]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-intel.c
projects / thirdparty / mdadm.git / blob
? search:


eaf5b0b8e6f979051ce48425a8222636f686eca1
[thirdparty/mdadm.git] / super-intel.c
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 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
53
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56 #define SECT_PER_MB_SHIFT 11
57
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
60 struct imsm_disk {
61 __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 __u32 status; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
71 };
72
73 /* RAID map configuration infos. */
74 struct imsm_map {
75 __u32 pba_of_lba0; /* start address of partition */
76 __u32 blocks_per_member;/* blocks per member */
77 __u32 num_data_stripes; /* number of data stripes */
78 __u16 blocks_per_strip;
79 __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
84 __u8 raid_level;
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members; /* number of member disks */
89 __u8 num_domains; /* number of parity domains */
90 __u8 failed_disk_num; /* valid only when state is degraded */
91 __u8 ddf;
92 __u32 filler[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
96 */
97 } __attribute__ ((packed));
98
99 struct imsm_vol {
100 __u32 curr_migr_unit;
101 __u32 checkpoint_id; /* id to access curr_migr_unit */
102 __u8 migr_state; /* Normal or Migrating */
103 #define MIGR_INIT 0
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type; /* Initializing, Rebuilding, ... */
110 __u8 dirty;
111 __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors; /* number of mismatches */
113 __u16 bad_blocks; /* number of bad blocks during verify */
114 __u32 filler[4];
115 struct imsm_map map[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed));
118
119 struct imsm_dev {
120 __u8 volume[MAX_RAID_SERIAL_LEN];
121 __u32 size_low;
122 __u32 size_high;
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status; /* Persistent RaidDev status */
137 __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
138 __u8 migr_priority;
139 __u8 num_sub_vols;
140 __u8 tid;
141 __u8 cng_master_disk;
142 __u16 cache_policy;
143 __u8 cng_state;
144 __u8 cng_sub_state;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler[IMSM_DEV_FILLERS];
147 struct imsm_vol vol;
148 } __attribute__ ((packed));
149
150 struct imsm_super {
151 __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
152 __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size; /* 0x30 - 0x33 in bytes */
157 __u32 attributes; /* 0x34 - 0x37 */
158 __u8 num_disks; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos; /* 0x3A */
161 __u8 fill[1]; /* 0x3B */
162 __u32 cache_size; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed));
172
173 #define BBM_LOG_MAX_ENTRIES 254
174
175 struct bbm_log_entry {
176 __u64 defective_block_start;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset;
179 __u16 remapped_marked_count;
180 __u16 disk_ordinal;
181 } __attribute__ ((__packed__));
182
183 struct bbm_log {
184 __u32 signature; /* 0xABADB10C */
185 __u32 entry_count;
186 __u32 reserved_spare_block_count; /* 0 */
187 __u32 reserved; /* 0xFFFF */
188 __u64 first_spare_lba;
189 struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
190 } __attribute__ ((__packed__));
191
192
193 #ifndef MDASSEMBLE
194 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
195 #endif
196
197 static __u8 migr_type(struct imsm_dev *dev)
198 {
199 if (dev->vol.migr_type == MIGR_VERIFY &&
200 dev->status & DEV_VERIFY_AND_FIX)
201 return MIGR_REPAIR;
202 else
203 return dev->vol.migr_type;
204 }
205
206 static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
207 {
208 /* for compatibility with older oroms convert MIGR_REPAIR, into
209 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
210 */
211 if (migr_type == MIGR_REPAIR) {
212 dev->vol.migr_type = MIGR_VERIFY;
213 dev->status |= DEV_VERIFY_AND_FIX;
214 } else {
215 dev->vol.migr_type = migr_type;
216 dev->status &= ~DEV_VERIFY_AND_FIX;
217 }
218 }
219
220 static unsigned int sector_count(__u32 bytes)
221 {
222 return ((bytes + (512-1)) & (~(512-1))) / 512;
223 }
224
225 static unsigned int mpb_sectors(struct imsm_super *mpb)
226 {
227 return sector_count(__le32_to_cpu(mpb->mpb_size));
228 }
229
230 struct intel_dev {
231 struct imsm_dev *dev;
232 struct intel_dev *next;
233 int index;
234 };
235
236 /* internal representation of IMSM metadata */
237 struct intel_super {
238 union {
239 void *buf; /* O_DIRECT buffer for reading/writing metadata */
240 struct imsm_super *anchor; /* immovable parameters */
241 };
242 size_t len; /* size of the 'buf' allocation */
243 void *next_buf; /* for realloc'ing buf from the manager */
244 size_t next_len;
245 int updates_pending; /* count of pending updates for mdmon */
246 int creating_imsm; /* flag to indicate container creation */
247 int current_vol; /* index of raid device undergoing creation */
248 __u32 create_offset; /* common start for 'current_vol' */
249 __u32 random; /* random data for seeding new family numbers */
250 struct intel_dev *devlist;
251 struct dl {
252 struct dl *next;
253 int index;
254 __u8 serial[MAX_RAID_SERIAL_LEN];
255 int major, minor;
256 char *devname;
257 struct imsm_disk disk;
258 int fd;
259 int extent_cnt;
260 struct extent *e; /* for determining freespace @ create */
261 int raiddisk; /* slot to fill in autolayout */
262 } *disks;
263 struct dl *add; /* list of disks to add while mdmon active */
264 struct dl *missing; /* disks removed while we weren't looking */
265 struct bbm_log *bbm_log;
266 const char *hba; /* device path of the raid controller for this metadata */
267 const struct imsm_orom *orom; /* platform firmware support */
268 struct intel_super *next; /* (temp) list for disambiguating family_num */
269 };
270
271 struct intel_disk {
272 struct imsm_disk disk;
273 #define IMSM_UNKNOWN_OWNER (-1)
274 int owner;
275 struct intel_disk *next;
276 };
277
278 struct extent {
279 unsigned long long start, size;
280 };
281
282 /* definition of messages passed to imsm_process_update */
283 enum imsm_update_type {
284 update_activate_spare,
285 update_create_array,
286 update_add_disk,
287 };
288
289 struct imsm_update_activate_spare {
290 enum imsm_update_type type;
291 struct dl *dl;
292 int slot;
293 int array;
294 struct imsm_update_activate_spare *next;
295 };
296
297 struct disk_info {
298 __u8 serial[MAX_RAID_SERIAL_LEN];
299 };
300
301 struct imsm_update_create_array {
302 enum imsm_update_type type;
303 int dev_idx;
304 struct imsm_dev dev;
305 };
306
307 struct imsm_update_add_disk {
308 enum imsm_update_type type;
309 };
310
311 static struct supertype *match_metadata_desc_imsm(char *arg)
312 {
313 struct supertype *st;
314
315 if (strcmp(arg, "imsm") != 0 &&
316 strcmp(arg, "default") != 0
317 )
318 return NULL;
319
320 st = malloc(sizeof(*st));
321 memset(st, 0, sizeof(*st));
322 st->ss = &super_imsm;
323 st->max_devs = IMSM_MAX_DEVICES;
324 st->minor_version = 0;
325 st->sb = NULL;
326 return st;
327 }
328
329 #ifndef MDASSEMBLE
330 static __u8 *get_imsm_version(struct imsm_super *mpb)
331 {
332 return &mpb->sig[MPB_SIG_LEN];
333 }
334 #endif
335
336 /* retrieve a disk directly from the anchor when the anchor is known to be
337 * up-to-date, currently only at load time
338 */
339 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
340 {
341 if (index >= mpb->num_disks)
342 return NULL;
343 return &mpb->disk[index];
344 }
345
346 #ifndef MDASSEMBLE
347 /* retrieve a disk from the parsed metadata */
348 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
349 {
350 struct dl *d;
351
352 for (d = super->disks; d; d = d->next)
353 if (d->index == index)
354 return &d->disk;
355
356 return NULL;
357 }
358 #endif
359
360 /* generate a checksum directly from the anchor when the anchor is known to be
361 * up-to-date, currently only at load or write_super after coalescing
362 */
363 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
364 {
365 __u32 end = mpb->mpb_size / sizeof(end);
366 __u32 *p = (__u32 *) mpb;
367 __u32 sum = 0;
368
369 while (end--) {
370 sum += __le32_to_cpu(*p);
371 p++;
372 }
373
374 return sum - __le32_to_cpu(mpb->check_sum);
375 }
376
377 static size_t sizeof_imsm_map(struct imsm_map *map)
378 {
379 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
380 }
381
382 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
383 {
384 struct imsm_map *map = &dev->vol.map[0];
385
386 if (second_map && !dev->vol.migr_state)
387 return NULL;
388 else if (second_map) {
389 void *ptr = map;
390
391 return ptr + sizeof_imsm_map(map);
392 } else
393 return map;
394
395 }
396
397 /* return the size of the device.
398 * migr_state increases the returned size if map[0] were to be duplicated
399 */
400 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
401 {
402 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
403 sizeof_imsm_map(get_imsm_map(dev, 0));
404
405 /* migrating means an additional map */
406 if (dev->vol.migr_state)
407 size += sizeof_imsm_map(get_imsm_map(dev, 1));
408 else if (migr_state)
409 size += sizeof_imsm_map(get_imsm_map(dev, 0));
410
411 return size;
412 }
413
414 #ifndef MDASSEMBLE
415 /* retrieve disk serial number list from a metadata update */
416 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
417 {
418 void *u = update;
419 struct disk_info *inf;
420
421 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
422 sizeof_imsm_dev(&update->dev, 0);
423
424 return inf;
425 }
426 #endif
427
428 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
429 {
430 int offset;
431 int i;
432 void *_mpb = mpb;
433
434 if (index >= mpb->num_raid_devs)
435 return NULL;
436
437 /* devices start after all disks */
438 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
439
440 for (i = 0; i <= index; i++)
441 if (i == index)
442 return _mpb + offset;
443 else
444 offset += sizeof_imsm_dev(_mpb + offset, 0);
445
446 return NULL;
447 }
448
449 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
450 {
451 struct intel_dev *dv;
452
453 if (index >= super->anchor->num_raid_devs)
454 return NULL;
455 for (dv = super->devlist; dv; dv = dv->next)
456 if (dv->index == index)
457 return dv->dev;
458 return NULL;
459 }
460
461 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, int slot)
462 {
463 struct imsm_map *map;
464
465 if (dev->vol.migr_state)
466 map = get_imsm_map(dev, 1);
467 else
468 map = get_imsm_map(dev, 0);
469
470 /* top byte identifies disk under rebuild */
471 return __le32_to_cpu(map->disk_ord_tbl[slot]);
472 }
473
474 #define ord_to_idx(ord) (((ord) << 8) >> 8)
475 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot)
476 {
477 __u32 ord = get_imsm_ord_tbl_ent(dev, slot);
478
479 return ord_to_idx(ord);
480 }
481
482 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
483 {
484 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
485 }
486
487 static int get_imsm_disk_slot(struct imsm_map *map, int idx)
488 {
489 int slot;
490 __u32 ord;
491
492 for (slot = 0; slot < map->num_members; slot++) {
493 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
494 if (ord_to_idx(ord) == idx)
495 return slot;
496 }
497
498 return -1;
499 }
500
501 static int get_imsm_raid_level(struct imsm_map *map)
502 {
503 if (map->raid_level == 1) {
504 if (map->num_members == 2)
505 return 1;
506 else
507 return 10;
508 }
509
510 return map->raid_level;
511 }
512
513 static int cmp_extent(const void *av, const void *bv)
514 {
515 const struct extent *a = av;
516 const struct extent *b = bv;
517 if (a->start < b->start)
518 return -1;
519 if (a->start > b->start)
520 return 1;
521 return 0;
522 }
523
524 static int count_memberships(struct dl *dl, struct intel_super *super)
525 {
526 int memberships = 0;
527 int i;
528
529 for (i = 0; i < super->anchor->num_raid_devs; i++) {
530 struct imsm_dev *dev = get_imsm_dev(super, i);
531 struct imsm_map *map = get_imsm_map(dev, 0);
532
533 if (get_imsm_disk_slot(map, dl->index) >= 0)
534 memberships++;
535 }
536
537 return memberships;
538 }
539
540 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
541 {
542 /* find a list of used extents on the given physical device */
543 struct extent *rv, *e;
544 int i;
545 int memberships = count_memberships(dl, super);
546 __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
547
548 rv = malloc(sizeof(struct extent) * (memberships + 1));
549 if (!rv)
550 return NULL;
551 e = rv;
552
553 for (i = 0; i < super->anchor->num_raid_devs; i++) {
554 struct imsm_dev *dev = get_imsm_dev(super, i);
555 struct imsm_map *map = get_imsm_map(dev, 0);
556
557 if (get_imsm_disk_slot(map, dl->index) >= 0) {
558 e->start = __le32_to_cpu(map->pba_of_lba0);
559 e->size = __le32_to_cpu(map->blocks_per_member);
560 e++;
561 }
562 }
563 qsort(rv, memberships, sizeof(*rv), cmp_extent);
564
565 /* determine the start of the metadata
566 * when no raid devices are defined use the default
567 * ...otherwise allow the metadata to truncate the value
568 * as is the case with older versions of imsm
569 */
570 if (memberships) {
571 struct extent *last = &rv[memberships - 1];
572 __u32 remainder;
573
574 remainder = __le32_to_cpu(dl->disk.total_blocks) -
575 (last->start + last->size);
576 /* round down to 1k block to satisfy precision of the kernel
577 * 'size' interface
578 */
579 remainder &= ~1UL;
580 /* make sure remainder is still sane */
581 if (remainder < ROUND_UP(super->len, 512) >> 9)
582 remainder = ROUND_UP(super->len, 512) >> 9;
583 if (reservation > remainder)
584 reservation = remainder;
585 }
586 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
587 e->size = 0;
588 return rv;
589 }
590
591 /* try to determine how much space is reserved for metadata from
592 * the last get_extents() entry, otherwise fallback to the
593 * default
594 */
595 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
596 {
597 struct extent *e;
598 int i;
599 __u32 rv;
600
601 /* for spares just return a minimal reservation which will grow
602 * once the spare is picked up by an array
603 */
604 if (dl->index == -1)
605 return MPB_SECTOR_CNT;
606
607 e = get_extents(super, dl);
608 if (!e)
609 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
610
611 /* scroll to last entry */
612 for (i = 0; e[i].size; i++)
613 continue;
614
615 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
616
617 free(e);
618
619 return rv;
620 }
621
622 #ifndef MDASSEMBLE
623 static int is_spare(struct imsm_disk *disk)
624 {
625 return (disk->status & SPARE_DISK) == SPARE_DISK;
626 }
627
628 static int is_configured(struct imsm_disk *disk)
629 {
630 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
631 }
632
633 static int is_failed(struct imsm_disk *disk)
634 {
635 return (disk->status & FAILED_DISK) == FAILED_DISK;
636 }
637
638 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
639 {
640 __u64 sz;
641 int slot;
642 struct imsm_map *map = get_imsm_map(dev, 0);
643 __u32 ord;
644
645 printf("\n");
646 printf("[%.16s]:\n", dev->volume);
647 printf(" UUID : %s\n", uuid);
648 printf(" RAID Level : %d\n", get_imsm_raid_level(map));
649 printf(" Members : %d\n", map->num_members);
650 slot = get_imsm_disk_slot(map, disk_idx);
651 if (slot >= 0) {
652 ord = get_imsm_ord_tbl_ent(dev, slot);
653 printf(" This Slot : %d%s\n", slot,
654 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
655 } else
656 printf(" This Slot : ?\n");
657 sz = __le32_to_cpu(dev->size_high);
658 sz <<= 32;
659 sz += __le32_to_cpu(dev->size_low);
660 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
661 human_size(sz * 512));
662 sz = __le32_to_cpu(map->blocks_per_member);
663 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
664 human_size(sz * 512));
665 printf(" Sector Offset : %u\n",
666 __le32_to_cpu(map->pba_of_lba0));
667 printf(" Num Stripes : %u\n",
668 __le32_to_cpu(map->num_data_stripes));
669 printf(" Chunk Size : %u KiB\n",
670 __le16_to_cpu(map->blocks_per_strip) / 2);
671 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
672 printf(" Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle\n");
673 if (dev->vol.migr_state) {
674 if (migr_type(dev) == MIGR_INIT)
675 printf(": initializing\n");
676 else if (migr_type(dev) == MIGR_REBUILD)
677 printf(": rebuilding\n");
678 else if (migr_type(dev) == MIGR_VERIFY)
679 printf(": check\n");
680 else if (migr_type(dev) == MIGR_GEN_MIGR)
681 printf(": general migration\n");
682 else if (migr_type(dev) == MIGR_STATE_CHANGE)
683 printf(": state change\n");
684 else if (migr_type(dev) == MIGR_REPAIR)
685 printf(": repair\n");
686 else
687 printf(": <unknown:%d>\n", migr_type(dev));
688 }
689 printf(" Map State : %s", map_state_str[map->map_state]);
690 if (dev->vol.migr_state) {
691 struct imsm_map *map = get_imsm_map(dev, 1);
692 printf(" <-- %s", map_state_str[map->map_state]);
693 }
694 printf("\n");
695 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
696 }
697
698 static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved)
699 {
700 struct imsm_disk *disk = __get_imsm_disk(mpb, index);
701 char str[MAX_RAID_SERIAL_LEN + 1];
702 __u64 sz;
703
704 if (index < 0)
705 return;
706
707 printf("\n");
708 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
709 printf(" Disk%02d Serial : %s\n", index, str);
710 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
711 is_configured(disk) ? " active" : "",
712 is_failed(disk) ? " failed" : "");
713 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
714 sz = __le32_to_cpu(disk->total_blocks) - reserved;
715 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
716 human_size(sz * 512));
717 }
718
719 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info);
720
721 static void examine_super_imsm(struct supertype *st, char *homehost)
722 {
723 struct intel_super *super = st->sb;
724 struct imsm_super *mpb = super->anchor;
725 char str[MAX_SIGNATURE_LENGTH];
726 int i;
727 struct mdinfo info;
728 char nbuf[64];
729 __u32 sum;
730 __u32 reserved = imsm_reserved_sectors(super, super->disks);
731
732
733 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
734 printf(" Magic : %s\n", str);
735 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
736 printf(" Version : %s\n", get_imsm_version(mpb));
737 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
738 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
739 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
740 getinfo_super_imsm(st, &info);
741 fname_from_uuid(st, &info, nbuf, ':');
742 printf(" UUID : %s\n", nbuf + 5);
743 sum = __le32_to_cpu(mpb->check_sum);
744 printf(" Checksum : %08x %s\n", sum,
745 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
746 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
747 printf(" Disks : %d\n", mpb->num_disks);
748 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
749 print_imsm_disk(mpb, super->disks->index, reserved);
750 if (super->bbm_log) {
751 struct bbm_log *log = super->bbm_log;
752
753 printf("\n");
754 printf("Bad Block Management Log:\n");
755 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
756 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
757 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
758 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
759 printf(" First Spare : %llx\n",
760 (unsigned long long) __le64_to_cpu(log->first_spare_lba));
761 }
762 for (i = 0; i < mpb->num_raid_devs; i++) {
763 struct mdinfo info;
764 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
765
766 super->current_vol = i;
767 getinfo_super_imsm(st, &info);
768 fname_from_uuid(st, &info, nbuf, ':');
769 print_imsm_dev(dev, nbuf + 5, super->disks->index);
770 }
771 for (i = 0; i < mpb->num_disks; i++) {
772 if (i == super->disks->index)
773 continue;
774 print_imsm_disk(mpb, i, reserved);
775 }
776 }
777
778 static void brief_examine_super_imsm(struct supertype *st, int verbose)
779 {
780 /* We just write a generic IMSM ARRAY entry */
781 struct mdinfo info;
782 char nbuf[64];
783 struct intel_super *super = st->sb;
784
785 if (!super->anchor->num_raid_devs) {
786 printf("ARRAY metadata=imsm\n");
787 return;
788 }
789
790 getinfo_super_imsm(st, &info);
791 fname_from_uuid(st, &info, nbuf, ':');
792 printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
793 }
794
795 static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
796 {
797 /* We just write a generic IMSM ARRAY entry */
798 struct mdinfo info;
799 char nbuf[64];
800 char nbuf1[64];
801 struct intel_super *super = st->sb;
802 int i;
803
804 if (!super->anchor->num_raid_devs)
805 return;
806
807 getinfo_super_imsm(st, &info);
808 fname_from_uuid(st, &info, nbuf, ':');
809 for (i = 0; i < super->anchor->num_raid_devs; i++) {
810 struct imsm_dev *dev = get_imsm_dev(super, i);
811
812 super->current_vol = i;
813 getinfo_super_imsm(st, &info);
814 fname_from_uuid(st, &info, nbuf1, ':');
815 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
816 dev->volume, nbuf + 5, i, nbuf1 + 5);
817 }
818 }
819
820 static void export_examine_super_imsm(struct supertype *st)
821 {
822 struct intel_super *super = st->sb;
823 struct imsm_super *mpb = super->anchor;
824 struct mdinfo info;
825 char nbuf[64];
826
827 getinfo_super_imsm(st, &info);
828 fname_from_uuid(st, &info, nbuf, ':');
829 printf("MD_METADATA=imsm\n");
830 printf("MD_LEVEL=container\n");
831 printf("MD_UUID=%s\n", nbuf+5);
832 printf("MD_DEVICES=%u\n", mpb->num_disks);
833 }
834
835 static void detail_super_imsm(struct supertype *st, char *homehost)
836 {
837 struct mdinfo info;
838 char nbuf[64];
839
840 getinfo_super_imsm(st, &info);
841 fname_from_uuid(st, &info, nbuf, ':');
842 printf("\n UUID : %s\n", nbuf + 5);
843 }
844
845 static void brief_detail_super_imsm(struct supertype *st)
846 {
847 struct mdinfo info;
848 char nbuf[64];
849 getinfo_super_imsm(st, &info);
850 fname_from_uuid(st, &info, nbuf, ':');
851 printf(" UUID=%s", nbuf + 5);
852 }
853
854 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
855 static void fd2devname(int fd, char *name);
856
857 static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
858 {
859 /* dump an unsorted list of devices attached to ahci, as well as
860 * non-connected ports
861 */
862 int hba_len = strlen(hba_path) + 1;
863 struct dirent *ent;
864 DIR *dir;
865 char *path = NULL;
866 int err = 0;
867 unsigned long port_mask = (1 << port_count) - 1;
868
869 if (port_count > sizeof(port_mask) * 8) {
870 if (verbose)
871 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
872 return 2;
873 }
874
875 /* scroll through /sys/dev/block looking for devices attached to
876 * this hba
877 */
878 dir = opendir("/sys/dev/block");
879 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
880 int fd;
881 char model[64];
882 char vendor[64];
883 char buf[1024];
884 int major, minor;
885 char *device;
886 char *c;
887 int port;
888 int type;
889
890 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
891 continue;
892 path = devt_to_devpath(makedev(major, minor));
893 if (!path)
894 continue;
895 if (!path_attached_to_hba(path, hba_path)) {
896 free(path);
897 path = NULL;
898 continue;
899 }
900
901 /* retrieve the scsi device type */
902 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
903 if (verbose)
904 fprintf(stderr, Name ": failed to allocate 'device'\n");
905 err = 2;
906 break;
907 }
908 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
909 if (load_sys(device, buf) != 0) {
910 if (verbose)
911 fprintf(stderr, Name ": failed to read device type for %s\n",
912 path);
913 err = 2;
914 free(device);
915 break;
916 }
917 type = strtoul(buf, NULL, 10);
918
919 /* if it's not a disk print the vendor and model */
920 if (!(type == 0 || type == 7 || type == 14)) {
921 vendor[0] = '\0';
922 model[0] = '\0';
923 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
924 if (load_sys(device, buf) == 0) {
925 strncpy(vendor, buf, sizeof(vendor));
926 vendor[sizeof(vendor) - 1] = '\0';
927 c = (char *) &vendor[sizeof(vendor) - 1];
928 while (isspace(*c) || *c == '\0')
929 *c-- = '\0';
930
931 }
932 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
933 if (load_sys(device, buf) == 0) {
934 strncpy(model, buf, sizeof(model));
935 model[sizeof(model) - 1] = '\0';
936 c = (char *) &model[sizeof(model) - 1];
937 while (isspace(*c) || *c == '\0')
938 *c-- = '\0';
939 }
940
941 if (vendor[0] && model[0])
942 sprintf(buf, "%.64s %.64s", vendor, model);
943 else
944 switch (type) { /* numbers from hald/linux/device.c */
945 case 1: sprintf(buf, "tape"); break;
946 case 2: sprintf(buf, "printer"); break;
947 case 3: sprintf(buf, "processor"); break;
948 case 4:
949 case 5: sprintf(buf, "cdrom"); break;
950 case 6: sprintf(buf, "scanner"); break;
951 case 8: sprintf(buf, "media_changer"); break;
952 case 9: sprintf(buf, "comm"); break;
953 case 12: sprintf(buf, "raid"); break;
954 default: sprintf(buf, "unknown");
955 }
956 } else
957 buf[0] = '\0';
958 free(device);
959
960 /* chop device path to 'host%d' and calculate the port number */
961 c = strchr(&path[hba_len], '/');
962 *c = '\0';
963 if (sscanf(&path[hba_len], "host%d", &port) == 1)
964 port -= host_base;
965 else {
966 if (verbose) {
967 *c = '/'; /* repair the full string */
968 fprintf(stderr, Name ": failed to determine port number for %s\n",
969 path);
970 }
971 err = 2;
972 break;
973 }
974
975 /* mark this port as used */
976 port_mask &= ~(1 << port);
977
978 /* print out the device information */
979 if (buf[0]) {
980 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
981 continue;
982 }
983
984 fd = dev_open(ent->d_name, O_RDONLY);
985 if (fd < 0)
986 printf(" Port%d : - disk info unavailable -\n", port);
987 else {
988 fd2devname(fd, buf);
989 printf(" Port%d : %s", port, buf);
990 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
991 printf(" (%s)\n", buf);
992 else
993 printf("()\n");
994 }
995 close(fd);
996 free(path);
997 path = NULL;
998 }
999 if (path)
1000 free(path);
1001 if (dir)
1002 closedir(dir);
1003 if (err == 0) {
1004 int i;
1005
1006 for (i = 0; i < port_count; i++)
1007 if (port_mask & (1 << i))
1008 printf(" Port%d : - no device attached -\n", i);
1009 }
1010
1011 return err;
1012 }
1013
1014 static int detail_platform_imsm(int verbose, int enumerate_only)
1015 {
1016 /* There are two components to imsm platform support, the ahci SATA
1017 * controller and the option-rom. To find the SATA controller we
1018 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1019 * controller with the Intel vendor id is present. This approach
1020 * allows mdadm to leverage the kernel's ahci detection logic, with the
1021 * caveat that if ahci.ko is not loaded mdadm will not be able to
1022 * detect platform raid capabilities. The option-rom resides in a
1023 * platform "Adapter ROM". We scan for its signature to retrieve the
1024 * platform capabilities. If raid support is disabled in the BIOS the
1025 * option-rom capability structure will not be available.
1026 */
1027 const struct imsm_orom *orom;
1028 struct sys_dev *list, *hba;
1029 DIR *dir;
1030 struct dirent *ent;
1031 const char *hba_path;
1032 int host_base = 0;
1033 int port_count = 0;
1034
1035 if (enumerate_only) {
1036 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1037 return 0;
1038 return 2;
1039 }
1040
1041 list = find_driver_devices("pci", "ahci");
1042 for (hba = list; hba; hba = hba->next)
1043 if (devpath_to_vendor(hba->path) == 0x8086)
1044 break;
1045
1046 if (!hba) {
1047 if (verbose)
1048 fprintf(stderr, Name ": unable to find active ahci controller\n");
1049 free_sys_dev(&list);
1050 return 2;
1051 } else if (verbose)
1052 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
1053 hba_path = hba->path;
1054 hba->path = NULL;
1055 free_sys_dev(&list);
1056
1057 orom = find_imsm_orom();
1058 if (!orom) {
1059 if (verbose)
1060 fprintf(stderr, Name ": imsm option-rom not found\n");
1061 return 2;
1062 }
1063
1064 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1065 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
1066 orom->hotfix_ver, orom->build);
1067 printf(" RAID Levels :%s%s%s%s%s\n",
1068 imsm_orom_has_raid0(orom) ? " raid0" : "",
1069 imsm_orom_has_raid1(orom) ? " raid1" : "",
1070 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
1071 imsm_orom_has_raid10(orom) ? " raid10" : "",
1072 imsm_orom_has_raid5(orom) ? " raid5" : "");
1073 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1074 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
1075 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
1076 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
1077 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
1078 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
1079 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
1080 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
1081 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
1082 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
1083 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
1084 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
1085 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
1086 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
1087 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
1088 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
1089 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1090 printf(" Max Disks : %d\n", orom->tds);
1091 printf(" Max Volumes : %d\n", orom->vpa);
1092 printf(" I/O Controller : %s\n", hba_path);
1093
1094 /* find the smallest scsi host number to determine a port number base */
1095 dir = opendir(hba_path);
1096 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1097 int host;
1098
1099 if (sscanf(ent->d_name, "host%d", &host) != 1)
1100 continue;
1101 if (port_count == 0)
1102 host_base = host;
1103 else if (host < host_base)
1104 host_base = host;
1105
1106 if (host + 1 > port_count + host_base)
1107 port_count = host + 1 - host_base;
1108
1109 }
1110 if (dir)
1111 closedir(dir);
1112
1113 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1114 host_base, verbose) != 0) {
1115 if (verbose)
1116 fprintf(stderr, Name ": failed to enumerate ports\n");
1117 return 2;
1118 }
1119
1120 return 0;
1121 }
1122 #endif
1123
1124 static int match_home_imsm(struct supertype *st, char *homehost)
1125 {
1126 /* the imsm metadata format does not specify any host
1127 * identification information. We return -1 since we can never
1128 * confirm nor deny whether a given array is "meant" for this
1129 * host. We rely on compare_super and the 'family_num' fields to
1130 * exclude member disks that do not belong, and we rely on
1131 * mdadm.conf to specify the arrays that should be assembled.
1132 * Auto-assembly may still pick up "foreign" arrays.
1133 */
1134
1135 return -1;
1136 }
1137
1138 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1139 {
1140 /* The uuid returned here is used for:
1141 * uuid to put into bitmap file (Create, Grow)
1142 * uuid for backup header when saving critical section (Grow)
1143 * comparing uuids when re-adding a device into an array
1144 * In these cases the uuid required is that of the data-array,
1145 * not the device-set.
1146 * uuid to recognise same set when adding a missing device back
1147 * to an array. This is a uuid for the device-set.
1148 *
1149 * For each of these we can make do with a truncated
1150 * or hashed uuid rather than the original, as long as
1151 * everyone agrees.
1152 * In each case the uuid required is that of the data-array,
1153 * not the device-set.
1154 */
1155 /* imsm does not track uuid's so we synthesis one using sha1 on
1156 * - The signature (Which is constant for all imsm array, but no matter)
1157 * - the orig_family_num of the container
1158 * - the index number of the volume
1159 * - the 'serial' number of the volume.
1160 * Hopefully these are all constant.
1161 */
1162 struct intel_super *super = st->sb;
1163
1164 char buf[20];
1165 struct sha1_ctx ctx;
1166 struct imsm_dev *dev = NULL;
1167 __u32 family_num;
1168
1169 /* some mdadm versions failed to set ->orig_family_num, in which
1170 * case fall back to ->family_num. orig_family_num will be
1171 * fixed up with the first metadata update.
1172 */
1173 family_num = super->anchor->orig_family_num;
1174 if (family_num == 0)
1175 family_num = super->anchor->family_num;
1176 sha1_init_ctx(&ctx);
1177 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1178 sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
1179 if (super->current_vol >= 0)
1180 dev = get_imsm_dev(super, super->current_vol);
1181 if (dev) {
1182 __u32 vol = super->current_vol;
1183 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1184 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1185 }
1186 sha1_finish_ctx(&ctx, buf);
1187 memcpy(uuid, buf, 4*4);
1188 }
1189
1190 #if 0
1191 static void
1192 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1193 {
1194 __u8 *v = get_imsm_version(mpb);
1195 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1196 char major[] = { 0, 0, 0 };
1197 char minor[] = { 0 ,0, 0 };
1198 char patch[] = { 0, 0, 0 };
1199 char *ver_parse[] = { major, minor, patch };
1200 int i, j;
1201
1202 i = j = 0;
1203 while (*v != '\0' && v < end) {
1204 if (*v != '.' && j < 2)
1205 ver_parse[i][j++] = *v;
1206 else {
1207 i++;
1208 j = 0;
1209 }
1210 v++;
1211 }
1212
1213 *m = strtol(minor, NULL, 0);
1214 *p = strtol(patch, NULL, 0);
1215 }
1216 #endif
1217
1218 static int imsm_level_to_layout(int level)
1219 {
1220 switch (level) {
1221 case 0:
1222 case 1:
1223 return 0;
1224 case 5:
1225 case 6:
1226 return ALGORITHM_LEFT_ASYMMETRIC;
1227 case 10:
1228 return 0x102;
1229 }
1230 return UnSet;
1231 }
1232
1233 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
1234 {
1235 struct intel_super *super = st->sb;
1236 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1237 struct imsm_map *map = get_imsm_map(dev, 0);
1238 struct dl *dl;
1239
1240 for (dl = super->disks; dl; dl = dl->next)
1241 if (dl->raiddisk == info->disk.raid_disk)
1242 break;
1243 info->container_member = super->current_vol;
1244 info->array.raid_disks = map->num_members;
1245 info->array.level = get_imsm_raid_level(map);
1246 info->array.layout = imsm_level_to_layout(info->array.level);
1247 info->array.md_minor = -1;
1248 info->array.ctime = 0;
1249 info->array.utime = 0;
1250 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1251 info->array.state = !dev->vol.dirty;
1252 info->custom_array_size = __le32_to_cpu(dev->size_high);
1253 info->custom_array_size <<= 32;
1254 info->custom_array_size |= __le32_to_cpu(dev->size_low);
1255
1256 info->disk.major = 0;
1257 info->disk.minor = 0;
1258 if (dl) {
1259 info->disk.major = dl->major;
1260 info->disk.minor = dl->minor;
1261 }
1262
1263 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1264 info->component_size = __le32_to_cpu(map->blocks_per_member);
1265 memset(info->uuid, 0, sizeof(info->uuid));
1266
1267 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
1268 info->resync_start = 0;
1269 else if (dev->vol.migr_state)
1270 /* FIXME add curr_migr_unit to resync_start conversion */
1271 info->resync_start = 0;
1272 else
1273 info->resync_start = ~0ULL;
1274
1275 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1276 info->name[MAX_RAID_SERIAL_LEN] = 0;
1277
1278 info->array.major_version = -1;
1279 info->array.minor_version = -2;
1280 sprintf(info->text_version, "/%s/%d",
1281 devnum2devname(st->container_dev),
1282 info->container_member);
1283 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1284 uuid_from_super_imsm(st, info->uuid);
1285 }
1286
1287 /* check the config file to see if we can return a real uuid for this spare */
1288 static void fixup_container_spare_uuid(struct mdinfo *inf)
1289 {
1290 struct mddev_ident_s *array_list;
1291
1292 if (inf->array.level != LEVEL_CONTAINER ||
1293 memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0)
1294 return;
1295
1296 array_list = conf_get_ident(NULL);
1297
1298 for (; array_list; array_list = array_list->next) {
1299 if (array_list->uuid_set) {
1300 struct supertype *_sst; /* spare supertype */
1301 struct supertype *_cst; /* container supertype */
1302
1303 _cst = array_list->st;
1304 if (_cst)
1305 _sst = _cst->ss->match_metadata_desc(inf->text_version);
1306 else
1307 _sst = NULL;
1308
1309 if (_sst) {
1310 memcpy(inf->uuid, array_list->uuid, sizeof(int[4]));
1311 free(_sst);
1312 break;
1313 }
1314 }
1315 }
1316 }
1317
1318 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
1319 {
1320 struct intel_super *super = st->sb;
1321 struct imsm_disk *disk;
1322
1323 if (super->current_vol >= 0) {
1324 getinfo_super_imsm_volume(st, info);
1325 return;
1326 }
1327
1328 /* Set raid_disks to zero so that Assemble will always pull in valid
1329 * spares
1330 */
1331 info->array.raid_disks = 0;
1332 info->array.level = LEVEL_CONTAINER;
1333 info->array.layout = 0;
1334 info->array.md_minor = -1;
1335 info->array.ctime = 0; /* N/A for imsm */
1336 info->array.utime = 0;
1337 info->array.chunk_size = 0;
1338
1339 info->disk.major = 0;
1340 info->disk.minor = 0;
1341 info->disk.raid_disk = -1;
1342 info->reshape_active = 0;
1343 info->array.major_version = -1;
1344 info->array.minor_version = -2;
1345 strcpy(info->text_version, "imsm");
1346 info->safe_mode_delay = 0;
1347 info->disk.number = -1;
1348 info->disk.state = 0;
1349 info->name[0] = 0;
1350
1351 if (super->disks) {
1352 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1353
1354 disk = &super->disks->disk;
1355 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1356 info->component_size = reserved;
1357 info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
1358 /* we don't change info->disk.raid_disk here because
1359 * this state will be finalized in mdmon after we have
1360 * found the 'most fresh' version of the metadata
1361 */
1362 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
1363 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
1364 }
1365
1366 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1367 * ->compare_super may have updated the 'num_raid_devs' field for spares
1368 */
1369 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1370 uuid_from_super_imsm(st, info->uuid);
1371 else {
1372 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1373 fixup_container_spare_uuid(info);
1374 }
1375 }
1376
1377 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1378 char *update, char *devname, int verbose,
1379 int uuid_set, char *homehost)
1380 {
1381 /* FIXME */
1382
1383 /* For 'assemble' and 'force' we need to return non-zero if any
1384 * change was made. For others, the return value is ignored.
1385 * Update options are:
1386 * force-one : This device looks a bit old but needs to be included,
1387 * update age info appropriately.
1388 * assemble: clear any 'faulty' flag to allow this device to
1389 * be assembled.
1390 * force-array: Array is degraded but being forced, mark it clean
1391 * if that will be needed to assemble it.
1392 *
1393 * newdev: not used ????
1394 * grow: Array has gained a new device - this is currently for
1395 * linear only
1396 * resync: mark as dirty so a resync will happen.
1397 * name: update the name - preserving the homehost
1398 *
1399 * Following are not relevant for this imsm:
1400 * sparc2.2 : update from old dodgey metadata
1401 * super-minor: change the preferred_minor number
1402 * summaries: update redundant counters.
1403 * uuid: Change the uuid of the array to match watch is given
1404 * homehost: update the recorded homehost
1405 * _reshape_progress: record new reshape_progress position.
1406 */
1407 int rv = 0;
1408 //struct intel_super *super = st->sb;
1409 //struct imsm_super *mpb = super->mpb;
1410
1411 if (strcmp(update, "grow") == 0) {
1412 }
1413 if (strcmp(update, "resync") == 0) {
1414 /* dev->vol.dirty = 1; */
1415 }
1416
1417 /* IMSM has no concept of UUID or homehost */
1418
1419 return rv;
1420 }
1421
1422 static size_t disks_to_mpb_size(int disks)
1423 {
1424 size_t size;
1425
1426 size = sizeof(struct imsm_super);
1427 size += (disks - 1) * sizeof(struct imsm_disk);
1428 size += 2 * sizeof(struct imsm_dev);
1429 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1430 size += (4 - 2) * sizeof(struct imsm_map);
1431 /* 4 possible disk_ord_tbl's */
1432 size += 4 * (disks - 1) * sizeof(__u32);
1433
1434 return size;
1435 }
1436
1437 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1438 {
1439 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1440 return 0;
1441
1442 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1443 }
1444
1445 static void free_devlist(struct intel_super *super)
1446 {
1447 struct intel_dev *dv;
1448
1449 while (super->devlist) {
1450 dv = super->devlist->next;
1451 free(super->devlist->dev);
1452 free(super->devlist);
1453 super->devlist = dv;
1454 }
1455 }
1456
1457 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1458 {
1459 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1460 }
1461
1462 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1463 {
1464 /*
1465 * return:
1466 * 0 same, or first was empty, and second was copied
1467 * 1 second had wrong number
1468 * 2 wrong uuid
1469 * 3 wrong other info
1470 */
1471 struct intel_super *first = st->sb;
1472 struct intel_super *sec = tst->sb;
1473
1474 if (!first) {
1475 st->sb = tst->sb;
1476 tst->sb = NULL;
1477 return 0;
1478 }
1479
1480 /* if an anchor does not have num_raid_devs set then it is a free
1481 * floating spare
1482 */
1483 if (first->anchor->num_raid_devs > 0 &&
1484 sec->anchor->num_raid_devs > 0) {
1485 /* Determine if these disks might ever have been
1486 * related. Further disambiguation can only take place
1487 * in load_super_imsm_all
1488 */
1489 __u32 first_family = first->anchor->orig_family_num;
1490 __u32 sec_family = sec->anchor->orig_family_num;
1491
1492 if (memcmp(first->anchor->sig, sec->anchor->sig,
1493 MAX_SIGNATURE_LENGTH) != 0)
1494 return 3;
1495
1496 if (first_family == 0)
1497 first_family = first->anchor->family_num;
1498 if (sec_family == 0)
1499 sec_family = sec->anchor->family_num;
1500
1501 if (first_family != sec_family)
1502 return 3;
1503
1504 }
1505
1506
1507 /* if 'first' is a spare promote it to a populated mpb with sec's
1508 * family number
1509 */
1510 if (first->anchor->num_raid_devs == 0 &&
1511 sec->anchor->num_raid_devs > 0) {
1512 int i;
1513 struct intel_dev *dv;
1514 struct imsm_dev *dev;
1515
1516 /* we need to copy raid device info from sec if an allocation
1517 * fails here we don't associate the spare
1518 */
1519 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1520 dv = malloc(sizeof(*dv));
1521 if (!dv)
1522 break;
1523 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1524 if (!dev) {
1525 free(dv);
1526 break;
1527 }
1528 dv->dev = dev;
1529 dv->index = i;
1530 dv->next = first->devlist;
1531 first->devlist = dv;
1532 }
1533 if (i < sec->anchor->num_raid_devs) {
1534 /* allocation failure */
1535 free_devlist(first);
1536 fprintf(stderr, "imsm: failed to associate spare\n");
1537 return 3;
1538 }
1539 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1540 first->anchor->orig_family_num = sec->anchor->orig_family_num;
1541 first->anchor->family_num = sec->anchor->family_num;
1542 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1543 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1544 }
1545
1546 return 0;
1547 }
1548
1549 static void fd2devname(int fd, char *name)
1550 {
1551 struct stat st;
1552 char path[256];
1553 char dname[100];
1554 char *nm;
1555 int rv;
1556
1557 name[0] = '\0';
1558 if (fstat(fd, &st) != 0)
1559 return;
1560 sprintf(path, "/sys/dev/block/%d:%d",
1561 major(st.st_rdev), minor(st.st_rdev));
1562
1563 rv = readlink(path, dname, sizeof(dname));
1564 if (rv <= 0)
1565 return;
1566
1567 dname[rv] = '\0';
1568 nm = strrchr(dname, '/');
1569 nm++;
1570 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1571 }
1572
1573 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1574
1575 static int imsm_read_serial(int fd, char *devname,
1576 __u8 serial[MAX_RAID_SERIAL_LEN])
1577 {
1578 unsigned char scsi_serial[255];
1579 int rv;
1580 int rsp_len;
1581 int len;
1582 char *dest;
1583 char *src;
1584 char *rsp_buf;
1585 int i;
1586
1587 memset(scsi_serial, 0, sizeof(scsi_serial));
1588
1589 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1590
1591 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1592 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1593 fd2devname(fd, (char *) serial);
1594 return 0;
1595 }
1596
1597 if (rv != 0) {
1598 if (devname)
1599 fprintf(stderr,
1600 Name ": Failed to retrieve serial for %s\n",
1601 devname);
1602 return rv;
1603 }
1604
1605 rsp_len = scsi_serial[3];
1606 if (!rsp_len) {
1607 if (devname)
1608 fprintf(stderr,
1609 Name ": Failed to retrieve serial for %s\n",
1610 devname);
1611 return 2;
1612 }
1613 rsp_buf = (char *) &scsi_serial[4];
1614
1615 /* trim all whitespace and non-printable characters and convert
1616 * ':' to ';'
1617 */
1618 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
1619 src = &rsp_buf[i];
1620 if (*src > 0x20) {
1621 /* ':' is reserved for use in placeholder serial
1622 * numbers for missing disks
1623 */
1624 if (*src == ':')
1625 *dest++ = ';';
1626 else
1627 *dest++ = *src;
1628 }
1629 }
1630 len = dest - rsp_buf;
1631 dest = rsp_buf;
1632
1633 /* truncate leading characters */
1634 if (len > MAX_RAID_SERIAL_LEN) {
1635 dest += len - MAX_RAID_SERIAL_LEN;
1636 len = MAX_RAID_SERIAL_LEN;
1637 }
1638
1639 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1640 memcpy(serial, dest, len);
1641
1642 return 0;
1643 }
1644
1645 static int serialcmp(__u8 *s1, __u8 *s2)
1646 {
1647 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1648 }
1649
1650 static void serialcpy(__u8 *dest, __u8 *src)
1651 {
1652 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1653 }
1654
1655 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1656 {
1657 struct dl *dl;
1658
1659 for (dl = super->disks; dl; dl = dl->next)
1660 if (serialcmp(dl->serial, serial) == 0)
1661 break;
1662
1663 return dl;
1664 }
1665
1666 static struct imsm_disk *
1667 __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
1668 {
1669 int i;
1670
1671 for (i = 0; i < mpb->num_disks; i++) {
1672 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
1673
1674 if (serialcmp(disk->serial, serial) == 0) {
1675 if (idx)
1676 *idx = i;
1677 return disk;
1678 }
1679 }
1680
1681 return NULL;
1682 }
1683
1684 static int
1685 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1686 {
1687 struct imsm_disk *disk;
1688 struct dl *dl;
1689 struct stat stb;
1690 int rv;
1691 char name[40];
1692 __u8 serial[MAX_RAID_SERIAL_LEN];
1693
1694 rv = imsm_read_serial(fd, devname, serial);
1695
1696 if (rv != 0)
1697 return 2;
1698
1699 dl = calloc(1, sizeof(*dl));
1700 if (!dl) {
1701 if (devname)
1702 fprintf(stderr,
1703 Name ": failed to allocate disk buffer for %s\n",
1704 devname);
1705 return 2;
1706 }
1707
1708 fstat(fd, &stb);
1709 dl->major = major(stb.st_rdev);
1710 dl->minor = minor(stb.st_rdev);
1711 dl->next = super->disks;
1712 dl->fd = keep_fd ? fd : -1;
1713 assert(super->disks == NULL);
1714 super->disks = dl;
1715 serialcpy(dl->serial, serial);
1716 dl->index = -2;
1717 dl->e = NULL;
1718 fd2devname(fd, name);
1719 if (devname)
1720 dl->devname = strdup(devname);
1721 else
1722 dl->devname = strdup(name);
1723
1724 /* look up this disk's index in the current anchor */
1725 disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
1726 if (disk) {
1727 dl->disk = *disk;
1728 /* only set index on disks that are a member of a
1729 * populated contianer, i.e. one with raid_devs
1730 */
1731 if (is_failed(&dl->disk))
1732 dl->index = -2;
1733 else if (is_spare(&dl->disk))
1734 dl->index = -1;
1735 }
1736
1737 return 0;
1738 }
1739
1740 #ifndef MDASSEMBLE
1741 /* When migrating map0 contains the 'destination' state while map1
1742 * contains the current state. When not migrating map0 contains the
1743 * current state. This routine assumes that map[0].map_state is set to
1744 * the current array state before being called.
1745 *
1746 * Migration is indicated by one of the following states
1747 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1748 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1749 * map1state=unitialized)
1750 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1751 * map1state=normal)
1752 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1753 * map1state=degraded)
1754 */
1755 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
1756 {
1757 struct imsm_map *dest;
1758 struct imsm_map *src = get_imsm_map(dev, 0);
1759
1760 dev->vol.migr_state = 1;
1761 set_migr_type(dev, migr_type);
1762 dev->vol.curr_migr_unit = 0;
1763 dest = get_imsm_map(dev, 1);
1764
1765 /* duplicate and then set the target end state in map[0] */
1766 memcpy(dest, src, sizeof_imsm_map(src));
1767 if (migr_type == MIGR_REBUILD) {
1768 __u32 ord;
1769 int i;
1770
1771 for (i = 0; i < src->num_members; i++) {
1772 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
1773 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
1774 }
1775 }
1776
1777 src->map_state = to_state;
1778 }
1779
1780 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1781 {
1782 struct imsm_map *map = get_imsm_map(dev, 0);
1783 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
1784 int i;
1785
1786 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1787 * completed in the last migration.
1788 *
1789 * FIXME add support for online capacity expansion and
1790 * raid-level-migration
1791 */
1792 for (i = 0; i < prev->num_members; i++)
1793 map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i];
1794
1795 dev->vol.migr_state = 0;
1796 dev->vol.curr_migr_unit = 0;
1797 map->map_state = map_state;
1798 }
1799 #endif
1800
1801 static int parse_raid_devices(struct intel_super *super)
1802 {
1803 int i;
1804 struct imsm_dev *dev_new;
1805 size_t len, len_migr;
1806 size_t space_needed = 0;
1807 struct imsm_super *mpb = super->anchor;
1808
1809 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1810 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1811 struct intel_dev *dv;
1812
1813 len = sizeof_imsm_dev(dev_iter, 0);
1814 len_migr = sizeof_imsm_dev(dev_iter, 1);
1815 if (len_migr > len)
1816 space_needed += len_migr - len;
1817
1818 dv = malloc(sizeof(*dv));
1819 if (!dv)
1820 return 1;
1821 dev_new = malloc(len_migr);
1822 if (!dev_new) {
1823 free(dv);
1824 return 1;
1825 }
1826 imsm_copy_dev(dev_new, dev_iter);
1827 dv->dev = dev_new;
1828 dv->index = i;
1829 dv->next = super->devlist;
1830 super->devlist = dv;
1831 }
1832
1833 /* ensure that super->buf is large enough when all raid devices
1834 * are migrating
1835 */
1836 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1837 void *buf;
1838
1839 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1840 if (posix_memalign(&buf, 512, len) != 0)
1841 return 1;
1842
1843 memcpy(buf, super->buf, super->len);
1844 memset(buf + super->len, 0, len - super->len);
1845 free(super->buf);
1846 super->buf = buf;
1847 super->len = len;
1848 }
1849
1850 return 0;
1851 }
1852
1853 /* retrieve a pointer to the bbm log which starts after all raid devices */
1854 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1855 {
1856 void *ptr = NULL;
1857
1858 if (__le32_to_cpu(mpb->bbm_log_size)) {
1859 ptr = mpb;
1860 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1861 }
1862
1863 return ptr;
1864 }
1865
1866 static void __free_imsm(struct intel_super *super, int free_disks);
1867
1868 /* load_imsm_mpb - read matrix metadata
1869 * allocates super->mpb to be freed by free_super
1870 */
1871 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1872 {
1873 unsigned long long dsize;
1874 unsigned long long sectors;
1875 struct stat;
1876 struct imsm_super *anchor;
1877 __u32 check_sum;
1878
1879 get_dev_size(fd, NULL, &dsize);
1880
1881 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1882 if (devname)
1883 fprintf(stderr,
1884 Name ": Cannot seek to anchor block on %s: %s\n",
1885 devname, strerror(errno));
1886 return 1;
1887 }
1888
1889 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1890 if (devname)
1891 fprintf(stderr,
1892 Name ": Failed to allocate imsm anchor buffer"
1893 " on %s\n", devname);
1894 return 1;
1895 }
1896 if (read(fd, anchor, 512) != 512) {
1897 if (devname)
1898 fprintf(stderr,
1899 Name ": Cannot read anchor block on %s: %s\n",
1900 devname, strerror(errno));
1901 free(anchor);
1902 return 1;
1903 }
1904
1905 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1906 if (devname)
1907 fprintf(stderr,
1908 Name ": no IMSM anchor on %s\n", devname);
1909 free(anchor);
1910 return 2;
1911 }
1912
1913 __free_imsm(super, 0);
1914 super->len = ROUND_UP(anchor->mpb_size, 512);
1915 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1916 if (devname)
1917 fprintf(stderr,
1918 Name ": unable to allocate %zu byte mpb buffer\n",
1919 super->len);
1920 free(anchor);
1921 return 2;
1922 }
1923 memcpy(super->buf, anchor, 512);
1924
1925 sectors = mpb_sectors(anchor) - 1;
1926 free(anchor);
1927 if (!sectors) {
1928 check_sum = __gen_imsm_checksum(super->anchor);
1929 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1930 if (devname)
1931 fprintf(stderr,
1932 Name ": IMSM checksum %x != %x on %s\n",
1933 check_sum,
1934 __le32_to_cpu(super->anchor->check_sum),
1935 devname);
1936 return 2;
1937 }
1938
1939 return 0;
1940 }
1941
1942 /* read the extended mpb */
1943 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1944 if (devname)
1945 fprintf(stderr,
1946 Name ": Cannot seek to extended mpb on %s: %s\n",
1947 devname, strerror(errno));
1948 return 1;
1949 }
1950
1951 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1952 if (devname)
1953 fprintf(stderr,
1954 Name ": Cannot read extended mpb on %s: %s\n",
1955 devname, strerror(errno));
1956 return 2;
1957 }
1958
1959 check_sum = __gen_imsm_checksum(super->anchor);
1960 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1961 if (devname)
1962 fprintf(stderr,
1963 Name ": IMSM checksum %x != %x on %s\n",
1964 check_sum, __le32_to_cpu(super->anchor->check_sum),
1965 devname);
1966 return 3;
1967 }
1968
1969 /* FIXME the BBM log is disk specific so we cannot use this global
1970 * buffer for all disks. Ok for now since we only look at the global
1971 * bbm_log_size parameter to gate assembly
1972 */
1973 super->bbm_log = __get_imsm_bbm_log(super->anchor);
1974
1975 return 0;
1976 }
1977
1978 static int
1979 load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
1980 {
1981 int err;
1982
1983 err = load_imsm_mpb(fd, super, devname);
1984 if (err)
1985 return err;
1986 err = load_imsm_disk(fd, super, devname, keep_fd);
1987 if (err)
1988 return err;
1989 err = parse_raid_devices(super);
1990
1991 return err;
1992 }
1993
1994 static void __free_imsm_disk(struct dl *d)
1995 {
1996 if (d->fd >= 0)
1997 close(d->fd);
1998 if (d->devname)
1999 free(d->devname);
2000 if (d->e)
2001 free(d->e);
2002 free(d);
2003
2004 }
2005 static void free_imsm_disks(struct intel_super *super)
2006 {
2007 struct dl *d;
2008
2009 while (super->disks) {
2010 d = super->disks;
2011 super->disks = d->next;
2012 __free_imsm_disk(d);
2013 }
2014 while (super->missing) {
2015 d = super->missing;
2016 super->missing = d->next;
2017 __free_imsm_disk(d);
2018 }
2019
2020 }
2021
2022 /* free all the pieces hanging off of a super pointer */
2023 static void __free_imsm(struct intel_super *super, int free_disks)
2024 {
2025 if (super->buf) {
2026 free(super->buf);
2027 super->buf = NULL;
2028 }
2029 if (free_disks)
2030 free_imsm_disks(super);
2031 free_devlist(super);
2032 if (super->hba) {
2033 free((void *) super->hba);
2034 super->hba = NULL;
2035 }
2036 }
2037
2038 static void free_imsm(struct intel_super *super)
2039 {
2040 __free_imsm(super, 1);
2041 free(super);
2042 }
2043
2044 static void free_super_imsm(struct supertype *st)
2045 {
2046 struct intel_super *super = st->sb;
2047
2048 if (!super)
2049 return;
2050
2051 free_imsm(super);
2052 st->sb = NULL;
2053 }
2054
2055 static struct intel_super *alloc_super(int creating_imsm)
2056 {
2057 struct intel_super *super = malloc(sizeof(*super));
2058
2059 if (super) {
2060 memset(super, 0, sizeof(*super));
2061 super->creating_imsm = creating_imsm;
2062 super->current_vol = -1;
2063 super->create_offset = ~((__u32 ) 0);
2064 if (!check_env("IMSM_NO_PLATFORM"))
2065 super->orom = find_imsm_orom();
2066 if (super->orom && !check_env("IMSM_TEST_OROM")) {
2067 struct sys_dev *list, *ent;
2068
2069 /* find the first intel ahci controller */
2070 list = find_driver_devices("pci", "ahci");
2071 for (ent = list; ent; ent = ent->next)
2072 if (devpath_to_vendor(ent->path) == 0x8086)
2073 break;
2074 if (ent) {
2075 super->hba = ent->path;
2076 ent->path = NULL;
2077 }
2078 free_sys_dev(&list);
2079 }
2080 }
2081
2082 return super;
2083 }
2084
2085 #ifndef MDASSEMBLE
2086 /* find_missing - helper routine for load_super_imsm_all that identifies
2087 * disks that have disappeared from the system. This routine relies on
2088 * the mpb being uptodate, which it is at load time.
2089 */
2090 static int find_missing(struct intel_super *super)
2091 {
2092 int i;
2093 struct imsm_super *mpb = super->anchor;
2094 struct dl *dl;
2095 struct imsm_disk *disk;
2096
2097 for (i = 0; i < mpb->num_disks; i++) {
2098 disk = __get_imsm_disk(mpb, i);
2099 dl = serial_to_dl(disk->serial, super);
2100 if (dl)
2101 continue;
2102
2103 dl = malloc(sizeof(*dl));
2104 if (!dl)
2105 return 1;
2106 dl->major = 0;
2107 dl->minor = 0;
2108 dl->fd = -1;
2109 dl->devname = strdup("missing");
2110 dl->index = i;
2111 serialcpy(dl->serial, disk->serial);
2112 dl->disk = *disk;
2113 dl->e = NULL;
2114 dl->next = super->missing;
2115 super->missing = dl;
2116 }
2117
2118 return 0;
2119 }
2120
2121 static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
2122 {
2123 struct intel_disk *idisk = disk_list;
2124
2125 while (idisk) {
2126 if (serialcmp(idisk->disk.serial, serial) == 0)
2127 break;
2128 idisk = idisk->next;
2129 }
2130
2131 return idisk;
2132 }
2133
2134 static int __prep_thunderdome(struct intel_super **table, int tbl_size,
2135 struct intel_super *super,
2136 struct intel_disk **disk_list)
2137 {
2138 struct imsm_disk *d = &super->disks->disk;
2139 struct imsm_super *mpb = super->anchor;
2140 int i, j;
2141
2142 for (i = 0; i < tbl_size; i++) {
2143 struct imsm_super *tbl_mpb = table[i]->anchor;
2144 struct imsm_disk *tbl_d = &table[i]->disks->disk;
2145
2146 if (tbl_mpb->family_num == mpb->family_num) {
2147 if (tbl_mpb->check_sum == mpb->check_sum) {
2148 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2149 __func__, super->disks->major,
2150 super->disks->minor,
2151 table[i]->disks->major,
2152 table[i]->disks->minor);
2153 break;
2154 }
2155
2156 if (((is_configured(d) && !is_configured(tbl_d)) ||
2157 is_configured(d) == is_configured(tbl_d)) &&
2158 tbl_mpb->generation_num < mpb->generation_num) {
2159 /* current version of the mpb is a
2160 * better candidate than the one in
2161 * super_table, but copy over "cross
2162 * generational" status
2163 */
2164 struct intel_disk *idisk;
2165
2166 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2167 __func__, super->disks->major,
2168 super->disks->minor,
2169 table[i]->disks->major,
2170 table[i]->disks->minor);
2171
2172 idisk = disk_list_get(tbl_d->serial, *disk_list);
2173 if (idisk && is_failed(&idisk->disk))
2174 tbl_d->status |= FAILED_DISK;
2175 break;
2176 } else {
2177 struct intel_disk *idisk;
2178 struct imsm_disk *disk;
2179
2180 /* tbl_mpb is more up to date, but copy
2181 * over cross generational status before
2182 * returning
2183 */
2184 disk = __serial_to_disk(d->serial, mpb, NULL);
2185 if (disk && is_failed(disk))
2186 d->status |= FAILED_DISK;
2187
2188 idisk = disk_list_get(d->serial, *disk_list);
2189 if (idisk) {
2190 idisk->owner = i;
2191 if (disk && is_configured(disk))
2192 idisk->disk.status |= CONFIGURED_DISK;
2193 }
2194
2195 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2196 __func__, super->disks->major,
2197 super->disks->minor,
2198 table[i]->disks->major,
2199 table[i]->disks->minor);
2200
2201 return tbl_size;
2202 }
2203 }
2204 }
2205
2206 if (i >= tbl_size)
2207 table[tbl_size++] = super;
2208 else
2209 table[i] = super;
2210
2211 /* update/extend the merged list of imsm_disk records */
2212 for (j = 0; j < mpb->num_disks; j++) {
2213 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
2214 struct intel_disk *idisk;
2215
2216 idisk = disk_list_get(disk->serial, *disk_list);
2217 if (idisk) {
2218 idisk->disk.status |= disk->status;
2219 if (is_configured(&idisk->disk) ||
2220 is_failed(&idisk->disk))
2221 idisk->disk.status &= ~(SPARE_DISK);
2222 } else {
2223 idisk = calloc(1, sizeof(*idisk));
2224 if (!idisk)
2225 return -1;
2226 idisk->owner = IMSM_UNKNOWN_OWNER;
2227 idisk->disk = *disk;
2228 idisk->next = *disk_list;
2229 *disk_list = idisk;
2230 }
2231
2232 if (serialcmp(idisk->disk.serial, d->serial) == 0)
2233 idisk->owner = i;
2234 }
2235
2236 return tbl_size;
2237 }
2238
2239 static struct intel_super *
2240 validate_members(struct intel_super *super, struct intel_disk *disk_list,
2241 const int owner)
2242 {
2243 struct imsm_super *mpb = super->anchor;
2244 int ok_count = 0;
2245 int i;
2246
2247 for (i = 0; i < mpb->num_disks; i++) {
2248 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2249 struct intel_disk *idisk;
2250
2251 idisk = disk_list_get(disk->serial, disk_list);
2252 if (idisk) {
2253 if (idisk->owner == owner ||
2254 idisk->owner == IMSM_UNKNOWN_OWNER)
2255 ok_count++;
2256 else
2257 dprintf("%s: '%.16s' owner %d != %d\n",
2258 __func__, disk->serial, idisk->owner,
2259 owner);
2260 } else {
2261 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2262 __func__, __le32_to_cpu(mpb->family_num), i,
2263 disk->serial);
2264 break;
2265 }
2266 }
2267
2268 if (ok_count == mpb->num_disks)
2269 return super;
2270 return NULL;
2271 }
2272
2273 static void show_conflicts(__u32 family_num, struct intel_super *super_list)
2274 {
2275 struct intel_super *s;
2276
2277 for (s = super_list; s; s = s->next) {
2278 if (family_num != s->anchor->family_num)
2279 continue;
2280 fprintf(stderr, "Conflict, offlining family %#x on '%s'\n",
2281 __le32_to_cpu(family_num), s->disks->devname);
2282 }
2283 }
2284
2285 static struct intel_super *
2286 imsm_thunderdome(struct intel_super **super_list, int len)
2287 {
2288 struct intel_super *super_table[len];
2289 struct intel_disk *disk_list = NULL;
2290 struct intel_super *champion, *spare;
2291 struct intel_super *s, **del;
2292 int tbl_size = 0;
2293 int conflict;
2294 int i;
2295
2296 memset(super_table, 0, sizeof(super_table));
2297 for (s = *super_list; s; s = s->next)
2298 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
2299
2300 for (i = 0; i < tbl_size; i++) {
2301 struct imsm_disk *d;
2302 struct intel_disk *idisk;
2303 struct imsm_super *mpb = super_table[i]->anchor;
2304
2305 s = super_table[i];
2306 d = &s->disks->disk;
2307
2308 /* 'd' must appear in merged disk list for its
2309 * configuration to be valid
2310 */
2311 idisk = disk_list_get(d->serial, disk_list);
2312 if (idisk && idisk->owner == i)
2313 s = validate_members(s, disk_list, i);
2314 else
2315 s = NULL;
2316
2317 if (!s)
2318 dprintf("%s: marking family: %#x from %d:%d offline\n",
2319 __func__, mpb->family_num,
2320 super_table[i]->disks->major,
2321 super_table[i]->disks->minor);
2322 super_table[i] = s;
2323 }
2324
2325 /* This is where the mdadm implementation differs from the Windows
2326 * driver which has no strict concept of a container. We can only
2327 * assemble one family from a container, so when returning a prodigal
2328 * array member to this system the code will not be able to disambiguate
2329 * the container contents that should be assembled ("foreign" versus
2330 * "local"). It requires user intervention to set the orig_family_num
2331 * to a new value to establish a new container. The Windows driver in
2332 * this situation fixes up the volume name in place and manages the
2333 * foreign array as an independent entity.
2334 */
2335 s = NULL;
2336 spare = NULL;
2337 conflict = 0;
2338 for (i = 0; i < tbl_size; i++) {
2339 struct intel_super *tbl_ent = super_table[i];
2340 int is_spare = 0;
2341
2342 if (!tbl_ent)
2343 continue;
2344
2345 if (tbl_ent->anchor->num_raid_devs == 0) {
2346 spare = tbl_ent;
2347 is_spare = 1;
2348 }
2349
2350 if (s && !is_spare) {
2351 show_conflicts(tbl_ent->anchor->family_num, *super_list);
2352 conflict++;
2353 } else if (!s && !is_spare)
2354 s = tbl_ent;
2355 }
2356
2357 if (!s)
2358 s = spare;
2359 if (!s) {
2360 champion = NULL;
2361 goto out;
2362 }
2363 champion = s;
2364
2365 if (conflict)
2366 fprintf(stderr, "Chose family %#x on '%s', "
2367 "assemble conflicts to new container with '--update=uuid'\n",
2368 __le32_to_cpu(s->anchor->family_num), s->disks->devname);
2369
2370 /* collect all dl's onto 'champion', and update them to
2371 * champion's version of the status
2372 */
2373 for (s = *super_list; s; s = s->next) {
2374 struct imsm_super *mpb = champion->anchor;
2375 struct dl *dl = s->disks;
2376
2377 if (s == champion)
2378 continue;
2379
2380 for (i = 0; i < mpb->num_disks; i++) {
2381 struct imsm_disk *disk;
2382
2383 disk = __serial_to_disk(dl->serial, mpb, &dl->index);
2384 if (disk) {
2385 dl->disk = *disk;
2386 /* only set index on disks that are a member of
2387 * a populated contianer, i.e. one with
2388 * raid_devs
2389 */
2390 if (is_failed(&dl->disk))
2391 dl->index = -2;
2392 else if (is_spare(&dl->disk))
2393 dl->index = -1;
2394 break;
2395 }
2396 }
2397
2398 if (i >= mpb->num_disks) {
2399 struct intel_disk *idisk;
2400
2401 idisk = disk_list_get(dl->serial, disk_list);
2402 if (is_spare(&idisk->disk) &&
2403 !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
2404 dl->index = -1;
2405 else {
2406 dl->index = -2;
2407 continue;
2408 }
2409 }
2410
2411 dl->next = champion->disks;
2412 champion->disks = dl;
2413 s->disks = NULL;
2414 }
2415
2416 /* delete 'champion' from super_list */
2417 for (del = super_list; *del; ) {
2418 if (*del == champion) {
2419 *del = (*del)->next;
2420 break;
2421 } else
2422 del = &(*del)->next;
2423 }
2424 champion->next = NULL;
2425
2426 out:
2427 while (disk_list) {
2428 struct intel_disk *idisk = disk_list;
2429
2430 disk_list = disk_list->next;
2431 free(idisk);
2432 }
2433
2434 return champion;
2435 }
2436
2437 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
2438 char *devname, int keep_fd)
2439 {
2440 struct mdinfo *sra;
2441 struct intel_super *super_list = NULL;
2442 struct intel_super *super = NULL;
2443 int devnum = fd2devnum(fd);
2444 struct mdinfo *sd;
2445 int retry;
2446 int err = 0;
2447 int i;
2448 enum sysfs_read_flags flags;
2449
2450 flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2451 if (mdmon_running(devnum))
2452 flags |= SKIP_GONE_DEVS;
2453
2454 /* check if 'fd' an opened container */
2455 sra = sysfs_read(fd, 0, flags);
2456 if (!sra)
2457 return 1;
2458
2459 if (sra->array.major_version != -1 ||
2460 sra->array.minor_version != -2 ||
2461 strcmp(sra->text_version, "imsm") != 0)
2462 return 1;
2463
2464 /* load all mpbs */
2465 for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
2466 struct intel_super *s = alloc_super(0);
2467 char nm[20];
2468 int dfd;
2469
2470 err = 1;
2471 if (!s)
2472 goto error;
2473 s->next = super_list;
2474 super_list = s;
2475
2476 err = 2;
2477 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2478 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
2479 if (dfd < 0)
2480 goto error;
2481
2482 err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
2483
2484 /* retry the load if we might have raced against mdmon */
2485 if (err == 3 && mdmon_running(devnum))
2486 for (retry = 0; retry < 3; retry++) {
2487 usleep(3000);
2488 err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
2489 if (err != 3)
2490 break;
2491 }
2492 if (!keep_fd)
2493 close(dfd);
2494 if (err)
2495 goto error;
2496 }
2497
2498 /* all mpbs enter, maybe one leaves */
2499 super = imsm_thunderdome(&super_list, i);
2500 if (!super) {
2501 err = 1;
2502 goto error;
2503 }
2504
2505 if (find_missing(super) != 0) {
2506 free_imsm(super);
2507 err = 2;
2508 goto error;
2509 }
2510
2511 if (st->subarray[0]) {
2512 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2513 super->current_vol = atoi(st->subarray);
2514 else {
2515 free_imsm(super);
2516 err = 1;
2517 goto error;
2518 }
2519 }
2520 err = 0;
2521
2522 error:
2523 while (super_list) {
2524 struct intel_super *s = super_list;
2525
2526 super_list = super_list->next;
2527 free_imsm(s);
2528 }
2529
2530 if (err)
2531 return err;
2532
2533 *sbp = super;
2534 st->container_dev = devnum;
2535 if (err == 0 && st->ss == NULL) {
2536 st->ss = &super_imsm;
2537 st->minor_version = 0;
2538 st->max_devs = IMSM_MAX_DEVICES;
2539 }
2540 st->loaded_container = 1;
2541
2542 return 0;
2543 }
2544 #endif
2545
2546 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2547 {
2548 struct intel_super *super;
2549 int rv;
2550
2551 #ifndef MDASSEMBLE
2552 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2553 return 0;
2554 #endif
2555
2556 free_super_imsm(st);
2557
2558 super = alloc_super(0);
2559 if (!super) {
2560 fprintf(stderr,
2561 Name ": malloc of %zu failed.\n",
2562 sizeof(*super));
2563 return 1;
2564 }
2565
2566 rv = load_and_parse_mpb(fd, super, devname, 0);
2567
2568 if (rv) {
2569 if (devname)
2570 fprintf(stderr,
2571 Name ": Failed to load all information "
2572 "sections on %s\n", devname);
2573 free_imsm(super);
2574 return rv;
2575 }
2576
2577 if (st->subarray[0]) {
2578 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2579 super->current_vol = atoi(st->subarray);
2580 else {
2581 free_imsm(super);
2582 return 1;
2583 }
2584 }
2585
2586 st->sb = super;
2587 if (st->ss == NULL) {
2588 st->ss = &super_imsm;
2589 st->minor_version = 0;
2590 st->max_devs = IMSM_MAX_DEVICES;
2591 }
2592 st->loaded_container = 0;
2593
2594 return 0;
2595 }
2596
2597 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2598 {
2599 if (info->level == 1)
2600 return 128;
2601 return info->chunk_size >> 9;
2602 }
2603
2604 static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
2605 {
2606 __u32 num_stripes;
2607
2608 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2609 num_stripes /= num_domains;
2610
2611 return num_stripes;
2612 }
2613
2614 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2615 {
2616 if (info->level == 1)
2617 return info->size * 2;
2618 else
2619 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2620 }
2621
2622 static void imsm_update_version_info(struct intel_super *super)
2623 {
2624 /* update the version and attributes */
2625 struct imsm_super *mpb = super->anchor;
2626 char *version;
2627 struct imsm_dev *dev;
2628 struct imsm_map *map;
2629 int i;
2630
2631 for (i = 0; i < mpb->num_raid_devs; i++) {
2632 dev = get_imsm_dev(super, i);
2633 map = get_imsm_map(dev, 0);
2634 if (__le32_to_cpu(dev->size_high) > 0)
2635 mpb->attributes |= MPB_ATTRIB_2TB;
2636
2637 /* FIXME detect when an array spans a port multiplier */
2638 #if 0
2639 mpb->attributes |= MPB_ATTRIB_PM;
2640 #endif
2641
2642 if (mpb->num_raid_devs > 1 ||
2643 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2644 version = MPB_VERSION_ATTRIBS;
2645 switch (get_imsm_raid_level(map)) {
2646 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2647 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2648 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2649 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2650 }
2651 } else {
2652 if (map->num_members >= 5)
2653 version = MPB_VERSION_5OR6_DISK_ARRAY;
2654 else if (dev->status == DEV_CLONE_N_GO)
2655 version = MPB_VERSION_CNG;
2656 else if (get_imsm_raid_level(map) == 5)
2657 version = MPB_VERSION_RAID5;
2658 else if (map->num_members >= 3)
2659 version = MPB_VERSION_3OR4_DISK_ARRAY;
2660 else if (get_imsm_raid_level(map) == 1)
2661 version = MPB_VERSION_RAID1;
2662 else
2663 version = MPB_VERSION_RAID0;
2664 }
2665 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2666 }
2667 }
2668
2669 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2670 unsigned long long size, char *name,
2671 char *homehost, int *uuid)
2672 {
2673 /* We are creating a volume inside a pre-existing container.
2674 * so st->sb is already set.
2675 */
2676 struct intel_super *super = st->sb;
2677 struct imsm_super *mpb = super->anchor;
2678 struct intel_dev *dv;
2679 struct imsm_dev *dev;
2680 struct imsm_vol *vol;
2681 struct imsm_map *map;
2682 int idx = mpb->num_raid_devs;
2683 int i;
2684 unsigned long long array_blocks;
2685 size_t size_old, size_new;
2686 __u32 num_data_stripes;
2687
2688 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2689 fprintf(stderr, Name": This imsm-container already has the "
2690 "maximum of %d volumes\n", super->orom->vpa);
2691 return 0;
2692 }
2693
2694 /* ensure the mpb is large enough for the new data */
2695 size_old = __le32_to_cpu(mpb->mpb_size);
2696 size_new = disks_to_mpb_size(info->nr_disks);
2697 if (size_new > size_old) {
2698 void *mpb_new;
2699 size_t size_round = ROUND_UP(size_new, 512);
2700
2701 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2702 fprintf(stderr, Name": could not allocate new mpb\n");
2703 return 0;
2704 }
2705 memcpy(mpb_new, mpb, size_old);
2706 free(mpb);
2707 mpb = mpb_new;
2708 super->anchor = mpb_new;
2709 mpb->mpb_size = __cpu_to_le32(size_new);
2710 memset(mpb_new + size_old, 0, size_round - size_old);
2711 }
2712 super->current_vol = idx;
2713 /* when creating the first raid device in this container set num_disks
2714 * to zero, i.e. delete this spare and add raid member devices in
2715 * add_to_super_imsm_volume()
2716 */
2717 if (super->current_vol == 0)
2718 mpb->num_disks = 0;
2719
2720 for (i = 0; i < super->current_vol; i++) {
2721 dev = get_imsm_dev(super, i);
2722 if (strncmp((char *) dev->volume, name,
2723 MAX_RAID_SERIAL_LEN) == 0) {
2724 fprintf(stderr, Name": '%s' is already defined for this container\n",
2725 name);
2726 return 0;
2727 }
2728 }
2729
2730 sprintf(st->subarray, "%d", idx);
2731 dv = malloc(sizeof(*dv));
2732 if (!dv) {
2733 fprintf(stderr, Name ": failed to allocate device list entry\n");
2734 return 0;
2735 }
2736 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2737 if (!dev) {
2738 free(dv);
2739 fprintf(stderr, Name": could not allocate raid device\n");
2740 return 0;
2741 }
2742 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2743 if (info->level == 1)
2744 array_blocks = info_to_blocks_per_member(info);
2745 else
2746 array_blocks = calc_array_size(info->level, info->raid_disks,
2747 info->layout, info->chunk_size,
2748 info->size*2);
2749 /* round array size down to closest MB */
2750 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
2751
2752 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2753 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2754 dev->status = __cpu_to_le32(0);
2755 dev->reserved_blocks = __cpu_to_le32(0);
2756 vol = &dev->vol;
2757 vol->migr_state = 0;
2758 set_migr_type(dev, MIGR_INIT);
2759 vol->dirty = 0;
2760 vol->curr_migr_unit = 0;
2761 map = get_imsm_map(dev, 0);
2762 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2763 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2764 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2765 map->failed_disk_num = ~0;
2766 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2767 IMSM_T_STATE_NORMAL;
2768 map->ddf = 1;
2769
2770 if (info->level == 1 && info->raid_disks > 2) {
2771 fprintf(stderr, Name": imsm does not support more than 2 disks"
2772 "in a raid1 volume\n");
2773 return 0;
2774 }
2775
2776 map->raid_level = info->level;
2777 if (info->level == 10) {
2778 map->raid_level = 1;
2779 map->num_domains = info->raid_disks / 2;
2780 } else if (info->level == 1)
2781 map->num_domains = info->raid_disks;
2782 else
2783 map->num_domains = 1;
2784
2785 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
2786 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
2787
2788 map->num_members = info->raid_disks;
2789 for (i = 0; i < map->num_members; i++) {
2790 /* initialized in add_to_super */
2791 set_imsm_ord_tbl_ent(map, i, 0);
2792 }
2793 mpb->num_raid_devs++;
2794
2795 dv->dev = dev;
2796 dv->index = super->current_vol;
2797 dv->next = super->devlist;
2798 super->devlist = dv;
2799
2800 imsm_update_version_info(super);
2801
2802 return 1;
2803 }
2804
2805 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2806 unsigned long long size, char *name,
2807 char *homehost, int *uuid)
2808 {
2809 /* This is primarily called by Create when creating a new array.
2810 * We will then get add_to_super called for each component, and then
2811 * write_init_super called to write it out to each device.
2812 * For IMSM, Create can create on fresh devices or on a pre-existing
2813 * array.
2814 * To create on a pre-existing array a different method will be called.
2815 * This one is just for fresh drives.
2816 */
2817 struct intel_super *super;
2818 struct imsm_super *mpb;
2819 size_t mpb_size;
2820 char *version;
2821
2822 if (st->sb)
2823 return init_super_imsm_volume(st, info, size, name, homehost, uuid);
2824
2825 if (info)
2826 mpb_size = disks_to_mpb_size(info->nr_disks);
2827 else
2828 mpb_size = 512;
2829
2830 super = alloc_super(1);
2831 if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) {
2832 free(super);
2833 super = NULL;
2834 }
2835 if (!super) {
2836 fprintf(stderr, Name
2837 ": %s could not allocate superblock\n", __func__);
2838 return 0;
2839 }
2840 memset(super->buf, 0, mpb_size);
2841 mpb = super->buf;
2842 mpb->mpb_size = __cpu_to_le32(mpb_size);
2843 st->sb = super;
2844
2845 if (info == NULL) {
2846 /* zeroing superblock */
2847 return 0;
2848 }
2849
2850 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2851
2852 version = (char *) mpb->sig;
2853 strcpy(version, MPB_SIGNATURE);
2854 version += strlen(MPB_SIGNATURE);
2855 strcpy(version, MPB_VERSION_RAID0);
2856
2857 return 1;
2858 }
2859
2860 #ifndef MDASSEMBLE
2861 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2862 int fd, char *devname)
2863 {
2864 struct intel_super *super = st->sb;
2865 struct imsm_super *mpb = super->anchor;
2866 struct dl *dl;
2867 struct imsm_dev *dev;
2868 struct imsm_map *map;
2869
2870 dev = get_imsm_dev(super, super->current_vol);
2871 map = get_imsm_map(dev, 0);
2872
2873 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2874 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2875 devname);
2876 return 1;
2877 }
2878
2879 if (fd == -1) {
2880 /* we're doing autolayout so grab the pre-marked (in
2881 * validate_geometry) raid_disk
2882 */
2883 for (dl = super->disks; dl; dl = dl->next)
2884 if (dl->raiddisk == dk->raid_disk)
2885 break;
2886 } else {
2887 for (dl = super->disks; dl ; dl = dl->next)
2888 if (dl->major == dk->major &&
2889 dl->minor == dk->minor)
2890 break;
2891 }
2892
2893 if (!dl) {
2894 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2895 return 1;
2896 }
2897
2898 /* add a pristine spare to the metadata */
2899 if (dl->index < 0) {
2900 dl->index = super->anchor->num_disks;
2901 super->anchor->num_disks++;
2902 }
2903 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2904 dl->disk.status = CONFIGURED_DISK;
2905
2906 /* if we are creating the first raid device update the family number */
2907 if (super->current_vol == 0) {
2908 __u32 sum;
2909 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2910 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2911
2912 *_dev = *dev;
2913 *_disk = dl->disk;
2914 sum = random32();
2915 sum += __gen_imsm_checksum(mpb);
2916 mpb->family_num = __cpu_to_le32(sum);
2917 mpb->orig_family_num = mpb->family_num;
2918 }
2919
2920 return 0;
2921 }
2922
2923 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2924 int fd, char *devname)
2925 {
2926 struct intel_super *super = st->sb;
2927 struct dl *dd;
2928 unsigned long long size;
2929 __u32 id;
2930 int rv;
2931 struct stat stb;
2932
2933 /* if we are on an RAID enabled platform check that the disk is
2934 * attached to the raid controller
2935 */
2936 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2937 fprintf(stderr,
2938 Name ": %s is not attached to the raid controller: %s\n",
2939 devname ? : "disk", super->hba);
2940 return 1;
2941 }
2942
2943 if (super->current_vol >= 0)
2944 return add_to_super_imsm_volume(st, dk, fd, devname);
2945
2946 fstat(fd, &stb);
2947 dd = malloc(sizeof(*dd));
2948 if (!dd) {
2949 fprintf(stderr,
2950 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2951 return 1;
2952 }
2953 memset(dd, 0, sizeof(*dd));
2954 dd->major = major(stb.st_rdev);
2955 dd->minor = minor(stb.st_rdev);
2956 dd->index = -1;
2957 dd->devname = devname ? strdup(devname) : NULL;
2958 dd->fd = fd;
2959 dd->e = NULL;
2960 rv = imsm_read_serial(fd, devname, dd->serial);
2961 if (rv) {
2962 fprintf(stderr,
2963 Name ": failed to retrieve scsi serial, aborting\n");
2964 free(dd);
2965 abort();
2966 }
2967
2968 get_dev_size(fd, NULL, &size);
2969 size /= 512;
2970 serialcpy(dd->disk.serial, dd->serial);
2971 dd->disk.total_blocks = __cpu_to_le32(size);
2972 dd->disk.status = SPARE_DISK;
2973 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
2974 dd->disk.scsi_id = __cpu_to_le32(id);
2975 else
2976 dd->disk.scsi_id = __cpu_to_le32(0);
2977
2978 if (st->update_tail) {
2979 dd->next = super->add;
2980 super->add = dd;
2981 } else {
2982 dd->next = super->disks;
2983 super->disks = dd;
2984 }
2985
2986 return 0;
2987 }
2988
2989 static int store_imsm_mpb(int fd, struct imsm_super *mpb);
2990
2991 static union {
2992 char buf[512];
2993 struct imsm_super anchor;
2994 } spare_record __attribute__ ((aligned(512)));
2995
2996 /* spare records have their own family number and do not have any defined raid
2997 * devices
2998 */
2999 static int write_super_imsm_spares(struct intel_super *super, int doclose)
3000 {
3001 struct imsm_super *mpb = super->anchor;
3002 struct imsm_super *spare = &spare_record.anchor;
3003 __u32 sum;
3004 struct dl *d;
3005
3006 spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)),
3007 spare->generation_num = __cpu_to_le32(1UL),
3008 spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
3009 spare->num_disks = 1,
3010 spare->num_raid_devs = 0,
3011 spare->cache_size = mpb->cache_size,
3012 spare->pwr_cycle_count = __cpu_to_le32(1),
3013
3014 snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
3015 MPB_SIGNATURE MPB_VERSION_RAID0);
3016
3017 for (d = super->disks; d; d = d->next) {
3018 if (d->index != -1)
3019 continue;
3020
3021 spare->disk[0] = d->disk;
3022 sum = __gen_imsm_checksum(spare);
3023 spare->family_num = __cpu_to_le32(sum);
3024 spare->orig_family_num = 0;
3025 sum = __gen_imsm_checksum(spare);
3026 spare->check_sum = __cpu_to_le32(sum);
3027
3028 if (store_imsm_mpb(d->fd, spare)) {
3029 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3030 __func__, d->major, d->minor, strerror(errno));
3031 return 1;
3032 }
3033 if (doclose) {
3034 close(d->fd);
3035 d->fd = -1;
3036 }
3037 }
3038
3039 return 0;
3040 }
3041
3042 static int write_super_imsm(struct intel_super *super, int doclose)
3043 {
3044 struct imsm_super *mpb = super->anchor;
3045 struct dl *d;
3046 __u32 generation;
3047 __u32 sum;
3048 int spares = 0;
3049 int i;
3050 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
3051
3052 /* 'generation' is incremented everytime the metadata is written */
3053 generation = __le32_to_cpu(mpb->generation_num);
3054 generation++;
3055 mpb->generation_num = __cpu_to_le32(generation);
3056
3057 /* fix up cases where previous mdadm releases failed to set
3058 * orig_family_num
3059 */
3060 if (mpb->orig_family_num == 0)
3061 mpb->orig_family_num = mpb->family_num;
3062
3063 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
3064 for (d = super->disks; d; d = d->next) {
3065 if (d->index == -1)
3066 spares++;
3067 else
3068 mpb->disk[d->index] = d->disk;
3069 }
3070 for (d = super->missing; d; d = d->next)
3071 mpb->disk[d->index] = d->disk;
3072
3073 for (i = 0; i < mpb->num_raid_devs; i++) {
3074 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
3075
3076 imsm_copy_dev(dev, get_imsm_dev(super, i));
3077 mpb_size += sizeof_imsm_dev(dev, 0);
3078 }
3079 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
3080 mpb->mpb_size = __cpu_to_le32(mpb_size);
3081
3082 /* recalculate checksum */
3083 sum = __gen_imsm_checksum(mpb);
3084 mpb->check_sum = __cpu_to_le32(sum);
3085
3086 /* write the mpb for disks that compose raid devices */
3087 for (d = super->disks; d ; d = d->next) {
3088 if (d->index < 0)
3089 continue;
3090 if (store_imsm_mpb(d->fd, mpb))
3091 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3092 __func__, d->major, d->minor, strerror(errno));
3093 if (doclose) {
3094 close(d->fd);
3095 d->fd = -1;
3096 }
3097 }
3098
3099 if (spares)
3100 return write_super_imsm_spares(super, doclose);
3101
3102 return 0;
3103 }
3104
3105
3106 static int create_array(struct supertype *st, int dev_idx)
3107 {
3108 size_t len;
3109 struct imsm_update_create_array *u;
3110 struct intel_super *super = st->sb;
3111 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
3112 struct imsm_map *map = get_imsm_map(dev, 0);
3113 struct disk_info *inf;
3114 struct imsm_disk *disk;
3115 int i;
3116
3117 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
3118 sizeof(*inf) * map->num_members;
3119 u = malloc(len);
3120 if (!u) {
3121 fprintf(stderr, "%s: failed to allocate update buffer\n",
3122 __func__);
3123 return 1;
3124 }
3125
3126 u->type = update_create_array;
3127 u->dev_idx = dev_idx;
3128 imsm_copy_dev(&u->dev, dev);
3129 inf = get_disk_info(u);
3130 for (i = 0; i < map->num_members; i++) {
3131 int idx = get_imsm_disk_idx(dev, i);
3132
3133 disk = get_imsm_disk(super, idx);
3134 serialcpy(inf[i].serial, disk->serial);
3135 }
3136 append_metadata_update(st, u, len);
3137
3138 return 0;
3139 }
3140
3141 static int _add_disk(struct supertype *st)
3142 {
3143 struct intel_super *super = st->sb;
3144 size_t len;
3145 struct imsm_update_add_disk *u;
3146
3147 if (!super->add)
3148 return 0;
3149
3150 len = sizeof(*u);
3151 u = malloc(len);
3152 if (!u) {
3153 fprintf(stderr, "%s: failed to allocate update buffer\n",
3154 __func__);
3155 return 1;
3156 }
3157
3158 u->type = update_add_disk;
3159 append_metadata_update(st, u, len);
3160
3161 return 0;
3162 }
3163
3164 static int write_init_super_imsm(struct supertype *st)
3165 {
3166 struct intel_super *super = st->sb;
3167 int current_vol = super->current_vol;
3168
3169 /* we are done with current_vol reset it to point st at the container */
3170 super->current_vol = -1;
3171
3172 if (st->update_tail) {
3173 /* queue the recently created array / added disk
3174 * as a metadata update */
3175 struct dl *d;
3176 int rv;
3177
3178 /* determine if we are creating a volume or adding a disk */
3179 if (current_vol < 0) {
3180 /* in the add disk case we are running in mdmon
3181 * context, so don't close fd's
3182 */
3183 return _add_disk(st);
3184 } else
3185 rv = create_array(st, current_vol);
3186
3187 for (d = super->disks; d ; d = d->next) {
3188 close(d->fd);
3189 d->fd = -1;
3190 }
3191
3192 return rv;
3193 } else
3194 return write_super_imsm(st->sb, 1);
3195 }
3196 #endif
3197
3198 static int store_super_imsm(struct supertype *st, int fd)
3199 {
3200 struct intel_super *super = st->sb;
3201 struct imsm_super *mpb = super ? super->anchor : NULL;
3202
3203 if (!mpb)
3204 return 1;
3205
3206 return store_imsm_mpb(fd, mpb);
3207 }
3208
3209 static int imsm_bbm_log_size(struct imsm_super *mpb)
3210 {
3211 return __le32_to_cpu(mpb->bbm_log_size);
3212 }
3213
3214 #ifndef MDASSEMBLE
3215 static int validate_geometry_imsm_container(struct supertype *st, int level,
3216 int layout, int raiddisks, int chunk,
3217 unsigned long long size, char *dev,
3218 unsigned long long *freesize,
3219 int verbose)
3220 {
3221 int fd;
3222 unsigned long long ldsize;
3223 const struct imsm_orom *orom;
3224
3225 if (level != LEVEL_CONTAINER)
3226 return 0;
3227 if (!dev)
3228 return 1;
3229
3230 if (check_env("IMSM_NO_PLATFORM"))
3231 orom = NULL;
3232 else
3233 orom = find_imsm_orom();
3234 if (orom && raiddisks > orom->tds) {
3235 if (verbose)
3236 fprintf(stderr, Name ": %d exceeds maximum number of"
3237 " platform supported disks: %d\n",
3238 raiddisks, orom->tds);
3239 return 0;
3240 }
3241
3242 fd = open(dev, O_RDONLY|O_EXCL, 0);
3243 if (fd < 0) {
3244 if (verbose)
3245 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
3246 dev, strerror(errno));
3247 return 0;
3248 }
3249 if (!get_dev_size(fd, dev, &ldsize)) {
3250 close(fd);
3251 return 0;
3252 }
3253 close(fd);
3254
3255 *freesize = avail_size_imsm(st, ldsize >> 9);
3256
3257 return 1;
3258 }
3259
3260 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
3261 {
3262 const unsigned long long base_start = e[*idx].start;
3263 unsigned long long end = base_start + e[*idx].size;
3264 int i;
3265
3266 if (base_start == end)
3267 return 0;
3268
3269 *idx = *idx + 1;
3270 for (i = *idx; i < num_extents; i++) {
3271 /* extend overlapping extents */
3272 if (e[i].start >= base_start &&
3273 e[i].start <= end) {
3274 if (e[i].size == 0)
3275 return 0;
3276 if (e[i].start + e[i].size > end)
3277 end = e[i].start + e[i].size;
3278 } else if (e[i].start > end) {
3279 *idx = i;
3280 break;
3281 }
3282 }
3283
3284 return end - base_start;
3285 }
3286
3287 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
3288 {
3289 /* build a composite disk with all known extents and generate a new
3290 * 'maxsize' given the "all disks in an array must share a common start
3291 * offset" constraint
3292 */
3293 struct extent *e = calloc(sum_extents, sizeof(*e));
3294 struct dl *dl;
3295 int i, j;
3296 int start_extent;
3297 unsigned long long pos;
3298 unsigned long long start = 0;
3299 unsigned long long maxsize;
3300 unsigned long reserve;
3301
3302 if (!e)
3303 return ~0ULL; /* error */
3304
3305 /* coalesce and sort all extents. also, check to see if we need to
3306 * reserve space between member arrays
3307 */
3308 j = 0;
3309 for (dl = super->disks; dl; dl = dl->next) {
3310 if (!dl->e)
3311 continue;
3312 for (i = 0; i < dl->extent_cnt; i++)
3313 e[j++] = dl->e[i];
3314 }
3315 qsort(e, sum_extents, sizeof(*e), cmp_extent);
3316
3317 /* merge extents */
3318 i = 0;
3319 j = 0;
3320 while (i < sum_extents) {
3321 e[j].start = e[i].start;
3322 e[j].size = find_size(e, &i, sum_extents);
3323 j++;
3324 if (e[j-1].size == 0)
3325 break;
3326 }
3327
3328 pos = 0;
3329 maxsize = 0;
3330 start_extent = 0;
3331 i = 0;
3332 do {
3333 unsigned long long esize;
3334
3335 esize = e[i].start - pos;
3336 if (esize >= maxsize) {
3337 maxsize = esize;
3338 start = pos;
3339 start_extent = i;
3340 }
3341 pos = e[i].start + e[i].size;
3342 i++;
3343 } while (e[i-1].size);
3344 free(e);
3345
3346 if (start_extent > 0)
3347 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
3348 else
3349 reserve = 0;
3350
3351 if (maxsize < reserve)
3352 return ~0ULL;
3353
3354 super->create_offset = ~((__u32) 0);
3355 if (start + reserve > super->create_offset)
3356 return ~0ULL; /* start overflows create_offset */
3357 super->create_offset = start + reserve;
3358
3359 return maxsize - reserve;
3360 }
3361
3362 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
3363 {
3364 if (level < 0 || level == 6 || level == 4)
3365 return 0;
3366
3367 /* if we have an orom prevent invalid raid levels */
3368 if (orom)
3369 switch (level) {
3370 case 0: return imsm_orom_has_raid0(orom);
3371 case 1:
3372 if (raiddisks > 2)
3373 return imsm_orom_has_raid1e(orom);
3374 return imsm_orom_has_raid1(orom) && raiddisks == 2;
3375 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
3376 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
3377 }
3378 else
3379 return 1; /* not on an Intel RAID platform so anything goes */
3380
3381 return 0;
3382 }
3383
3384 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3385 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3386 * FIX ME add ahci details
3387 */
3388 static int validate_geometry_imsm_volume(struct supertype *st, int level,
3389 int layout, int raiddisks, int chunk,
3390 unsigned long long size, char *dev,
3391 unsigned long long *freesize,
3392 int verbose)
3393 {
3394 struct stat stb;
3395 struct intel_super *super = st->sb;
3396 struct imsm_super *mpb = super->anchor;
3397 struct dl *dl;
3398 unsigned long long pos = 0;
3399 unsigned long long maxsize;
3400 struct extent *e;
3401 int i;
3402
3403 /* We must have the container info already read in. */
3404 if (!super)
3405 return 0;
3406
3407 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
3408 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3409 level, raiddisks, raiddisks > 1 ? "s" : "");
3410 return 0;
3411 }
3412 if (super->orom && level != 1 &&
3413 !imsm_orom_has_chunk(super->orom, chunk)) {
3414 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
3415 return 0;
3416 }
3417 if (layout != imsm_level_to_layout(level)) {
3418 if (level == 5)
3419 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3420 else if (level == 10)
3421 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3422 else
3423 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3424 layout, level);
3425 return 0;
3426 }
3427
3428 if (!dev) {
3429 /* General test: make sure there is space for
3430 * 'raiddisks' device extents of size 'size' at a given
3431 * offset
3432 */
3433 unsigned long long minsize = size;
3434 unsigned long long start_offset = ~0ULL;
3435 int dcnt = 0;
3436 if (minsize == 0)
3437 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
3438 for (dl = super->disks; dl ; dl = dl->next) {
3439 int found = 0;
3440
3441 pos = 0;
3442 i = 0;
3443 e = get_extents(super, dl);
3444 if (!e) continue;
3445 do {
3446 unsigned long long esize;
3447 esize = e[i].start - pos;
3448 if (esize >= minsize)
3449 found = 1;
3450 if (found && start_offset == ~0ULL) {
3451 start_offset = pos;
3452 break;
3453 } else if (found && pos != start_offset) {
3454 found = 0;
3455 break;
3456 }
3457 pos = e[i].start + e[i].size;
3458 i++;
3459 } while (e[i-1].size);
3460 if (found)
3461 dcnt++;
3462 free(e);
3463 }
3464 if (dcnt < raiddisks) {
3465 if (verbose)
3466 fprintf(stderr, Name ": imsm: Not enough "
3467 "devices with space for this array "
3468 "(%d < %d)\n",
3469 dcnt, raiddisks);
3470 return 0;
3471 }
3472 return 1;
3473 }
3474
3475 /* This device must be a member of the set */
3476 if (stat(dev, &stb) < 0)
3477 return 0;
3478 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3479 return 0;
3480 for (dl = super->disks ; dl ; dl = dl->next) {
3481 if (dl->major == major(stb.st_rdev) &&
3482 dl->minor == minor(stb.st_rdev))
3483 break;
3484 }
3485 if (!dl) {
3486 if (verbose)
3487 fprintf(stderr, Name ": %s is not in the "
3488 "same imsm set\n", dev);
3489 return 0;
3490 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
3491 /* If a volume is present then the current creation attempt
3492 * cannot incorporate new spares because the orom may not
3493 * understand this configuration (all member disks must be
3494 * members of each array in the container).
3495 */
3496 fprintf(stderr, Name ": %s is a spare and a volume"
3497 " is already defined for this container\n", dev);
3498 fprintf(stderr, Name ": The option-rom requires all member"
3499 " disks to be a member of all volumes\n");
3500 return 0;
3501 }
3502
3503 /* retrieve the largest free space block */
3504 e = get_extents(super, dl);
3505 maxsize = 0;
3506 i = 0;
3507 if (e) {
3508 do {
3509 unsigned long long esize;
3510
3511 esize = e[i].start - pos;
3512 if (esize >= maxsize)
3513 maxsize = esize;
3514 pos = e[i].start + e[i].size;
3515 i++;
3516 } while (e[i-1].size);
3517 dl->e = e;
3518 dl->extent_cnt = i;
3519 } else {
3520 if (verbose)
3521 fprintf(stderr, Name ": unable to determine free space for: %s\n",
3522 dev);
3523 return 0;
3524 }
3525 if (maxsize < size) {
3526 if (verbose)
3527 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
3528 dev, maxsize, size);
3529 return 0;
3530 }
3531
3532 /* count total number of extents for merge */
3533 i = 0;
3534 for (dl = super->disks; dl; dl = dl->next)
3535 if (dl->e)
3536 i += dl->extent_cnt;
3537
3538 maxsize = merge_extents(super, i);
3539 if (maxsize < size) {
3540 if (verbose)
3541 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
3542 maxsize, size);
3543 return 0;
3544 } else if (maxsize == ~0ULL) {
3545 if (verbose)
3546 fprintf(stderr, Name ": failed to merge %d extents\n", i);
3547 return 0;
3548 }
3549
3550 *freesize = maxsize;
3551
3552 return 1;
3553 }
3554
3555 static int reserve_space(struct supertype *st, int raiddisks,
3556 unsigned long long size, int chunk,
3557 unsigned long long *freesize)
3558 {
3559 struct intel_super *super = st->sb;
3560 struct imsm_super *mpb = super->anchor;
3561 struct dl *dl;
3562 int i;
3563 int extent_cnt;
3564 struct extent *e;
3565 unsigned long long maxsize;
3566 unsigned long long minsize;
3567 int cnt;
3568 int used;
3569
3570 /* find the largest common start free region of the possible disks */
3571 used = 0;
3572 extent_cnt = 0;
3573 cnt = 0;
3574 for (dl = super->disks; dl; dl = dl->next) {
3575 dl->raiddisk = -1;
3576
3577 if (dl->index >= 0)
3578 used++;
3579
3580 /* don't activate new spares if we are orom constrained
3581 * and there is already a volume active in the container
3582 */
3583 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
3584 continue;
3585
3586 e = get_extents(super, dl);
3587 if (!e)
3588 continue;
3589 for (i = 1; e[i-1].size; i++)
3590 ;
3591 dl->e = e;
3592 dl->extent_cnt = i;
3593 extent_cnt += i;
3594 cnt++;
3595 }
3596
3597 maxsize = merge_extents(super, extent_cnt);
3598 minsize = size;
3599 if (size == 0)
3600 minsize = chunk;
3601
3602 if (cnt < raiddisks ||
3603 (super->orom && used && used != raiddisks) ||
3604 maxsize < minsize) {
3605 fprintf(stderr, Name ": not enough devices with space to create array.\n");
3606 return 0; /* No enough free spaces large enough */
3607 }
3608
3609 if (size == 0) {
3610 size = maxsize;
3611 if (chunk) {
3612 size /= chunk;
3613 size *= chunk;
3614 }
3615 }
3616
3617 cnt = 0;
3618 for (dl = super->disks; dl; dl = dl->next)
3619 if (dl->e)
3620 dl->raiddisk = cnt++;
3621
3622 *freesize = size;
3623
3624 return 1;
3625 }
3626
3627 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
3628 int raiddisks, int chunk, unsigned long long size,
3629 char *dev, unsigned long long *freesize,
3630 int verbose)
3631 {
3632 int fd, cfd;
3633 struct mdinfo *sra;
3634
3635 /* if given unused devices create a container
3636 * if given given devices in a container create a member volume
3637 */
3638 if (level == LEVEL_CONTAINER) {
3639 /* Must be a fresh device to add to a container */
3640 return validate_geometry_imsm_container(st, level, layout,
3641 raiddisks, chunk, size,
3642 dev, freesize,
3643 verbose);
3644 }
3645
3646 if (!dev) {
3647 if (st->sb && freesize) {
3648 /* we are being asked to automatically layout a
3649 * new volume based on the current contents of
3650 * the container. If the the parameters can be
3651 * satisfied reserve_space will record the disks,
3652 * start offset, and size of the volume to be
3653 * created. add_to_super and getinfo_super
3654 * detect when autolayout is in progress.
3655 */
3656 return reserve_space(st, raiddisks, size, chunk, freesize);
3657 }
3658 return 1;
3659 }
3660 if (st->sb) {
3661 /* creating in a given container */
3662 return validate_geometry_imsm_volume(st, level, layout,
3663 raiddisks, chunk, size,
3664 dev, freesize, verbose);
3665 }
3666
3667 /* limit creation to the following levels */
3668 if (!dev)
3669 switch (level) {
3670 case 0:
3671 case 1:
3672 case 10:
3673 case 5:
3674 break;
3675 default:
3676 return 1;
3677 }
3678
3679 /* This device needs to be a device in an 'imsm' container */
3680 fd = open(dev, O_RDONLY|O_EXCL, 0);
3681 if (fd >= 0) {
3682 if (verbose)
3683 fprintf(stderr,
3684 Name ": Cannot create this array on device %s\n",
3685 dev);
3686 close(fd);
3687 return 0;
3688 }
3689 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3690 if (verbose)
3691 fprintf(stderr, Name ": Cannot open %s: %s\n",
3692 dev, strerror(errno));
3693 return 0;
3694 }
3695 /* Well, it is in use by someone, maybe an 'imsm' container. */
3696 cfd = open_container(fd);
3697 if (cfd < 0) {
3698 close(fd);
3699 if (verbose)
3700 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3701 dev);
3702 return 0;
3703 }
3704 sra = sysfs_read(cfd, 0, GET_VERSION);
3705 close(fd);
3706 if (sra && sra->array.major_version == -1 &&
3707 strcmp(sra->text_version, "imsm") == 0) {
3708 /* This is a member of a imsm container. Load the container
3709 * and try to create a volume
3710 */
3711 struct intel_super *super;
3712
3713 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3714 st->sb = super;
3715 st->container_dev = fd2devnum(cfd);
3716 close(cfd);
3717 return validate_geometry_imsm_volume(st, level, layout,
3718 raiddisks, chunk,
3719 size, dev,
3720 freesize, verbose);
3721 }
3722 close(cfd);
3723 } else /* may belong to another container */
3724 return 0;
3725
3726 return 1;
3727 }
3728 #endif /* MDASSEMBLE */
3729
3730 static struct mdinfo *container_content_imsm(struct supertype *st)
3731 {
3732 /* Given a container loaded by load_super_imsm_all,
3733 * extract information about all the arrays into
3734 * an mdinfo tree.
3735 *
3736 * For each imsm_dev create an mdinfo, fill it in,
3737 * then look for matching devices in super->disks
3738 * and create appropriate device mdinfo.
3739 */
3740 struct intel_super *super = st->sb;
3741 struct imsm_super *mpb = super->anchor;
3742 struct mdinfo *rest = NULL;
3743 int i;
3744
3745 /* do not assemble arrays that might have bad blocks */
3746 if (imsm_bbm_log_size(super->anchor)) {
3747 fprintf(stderr, Name ": BBM log found in metadata. "
3748 "Cannot activate array(s).\n");
3749 return NULL;
3750 }
3751
3752 for (i = 0; i < mpb->num_raid_devs; i++) {
3753 struct imsm_dev *dev = get_imsm_dev(super, i);
3754 struct imsm_map *map = get_imsm_map(dev, 0);
3755 struct mdinfo *this;
3756 int slot;
3757
3758 /* do not publish arrays that are in the middle of an
3759 * unsupported migration
3760 */
3761 if (dev->vol.migr_state &&
3762 (migr_type(dev) == MIGR_GEN_MIGR ||
3763 migr_type(dev) == MIGR_STATE_CHANGE)) {
3764 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
3765 " unsupported migration in progress\n",
3766 dev->volume);
3767 continue;
3768 }
3769
3770 this = malloc(sizeof(*this));
3771 memset(this, 0, sizeof(*this));
3772 this->next = rest;
3773
3774 super->current_vol = i;
3775 getinfo_super_imsm_volume(st, this);
3776 for (slot = 0 ; slot < map->num_members; slot++) {
3777 struct mdinfo *info_d;
3778 struct dl *d;
3779 int idx;
3780 int skip;
3781 __u32 ord;
3782
3783 skip = 0;
3784 idx = get_imsm_disk_idx(dev, slot);
3785 ord = get_imsm_ord_tbl_ent(dev, slot);
3786 for (d = super->disks; d ; d = d->next)
3787 if (d->index == idx)
3788 break;
3789
3790 if (d == NULL)
3791 skip = 1;
3792 if (d && is_failed(&d->disk))
3793 skip = 1;
3794 if (ord & IMSM_ORD_REBUILD)
3795 skip = 1;
3796
3797 /*
3798 * if we skip some disks the array will be assmebled degraded;
3799 * reset resync start to avoid a dirty-degraded situation
3800 *
3801 * FIXME handle dirty degraded
3802 */
3803 if (skip && !dev->vol.dirty)
3804 this->resync_start = ~0ULL;
3805 if (skip)
3806 continue;
3807
3808 info_d = malloc(sizeof(*info_d));
3809 if (!info_d) {
3810 fprintf(stderr, Name ": failed to allocate disk"
3811 " for volume %.16s\n", dev->volume);
3812 free(this);
3813 this = rest;
3814 break;
3815 }
3816 memset(info_d, 0, sizeof(*info_d));
3817 info_d->next = this->devs;
3818 this->devs = info_d;
3819
3820 info_d->disk.number = d->index;
3821 info_d->disk.major = d->major;
3822 info_d->disk.minor = d->minor;
3823 info_d->disk.raid_disk = slot;
3824
3825 this->array.working_disks++;
3826
3827 info_d->events = __le32_to_cpu(mpb->generation_num);
3828 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3829 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3830 if (d->devname)
3831 strcpy(info_d->name, d->devname);
3832 }
3833 rest = this;
3834 }
3835
3836 return rest;
3837 }
3838
3839
3840 #ifndef MDASSEMBLE
3841 static int imsm_open_new(struct supertype *c, struct active_array *a,
3842 char *inst)
3843 {
3844 struct intel_super *super = c->sb;
3845 struct imsm_super *mpb = super->anchor;
3846
3847 if (atoi(inst) >= mpb->num_raid_devs) {
3848 fprintf(stderr, "%s: subarry index %d, out of range\n",
3849 __func__, atoi(inst));
3850 return -ENODEV;
3851 }
3852
3853 dprintf("imsm: open_new %s\n", inst);
3854 a->info.container_member = atoi(inst);
3855 return 0;
3856 }
3857
3858 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3859 {
3860 struct imsm_map *map = get_imsm_map(dev, 0);
3861
3862 if (!failed)
3863 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3864 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3865
3866 switch (get_imsm_raid_level(map)) {
3867 case 0:
3868 return IMSM_T_STATE_FAILED;
3869 break;
3870 case 1:
3871 if (failed < map->num_members)
3872 return IMSM_T_STATE_DEGRADED;
3873 else
3874 return IMSM_T_STATE_FAILED;
3875 break;
3876 case 10:
3877 {
3878 /**
3879 * check to see if any mirrors have failed, otherwise we
3880 * are degraded. Even numbered slots are mirrored on
3881 * slot+1
3882 */
3883 int i;
3884 /* gcc -Os complains that this is unused */
3885 int insync = insync;
3886
3887 for (i = 0; i < map->num_members; i++) {
3888 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3889 int idx = ord_to_idx(ord);
3890 struct imsm_disk *disk;
3891
3892 /* reset the potential in-sync count on even-numbered
3893 * slots. num_copies is always 2 for imsm raid10
3894 */
3895 if ((i & 1) == 0)
3896 insync = 2;
3897
3898 disk = get_imsm_disk(super, idx);
3899 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
3900 insync--;
3901
3902 /* no in-sync disks left in this mirror the
3903 * array has failed
3904 */
3905 if (insync == 0)
3906 return IMSM_T_STATE_FAILED;
3907 }
3908
3909 return IMSM_T_STATE_DEGRADED;
3910 }
3911 case 5:
3912 if (failed < 2)
3913 return IMSM_T_STATE_DEGRADED;
3914 else
3915 return IMSM_T_STATE_FAILED;
3916 break;
3917 default:
3918 break;
3919 }
3920
3921 return map->map_state;
3922 }
3923
3924 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3925 {
3926 int i;
3927 int failed = 0;
3928 struct imsm_disk *disk;
3929 struct imsm_map *map = get_imsm_map(dev, 0);
3930 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3931 __u32 ord;
3932 int idx;
3933
3934 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3935 * disks that are being rebuilt. New failures are recorded to
3936 * map[0]. So we look through all the disks we started with and
3937 * see if any failures are still present, or if any new ones
3938 * have arrived
3939 *
3940 * FIXME add support for online capacity expansion and
3941 * raid-level-migration
3942 */
3943 for (i = 0; i < prev->num_members; i++) {
3944 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
3945 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
3946 idx = ord_to_idx(ord);
3947
3948 disk = get_imsm_disk(super, idx);
3949 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
3950 failed++;
3951 }
3952
3953 return failed;
3954 }
3955
3956 static int is_resyncing(struct imsm_dev *dev)
3957 {
3958 struct imsm_map *migr_map;
3959
3960 if (!dev->vol.migr_state)
3961 return 0;
3962
3963 if (migr_type(dev) == MIGR_INIT ||
3964 migr_type(dev) == MIGR_REPAIR)
3965 return 1;
3966
3967 migr_map = get_imsm_map(dev, 1);
3968
3969 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
3970 return 1;
3971 else
3972 return 0;
3973 }
3974
3975 static int is_rebuilding(struct imsm_dev *dev)
3976 {
3977 struct imsm_map *migr_map;
3978
3979 if (!dev->vol.migr_state)
3980 return 0;
3981
3982 if (migr_type(dev) != MIGR_REBUILD)
3983 return 0;
3984
3985 migr_map = get_imsm_map(dev, 1);
3986
3987 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
3988 return 1;
3989 else
3990 return 0;
3991 }
3992
3993 /* return true if we recorded new information */
3994 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
3995 {
3996 __u32 ord;
3997 int slot;
3998 struct imsm_map *map;
3999
4000 /* new failures are always set in map[0] */
4001 map = get_imsm_map(dev, 0);
4002
4003 slot = get_imsm_disk_slot(map, idx);
4004 if (slot < 0)
4005 return 0;
4006
4007 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
4008 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
4009 return 0;
4010
4011 disk->status |= FAILED_DISK;
4012 disk->status &= ~CONFIGURED_DISK;
4013 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
4014 if (~map->failed_disk_num == 0)
4015 map->failed_disk_num = slot;
4016 return 1;
4017 }
4018
4019 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4020 {
4021 mark_failure(dev, disk, idx);
4022
4023 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
4024 return;
4025
4026 disk->scsi_id = __cpu_to_le32(~(__u32)0);
4027 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
4028 }
4029
4030 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4031 * states are handled in imsm_set_disk() with one exception, when a
4032 * resync is stopped due to a new failure this routine will set the
4033 * 'degraded' state for the array.
4034 */
4035 static int imsm_set_array_state(struct active_array *a, int consistent)
4036 {
4037 int inst = a->info.container_member;
4038 struct intel_super *super = a->container->sb;
4039 struct imsm_dev *dev = get_imsm_dev(super, inst);
4040 struct imsm_map *map = get_imsm_map(dev, 0);
4041 int failed = imsm_count_failed(super, dev);
4042 __u8 map_state = imsm_check_degraded(super, dev, failed);
4043
4044 /* before we activate this array handle any missing disks */
4045 if (consistent == 2 && super->missing) {
4046 struct dl *dl;
4047
4048 dprintf("imsm: mark missing\n");
4049 end_migration(dev, map_state);
4050 for (dl = super->missing; dl; dl = dl->next)
4051 mark_missing(dev, &dl->disk, dl->index);
4052 super->updates_pending++;
4053 }
4054
4055 if (consistent == 2 &&
4056 (!is_resync_complete(a) ||
4057 map_state != IMSM_T_STATE_NORMAL ||
4058 dev->vol.migr_state))
4059 consistent = 0;
4060
4061 if (is_resync_complete(a)) {
4062 /* complete intialization / resync,
4063 * recovery and interrupted recovery is completed in
4064 * ->set_disk
4065 */
4066 if (is_resyncing(dev)) {
4067 dprintf("imsm: mark resync done\n");
4068 end_migration(dev, map_state);
4069 super->updates_pending++;
4070 }
4071 } else if (!is_resyncing(dev) && !failed) {
4072 /* mark the start of the init process if nothing is failed */
4073 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
4074 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
4075 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
4076 else
4077 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
4078 super->updates_pending++;
4079 }
4080
4081 /* FIXME check if we can update curr_migr_unit from resync_start */
4082
4083 /* mark dirty / clean */
4084 if (dev->vol.dirty != !consistent) {
4085 dprintf("imsm: mark '%s' (%llu)\n",
4086 consistent ? "clean" : "dirty", a->resync_start);
4087 if (consistent)
4088 dev->vol.dirty = 0;
4089 else
4090 dev->vol.dirty = 1;
4091 super->updates_pending++;
4092 }
4093 return consistent;
4094 }
4095
4096 static void imsm_set_disk(struct active_array *a, int n, int state)
4097 {
4098 int inst = a->info.container_member;
4099 struct intel_super *super = a->container->sb;
4100 struct imsm_dev *dev = get_imsm_dev(super, inst);
4101 struct imsm_map *map = get_imsm_map(dev, 0);
4102 struct imsm_disk *disk;
4103 int failed;
4104 __u32 ord;
4105 __u8 map_state;
4106
4107 if (n > map->num_members)
4108 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
4109 n, map->num_members - 1);
4110
4111 if (n < 0)
4112 return;
4113
4114 dprintf("imsm: set_disk %d:%x\n", n, state);
4115
4116 ord = get_imsm_ord_tbl_ent(dev, n);
4117 disk = get_imsm_disk(super, ord_to_idx(ord));
4118
4119 /* check for new failures */
4120 if (state & DS_FAULTY) {
4121 if (mark_failure(dev, disk, ord_to_idx(ord)))
4122 super->updates_pending++;
4123 }
4124
4125 /* check if in_sync */
4126 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
4127 struct imsm_map *migr_map = get_imsm_map(dev, 1);
4128
4129 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
4130 super->updates_pending++;
4131 }
4132
4133 failed = imsm_count_failed(super, dev);
4134 map_state = imsm_check_degraded(super, dev, failed);
4135
4136 /* check if recovery complete, newly degraded, or failed */
4137 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
4138 end_migration(dev, map_state);
4139 map = get_imsm_map(dev, 0);
4140 map->failed_disk_num = ~0;
4141 super->updates_pending++;
4142 } else if (map_state == IMSM_T_STATE_DEGRADED &&
4143 map->map_state != map_state &&
4144 !dev->vol.migr_state) {
4145 dprintf("imsm: mark degraded\n");
4146 map->map_state = map_state;
4147 super->updates_pending++;
4148 } else if (map_state == IMSM_T_STATE_FAILED &&
4149 map->map_state != map_state) {
4150 dprintf("imsm: mark failed\n");
4151 end_migration(dev, map_state);
4152 super->updates_pending++;
4153 }
4154 }
4155
4156 static int store_imsm_mpb(int fd, struct imsm_super *mpb)
4157 {
4158 void *buf = mpb;
4159 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
4160 unsigned long long dsize;
4161 unsigned long long sectors;
4162
4163 get_dev_size(fd, NULL, &dsize);
4164
4165 if (mpb_size > 512) {
4166 /* -1 to account for anchor */
4167 sectors = mpb_sectors(mpb) - 1;
4168
4169 /* write the extended mpb to the sectors preceeding the anchor */
4170 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
4171 return 1;
4172
4173 if (write(fd, buf + 512, 512 * sectors) != 512 * sectors)
4174 return 1;
4175 }
4176
4177 /* first block is stored on second to last sector of the disk */
4178 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
4179 return 1;
4180
4181 if (write(fd, buf, 512) != 512)
4182 return 1;
4183
4184 return 0;
4185 }
4186
4187 static void imsm_sync_metadata(struct supertype *container)
4188 {
4189 struct intel_super *super = container->sb;
4190
4191 if (!super->updates_pending)
4192 return;
4193
4194 write_super_imsm(super, 0);
4195
4196 super->updates_pending = 0;
4197 }
4198
4199 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
4200 {
4201 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4202 int i = get_imsm_disk_idx(dev, idx);
4203 struct dl *dl;
4204
4205 for (dl = super->disks; dl; dl = dl->next)
4206 if (dl->index == i)
4207 break;
4208
4209 if (dl && is_failed(&dl->disk))
4210 dl = NULL;
4211
4212 if (dl)
4213 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
4214
4215 return dl;
4216 }
4217
4218 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
4219 struct active_array *a, int activate_new)
4220 {
4221 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4222 int idx = get_imsm_disk_idx(dev, slot);
4223 struct imsm_super *mpb = super->anchor;
4224 struct imsm_map *map;
4225 unsigned long long pos;
4226 struct mdinfo *d;
4227 struct extent *ex;
4228 int i, j;
4229 int found;
4230 __u32 array_start;
4231 __u32 array_end;
4232 struct dl *dl;
4233
4234 for (dl = super->disks; dl; dl = dl->next) {
4235 /* If in this array, skip */
4236 for (d = a->info.devs ; d ; d = d->next)
4237 if (d->state_fd >= 0 &&
4238 d->disk.major == dl->major &&
4239 d->disk.minor == dl->minor) {
4240 dprintf("%x:%x already in array\n", dl->major, dl->minor);
4241 break;
4242 }
4243 if (d)
4244 continue;
4245
4246 /* skip in use or failed drives */
4247 if (is_failed(&dl->disk) || idx == dl->index ||
4248 dl->index == -2) {
4249 dprintf("%x:%x status (failed: %d index: %d)\n",
4250 dl->major, dl->minor, is_failed(&dl->disk), idx);
4251 continue;
4252 }
4253
4254 /* skip pure spares when we are looking for partially
4255 * assimilated drives
4256 */
4257 if (dl->index == -1 && !activate_new)
4258 continue;
4259
4260 /* Does this unused device have the requisite free space?
4261 * It needs to be able to cover all member volumes
4262 */
4263 ex = get_extents(super, dl);
4264 if (!ex) {
4265 dprintf("cannot get extents\n");
4266 continue;
4267 }
4268 for (i = 0; i < mpb->num_raid_devs; i++) {
4269 dev = get_imsm_dev(super, i);
4270 map = get_imsm_map(dev, 0);
4271
4272 /* check if this disk is already a member of
4273 * this array
4274 */
4275 if (get_imsm_disk_slot(map, dl->index) >= 0)
4276 continue;
4277
4278 found = 0;
4279 j = 0;
4280 pos = 0;
4281 array_start = __le32_to_cpu(map->pba_of_lba0);
4282 array_end = array_start +
4283 __le32_to_cpu(map->blocks_per_member) - 1;
4284
4285 do {
4286 /* check that we can start at pba_of_lba0 with
4287 * blocks_per_member of space
4288 */
4289 if (array_start >= pos && array_end < ex[j].start) {
4290 found = 1;
4291 break;
4292 }
4293 pos = ex[j].start + ex[j].size;
4294 j++;
4295 } while (ex[j-1].size);
4296
4297 if (!found)
4298 break;
4299 }
4300
4301 free(ex);
4302 if (i < mpb->num_raid_devs) {
4303 dprintf("%x:%x does not have %u to %u available\n",
4304 dl->major, dl->minor, array_start, array_end);
4305 /* No room */
4306 continue;
4307 }
4308 return dl;
4309 }
4310
4311 return dl;
4312 }
4313
4314 static struct mdinfo *imsm_activate_spare(struct active_array *a,
4315 struct metadata_update **updates)
4316 {
4317 /**
4318 * Find a device with unused free space and use it to replace a
4319 * failed/vacant region in an array. We replace failed regions one a
4320 * array at a time. The result is that a new spare disk will be added
4321 * to the first failed array and after the monitor has finished
4322 * propagating failures the remainder will be consumed.
4323 *
4324 * FIXME add a capability for mdmon to request spares from another
4325 * container.
4326 */
4327
4328 struct intel_super *super = a->container->sb;
4329 int inst = a->info.container_member;
4330 struct imsm_dev *dev = get_imsm_dev(super, inst);
4331 struct imsm_map *map = get_imsm_map(dev, 0);
4332 int failed = a->info.array.raid_disks;
4333 struct mdinfo *rv = NULL;
4334 struct mdinfo *d;
4335 struct mdinfo *di;
4336 struct metadata_update *mu;
4337 struct dl *dl;
4338 struct imsm_update_activate_spare *u;
4339 int num_spares = 0;
4340 int i;
4341
4342 for (d = a->info.devs ; d ; d = d->next) {
4343 if ((d->curr_state & DS_FAULTY) &&
4344 d->state_fd >= 0)
4345 /* wait for Removal to happen */
4346 return NULL;
4347 if (d->state_fd >= 0)
4348 failed--;
4349 }
4350
4351 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4352 inst, failed, a->info.array.raid_disks, a->info.array.level);
4353 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
4354 return NULL;
4355
4356 /* For each slot, if it is not working, find a spare */
4357 for (i = 0; i < a->info.array.raid_disks; i++) {
4358 for (d = a->info.devs ; d ; d = d->next)
4359 if (d->disk.raid_disk == i)
4360 break;
4361 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
4362 if (d && (d->state_fd >= 0))
4363 continue;
4364
4365 /*
4366 * OK, this device needs recovery. Try to re-add the
4367 * previous occupant of this slot, if this fails see if
4368 * we can continue the assimilation of a spare that was
4369 * partially assimilated, finally try to activate a new
4370 * spare.
4371 */
4372 dl = imsm_readd(super, i, a);
4373 if (!dl)
4374 dl = imsm_add_spare(super, i, a, 0);
4375 if (!dl)
4376 dl = imsm_add_spare(super, i, a, 1);
4377 if (!dl)
4378 continue;
4379
4380 /* found a usable disk with enough space */
4381 di = malloc(sizeof(*di));
4382 if (!di)
4383 continue;
4384 memset(di, 0, sizeof(*di));
4385
4386 /* dl->index will be -1 in the case we are activating a
4387 * pristine spare. imsm_process_update() will create a
4388 * new index in this case. Once a disk is found to be
4389 * failed in all member arrays it is kicked from the
4390 * metadata
4391 */
4392 di->disk.number = dl->index;
4393
4394 /* (ab)use di->devs to store a pointer to the device
4395 * we chose
4396 */
4397 di->devs = (struct mdinfo *) dl;
4398
4399 di->disk.raid_disk = i;
4400 di->disk.major = dl->major;
4401 di->disk.minor = dl->minor;
4402 di->disk.state = 0;
4403 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
4404 di->component_size = a->info.component_size;
4405 di->container_member = inst;
4406 super->random = random32();
4407 di->next = rv;
4408 rv = di;
4409 num_spares++;
4410 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
4411 i, di->data_offset);
4412
4413 break;
4414 }
4415
4416 if (!rv)
4417 /* No spares found */
4418 return rv;
4419 /* Now 'rv' has a list of devices to return.
4420 * Create a metadata_update record to update the
4421 * disk_ord_tbl for the array
4422 */
4423 mu = malloc(sizeof(*mu));
4424 if (mu) {
4425 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
4426 if (mu->buf == NULL) {
4427 free(mu);
4428 mu = NULL;
4429 }
4430 }
4431 if (!mu) {
4432 while (rv) {
4433 struct mdinfo *n = rv->next;
4434
4435 free(rv);
4436 rv = n;
4437 }
4438 return NULL;
4439 }
4440
4441 mu->space = NULL;
4442 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
4443 mu->next = *updates;
4444 u = (struct imsm_update_activate_spare *) mu->buf;
4445
4446 for (di = rv ; di ; di = di->next) {
4447 u->type = update_activate_spare;
4448 u->dl = (struct dl *) di->devs;
4449 di->devs = NULL;
4450 u->slot = di->disk.raid_disk;
4451 u->array = inst;
4452 u->next = u + 1;
4453 u++;
4454 }
4455 (u-1)->next = NULL;
4456 *updates = mu;
4457
4458 return rv;
4459 }
4460
4461 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
4462 {
4463 struct imsm_dev *dev = get_imsm_dev(super, idx);
4464 struct imsm_map *map = get_imsm_map(dev, 0);
4465 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
4466 struct disk_info *inf = get_disk_info(u);
4467 struct imsm_disk *disk;
4468 int i;
4469 int j;
4470
4471 for (i = 0; i < map->num_members; i++) {
4472 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4473 for (j = 0; j < new_map->num_members; j++)
4474 if (serialcmp(disk->serial, inf[j].serial) == 0)
4475 return 1;
4476 }
4477
4478 return 0;
4479 }
4480
4481 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
4482
4483 static void imsm_process_update(struct supertype *st,
4484 struct metadata_update *update)
4485 {
4486 /**
4487 * crack open the metadata_update envelope to find the update record
4488 * update can be one of:
4489 * update_activate_spare - a spare device has replaced a failed
4490 * device in an array, update the disk_ord_tbl. If this disk is
4491 * present in all member arrays then also clear the SPARE_DISK
4492 * flag
4493 */
4494 struct intel_super *super = st->sb;
4495 struct imsm_super *mpb;
4496 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4497
4498 /* update requires a larger buf but the allocation failed */
4499 if (super->next_len && !super->next_buf) {
4500 super->next_len = 0;
4501 return;
4502 }
4503
4504 if (super->next_buf) {
4505 memcpy(super->next_buf, super->buf, super->len);
4506 free(super->buf);
4507 super->len = super->next_len;
4508 super->buf = super->next_buf;
4509
4510 super->next_len = 0;
4511 super->next_buf = NULL;
4512 }
4513
4514 mpb = super->anchor;
4515
4516 switch (type) {
4517 case update_activate_spare: {
4518 struct imsm_update_activate_spare *u = (void *) update->buf;
4519 struct imsm_dev *dev = get_imsm_dev(super, u->array);
4520 struct imsm_map *map = get_imsm_map(dev, 0);
4521 struct imsm_map *migr_map;
4522 struct active_array *a;
4523 struct imsm_disk *disk;
4524 __u8 to_state;
4525 struct dl *dl;
4526 unsigned int found;
4527 int failed;
4528 int victim = get_imsm_disk_idx(dev, u->slot);
4529 int i;
4530
4531 for (dl = super->disks; dl; dl = dl->next)
4532 if (dl == u->dl)
4533 break;
4534
4535 if (!dl) {
4536 fprintf(stderr, "error: imsm_activate_spare passed "
4537 "an unknown disk (index: %d)\n",
4538 u->dl->index);
4539 return;
4540 }
4541
4542 super->updates_pending++;
4543
4544 /* count failures (excluding rebuilds and the victim)
4545 * to determine map[0] state
4546 */
4547 failed = 0;
4548 for (i = 0; i < map->num_members; i++) {
4549 if (i == u->slot)
4550 continue;
4551 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4552 if (!disk || is_failed(disk))
4553 failed++;
4554 }
4555
4556 /* adding a pristine spare, assign a new index */
4557 if (dl->index < 0) {
4558 dl->index = super->anchor->num_disks;
4559 super->anchor->num_disks++;
4560 }
4561 disk = &dl->disk;
4562 disk->status |= CONFIGURED_DISK;
4563 disk->status &= ~SPARE_DISK;
4564
4565 /* mark rebuild */
4566 to_state = imsm_check_degraded(super, dev, failed);
4567 map->map_state = IMSM_T_STATE_DEGRADED;
4568 migrate(dev, to_state, MIGR_REBUILD);
4569 migr_map = get_imsm_map(dev, 1);
4570 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
4571 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
4572
4573 /* update the family_num to mark a new container
4574 * generation, being careful to record the existing
4575 * family_num in orig_family_num to clean up after
4576 * earlier mdadm versions that neglected to set it.
4577 */
4578 if (mpb->orig_family_num == 0)
4579 mpb->orig_family_num = mpb->family_num;
4580 mpb->family_num += super->random;
4581
4582 /* count arrays using the victim in the metadata */
4583 found = 0;
4584 for (a = st->arrays; a ; a = a->next) {
4585 dev = get_imsm_dev(super, a->info.container_member);
4586 map = get_imsm_map(dev, 0);
4587
4588 if (get_imsm_disk_slot(map, victim) >= 0)
4589 found++;
4590 }
4591
4592 /* delete the victim if it is no longer being
4593 * utilized anywhere
4594 */
4595 if (!found) {
4596 struct dl **dlp;
4597
4598 /* We know that 'manager' isn't touching anything,
4599 * so it is safe to delete
4600 */
4601 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
4602 if ((*dlp)->index == victim)
4603 break;
4604
4605 /* victim may be on the missing list */
4606 if (!*dlp)
4607 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
4608 if ((*dlp)->index == victim)
4609 break;
4610 imsm_delete(super, dlp, victim);
4611 }
4612 break;
4613 }
4614 case update_create_array: {
4615 /* someone wants to create a new array, we need to be aware of
4616 * a few races/collisions:
4617 * 1/ 'Create' called by two separate instances of mdadm
4618 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4619 * devices that have since been assimilated via
4620 * activate_spare.
4621 * In the event this update can not be carried out mdadm will
4622 * (FIX ME) notice that its update did not take hold.
4623 */
4624 struct imsm_update_create_array *u = (void *) update->buf;
4625 struct intel_dev *dv;
4626 struct imsm_dev *dev;
4627 struct imsm_map *map, *new_map;
4628 unsigned long long start, end;
4629 unsigned long long new_start, new_end;
4630 int i;
4631 struct disk_info *inf;
4632 struct dl *dl;
4633
4634 /* handle racing creates: first come first serve */
4635 if (u->dev_idx < mpb->num_raid_devs) {
4636 dprintf("%s: subarray %d already defined\n",
4637 __func__, u->dev_idx);
4638 goto create_error;
4639 }
4640
4641 /* check update is next in sequence */
4642 if (u->dev_idx != mpb->num_raid_devs) {
4643 dprintf("%s: can not create array %d expected index %d\n",
4644 __func__, u->dev_idx, mpb->num_raid_devs);
4645 goto create_error;
4646 }
4647
4648 new_map = get_imsm_map(&u->dev, 0);
4649 new_start = __le32_to_cpu(new_map->pba_of_lba0);
4650 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
4651 inf = get_disk_info(u);
4652
4653 /* handle activate_spare versus create race:
4654 * check to make sure that overlapping arrays do not include
4655 * overalpping disks
4656 */
4657 for (i = 0; i < mpb->num_raid_devs; i++) {
4658 dev = get_imsm_dev(super, i);
4659 map = get_imsm_map(dev, 0);
4660 start = __le32_to_cpu(map->pba_of_lba0);
4661 end = start + __le32_to_cpu(map->blocks_per_member);
4662 if ((new_start >= start && new_start <= end) ||
4663 (start >= new_start && start <= new_end))
4664 /* overlap */;
4665 else
4666 continue;
4667
4668 if (disks_overlap(super, i, u)) {
4669 dprintf("%s: arrays overlap\n", __func__);
4670 goto create_error;
4671 }
4672 }
4673
4674 /* check that prepare update was successful */
4675 if (!update->space) {
4676 dprintf("%s: prepare update failed\n", __func__);
4677 goto create_error;
4678 }
4679
4680 /* check that all disks are still active before committing
4681 * changes. FIXME: could we instead handle this by creating a
4682 * degraded array? That's probably not what the user expects,
4683 * so better to drop this update on the floor.
4684 */
4685 for (i = 0; i < new_map->num_members; i++) {
4686 dl = serial_to_dl(inf[i].serial, super);
4687 if (!dl) {
4688 dprintf("%s: disk disappeared\n", __func__);
4689 goto create_error;
4690 }
4691 }
4692
4693 super->updates_pending++;
4694
4695 /* convert spares to members and fixup ord_tbl */
4696 for (i = 0; i < new_map->num_members; i++) {
4697 dl = serial_to_dl(inf[i].serial, super);
4698 if (dl->index == -1) {
4699 dl->index = mpb->num_disks;
4700 mpb->num_disks++;
4701 dl->disk.status |= CONFIGURED_DISK;
4702 dl->disk.status &= ~SPARE_DISK;
4703 }
4704 set_imsm_ord_tbl_ent(new_map, i, dl->index);
4705 }
4706
4707 dv = update->space;
4708 dev = dv->dev;
4709 update->space = NULL;
4710 imsm_copy_dev(dev, &u->dev);
4711 dv->index = u->dev_idx;
4712 dv->next = super->devlist;
4713 super->devlist = dv;
4714 mpb->num_raid_devs++;
4715
4716 imsm_update_version_info(super);
4717 break;
4718 create_error:
4719 /* mdmon knows how to release update->space, but not
4720 * ((struct intel_dev *) update->space)->dev
4721 */
4722 if (update->space) {
4723 dv = update->space;
4724 free(dv->dev);
4725 }
4726 break;
4727 }
4728 case update_add_disk:
4729
4730 /* we may be able to repair some arrays if disks are
4731 * being added */
4732 if (super->add) {
4733 struct active_array *a;
4734
4735 super->updates_pending++;
4736 for (a = st->arrays; a; a = a->next)
4737 a->check_degraded = 1;
4738 }
4739 /* add some spares to the metadata */
4740 while (super->add) {
4741 struct dl *al;
4742
4743 al = super->add;
4744 super->add = al->next;
4745 al->next = super->disks;
4746 super->disks = al;
4747 dprintf("%s: added %x:%x\n",
4748 __func__, al->major, al->minor);
4749 }
4750
4751 break;
4752 }
4753 }
4754
4755 static void imsm_prepare_update(struct supertype *st,
4756 struct metadata_update *update)
4757 {
4758 /**
4759 * Allocate space to hold new disk entries, raid-device entries or a new
4760 * mpb if necessary. The manager synchronously waits for updates to
4761 * complete in the monitor, so new mpb buffers allocated here can be
4762 * integrated by the monitor thread without worrying about live pointers
4763 * in the manager thread.
4764 */
4765 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4766 struct intel_super *super = st->sb;
4767 struct imsm_super *mpb = super->anchor;
4768 size_t buf_len;
4769 size_t len = 0;
4770
4771 switch (type) {
4772 case update_create_array: {
4773 struct imsm_update_create_array *u = (void *) update->buf;
4774 struct intel_dev *dv;
4775 struct imsm_dev *dev = &u->dev;
4776 struct imsm_map *map = get_imsm_map(dev, 0);
4777 struct dl *dl;
4778 struct disk_info *inf;
4779 int i;
4780 int activate = 0;
4781
4782 inf = get_disk_info(u);
4783 len = sizeof_imsm_dev(dev, 1);
4784 /* allocate a new super->devlist entry */
4785 dv = malloc(sizeof(*dv));
4786 if (dv) {
4787 dv->dev = malloc(len);
4788 if (dv->dev)
4789 update->space = dv;
4790 else {
4791 free(dv);
4792 update->space = NULL;
4793 }
4794 }
4795
4796 /* count how many spares will be converted to members */
4797 for (i = 0; i < map->num_members; i++) {
4798 dl = serial_to_dl(inf[i].serial, super);
4799 if (!dl) {
4800 /* hmm maybe it failed?, nothing we can do about
4801 * it here
4802 */
4803 continue;
4804 }
4805 if (count_memberships(dl, super) == 0)
4806 activate++;
4807 }
4808 len += activate * sizeof(struct imsm_disk);
4809 break;
4810 default:
4811 break;
4812 }
4813 }
4814
4815 /* check if we need a larger metadata buffer */
4816 if (super->next_buf)
4817 buf_len = super->next_len;
4818 else
4819 buf_len = super->len;
4820
4821 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4822 /* ok we need a larger buf than what is currently allocated
4823 * if this allocation fails process_update will notice that
4824 * ->next_len is set and ->next_buf is NULL
4825 */
4826 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4827 if (super->next_buf)
4828 free(super->next_buf);
4829
4830 super->next_len = buf_len;
4831 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
4832 memset(super->next_buf, 0, buf_len);
4833 else
4834 super->next_buf = NULL;
4835 }
4836 }
4837
4838 /* must be called while manager is quiesced */
4839 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4840 {
4841 struct imsm_super *mpb = super->anchor;
4842 struct dl *iter;
4843 struct imsm_dev *dev;
4844 struct imsm_map *map;
4845 int i, j, num_members;
4846 __u32 ord;
4847
4848 dprintf("%s: deleting device[%d] from imsm_super\n",
4849 __func__, index);
4850
4851 /* shift all indexes down one */
4852 for (iter = super->disks; iter; iter = iter->next)
4853 if (iter->index > index)
4854 iter->index--;
4855 for (iter = super->missing; iter; iter = iter->next)
4856 if (iter->index > index)
4857 iter->index--;
4858
4859 for (i = 0; i < mpb->num_raid_devs; i++) {
4860 dev = get_imsm_dev(super, i);
4861 map = get_imsm_map(dev, 0);
4862 num_members = map->num_members;
4863 for (j = 0; j < num_members; j++) {
4864 /* update ord entries being careful not to propagate
4865 * ord-flags to the first map
4866 */
4867 ord = get_imsm_ord_tbl_ent(dev, j);
4868
4869 if (ord_to_idx(ord) <= index)
4870 continue;
4871
4872 map = get_imsm_map(dev, 0);
4873 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4874 map = get_imsm_map(dev, 1);
4875 if (map)
4876 set_imsm_ord_tbl_ent(map, j, ord - 1);
4877 }
4878 }
4879
4880 mpb->num_disks--;
4881 super->updates_pending++;
4882 if (*dlp) {
4883 struct dl *dl = *dlp;
4884
4885 *dlp = (*dlp)->next;
4886 __free_imsm_disk(dl);
4887 }
4888 }
4889 #endif /* MDASSEMBLE */
4890
4891 struct superswitch super_imsm = {
4892 #ifndef MDASSEMBLE
4893 .examine_super = examine_super_imsm,
4894 .brief_examine_super = brief_examine_super_imsm,
4895 .brief_examine_subarrays = brief_examine_subarrays_imsm,
4896 .export_examine_super = export_examine_super_imsm,
4897 .detail_super = detail_super_imsm,
4898 .brief_detail_super = brief_detail_super_imsm,
4899 .write_init_super = write_init_super_imsm,
4900 .validate_geometry = validate_geometry_imsm,
4901 .add_to_super = add_to_super_imsm,
4902 .detail_platform = detail_platform_imsm,
4903 #endif
4904 .match_home = match_home_imsm,
4905 .uuid_from_super= uuid_from_super_imsm,
4906 .getinfo_super = getinfo_super_imsm,
4907 .update_super = update_super_imsm,
4908
4909 .avail_size = avail_size_imsm,
4910
4911 .compare_super = compare_super_imsm,
4912
4913 .load_super = load_super_imsm,
4914 .init_super = init_super_imsm,
4915 .store_super = store_super_imsm,
4916 .free_super = free_super_imsm,
4917 .match_metadata_desc = match_metadata_desc_imsm,
4918 .container_content = container_content_imsm,
4919 .default_layout = imsm_level_to_layout,
4920
4921 .external = 1,
4922 .name = "imsm",
4923
4924 #ifndef MDASSEMBLE
4925 /* for mdmon */
4926 .open_new = imsm_open_new,
4927 .load_super = load_super_imsm,
4928 .set_array_state= imsm_set_array_state,
4929 .set_disk = imsm_set_disk,
4930 .sync_metadata = imsm_sync_metadata,
4931 .activate_spare = imsm_activate_spare,
4932 .process_update = imsm_process_update,
4933 .prepare_update = imsm_prepare_update,
4934 #endif /* MDASSEMBLE */
4935 };
Unnamed repository; edit this file 'description' to name the repository.