]> 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
Create: warn when a metadata format's platform components are missing
[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, int enumerate_only)
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 if (enumerate_only) {
941 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
942 return 0;
943 return 2;
944 }
945
946 list = find_driver_devices("pci", "ahci");
947 for (hba = list; hba; hba = hba->next)
948 if (devpath_to_vendor(hba->path) == 0x8086)
949 break;
950
951 if (!hba) {
952 if (verbose)
953 fprintf(stderr, Name ": unable to find active ahci controller\n");
954 free_sys_dev(&list);
955 return 2;
956 } else if (verbose)
957 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
958 hba_path = hba->path;
959 hba->path = NULL;
960 free_sys_dev(&list);
961
962 orom = find_imsm_orom();
963 if (!orom) {
964 if (verbose)
965 fprintf(stderr, Name ": imsm option-rom not found\n");
966 return 2;
967 }
968
969 printf(" Platform : Intel(R) Matrix Storage Manager\n");
970 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
971 orom->hotfix_ver, orom->build);
972 printf(" RAID Levels :%s%s%s%s%s\n",
973 imsm_orom_has_raid0(orom) ? " raid0" : "",
974 imsm_orom_has_raid1(orom) ? " raid1" : "",
975 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
976 imsm_orom_has_raid10(orom) ? " raid10" : "",
977 imsm_orom_has_raid5(orom) ? " raid5" : "");
978 printf(" Max Disks : %d\n", orom->tds);
979 printf(" Max Volumes : %d\n", orom->vpa);
980 printf(" I/O Controller : %s\n", hba_path);
981
982 /* find the smallest scsi host number to determine a port number base */
983 dir = opendir(hba_path);
984 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
985 int host;
986
987 if (sscanf(ent->d_name, "host%d", &host) != 1)
988 continue;
989 if (port_count == 0)
990 host_base = host;
991 else if (host < host_base)
992 host_base = host;
993
994 if (host + 1 > port_count + host_base)
995 port_count = host + 1 - host_base;
996
997 }
998 if (dir)
999 closedir(dir);
1000
1001 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1002 host_base, verbose) != 0) {
1003 if (verbose)
1004 fprintf(stderr, Name ": failed to enumerate ports\n");
1005 return 2;
1006 }
1007
1008 return 0;
1009 }
1010 #endif
1011
1012 static int match_home_imsm(struct supertype *st, char *homehost)
1013 {
1014 /* the imsm metadata format does not specify any host
1015 * identification information. We return -1 since we can never
1016 * confirm nor deny whether a given array is "meant" for this
1017 * host. We rely on compare_super and the 'family_num' field to
1018 * exclude member disks that do not belong, and we rely on
1019 * mdadm.conf to specify the arrays that should be assembled.
1020 * Auto-assembly may still pick up "foreign" arrays.
1021 */
1022
1023 return -1;
1024 }
1025
1026 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1027 {
1028 /* The uuid returned here is used for:
1029 * uuid to put into bitmap file (Create, Grow)
1030 * uuid for backup header when saving critical section (Grow)
1031 * comparing uuids when re-adding a device into an array
1032 * In these cases the uuid required is that of the data-array,
1033 * not the device-set.
1034 * uuid to recognise same set when adding a missing device back
1035 * to an array. This is a uuid for the device-set.
1036 *
1037 * For each of these we can make do with a truncated
1038 * or hashed uuid rather than the original, as long as
1039 * everyone agrees.
1040 * In each case the uuid required is that of the data-array,
1041 * not the device-set.
1042 */
1043 /* imsm does not track uuid's so we synthesis one using sha1 on
1044 * - The signature (Which is constant for all imsm array, but no matter)
1045 * - the family_num of the container
1046 * - the index number of the volume
1047 * - the 'serial' number of the volume.
1048 * Hopefully these are all constant.
1049 */
1050 struct intel_super *super = st->sb;
1051
1052 char buf[20];
1053 struct sha1_ctx ctx;
1054 struct imsm_dev *dev = NULL;
1055
1056 sha1_init_ctx(&ctx);
1057 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1058 sha1_process_bytes(&super->anchor->family_num, sizeof(__u32), &ctx);
1059 if (super->current_vol >= 0)
1060 dev = get_imsm_dev(super, super->current_vol);
1061 if (dev) {
1062 __u32 vol = super->current_vol;
1063 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1064 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1065 }
1066 sha1_finish_ctx(&ctx, buf);
1067 memcpy(uuid, buf, 4*4);
1068 }
1069
1070 #if 0
1071 static void
1072 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1073 {
1074 __u8 *v = get_imsm_version(mpb);
1075 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1076 char major[] = { 0, 0, 0 };
1077 char minor[] = { 0 ,0, 0 };
1078 char patch[] = { 0, 0, 0 };
1079 char *ver_parse[] = { major, minor, patch };
1080 int i, j;
1081
1082 i = j = 0;
1083 while (*v != '\0' && v < end) {
1084 if (*v != '.' && j < 2)
1085 ver_parse[i][j++] = *v;
1086 else {
1087 i++;
1088 j = 0;
1089 }
1090 v++;
1091 }
1092
1093 *m = strtol(minor, NULL, 0);
1094 *p = strtol(patch, NULL, 0);
1095 }
1096 #endif
1097
1098 static int imsm_level_to_layout(int level)
1099 {
1100 switch (level) {
1101 case 0:
1102 case 1:
1103 return 0;
1104 case 5:
1105 case 6:
1106 return ALGORITHM_LEFT_ASYMMETRIC;
1107 case 10:
1108 return 0x102;
1109 }
1110 return UnSet;
1111 }
1112
1113 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
1114 {
1115 struct intel_super *super = st->sb;
1116 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1117 struct imsm_map *map = get_imsm_map(dev, 0);
1118
1119 info->container_member = super->current_vol;
1120 info->array.raid_disks = map->num_members;
1121 info->array.level = get_imsm_raid_level(map);
1122 info->array.layout = imsm_level_to_layout(info->array.level);
1123 info->array.md_minor = -1;
1124 info->array.ctime = 0;
1125 info->array.utime = 0;
1126 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1127 info->array.state = !dev->vol.dirty;
1128
1129 info->disk.major = 0;
1130 info->disk.minor = 0;
1131
1132 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1133 info->component_size = __le32_to_cpu(map->blocks_per_member);
1134 memset(info->uuid, 0, sizeof(info->uuid));
1135
1136 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
1137 info->resync_start = 0;
1138 else if (dev->vol.migr_state)
1139 info->resync_start = __le32_to_cpu(dev->vol.curr_migr_unit);
1140 else
1141 info->resync_start = ~0ULL;
1142
1143 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1144 info->name[MAX_RAID_SERIAL_LEN] = 0;
1145
1146 info->array.major_version = -1;
1147 info->array.minor_version = -2;
1148 sprintf(info->text_version, "/%s/%d",
1149 devnum2devname(st->container_dev),
1150 info->container_member);
1151 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1152 uuid_from_super_imsm(st, info->uuid);
1153 }
1154
1155
1156 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
1157 {
1158 struct intel_super *super = st->sb;
1159 struct imsm_disk *disk;
1160 __u32 s;
1161
1162 if (super->current_vol >= 0) {
1163 getinfo_super_imsm_volume(st, info);
1164 return;
1165 }
1166
1167 /* Set raid_disks to zero so that Assemble will always pull in valid
1168 * spares
1169 */
1170 info->array.raid_disks = 0;
1171 info->array.level = LEVEL_CONTAINER;
1172 info->array.layout = 0;
1173 info->array.md_minor = -1;
1174 info->array.ctime = 0; /* N/A for imsm */
1175 info->array.utime = 0;
1176 info->array.chunk_size = 0;
1177
1178 info->disk.major = 0;
1179 info->disk.minor = 0;
1180 info->disk.raid_disk = -1;
1181 info->reshape_active = 0;
1182 info->array.major_version = -1;
1183 info->array.minor_version = -2;
1184 strcpy(info->text_version, "imsm");
1185 info->safe_mode_delay = 0;
1186 info->disk.number = -1;
1187 info->disk.state = 0;
1188 info->name[0] = 0;
1189
1190 if (super->disks) {
1191 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1192
1193 disk = &super->disks->disk;
1194 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1195 info->component_size = reserved;
1196 s = disk->status;
1197 info->disk.state = s & CONFIGURED_DISK ? (1 << MD_DISK_ACTIVE) : 0;
1198 info->disk.state |= s & FAILED_DISK ? (1 << MD_DISK_FAULTY) : 0;
1199 info->disk.state |= s & SPARE_DISK ? 0 : (1 << MD_DISK_SYNC);
1200 }
1201
1202 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1203 * ->compare_super may have updated the 'num_raid_devs' field for spares
1204 */
1205 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1206 uuid_from_super_imsm(st, info->uuid);
1207 else
1208 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1209 }
1210
1211 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1212 char *update, char *devname, int verbose,
1213 int uuid_set, char *homehost)
1214 {
1215 /* FIXME */
1216
1217 /* For 'assemble' and 'force' we need to return non-zero if any
1218 * change was made. For others, the return value is ignored.
1219 * Update options are:
1220 * force-one : This device looks a bit old but needs to be included,
1221 * update age info appropriately.
1222 * assemble: clear any 'faulty' flag to allow this device to
1223 * be assembled.
1224 * force-array: Array is degraded but being forced, mark it clean
1225 * if that will be needed to assemble it.
1226 *
1227 * newdev: not used ????
1228 * grow: Array has gained a new device - this is currently for
1229 * linear only
1230 * resync: mark as dirty so a resync will happen.
1231 * name: update the name - preserving the homehost
1232 *
1233 * Following are not relevant for this imsm:
1234 * sparc2.2 : update from old dodgey metadata
1235 * super-minor: change the preferred_minor number
1236 * summaries: update redundant counters.
1237 * uuid: Change the uuid of the array to match watch is given
1238 * homehost: update the recorded homehost
1239 * _reshape_progress: record new reshape_progress position.
1240 */
1241 int rv = 0;
1242 //struct intel_super *super = st->sb;
1243 //struct imsm_super *mpb = super->mpb;
1244
1245 if (strcmp(update, "grow") == 0) {
1246 }
1247 if (strcmp(update, "resync") == 0) {
1248 /* dev->vol.dirty = 1; */
1249 }
1250
1251 /* IMSM has no concept of UUID or homehost */
1252
1253 return rv;
1254 }
1255
1256 static size_t disks_to_mpb_size(int disks)
1257 {
1258 size_t size;
1259
1260 size = sizeof(struct imsm_super);
1261 size += (disks - 1) * sizeof(struct imsm_disk);
1262 size += 2 * sizeof(struct imsm_dev);
1263 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1264 size += (4 - 2) * sizeof(struct imsm_map);
1265 /* 4 possible disk_ord_tbl's */
1266 size += 4 * (disks - 1) * sizeof(__u32);
1267
1268 return size;
1269 }
1270
1271 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1272 {
1273 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1274 return 0;
1275
1276 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1277 }
1278
1279 static void free_devlist(struct intel_super *super)
1280 {
1281 struct intel_dev *dv;
1282
1283 while (super->devlist) {
1284 dv = super->devlist->next;
1285 free(super->devlist->dev);
1286 free(super->devlist);
1287 super->devlist = dv;
1288 }
1289 }
1290
1291 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1292 {
1293 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1294 }
1295
1296 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1297 {
1298 /*
1299 * return:
1300 * 0 same, or first was empty, and second was copied
1301 * 1 second had wrong number
1302 * 2 wrong uuid
1303 * 3 wrong other info
1304 */
1305 struct intel_super *first = st->sb;
1306 struct intel_super *sec = tst->sb;
1307
1308 if (!first) {
1309 st->sb = tst->sb;
1310 tst->sb = NULL;
1311 return 0;
1312 }
1313
1314 if (memcmp(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH) != 0)
1315 return 3;
1316
1317 /* if an anchor does not have num_raid_devs set then it is a free
1318 * floating spare
1319 */
1320 if (first->anchor->num_raid_devs > 0 &&
1321 sec->anchor->num_raid_devs > 0) {
1322 if (first->anchor->family_num != sec->anchor->family_num)
1323 return 3;
1324 }
1325
1326 /* if 'first' is a spare promote it to a populated mpb with sec's
1327 * family number
1328 */
1329 if (first->anchor->num_raid_devs == 0 &&
1330 sec->anchor->num_raid_devs > 0) {
1331 int i;
1332 struct intel_dev *dv;
1333 struct imsm_dev *dev;
1334
1335 /* we need to copy raid device info from sec if an allocation
1336 * fails here we don't associate the spare
1337 */
1338 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1339 dv = malloc(sizeof(*dv));
1340 if (!dv)
1341 break;
1342 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1343 if (!dev) {
1344 free(dv);
1345 break;
1346 }
1347 dv->dev = dev;
1348 dv->index = i;
1349 dv->next = first->devlist;
1350 first->devlist = dv;
1351 }
1352 if (i <= sec->anchor->num_raid_devs) {
1353 /* allocation failure */
1354 free_devlist(first);
1355 fprintf(stderr, "imsm: failed to associate spare\n");
1356 return 3;
1357 }
1358 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1359 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1360
1361 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1362 first->anchor->family_num = sec->anchor->family_num;
1363 }
1364
1365 return 0;
1366 }
1367
1368 static void fd2devname(int fd, char *name)
1369 {
1370 struct stat st;
1371 char path[256];
1372 char dname[100];
1373 char *nm;
1374 int rv;
1375
1376 name[0] = '\0';
1377 if (fstat(fd, &st) != 0)
1378 return;
1379 sprintf(path, "/sys/dev/block/%d:%d",
1380 major(st.st_rdev), minor(st.st_rdev));
1381
1382 rv = readlink(path, dname, sizeof(dname));
1383 if (rv <= 0)
1384 return;
1385
1386 dname[rv] = '\0';
1387 nm = strrchr(dname, '/');
1388 nm++;
1389 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1390 }
1391
1392
1393 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1394
1395 static int imsm_read_serial(int fd, char *devname,
1396 __u8 serial[MAX_RAID_SERIAL_LEN])
1397 {
1398 unsigned char scsi_serial[255];
1399 int rv;
1400 int rsp_len;
1401 int len;
1402 char *c, *rsp_buf;
1403
1404 memset(scsi_serial, 0, sizeof(scsi_serial));
1405
1406 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1407
1408 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1409 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1410 fd2devname(fd, (char *) serial);
1411 return 0;
1412 }
1413
1414 if (rv != 0) {
1415 if (devname)
1416 fprintf(stderr,
1417 Name ": Failed to retrieve serial for %s\n",
1418 devname);
1419 return rv;
1420 }
1421
1422 /* trim leading whitespace */
1423 rsp_len = scsi_serial[3];
1424 if (!rsp_len) {
1425 if (devname)
1426 fprintf(stderr,
1427 Name ": Failed to retrieve serial for %s\n",
1428 devname);
1429 return 2;
1430 }
1431 rsp_buf = (char *) &scsi_serial[4];
1432 c = rsp_buf;
1433 while (isspace(*c))
1434 c++;
1435
1436 /* truncate len to the end of rsp_buf if necessary */
1437 if (c + MAX_RAID_SERIAL_LEN > rsp_buf + rsp_len)
1438 len = rsp_len - (c - rsp_buf);
1439 else
1440 len = MAX_RAID_SERIAL_LEN;
1441
1442 /* initialize the buffer and copy rsp_buf characters */
1443 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1444 memcpy(serial, c, len);
1445
1446 /* trim trailing whitespace starting with the last character copied */
1447 c = (char *) &serial[len - 1];
1448 while (isspace(*c) || *c == '\0')
1449 *c-- = '\0';
1450
1451 return 0;
1452 }
1453
1454 static int serialcmp(__u8 *s1, __u8 *s2)
1455 {
1456 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1457 }
1458
1459 static void serialcpy(__u8 *dest, __u8 *src)
1460 {
1461 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1462 }
1463
1464 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1465 {
1466 struct dl *dl;
1467
1468 for (dl = super->disks; dl; dl = dl->next)
1469 if (serialcmp(dl->serial, serial) == 0)
1470 break;
1471
1472 return dl;
1473 }
1474
1475 static int
1476 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1477 {
1478 struct dl *dl;
1479 struct stat stb;
1480 int rv;
1481 int i;
1482 int alloc = 1;
1483 __u8 serial[MAX_RAID_SERIAL_LEN];
1484
1485 rv = imsm_read_serial(fd, devname, serial);
1486
1487 if (rv != 0)
1488 return 2;
1489
1490 /* check if this is a disk we have seen before. it may be a spare in
1491 * super->disks while the current anchor believes it is a raid member,
1492 * check if we need to update dl->index
1493 */
1494 dl = serial_to_dl(serial, super);
1495 if (!dl)
1496 dl = malloc(sizeof(*dl));
1497 else
1498 alloc = 0;
1499
1500 if (!dl) {
1501 if (devname)
1502 fprintf(stderr,
1503 Name ": failed to allocate disk buffer for %s\n",
1504 devname);
1505 return 2;
1506 }
1507
1508 if (alloc) {
1509 fstat(fd, &stb);
1510 dl->major = major(stb.st_rdev);
1511 dl->minor = minor(stb.st_rdev);
1512 dl->next = super->disks;
1513 dl->fd = keep_fd ? fd : -1;
1514 dl->devname = devname ? strdup(devname) : NULL;
1515 serialcpy(dl->serial, serial);
1516 dl->index = -2;
1517 dl->e = NULL;
1518 } else if (keep_fd) {
1519 close(dl->fd);
1520 dl->fd = fd;
1521 }
1522
1523 /* look up this disk's index in the current anchor */
1524 for (i = 0; i < super->anchor->num_disks; i++) {
1525 struct imsm_disk *disk_iter;
1526
1527 disk_iter = __get_imsm_disk(super->anchor, i);
1528
1529 if (serialcmp(disk_iter->serial, dl->serial) == 0) {
1530 dl->disk = *disk_iter;
1531 /* only set index on disks that are a member of a
1532 * populated contianer, i.e. one with raid_devs
1533 */
1534 if (dl->disk.status & FAILED_DISK)
1535 dl->index = -2;
1536 else if (dl->disk.status & SPARE_DISK)
1537 dl->index = -1;
1538 else
1539 dl->index = i;
1540
1541 break;
1542 }
1543 }
1544
1545 /* no match, maybe a stale failed drive */
1546 if (i == super->anchor->num_disks && dl->index >= 0) {
1547 dl->disk = *__get_imsm_disk(super->anchor, dl->index);
1548 if (dl->disk.status & FAILED_DISK)
1549 dl->index = -2;
1550 }
1551
1552 if (alloc)
1553 super->disks = dl;
1554
1555 return 0;
1556 }
1557
1558 #ifndef MDASSEMBLE
1559 /* When migrating map0 contains the 'destination' state while map1
1560 * contains the current state. When not migrating map0 contains the
1561 * current state. This routine assumes that map[0].map_state is set to
1562 * the current array state before being called.
1563 *
1564 * Migration is indicated by one of the following states
1565 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1566 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1567 * map1state=unitialized)
1568 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1569 * map1state=normal)
1570 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1571 * map1state=degraded)
1572 */
1573 static void migrate(struct imsm_dev *dev, __u8 to_state, int rebuild_resync)
1574 {
1575 struct imsm_map *dest;
1576 struct imsm_map *src = get_imsm_map(dev, 0);
1577
1578 dev->vol.migr_state = 1;
1579 dev->vol.migr_type = rebuild_resync;
1580 dev->vol.curr_migr_unit = 0;
1581 dest = get_imsm_map(dev, 1);
1582
1583 memcpy(dest, src, sizeof_imsm_map(src));
1584 src->map_state = to_state;
1585 }
1586
1587 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1588 {
1589 struct imsm_map *map = get_imsm_map(dev, 0);
1590
1591 dev->vol.migr_state = 0;
1592 dev->vol.curr_migr_unit = 0;
1593 map->map_state = map_state;
1594 }
1595 #endif
1596
1597 static int parse_raid_devices(struct intel_super *super)
1598 {
1599 int i;
1600 struct imsm_dev *dev_new;
1601 size_t len, len_migr;
1602 size_t space_needed = 0;
1603 struct imsm_super *mpb = super->anchor;
1604
1605 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1606 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1607 struct intel_dev *dv;
1608
1609 len = sizeof_imsm_dev(dev_iter, 0);
1610 len_migr = sizeof_imsm_dev(dev_iter, 1);
1611 if (len_migr > len)
1612 space_needed += len_migr - len;
1613
1614 dv = malloc(sizeof(*dv));
1615 if (!dv)
1616 return 1;
1617 dev_new = malloc(len_migr);
1618 if (!dev_new) {
1619 free(dv);
1620 return 1;
1621 }
1622 imsm_copy_dev(dev_new, dev_iter);
1623 dv->dev = dev_new;
1624 dv->index = i;
1625 dv->next = super->devlist;
1626 super->devlist = dv;
1627 }
1628
1629 /* ensure that super->buf is large enough when all raid devices
1630 * are migrating
1631 */
1632 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1633 void *buf;
1634
1635 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1636 if (posix_memalign(&buf, 512, len) != 0)
1637 return 1;
1638
1639 memcpy(buf, super->buf, len);
1640 free(super->buf);
1641 super->buf = buf;
1642 super->len = len;
1643 }
1644
1645 return 0;
1646 }
1647
1648 /* retrieve a pointer to the bbm log which starts after all raid devices */
1649 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1650 {
1651 void *ptr = NULL;
1652
1653 if (__le32_to_cpu(mpb->bbm_log_size)) {
1654 ptr = mpb;
1655 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1656 }
1657
1658 return ptr;
1659 }
1660
1661 static void __free_imsm(struct intel_super *super, int free_disks);
1662
1663 /* load_imsm_mpb - read matrix metadata
1664 * allocates super->mpb to be freed by free_super
1665 */
1666 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1667 {
1668 unsigned long long dsize;
1669 unsigned long long sectors;
1670 struct stat;
1671 struct imsm_super *anchor;
1672 __u32 check_sum;
1673 int rc;
1674
1675 get_dev_size(fd, NULL, &dsize);
1676
1677 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1678 if (devname)
1679 fprintf(stderr,
1680 Name ": Cannot seek to anchor block on %s: %s\n",
1681 devname, strerror(errno));
1682 return 1;
1683 }
1684
1685 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1686 if (devname)
1687 fprintf(stderr,
1688 Name ": Failed to allocate imsm anchor buffer"
1689 " on %s\n", devname);
1690 return 1;
1691 }
1692 if (read(fd, anchor, 512) != 512) {
1693 if (devname)
1694 fprintf(stderr,
1695 Name ": Cannot read anchor block on %s: %s\n",
1696 devname, strerror(errno));
1697 free(anchor);
1698 return 1;
1699 }
1700
1701 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1702 if (devname)
1703 fprintf(stderr,
1704 Name ": no IMSM anchor on %s\n", devname);
1705 free(anchor);
1706 return 2;
1707 }
1708
1709 __free_imsm(super, 0);
1710 super->len = ROUND_UP(anchor->mpb_size, 512);
1711 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1712 if (devname)
1713 fprintf(stderr,
1714 Name ": unable to allocate %zu byte mpb buffer\n",
1715 super->len);
1716 free(anchor);
1717 return 2;
1718 }
1719 memcpy(super->buf, anchor, 512);
1720
1721 sectors = mpb_sectors(anchor) - 1;
1722 free(anchor);
1723 if (!sectors) {
1724 rc = load_imsm_disk(fd, super, devname, 0);
1725 if (rc == 0)
1726 rc = parse_raid_devices(super);
1727 return rc;
1728 }
1729
1730 /* read the extended mpb */
1731 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1732 if (devname)
1733 fprintf(stderr,
1734 Name ": Cannot seek to extended mpb on %s: %s\n",
1735 devname, strerror(errno));
1736 return 1;
1737 }
1738
1739 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1740 if (devname)
1741 fprintf(stderr,
1742 Name ": Cannot read extended mpb on %s: %s\n",
1743 devname, strerror(errno));
1744 return 2;
1745 }
1746
1747 check_sum = __gen_imsm_checksum(super->anchor);
1748 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1749 if (devname)
1750 fprintf(stderr,
1751 Name ": IMSM checksum %x != %x on %s\n",
1752 check_sum, __le32_to_cpu(super->anchor->check_sum),
1753 devname);
1754 return 2;
1755 }
1756
1757 /* FIXME the BBM log is disk specific so we cannot use this global
1758 * buffer for all disks. Ok for now since we only look at the global
1759 * bbm_log_size parameter to gate assembly
1760 */
1761 super->bbm_log = __get_imsm_bbm_log(super->anchor);
1762
1763 rc = load_imsm_disk(fd, super, devname, 0);
1764 if (rc == 0)
1765 rc = parse_raid_devices(super);
1766
1767 return rc;
1768 }
1769
1770 static void __free_imsm_disk(struct dl *d)
1771 {
1772 if (d->fd >= 0)
1773 close(d->fd);
1774 if (d->devname)
1775 free(d->devname);
1776 if (d->e)
1777 free(d->e);
1778 free(d);
1779
1780 }
1781 static void free_imsm_disks(struct intel_super *super)
1782 {
1783 struct dl *d;
1784
1785 while (super->disks) {
1786 d = super->disks;
1787 super->disks = d->next;
1788 __free_imsm_disk(d);
1789 }
1790 while (super->missing) {
1791 d = super->missing;
1792 super->missing = d->next;
1793 __free_imsm_disk(d);
1794 }
1795
1796 }
1797
1798 /* free all the pieces hanging off of a super pointer */
1799 static void __free_imsm(struct intel_super *super, int free_disks)
1800 {
1801 if (super->buf) {
1802 free(super->buf);
1803 super->buf = NULL;
1804 }
1805 if (free_disks)
1806 free_imsm_disks(super);
1807 free_devlist(super);
1808 if (super->hba) {
1809 free((void *) super->hba);
1810 super->hba = NULL;
1811 }
1812 }
1813
1814 static void free_imsm(struct intel_super *super)
1815 {
1816 __free_imsm(super, 1);
1817 free(super);
1818 }
1819
1820 static void free_super_imsm(struct supertype *st)
1821 {
1822 struct intel_super *super = st->sb;
1823
1824 if (!super)
1825 return;
1826
1827 free_imsm(super);
1828 st->sb = NULL;
1829 }
1830
1831 static struct intel_super *alloc_super(int creating_imsm)
1832 {
1833 struct intel_super *super = malloc(sizeof(*super));
1834
1835 if (super) {
1836 memset(super, 0, sizeof(*super));
1837 super->creating_imsm = creating_imsm;
1838 super->current_vol = -1;
1839 super->create_offset = ~((__u32 ) 0);
1840 if (!check_env("IMSM_NO_PLATFORM"))
1841 super->orom = find_imsm_orom();
1842 if (super->orom) {
1843 struct sys_dev *list, *ent;
1844
1845 /* find the first intel ahci controller */
1846 list = find_driver_devices("pci", "ahci");
1847 for (ent = list; ent; ent = ent->next)
1848 if (devpath_to_vendor(ent->path) == 0x8086)
1849 break;
1850 if (ent) {
1851 super->hba = ent->path;
1852 ent->path = NULL;
1853 }
1854 free_sys_dev(&list);
1855 }
1856 }
1857
1858 return super;
1859 }
1860
1861 #ifndef MDASSEMBLE
1862 /* find_missing - helper routine for load_super_imsm_all that identifies
1863 * disks that have disappeared from the system. This routine relies on
1864 * the mpb being uptodate, which it is at load time.
1865 */
1866 static int find_missing(struct intel_super *super)
1867 {
1868 int i;
1869 struct imsm_super *mpb = super->anchor;
1870 struct dl *dl;
1871 struct imsm_disk *disk;
1872
1873 for (i = 0; i < mpb->num_disks; i++) {
1874 disk = __get_imsm_disk(mpb, i);
1875 dl = serial_to_dl(disk->serial, super);
1876 if (dl)
1877 continue;
1878 /* ok we have a 'disk' without a live entry in
1879 * super->disks
1880 */
1881 if (disk->status & FAILED_DISK || !(disk->status & USABLE_DISK))
1882 continue; /* never mind, already marked */
1883
1884 dl = malloc(sizeof(*dl));
1885 if (!dl)
1886 return 1;
1887 dl->major = 0;
1888 dl->minor = 0;
1889 dl->fd = -1;
1890 dl->devname = strdup("missing");
1891 dl->index = i;
1892 serialcpy(dl->serial, disk->serial);
1893 dl->disk = *disk;
1894 dl->next = super->missing;
1895 super->missing = dl;
1896 }
1897
1898 return 0;
1899 }
1900
1901 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
1902 char *devname, int keep_fd)
1903 {
1904 struct mdinfo *sra;
1905 struct intel_super *super;
1906 struct mdinfo *sd, *best = NULL;
1907 __u32 bestgen = 0;
1908 __u32 gen;
1909 char nm[20];
1910 int dfd;
1911 int rv;
1912
1913 /* check if this disk is a member of an active array */
1914 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
1915 if (!sra)
1916 return 1;
1917
1918 if (sra->array.major_version != -1 ||
1919 sra->array.minor_version != -2 ||
1920 strcmp(sra->text_version, "imsm") != 0)
1921 return 1;
1922
1923 super = alloc_super(0);
1924 if (!super)
1925 return 1;
1926
1927 /* find the most up to date disk in this array, skipping spares */
1928 for (sd = sra->devs; sd; sd = sd->next) {
1929 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
1930 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
1931 if (dfd < 0) {
1932 free_imsm(super);
1933 return 2;
1934 }
1935 rv = load_imsm_mpb(dfd, super, NULL);
1936 if (!keep_fd)
1937 close(dfd);
1938 if (rv == 0) {
1939 if (super->anchor->num_raid_devs == 0)
1940 gen = 0;
1941 else
1942 gen = __le32_to_cpu(super->anchor->generation_num);
1943 if (!best || gen > bestgen) {
1944 bestgen = gen;
1945 best = sd;
1946 }
1947 } else {
1948 free_imsm(super);
1949 return 2;
1950 }
1951 }
1952
1953 if (!best) {
1954 free_imsm(super);
1955 return 1;
1956 }
1957
1958 /* load the most up to date anchor */
1959 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
1960 dfd = dev_open(nm, O_RDONLY);
1961 if (dfd < 0) {
1962 free_imsm(super);
1963 return 1;
1964 }
1965 rv = load_imsm_mpb(dfd, super, NULL);
1966 close(dfd);
1967 if (rv != 0) {
1968 free_imsm(super);
1969 return 2;
1970 }
1971
1972 /* re-parse the disk list with the current anchor */
1973 for (sd = sra->devs ; sd ; sd = sd->next) {
1974 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
1975 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
1976 if (dfd < 0) {
1977 free_imsm(super);
1978 return 2;
1979 }
1980 load_imsm_disk(dfd, super, NULL, keep_fd);
1981 if (!keep_fd)
1982 close(dfd);
1983 }
1984
1985
1986 if (find_missing(super) != 0) {
1987 free_imsm(super);
1988 return 2;
1989 }
1990
1991 if (st->subarray[0]) {
1992 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
1993 super->current_vol = atoi(st->subarray);
1994 else
1995 return 1;
1996 }
1997
1998 *sbp = super;
1999 st->container_dev = fd2devnum(fd);
2000 if (st->ss == NULL) {
2001 st->ss = &super_imsm;
2002 st->minor_version = 0;
2003 st->max_devs = IMSM_MAX_DEVICES;
2004 }
2005 st->loaded_container = 1;
2006
2007 return 0;
2008 }
2009 #endif
2010
2011 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2012 {
2013 struct intel_super *super;
2014 int rv;
2015
2016 #ifndef MDASSEMBLE
2017 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2018 return 0;
2019 #endif
2020 if (st->subarray[0])
2021 return 1; /* FIXME */
2022
2023 super = alloc_super(0);
2024 if (!super) {
2025 fprintf(stderr,
2026 Name ": malloc of %zu failed.\n",
2027 sizeof(*super));
2028 return 1;
2029 }
2030
2031 rv = load_imsm_mpb(fd, super, devname);
2032
2033 if (rv) {
2034 if (devname)
2035 fprintf(stderr,
2036 Name ": Failed to load all information "
2037 "sections on %s\n", devname);
2038 free_imsm(super);
2039 return rv;
2040 }
2041
2042 st->sb = super;
2043 if (st->ss == NULL) {
2044 st->ss = &super_imsm;
2045 st->minor_version = 0;
2046 st->max_devs = IMSM_MAX_DEVICES;
2047 }
2048 st->loaded_container = 0;
2049
2050 return 0;
2051 }
2052
2053 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2054 {
2055 if (info->level == 1)
2056 return 128;
2057 return info->chunk_size >> 9;
2058 }
2059
2060 static __u32 info_to_num_data_stripes(mdu_array_info_t *info)
2061 {
2062 __u32 num_stripes;
2063
2064 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2065 if (info->level == 1)
2066 num_stripes /= 2;
2067
2068 return num_stripes;
2069 }
2070
2071 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2072 {
2073 if (info->level == 1)
2074 return info->size * 2;
2075 else
2076 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2077 }
2078
2079 static void imsm_update_version_info(struct intel_super *super)
2080 {
2081 /* update the version and attributes */
2082 struct imsm_super *mpb = super->anchor;
2083 char *version;
2084 struct imsm_dev *dev;
2085 struct imsm_map *map;
2086 int i;
2087
2088 for (i = 0; i < mpb->num_raid_devs; i++) {
2089 dev = get_imsm_dev(super, i);
2090 map = get_imsm_map(dev, 0);
2091 if (__le32_to_cpu(dev->size_high) > 0)
2092 mpb->attributes |= MPB_ATTRIB_2TB;
2093
2094 /* FIXME detect when an array spans a port multiplier */
2095 #if 0
2096 mpb->attributes |= MPB_ATTRIB_PM;
2097 #endif
2098
2099 if (mpb->num_raid_devs > 1 ||
2100 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2101 version = MPB_VERSION_ATTRIBS;
2102 switch (get_imsm_raid_level(map)) {
2103 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2104 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2105 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2106 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2107 }
2108 } else {
2109 if (map->num_members >= 5)
2110 version = MPB_VERSION_5OR6_DISK_ARRAY;
2111 else if (dev->status == DEV_CLONE_N_GO)
2112 version = MPB_VERSION_CNG;
2113 else if (get_imsm_raid_level(map) == 5)
2114 version = MPB_VERSION_RAID5;
2115 else if (map->num_members >= 3)
2116 version = MPB_VERSION_3OR4_DISK_ARRAY;
2117 else if (get_imsm_raid_level(map) == 1)
2118 version = MPB_VERSION_RAID1;
2119 else
2120 version = MPB_VERSION_RAID0;
2121 }
2122 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2123 }
2124 }
2125
2126 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2127 unsigned long long size, char *name,
2128 char *homehost, int *uuid)
2129 {
2130 /* We are creating a volume inside a pre-existing container.
2131 * so st->sb is already set.
2132 */
2133 struct intel_super *super = st->sb;
2134 struct imsm_super *mpb = super->anchor;
2135 struct intel_dev *dv;
2136 struct imsm_dev *dev;
2137 struct imsm_vol *vol;
2138 struct imsm_map *map;
2139 int idx = mpb->num_raid_devs;
2140 int i;
2141 unsigned long long array_blocks;
2142 size_t size_old, size_new;
2143
2144 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2145 fprintf(stderr, Name": This imsm-container already has the "
2146 "maximum of %d volumes\n", super->orom->vpa);
2147 return 0;
2148 }
2149
2150 /* ensure the mpb is large enough for the new data */
2151 size_old = __le32_to_cpu(mpb->mpb_size);
2152 size_new = disks_to_mpb_size(info->nr_disks);
2153 if (size_new > size_old) {
2154 void *mpb_new;
2155 size_t size_round = ROUND_UP(size_new, 512);
2156
2157 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2158 fprintf(stderr, Name": could not allocate new mpb\n");
2159 return 0;
2160 }
2161 memcpy(mpb_new, mpb, size_old);
2162 free(mpb);
2163 mpb = mpb_new;
2164 super->anchor = mpb_new;
2165 mpb->mpb_size = __cpu_to_le32(size_new);
2166 memset(mpb_new + size_old, 0, size_round - size_old);
2167 }
2168 super->current_vol = idx;
2169 /* when creating the first raid device in this container set num_disks
2170 * to zero, i.e. delete this spare and add raid member devices in
2171 * add_to_super_imsm_volume()
2172 */
2173 if (super->current_vol == 0)
2174 mpb->num_disks = 0;
2175 sprintf(st->subarray, "%d", idx);
2176 dv = malloc(sizeof(*dv));
2177 if (!dv) {
2178 fprintf(stderr, Name ": failed to allocate device list entry\n");
2179 return 0;
2180 }
2181 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2182 if (!dev) {
2183 free(dv);
2184 fprintf(stderr, Name": could not allocate raid device\n");
2185 return 0;
2186 }
2187 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2188 if (info->level == 1)
2189 array_blocks = info_to_blocks_per_member(info);
2190 else
2191 array_blocks = calc_array_size(info->level, info->raid_disks,
2192 info->layout, info->chunk_size,
2193 info->size*2);
2194 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2195 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2196 dev->status = __cpu_to_le32(0);
2197 dev->reserved_blocks = __cpu_to_le32(0);
2198 vol = &dev->vol;
2199 vol->migr_state = 0;
2200 vol->migr_type = MIGR_INIT;
2201 vol->dirty = 0;
2202 vol->curr_migr_unit = 0;
2203 map = get_imsm_map(dev, 0);
2204 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2205 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2206 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2207 map->num_data_stripes = __cpu_to_le32(info_to_num_data_stripes(info));
2208 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2209 IMSM_T_STATE_NORMAL;
2210
2211 if (info->level == 1 && info->raid_disks > 2) {
2212 fprintf(stderr, Name": imsm does not support more than 2 disks"
2213 "in a raid1 volume\n");
2214 return 0;
2215 }
2216 if (info->level == 10) {
2217 map->raid_level = 1;
2218 map->num_domains = info->raid_disks / 2;
2219 } else {
2220 map->raid_level = info->level;
2221 map->num_domains = !!map->raid_level;
2222 }
2223
2224 map->num_members = info->raid_disks;
2225 for (i = 0; i < map->num_members; i++) {
2226 /* initialized in add_to_super */
2227 set_imsm_ord_tbl_ent(map, i, 0);
2228 }
2229 mpb->num_raid_devs++;
2230
2231 dv->dev = dev;
2232 dv->index = super->current_vol;
2233 dv->next = super->devlist;
2234 super->devlist = dv;
2235
2236 imsm_update_version_info(super);
2237
2238 return 1;
2239 }
2240
2241 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2242 unsigned long long size, char *name,
2243 char *homehost, int *uuid)
2244 {
2245 /* This is primarily called by Create when creating a new array.
2246 * We will then get add_to_super called for each component, and then
2247 * write_init_super called to write it out to each device.
2248 * For IMSM, Create can create on fresh devices or on a pre-existing
2249 * array.
2250 * To create on a pre-existing array a different method will be called.
2251 * This one is just for fresh drives.
2252 */
2253 struct intel_super *super;
2254 struct imsm_super *mpb;
2255 size_t mpb_size;
2256 char *version;
2257
2258 if (!info) {
2259 st->sb = NULL;
2260 return 0;
2261 }
2262 if (st->sb)
2263 return init_super_imsm_volume(st, info, size, name, homehost,
2264 uuid);
2265
2266 super = alloc_super(1);
2267 if (!super)
2268 return 0;
2269 mpb_size = disks_to_mpb_size(info->nr_disks);
2270 if (posix_memalign(&super->buf, 512, mpb_size) != 0) {
2271 free(super);
2272 return 0;
2273 }
2274 mpb = super->buf;
2275 memset(mpb, 0, mpb_size);
2276
2277 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2278
2279 version = (char *) mpb->sig;
2280 strcpy(version, MPB_SIGNATURE);
2281 version += strlen(MPB_SIGNATURE);
2282 strcpy(version, MPB_VERSION_RAID0);
2283 mpb->mpb_size = mpb_size;
2284
2285 st->sb = super;
2286 return 1;
2287 }
2288
2289 #ifndef MDASSEMBLE
2290 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2291 int fd, char *devname)
2292 {
2293 struct intel_super *super = st->sb;
2294 struct imsm_super *mpb = super->anchor;
2295 struct dl *dl;
2296 struct imsm_dev *dev;
2297 struct imsm_map *map;
2298
2299 dev = get_imsm_dev(super, super->current_vol);
2300 map = get_imsm_map(dev, 0);
2301
2302 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2303 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2304 devname);
2305 return 1;
2306 }
2307
2308 for (dl = super->disks; dl ; dl = dl->next)
2309 if (dl->major == dk->major &&
2310 dl->minor == dk->minor)
2311 break;
2312
2313 if (!dl) {
2314 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2315 return 1;
2316 }
2317
2318 /* add a pristine spare to the metadata */
2319 if (dl->index < 0) {
2320 dl->index = super->anchor->num_disks;
2321 super->anchor->num_disks++;
2322 }
2323 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2324 dl->disk.status = CONFIGURED_DISK | USABLE_DISK;
2325
2326 /* if we are creating the first raid device update the family number */
2327 if (super->current_vol == 0) {
2328 __u32 sum;
2329 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2330 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2331
2332 *_dev = *dev;
2333 *_disk = dl->disk;
2334 sum = __gen_imsm_checksum(mpb);
2335 mpb->family_num = __cpu_to_le32(sum);
2336 }
2337
2338 return 0;
2339 }
2340
2341 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2342 int fd, char *devname)
2343 {
2344 struct intel_super *super = st->sb;
2345 struct dl *dd;
2346 unsigned long long size;
2347 __u32 id;
2348 int rv;
2349 struct stat stb;
2350
2351 /* if we are on an RAID enabled platform check that the disk is
2352 * attached to the raid controller
2353 */
2354 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2355 fprintf(stderr,
2356 Name ": %s is not attached to the raid controller: %s\n",
2357 devname ? : "disk", super->hba);
2358 return 1;
2359 }
2360
2361 if (super->current_vol >= 0)
2362 return add_to_super_imsm_volume(st, dk, fd, devname);
2363
2364 fstat(fd, &stb);
2365 dd = malloc(sizeof(*dd));
2366 if (!dd) {
2367 fprintf(stderr,
2368 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2369 return 1;
2370 }
2371 memset(dd, 0, sizeof(*dd));
2372 dd->major = major(stb.st_rdev);
2373 dd->minor = minor(stb.st_rdev);
2374 dd->index = -1;
2375 dd->devname = devname ? strdup(devname) : NULL;
2376 dd->fd = fd;
2377 rv = imsm_read_serial(fd, devname, dd->serial);
2378 if (rv) {
2379 fprintf(stderr,
2380 Name ": failed to retrieve scsi serial, aborting\n");
2381 free(dd);
2382 abort();
2383 }
2384
2385 get_dev_size(fd, NULL, &size);
2386 size /= 512;
2387 serialcpy(dd->disk.serial, dd->serial);
2388 dd->disk.total_blocks = __cpu_to_le32(size);
2389 dd->disk.status = USABLE_DISK | SPARE_DISK;
2390 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
2391 dd->disk.scsi_id = __cpu_to_le32(id);
2392 else
2393 dd->disk.scsi_id = __cpu_to_le32(0);
2394
2395 if (st->update_tail) {
2396 dd->next = super->add;
2397 super->add = dd;
2398 } else {
2399 dd->next = super->disks;
2400 super->disks = dd;
2401 }
2402
2403 return 0;
2404 }
2405
2406 static int store_imsm_mpb(int fd, struct intel_super *super);
2407
2408 /* spare records have their own family number and do not have any defined raid
2409 * devices
2410 */
2411 static int write_super_imsm_spares(struct intel_super *super, int doclose)
2412 {
2413 struct imsm_super mpb_save;
2414 struct imsm_super *mpb = super->anchor;
2415 __u32 sum;
2416 struct dl *d;
2417
2418 mpb_save = *mpb;
2419 mpb->num_raid_devs = 0;
2420 mpb->num_disks = 1;
2421 mpb->mpb_size = sizeof(struct imsm_super);
2422 mpb->generation_num = __cpu_to_le32(1UL);
2423
2424 for (d = super->disks; d; d = d->next) {
2425 if (d->index != -1)
2426 continue;
2427
2428 mpb->disk[0] = d->disk;
2429 sum = __gen_imsm_checksum(mpb);
2430 mpb->family_num = __cpu_to_le32(sum);
2431 sum = __gen_imsm_checksum(mpb);
2432 mpb->check_sum = __cpu_to_le32(sum);
2433
2434 if (store_imsm_mpb(d->fd, super)) {
2435 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2436 __func__, d->major, d->minor, strerror(errno));
2437 *mpb = mpb_save;
2438 return 1;
2439 }
2440 if (doclose) {
2441 close(d->fd);
2442 d->fd = -1;
2443 }
2444 }
2445
2446 *mpb = mpb_save;
2447 return 0;
2448 }
2449
2450 static int write_super_imsm(struct intel_super *super, int doclose)
2451 {
2452 struct imsm_super *mpb = super->anchor;
2453 struct dl *d;
2454 __u32 generation;
2455 __u32 sum;
2456 int spares = 0;
2457 int i;
2458 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
2459
2460 /* 'generation' is incremented everytime the metadata is written */
2461 generation = __le32_to_cpu(mpb->generation_num);
2462 generation++;
2463 mpb->generation_num = __cpu_to_le32(generation);
2464
2465 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
2466 for (d = super->disks; d; d = d->next) {
2467 if (d->index == -1)
2468 spares++;
2469 else
2470 mpb->disk[d->index] = d->disk;
2471 }
2472 for (d = super->missing; d; d = d->next)
2473 mpb->disk[d->index] = d->disk;
2474
2475 for (i = 0; i < mpb->num_raid_devs; i++) {
2476 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
2477
2478 imsm_copy_dev(dev, get_imsm_dev(super, i));
2479 mpb_size += sizeof_imsm_dev(dev, 0);
2480 }
2481 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
2482 mpb->mpb_size = __cpu_to_le32(mpb_size);
2483
2484 /* recalculate checksum */
2485 sum = __gen_imsm_checksum(mpb);
2486 mpb->check_sum = __cpu_to_le32(sum);
2487
2488 /* write the mpb for disks that compose raid devices */
2489 for (d = super->disks; d ; d = d->next) {
2490 if (d->index < 0)
2491 continue;
2492 if (store_imsm_mpb(d->fd, super))
2493 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2494 __func__, d->major, d->minor, strerror(errno));
2495 if (doclose) {
2496 close(d->fd);
2497 d->fd = -1;
2498 }
2499 }
2500
2501 if (spares)
2502 return write_super_imsm_spares(super, doclose);
2503
2504 return 0;
2505 }
2506
2507
2508 static int create_array(struct supertype *st)
2509 {
2510 size_t len;
2511 struct imsm_update_create_array *u;
2512 struct intel_super *super = st->sb;
2513 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
2514 struct imsm_map *map = get_imsm_map(dev, 0);
2515 struct disk_info *inf;
2516 struct imsm_disk *disk;
2517 int i;
2518 int idx;
2519
2520 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
2521 sizeof(*inf) * map->num_members;
2522 u = malloc(len);
2523 if (!u) {
2524 fprintf(stderr, "%s: failed to allocate update buffer\n",
2525 __func__);
2526 return 1;
2527 }
2528
2529 u->type = update_create_array;
2530 u->dev_idx = super->current_vol;
2531 imsm_copy_dev(&u->dev, dev);
2532 inf = get_disk_info(u);
2533 for (i = 0; i < map->num_members; i++) {
2534 idx = get_imsm_disk_idx(dev, i);
2535 disk = get_imsm_disk(super, idx);
2536 serialcpy(inf[i].serial, disk->serial);
2537 }
2538 append_metadata_update(st, u, len);
2539
2540 return 0;
2541 }
2542
2543 static int _add_disk(struct supertype *st)
2544 {
2545 struct intel_super *super = st->sb;
2546 size_t len;
2547 struct imsm_update_add_disk *u;
2548
2549 if (!super->add)
2550 return 0;
2551
2552 len = sizeof(*u);
2553 u = malloc(len);
2554 if (!u) {
2555 fprintf(stderr, "%s: failed to allocate update buffer\n",
2556 __func__);
2557 return 1;
2558 }
2559
2560 u->type = update_add_disk;
2561 append_metadata_update(st, u, len);
2562
2563 return 0;
2564 }
2565
2566 static int write_init_super_imsm(struct supertype *st)
2567 {
2568 if (st->update_tail) {
2569 /* queue the recently created array / added disk
2570 * as a metadata update */
2571 struct intel_super *super = st->sb;
2572 struct dl *d;
2573 int rv;
2574
2575 /* determine if we are creating a volume or adding a disk */
2576 if (super->current_vol < 0) {
2577 /* in the add disk case we are running in mdmon
2578 * context, so don't close fd's
2579 */
2580 return _add_disk(st);
2581 } else
2582 rv = create_array(st);
2583
2584 for (d = super->disks; d ; d = d->next) {
2585 close(d->fd);
2586 d->fd = -1;
2587 }
2588
2589 return rv;
2590 } else
2591 return write_super_imsm(st->sb, 1);
2592 }
2593 #endif
2594
2595 static int store_zero_imsm(struct supertype *st, int fd)
2596 {
2597 unsigned long long dsize;
2598 void *buf;
2599
2600 get_dev_size(fd, NULL, &dsize);
2601
2602 /* first block is stored on second to last sector of the disk */
2603 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
2604 return 1;
2605
2606 if (posix_memalign(&buf, 512, 512) != 0)
2607 return 1;
2608
2609 memset(buf, 0, 512);
2610 if (write(fd, buf, 512) != 512)
2611 return 1;
2612 return 0;
2613 }
2614
2615 static int imsm_bbm_log_size(struct imsm_super *mpb)
2616 {
2617 return __le32_to_cpu(mpb->bbm_log_size);
2618 }
2619
2620 #ifndef MDASSEMBLE
2621 static int validate_geometry_imsm_container(struct supertype *st, int level,
2622 int layout, int raiddisks, int chunk,
2623 unsigned long long size, char *dev,
2624 unsigned long long *freesize,
2625 int verbose)
2626 {
2627 int fd;
2628 unsigned long long ldsize;
2629 const struct imsm_orom *orom;
2630
2631 if (level != LEVEL_CONTAINER)
2632 return 0;
2633 if (!dev)
2634 return 1;
2635
2636 if (check_env("IMSM_NO_PLATFORM"))
2637 orom = NULL;
2638 else
2639 orom = find_imsm_orom();
2640 if (orom && raiddisks > orom->tds) {
2641 if (verbose)
2642 fprintf(stderr, Name ": %d exceeds maximum number of"
2643 " platform supported disks: %d\n",
2644 raiddisks, orom->tds);
2645 return 0;
2646 }
2647
2648 fd = open(dev, O_RDONLY|O_EXCL, 0);
2649 if (fd < 0) {
2650 if (verbose)
2651 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
2652 dev, strerror(errno));
2653 return 0;
2654 }
2655 if (!get_dev_size(fd, dev, &ldsize)) {
2656 close(fd);
2657 return 0;
2658 }
2659 close(fd);
2660
2661 *freesize = avail_size_imsm(st, ldsize >> 9);
2662
2663 return 1;
2664 }
2665
2666 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
2667 {
2668 const unsigned long long base_start = e[*idx].start;
2669 unsigned long long end = base_start + e[*idx].size;
2670 int i;
2671
2672 if (base_start == end)
2673 return 0;
2674
2675 *idx = *idx + 1;
2676 for (i = *idx; i < num_extents; i++) {
2677 /* extend overlapping extents */
2678 if (e[i].start >= base_start &&
2679 e[i].start <= end) {
2680 if (e[i].size == 0)
2681 return 0;
2682 if (e[i].start + e[i].size > end)
2683 end = e[i].start + e[i].size;
2684 } else if (e[i].start > end) {
2685 *idx = i;
2686 break;
2687 }
2688 }
2689
2690 return end - base_start;
2691 }
2692
2693 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
2694 {
2695 /* build a composite disk with all known extents and generate a new
2696 * 'maxsize' given the "all disks in an array must share a common start
2697 * offset" constraint
2698 */
2699 struct extent *e = calloc(sum_extents, sizeof(*e));
2700 struct dl *dl;
2701 int i, j;
2702 int start_extent;
2703 unsigned long long pos;
2704 unsigned long long start;
2705 unsigned long long maxsize;
2706 unsigned long reserve;
2707
2708 if (!e)
2709 return ~0ULL; /* error */
2710
2711 /* coalesce and sort all extents. also, check to see if we need to
2712 * reserve space between member arrays
2713 */
2714 j = 0;
2715 for (dl = super->disks; dl; dl = dl->next) {
2716 if (!dl->e)
2717 continue;
2718 for (i = 0; i < dl->extent_cnt; i++)
2719 e[j++] = dl->e[i];
2720 }
2721 qsort(e, sum_extents, sizeof(*e), cmp_extent);
2722
2723 /* merge extents */
2724 i = 0;
2725 j = 0;
2726 while (i < sum_extents) {
2727 e[j].start = e[i].start;
2728 e[j].size = find_size(e, &i, sum_extents);
2729 j++;
2730 if (e[j-1].size == 0)
2731 break;
2732 }
2733
2734 pos = 0;
2735 maxsize = 0;
2736 start_extent = 0;
2737 i = 0;
2738 do {
2739 unsigned long long esize;
2740
2741 esize = e[i].start - pos;
2742 if (esize >= maxsize) {
2743 maxsize = esize;
2744 start = pos;
2745 start_extent = i;
2746 }
2747 pos = e[i].start + e[i].size;
2748 i++;
2749 } while (e[i-1].size);
2750 free(e);
2751
2752 if (start_extent > 0)
2753 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
2754 else
2755 reserve = 0;
2756
2757 if (maxsize < reserve)
2758 return ~0ULL;
2759
2760 super->create_offset = ~((__u32) 0);
2761 if (start + reserve > super->create_offset)
2762 return ~0ULL; /* start overflows create_offset */
2763 super->create_offset = start + reserve;
2764
2765 return maxsize - reserve;
2766 }
2767
2768 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
2769 {
2770 if (level < 0 || level == 6 || level == 4)
2771 return 0;
2772
2773 /* if we have an orom prevent invalid raid levels */
2774 if (orom)
2775 switch (level) {
2776 case 0: return imsm_orom_has_raid0(orom);
2777 case 1:
2778 if (raiddisks > 2)
2779 return imsm_orom_has_raid1e(orom);
2780 return imsm_orom_has_raid1(orom) && raiddisks == 2;
2781 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
2782 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
2783 }
2784 else
2785 return 1; /* not on an Intel RAID platform so anything goes */
2786
2787 return 0;
2788 }
2789
2790 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2791 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2792 * FIX ME add ahci details
2793 */
2794 static int validate_geometry_imsm_volume(struct supertype *st, int level,
2795 int layout, int raiddisks, int chunk,
2796 unsigned long long size, char *dev,
2797 unsigned long long *freesize,
2798 int verbose)
2799 {
2800 struct stat stb;
2801 struct intel_super *super = st->sb;
2802 struct imsm_super *mpb = super->anchor;
2803 struct dl *dl;
2804 unsigned long long pos = 0;
2805 unsigned long long maxsize;
2806 struct extent *e;
2807 int i;
2808
2809 /* We must have the container info already read in. */
2810 if (!super)
2811 return 0;
2812
2813 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
2814 pr_vrb(": platform does not support raid%d with %d disk%s\n",
2815 level, raiddisks, raiddisks > 1 ? "s" : "");
2816 return 0;
2817 }
2818 if (super->orom && !imsm_orom_has_chunk(super->orom, chunk)) {
2819 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
2820 return 0;
2821 }
2822 if (layout != imsm_level_to_layout(level)) {
2823 if (level == 5)
2824 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
2825 else if (level == 10)
2826 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
2827 else
2828 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
2829 layout, level);
2830 return 0;
2831 }
2832
2833 if (!dev) {
2834 /* General test: make sure there is space for
2835 * 'raiddisks' device extents of size 'size' at a given
2836 * offset
2837 */
2838 unsigned long long minsize = size;
2839 unsigned long long start_offset = ~0ULL;
2840 int dcnt = 0;
2841 if (minsize == 0)
2842 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
2843 for (dl = super->disks; dl ; dl = dl->next) {
2844 int found = 0;
2845
2846 pos = 0;
2847 i = 0;
2848 e = get_extents(super, dl);
2849 if (!e) continue;
2850 do {
2851 unsigned long long esize;
2852 esize = e[i].start - pos;
2853 if (esize >= minsize)
2854 found = 1;
2855 if (found && start_offset == ~0ULL) {
2856 start_offset = pos;
2857 break;
2858 } else if (found && pos != start_offset) {
2859 found = 0;
2860 break;
2861 }
2862 pos = e[i].start + e[i].size;
2863 i++;
2864 } while (e[i-1].size);
2865 if (found)
2866 dcnt++;
2867 free(e);
2868 }
2869 if (dcnt < raiddisks) {
2870 if (verbose)
2871 fprintf(stderr, Name ": imsm: Not enough "
2872 "devices with space for this array "
2873 "(%d < %d)\n",
2874 dcnt, raiddisks);
2875 return 0;
2876 }
2877 return 1;
2878 }
2879
2880 /* This device must be a member of the set */
2881 if (stat(dev, &stb) < 0)
2882 return 0;
2883 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2884 return 0;
2885 for (dl = super->disks ; dl ; dl = dl->next) {
2886 if (dl->major == major(stb.st_rdev) &&
2887 dl->minor == minor(stb.st_rdev))
2888 break;
2889 }
2890 if (!dl) {
2891 if (verbose)
2892 fprintf(stderr, Name ": %s is not in the "
2893 "same imsm set\n", dev);
2894 return 0;
2895 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
2896 /* If a volume is present then the current creation attempt
2897 * cannot incorporate new spares because the orom may not
2898 * understand this configuration (all member disks must be
2899 * members of each array in the container).
2900 */
2901 fprintf(stderr, Name ": %s is a spare and a volume"
2902 " is already defined for this container\n", dev);
2903 fprintf(stderr, Name ": The option-rom requires all member"
2904 " disks to be a member of all volumes\n");
2905 return 0;
2906 }
2907
2908 /* retrieve the largest free space block */
2909 e = get_extents(super, dl);
2910 maxsize = 0;
2911 i = 0;
2912 if (e) {
2913 do {
2914 unsigned long long esize;
2915
2916 esize = e[i].start - pos;
2917 if (esize >= maxsize)
2918 maxsize = esize;
2919 pos = e[i].start + e[i].size;
2920 i++;
2921 } while (e[i-1].size);
2922 dl->e = e;
2923 dl->extent_cnt = i;
2924 } else {
2925 if (verbose)
2926 fprintf(stderr, Name ": unable to determine free space for: %s\n",
2927 dev);
2928 return 0;
2929 }
2930 if (maxsize < size) {
2931 if (verbose)
2932 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
2933 dev, maxsize, size);
2934 return 0;
2935 }
2936
2937 /* count total number of extents for merge */
2938 i = 0;
2939 for (dl = super->disks; dl; dl = dl->next)
2940 if (dl->e)
2941 i += dl->extent_cnt;
2942
2943 maxsize = merge_extents(super, i);
2944 if (maxsize < size) {
2945 if (verbose)
2946 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
2947 maxsize, size);
2948 return 0;
2949 } else if (maxsize == ~0ULL) {
2950 if (verbose)
2951 fprintf(stderr, Name ": failed to merge %d extents\n", i);
2952 return 0;
2953 }
2954
2955 *freesize = maxsize;
2956
2957 return 1;
2958 }
2959
2960 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
2961 int raiddisks, int chunk, unsigned long long size,
2962 char *dev, unsigned long long *freesize,
2963 int verbose)
2964 {
2965 int fd, cfd;
2966 struct mdinfo *sra;
2967
2968 /* if given unused devices create a container
2969 * if given given devices in a container create a member volume
2970 */
2971 if (level == LEVEL_CONTAINER) {
2972 /* Must be a fresh device to add to a container */
2973 return validate_geometry_imsm_container(st, level, layout,
2974 raiddisks, chunk, size,
2975 dev, freesize,
2976 verbose);
2977 }
2978
2979 if (!dev) {
2980 if (st->sb && freesize) {
2981 /* Should do auto-layout here */
2982 fprintf(stderr, Name ": IMSM does not support auto-layout yet\n");
2983 return 0;
2984 }
2985 return 1;
2986 }
2987 if (st->sb) {
2988 /* creating in a given container */
2989 return validate_geometry_imsm_volume(st, level, layout,
2990 raiddisks, chunk, size,
2991 dev, freesize, verbose);
2992 }
2993
2994 /* limit creation to the following levels */
2995 if (!dev)
2996 switch (level) {
2997 case 0:
2998 case 1:
2999 case 10:
3000 case 5:
3001 break;
3002 default:
3003 return 1;
3004 }
3005
3006 /* This device needs to be a device in an 'imsm' container */
3007 fd = open(dev, O_RDONLY|O_EXCL, 0);
3008 if (fd >= 0) {
3009 if (verbose)
3010 fprintf(stderr,
3011 Name ": Cannot create this array on device %s\n",
3012 dev);
3013 close(fd);
3014 return 0;
3015 }
3016 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3017 if (verbose)
3018 fprintf(stderr, Name ": Cannot open %s: %s\n",
3019 dev, strerror(errno));
3020 return 0;
3021 }
3022 /* Well, it is in use by someone, maybe an 'imsm' container. */
3023 cfd = open_container(fd);
3024 if (cfd < 0) {
3025 close(fd);
3026 if (verbose)
3027 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3028 dev);
3029 return 0;
3030 }
3031 sra = sysfs_read(cfd, 0, GET_VERSION);
3032 close(fd);
3033 if (sra && sra->array.major_version == -1 &&
3034 strcmp(sra->text_version, "imsm") == 0) {
3035 /* This is a member of a imsm container. Load the container
3036 * and try to create a volume
3037 */
3038 struct intel_super *super;
3039
3040 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3041 st->sb = super;
3042 st->container_dev = fd2devnum(cfd);
3043 close(cfd);
3044 return validate_geometry_imsm_volume(st, level, layout,
3045 raiddisks, chunk,
3046 size, dev,
3047 freesize, verbose);
3048 }
3049 close(cfd);
3050 } else /* may belong to another container */
3051 return 0;
3052
3053 return 1;
3054 }
3055 #endif /* MDASSEMBLE */
3056
3057 static struct mdinfo *container_content_imsm(struct supertype *st)
3058 {
3059 /* Given a container loaded by load_super_imsm_all,
3060 * extract information about all the arrays into
3061 * an mdinfo tree.
3062 *
3063 * For each imsm_dev create an mdinfo, fill it in,
3064 * then look for matching devices in super->disks
3065 * and create appropriate device mdinfo.
3066 */
3067 struct intel_super *super = st->sb;
3068 struct imsm_super *mpb = super->anchor;
3069 struct mdinfo *rest = NULL;
3070 int i;
3071
3072 /* do not assemble arrays that might have bad blocks */
3073 if (imsm_bbm_log_size(super->anchor)) {
3074 fprintf(stderr, Name ": BBM log found in metadata. "
3075 "Cannot activate array(s).\n");
3076 return NULL;
3077 }
3078
3079 for (i = 0; i < mpb->num_raid_devs; i++) {
3080 struct imsm_dev *dev = get_imsm_dev(super, i);
3081 struct imsm_map *map = get_imsm_map(dev, 0);
3082 struct mdinfo *this;
3083 int slot;
3084
3085 this = malloc(sizeof(*this));
3086 memset(this, 0, sizeof(*this));
3087 this->next = rest;
3088
3089 super->current_vol = i;
3090 getinfo_super_imsm_volume(st, this);
3091 for (slot = 0 ; slot < map->num_members; slot++) {
3092 struct mdinfo *info_d;
3093 struct dl *d;
3094 int idx;
3095 int skip;
3096 __u32 s;
3097 __u32 ord;
3098
3099 skip = 0;
3100 idx = get_imsm_disk_idx(dev, slot);
3101 ord = get_imsm_ord_tbl_ent(dev, slot);
3102 for (d = super->disks; d ; d = d->next)
3103 if (d->index == idx)
3104 break;
3105
3106 if (d == NULL)
3107 skip = 1;
3108
3109 s = d ? d->disk.status : 0;
3110 if (s & FAILED_DISK)
3111 skip = 1;
3112 if (!(s & USABLE_DISK))
3113 skip = 1;
3114 if (ord & IMSM_ORD_REBUILD)
3115 skip = 1;
3116
3117 /*
3118 * if we skip some disks the array will be assmebled degraded;
3119 * reset resync start to avoid a dirty-degraded situation
3120 *
3121 * FIXME handle dirty degraded
3122 */
3123 if (skip && !dev->vol.dirty)
3124 this->resync_start = ~0ULL;
3125 if (skip)
3126 continue;
3127
3128 info_d = malloc(sizeof(*info_d));
3129 if (!info_d) {
3130 fprintf(stderr, Name ": failed to allocate disk"
3131 " for volume %s\n", (char *) dev->volume);
3132 free(this);
3133 this = rest;
3134 break;
3135 }
3136 memset(info_d, 0, sizeof(*info_d));
3137 info_d->next = this->devs;
3138 this->devs = info_d;
3139
3140 info_d->disk.number = d->index;
3141 info_d->disk.major = d->major;
3142 info_d->disk.minor = d->minor;
3143 info_d->disk.raid_disk = slot;
3144
3145 this->array.working_disks++;
3146
3147 info_d->events = __le32_to_cpu(mpb->generation_num);
3148 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3149 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3150 if (d->devname)
3151 strcpy(info_d->name, d->devname);
3152 }
3153 rest = this;
3154 }
3155
3156 return rest;
3157 }
3158
3159
3160 #ifndef MDASSEMBLE
3161 static int imsm_open_new(struct supertype *c, struct active_array *a,
3162 char *inst)
3163 {
3164 struct intel_super *super = c->sb;
3165 struct imsm_super *mpb = super->anchor;
3166
3167 if (atoi(inst) >= mpb->num_raid_devs) {
3168 fprintf(stderr, "%s: subarry index %d, out of range\n",
3169 __func__, atoi(inst));
3170 return -ENODEV;
3171 }
3172
3173 dprintf("imsm: open_new %s\n", inst);
3174 a->info.container_member = atoi(inst);
3175 return 0;
3176 }
3177
3178 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3179 {
3180 struct imsm_map *map = get_imsm_map(dev, 0);
3181
3182 if (!failed)
3183 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3184 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3185
3186 switch (get_imsm_raid_level(map)) {
3187 case 0:
3188 return IMSM_T_STATE_FAILED;
3189 break;
3190 case 1:
3191 if (failed < map->num_members)
3192 return IMSM_T_STATE_DEGRADED;
3193 else
3194 return IMSM_T_STATE_FAILED;
3195 break;
3196 case 10:
3197 {
3198 /**
3199 * check to see if any mirrors have failed, otherwise we
3200 * are degraded. Even numbered slots are mirrored on
3201 * slot+1
3202 */
3203 int i;
3204 /* gcc -Os complains that this is unused */
3205 int insync = insync;
3206
3207 for (i = 0; i < map->num_members; i++) {
3208 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3209 int idx = ord_to_idx(ord);
3210 struct imsm_disk *disk;
3211
3212 /* reset the potential in-sync count on even-numbered
3213 * slots. num_copies is always 2 for imsm raid10
3214 */
3215 if ((i & 1) == 0)
3216 insync = 2;
3217
3218 disk = get_imsm_disk(super, idx);
3219 if (!disk || disk->status & FAILED_DISK ||
3220 ord & IMSM_ORD_REBUILD)
3221 insync--;
3222
3223 /* no in-sync disks left in this mirror the
3224 * array has failed
3225 */
3226 if (insync == 0)
3227 return IMSM_T_STATE_FAILED;
3228 }
3229
3230 return IMSM_T_STATE_DEGRADED;
3231 }
3232 case 5:
3233 if (failed < 2)
3234 return IMSM_T_STATE_DEGRADED;
3235 else
3236 return IMSM_T_STATE_FAILED;
3237 break;
3238 default:
3239 break;
3240 }
3241
3242 return map->map_state;
3243 }
3244
3245 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3246 {
3247 int i;
3248 int failed = 0;
3249 struct imsm_disk *disk;
3250 struct imsm_map *map = get_imsm_map(dev, 0);
3251
3252 for (i = 0; i < map->num_members; i++) {
3253 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3254 int idx = ord_to_idx(ord);
3255
3256 disk = get_imsm_disk(super, idx);
3257 if (!disk || disk->status & FAILED_DISK ||
3258 ord & IMSM_ORD_REBUILD)
3259 failed++;
3260 }
3261
3262 return failed;
3263 }
3264
3265 static int is_resyncing(struct imsm_dev *dev)
3266 {
3267 struct imsm_map *migr_map;
3268
3269 if (!dev->vol.migr_state)
3270 return 0;
3271
3272 if (dev->vol.migr_type == MIGR_INIT)
3273 return 1;
3274
3275 migr_map = get_imsm_map(dev, 1);
3276
3277 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
3278 return 1;
3279 else
3280 return 0;
3281 }
3282
3283 static int is_rebuilding(struct imsm_dev *dev)
3284 {
3285 struct imsm_map *migr_map;
3286
3287 if (!dev->vol.migr_state)
3288 return 0;
3289
3290 if (dev->vol.migr_type != MIGR_REBUILD)
3291 return 0;
3292
3293 migr_map = get_imsm_map(dev, 1);
3294
3295 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
3296 return 1;
3297 else
3298 return 0;
3299 }
3300
3301 static void mark_failure(struct imsm_disk *disk)
3302 {
3303 if (disk->status & FAILED_DISK)
3304 return;
3305 disk->status |= FAILED_DISK;
3306 disk->scsi_id = __cpu_to_le32(~(__u32)0);
3307 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
3308 }
3309
3310 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3311 * states are handled in imsm_set_disk() with one exception, when a
3312 * resync is stopped due to a new failure this routine will set the
3313 * 'degraded' state for the array.
3314 */
3315 static int imsm_set_array_state(struct active_array *a, int consistent)
3316 {
3317 int inst = a->info.container_member;
3318 struct intel_super *super = a->container->sb;
3319 struct imsm_dev *dev = get_imsm_dev(super, inst);
3320 struct imsm_map *map = get_imsm_map(dev, 0);
3321 int failed = imsm_count_failed(super, dev);
3322 __u8 map_state = imsm_check_degraded(super, dev, failed);
3323
3324 /* before we activate this array handle any missing disks */
3325 if (consistent == 2 && super->missing) {
3326 struct dl *dl;
3327
3328 dprintf("imsm: mark missing\n");
3329 end_migration(dev, map_state);
3330 for (dl = super->missing; dl; dl = dl->next)
3331 mark_failure(&dl->disk);
3332 super->updates_pending++;
3333 }
3334
3335 if (consistent == 2 &&
3336 (!is_resync_complete(a) ||
3337 map_state != IMSM_T_STATE_NORMAL ||
3338 dev->vol.migr_state))
3339 consistent = 0;
3340
3341 if (is_resync_complete(a)) {
3342 /* complete intialization / resync,
3343 * recovery is completed in ->set_disk
3344 */
3345 if (is_resyncing(dev)) {
3346 dprintf("imsm: mark resync done\n");
3347 end_migration(dev, map_state);
3348 super->updates_pending++;
3349 }
3350 } else if (!is_resyncing(dev) && !failed) {
3351 /* mark the start of the init process if nothing is failed */
3352 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
3353 if (map->map_state == IMSM_T_STATE_NORMAL)
3354 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REBUILD);
3355 else
3356 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
3357 super->updates_pending++;
3358 }
3359
3360 /* check if we can update the migration checkpoint */
3361 if (dev->vol.migr_state &&
3362 __le32_to_cpu(dev->vol.curr_migr_unit) != a->resync_start) {
3363 dprintf("imsm: checkpoint migration (%llu)\n", a->resync_start);
3364 dev->vol.curr_migr_unit = __cpu_to_le32(a->resync_start);
3365 super->updates_pending++;
3366 }
3367
3368 /* mark dirty / clean */
3369 if (dev->vol.dirty != !consistent) {
3370 dprintf("imsm: mark '%s' (%llu)\n",
3371 consistent ? "clean" : "dirty", a->resync_start);
3372 if (consistent)
3373 dev->vol.dirty = 0;
3374 else
3375 dev->vol.dirty = 1;
3376 super->updates_pending++;
3377 }
3378 return consistent;
3379 }
3380
3381 static void imsm_set_disk(struct active_array *a, int n, int state)
3382 {
3383 int inst = a->info.container_member;
3384 struct intel_super *super = a->container->sb;
3385 struct imsm_dev *dev = get_imsm_dev(super, inst);
3386 struct imsm_map *map = get_imsm_map(dev, 0);
3387 struct imsm_disk *disk;
3388 int failed;
3389 __u32 ord;
3390 __u8 map_state;
3391
3392 if (n > map->num_members)
3393 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
3394 n, map->num_members - 1);
3395
3396 if (n < 0)
3397 return;
3398
3399 dprintf("imsm: set_disk %d:%x\n", n, state);
3400
3401 ord = get_imsm_ord_tbl_ent(dev, n);
3402 disk = get_imsm_disk(super, ord_to_idx(ord));
3403
3404 /* check for new failures */
3405 if ((state & DS_FAULTY) && !(disk->status & FAILED_DISK)) {
3406 mark_failure(disk);
3407 super->updates_pending++;
3408 }
3409
3410 /* check if in_sync */
3411 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD) {
3412 struct imsm_map *migr_map = get_imsm_map(dev, 1);
3413
3414 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
3415 super->updates_pending++;
3416 }
3417
3418 failed = imsm_count_failed(super, dev);
3419 map_state = imsm_check_degraded(super, dev, failed);
3420
3421 /* check if recovery complete, newly degraded, or failed */
3422 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
3423 end_migration(dev, map_state);
3424 super->updates_pending++;
3425 } else if (map_state == IMSM_T_STATE_DEGRADED &&
3426 map->map_state != map_state &&
3427 !dev->vol.migr_state) {
3428 dprintf("imsm: mark degraded\n");
3429 map->map_state = map_state;
3430 super->updates_pending++;
3431 } else if (map_state == IMSM_T_STATE_FAILED &&
3432 map->map_state != map_state) {
3433 dprintf("imsm: mark failed\n");
3434 end_migration(dev, map_state);
3435 super->updates_pending++;
3436 }
3437 }
3438
3439 static int store_imsm_mpb(int fd, struct intel_super *super)
3440 {
3441 struct imsm_super *mpb = super->anchor;
3442 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
3443 unsigned long long dsize;
3444 unsigned long long sectors;
3445
3446 get_dev_size(fd, NULL, &dsize);
3447
3448 if (mpb_size > 512) {
3449 /* -1 to account for anchor */
3450 sectors = mpb_sectors(mpb) - 1;
3451
3452 /* write the extended mpb to the sectors preceeding the anchor */
3453 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
3454 return 1;
3455
3456 if (write(fd, super->buf + 512, 512 * sectors) != 512 * sectors)
3457 return 1;
3458 }
3459
3460 /* first block is stored on second to last sector of the disk */
3461 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
3462 return 1;
3463
3464 if (write(fd, super->buf, 512) != 512)
3465 return 1;
3466
3467 return 0;
3468 }
3469
3470 static void imsm_sync_metadata(struct supertype *container)
3471 {
3472 struct intel_super *super = container->sb;
3473
3474 if (!super->updates_pending)
3475 return;
3476
3477 write_super_imsm(super, 0);
3478
3479 super->updates_pending = 0;
3480 }
3481
3482 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
3483 {
3484 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3485 int i = get_imsm_disk_idx(dev, idx);
3486 struct dl *dl;
3487
3488 for (dl = super->disks; dl; dl = dl->next)
3489 if (dl->index == i)
3490 break;
3491
3492 if (dl && dl->disk.status & FAILED_DISK)
3493 dl = NULL;
3494
3495 if (dl)
3496 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
3497
3498 return dl;
3499 }
3500
3501 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
3502 struct active_array *a, int activate_new)
3503 {
3504 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3505 int idx = get_imsm_disk_idx(dev, slot);
3506 struct imsm_super *mpb = super->anchor;
3507 struct imsm_map *map;
3508 unsigned long long esize;
3509 unsigned long long pos;
3510 struct mdinfo *d;
3511 struct extent *ex;
3512 int i, j;
3513 int found;
3514 __u32 array_start;
3515 __u32 blocks;
3516 struct dl *dl;
3517
3518 for (dl = super->disks; dl; dl = dl->next) {
3519 /* If in this array, skip */
3520 for (d = a->info.devs ; d ; d = d->next)
3521 if (d->state_fd >= 0 &&
3522 d->disk.major == dl->major &&
3523 d->disk.minor == dl->minor) {
3524 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3525 break;
3526 }
3527 if (d)
3528 continue;
3529
3530 /* skip in use or failed drives */
3531 if (dl->disk.status & FAILED_DISK || idx == dl->index) {
3532 dprintf("%x:%x status ( %s%s)\n",
3533 dl->major, dl->minor,
3534 dl->disk.status & FAILED_DISK ? "failed " : "",
3535 idx == dl->index ? "in use " : "");
3536 continue;
3537 }
3538
3539 /* skip pure spares when we are looking for partially
3540 * assimilated drives
3541 */
3542 if (dl->index == -1 && !activate_new)
3543 continue;
3544
3545 /* Does this unused device have the requisite free space?
3546 * It needs to be able to cover all member volumes
3547 */
3548 ex = get_extents(super, dl);
3549 if (!ex) {
3550 dprintf("cannot get extents\n");
3551 continue;
3552 }
3553 for (i = 0; i < mpb->num_raid_devs; i++) {
3554 dev = get_imsm_dev(super, i);
3555 map = get_imsm_map(dev, 0);
3556
3557 /* check if this disk is already a member of
3558 * this array
3559 */
3560 for (j = 0; j < map->num_members; j++)
3561 if (get_imsm_disk_idx(dev, j) == dl->index)
3562 break;
3563 if (j < map->num_members)
3564 continue;
3565
3566 found = 0;
3567 j = 0;
3568 pos = 0;
3569 array_start = __le32_to_cpu(map->pba_of_lba0);
3570 blocks = __le32_to_cpu(map->blocks_per_member);
3571
3572 do {
3573 /* check that we can start at pba_of_lba0 with
3574 * blocks_per_member of space
3575 */
3576 esize = ex[j].start - pos;
3577 if (array_start >= pos &&
3578 array_start + blocks < ex[j].start) {
3579 found = 1;
3580 break;
3581 }
3582 pos = ex[j].start + ex[j].size;
3583 j++;
3584 } while (ex[j-1].size);
3585
3586 if (!found)
3587 break;
3588 }
3589
3590 free(ex);
3591 if (i < mpb->num_raid_devs) {
3592 dprintf("%x:%x does not have %u at %u\n",
3593 dl->major, dl->minor,
3594 blocks, array_start);
3595 /* No room */
3596 continue;
3597 }
3598 return dl;
3599 }
3600
3601 return dl;
3602 }
3603
3604 static struct mdinfo *imsm_activate_spare(struct active_array *a,
3605 struct metadata_update **updates)
3606 {
3607 /**
3608 * Find a device with unused free space and use it to replace a
3609 * failed/vacant region in an array. We replace failed regions one a
3610 * array at a time. The result is that a new spare disk will be added
3611 * to the first failed array and after the monitor has finished
3612 * propagating failures the remainder will be consumed.
3613 *
3614 * FIXME add a capability for mdmon to request spares from another
3615 * container.
3616 */
3617
3618 struct intel_super *super = a->container->sb;
3619 int inst = a->info.container_member;
3620 struct imsm_dev *dev = get_imsm_dev(super, inst);
3621 struct imsm_map *map = get_imsm_map(dev, 0);
3622 int failed = a->info.array.raid_disks;
3623 struct mdinfo *rv = NULL;
3624 struct mdinfo *d;
3625 struct mdinfo *di;
3626 struct metadata_update *mu;
3627 struct dl *dl;
3628 struct imsm_update_activate_spare *u;
3629 int num_spares = 0;
3630 int i;
3631
3632 for (d = a->info.devs ; d ; d = d->next) {
3633 if ((d->curr_state & DS_FAULTY) &&
3634 d->state_fd >= 0)
3635 /* wait for Removal to happen */
3636 return NULL;
3637 if (d->state_fd >= 0)
3638 failed--;
3639 }
3640
3641 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3642 inst, failed, a->info.array.raid_disks, a->info.array.level);
3643 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
3644 return NULL;
3645
3646 /* For each slot, if it is not working, find a spare */
3647 for (i = 0; i < a->info.array.raid_disks; i++) {
3648 for (d = a->info.devs ; d ; d = d->next)
3649 if (d->disk.raid_disk == i)
3650 break;
3651 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3652 if (d && (d->state_fd >= 0))
3653 continue;
3654
3655 /*
3656 * OK, this device needs recovery. Try to re-add the
3657 * previous occupant of this slot, if this fails see if
3658 * we can continue the assimilation of a spare that was
3659 * partially assimilated, finally try to activate a new
3660 * spare.
3661 */
3662 dl = imsm_readd(super, i, a);
3663 if (!dl)
3664 dl = imsm_add_spare(super, i, a, 0);
3665 if (!dl)
3666 dl = imsm_add_spare(super, i, a, 1);
3667 if (!dl)
3668 continue;
3669
3670 /* found a usable disk with enough space */
3671 di = malloc(sizeof(*di));
3672 if (!di)
3673 continue;
3674 memset(di, 0, sizeof(*di));
3675
3676 /* dl->index will be -1 in the case we are activating a
3677 * pristine spare. imsm_process_update() will create a
3678 * new index in this case. Once a disk is found to be
3679 * failed in all member arrays it is kicked from the
3680 * metadata
3681 */
3682 di->disk.number = dl->index;
3683
3684 /* (ab)use di->devs to store a pointer to the device
3685 * we chose
3686 */
3687 di->devs = (struct mdinfo *) dl;
3688
3689 di->disk.raid_disk = i;
3690 di->disk.major = dl->major;
3691 di->disk.minor = dl->minor;
3692 di->disk.state = 0;
3693 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
3694 di->component_size = a->info.component_size;
3695 di->container_member = inst;
3696 di->next = rv;
3697 rv = di;
3698 num_spares++;
3699 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3700 i, di->data_offset);
3701
3702 break;
3703 }
3704
3705 if (!rv)
3706 /* No spares found */
3707 return rv;
3708 /* Now 'rv' has a list of devices to return.
3709 * Create a metadata_update record to update the
3710 * disk_ord_tbl for the array
3711 */
3712 mu = malloc(sizeof(*mu));
3713 if (mu) {
3714 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
3715 if (mu->buf == NULL) {
3716 free(mu);
3717 mu = NULL;
3718 }
3719 }
3720 if (!mu) {
3721 while (rv) {
3722 struct mdinfo *n = rv->next;
3723
3724 free(rv);
3725 rv = n;
3726 }
3727 return NULL;
3728 }
3729
3730 mu->space = NULL;
3731 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
3732 mu->next = *updates;
3733 u = (struct imsm_update_activate_spare *) mu->buf;
3734
3735 for (di = rv ; di ; di = di->next) {
3736 u->type = update_activate_spare;
3737 u->dl = (struct dl *) di->devs;
3738 di->devs = NULL;
3739 u->slot = di->disk.raid_disk;
3740 u->array = inst;
3741 u->next = u + 1;
3742 u++;
3743 }
3744 (u-1)->next = NULL;
3745 *updates = mu;
3746
3747 return rv;
3748 }
3749
3750 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
3751 {
3752 struct imsm_dev *dev = get_imsm_dev(super, idx);
3753 struct imsm_map *map = get_imsm_map(dev, 0);
3754 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
3755 struct disk_info *inf = get_disk_info(u);
3756 struct imsm_disk *disk;
3757 int i;
3758 int j;
3759
3760 for (i = 0; i < map->num_members; i++) {
3761 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
3762 for (j = 0; j < new_map->num_members; j++)
3763 if (serialcmp(disk->serial, inf[j].serial) == 0)
3764 return 1;
3765 }
3766
3767 return 0;
3768 }
3769
3770 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
3771
3772 static void imsm_process_update(struct supertype *st,
3773 struct metadata_update *update)
3774 {
3775 /**
3776 * crack open the metadata_update envelope to find the update record
3777 * update can be one of:
3778 * update_activate_spare - a spare device has replaced a failed
3779 * device in an array, update the disk_ord_tbl. If this disk is
3780 * present in all member arrays then also clear the SPARE_DISK
3781 * flag
3782 */
3783 struct intel_super *super = st->sb;
3784 struct imsm_super *mpb;
3785 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
3786
3787 /* update requires a larger buf but the allocation failed */
3788 if (super->next_len && !super->next_buf) {
3789 super->next_len = 0;
3790 return;
3791 }
3792
3793 if (super->next_buf) {
3794 memcpy(super->next_buf, super->buf, super->len);
3795 free(super->buf);
3796 super->len = super->next_len;
3797 super->buf = super->next_buf;
3798
3799 super->next_len = 0;
3800 super->next_buf = NULL;
3801 }
3802
3803 mpb = super->anchor;
3804
3805 switch (type) {
3806 case update_activate_spare: {
3807 struct imsm_update_activate_spare *u = (void *) update->buf;
3808 struct imsm_dev *dev = get_imsm_dev(super, u->array);
3809 struct imsm_map *map = get_imsm_map(dev, 0);
3810 struct imsm_map *migr_map;
3811 struct active_array *a;
3812 struct imsm_disk *disk;
3813 __u8 to_state;
3814 struct dl *dl;
3815 unsigned int found;
3816 int failed;
3817 int victim = get_imsm_disk_idx(dev, u->slot);
3818 int i;
3819
3820 for (dl = super->disks; dl; dl = dl->next)
3821 if (dl == u->dl)
3822 break;
3823
3824 if (!dl) {
3825 fprintf(stderr, "error: imsm_activate_spare passed "
3826 "an unknown disk (index: %d)\n",
3827 u->dl->index);
3828 return;
3829 }
3830
3831 super->updates_pending++;
3832
3833 /* count failures (excluding rebuilds and the victim)
3834 * to determine map[0] state
3835 */
3836 failed = 0;
3837 for (i = 0; i < map->num_members; i++) {
3838 if (i == u->slot)
3839 continue;
3840 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
3841 if (!disk || disk->status & FAILED_DISK)
3842 failed++;
3843 }
3844
3845 /* adding a pristine spare, assign a new index */
3846 if (dl->index < 0) {
3847 dl->index = super->anchor->num_disks;
3848 super->anchor->num_disks++;
3849 }
3850 disk = &dl->disk;
3851 disk->status |= CONFIGURED_DISK;
3852 disk->status &= ~SPARE_DISK;
3853
3854 /* mark rebuild */
3855 to_state = imsm_check_degraded(super, dev, failed);
3856 map->map_state = IMSM_T_STATE_DEGRADED;
3857 migrate(dev, to_state, MIGR_REBUILD);
3858 migr_map = get_imsm_map(dev, 1);
3859 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
3860 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
3861
3862 /* count arrays using the victim in the metadata */
3863 found = 0;
3864 for (a = st->arrays; a ; a = a->next) {
3865 dev = get_imsm_dev(super, a->info.container_member);
3866 for (i = 0; i < map->num_members; i++)
3867 if (victim == get_imsm_disk_idx(dev, i))
3868 found++;
3869 }
3870
3871 /* delete the victim if it is no longer being
3872 * utilized anywhere
3873 */
3874 if (!found) {
3875 struct dl **dlp;
3876
3877 /* We know that 'manager' isn't touching anything,
3878 * so it is safe to delete
3879 */
3880 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
3881 if ((*dlp)->index == victim)
3882 break;
3883
3884 /* victim may be on the missing list */
3885 if (!*dlp)
3886 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
3887 if ((*dlp)->index == victim)
3888 break;
3889 imsm_delete(super, dlp, victim);
3890 }
3891 break;
3892 }
3893 case update_create_array: {
3894 /* someone wants to create a new array, we need to be aware of
3895 * a few races/collisions:
3896 * 1/ 'Create' called by two separate instances of mdadm
3897 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
3898 * devices that have since been assimilated via
3899 * activate_spare.
3900 * In the event this update can not be carried out mdadm will
3901 * (FIX ME) notice that its update did not take hold.
3902 */
3903 struct imsm_update_create_array *u = (void *) update->buf;
3904 struct intel_dev *dv;
3905 struct imsm_dev *dev;
3906 struct imsm_map *map, *new_map;
3907 unsigned long long start, end;
3908 unsigned long long new_start, new_end;
3909 int i;
3910 struct disk_info *inf;
3911 struct dl *dl;
3912
3913 /* handle racing creates: first come first serve */
3914 if (u->dev_idx < mpb->num_raid_devs) {
3915 dprintf("%s: subarray %d already defined\n",
3916 __func__, u->dev_idx);
3917 goto create_error;
3918 }
3919
3920 /* check update is next in sequence */
3921 if (u->dev_idx != mpb->num_raid_devs) {
3922 dprintf("%s: can not create array %d expected index %d\n",
3923 __func__, u->dev_idx, mpb->num_raid_devs);
3924 goto create_error;
3925 }
3926
3927 new_map = get_imsm_map(&u->dev, 0);
3928 new_start = __le32_to_cpu(new_map->pba_of_lba0);
3929 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
3930 inf = get_disk_info(u);
3931
3932 /* handle activate_spare versus create race:
3933 * check to make sure that overlapping arrays do not include
3934 * overalpping disks
3935 */
3936 for (i = 0; i < mpb->num_raid_devs; i++) {
3937 dev = get_imsm_dev(super, i);
3938 map = get_imsm_map(dev, 0);
3939 start = __le32_to_cpu(map->pba_of_lba0);
3940 end = start + __le32_to_cpu(map->blocks_per_member);
3941 if ((new_start >= start && new_start <= end) ||
3942 (start >= new_start && start <= new_end))
3943 /* overlap */;
3944 else
3945 continue;
3946
3947 if (disks_overlap(super, i, u)) {
3948 dprintf("%s: arrays overlap\n", __func__);
3949 goto create_error;
3950 }
3951 }
3952
3953 /* check that prepare update was successful */
3954 if (!update->space) {
3955 dprintf("%s: prepare update failed\n", __func__);
3956 goto create_error;
3957 }
3958
3959 /* check that all disks are still active before committing
3960 * changes. FIXME: could we instead handle this by creating a
3961 * degraded array? That's probably not what the user expects,
3962 * so better to drop this update on the floor.
3963 */
3964 for (i = 0; i < new_map->num_members; i++) {
3965 dl = serial_to_dl(inf[i].serial, super);
3966 if (!dl) {
3967 dprintf("%s: disk disappeared\n", __func__);
3968 goto create_error;
3969 }
3970 }
3971
3972 super->updates_pending++;
3973
3974 /* convert spares to members and fixup ord_tbl */
3975 for (i = 0; i < new_map->num_members; i++) {
3976 dl = serial_to_dl(inf[i].serial, super);
3977 if (dl->index == -1) {
3978 dl->index = mpb->num_disks;
3979 mpb->num_disks++;
3980 dl->disk.status |= CONFIGURED_DISK;
3981 dl->disk.status &= ~SPARE_DISK;
3982 }
3983 set_imsm_ord_tbl_ent(new_map, i, dl->index);
3984 }
3985
3986 dv = update->space;
3987 dev = dv->dev;
3988 update->space = NULL;
3989 imsm_copy_dev(dev, &u->dev);
3990 dv->index = u->dev_idx;
3991 dv->next = super->devlist;
3992 super->devlist = dv;
3993 mpb->num_raid_devs++;
3994
3995 imsm_update_version_info(super);
3996 break;
3997 create_error:
3998 /* mdmon knows how to release update->space, but not
3999 * ((struct intel_dev *) update->space)->dev
4000 */
4001 if (update->space) {
4002 dv = update->space;
4003 free(dv->dev);
4004 }
4005 break;
4006 }
4007 case update_add_disk:
4008
4009 /* we may be able to repair some arrays if disks are
4010 * being added */
4011 if (super->add) {
4012 struct active_array *a;
4013
4014 super->updates_pending++;
4015 for (a = st->arrays; a; a = a->next)
4016 a->check_degraded = 1;
4017 }
4018 /* add some spares to the metadata */
4019 while (super->add) {
4020 struct dl *al;
4021
4022 al = super->add;
4023 super->add = al->next;
4024 al->next = super->disks;
4025 super->disks = al;
4026 dprintf("%s: added %x:%x\n",
4027 __func__, al->major, al->minor);
4028 }
4029
4030 break;
4031 }
4032 }
4033
4034 static void imsm_prepare_update(struct supertype *st,
4035 struct metadata_update *update)
4036 {
4037 /**
4038 * Allocate space to hold new disk entries, raid-device entries or a new
4039 * mpb if necessary. The manager synchronously waits for updates to
4040 * complete in the monitor, so new mpb buffers allocated here can be
4041 * integrated by the monitor thread without worrying about live pointers
4042 * in the manager thread.
4043 */
4044 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4045 struct intel_super *super = st->sb;
4046 struct imsm_super *mpb = super->anchor;
4047 size_t buf_len;
4048 size_t len = 0;
4049
4050 switch (type) {
4051 case update_create_array: {
4052 struct imsm_update_create_array *u = (void *) update->buf;
4053 struct intel_dev *dv;
4054 struct imsm_dev *dev = &u->dev;
4055 struct imsm_map *map = get_imsm_map(dev, 0);
4056 struct dl *dl;
4057 struct disk_info *inf;
4058 int i;
4059 int activate = 0;
4060
4061 inf = get_disk_info(u);
4062 len = sizeof_imsm_dev(dev, 1);
4063 /* allocate a new super->devlist entry */
4064 dv = malloc(sizeof(*dv));
4065 if (dv) {
4066 dv->dev = malloc(len);
4067 if (dv->dev)
4068 update->space = dv;
4069 else {
4070 free(dv);
4071 update->space = NULL;
4072 }
4073 }
4074
4075 /* count how many spares will be converted to members */
4076 for (i = 0; i < map->num_members; i++) {
4077 dl = serial_to_dl(inf[i].serial, super);
4078 if (!dl) {
4079 /* hmm maybe it failed?, nothing we can do about
4080 * it here
4081 */
4082 continue;
4083 }
4084 if (count_memberships(dl, super) == 0)
4085 activate++;
4086 }
4087 len += activate * sizeof(struct imsm_disk);
4088 break;
4089 default:
4090 break;
4091 }
4092 }
4093
4094 /* check if we need a larger metadata buffer */
4095 if (super->next_buf)
4096 buf_len = super->next_len;
4097 else
4098 buf_len = super->len;
4099
4100 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4101 /* ok we need a larger buf than what is currently allocated
4102 * if this allocation fails process_update will notice that
4103 * ->next_len is set and ->next_buf is NULL
4104 */
4105 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4106 if (super->next_buf)
4107 free(super->next_buf);
4108
4109 super->next_len = buf_len;
4110 if (posix_memalign(&super->next_buf, 512, buf_len) != 0)
4111 super->next_buf = NULL;
4112 }
4113 }
4114
4115 /* must be called while manager is quiesced */
4116 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4117 {
4118 struct imsm_super *mpb = super->anchor;
4119 struct dl *iter;
4120 struct imsm_dev *dev;
4121 struct imsm_map *map;
4122 int i, j, num_members;
4123 __u32 ord;
4124
4125 dprintf("%s: deleting device[%d] from imsm_super\n",
4126 __func__, index);
4127
4128 /* shift all indexes down one */
4129 for (iter = super->disks; iter; iter = iter->next)
4130 if (iter->index > index)
4131 iter->index--;
4132 for (iter = super->missing; iter; iter = iter->next)
4133 if (iter->index > index)
4134 iter->index--;
4135
4136 for (i = 0; i < mpb->num_raid_devs; i++) {
4137 dev = get_imsm_dev(super, i);
4138 map = get_imsm_map(dev, 0);
4139 num_members = map->num_members;
4140 for (j = 0; j < num_members; j++) {
4141 /* update ord entries being careful not to propagate
4142 * ord-flags to the first map
4143 */
4144 ord = get_imsm_ord_tbl_ent(dev, j);
4145
4146 if (ord_to_idx(ord) <= index)
4147 continue;
4148
4149 map = get_imsm_map(dev, 0);
4150 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4151 map = get_imsm_map(dev, 1);
4152 if (map)
4153 set_imsm_ord_tbl_ent(map, j, ord - 1);
4154 }
4155 }
4156
4157 mpb->num_disks--;
4158 super->updates_pending++;
4159 if (*dlp) {
4160 struct dl *dl = *dlp;
4161
4162 *dlp = (*dlp)->next;
4163 __free_imsm_disk(dl);
4164 }
4165 }
4166 #endif /* MDASSEMBLE */
4167
4168 struct superswitch super_imsm = {
4169 #ifndef MDASSEMBLE
4170 .examine_super = examine_super_imsm,
4171 .brief_examine_super = brief_examine_super_imsm,
4172 .detail_super = detail_super_imsm,
4173 .brief_detail_super = brief_detail_super_imsm,
4174 .write_init_super = write_init_super_imsm,
4175 .validate_geometry = validate_geometry_imsm,
4176 .add_to_super = add_to_super_imsm,
4177 .detail_platform = detail_platform_imsm,
4178 #endif
4179 .match_home = match_home_imsm,
4180 .uuid_from_super= uuid_from_super_imsm,
4181 .getinfo_super = getinfo_super_imsm,
4182 .update_super = update_super_imsm,
4183
4184 .avail_size = avail_size_imsm,
4185
4186 .compare_super = compare_super_imsm,
4187
4188 .load_super = load_super_imsm,
4189 .init_super = init_super_imsm,
4190 .store_super = store_zero_imsm,
4191 .free_super = free_super_imsm,
4192 .match_metadata_desc = match_metadata_desc_imsm,
4193 .container_content = container_content_imsm,
4194 .default_layout = imsm_level_to_layout,
4195
4196 .external = 1,
4197 .name = "imsm",
4198
4199 #ifndef MDASSEMBLE
4200 /* for mdmon */
4201 .open_new = imsm_open_new,
4202 .load_super = load_super_imsm,
4203 .set_array_state= imsm_set_array_state,
4204 .set_disk = imsm_set_disk,
4205 .sync_metadata = imsm_sync_metadata,
4206 .activate_spare = imsm_activate_spare,
4207 .process_update = imsm_process_update,
4208 .prepare_update = imsm_prepare_update,
4209 #endif /* MDASSEMBLE */
4210 };
Unnamed repository; edit this file 'description' to name the repository.