transition */
9000 if (geo->level == 10)
9001 u->direction = R0_TO_R10;
9002
9003 /* update metadata locally */
9004 imsm_update_metadata_locally(st, u,
9005 sizeof(struct imsm_update_takeover));
9006 /* and possibly remotely */
9007 if (st->update_tail)
9008 append_metadata_update(st, u,
9009 sizeof(struct imsm_update_takeover));
9010 else
9011 free(u);
9012
9013 return 0;
9014}
9015
9016static int imsm_reshape_super(struct supertype *st, long long size, int level,
9017 int layout, int chunksize, int raid_disks,
9018 int delta_disks, char *backup, char *dev,
9019 int verbose)
9020{
9021 int ret_val = 1;
9022 struct geo_params geo;
9023
9024 dprintf("imsm: reshape_super called.\n");
9025
9026 memset(&geo, 0, sizeof(struct geo_params));
9027
9028 geo.dev_name = dev;
9029 geo.dev_id = st->devnum;
9030 geo.size = size;
9031 geo.level = level;
9032 geo.layout = layout;
9033 geo.chunksize = chunksize;
9034 geo.raid_disks = raid_disks;
9035 if (delta_disks != UnSet)
9036 geo.raid_disks += delta_disks;
9037
9038 dprintf("\tfor level : %i\n", geo.level);
9039 dprintf("\tfor raid_disks : %i\n", geo.raid_disks);
9040
9041 if (experimental() == 0)
9042 return ret_val;
9043
9044 if (st->container_dev == st->devnum) {
9045 /* On container level we can only increase number of devices. */
9046 dprintf("imsm: info: Container operation\n");
9047 int old_raid_disks = 0;
9048
9049 if (imsm_reshape_is_allowed_on_container(
9050 st, &geo, &old_raid_disks)) {
9051 struct imsm_update_reshape *u = NULL;
9052 int len;
9053
9054 len = imsm_create_metadata_update_for_reshape(
9055 st, &geo, old_raid_disks, &u);
9056
9057 if (len <= 0) {
9058 dprintf("imsm: Cannot prepare update\n");
9059 goto exit_imsm_reshape_super;
9060 }
9061
9062 ret_val = 0;
9063 /* update metadata locally */
9064 imsm_update_metadata_locally(st, u, len);
9065 /* and possibly remotely */
9066 if (st->update_tail)
9067 append_metadata_update(st, u, len);
9068 else
9069 free(u);
9070
9071 } else {
9072 fprintf(stderr, Name ": (imsm) Operation "
9073 "is not allowed on this container\n");
9074 }
9075 } else {
9076 /* On volume level we support following operations
9077 * - takeover: raid10 -> raid0; raid0 -> raid10
9078 * - chunk size migration
9079 * - migration: raid5 -> raid0; raid0 -> raid5
9080 */
9081 struct intel_super *super = st->sb;
9082 struct intel_dev *dev = super->devlist;
9083 int change, devnum;
9084 dprintf("imsm: info: Volume operation\n");
9085 /* find requested device */
9086 while (dev) {
9087 if (imsm_find_array_minor_by_subdev(
9088 dev->index, st->container_dev, &devnum) == 0
9089 && devnum == geo.dev_id)
9090 break;
9091 dev = dev->next;
9092 }
9093 if (dev == NULL) {
9094 fprintf(stderr, Name " Cannot find %s (%i) subarray\n",
9095 geo.dev_name, geo.dev_id);
9096 goto exit_imsm_reshape_super;
9097 }
9098 super->current_vol = dev->index;
9099 change = imsm_analyze_change(st, &geo);
9100 switch (change) {
9101 case CH_TAKEOVER:
9102 ret_val = imsm_takeover(st, &geo);
9103 break;
9104 case CH_MIGRATION: {
9105 struct imsm_update_reshape_migration *u = NULL;
9106 int len =
9107 imsm_create_metadata_update_for_migration(
9108 st, &geo, &u);
9109 if (len < 1) {
9110 dprintf("imsm: "
9111 "Cannot prepare update\n");
9112 break;
9113 }
9114 ret_val = 0;
9115 /* update metadata locally */
9116 imsm_update_metadata_locally(st, u, len);
9117 /* and possibly remotely */
9118 if (st->update_tail)
9119 append_metadata_update(st, u, len);
9120 else
9121 free(u);
9122 }
9123 break;
9124 default:
9125 ret_val = 1;
9126 }
9127 }
9128
9129exit_imsm_reshape_super:
9130 dprintf("imsm: reshape_super Exit code = %i\n", ret_val);
9131 return ret_val;
9132}
9133
9134/*******************************************************************************
9135 * Function: wait_for_reshape_imsm
9136 * Description: Function writes new sync_max value and waits until
9137 * reshape process reach new position
9138 * Parameters:
9139 * sra : general array info
9140 * ndata : number of disks in new array's layout
9141 * Returns:
9142 * 0 : success,
9143 * 1 : there is no reshape in progress,
9144 * -1 : fail
9145 ******************************************************************************/
9146int wait_for_reshape_imsm(struct mdinfo *sra, int ndata)
9147{
9148 int fd = sysfs_get_fd(sra, NULL, "reshape_position");
9149 unsigned long long completed;
9150 /* to_complete : new sync_max position */
9151 unsigned long long to_complete = sra->reshape_progress;
9152 unsigned long long position_to_set = to_complete / ndata;
9153
9154 if (fd < 0) {
9155 dprintf("imsm: wait_for_reshape_imsm() "
9156 "cannot open reshape_position\n");
9157 return 1;
9158 }
9159
9160 if (sysfs_fd_get_ll(fd, &completed) < 0) {
9161 dprintf("imsm: wait_for_reshape_imsm() "
9162 "cannot read reshape_position (no reshape in progres)\n");
9163 close(fd);
9164 return 0;
9165 }
9166
9167 if (completed > to_complete) {
9168 dprintf("imsm: wait_for_reshape_imsm() "
9169 "wrong next position to set %llu (%llu)\n",
9170 to_complete, completed);
9171 close(fd);
9172 return -1;
9173 }
9174 dprintf("Position set: %llu\n", position_to_set);
9175 if (sysfs_set_num(sra, NULL, "sync_max",
9176 position_to_set) != 0) {
9177 dprintf("imsm: wait_for_reshape_imsm() "
9178 "cannot set reshape position to %llu\n",
9179 position_to_set);
9180 close(fd);
9181 return -1;
9182 }
9183
9184 do {
9185 char action[20];
9186 fd_set rfds;
9187 FD_ZERO(&rfds);
9188 FD_SET(fd, &rfds);
9189 select(fd+1, &rfds, NULL, NULL, NULL);
9190 if (sysfs_get_str(sra, NULL, "sync_action",
9191 action, 20) > 0 &&
9192 strncmp(action, "reshape", 7) != 0)
9193 break;
9194 if (sysfs_fd_get_ll(fd, &completed) < 0) {
9195 dprintf("imsm: wait_for_reshape_imsm() "
9196 "cannot read reshape_position (in loop)\n");
9197 close(fd);
9198 return 1;
9199 }
9200 } while (completed < to_complete);
9201 close(fd);
9202 return 0;
9203
9204}
9205
9206/*******************************************************************************
9207 * Function: check_degradation_change
9208 * Description: Check that array hasn't become failed.
9209 * Parameters:
9210 * info : for sysfs access
9211 * sources : source disks descriptors
9212 * degraded: previous degradation level
9213 * Returns:
9214 * degradation level
9215 ******************************************************************************/
9216int check_degradation_change(struct mdinfo *info,
9217 int *sources,
9218 int degraded)
9219{
9220 unsigned long long new_degraded;
9221 sysfs_get_ll(info, NULL, "degraded", &new_degraded);
9222 if (new_degraded != (unsigned long long)degraded) {
9223 /* check each device to ensure it is still working */
9224 struct mdinfo *sd;
9225 new_degraded = 0;
9226 for (sd = info->devs ; sd ; sd = sd->next) {
9227 if (sd->disk.state & (1<<MD_DISK_FAULTY))
9228 continue;
9229 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
9230 char sbuf[20];
9231 if (sysfs_get_str(info,
9232 sd, "state", sbuf, 20) < 0 ||
9233 strstr(sbuf, "faulty") ||
9234 strstr(sbuf, "in_sync") == NULL) {
9235 /* this device is dead */
9236 sd->disk.state = (1<<MD_DISK_FAULTY);
9237 if (sd->disk.raid_disk >= 0 &&
9238 sources[sd->disk.raid_disk] >= 0) {
9239 close(sources[
9240 sd->disk.raid_disk]);
9241 sources[sd->disk.raid_disk] =
9242 -1;
9243 }
9244 new_degraded++;
9245 }
9246 }
9247 }
9248 }
9249
9250 return new_degraded;
9251}
9252
9253/*******************************************************************************
9254 * Function: imsm_manage_reshape
9255 * Description: Function finds array under reshape and it manages reshape
9256 * process. It creates stripes backups (if required) and sets
9257 * checheckpoits.
9258 * Parameters:
9259 * afd : Backup handle (nattive) - not used
9260 * sra : general array info
9261 * reshape : reshape parameters - not used
9262 * st : supertype structure
9263 * blocks : size of critical section [blocks]
9264 * fds : table of source device descriptor
9265 * offsets : start of array (offest per devices)
9266 * dests : not used
9267 * destfd : table of destination device descriptor
9268 * destoffsets : table of destination offsets (per device)
9269 * Returns:
9270 * 1 : success, reshape is done
9271 * 0 : fail
9272 ******************************************************************************/
9273static int imsm_manage_reshape(
9274 int afd, struct mdinfo *sra, struct reshape *reshape,
9275 struct supertype *st, unsigned long backup_blocks,
9276 int *fds, unsigned long long *offsets,
9277 int dests, int *destfd, unsigned long long *destoffsets)
9278{
9279 int ret_val = 0;
9280 struct intel_super *super = st->sb;
9281 struct intel_dev *dv = NULL;
9282 struct imsm_dev *dev = NULL;
9283 struct imsm_map *map_src;
9284 int migr_vol_qan = 0;
9285 int ndata, odata; /* [bytes] */
9286 int chunk; /* [bytes] */
9287 struct migr_record *migr_rec;
9288 char *buf = NULL;
9289 unsigned int buf_size; /* [bytes] */
9290 unsigned long long max_position; /* array size [bytes] */
9291 unsigned long long next_step; /* [blocks]/[bytes] */
9292 unsigned long long old_data_stripe_length;
9293 unsigned long long start_src; /* [bytes] */
9294 unsigned long long start; /* [bytes] */
9295 unsigned long long start_buf_shift; /* [bytes] */
9296 int degraded = 0;
9297 int source_layout = 0;
9298
9299 if (!fds || !offsets || !sra)
9300 goto abort;
9301
9302 /* Find volume during the reshape */
9303 for (dv = super->devlist; dv; dv = dv->next) {
9304 if (dv->dev->vol.migr_type == MIGR_GEN_MIGR
9305 && dv->dev->vol.migr_state == 1) {
9306 dev = dv->dev;
9307 migr_vol_qan++;
9308 }
9309 }
9310 /* Only one volume can migrate at the same time */
9311 if (migr_vol_qan != 1) {
9312 fprintf(stderr, Name " : %s", migr_vol_qan ?
9313 "Number of migrating volumes greater than 1\n" :
9314 "There is no volume during migrationg\n");
9315 goto abort;
9316 }
9317
9318 map_src = get_imsm_map(dev, 1);
9319 if (map_src == NULL)
9320 goto abort;
9321
9322 ndata = imsm_num_data_members(dev, 0);
9323 odata = imsm_num_data_members(dev, 1);
9324
9325 chunk = __le16_to_cpu(map_src->blocks_per_strip) * 512;
9326 old_data_stripe_length = odata * chunk;
9327
9328 migr_rec = super->migr_rec;
9329
9330 /* initialize migration record for start condition */
9331 if (sra->reshape_progress == 0)
9332 init_migr_record_imsm(st, dev, sra);
9333 else {
9334 if (__le32_to_cpu(migr_rec->rec_status) != UNIT_SRC_NORMAL) {
9335 dprintf("imsm: cannot restart migration when data "
9336 "are present in copy area.\n");
9337 goto abort;
9338 }
9339 }
9340
9341 /* size for data */
9342 buf_size = __le32_to_cpu(migr_rec->blocks_per_unit) * 512;
9343 /* extend buffer size for parity disk */
9344 buf_size += __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512;
9345 /* add space for stripe aligment */
9346 buf_size += old_data_stripe_length;
9347 if (posix_memalign((void **)&buf, 4096, buf_size)) {
9348 dprintf("imsm: Cannot allocate checpoint buffer\n");
9349 goto abort;
9350 }
9351
9352 max_position = sra->component_size * ndata;
9353 source_layout = imsm_level_to_layout(map_src->raid_level);
9354
9355 while (__le32_to_cpu(migr_rec->curr_migr_unit) <
9356 __le32_to_cpu(migr_rec->num_migr_units)) {
9357 /* current reshape position [blocks] */
9358 unsigned long long current_position =
9359 __le32_to_cpu(migr_rec->blocks_per_unit)
9360 * __le32_to_cpu(migr_rec->curr_migr_unit);
9361 unsigned long long border;
9362
9363 /* Check that array hasn't become failed.
9364 */
9365 degraded = check_degradation_change(sra, fds, degraded);
9366 if (degraded > 1) {
9367 dprintf("imsm: Abort reshape due to degradation"
9368 " level (%i)\n", degraded);
9369 goto abort;
9370 }
9371
9372 next_step = __le32_to_cpu(migr_rec->blocks_per_unit);
9373
9374 if ((current_position + next_step) > max_position)
9375 next_step = max_position - current_position;
9376
9377 start = current_position * 512;
9378
9379 /* allign reading start to old geometry */
9380 start_buf_shift = start % old_data_stripe_length;
9381 start_src = start - start_buf_shift;
9382
9383 border = (start_src / odata) - (start / ndata);
9384 border /= 512;
9385 if (border <= __le32_to_cpu(migr_rec->dest_depth_per_unit)) {
9386 /* save critical stripes to buf
9387 * start - start address of current unit
9388 * to backup [bytes]
9389 * start_src - start address of current unit
9390 * to backup alligned to source array
9391 * [bytes]
9392 */
9393 unsigned long long next_step_filler = 0;
9394 unsigned long long copy_length = next_step * 512;
9395
9396 /* allign copy area length to stripe in old geometry */
9397 next_step_filler = ((copy_length + start_buf_shift)
9398 % old_data_stripe_length);
9399 if (next_step_filler)
9400 next_step_filler = (old_data_stripe_length
9401 - next_step_filler);
9402 dprintf("save_stripes() parameters: start = %llu,"
9403 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9404 "\tstart_in_buf_shift = %llu,"
9405 "\tnext_step_filler = %llu\n",
9406 start, start_src, copy_length,
9407 start_buf_shift, next_step_filler);
9408
9409 if (save_stripes(fds, offsets, map_src->num_members,
9410 chunk, map_src->raid_level,
9411 source_layout, 0, NULL, start_src,
9412 copy_length +
9413 next_step_filler + start_buf_shift,
9414 buf)) {
9415 dprintf("imsm: Cannot save stripes"
9416 " to buffer\n");
9417 goto abort;
9418 }
9419 /* Convert data to destination format and store it
9420 * in backup general migration area
9421 */
9422 if (save_backup_imsm(st, dev, sra,
9423 buf + start_buf_shift, copy_length)) {
9424 dprintf("imsm: Cannot save stripes to "
9425 "target devices\n");
9426 goto abort;
9427 }
9428 if (save_checkpoint_imsm(st, sra,
9429 UNIT_SRC_IN_CP_AREA)) {
9430 dprintf("imsm: Cannot write checkpoint to "
9431 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9432 goto abort;
9433 }
9434 } else {
9435 /* set next step to use whole border area */
9436 border /= next_step;
9437 if (border > 1)
9438 next_step *= border;
9439 }
9440 /* When data backed up, checkpoint stored,
9441 * kick the kernel to reshape unit of data
9442 */
9443 next_step = next_step + sra->reshape_progress;
9444 /* limit next step to array max position */
9445 if (next_step > max_position)
9446 next_step = max_position;
9447 sysfs_set_num(sra, NULL, "suspend_lo", sra->reshape_progress);
9448 sysfs_set_num(sra, NULL, "suspend_hi", next_step);
9449 sra->reshape_progress = next_step;
9450
9451 /* wait until reshape finish */
9452 if (wait_for_reshape_imsm(sra, ndata) < 0) {
9453 dprintf("wait_for_reshape_imsm returned error!\n");
9454 goto abort;
9455 }
9456
9457 if (save_checkpoint_imsm(st, sra, UNIT_SRC_NORMAL) == 1) {
9458 /* ignore error == 2, this can mean end of reshape here
9459 */
9460 dprintf("imsm: Cannot write checkpoint to "
9461 "migration record (UNIT_SRC_NORMAL)\n");
9462 goto abort;
9463 }
9464
9465 }
9466
9467 /* return '1' if done */
9468 ret_val = 1;
9469abort:
9470 free(buf);
9471 abort_reshape(sra);
9472
9473 return ret_val;
9474}
9475#endif /* MDASSEMBLE */
9476
9477struct superswitch super_imsm = {
9478#ifndef MDASSEMBLE
9479 .examine_super = examine_super_imsm,
9480 .brief_examine_super = brief_examine_super_imsm,
9481 .brief_examine_subarrays = brief_examine_subarrays_imsm,
9482 .export_examine_super = export_examine_super_imsm,
9483 .detail_super = detail_super_imsm,
9484 .brief_detail_super = brief_detail_super_imsm,
9485 .write_init_super = write_init_super_imsm,
9486 .validate_geometry = validate_geometry_imsm,
9487 .add_to_super = add_to_super_imsm,
9488 .remove_from_super = remove_from_super_imsm,
9489 .detail_platform = detail_platform_imsm,
9490 .kill_subarray = kill_subarray_imsm,
9491 .update_subarray = update_subarray_imsm,
9492 .load_container = load_container_imsm,
9493 .default_geometry = default_geometry_imsm,
9494 .get_disk_controller_domain = imsm_get_disk_controller_domain,
9495 .reshape_super = imsm_reshape_super,
9496 .manage_reshape = imsm_manage_reshape,
9497 .recover_backup = recover_backup_imsm,
9498#endif
9499 .match_home = match_home_imsm,
9500 .uuid_from_super= uuid_from_super_imsm,
9501 .getinfo_super = getinfo_super_imsm,
9502 .getinfo_super_disks = getinfo_super_disks_imsm,
9503 .update_super = update_super_imsm,
9504
9505 .avail_size = avail_size_imsm,
9506 .min_acceptable_spare_size = min_acceptable_spare_size_imsm,
9507
9508 .compare_super = compare_super_imsm,
9509
9510 .load_super = load_super_imsm,
9511 .init_super = init_super_imsm,
9512 .store_super = store_super_imsm,
9513 .free_super = free_super_imsm,
9514 .match_metadata_desc = match_metadata_desc_imsm,
9515 .container_content = container_content_imsm,
9516
9517
9518 .external = 1,
9519 .name = "imsm",
9520
9521#ifndef MDASSEMBLE
9522/* for mdmon */
9523 .open_new = imsm_open_new,
9524 .set_array_state= imsm_set_array_state,
9525 .set_disk = imsm_set_disk,
9526 .sync_metadata = imsm_sync_metadata,
9527 .activate_spare = imsm_activate_spare,
9528 .process_update = imsm_process_update,
9529 .prepare_update = imsm_prepare_update,
9530#endif /* MDASSEMBLE */
9531};