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