]> 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
test: fix a call to udevsettle
[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 & SPARE_DISK ? 0 : (1 << MD_DISK_SYNC);
1199 if (s & FAILED_DISK || super->disks->index == -2) {
1200 info->disk.state |= 1 << MD_DISK_FAULTY;
1201 info->disk.raid_disk = -2;
1202 }
1203 }
1204
1205 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1206 * ->compare_super may have updated the 'num_raid_devs' field for spares
1207 */
1208 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1209 uuid_from_super_imsm(st, info->uuid);
1210 else
1211 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1212 }
1213
1214 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1215 char *update, char *devname, int verbose,
1216 int uuid_set, char *homehost)
1217 {
1218 /* FIXME */
1219
1220 /* For 'assemble' and 'force' we need to return non-zero if any
1221 * change was made. For others, the return value is ignored.
1222 * Update options are:
1223 * force-one : This device looks a bit old but needs to be included,
1224 * update age info appropriately.
1225 * assemble: clear any 'faulty' flag to allow this device to
1226 * be assembled.
1227 * force-array: Array is degraded but being forced, mark it clean
1228 * if that will be needed to assemble it.
1229 *
1230 * newdev: not used ????
1231 * grow: Array has gained a new device - this is currently for
1232 * linear only
1233 * resync: mark as dirty so a resync will happen.
1234 * name: update the name - preserving the homehost
1235 *
1236 * Following are not relevant for this imsm:
1237 * sparc2.2 : update from old dodgey metadata
1238 * super-minor: change the preferred_minor number
1239 * summaries: update redundant counters.
1240 * uuid: Change the uuid of the array to match watch is given
1241 * homehost: update the recorded homehost
1242 * _reshape_progress: record new reshape_progress position.
1243 */
1244 int rv = 0;
1245 //struct intel_super *super = st->sb;
1246 //struct imsm_super *mpb = super->mpb;
1247
1248 if (strcmp(update, "grow") == 0) {
1249 }
1250 if (strcmp(update, "resync") == 0) {
1251 /* dev->vol.dirty = 1; */
1252 }
1253
1254 /* IMSM has no concept of UUID or homehost */
1255
1256 return rv;
1257 }
1258
1259 static size_t disks_to_mpb_size(int disks)
1260 {
1261 size_t size;
1262
1263 size = sizeof(struct imsm_super);
1264 size += (disks - 1) * sizeof(struct imsm_disk);
1265 size += 2 * sizeof(struct imsm_dev);
1266 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1267 size += (4 - 2) * sizeof(struct imsm_map);
1268 /* 4 possible disk_ord_tbl's */
1269 size += 4 * (disks - 1) * sizeof(__u32);
1270
1271 return size;
1272 }
1273
1274 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1275 {
1276 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1277 return 0;
1278
1279 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1280 }
1281
1282 static void free_devlist(struct intel_super *super)
1283 {
1284 struct intel_dev *dv;
1285
1286 while (super->devlist) {
1287 dv = super->devlist->next;
1288 free(super->devlist->dev);
1289 free(super->devlist);
1290 super->devlist = dv;
1291 }
1292 }
1293
1294 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1295 {
1296 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1297 }
1298
1299 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1300 {
1301 /*
1302 * return:
1303 * 0 same, or first was empty, and second was copied
1304 * 1 second had wrong number
1305 * 2 wrong uuid
1306 * 3 wrong other info
1307 */
1308 struct intel_super *first = st->sb;
1309 struct intel_super *sec = tst->sb;
1310
1311 if (!first) {
1312 st->sb = tst->sb;
1313 tst->sb = NULL;
1314 return 0;
1315 }
1316
1317 if (memcmp(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH) != 0)
1318 return 3;
1319
1320 /* if an anchor does not have num_raid_devs set then it is a free
1321 * floating spare
1322 */
1323 if (first->anchor->num_raid_devs > 0 &&
1324 sec->anchor->num_raid_devs > 0) {
1325 if (first->anchor->family_num != sec->anchor->family_num)
1326 return 3;
1327 }
1328
1329 /* if 'first' is a spare promote it to a populated mpb with sec's
1330 * family number
1331 */
1332 if (first->anchor->num_raid_devs == 0 &&
1333 sec->anchor->num_raid_devs > 0) {
1334 int i;
1335 struct intel_dev *dv;
1336 struct imsm_dev *dev;
1337
1338 /* we need to copy raid device info from sec if an allocation
1339 * fails here we don't associate the spare
1340 */
1341 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1342 dv = malloc(sizeof(*dv));
1343 if (!dv)
1344 break;
1345 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1346 if (!dev) {
1347 free(dv);
1348 break;
1349 }
1350 dv->dev = dev;
1351 dv->index = i;
1352 dv->next = first->devlist;
1353 first->devlist = dv;
1354 }
1355 if (i <= sec->anchor->num_raid_devs) {
1356 /* allocation failure */
1357 free_devlist(first);
1358 fprintf(stderr, "imsm: failed to associate spare\n");
1359 return 3;
1360 }
1361 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1362 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1363
1364 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1365 first->anchor->family_num = sec->anchor->family_num;
1366 }
1367
1368 return 0;
1369 }
1370
1371 static void fd2devname(int fd, char *name)
1372 {
1373 struct stat st;
1374 char path[256];
1375 char dname[100];
1376 char *nm;
1377 int rv;
1378
1379 name[0] = '\0';
1380 if (fstat(fd, &st) != 0)
1381 return;
1382 sprintf(path, "/sys/dev/block/%d:%d",
1383 major(st.st_rdev), minor(st.st_rdev));
1384
1385 rv = readlink(path, dname, sizeof(dname));
1386 if (rv <= 0)
1387 return;
1388
1389 dname[rv] = '\0';
1390 nm = strrchr(dname, '/');
1391 nm++;
1392 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1393 }
1394
1395
1396 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1397
1398 static int imsm_read_serial(int fd, char *devname,
1399 __u8 serial[MAX_RAID_SERIAL_LEN])
1400 {
1401 unsigned char scsi_serial[255];
1402 int rv;
1403 int rsp_len;
1404 int len;
1405 char *c, *rsp_buf;
1406
1407 memset(scsi_serial, 0, sizeof(scsi_serial));
1408
1409 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1410
1411 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1412 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1413 fd2devname(fd, (char *) serial);
1414 return 0;
1415 }
1416
1417 if (rv != 0) {
1418 if (devname)
1419 fprintf(stderr,
1420 Name ": Failed to retrieve serial for %s\n",
1421 devname);
1422 return rv;
1423 }
1424
1425 /* trim leading whitespace */
1426 rsp_len = scsi_serial[3];
1427 if (!rsp_len) {
1428 if (devname)
1429 fprintf(stderr,
1430 Name ": Failed to retrieve serial for %s\n",
1431 devname);
1432 return 2;
1433 }
1434 rsp_buf = (char *) &scsi_serial[4];
1435 c = rsp_buf;
1436 while (isspace(*c))
1437 c++;
1438
1439 /* truncate len to the end of rsp_buf if necessary */
1440 if (c + MAX_RAID_SERIAL_LEN > rsp_buf + rsp_len)
1441 len = rsp_len - (c - rsp_buf);
1442 else
1443 len = MAX_RAID_SERIAL_LEN;
1444
1445 /* initialize the buffer and copy rsp_buf characters */
1446 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1447 memcpy(serial, c, len);
1448
1449 /* trim trailing whitespace starting with the last character copied */
1450 c = (char *) &serial[len - 1];
1451 while (isspace(*c) || *c == '\0')
1452 *c-- = '\0';
1453
1454 return 0;
1455 }
1456
1457 static int serialcmp(__u8 *s1, __u8 *s2)
1458 {
1459 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1460 }
1461
1462 static void serialcpy(__u8 *dest, __u8 *src)
1463 {
1464 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1465 }
1466
1467 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1468 {
1469 struct dl *dl;
1470
1471 for (dl = super->disks; dl; dl = dl->next)
1472 if (serialcmp(dl->serial, serial) == 0)
1473 break;
1474
1475 return dl;
1476 }
1477
1478 static int
1479 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1480 {
1481 struct dl *dl;
1482 struct stat stb;
1483 int rv;
1484 int i;
1485 int alloc = 1;
1486 __u8 serial[MAX_RAID_SERIAL_LEN];
1487
1488 rv = imsm_read_serial(fd, devname, serial);
1489
1490 if (rv != 0)
1491 return 2;
1492
1493 /* check if this is a disk we have seen before. it may be a spare in
1494 * super->disks while the current anchor believes it is a raid member,
1495 * check if we need to update dl->index
1496 */
1497 dl = serial_to_dl(serial, super);
1498 if (!dl)
1499 dl = malloc(sizeof(*dl));
1500 else
1501 alloc = 0;
1502
1503 if (!dl) {
1504 if (devname)
1505 fprintf(stderr,
1506 Name ": failed to allocate disk buffer for %s\n",
1507 devname);
1508 return 2;
1509 }
1510
1511 if (alloc) {
1512 fstat(fd, &stb);
1513 dl->major = major(stb.st_rdev);
1514 dl->minor = minor(stb.st_rdev);
1515 dl->next = super->disks;
1516 dl->fd = keep_fd ? fd : -1;
1517 dl->devname = devname ? strdup(devname) : NULL;
1518 serialcpy(dl->serial, serial);
1519 dl->index = -2;
1520 dl->e = NULL;
1521 } else if (keep_fd) {
1522 close(dl->fd);
1523 dl->fd = fd;
1524 }
1525
1526 /* look up this disk's index in the current anchor */
1527 for (i = 0; i < super->anchor->num_disks; i++) {
1528 struct imsm_disk *disk_iter;
1529
1530 disk_iter = __get_imsm_disk(super->anchor, i);
1531
1532 if (serialcmp(disk_iter->serial, dl->serial) == 0) {
1533 dl->disk = *disk_iter;
1534 /* only set index on disks that are a member of a
1535 * populated contianer, i.e. one with raid_devs
1536 */
1537 if (dl->disk.status & FAILED_DISK)
1538 dl->index = -2;
1539 else if (dl->disk.status & SPARE_DISK)
1540 dl->index = -1;
1541 else
1542 dl->index = i;
1543
1544 break;
1545 }
1546 }
1547
1548 /* no match, maybe a stale failed drive */
1549 if (i == super->anchor->num_disks && dl->index >= 0) {
1550 dl->disk = *__get_imsm_disk(super->anchor, dl->index);
1551 if (dl->disk.status & FAILED_DISK)
1552 dl->index = -2;
1553 }
1554
1555 if (alloc)
1556 super->disks = dl;
1557
1558 return 0;
1559 }
1560
1561 #ifndef MDASSEMBLE
1562 /* When migrating map0 contains the 'destination' state while map1
1563 * contains the current state. When not migrating map0 contains the
1564 * current state. This routine assumes that map[0].map_state is set to
1565 * the current array state before being called.
1566 *
1567 * Migration is indicated by one of the following states
1568 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1569 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1570 * map1state=unitialized)
1571 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1572 * map1state=normal)
1573 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1574 * map1state=degraded)
1575 */
1576 static void migrate(struct imsm_dev *dev, __u8 to_state, int rebuild_resync)
1577 {
1578 struct imsm_map *dest;
1579 struct imsm_map *src = get_imsm_map(dev, 0);
1580
1581 dev->vol.migr_state = 1;
1582 dev->vol.migr_type = rebuild_resync;
1583 dev->vol.curr_migr_unit = 0;
1584 dest = get_imsm_map(dev, 1);
1585
1586 memcpy(dest, src, sizeof_imsm_map(src));
1587 src->map_state = to_state;
1588 }
1589
1590 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1591 {
1592 struct imsm_map *map = get_imsm_map(dev, 0);
1593
1594 dev->vol.migr_state = 0;
1595 dev->vol.curr_migr_unit = 0;
1596 map->map_state = map_state;
1597 }
1598 #endif
1599
1600 static int parse_raid_devices(struct intel_super *super)
1601 {
1602 int i;
1603 struct imsm_dev *dev_new;
1604 size_t len, len_migr;
1605 size_t space_needed = 0;
1606 struct imsm_super *mpb = super->anchor;
1607
1608 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1609 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1610 struct intel_dev *dv;
1611
1612 len = sizeof_imsm_dev(dev_iter, 0);
1613 len_migr = sizeof_imsm_dev(dev_iter, 1);
1614 if (len_migr > len)
1615 space_needed += len_migr - len;
1616
1617 dv = malloc(sizeof(*dv));
1618 if (!dv)
1619 return 1;
1620 dev_new = malloc(len_migr);
1621 if (!dev_new) {
1622 free(dv);
1623 return 1;
1624 }
1625 imsm_copy_dev(dev_new, dev_iter);
1626 dv->dev = dev_new;
1627 dv->index = i;
1628 dv->next = super->devlist;
1629 super->devlist = dv;
1630 }
1631
1632 /* ensure that super->buf is large enough when all raid devices
1633 * are migrating
1634 */
1635 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1636 void *buf;
1637
1638 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1639 if (posix_memalign(&buf, 512, len) != 0)
1640 return 1;
1641
1642 memcpy(buf, super->buf, len);
1643 free(super->buf);
1644 super->buf = buf;
1645 super->len = len;
1646 }
1647
1648 return 0;
1649 }
1650
1651 /* retrieve a pointer to the bbm log which starts after all raid devices */
1652 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1653 {
1654 void *ptr = NULL;
1655
1656 if (__le32_to_cpu(mpb->bbm_log_size)) {
1657 ptr = mpb;
1658 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1659 }
1660
1661 return ptr;
1662 }
1663
1664 static void __free_imsm(struct intel_super *super, int free_disks);
1665
1666 /* load_imsm_mpb - read matrix metadata
1667 * allocates super->mpb to be freed by free_super
1668 */
1669 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1670 {
1671 unsigned long long dsize;
1672 unsigned long long sectors;
1673 struct stat;
1674 struct imsm_super *anchor;
1675 __u32 check_sum;
1676 int rc;
1677
1678 get_dev_size(fd, NULL, &dsize);
1679
1680 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1681 if (devname)
1682 fprintf(stderr,
1683 Name ": Cannot seek to anchor block on %s: %s\n",
1684 devname, strerror(errno));
1685 return 1;
1686 }
1687
1688 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1689 if (devname)
1690 fprintf(stderr,
1691 Name ": Failed to allocate imsm anchor buffer"
1692 " on %s\n", devname);
1693 return 1;
1694 }
1695 if (read(fd, anchor, 512) != 512) {
1696 if (devname)
1697 fprintf(stderr,
1698 Name ": Cannot read anchor block on %s: %s\n",
1699 devname, strerror(errno));
1700 free(anchor);
1701 return 1;
1702 }
1703
1704 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1705 if (devname)
1706 fprintf(stderr,
1707 Name ": no IMSM anchor on %s\n", devname);
1708 free(anchor);
1709 return 2;
1710 }
1711
1712 __free_imsm(super, 0);
1713 super->len = ROUND_UP(anchor->mpb_size, 512);
1714 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1715 if (devname)
1716 fprintf(stderr,
1717 Name ": unable to allocate %zu byte mpb buffer\n",
1718 super->len);
1719 free(anchor);
1720 return 2;
1721 }
1722 memcpy(super->buf, anchor, 512);
1723
1724 sectors = mpb_sectors(anchor) - 1;
1725 free(anchor);
1726 if (!sectors) {
1727 rc = load_imsm_disk(fd, super, devname, 0);
1728 if (rc == 0)
1729 rc = parse_raid_devices(super);
1730 return rc;
1731 }
1732
1733 /* read the extended mpb */
1734 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1735 if (devname)
1736 fprintf(stderr,
1737 Name ": Cannot seek to extended mpb on %s: %s\n",
1738 devname, strerror(errno));
1739 return 1;
1740 }
1741
1742 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1743 if (devname)
1744 fprintf(stderr,
1745 Name ": Cannot read extended mpb on %s: %s\n",
1746 devname, strerror(errno));
1747 return 2;
1748 }
1749
1750 check_sum = __gen_imsm_checksum(super->anchor);
1751 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1752 if (devname)
1753 fprintf(stderr,
1754 Name ": IMSM checksum %x != %x on %s\n",
1755 check_sum, __le32_to_cpu(super->anchor->check_sum),
1756 devname);
1757 return 2;
1758 }
1759
1760 /* FIXME the BBM log is disk specific so we cannot use this global
1761 * buffer for all disks. Ok for now since we only look at the global
1762 * bbm_log_size parameter to gate assembly
1763 */
1764 super->bbm_log = __get_imsm_bbm_log(super->anchor);
1765
1766 rc = load_imsm_disk(fd, super, devname, 0);
1767 if (rc == 0)
1768 rc = parse_raid_devices(super);
1769
1770 return rc;
1771 }
1772
1773 static void __free_imsm_disk(struct dl *d)
1774 {
1775 if (d->fd >= 0)
1776 close(d->fd);
1777 if (d->devname)
1778 free(d->devname);
1779 if (d->e)
1780 free(d->e);
1781 free(d);
1782
1783 }
1784 static void free_imsm_disks(struct intel_super *super)
1785 {
1786 struct dl *d;
1787
1788 while (super->disks) {
1789 d = super->disks;
1790 super->disks = d->next;
1791 __free_imsm_disk(d);
1792 }
1793 while (super->missing) {
1794 d = super->missing;
1795 super->missing = d->next;
1796 __free_imsm_disk(d);
1797 }
1798
1799 }
1800
1801 /* free all the pieces hanging off of a super pointer */
1802 static void __free_imsm(struct intel_super *super, int free_disks)
1803 {
1804 if (super->buf) {
1805 free(super->buf);
1806 super->buf = NULL;
1807 }
1808 if (free_disks)
1809 free_imsm_disks(super);
1810 free_devlist(super);
1811 if (super->hba) {
1812 free((void *) super->hba);
1813 super->hba = NULL;
1814 }
1815 }
1816
1817 static void free_imsm(struct intel_super *super)
1818 {
1819 __free_imsm(super, 1);
1820 free(super);
1821 }
1822
1823 static void free_super_imsm(struct supertype *st)
1824 {
1825 struct intel_super *super = st->sb;
1826
1827 if (!super)
1828 return;
1829
1830 free_imsm(super);
1831 st->sb = NULL;
1832 }
1833
1834 static struct intel_super *alloc_super(int creating_imsm)
1835 {
1836 struct intel_super *super = malloc(sizeof(*super));
1837
1838 if (super) {
1839 memset(super, 0, sizeof(*super));
1840 super->creating_imsm = creating_imsm;
1841 super->current_vol = -1;
1842 super->create_offset = ~((__u32 ) 0);
1843 if (!check_env("IMSM_NO_PLATFORM"))
1844 super->orom = find_imsm_orom();
1845 if (super->orom && !check_env("IMSM_TEST_OROM")) {
1846 struct sys_dev *list, *ent;
1847
1848 /* find the first intel ahci controller */
1849 list = find_driver_devices("pci", "ahci");
1850 for (ent = list; ent; ent = ent->next)
1851 if (devpath_to_vendor(ent->path) == 0x8086)
1852 break;
1853 if (ent) {
1854 super->hba = ent->path;
1855 ent->path = NULL;
1856 }
1857 free_sys_dev(&list);
1858 }
1859 }
1860
1861 return super;
1862 }
1863
1864 #ifndef MDASSEMBLE
1865 /* find_missing - helper routine for load_super_imsm_all that identifies
1866 * disks that have disappeared from the system. This routine relies on
1867 * the mpb being uptodate, which it is at load time.
1868 */
1869 static int find_missing(struct intel_super *super)
1870 {
1871 int i;
1872 struct imsm_super *mpb = super->anchor;
1873 struct dl *dl;
1874 struct imsm_disk *disk;
1875
1876 for (i = 0; i < mpb->num_disks; i++) {
1877 disk = __get_imsm_disk(mpb, i);
1878 dl = serial_to_dl(disk->serial, super);
1879 if (dl)
1880 continue;
1881 /* ok we have a 'disk' without a live entry in
1882 * super->disks
1883 */
1884 if (disk->status & FAILED_DISK || !(disk->status & USABLE_DISK))
1885 continue; /* never mind, already marked */
1886
1887 dl = malloc(sizeof(*dl));
1888 if (!dl)
1889 return 1;
1890 dl->major = 0;
1891 dl->minor = 0;
1892 dl->fd = -1;
1893 dl->devname = strdup("missing");
1894 dl->index = i;
1895 serialcpy(dl->serial, disk->serial);
1896 dl->disk = *disk;
1897 dl->next = super->missing;
1898 super->missing = dl;
1899 }
1900
1901 return 0;
1902 }
1903
1904 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
1905 char *devname, int keep_fd)
1906 {
1907 struct mdinfo *sra;
1908 struct intel_super *super;
1909 struct mdinfo *sd, *best = NULL;
1910 __u32 bestgen = 0;
1911 __u32 gen;
1912 char nm[20];
1913 int dfd;
1914 int rv;
1915
1916 /* check if this disk is a member of an active array */
1917 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
1918 if (!sra)
1919 return 1;
1920
1921 if (sra->array.major_version != -1 ||
1922 sra->array.minor_version != -2 ||
1923 strcmp(sra->text_version, "imsm") != 0)
1924 return 1;
1925
1926 super = alloc_super(0);
1927 if (!super)
1928 return 1;
1929
1930 /* find the most up to date disk in this array, skipping spares */
1931 for (sd = sra->devs; sd; sd = sd->next) {
1932 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
1933 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
1934 if (dfd < 0) {
1935 free_imsm(super);
1936 return 2;
1937 }
1938 rv = load_imsm_mpb(dfd, super, NULL);
1939 if (!keep_fd)
1940 close(dfd);
1941 if (rv == 0) {
1942 if (super->anchor->num_raid_devs == 0)
1943 gen = 0;
1944 else
1945 gen = __le32_to_cpu(super->anchor->generation_num);
1946 if (!best || gen > bestgen) {
1947 bestgen = gen;
1948 best = sd;
1949 }
1950 } else {
1951 free_imsm(super);
1952 return 2;
1953 }
1954 }
1955
1956 if (!best) {
1957 free_imsm(super);
1958 return 1;
1959 }
1960
1961 /* load the most up to date anchor */
1962 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
1963 dfd = dev_open(nm, O_RDONLY);
1964 if (dfd < 0) {
1965 free_imsm(super);
1966 return 1;
1967 }
1968 rv = load_imsm_mpb(dfd, super, NULL);
1969 close(dfd);
1970 if (rv != 0) {
1971 free_imsm(super);
1972 return 2;
1973 }
1974
1975 /* re-parse the disk list with the current anchor */
1976 for (sd = sra->devs ; sd ; sd = sd->next) {
1977 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
1978 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
1979 if (dfd < 0) {
1980 free_imsm(super);
1981 return 2;
1982 }
1983 load_imsm_disk(dfd, super, NULL, keep_fd);
1984 if (!keep_fd)
1985 close(dfd);
1986 }
1987
1988
1989 if (find_missing(super) != 0) {
1990 free_imsm(super);
1991 return 2;
1992 }
1993
1994 if (st->subarray[0]) {
1995 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
1996 super->current_vol = atoi(st->subarray);
1997 else
1998 return 1;
1999 }
2000
2001 *sbp = super;
2002 st->container_dev = fd2devnum(fd);
2003 if (st->ss == NULL) {
2004 st->ss = &super_imsm;
2005 st->minor_version = 0;
2006 st->max_devs = IMSM_MAX_DEVICES;
2007 }
2008 st->loaded_container = 1;
2009
2010 return 0;
2011 }
2012 #endif
2013
2014 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2015 {
2016 struct intel_super *super;
2017 int rv;
2018
2019 #ifndef MDASSEMBLE
2020 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2021 return 0;
2022 #endif
2023 if (st->subarray[0])
2024 return 1; /* FIXME */
2025
2026 super = alloc_super(0);
2027 if (!super) {
2028 fprintf(stderr,
2029 Name ": malloc of %zu failed.\n",
2030 sizeof(*super));
2031 return 1;
2032 }
2033
2034 rv = load_imsm_mpb(fd, super, devname);
2035
2036 if (rv) {
2037 if (devname)
2038 fprintf(stderr,
2039 Name ": Failed to load all information "
2040 "sections on %s\n", devname);
2041 free_imsm(super);
2042 return rv;
2043 }
2044
2045 st->sb = super;
2046 if (st->ss == NULL) {
2047 st->ss = &super_imsm;
2048 st->minor_version = 0;
2049 st->max_devs = IMSM_MAX_DEVICES;
2050 }
2051 st->loaded_container = 0;
2052
2053 return 0;
2054 }
2055
2056 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2057 {
2058 if (info->level == 1)
2059 return 128;
2060 return info->chunk_size >> 9;
2061 }
2062
2063 static __u32 info_to_num_data_stripes(mdu_array_info_t *info)
2064 {
2065 __u32 num_stripes;
2066
2067 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2068 if (info->level == 1)
2069 num_stripes /= 2;
2070
2071 return num_stripes;
2072 }
2073
2074 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2075 {
2076 if (info->level == 1)
2077 return info->size * 2;
2078 else
2079 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2080 }
2081
2082 static void imsm_update_version_info(struct intel_super *super)
2083 {
2084 /* update the version and attributes */
2085 struct imsm_super *mpb = super->anchor;
2086 char *version;
2087 struct imsm_dev *dev;
2088 struct imsm_map *map;
2089 int i;
2090
2091 for (i = 0; i < mpb->num_raid_devs; i++) {
2092 dev = get_imsm_dev(super, i);
2093 map = get_imsm_map(dev, 0);
2094 if (__le32_to_cpu(dev->size_high) > 0)
2095 mpb->attributes |= MPB_ATTRIB_2TB;
2096
2097 /* FIXME detect when an array spans a port multiplier */
2098 #if 0
2099 mpb->attributes |= MPB_ATTRIB_PM;
2100 #endif
2101
2102 if (mpb->num_raid_devs > 1 ||
2103 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2104 version = MPB_VERSION_ATTRIBS;
2105 switch (get_imsm_raid_level(map)) {
2106 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2107 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2108 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2109 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2110 }
2111 } else {
2112 if (map->num_members >= 5)
2113 version = MPB_VERSION_5OR6_DISK_ARRAY;
2114 else if (dev->status == DEV_CLONE_N_GO)
2115 version = MPB_VERSION_CNG;
2116 else if (get_imsm_raid_level(map) == 5)
2117 version = MPB_VERSION_RAID5;
2118 else if (map->num_members >= 3)
2119 version = MPB_VERSION_3OR4_DISK_ARRAY;
2120 else if (get_imsm_raid_level(map) == 1)
2121 version = MPB_VERSION_RAID1;
2122 else
2123 version = MPB_VERSION_RAID0;
2124 }
2125 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2126 }
2127 }
2128
2129 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2130 unsigned long long size, char *name,
2131 char *homehost, int *uuid)
2132 {
2133 /* We are creating a volume inside a pre-existing container.
2134 * so st->sb is already set.
2135 */
2136 struct intel_super *super = st->sb;
2137 struct imsm_super *mpb = super->anchor;
2138 struct intel_dev *dv;
2139 struct imsm_dev *dev;
2140 struct imsm_vol *vol;
2141 struct imsm_map *map;
2142 int idx = mpb->num_raid_devs;
2143 int i;
2144 unsigned long long array_blocks;
2145 size_t size_old, size_new;
2146
2147 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2148 fprintf(stderr, Name": This imsm-container already has the "
2149 "maximum of %d volumes\n", super->orom->vpa);
2150 return 0;
2151 }
2152
2153 /* ensure the mpb is large enough for the new data */
2154 size_old = __le32_to_cpu(mpb->mpb_size);
2155 size_new = disks_to_mpb_size(info->nr_disks);
2156 if (size_new > size_old) {
2157 void *mpb_new;
2158 size_t size_round = ROUND_UP(size_new, 512);
2159
2160 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2161 fprintf(stderr, Name": could not allocate new mpb\n");
2162 return 0;
2163 }
2164 memcpy(mpb_new, mpb, size_old);
2165 free(mpb);
2166 mpb = mpb_new;
2167 super->anchor = mpb_new;
2168 mpb->mpb_size = __cpu_to_le32(size_new);
2169 memset(mpb_new + size_old, 0, size_round - size_old);
2170 }
2171 super->current_vol = idx;
2172 /* when creating the first raid device in this container set num_disks
2173 * to zero, i.e. delete this spare and add raid member devices in
2174 * add_to_super_imsm_volume()
2175 */
2176 if (super->current_vol == 0)
2177 mpb->num_disks = 0;
2178
2179 for (i = 0; i < super->current_vol; i++) {
2180 dev = get_imsm_dev(super, i);
2181 if (strncmp((char *) dev->volume, name,
2182 MAX_RAID_SERIAL_LEN) == 0) {
2183 fprintf(stderr, Name": '%s' is already defined for this container\n",
2184 name);
2185 return 0;
2186 }
2187 }
2188
2189 sprintf(st->subarray, "%d", idx);
2190 dv = malloc(sizeof(*dv));
2191 if (!dv) {
2192 fprintf(stderr, Name ": failed to allocate device list entry\n");
2193 return 0;
2194 }
2195 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2196 if (!dev) {
2197 free(dv);
2198 fprintf(stderr, Name": could not allocate raid device\n");
2199 return 0;
2200 }
2201 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2202 if (info->level == 1)
2203 array_blocks = info_to_blocks_per_member(info);
2204 else
2205 array_blocks = calc_array_size(info->level, info->raid_disks,
2206 info->layout, info->chunk_size,
2207 info->size*2);
2208 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2209 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2210 dev->status = __cpu_to_le32(0);
2211 dev->reserved_blocks = __cpu_to_le32(0);
2212 vol = &dev->vol;
2213 vol->migr_state = 0;
2214 vol->migr_type = MIGR_INIT;
2215 vol->dirty = 0;
2216 vol->curr_migr_unit = 0;
2217 map = get_imsm_map(dev, 0);
2218 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2219 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2220 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2221 map->num_data_stripes = __cpu_to_le32(info_to_num_data_stripes(info));
2222 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2223 IMSM_T_STATE_NORMAL;
2224
2225 if (info->level == 1 && info->raid_disks > 2) {
2226 fprintf(stderr, Name": imsm does not support more than 2 disks"
2227 "in a raid1 volume\n");
2228 return 0;
2229 }
2230 if (info->level == 10) {
2231 map->raid_level = 1;
2232 map->num_domains = info->raid_disks / 2;
2233 } else {
2234 map->raid_level = info->level;
2235 map->num_domains = !!map->raid_level;
2236 }
2237
2238 map->num_members = info->raid_disks;
2239 for (i = 0; i < map->num_members; i++) {
2240 /* initialized in add_to_super */
2241 set_imsm_ord_tbl_ent(map, i, 0);
2242 }
2243 mpb->num_raid_devs++;
2244
2245 dv->dev = dev;
2246 dv->index = super->current_vol;
2247 dv->next = super->devlist;
2248 super->devlist = dv;
2249
2250 imsm_update_version_info(super);
2251
2252 return 1;
2253 }
2254
2255 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2256 unsigned long long size, char *name,
2257 char *homehost, int *uuid)
2258 {
2259 /* This is primarily called by Create when creating a new array.
2260 * We will then get add_to_super called for each component, and then
2261 * write_init_super called to write it out to each device.
2262 * For IMSM, Create can create on fresh devices or on a pre-existing
2263 * array.
2264 * To create on a pre-existing array a different method will be called.
2265 * This one is just for fresh drives.
2266 */
2267 struct intel_super *super;
2268 struct imsm_super *mpb;
2269 size_t mpb_size;
2270 char *version;
2271
2272 if (!info) {
2273 st->sb = NULL;
2274 return 0;
2275 }
2276 if (st->sb)
2277 return init_super_imsm_volume(st, info, size, name, homehost,
2278 uuid);
2279
2280 super = alloc_super(1);
2281 if (!super)
2282 return 0;
2283 mpb_size = disks_to_mpb_size(info->nr_disks);
2284 if (posix_memalign(&super->buf, 512, mpb_size) != 0) {
2285 free(super);
2286 return 0;
2287 }
2288 mpb = super->buf;
2289 memset(mpb, 0, mpb_size);
2290
2291 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2292
2293 version = (char *) mpb->sig;
2294 strcpy(version, MPB_SIGNATURE);
2295 version += strlen(MPB_SIGNATURE);
2296 strcpy(version, MPB_VERSION_RAID0);
2297 mpb->mpb_size = mpb_size;
2298
2299 st->sb = super;
2300 return 1;
2301 }
2302
2303 #ifndef MDASSEMBLE
2304 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2305 int fd, char *devname)
2306 {
2307 struct intel_super *super = st->sb;
2308 struct imsm_super *mpb = super->anchor;
2309 struct dl *dl;
2310 struct imsm_dev *dev;
2311 struct imsm_map *map;
2312
2313 dev = get_imsm_dev(super, super->current_vol);
2314 map = get_imsm_map(dev, 0);
2315
2316 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2317 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2318 devname);
2319 return 1;
2320 }
2321
2322 for (dl = super->disks; dl ; dl = dl->next)
2323 if (dl->major == dk->major &&
2324 dl->minor == dk->minor)
2325 break;
2326
2327 if (!dl) {
2328 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2329 return 1;
2330 }
2331
2332 /* add a pristine spare to the metadata */
2333 if (dl->index < 0) {
2334 dl->index = super->anchor->num_disks;
2335 super->anchor->num_disks++;
2336 }
2337 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2338 dl->disk.status = CONFIGURED_DISK | USABLE_DISK;
2339
2340 /* if we are creating the first raid device update the family number */
2341 if (super->current_vol == 0) {
2342 __u32 sum;
2343 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2344 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2345
2346 *_dev = *dev;
2347 *_disk = dl->disk;
2348 sum = __gen_imsm_checksum(mpb);
2349 mpb->family_num = __cpu_to_le32(sum);
2350 }
2351
2352 return 0;
2353 }
2354
2355 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2356 int fd, char *devname)
2357 {
2358 struct intel_super *super = st->sb;
2359 struct dl *dd;
2360 unsigned long long size;
2361 __u32 id;
2362 int rv;
2363 struct stat stb;
2364
2365 /* if we are on an RAID enabled platform check that the disk is
2366 * attached to the raid controller
2367 */
2368 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2369 fprintf(stderr,
2370 Name ": %s is not attached to the raid controller: %s\n",
2371 devname ? : "disk", super->hba);
2372 return 1;
2373 }
2374
2375 if (super->current_vol >= 0)
2376 return add_to_super_imsm_volume(st, dk, fd, devname);
2377
2378 fstat(fd, &stb);
2379 dd = malloc(sizeof(*dd));
2380 if (!dd) {
2381 fprintf(stderr,
2382 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2383 return 1;
2384 }
2385 memset(dd, 0, sizeof(*dd));
2386 dd->major = major(stb.st_rdev);
2387 dd->minor = minor(stb.st_rdev);
2388 dd->index = -1;
2389 dd->devname = devname ? strdup(devname) : NULL;
2390 dd->fd = fd;
2391 rv = imsm_read_serial(fd, devname, dd->serial);
2392 if (rv) {
2393 fprintf(stderr,
2394 Name ": failed to retrieve scsi serial, aborting\n");
2395 free(dd);
2396 abort();
2397 }
2398
2399 get_dev_size(fd, NULL, &size);
2400 size /= 512;
2401 serialcpy(dd->disk.serial, dd->serial);
2402 dd->disk.total_blocks = __cpu_to_le32(size);
2403 dd->disk.status = USABLE_DISK | SPARE_DISK;
2404 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
2405 dd->disk.scsi_id = __cpu_to_le32(id);
2406 else
2407 dd->disk.scsi_id = __cpu_to_le32(0);
2408
2409 if (st->update_tail) {
2410 dd->next = super->add;
2411 super->add = dd;
2412 } else {
2413 dd->next = super->disks;
2414 super->disks = dd;
2415 }
2416
2417 return 0;
2418 }
2419
2420 static int store_imsm_mpb(int fd, struct intel_super *super);
2421
2422 /* spare records have their own family number and do not have any defined raid
2423 * devices
2424 */
2425 static int write_super_imsm_spares(struct intel_super *super, int doclose)
2426 {
2427 struct imsm_super mpb_save;
2428 struct imsm_super *mpb = super->anchor;
2429 __u32 sum;
2430 struct dl *d;
2431
2432 mpb_save = *mpb;
2433 mpb->num_raid_devs = 0;
2434 mpb->num_disks = 1;
2435 mpb->mpb_size = sizeof(struct imsm_super);
2436 mpb->generation_num = __cpu_to_le32(1UL);
2437
2438 for (d = super->disks; d; d = d->next) {
2439 if (d->index != -1)
2440 continue;
2441
2442 mpb->disk[0] = d->disk;
2443 sum = __gen_imsm_checksum(mpb);
2444 mpb->family_num = __cpu_to_le32(sum);
2445 sum = __gen_imsm_checksum(mpb);
2446 mpb->check_sum = __cpu_to_le32(sum);
2447
2448 if (store_imsm_mpb(d->fd, super)) {
2449 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2450 __func__, d->major, d->minor, strerror(errno));
2451 *mpb = mpb_save;
2452 return 1;
2453 }
2454 if (doclose) {
2455 close(d->fd);
2456 d->fd = -1;
2457 }
2458 }
2459
2460 *mpb = mpb_save;
2461 return 0;
2462 }
2463
2464 static int write_super_imsm(struct intel_super *super, int doclose)
2465 {
2466 struct imsm_super *mpb = super->anchor;
2467 struct dl *d;
2468 __u32 generation;
2469 __u32 sum;
2470 int spares = 0;
2471 int i;
2472 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
2473
2474 /* 'generation' is incremented everytime the metadata is written */
2475 generation = __le32_to_cpu(mpb->generation_num);
2476 generation++;
2477 mpb->generation_num = __cpu_to_le32(generation);
2478
2479 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
2480 for (d = super->disks; d; d = d->next) {
2481 if (d->index == -1)
2482 spares++;
2483 else
2484 mpb->disk[d->index] = d->disk;
2485 }
2486 for (d = super->missing; d; d = d->next)
2487 mpb->disk[d->index] = d->disk;
2488
2489 for (i = 0; i < mpb->num_raid_devs; i++) {
2490 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
2491
2492 imsm_copy_dev(dev, get_imsm_dev(super, i));
2493 mpb_size += sizeof_imsm_dev(dev, 0);
2494 }
2495 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
2496 mpb->mpb_size = __cpu_to_le32(mpb_size);
2497
2498 /* recalculate checksum */
2499 sum = __gen_imsm_checksum(mpb);
2500 mpb->check_sum = __cpu_to_le32(sum);
2501
2502 /* write the mpb for disks that compose raid devices */
2503 for (d = super->disks; d ; d = d->next) {
2504 if (d->index < 0)
2505 continue;
2506 if (store_imsm_mpb(d->fd, super))
2507 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2508 __func__, d->major, d->minor, strerror(errno));
2509 if (doclose) {
2510 close(d->fd);
2511 d->fd = -1;
2512 }
2513 }
2514
2515 if (spares)
2516 return write_super_imsm_spares(super, doclose);
2517
2518 return 0;
2519 }
2520
2521
2522 static int create_array(struct supertype *st)
2523 {
2524 size_t len;
2525 struct imsm_update_create_array *u;
2526 struct intel_super *super = st->sb;
2527 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
2528 struct imsm_map *map = get_imsm_map(dev, 0);
2529 struct disk_info *inf;
2530 struct imsm_disk *disk;
2531 int i;
2532 int idx;
2533
2534 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
2535 sizeof(*inf) * map->num_members;
2536 u = malloc(len);
2537 if (!u) {
2538 fprintf(stderr, "%s: failed to allocate update buffer\n",
2539 __func__);
2540 return 1;
2541 }
2542
2543 u->type = update_create_array;
2544 u->dev_idx = super->current_vol;
2545 imsm_copy_dev(&u->dev, dev);
2546 inf = get_disk_info(u);
2547 for (i = 0; i < map->num_members; i++) {
2548 idx = get_imsm_disk_idx(dev, i);
2549 disk = get_imsm_disk(super, idx);
2550 serialcpy(inf[i].serial, disk->serial);
2551 }
2552 append_metadata_update(st, u, len);
2553
2554 return 0;
2555 }
2556
2557 static int _add_disk(struct supertype *st)
2558 {
2559 struct intel_super *super = st->sb;
2560 size_t len;
2561 struct imsm_update_add_disk *u;
2562
2563 if (!super->add)
2564 return 0;
2565
2566 len = sizeof(*u);
2567 u = malloc(len);
2568 if (!u) {
2569 fprintf(stderr, "%s: failed to allocate update buffer\n",
2570 __func__);
2571 return 1;
2572 }
2573
2574 u->type = update_add_disk;
2575 append_metadata_update(st, u, len);
2576
2577 return 0;
2578 }
2579
2580 static int write_init_super_imsm(struct supertype *st)
2581 {
2582 if (st->update_tail) {
2583 /* queue the recently created array / added disk
2584 * as a metadata update */
2585 struct intel_super *super = st->sb;
2586 struct dl *d;
2587 int rv;
2588
2589 /* determine if we are creating a volume or adding a disk */
2590 if (super->current_vol < 0) {
2591 /* in the add disk case we are running in mdmon
2592 * context, so don't close fd's
2593 */
2594 return _add_disk(st);
2595 } else
2596 rv = create_array(st);
2597
2598 for (d = super->disks; d ; d = d->next) {
2599 close(d->fd);
2600 d->fd = -1;
2601 }
2602
2603 return rv;
2604 } else
2605 return write_super_imsm(st->sb, 1);
2606 }
2607 #endif
2608
2609 static int store_zero_imsm(struct supertype *st, int fd)
2610 {
2611 unsigned long long dsize;
2612 void *buf;
2613
2614 get_dev_size(fd, NULL, &dsize);
2615
2616 /* first block is stored on second to last sector of the disk */
2617 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
2618 return 1;
2619
2620 if (posix_memalign(&buf, 512, 512) != 0)
2621 return 1;
2622
2623 memset(buf, 0, 512);
2624 if (write(fd, buf, 512) != 512)
2625 return 1;
2626 return 0;
2627 }
2628
2629 static int imsm_bbm_log_size(struct imsm_super *mpb)
2630 {
2631 return __le32_to_cpu(mpb->bbm_log_size);
2632 }
2633
2634 #ifndef MDASSEMBLE
2635 static int validate_geometry_imsm_container(struct supertype *st, int level,
2636 int layout, int raiddisks, int chunk,
2637 unsigned long long size, char *dev,
2638 unsigned long long *freesize,
2639 int verbose)
2640 {
2641 int fd;
2642 unsigned long long ldsize;
2643 const struct imsm_orom *orom;
2644
2645 if (level != LEVEL_CONTAINER)
2646 return 0;
2647 if (!dev)
2648 return 1;
2649
2650 if (check_env("IMSM_NO_PLATFORM"))
2651 orom = NULL;
2652 else
2653 orom = find_imsm_orom();
2654 if (orom && raiddisks > orom->tds) {
2655 if (verbose)
2656 fprintf(stderr, Name ": %d exceeds maximum number of"
2657 " platform supported disks: %d\n",
2658 raiddisks, orom->tds);
2659 return 0;
2660 }
2661
2662 fd = open(dev, O_RDONLY|O_EXCL, 0);
2663 if (fd < 0) {
2664 if (verbose)
2665 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
2666 dev, strerror(errno));
2667 return 0;
2668 }
2669 if (!get_dev_size(fd, dev, &ldsize)) {
2670 close(fd);
2671 return 0;
2672 }
2673 close(fd);
2674
2675 *freesize = avail_size_imsm(st, ldsize >> 9);
2676
2677 return 1;
2678 }
2679
2680 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
2681 {
2682 const unsigned long long base_start = e[*idx].start;
2683 unsigned long long end = base_start + e[*idx].size;
2684 int i;
2685
2686 if (base_start == end)
2687 return 0;
2688
2689 *idx = *idx + 1;
2690 for (i = *idx; i < num_extents; i++) {
2691 /* extend overlapping extents */
2692 if (e[i].start >= base_start &&
2693 e[i].start <= end) {
2694 if (e[i].size == 0)
2695 return 0;
2696 if (e[i].start + e[i].size > end)
2697 end = e[i].start + e[i].size;
2698 } else if (e[i].start > end) {
2699 *idx = i;
2700 break;
2701 }
2702 }
2703
2704 return end - base_start;
2705 }
2706
2707 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
2708 {
2709 /* build a composite disk with all known extents and generate a new
2710 * 'maxsize' given the "all disks in an array must share a common start
2711 * offset" constraint
2712 */
2713 struct extent *e = calloc(sum_extents, sizeof(*e));
2714 struct dl *dl;
2715 int i, j;
2716 int start_extent;
2717 unsigned long long pos;
2718 unsigned long long start;
2719 unsigned long long maxsize;
2720 unsigned long reserve;
2721
2722 if (!e)
2723 return ~0ULL; /* error */
2724
2725 /* coalesce and sort all extents. also, check to see if we need to
2726 * reserve space between member arrays
2727 */
2728 j = 0;
2729 for (dl = super->disks; dl; dl = dl->next) {
2730 if (!dl->e)
2731 continue;
2732 for (i = 0; i < dl->extent_cnt; i++)
2733 e[j++] = dl->e[i];
2734 }
2735 qsort(e, sum_extents, sizeof(*e), cmp_extent);
2736
2737 /* merge extents */
2738 i = 0;
2739 j = 0;
2740 while (i < sum_extents) {
2741 e[j].start = e[i].start;
2742 e[j].size = find_size(e, &i, sum_extents);
2743 j++;
2744 if (e[j-1].size == 0)
2745 break;
2746 }
2747
2748 pos = 0;
2749 maxsize = 0;
2750 start_extent = 0;
2751 i = 0;
2752 do {
2753 unsigned long long esize;
2754
2755 esize = e[i].start - pos;
2756 if (esize >= maxsize) {
2757 maxsize = esize;
2758 start = pos;
2759 start_extent = i;
2760 }
2761 pos = e[i].start + e[i].size;
2762 i++;
2763 } while (e[i-1].size);
2764 free(e);
2765
2766 if (start_extent > 0)
2767 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
2768 else
2769 reserve = 0;
2770
2771 if (maxsize < reserve)
2772 return ~0ULL;
2773
2774 super->create_offset = ~((__u32) 0);
2775 if (start + reserve > super->create_offset)
2776 return ~0ULL; /* start overflows create_offset */
2777 super->create_offset = start + reserve;
2778
2779 return maxsize - reserve;
2780 }
2781
2782 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
2783 {
2784 if (level < 0 || level == 6 || level == 4)
2785 return 0;
2786
2787 /* if we have an orom prevent invalid raid levels */
2788 if (orom)
2789 switch (level) {
2790 case 0: return imsm_orom_has_raid0(orom);
2791 case 1:
2792 if (raiddisks > 2)
2793 return imsm_orom_has_raid1e(orom);
2794 return imsm_orom_has_raid1(orom) && raiddisks == 2;
2795 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
2796 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
2797 }
2798 else
2799 return 1; /* not on an Intel RAID platform so anything goes */
2800
2801 return 0;
2802 }
2803
2804 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2805 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2806 * FIX ME add ahci details
2807 */
2808 static int validate_geometry_imsm_volume(struct supertype *st, int level,
2809 int layout, int raiddisks, int chunk,
2810 unsigned long long size, char *dev,
2811 unsigned long long *freesize,
2812 int verbose)
2813 {
2814 struct stat stb;
2815 struct intel_super *super = st->sb;
2816 struct imsm_super *mpb = super->anchor;
2817 struct dl *dl;
2818 unsigned long long pos = 0;
2819 unsigned long long maxsize;
2820 struct extent *e;
2821 int i;
2822
2823 /* We must have the container info already read in. */
2824 if (!super)
2825 return 0;
2826
2827 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
2828 pr_vrb(": platform does not support raid%d with %d disk%s\n",
2829 level, raiddisks, raiddisks > 1 ? "s" : "");
2830 return 0;
2831 }
2832 if (super->orom && level != 1 &&
2833 !imsm_orom_has_chunk(super->orom, chunk)) {
2834 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
2835 return 0;
2836 }
2837 if (layout != imsm_level_to_layout(level)) {
2838 if (level == 5)
2839 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
2840 else if (level == 10)
2841 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
2842 else
2843 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
2844 layout, level);
2845 return 0;
2846 }
2847
2848 if (!dev) {
2849 /* General test: make sure there is space for
2850 * 'raiddisks' device extents of size 'size' at a given
2851 * offset
2852 */
2853 unsigned long long minsize = size;
2854 unsigned long long start_offset = ~0ULL;
2855 int dcnt = 0;
2856 if (minsize == 0)
2857 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
2858 for (dl = super->disks; dl ; dl = dl->next) {
2859 int found = 0;
2860
2861 pos = 0;
2862 i = 0;
2863 e = get_extents(super, dl);
2864 if (!e) continue;
2865 do {
2866 unsigned long long esize;
2867 esize = e[i].start - pos;
2868 if (esize >= minsize)
2869 found = 1;
2870 if (found && start_offset == ~0ULL) {
2871 start_offset = pos;
2872 break;
2873 } else if (found && pos != start_offset) {
2874 found = 0;
2875 break;
2876 }
2877 pos = e[i].start + e[i].size;
2878 i++;
2879 } while (e[i-1].size);
2880 if (found)
2881 dcnt++;
2882 free(e);
2883 }
2884 if (dcnt < raiddisks) {
2885 if (verbose)
2886 fprintf(stderr, Name ": imsm: Not enough "
2887 "devices with space for this array "
2888 "(%d < %d)\n",
2889 dcnt, raiddisks);
2890 return 0;
2891 }
2892 return 1;
2893 }
2894
2895 /* This device must be a member of the set */
2896 if (stat(dev, &stb) < 0)
2897 return 0;
2898 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2899 return 0;
2900 for (dl = super->disks ; dl ; dl = dl->next) {
2901 if (dl->major == major(stb.st_rdev) &&
2902 dl->minor == minor(stb.st_rdev))
2903 break;
2904 }
2905 if (!dl) {
2906 if (verbose)
2907 fprintf(stderr, Name ": %s is not in the "
2908 "same imsm set\n", dev);
2909 return 0;
2910 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
2911 /* If a volume is present then the current creation attempt
2912 * cannot incorporate new spares because the orom may not
2913 * understand this configuration (all member disks must be
2914 * members of each array in the container).
2915 */
2916 fprintf(stderr, Name ": %s is a spare and a volume"
2917 " is already defined for this container\n", dev);
2918 fprintf(stderr, Name ": The option-rom requires all member"
2919 " disks to be a member of all volumes\n");
2920 return 0;
2921 }
2922
2923 /* retrieve the largest free space block */
2924 e = get_extents(super, dl);
2925 maxsize = 0;
2926 i = 0;
2927 if (e) {
2928 do {
2929 unsigned long long esize;
2930
2931 esize = e[i].start - pos;
2932 if (esize >= maxsize)
2933 maxsize = esize;
2934 pos = e[i].start + e[i].size;
2935 i++;
2936 } while (e[i-1].size);
2937 dl->e = e;
2938 dl->extent_cnt = i;
2939 } else {
2940 if (verbose)
2941 fprintf(stderr, Name ": unable to determine free space for: %s\n",
2942 dev);
2943 return 0;
2944 }
2945 if (maxsize < size) {
2946 if (verbose)
2947 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
2948 dev, maxsize, size);
2949 return 0;
2950 }
2951
2952 /* count total number of extents for merge */
2953 i = 0;
2954 for (dl = super->disks; dl; dl = dl->next)
2955 if (dl->e)
2956 i += dl->extent_cnt;
2957
2958 maxsize = merge_extents(super, i);
2959 if (maxsize < size) {
2960 if (verbose)
2961 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
2962 maxsize, size);
2963 return 0;
2964 } else if (maxsize == ~0ULL) {
2965 if (verbose)
2966 fprintf(stderr, Name ": failed to merge %d extents\n", i);
2967 return 0;
2968 }
2969
2970 *freesize = maxsize;
2971
2972 return 1;
2973 }
2974
2975 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
2976 int raiddisks, int chunk, unsigned long long size,
2977 char *dev, unsigned long long *freesize,
2978 int verbose)
2979 {
2980 int fd, cfd;
2981 struct mdinfo *sra;
2982
2983 /* if given unused devices create a container
2984 * if given given devices in a container create a member volume
2985 */
2986 if (level == LEVEL_CONTAINER) {
2987 /* Must be a fresh device to add to a container */
2988 return validate_geometry_imsm_container(st, level, layout,
2989 raiddisks, chunk, size,
2990 dev, freesize,
2991 verbose);
2992 }
2993
2994 if (!dev) {
2995 if (st->sb && freesize) {
2996 /* Should do auto-layout here */
2997 fprintf(stderr, Name ": IMSM does not support auto-layout yet\n");
2998 return 0;
2999 }
3000 return 1;
3001 }
3002 if (st->sb) {
3003 /* creating in a given container */
3004 return validate_geometry_imsm_volume(st, level, layout,
3005 raiddisks, chunk, size,
3006 dev, freesize, verbose);
3007 }
3008
3009 /* limit creation to the following levels */
3010 if (!dev)
3011 switch (level) {
3012 case 0:
3013 case 1:
3014 case 10:
3015 case 5:
3016 break;
3017 default:
3018 return 1;
3019 }
3020
3021 /* This device needs to be a device in an 'imsm' container */
3022 fd = open(dev, O_RDONLY|O_EXCL, 0);
3023 if (fd >= 0) {
3024 if (verbose)
3025 fprintf(stderr,
3026 Name ": Cannot create this array on device %s\n",
3027 dev);
3028 close(fd);
3029 return 0;
3030 }
3031 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3032 if (verbose)
3033 fprintf(stderr, Name ": Cannot open %s: %s\n",
3034 dev, strerror(errno));
3035 return 0;
3036 }
3037 /* Well, it is in use by someone, maybe an 'imsm' container. */
3038 cfd = open_container(fd);
3039 if (cfd < 0) {
3040 close(fd);
3041 if (verbose)
3042 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3043 dev);
3044 return 0;
3045 }
3046 sra = sysfs_read(cfd, 0, GET_VERSION);
3047 close(fd);
3048 if (sra && sra->array.major_version == -1 &&
3049 strcmp(sra->text_version, "imsm") == 0) {
3050 /* This is a member of a imsm container. Load the container
3051 * and try to create a volume
3052 */
3053 struct intel_super *super;
3054
3055 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3056 st->sb = super;
3057 st->container_dev = fd2devnum(cfd);
3058 close(cfd);
3059 return validate_geometry_imsm_volume(st, level, layout,
3060 raiddisks, chunk,
3061 size, dev,
3062 freesize, verbose);
3063 }
3064 close(cfd);
3065 } else /* may belong to another container */
3066 return 0;
3067
3068 return 1;
3069 }
3070 #endif /* MDASSEMBLE */
3071
3072 static struct mdinfo *container_content_imsm(struct supertype *st)
3073 {
3074 /* Given a container loaded by load_super_imsm_all,
3075 * extract information about all the arrays into
3076 * an mdinfo tree.
3077 *
3078 * For each imsm_dev create an mdinfo, fill it in,
3079 * then look for matching devices in super->disks
3080 * and create appropriate device mdinfo.
3081 */
3082 struct intel_super *super = st->sb;
3083 struct imsm_super *mpb = super->anchor;
3084 struct mdinfo *rest = NULL;
3085 int i;
3086
3087 /* do not assemble arrays that might have bad blocks */
3088 if (imsm_bbm_log_size(super->anchor)) {
3089 fprintf(stderr, Name ": BBM log found in metadata. "
3090 "Cannot activate array(s).\n");
3091 return NULL;
3092 }
3093
3094 for (i = 0; i < mpb->num_raid_devs; i++) {
3095 struct imsm_dev *dev = get_imsm_dev(super, i);
3096 struct imsm_map *map = get_imsm_map(dev, 0);
3097 struct mdinfo *this;
3098 int slot;
3099
3100 this = malloc(sizeof(*this));
3101 memset(this, 0, sizeof(*this));
3102 this->next = rest;
3103
3104 super->current_vol = i;
3105 getinfo_super_imsm_volume(st, this);
3106 for (slot = 0 ; slot < map->num_members; slot++) {
3107 struct mdinfo *info_d;
3108 struct dl *d;
3109 int idx;
3110 int skip;
3111 __u32 s;
3112 __u32 ord;
3113
3114 skip = 0;
3115 idx = get_imsm_disk_idx(dev, slot);
3116 ord = get_imsm_ord_tbl_ent(dev, slot);
3117 for (d = super->disks; d ; d = d->next)
3118 if (d->index == idx)
3119 break;
3120
3121 if (d == NULL)
3122 skip = 1;
3123
3124 s = d ? d->disk.status : 0;
3125 if (s & FAILED_DISK)
3126 skip = 1;
3127 if (!(s & USABLE_DISK))
3128 skip = 1;
3129 if (ord & IMSM_ORD_REBUILD)
3130 skip = 1;
3131
3132 /*
3133 * if we skip some disks the array will be assmebled degraded;
3134 * reset resync start to avoid a dirty-degraded situation
3135 *
3136 * FIXME handle dirty degraded
3137 */
3138 if (skip && !dev->vol.dirty)
3139 this->resync_start = ~0ULL;
3140 if (skip)
3141 continue;
3142
3143 info_d = malloc(sizeof(*info_d));
3144 if (!info_d) {
3145 fprintf(stderr, Name ": failed to allocate disk"
3146 " for volume %s\n", (char *) dev->volume);
3147 free(this);
3148 this = rest;
3149 break;
3150 }
3151 memset(info_d, 0, sizeof(*info_d));
3152 info_d->next = this->devs;
3153 this->devs = info_d;
3154
3155 info_d->disk.number = d->index;
3156 info_d->disk.major = d->major;
3157 info_d->disk.minor = d->minor;
3158 info_d->disk.raid_disk = slot;
3159
3160 this->array.working_disks++;
3161
3162 info_d->events = __le32_to_cpu(mpb->generation_num);
3163 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3164 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3165 if (d->devname)
3166 strcpy(info_d->name, d->devname);
3167 }
3168 rest = this;
3169 }
3170
3171 return rest;
3172 }
3173
3174
3175 #ifndef MDASSEMBLE
3176 static int imsm_open_new(struct supertype *c, struct active_array *a,
3177 char *inst)
3178 {
3179 struct intel_super *super = c->sb;
3180 struct imsm_super *mpb = super->anchor;
3181
3182 if (atoi(inst) >= mpb->num_raid_devs) {
3183 fprintf(stderr, "%s: subarry index %d, out of range\n",
3184 __func__, atoi(inst));
3185 return -ENODEV;
3186 }
3187
3188 dprintf("imsm: open_new %s\n", inst);
3189 a->info.container_member = atoi(inst);
3190 return 0;
3191 }
3192
3193 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3194 {
3195 struct imsm_map *map = get_imsm_map(dev, 0);
3196
3197 if (!failed)
3198 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3199 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3200
3201 switch (get_imsm_raid_level(map)) {
3202 case 0:
3203 return IMSM_T_STATE_FAILED;
3204 break;
3205 case 1:
3206 if (failed < map->num_members)
3207 return IMSM_T_STATE_DEGRADED;
3208 else
3209 return IMSM_T_STATE_FAILED;
3210 break;
3211 case 10:
3212 {
3213 /**
3214 * check to see if any mirrors have failed, otherwise we
3215 * are degraded. Even numbered slots are mirrored on
3216 * slot+1
3217 */
3218 int i;
3219 /* gcc -Os complains that this is unused */
3220 int insync = insync;
3221
3222 for (i = 0; i < map->num_members; i++) {
3223 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3224 int idx = ord_to_idx(ord);
3225 struct imsm_disk *disk;
3226
3227 /* reset the potential in-sync count on even-numbered
3228 * slots. num_copies is always 2 for imsm raid10
3229 */
3230 if ((i & 1) == 0)
3231 insync = 2;
3232
3233 disk = get_imsm_disk(super, idx);
3234 if (!disk || disk->status & FAILED_DISK ||
3235 ord & IMSM_ORD_REBUILD)
3236 insync--;
3237
3238 /* no in-sync disks left in this mirror the
3239 * array has failed
3240 */
3241 if (insync == 0)
3242 return IMSM_T_STATE_FAILED;
3243 }
3244
3245 return IMSM_T_STATE_DEGRADED;
3246 }
3247 case 5:
3248 if (failed < 2)
3249 return IMSM_T_STATE_DEGRADED;
3250 else
3251 return IMSM_T_STATE_FAILED;
3252 break;
3253 default:
3254 break;
3255 }
3256
3257 return map->map_state;
3258 }
3259
3260 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3261 {
3262 int i;
3263 int failed = 0;
3264 struct imsm_disk *disk;
3265 struct imsm_map *map = get_imsm_map(dev, 0);
3266
3267 for (i = 0; i < map->num_members; i++) {
3268 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3269 int idx = ord_to_idx(ord);
3270
3271 disk = get_imsm_disk(super, idx);
3272 if (!disk || disk->status & FAILED_DISK ||
3273 ord & IMSM_ORD_REBUILD)
3274 failed++;
3275 }
3276
3277 return failed;
3278 }
3279
3280 static int is_resyncing(struct imsm_dev *dev)
3281 {
3282 struct imsm_map *migr_map;
3283
3284 if (!dev->vol.migr_state)
3285 return 0;
3286
3287 if (dev->vol.migr_type == MIGR_INIT)
3288 return 1;
3289
3290 migr_map = get_imsm_map(dev, 1);
3291
3292 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
3293 return 1;
3294 else
3295 return 0;
3296 }
3297
3298 static int is_rebuilding(struct imsm_dev *dev)
3299 {
3300 struct imsm_map *migr_map;
3301
3302 if (!dev->vol.migr_state)
3303 return 0;
3304
3305 if (dev->vol.migr_type != MIGR_REBUILD)
3306 return 0;
3307
3308 migr_map = get_imsm_map(dev, 1);
3309
3310 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
3311 return 1;
3312 else
3313 return 0;
3314 }
3315
3316 static void mark_failure(struct imsm_disk *disk)
3317 {
3318 if (disk->status & FAILED_DISK)
3319 return;
3320 disk->status |= FAILED_DISK;
3321 disk->scsi_id = __cpu_to_le32(~(__u32)0);
3322 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
3323 }
3324
3325 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3326 * states are handled in imsm_set_disk() with one exception, when a
3327 * resync is stopped due to a new failure this routine will set the
3328 * 'degraded' state for the array.
3329 */
3330 static int imsm_set_array_state(struct active_array *a, int consistent)
3331 {
3332 int inst = a->info.container_member;
3333 struct intel_super *super = a->container->sb;
3334 struct imsm_dev *dev = get_imsm_dev(super, inst);
3335 struct imsm_map *map = get_imsm_map(dev, 0);
3336 int failed = imsm_count_failed(super, dev);
3337 __u8 map_state = imsm_check_degraded(super, dev, failed);
3338
3339 /* before we activate this array handle any missing disks */
3340 if (consistent == 2 && super->missing) {
3341 struct dl *dl;
3342
3343 dprintf("imsm: mark missing\n");
3344 end_migration(dev, map_state);
3345 for (dl = super->missing; dl; dl = dl->next)
3346 mark_failure(&dl->disk);
3347 super->updates_pending++;
3348 }
3349
3350 if (consistent == 2 &&
3351 (!is_resync_complete(a) ||
3352 map_state != IMSM_T_STATE_NORMAL ||
3353 dev->vol.migr_state))
3354 consistent = 0;
3355
3356 if (is_resync_complete(a)) {
3357 /* complete intialization / resync,
3358 * recovery is completed in ->set_disk
3359 */
3360 if (is_resyncing(dev)) {
3361 dprintf("imsm: mark resync done\n");
3362 end_migration(dev, map_state);
3363 super->updates_pending++;
3364 }
3365 } else if (!is_resyncing(dev) && !failed) {
3366 /* mark the start of the init process if nothing is failed */
3367 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
3368 if (map->map_state == IMSM_T_STATE_NORMAL)
3369 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REBUILD);
3370 else
3371 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
3372 super->updates_pending++;
3373 }
3374
3375 /* check if we can update the migration checkpoint */
3376 if (dev->vol.migr_state &&
3377 __le32_to_cpu(dev->vol.curr_migr_unit) != a->resync_start) {
3378 dprintf("imsm: checkpoint migration (%llu)\n", a->resync_start);
3379 dev->vol.curr_migr_unit = __cpu_to_le32(a->resync_start);
3380 super->updates_pending++;
3381 }
3382
3383 /* mark dirty / clean */
3384 if (dev->vol.dirty != !consistent) {
3385 dprintf("imsm: mark '%s' (%llu)\n",
3386 consistent ? "clean" : "dirty", a->resync_start);
3387 if (consistent)
3388 dev->vol.dirty = 0;
3389 else
3390 dev->vol.dirty = 1;
3391 super->updates_pending++;
3392 }
3393 return consistent;
3394 }
3395
3396 static void imsm_set_disk(struct active_array *a, int n, int state)
3397 {
3398 int inst = a->info.container_member;
3399 struct intel_super *super = a->container->sb;
3400 struct imsm_dev *dev = get_imsm_dev(super, inst);
3401 struct imsm_map *map = get_imsm_map(dev, 0);
3402 struct imsm_disk *disk;
3403 int failed;
3404 __u32 ord;
3405 __u8 map_state;
3406
3407 if (n > map->num_members)
3408 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
3409 n, map->num_members - 1);
3410
3411 if (n < 0)
3412 return;
3413
3414 dprintf("imsm: set_disk %d:%x\n", n, state);
3415
3416 ord = get_imsm_ord_tbl_ent(dev, n);
3417 disk = get_imsm_disk(super, ord_to_idx(ord));
3418
3419 /* check for new failures */
3420 if ((state & DS_FAULTY) && !(disk->status & FAILED_DISK)) {
3421 mark_failure(disk);
3422 super->updates_pending++;
3423 }
3424
3425 /* check if in_sync */
3426 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD) {
3427 struct imsm_map *migr_map = get_imsm_map(dev, 1);
3428
3429 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
3430 super->updates_pending++;
3431 }
3432
3433 failed = imsm_count_failed(super, dev);
3434 map_state = imsm_check_degraded(super, dev, failed);
3435
3436 /* check if recovery complete, newly degraded, or failed */
3437 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
3438 end_migration(dev, map_state);
3439 super->updates_pending++;
3440 } else if (map_state == IMSM_T_STATE_DEGRADED &&
3441 map->map_state != map_state &&
3442 !dev->vol.migr_state) {
3443 dprintf("imsm: mark degraded\n");
3444 map->map_state = map_state;
3445 super->updates_pending++;
3446 } else if (map_state == IMSM_T_STATE_FAILED &&
3447 map->map_state != map_state) {
3448 dprintf("imsm: mark failed\n");
3449 end_migration(dev, map_state);
3450 super->updates_pending++;
3451 }
3452 }
3453
3454 static int store_imsm_mpb(int fd, struct intel_super *super)
3455 {
3456 struct imsm_super *mpb = super->anchor;
3457 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
3458 unsigned long long dsize;
3459 unsigned long long sectors;
3460
3461 get_dev_size(fd, NULL, &dsize);
3462
3463 if (mpb_size > 512) {
3464 /* -1 to account for anchor */
3465 sectors = mpb_sectors(mpb) - 1;
3466
3467 /* write the extended mpb to the sectors preceeding the anchor */
3468 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
3469 return 1;
3470
3471 if (write(fd, super->buf + 512, 512 * sectors) != 512 * sectors)
3472 return 1;
3473 }
3474
3475 /* first block is stored on second to last sector of the disk */
3476 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
3477 return 1;
3478
3479 if (write(fd, super->buf, 512) != 512)
3480 return 1;
3481
3482 return 0;
3483 }
3484
3485 static void imsm_sync_metadata(struct supertype *container)
3486 {
3487 struct intel_super *super = container->sb;
3488
3489 if (!super->updates_pending)
3490 return;
3491
3492 write_super_imsm(super, 0);
3493
3494 super->updates_pending = 0;
3495 }
3496
3497 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
3498 {
3499 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3500 int i = get_imsm_disk_idx(dev, idx);
3501 struct dl *dl;
3502
3503 for (dl = super->disks; dl; dl = dl->next)
3504 if (dl->index == i)
3505 break;
3506
3507 if (dl && dl->disk.status & FAILED_DISK)
3508 dl = NULL;
3509
3510 if (dl)
3511 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
3512
3513 return dl;
3514 }
3515
3516 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
3517 struct active_array *a, int activate_new)
3518 {
3519 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3520 int idx = get_imsm_disk_idx(dev, slot);
3521 struct imsm_super *mpb = super->anchor;
3522 struct imsm_map *map;
3523 unsigned long long esize;
3524 unsigned long long pos;
3525 struct mdinfo *d;
3526 struct extent *ex;
3527 int i, j;
3528 int found;
3529 __u32 array_start;
3530 __u32 blocks;
3531 struct dl *dl;
3532
3533 for (dl = super->disks; dl; dl = dl->next) {
3534 /* If in this array, skip */
3535 for (d = a->info.devs ; d ; d = d->next)
3536 if (d->state_fd >= 0 &&
3537 d->disk.major == dl->major &&
3538 d->disk.minor == dl->minor) {
3539 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3540 break;
3541 }
3542 if (d)
3543 continue;
3544
3545 /* skip in use or failed drives */
3546 if (dl->disk.status & FAILED_DISK || idx == dl->index) {
3547 dprintf("%x:%x status ( %s%s)\n",
3548 dl->major, dl->minor,
3549 dl->disk.status & FAILED_DISK ? "failed " : "",
3550 idx == dl->index ? "in use " : "");
3551 continue;
3552 }
3553
3554 /* skip pure spares when we are looking for partially
3555 * assimilated drives
3556 */
3557 if (dl->index == -1 && !activate_new)
3558 continue;
3559
3560 /* Does this unused device have the requisite free space?
3561 * It needs to be able to cover all member volumes
3562 */
3563 ex = get_extents(super, dl);
3564 if (!ex) {
3565 dprintf("cannot get extents\n");
3566 continue;
3567 }
3568 for (i = 0; i < mpb->num_raid_devs; i++) {
3569 dev = get_imsm_dev(super, i);
3570 map = get_imsm_map(dev, 0);
3571
3572 /* check if this disk is already a member of
3573 * this array
3574 */
3575 for (j = 0; j < map->num_members; j++)
3576 if (get_imsm_disk_idx(dev, j) == dl->index)
3577 break;
3578 if (j < map->num_members)
3579 continue;
3580
3581 found = 0;
3582 j = 0;
3583 pos = 0;
3584 array_start = __le32_to_cpu(map->pba_of_lba0);
3585 blocks = __le32_to_cpu(map->blocks_per_member);
3586
3587 do {
3588 /* check that we can start at pba_of_lba0 with
3589 * blocks_per_member of space
3590 */
3591 esize = ex[j].start - pos;
3592 if (array_start >= pos &&
3593 array_start + blocks < ex[j].start) {
3594 found = 1;
3595 break;
3596 }
3597 pos = ex[j].start + ex[j].size;
3598 j++;
3599 } while (ex[j-1].size);
3600
3601 if (!found)
3602 break;
3603 }
3604
3605 free(ex);
3606 if (i < mpb->num_raid_devs) {
3607 dprintf("%x:%x does not have %u at %u\n",
3608 dl->major, dl->minor,
3609 blocks, array_start);
3610 /* No room */
3611 continue;
3612 }
3613 return dl;
3614 }
3615
3616 return dl;
3617 }
3618
3619 static struct mdinfo *imsm_activate_spare(struct active_array *a,
3620 struct metadata_update **updates)
3621 {
3622 /**
3623 * Find a device with unused free space and use it to replace a
3624 * failed/vacant region in an array. We replace failed regions one a
3625 * array at a time. The result is that a new spare disk will be added
3626 * to the first failed array and after the monitor has finished
3627 * propagating failures the remainder will be consumed.
3628 *
3629 * FIXME add a capability for mdmon to request spares from another
3630 * container.
3631 */
3632
3633 struct intel_super *super = a->container->sb;
3634 int inst = a->info.container_member;
3635 struct imsm_dev *dev = get_imsm_dev(super, inst);
3636 struct imsm_map *map = get_imsm_map(dev, 0);
3637 int failed = a->info.array.raid_disks;
3638 struct mdinfo *rv = NULL;
3639 struct mdinfo *d;
3640 struct mdinfo *di;
3641 struct metadata_update *mu;
3642 struct dl *dl;
3643 struct imsm_update_activate_spare *u;
3644 int num_spares = 0;
3645 int i;
3646
3647 for (d = a->info.devs ; d ; d = d->next) {
3648 if ((d->curr_state & DS_FAULTY) &&
3649 d->state_fd >= 0)
3650 /* wait for Removal to happen */
3651 return NULL;
3652 if (d->state_fd >= 0)
3653 failed--;
3654 }
3655
3656 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3657 inst, failed, a->info.array.raid_disks, a->info.array.level);
3658 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
3659 return NULL;
3660
3661 /* For each slot, if it is not working, find a spare */
3662 for (i = 0; i < a->info.array.raid_disks; i++) {
3663 for (d = a->info.devs ; d ; d = d->next)
3664 if (d->disk.raid_disk == i)
3665 break;
3666 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3667 if (d && (d->state_fd >= 0))
3668 continue;
3669
3670 /*
3671 * OK, this device needs recovery. Try to re-add the
3672 * previous occupant of this slot, if this fails see if
3673 * we can continue the assimilation of a spare that was
3674 * partially assimilated, finally try to activate a new
3675 * spare.
3676 */
3677 dl = imsm_readd(super, i, a);
3678 if (!dl)
3679 dl = imsm_add_spare(super, i, a, 0);
3680 if (!dl)
3681 dl = imsm_add_spare(super, i, a, 1);
3682 if (!dl)
3683 continue;
3684
3685 /* found a usable disk with enough space */
3686 di = malloc(sizeof(*di));
3687 if (!di)
3688 continue;
3689 memset(di, 0, sizeof(*di));
3690
3691 /* dl->index will be -1 in the case we are activating a
3692 * pristine spare. imsm_process_update() will create a
3693 * new index in this case. Once a disk is found to be
3694 * failed in all member arrays it is kicked from the
3695 * metadata
3696 */
3697 di->disk.number = dl->index;
3698
3699 /* (ab)use di->devs to store a pointer to the device
3700 * we chose
3701 */
3702 di->devs = (struct mdinfo *) dl;
3703
3704 di->disk.raid_disk = i;
3705 di->disk.major = dl->major;
3706 di->disk.minor = dl->minor;
3707 di->disk.state = 0;
3708 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
3709 di->component_size = a->info.component_size;
3710 di->container_member = inst;
3711 di->next = rv;
3712 rv = di;
3713 num_spares++;
3714 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3715 i, di->data_offset);
3716
3717 break;
3718 }
3719
3720 if (!rv)
3721 /* No spares found */
3722 return rv;
3723 /* Now 'rv' has a list of devices to return.
3724 * Create a metadata_update record to update the
3725 * disk_ord_tbl for the array
3726 */
3727 mu = malloc(sizeof(*mu));
3728 if (mu) {
3729 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
3730 if (mu->buf == NULL) {
3731 free(mu);
3732 mu = NULL;
3733 }
3734 }
3735 if (!mu) {
3736 while (rv) {
3737 struct mdinfo *n = rv->next;
3738
3739 free(rv);
3740 rv = n;
3741 }
3742 return NULL;
3743 }
3744
3745 mu->space = NULL;
3746 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
3747 mu->next = *updates;
3748 u = (struct imsm_update_activate_spare *) mu->buf;
3749
3750 for (di = rv ; di ; di = di->next) {
3751 u->type = update_activate_spare;
3752 u->dl = (struct dl *) di->devs;
3753 di->devs = NULL;
3754 u->slot = di->disk.raid_disk;
3755 u->array = inst;
3756 u->next = u + 1;
3757 u++;
3758 }
3759 (u-1)->next = NULL;
3760 *updates = mu;
3761
3762 return rv;
3763 }
3764
3765 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
3766 {
3767 struct imsm_dev *dev = get_imsm_dev(super, idx);
3768 struct imsm_map *map = get_imsm_map(dev, 0);
3769 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
3770 struct disk_info *inf = get_disk_info(u);
3771 struct imsm_disk *disk;
3772 int i;
3773 int j;
3774
3775 for (i = 0; i < map->num_members; i++) {
3776 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
3777 for (j = 0; j < new_map->num_members; j++)
3778 if (serialcmp(disk->serial, inf[j].serial) == 0)
3779 return 1;
3780 }
3781
3782 return 0;
3783 }
3784
3785 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
3786
3787 static void imsm_process_update(struct supertype *st,
3788 struct metadata_update *update)
3789 {
3790 /**
3791 * crack open the metadata_update envelope to find the update record
3792 * update can be one of:
3793 * update_activate_spare - a spare device has replaced a failed
3794 * device in an array, update the disk_ord_tbl. If this disk is
3795 * present in all member arrays then also clear the SPARE_DISK
3796 * flag
3797 */
3798 struct intel_super *super = st->sb;
3799 struct imsm_super *mpb;
3800 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
3801
3802 /* update requires a larger buf but the allocation failed */
3803 if (super->next_len && !super->next_buf) {
3804 super->next_len = 0;
3805 return;
3806 }
3807
3808 if (super->next_buf) {
3809 memcpy(super->next_buf, super->buf, super->len);
3810 free(super->buf);
3811 super->len = super->next_len;
3812 super->buf = super->next_buf;
3813
3814 super->next_len = 0;
3815 super->next_buf = NULL;
3816 }
3817
3818 mpb = super->anchor;
3819
3820 switch (type) {
3821 case update_activate_spare: {
3822 struct imsm_update_activate_spare *u = (void *) update->buf;
3823 struct imsm_dev *dev = get_imsm_dev(super, u->array);
3824 struct imsm_map *map = get_imsm_map(dev, 0);
3825 struct imsm_map *migr_map;
3826 struct active_array *a;
3827 struct imsm_disk *disk;
3828 __u8 to_state;
3829 struct dl *dl;
3830 unsigned int found;
3831 int failed;
3832 int victim = get_imsm_disk_idx(dev, u->slot);
3833 int i;
3834
3835 for (dl = super->disks; dl; dl = dl->next)
3836 if (dl == u->dl)
3837 break;
3838
3839 if (!dl) {
3840 fprintf(stderr, "error: imsm_activate_spare passed "
3841 "an unknown disk (index: %d)\n",
3842 u->dl->index);
3843 return;
3844 }
3845
3846 super->updates_pending++;
3847
3848 /* count failures (excluding rebuilds and the victim)
3849 * to determine map[0] state
3850 */
3851 failed = 0;
3852 for (i = 0; i < map->num_members; i++) {
3853 if (i == u->slot)
3854 continue;
3855 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
3856 if (!disk || disk->status & FAILED_DISK)
3857 failed++;
3858 }
3859
3860 /* adding a pristine spare, assign a new index */
3861 if (dl->index < 0) {
3862 dl->index = super->anchor->num_disks;
3863 super->anchor->num_disks++;
3864 }
3865 disk = &dl->disk;
3866 disk->status |= CONFIGURED_DISK;
3867 disk->status &= ~SPARE_DISK;
3868
3869 /* mark rebuild */
3870 to_state = imsm_check_degraded(super, dev, failed);
3871 map->map_state = IMSM_T_STATE_DEGRADED;
3872 migrate(dev, to_state, MIGR_REBUILD);
3873 migr_map = get_imsm_map(dev, 1);
3874 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
3875 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
3876
3877 /* count arrays using the victim in the metadata */
3878 found = 0;
3879 for (a = st->arrays; a ; a = a->next) {
3880 dev = get_imsm_dev(super, a->info.container_member);
3881 for (i = 0; i < map->num_members; i++)
3882 if (victim == get_imsm_disk_idx(dev, i))
3883 found++;
3884 }
3885
3886 /* delete the victim if it is no longer being
3887 * utilized anywhere
3888 */
3889 if (!found) {
3890 struct dl **dlp;
3891
3892 /* We know that 'manager' isn't touching anything,
3893 * so it is safe to delete
3894 */
3895 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
3896 if ((*dlp)->index == victim)
3897 break;
3898
3899 /* victim may be on the missing list */
3900 if (!*dlp)
3901 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
3902 if ((*dlp)->index == victim)
3903 break;
3904 imsm_delete(super, dlp, victim);
3905 }
3906 break;
3907 }
3908 case update_create_array: {
3909 /* someone wants to create a new array, we need to be aware of
3910 * a few races/collisions:
3911 * 1/ 'Create' called by two separate instances of mdadm
3912 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
3913 * devices that have since been assimilated via
3914 * activate_spare.
3915 * In the event this update can not be carried out mdadm will
3916 * (FIX ME) notice that its update did not take hold.
3917 */
3918 struct imsm_update_create_array *u = (void *) update->buf;
3919 struct intel_dev *dv;
3920 struct imsm_dev *dev;
3921 struct imsm_map *map, *new_map;
3922 unsigned long long start, end;
3923 unsigned long long new_start, new_end;
3924 int i;
3925 struct disk_info *inf;
3926 struct dl *dl;
3927
3928 /* handle racing creates: first come first serve */
3929 if (u->dev_idx < mpb->num_raid_devs) {
3930 dprintf("%s: subarray %d already defined\n",
3931 __func__, u->dev_idx);
3932 goto create_error;
3933 }
3934
3935 /* check update is next in sequence */
3936 if (u->dev_idx != mpb->num_raid_devs) {
3937 dprintf("%s: can not create array %d expected index %d\n",
3938 __func__, u->dev_idx, mpb->num_raid_devs);
3939 goto create_error;
3940 }
3941
3942 new_map = get_imsm_map(&u->dev, 0);
3943 new_start = __le32_to_cpu(new_map->pba_of_lba0);
3944 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
3945 inf = get_disk_info(u);
3946
3947 /* handle activate_spare versus create race:
3948 * check to make sure that overlapping arrays do not include
3949 * overalpping disks
3950 */
3951 for (i = 0; i < mpb->num_raid_devs; i++) {
3952 dev = get_imsm_dev(super, i);
3953 map = get_imsm_map(dev, 0);
3954 start = __le32_to_cpu(map->pba_of_lba0);
3955 end = start + __le32_to_cpu(map->blocks_per_member);
3956 if ((new_start >= start && new_start <= end) ||
3957 (start >= new_start && start <= new_end))
3958 /* overlap */;
3959 else
3960 continue;
3961
3962 if (disks_overlap(super, i, u)) {
3963 dprintf("%s: arrays overlap\n", __func__);
3964 goto create_error;
3965 }
3966 }
3967
3968 /* check that prepare update was successful */
3969 if (!update->space) {
3970 dprintf("%s: prepare update failed\n", __func__);
3971 goto create_error;
3972 }
3973
3974 /* check that all disks are still active before committing
3975 * changes. FIXME: could we instead handle this by creating a
3976 * degraded array? That's probably not what the user expects,
3977 * so better to drop this update on the floor.
3978 */
3979 for (i = 0; i < new_map->num_members; i++) {
3980 dl = serial_to_dl(inf[i].serial, super);
3981 if (!dl) {
3982 dprintf("%s: disk disappeared\n", __func__);
3983 goto create_error;
3984 }
3985 }
3986
3987 super->updates_pending++;
3988
3989 /* convert spares to members and fixup ord_tbl */
3990 for (i = 0; i < new_map->num_members; i++) {
3991 dl = serial_to_dl(inf[i].serial, super);
3992 if (dl->index == -1) {
3993 dl->index = mpb->num_disks;
3994 mpb->num_disks++;
3995 dl->disk.status |= CONFIGURED_DISK;
3996 dl->disk.status &= ~SPARE_DISK;
3997 }
3998 set_imsm_ord_tbl_ent(new_map, i, dl->index);
3999 }
4000
4001 dv = update->space;
4002 dev = dv->dev;
4003 update->space = NULL;
4004 imsm_copy_dev(dev, &u->dev);
4005 dv->index = u->dev_idx;
4006 dv->next = super->devlist;
4007 super->devlist = dv;
4008 mpb->num_raid_devs++;
4009
4010 imsm_update_version_info(super);
4011 break;
4012 create_error:
4013 /* mdmon knows how to release update->space, but not
4014 * ((struct intel_dev *) update->space)->dev
4015 */
4016 if (update->space) {
4017 dv = update->space;
4018 free(dv->dev);
4019 }
4020 break;
4021 }
4022 case update_add_disk:
4023
4024 /* we may be able to repair some arrays if disks are
4025 * being added */
4026 if (super->add) {
4027 struct active_array *a;
4028
4029 super->updates_pending++;
4030 for (a = st->arrays; a; a = a->next)
4031 a->check_degraded = 1;
4032 }
4033 /* add some spares to the metadata */
4034 while (super->add) {
4035 struct dl *al;
4036
4037 al = super->add;
4038 super->add = al->next;
4039 al->next = super->disks;
4040 super->disks = al;
4041 dprintf("%s: added %x:%x\n",
4042 __func__, al->major, al->minor);
4043 }
4044
4045 break;
4046 }
4047 }
4048
4049 static void imsm_prepare_update(struct supertype *st,
4050 struct metadata_update *update)
4051 {
4052 /**
4053 * Allocate space to hold new disk entries, raid-device entries or a new
4054 * mpb if necessary. The manager synchronously waits for updates to
4055 * complete in the monitor, so new mpb buffers allocated here can be
4056 * integrated by the monitor thread without worrying about live pointers
4057 * in the manager thread.
4058 */
4059 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4060 struct intel_super *super = st->sb;
4061 struct imsm_super *mpb = super->anchor;
4062 size_t buf_len;
4063 size_t len = 0;
4064
4065 switch (type) {
4066 case update_create_array: {
4067 struct imsm_update_create_array *u = (void *) update->buf;
4068 struct intel_dev *dv;
4069 struct imsm_dev *dev = &u->dev;
4070 struct imsm_map *map = get_imsm_map(dev, 0);
4071 struct dl *dl;
4072 struct disk_info *inf;
4073 int i;
4074 int activate = 0;
4075
4076 inf = get_disk_info(u);
4077 len = sizeof_imsm_dev(dev, 1);
4078 /* allocate a new super->devlist entry */
4079 dv = malloc(sizeof(*dv));
4080 if (dv) {
4081 dv->dev = malloc(len);
4082 if (dv->dev)
4083 update->space = dv;
4084 else {
4085 free(dv);
4086 update->space = NULL;
4087 }
4088 }
4089
4090 /* count how many spares will be converted to members */
4091 for (i = 0; i < map->num_members; i++) {
4092 dl = serial_to_dl(inf[i].serial, super);
4093 if (!dl) {
4094 /* hmm maybe it failed?, nothing we can do about
4095 * it here
4096 */
4097 continue;
4098 }
4099 if (count_memberships(dl, super) == 0)
4100 activate++;
4101 }
4102 len += activate * sizeof(struct imsm_disk);
4103 break;
4104 default:
4105 break;
4106 }
4107 }
4108
4109 /* check if we need a larger metadata buffer */
4110 if (super->next_buf)
4111 buf_len = super->next_len;
4112 else
4113 buf_len = super->len;
4114
4115 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4116 /* ok we need a larger buf than what is currently allocated
4117 * if this allocation fails process_update will notice that
4118 * ->next_len is set and ->next_buf is NULL
4119 */
4120 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4121 if (super->next_buf)
4122 free(super->next_buf);
4123
4124 super->next_len = buf_len;
4125 if (posix_memalign(&super->next_buf, 512, buf_len) != 0)
4126 super->next_buf = NULL;
4127 }
4128 }
4129
4130 /* must be called while manager is quiesced */
4131 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4132 {
4133 struct imsm_super *mpb = super->anchor;
4134 struct dl *iter;
4135 struct imsm_dev *dev;
4136 struct imsm_map *map;
4137 int i, j, num_members;
4138 __u32 ord;
4139
4140 dprintf("%s: deleting device[%d] from imsm_super\n",
4141 __func__, index);
4142
4143 /* shift all indexes down one */
4144 for (iter = super->disks; iter; iter = iter->next)
4145 if (iter->index > index)
4146 iter->index--;
4147 for (iter = super->missing; iter; iter = iter->next)
4148 if (iter->index > index)
4149 iter->index--;
4150
4151 for (i = 0; i < mpb->num_raid_devs; i++) {
4152 dev = get_imsm_dev(super, i);
4153 map = get_imsm_map(dev, 0);
4154 num_members = map->num_members;
4155 for (j = 0; j < num_members; j++) {
4156 /* update ord entries being careful not to propagate
4157 * ord-flags to the first map
4158 */
4159 ord = get_imsm_ord_tbl_ent(dev, j);
4160
4161 if (ord_to_idx(ord) <= index)
4162 continue;
4163
4164 map = get_imsm_map(dev, 0);
4165 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4166 map = get_imsm_map(dev, 1);
4167 if (map)
4168 set_imsm_ord_tbl_ent(map, j, ord - 1);
4169 }
4170 }
4171
4172 mpb->num_disks--;
4173 super->updates_pending++;
4174 if (*dlp) {
4175 struct dl *dl = *dlp;
4176
4177 *dlp = (*dlp)->next;
4178 __free_imsm_disk(dl);
4179 }
4180 }
4181 #endif /* MDASSEMBLE */
4182
4183 struct superswitch super_imsm = {
4184 #ifndef MDASSEMBLE
4185 .examine_super = examine_super_imsm,
4186 .brief_examine_super = brief_examine_super_imsm,
4187 .detail_super = detail_super_imsm,
4188 .brief_detail_super = brief_detail_super_imsm,
4189 .write_init_super = write_init_super_imsm,
4190 .validate_geometry = validate_geometry_imsm,
4191 .add_to_super = add_to_super_imsm,
4192 .detail_platform = detail_platform_imsm,
4193 #endif
4194 .match_home = match_home_imsm,
4195 .uuid_from_super= uuid_from_super_imsm,
4196 .getinfo_super = getinfo_super_imsm,
4197 .update_super = update_super_imsm,
4198
4199 .avail_size = avail_size_imsm,
4200
4201 .compare_super = compare_super_imsm,
4202
4203 .load_super = load_super_imsm,
4204 .init_super = init_super_imsm,
4205 .store_super = store_zero_imsm,
4206 .free_super = free_super_imsm,
4207 .match_metadata_desc = match_metadata_desc_imsm,
4208 .container_content = container_content_imsm,
4209 .default_layout = imsm_level_to_layout,
4210
4211 .external = 1,
4212 .name = "imsm",
4213
4214 #ifndef MDASSEMBLE
4215 /* for mdmon */
4216 .open_new = imsm_open_new,
4217 .load_super = load_super_imsm,
4218 .set_array_state= imsm_set_array_state,
4219 .set_disk = imsm_set_disk,
4220 .sync_metadata = imsm_sync_metadata,
4221 .activate_spare = imsm_activate_spare,
4222 .process_update = imsm_process_update,
4223 .prepare_update = imsm_prepare_update,
4224 #endif /* MDASSEMBLE */
4225 };
Unnamed repository; edit this file 'description' to name the repository.