]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-intel.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
imsm: fix spare record writeout race
[thirdparty/mdadm.git] / super-intel.c
1 /*
2 * mdadm - Intel(R) Matrix Storage Manager Support
3 *
4 * Copyright (C) 2002-2008 Intel Corporation
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define HAVE_STDINT_H 1
21 #include "mdadm.h"
22 #include "mdmon.h"
23 #include "sha1.h"
24 #include "platform-intel.h"
25 #include <values.h>
26 #include <scsi/sg.h>
27 #include <ctype.h>
28 #include <dirent.h>
29
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
43
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
53
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56 #define SECT_PER_MB_SHIFT 11
57
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
60 struct imsm_disk {
61 __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 __u32 status; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
71 };
72
73 /* RAID map configuration infos. */
74 struct imsm_map {
75 __u32 pba_of_lba0; /* start address of partition */
76 __u32 blocks_per_member;/* blocks per member */
77 __u32 num_data_stripes; /* number of data stripes */
78 __u16 blocks_per_strip;
79 __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
84 __u8 raid_level;
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members; /* number of member disks */
89 __u8 num_domains; /* number of parity domains */
90 __u8 failed_disk_num; /* valid only when state is degraded */
91 __u8 ddf;
92 __u32 filler[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
96 */
97 } __attribute__ ((packed));
98
99 struct imsm_vol {
100 __u32 curr_migr_unit;
101 __u32 checkpoint_id; /* id to access curr_migr_unit */
102 __u8 migr_state; /* Normal or Migrating */
103 #define MIGR_INIT 0
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type; /* Initializing, Rebuilding, ... */
110 __u8 dirty;
111 __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors; /* number of mismatches */
113 __u16 bad_blocks; /* number of bad blocks during verify */
114 __u32 filler[4];
115 struct imsm_map map[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed));
118
119 struct imsm_dev {
120 __u8 volume[MAX_RAID_SERIAL_LEN];
121 __u32 size_low;
122 __u32 size_high;
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status; /* Persistent RaidDev status */
137 __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
138 __u8 migr_priority;
139 __u8 num_sub_vols;
140 __u8 tid;
141 __u8 cng_master_disk;
142 __u16 cache_policy;
143 __u8 cng_state;
144 __u8 cng_sub_state;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler[IMSM_DEV_FILLERS];
147 struct imsm_vol vol;
148 } __attribute__ ((packed));
149
150 struct imsm_super {
151 __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
152 __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size; /* 0x30 - 0x33 in bytes */
157 __u32 attributes; /* 0x34 - 0x37 */
158 __u8 num_disks; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos; /* 0x3A */
161 __u8 fill[1]; /* 0x3B */
162 __u32 cache_size; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed));
172
173 #define BBM_LOG_MAX_ENTRIES 254
174
175 struct bbm_log_entry {
176 __u64 defective_block_start;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset;
179 __u16 remapped_marked_count;
180 __u16 disk_ordinal;
181 } __attribute__ ((__packed__));
182
183 struct bbm_log {
184 __u32 signature; /* 0xABADB10C */
185 __u32 entry_count;
186 __u32 reserved_spare_block_count; /* 0 */
187 __u32 reserved; /* 0xFFFF */
188 __u64 first_spare_lba;
189 struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
190 } __attribute__ ((__packed__));
191
192
193 #ifndef MDASSEMBLE
194 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
195 #endif
196
197 static __u8 migr_type(struct imsm_dev *dev)
198 {
199 if (dev->vol.migr_type == MIGR_VERIFY &&
200 dev->status & DEV_VERIFY_AND_FIX)
201 return MIGR_REPAIR;
202 else
203 return dev->vol.migr_type;
204 }
205
206 static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
207 {
208 /* for compatibility with older oroms convert MIGR_REPAIR, into
209 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
210 */
211 if (migr_type == MIGR_REPAIR) {
212 dev->vol.migr_type = MIGR_VERIFY;
213 dev->status |= DEV_VERIFY_AND_FIX;
214 } else {
215 dev->vol.migr_type = migr_type;
216 dev->status &= ~DEV_VERIFY_AND_FIX;
217 }
218 }
219
220 static unsigned int sector_count(__u32 bytes)
221 {
222 return ((bytes + (512-1)) & (~(512-1))) / 512;
223 }
224
225 static unsigned int mpb_sectors(struct imsm_super *mpb)
226 {
227 return sector_count(__le32_to_cpu(mpb->mpb_size));
228 }
229
230 struct intel_dev {
231 struct imsm_dev *dev;
232 struct intel_dev *next;
233 int index;
234 };
235
236 /* internal representation of IMSM metadata */
237 struct intel_super {
238 union {
239 void *buf; /* O_DIRECT buffer for reading/writing metadata */
240 struct imsm_super *anchor; /* immovable parameters */
241 };
242 size_t len; /* size of the 'buf' allocation */
243 void *next_buf; /* for realloc'ing buf from the manager */
244 size_t next_len;
245 int updates_pending; /* count of pending updates for mdmon */
246 int creating_imsm; /* flag to indicate container creation */
247 int current_vol; /* index of raid device undergoing creation */
248 __u32 create_offset; /* common start for 'current_vol' */
249 __u32 random; /* random data for seeding new family numbers */
250 struct intel_dev *devlist;
251 struct dl {
252 struct dl *next;
253 int index;
254 __u8 serial[MAX_RAID_SERIAL_LEN];
255 int major, minor;
256 char *devname;
257 struct imsm_disk disk;
258 int fd;
259 int extent_cnt;
260 struct extent *e; /* for determining freespace @ create */
261 int raiddisk; /* slot to fill in autolayout */
262 } *disks;
263 struct dl *add; /* list of disks to add while mdmon active */
264 struct dl *missing; /* disks removed while we weren't looking */
265 struct bbm_log *bbm_log;
266 const char *hba; /* device path of the raid controller for this metadata */
267 const struct imsm_orom *orom; /* platform firmware support */
268 struct intel_super *next; /* (temp) list for disambiguating family_num */
269 };
270
271 struct intel_disk {
272 struct imsm_disk disk;
273 #define IMSM_UNKNOWN_OWNER (-1)
274 int owner;
275 struct intel_disk *next;
276 };
277
278 struct extent {
279 unsigned long long start, size;
280 };
281
282 /* definition of messages passed to imsm_process_update */
283 enum imsm_update_type {
284 update_activate_spare,
285 update_create_array,
286 update_add_disk,
287 };
288
289 struct imsm_update_activate_spare {
290 enum imsm_update_type type;
291 struct dl *dl;
292 int slot;
293 int array;
294 struct imsm_update_activate_spare *next;
295 };
296
297 struct disk_info {
298 __u8 serial[MAX_RAID_SERIAL_LEN];
299 };
300
301 struct imsm_update_create_array {
302 enum imsm_update_type type;
303 int dev_idx;
304 struct imsm_dev dev;
305 };
306
307 struct imsm_update_add_disk {
308 enum imsm_update_type type;
309 };
310
311 static struct supertype *match_metadata_desc_imsm(char *arg)
312 {
313 struct supertype *st;
314
315 if (strcmp(arg, "imsm") != 0 &&
316 strcmp(arg, "default") != 0
317 )
318 return NULL;
319
320 st = malloc(sizeof(*st));
321 memset(st, 0, sizeof(*st));
322 st->ss = &super_imsm;
323 st->max_devs = IMSM_MAX_DEVICES;
324 st->minor_version = 0;
325 st->sb = NULL;
326 return st;
327 }
328
329 #ifndef MDASSEMBLE
330 static __u8 *get_imsm_version(struct imsm_super *mpb)
331 {
332 return &mpb->sig[MPB_SIG_LEN];
333 }
334 #endif
335
336 /* retrieve a disk directly from the anchor when the anchor is known to be
337 * up-to-date, currently only at load time
338 */
339 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
340 {
341 if (index >= mpb->num_disks)
342 return NULL;
343 return &mpb->disk[index];
344 }
345
346 #ifndef MDASSEMBLE
347 /* retrieve a disk from the parsed metadata */
348 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
349 {
350 struct dl *d;
351
352 for (d = super->disks; d; d = d->next)
353 if (d->index == index)
354 return &d->disk;
355
356 return NULL;
357 }
358 #endif
359
360 /* generate a checksum directly from the anchor when the anchor is known to be
361 * up-to-date, currently only at load or write_super after coalescing
362 */
363 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
364 {
365 __u32 end = mpb->mpb_size / sizeof(end);
366 __u32 *p = (__u32 *) mpb;
367 __u32 sum = 0;
368
369 while (end--) {
370 sum += __le32_to_cpu(*p);
371 p++;
372 }
373
374 return sum - __le32_to_cpu(mpb->check_sum);
375 }
376
377 static size_t sizeof_imsm_map(struct imsm_map *map)
378 {
379 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
380 }
381
382 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
383 {
384 struct imsm_map *map = &dev->vol.map[0];
385
386 if (second_map && !dev->vol.migr_state)
387 return NULL;
388 else if (second_map) {
389 void *ptr = map;
390
391 return ptr + sizeof_imsm_map(map);
392 } else
393 return map;
394
395 }
396
397 /* return the size of the device.
398 * migr_state increases the returned size if map[0] were to be duplicated
399 */
400 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
401 {
402 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
403 sizeof_imsm_map(get_imsm_map(dev, 0));
404
405 /* migrating means an additional map */
406 if (dev->vol.migr_state)
407 size += sizeof_imsm_map(get_imsm_map(dev, 1));
408 else if (migr_state)
409 size += sizeof_imsm_map(get_imsm_map(dev, 0));
410
411 return size;
412 }
413
414 #ifndef MDASSEMBLE
415 /* retrieve disk serial number list from a metadata update */
416 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
417 {
418 void *u = update;
419 struct disk_info *inf;
420
421 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
422 sizeof_imsm_dev(&update->dev, 0);
423
424 return inf;
425 }
426 #endif
427
428 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
429 {
430 int offset;
431 int i;
432 void *_mpb = mpb;
433
434 if (index >= mpb->num_raid_devs)
435 return NULL;
436
437 /* devices start after all disks */
438 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
439
440 for (i = 0; i <= index; i++)
441 if (i == index)
442 return _mpb + offset;
443 else
444 offset += sizeof_imsm_dev(_mpb + offset, 0);
445
446 return NULL;
447 }
448
449 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
450 {
451 struct intel_dev *dv;
452
453 if (index >= super->anchor->num_raid_devs)
454 return NULL;
455 for (dv = super->devlist; dv; dv = dv->next)
456 if (dv->index == index)
457 return dv->dev;
458 return NULL;
459 }
460
461 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, int slot)
462 {
463 struct imsm_map *map;
464
465 if (dev->vol.migr_state)
466 map = get_imsm_map(dev, 1);
467 else
468 map = get_imsm_map(dev, 0);
469
470 /* top byte identifies disk under rebuild */
471 return __le32_to_cpu(map->disk_ord_tbl[slot]);
472 }
473
474 #define ord_to_idx(ord) (((ord) << 8) >> 8)
475 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot)
476 {
477 __u32 ord = get_imsm_ord_tbl_ent(dev, slot);
478
479 return ord_to_idx(ord);
480 }
481
482 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
483 {
484 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
485 }
486
487 static int get_imsm_disk_slot(struct imsm_map *map, int idx)
488 {
489 int slot;
490 __u32 ord;
491
492 for (slot = 0; slot < map->num_members; slot++) {
493 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
494 if (ord_to_idx(ord) == idx)
495 return slot;
496 }
497
498 return -1;
499 }
500
501 static int get_imsm_raid_level(struct imsm_map *map)
502 {
503 if (map->raid_level == 1) {
504 if (map->num_members == 2)
505 return 1;
506 else
507 return 10;
508 }
509
510 return map->raid_level;
511 }
512
513 static int cmp_extent(const void *av, const void *bv)
514 {
515 const struct extent *a = av;
516 const struct extent *b = bv;
517 if (a->start < b->start)
518 return -1;
519 if (a->start > b->start)
520 return 1;
521 return 0;
522 }
523
524 static int count_memberships(struct dl *dl, struct intel_super *super)
525 {
526 int memberships = 0;
527 int i;
528
529 for (i = 0; i < super->anchor->num_raid_devs; i++) {
530 struct imsm_dev *dev = get_imsm_dev(super, i);
531 struct imsm_map *map = get_imsm_map(dev, 0);
532
533 if (get_imsm_disk_slot(map, dl->index) >= 0)
534 memberships++;
535 }
536
537 return memberships;
538 }
539
540 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
541 {
542 /* find a list of used extents on the given physical device */
543 struct extent *rv, *e;
544 int i;
545 int memberships = count_memberships(dl, super);
546 __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
547
548 rv = malloc(sizeof(struct extent) * (memberships + 1));
549 if (!rv)
550 return NULL;
551 e = rv;
552
553 for (i = 0; i < super->anchor->num_raid_devs; i++) {
554 struct imsm_dev *dev = get_imsm_dev(super, i);
555 struct imsm_map *map = get_imsm_map(dev, 0);
556
557 if (get_imsm_disk_slot(map, dl->index) >= 0) {
558 e->start = __le32_to_cpu(map->pba_of_lba0);
559 e->size = __le32_to_cpu(map->blocks_per_member);
560 e++;
561 }
562 }
563 qsort(rv, memberships, sizeof(*rv), cmp_extent);
564
565 /* determine the start of the metadata
566 * when no raid devices are defined use the default
567 * ...otherwise allow the metadata to truncate the value
568 * as is the case with older versions of imsm
569 */
570 if (memberships) {
571 struct extent *last = &rv[memberships - 1];
572 __u32 remainder;
573
574 remainder = __le32_to_cpu(dl->disk.total_blocks) -
575 (last->start + last->size);
576 /* round down to 1k block to satisfy precision of the kernel
577 * 'size' interface
578 */
579 remainder &= ~1UL;
580 /* make sure remainder is still sane */
581 if (remainder < ROUND_UP(super->len, 512) >> 9)
582 remainder = ROUND_UP(super->len, 512) >> 9;
583 if (reservation > remainder)
584 reservation = remainder;
585 }
586 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
587 e->size = 0;
588 return rv;
589 }
590
591 /* try to determine how much space is reserved for metadata from
592 * the last get_extents() entry, otherwise fallback to the
593 * default
594 */
595 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
596 {
597 struct extent *e;
598 int i;
599 __u32 rv;
600
601 /* for spares just return a minimal reservation which will grow
602 * once the spare is picked up by an array
603 */
604 if (dl->index == -1)
605 return MPB_SECTOR_CNT;
606
607 e = get_extents(super, dl);
608 if (!e)
609 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
610
611 /* scroll to last entry */
612 for (i = 0; e[i].size; i++)
613 continue;
614
615 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
616
617 free(e);
618
619 return rv;
620 }
621
622 #ifndef MDASSEMBLE
623 static int is_spare(struct imsm_disk *disk)
624 {
625 return (disk->status & SPARE_DISK) == SPARE_DISK;
626 }
627
628 static int is_configured(struct imsm_disk *disk)
629 {
630 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
631 }
632
633 static int is_failed(struct imsm_disk *disk)
634 {
635 return (disk->status & FAILED_DISK) == FAILED_DISK;
636 }
637
638 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
639 {
640 __u64 sz;
641 int slot;
642 struct imsm_map *map = get_imsm_map(dev, 0);
643 __u32 ord;
644
645 printf("\n");
646 printf("[%.16s]:\n", dev->volume);
647 printf(" UUID : %s\n", uuid);
648 printf(" RAID Level : %d\n", get_imsm_raid_level(map));
649 printf(" Members : %d\n", map->num_members);
650 slot = get_imsm_disk_slot(map, disk_idx);
651 if (slot >= 0) {
652 ord = get_imsm_ord_tbl_ent(dev, slot);
653 printf(" This Slot : %d%s\n", slot,
654 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
655 } else
656 printf(" This Slot : ?\n");
657 sz = __le32_to_cpu(dev->size_high);
658 sz <<= 32;
659 sz += __le32_to_cpu(dev->size_low);
660 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
661 human_size(sz * 512));
662 sz = __le32_to_cpu(map->blocks_per_member);
663 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
664 human_size(sz * 512));
665 printf(" Sector Offset : %u\n",
666 __le32_to_cpu(map->pba_of_lba0));
667 printf(" Num Stripes : %u\n",
668 __le32_to_cpu(map->num_data_stripes));
669 printf(" Chunk Size : %u KiB\n",
670 __le16_to_cpu(map->blocks_per_strip) / 2);
671 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
672 printf(" Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle\n");
673 if (dev->vol.migr_state) {
674 if (migr_type(dev) == MIGR_INIT)
675 printf(": initializing\n");
676 else if (migr_type(dev) == MIGR_REBUILD)
677 printf(": rebuilding\n");
678 else if (migr_type(dev) == MIGR_VERIFY)
679 printf(": check\n");
680 else if (migr_type(dev) == MIGR_GEN_MIGR)
681 printf(": general migration\n");
682 else if (migr_type(dev) == MIGR_STATE_CHANGE)
683 printf(": state change\n");
684 else if (migr_type(dev) == MIGR_REPAIR)
685 printf(": repair\n");
686 else
687 printf(": <unknown:%d>\n", migr_type(dev));
688 }
689 printf(" Map State : %s", map_state_str[map->map_state]);
690 if (dev->vol.migr_state) {
691 struct imsm_map *map = get_imsm_map(dev, 1);
692 printf(" <-- %s", map_state_str[map->map_state]);
693 }
694 printf("\n");
695 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
696 }
697
698 static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved)
699 {
700 struct imsm_disk *disk = __get_imsm_disk(mpb, index);
701 char str[MAX_RAID_SERIAL_LEN + 1];
702 __u64 sz;
703
704 if (index < 0)
705 return;
706
707 printf("\n");
708 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
709 printf(" Disk%02d Serial : %s\n", index, str);
710 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
711 is_configured(disk) ? " active" : "",
712 is_failed(disk) ? " failed" : "");
713 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
714 sz = __le32_to_cpu(disk->total_blocks) - reserved;
715 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
716 human_size(sz * 512));
717 }
718
719 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info);
720
721 static void examine_super_imsm(struct supertype *st, char *homehost)
722 {
723 struct intel_super *super = st->sb;
724 struct imsm_super *mpb = super->anchor;
725 char str[MAX_SIGNATURE_LENGTH];
726 int i;
727 struct mdinfo info;
728 char nbuf[64];
729 __u32 sum;
730 __u32 reserved = imsm_reserved_sectors(super, super->disks);
731
732
733 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
734 printf(" Magic : %s\n", str);
735 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
736 printf(" Version : %s\n", get_imsm_version(mpb));
737 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
738 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
739 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
740 getinfo_super_imsm(st, &info);
741 fname_from_uuid(st, &info, nbuf, ':');
742 printf(" UUID : %s\n", nbuf + 5);
743 sum = __le32_to_cpu(mpb->check_sum);
744 printf(" Checksum : %08x %s\n", sum,
745 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
746 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
747 printf(" Disks : %d\n", mpb->num_disks);
748 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
749 print_imsm_disk(mpb, super->disks->index, reserved);
750 if (super->bbm_log) {
751 struct bbm_log *log = super->bbm_log;
752
753 printf("\n");
754 printf("Bad Block Management Log:\n");
755 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
756 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
757 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
758 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
759 printf(" First Spare : %llx\n",
760 (unsigned long long) __le64_to_cpu(log->first_spare_lba));
761 }
762 for (i = 0; i < mpb->num_raid_devs; i++) {
763 struct mdinfo info;
764 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
765
766 super->current_vol = i;
767 getinfo_super_imsm(st, &info);
768 fname_from_uuid(st, &info, nbuf, ':');
769 print_imsm_dev(dev, nbuf + 5, super->disks->index);
770 }
771 for (i = 0; i < mpb->num_disks; i++) {
772 if (i == super->disks->index)
773 continue;
774 print_imsm_disk(mpb, i, reserved);
775 }
776 }
777
778 static void brief_examine_super_imsm(struct supertype *st, int verbose)
779 {
780 /* We just write a generic IMSM ARRAY entry */
781 struct mdinfo info;
782 char nbuf[64];
783 struct intel_super *super = st->sb;
784
785 if (!super->anchor->num_raid_devs) {
786 printf("ARRAY metadata=imsm\n");
787 return;
788 }
789
790 getinfo_super_imsm(st, &info);
791 fname_from_uuid(st, &info, nbuf, ':');
792 printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
793 }
794
795 static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
796 {
797 /* We just write a generic IMSM ARRAY entry */
798 struct mdinfo info;
799 char nbuf[64];
800 char nbuf1[64];
801 struct intel_super *super = st->sb;
802 int i;
803
804 if (!super->anchor->num_raid_devs)
805 return;
806
807 getinfo_super_imsm(st, &info);
808 fname_from_uuid(st, &info, nbuf, ':');
809 for (i = 0; i < super->anchor->num_raid_devs; i++) {
810 struct imsm_dev *dev = get_imsm_dev(super, i);
811
812 super->current_vol = i;
813 getinfo_super_imsm(st, &info);
814 fname_from_uuid(st, &info, nbuf1, ':');
815 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
816 dev->volume, nbuf + 5, i, nbuf1 + 5);
817 }
818 }
819
820 static void export_examine_super_imsm(struct supertype *st)
821 {
822 struct intel_super *super = st->sb;
823 struct imsm_super *mpb = super->anchor;
824 struct mdinfo info;
825 char nbuf[64];
826
827 getinfo_super_imsm(st, &info);
828 fname_from_uuid(st, &info, nbuf, ':');
829 printf("MD_METADATA=imsm\n");
830 printf("MD_LEVEL=container\n");
831 printf("MD_UUID=%s\n", nbuf+5);
832 printf("MD_DEVICES=%u\n", mpb->num_disks);
833 }
834
835 static void detail_super_imsm(struct supertype *st, char *homehost)
836 {
837 struct mdinfo info;
838 char nbuf[64];
839
840 getinfo_super_imsm(st, &info);
841 fname_from_uuid(st, &info, nbuf, ':');
842 printf("\n UUID : %s\n", nbuf + 5);
843 }
844
845 static void brief_detail_super_imsm(struct supertype *st)
846 {
847 struct mdinfo info;
848 char nbuf[64];
849 getinfo_super_imsm(st, &info);
850 fname_from_uuid(st, &info, nbuf, ':');
851 printf(" UUID=%s", nbuf + 5);
852 }
853
854 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
855 static void fd2devname(int fd, char *name);
856
857 static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
858 {
859 /* dump an unsorted list of devices attached to ahci, as well as
860 * non-connected ports
861 */
862 int hba_len = strlen(hba_path) + 1;
863 struct dirent *ent;
864 DIR *dir;
865 char *path = NULL;
866 int err = 0;
867 unsigned long port_mask = (1 << port_count) - 1;
868
869 if (port_count > sizeof(port_mask) * 8) {
870 if (verbose)
871 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
872 return 2;
873 }
874
875 /* scroll through /sys/dev/block looking for devices attached to
876 * this hba
877 */
878 dir = opendir("/sys/dev/block");
879 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
880 int fd;
881 char model[64];
882 char vendor[64];
883 char buf[1024];
884 int major, minor;
885 char *device;
886 char *c;
887 int port;
888 int type;
889
890 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
891 continue;
892 path = devt_to_devpath(makedev(major, minor));
893 if (!path)
894 continue;
895 if (!path_attached_to_hba(path, hba_path)) {
896 free(path);
897 path = NULL;
898 continue;
899 }
900
901 /* retrieve the scsi device type */
902 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
903 if (verbose)
904 fprintf(stderr, Name ": failed to allocate 'device'\n");
905 err = 2;
906 break;
907 }
908 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
909 if (load_sys(device, buf) != 0) {
910 if (verbose)
911 fprintf(stderr, Name ": failed to read device type for %s\n",
912 path);
913 err = 2;
914 free(device);
915 break;
916 }
917 type = strtoul(buf, NULL, 10);
918
919 /* if it's not a disk print the vendor and model */
920 if (!(type == 0 || type == 7 || type == 14)) {
921 vendor[0] = '\0';
922 model[0] = '\0';
923 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
924 if (load_sys(device, buf) == 0) {
925 strncpy(vendor, buf, sizeof(vendor));
926 vendor[sizeof(vendor) - 1] = '\0';
927 c = (char *) &vendor[sizeof(vendor) - 1];
928 while (isspace(*c) || *c == '\0')
929 *c-- = '\0';
930
931 }
932 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
933 if (load_sys(device, buf) == 0) {
934 strncpy(model, buf, sizeof(model));
935 model[sizeof(model) - 1] = '\0';
936 c = (char *) &model[sizeof(model) - 1];
937 while (isspace(*c) || *c == '\0')
938 *c-- = '\0';
939 }
940
941 if (vendor[0] && model[0])
942 sprintf(buf, "%.64s %.64s", vendor, model);
943 else
944 switch (type) { /* numbers from hald/linux/device.c */
945 case 1: sprintf(buf, "tape"); break;
946 case 2: sprintf(buf, "printer"); break;
947 case 3: sprintf(buf, "processor"); break;
948 case 4:
949 case 5: sprintf(buf, "cdrom"); break;
950 case 6: sprintf(buf, "scanner"); break;
951 case 8: sprintf(buf, "media_changer"); break;
952 case 9: sprintf(buf, "comm"); break;
953 case 12: sprintf(buf, "raid"); break;
954 default: sprintf(buf, "unknown");
955 }
956 } else
957 buf[0] = '\0';
958 free(device);
959
960 /* chop device path to 'host%d' and calculate the port number */
961 c = strchr(&path[hba_len], '/');
962 *c = '\0';
963 if (sscanf(&path[hba_len], "host%d", &port) == 1)
964 port -= host_base;
965 else {
966 if (verbose) {
967 *c = '/'; /* repair the full string */
968 fprintf(stderr, Name ": failed to determine port number for %s\n",
969 path);
970 }
971 err = 2;
972 break;
973 }
974
975 /* mark this port as used */
976 port_mask &= ~(1 << port);
977
978 /* print out the device information */
979 if (buf[0]) {
980 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
981 continue;
982 }
983
984 fd = dev_open(ent->d_name, O_RDONLY);
985 if (fd < 0)
986 printf(" Port%d : - disk info unavailable -\n", port);
987 else {
988 fd2devname(fd, buf);
989 printf(" Port%d : %s", port, buf);
990 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
991 printf(" (%s)\n", buf);
992 else
993 printf("()\n");
994 }
995 close(fd);
996 free(path);
997 path = NULL;
998 }
999 if (path)
1000 free(path);
1001 if (dir)
1002 closedir(dir);
1003 if (err == 0) {
1004 int i;
1005
1006 for (i = 0; i < port_count; i++)
1007 if (port_mask & (1 << i))
1008 printf(" Port%d : - no device attached -\n", i);
1009 }
1010
1011 return err;
1012 }
1013
1014 static int detail_platform_imsm(int verbose, int enumerate_only)
1015 {
1016 /* There are two components to imsm platform support, the ahci SATA
1017 * controller and the option-rom. To find the SATA controller we
1018 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1019 * controller with the Intel vendor id is present. This approach
1020 * allows mdadm to leverage the kernel's ahci detection logic, with the
1021 * caveat that if ahci.ko is not loaded mdadm will not be able to
1022 * detect platform raid capabilities. The option-rom resides in a
1023 * platform "Adapter ROM". We scan for its signature to retrieve the
1024 * platform capabilities. If raid support is disabled in the BIOS the
1025 * option-rom capability structure will not be available.
1026 */
1027 const struct imsm_orom *orom;
1028 struct sys_dev *list, *hba;
1029 DIR *dir;
1030 struct dirent *ent;
1031 const char *hba_path;
1032 int host_base = 0;
1033 int port_count = 0;
1034
1035 if (enumerate_only) {
1036 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1037 return 0;
1038 return 2;
1039 }
1040
1041 list = find_driver_devices("pci", "ahci");
1042 for (hba = list; hba; hba = hba->next)
1043 if (devpath_to_vendor(hba->path) == 0x8086)
1044 break;
1045
1046 if (!hba) {
1047 if (verbose)
1048 fprintf(stderr, Name ": unable to find active ahci controller\n");
1049 free_sys_dev(&list);
1050 return 2;
1051 } else if (verbose)
1052 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
1053 hba_path = hba->path;
1054 hba->path = NULL;
1055 free_sys_dev(&list);
1056
1057 orom = find_imsm_orom();
1058 if (!orom) {
1059 if (verbose)
1060 fprintf(stderr, Name ": imsm option-rom not found\n");
1061 return 2;
1062 }
1063
1064 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1065 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
1066 orom->hotfix_ver, orom->build);
1067 printf(" RAID Levels :%s%s%s%s%s\n",
1068 imsm_orom_has_raid0(orom) ? " raid0" : "",
1069 imsm_orom_has_raid1(orom) ? " raid1" : "",
1070 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
1071 imsm_orom_has_raid10(orom) ? " raid10" : "",
1072 imsm_orom_has_raid5(orom) ? " raid5" : "");
1073 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1074 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
1075 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
1076 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
1077 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
1078 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
1079 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
1080 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
1081 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
1082 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
1083 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
1084 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
1085 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
1086 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
1087 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
1088 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
1089 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1090 printf(" Max Disks : %d\n", orom->tds);
1091 printf(" Max Volumes : %d\n", orom->vpa);
1092 printf(" I/O Controller : %s\n", hba_path);
1093
1094 /* find the smallest scsi host number to determine a port number base */
1095 dir = opendir(hba_path);
1096 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1097 int host;
1098
1099 if (sscanf(ent->d_name, "host%d", &host) != 1)
1100 continue;
1101 if (port_count == 0)
1102 host_base = host;
1103 else if (host < host_base)
1104 host_base = host;
1105
1106 if (host + 1 > port_count + host_base)
1107 port_count = host + 1 - host_base;
1108
1109 }
1110 if (dir)
1111 closedir(dir);
1112
1113 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1114 host_base, verbose) != 0) {
1115 if (verbose)
1116 fprintf(stderr, Name ": failed to enumerate ports\n");
1117 return 2;
1118 }
1119
1120 return 0;
1121 }
1122 #endif
1123
1124 static int match_home_imsm(struct supertype *st, char *homehost)
1125 {
1126 /* the imsm metadata format does not specify any host
1127 * identification information. We return -1 since we can never
1128 * confirm nor deny whether a given array is "meant" for this
1129 * host. We rely on compare_super and the 'family_num' fields to
1130 * exclude member disks that do not belong, and we rely on
1131 * mdadm.conf to specify the arrays that should be assembled.
1132 * Auto-assembly may still pick up "foreign" arrays.
1133 */
1134
1135 return -1;
1136 }
1137
1138 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1139 {
1140 /* The uuid returned here is used for:
1141 * uuid to put into bitmap file (Create, Grow)
1142 * uuid for backup header when saving critical section (Grow)
1143 * comparing uuids when re-adding a device into an array
1144 * In these cases the uuid required is that of the data-array,
1145 * not the device-set.
1146 * uuid to recognise same set when adding a missing device back
1147 * to an array. This is a uuid for the device-set.
1148 *
1149 * For each of these we can make do with a truncated
1150 * or hashed uuid rather than the original, as long as
1151 * everyone agrees.
1152 * In each case the uuid required is that of the data-array,
1153 * not the device-set.
1154 */
1155 /* imsm does not track uuid's so we synthesis one using sha1 on
1156 * - The signature (Which is constant for all imsm array, but no matter)
1157 * - the orig_family_num of the container
1158 * - the index number of the volume
1159 * - the 'serial' number of the volume.
1160 * Hopefully these are all constant.
1161 */
1162 struct intel_super *super = st->sb;
1163
1164 char buf[20];
1165 struct sha1_ctx ctx;
1166 struct imsm_dev *dev = NULL;
1167 __u32 family_num;
1168
1169 /* some mdadm versions failed to set ->orig_family_num, in which
1170 * case fall back to ->family_num. orig_family_num will be
1171 * fixed up with the first metadata update.
1172 */
1173 family_num = super->anchor->orig_family_num;
1174 if (family_num == 0)
1175 family_num = super->anchor->family_num;
1176 sha1_init_ctx(&ctx);
1177 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1178 sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
1179 if (super->current_vol >= 0)
1180 dev = get_imsm_dev(super, super->current_vol);
1181 if (dev) {
1182 __u32 vol = super->current_vol;
1183 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1184 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1185 }
1186 sha1_finish_ctx(&ctx, buf);
1187 memcpy(uuid, buf, 4*4);
1188 }
1189
1190 #if 0
1191 static void
1192 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1193 {
1194 __u8 *v = get_imsm_version(mpb);
1195 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1196 char major[] = { 0, 0, 0 };
1197 char minor[] = { 0 ,0, 0 };
1198 char patch[] = { 0, 0, 0 };
1199 char *ver_parse[] = { major, minor, patch };
1200 int i, j;
1201
1202 i = j = 0;
1203 while (*v != '\0' && v < end) {
1204 if (*v != '.' && j < 2)
1205 ver_parse[i][j++] = *v;
1206 else {
1207 i++;
1208 j = 0;
1209 }
1210 v++;
1211 }
1212
1213 *m = strtol(minor, NULL, 0);
1214 *p = strtol(patch, NULL, 0);
1215 }
1216 #endif
1217
1218 static int imsm_level_to_layout(int level)
1219 {
1220 switch (level) {
1221 case 0:
1222 case 1:
1223 return 0;
1224 case 5:
1225 case 6:
1226 return ALGORITHM_LEFT_ASYMMETRIC;
1227 case 10:
1228 return 0x102;
1229 }
1230 return UnSet;
1231 }
1232
1233 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
1234 {
1235 struct intel_super *super = st->sb;
1236 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1237 struct imsm_map *map = get_imsm_map(dev, 0);
1238 struct dl *dl;
1239
1240 for (dl = super->disks; dl; dl = dl->next)
1241 if (dl->raiddisk == info->disk.raid_disk)
1242 break;
1243 info->container_member = super->current_vol;
1244 info->array.raid_disks = map->num_members;
1245 info->array.level = get_imsm_raid_level(map);
1246 info->array.layout = imsm_level_to_layout(info->array.level);
1247 info->array.md_minor = -1;
1248 info->array.ctime = 0;
1249 info->array.utime = 0;
1250 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1251 info->array.state = !dev->vol.dirty;
1252 info->custom_array_size = __le32_to_cpu(dev->size_high);
1253 info->custom_array_size <<= 32;
1254 info->custom_array_size |= __le32_to_cpu(dev->size_low);
1255
1256 info->disk.major = 0;
1257 info->disk.minor = 0;
1258 if (dl) {
1259 info->disk.major = dl->major;
1260 info->disk.minor = dl->minor;
1261 }
1262
1263 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1264 info->component_size = __le32_to_cpu(map->blocks_per_member);
1265 memset(info->uuid, 0, sizeof(info->uuid));
1266
1267 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
1268 info->resync_start = 0;
1269 else if (dev->vol.migr_state)
1270 /* FIXME add curr_migr_unit to resync_start conversion */
1271 info->resync_start = 0;
1272 else
1273 info->resync_start = ~0ULL;
1274
1275 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1276 info->name[MAX_RAID_SERIAL_LEN] = 0;
1277
1278 info->array.major_version = -1;
1279 info->array.minor_version = -2;
1280 sprintf(info->text_version, "/%s/%d",
1281 devnum2devname(st->container_dev),
1282 info->container_member);
1283 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1284 uuid_from_super_imsm(st, info->uuid);
1285 }
1286
1287 /* check the config file to see if we can return a real uuid for this spare */
1288 static void fixup_container_spare_uuid(struct mdinfo *inf)
1289 {
1290 struct mddev_ident_s *array_list;
1291
1292 if (inf->array.level != LEVEL_CONTAINER ||
1293 memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0)
1294 return;
1295
1296 array_list = conf_get_ident(NULL);
1297
1298 for (; array_list; array_list = array_list->next) {
1299 if (array_list->uuid_set) {
1300 struct supertype *_sst; /* spare supertype */
1301 struct supertype *_cst; /* container supertype */
1302
1303 _cst = array_list->st;
1304 if (_cst)
1305 _sst = _cst->ss->match_metadata_desc(inf->text_version);
1306 else
1307 _sst = NULL;
1308
1309 if (_sst) {
1310 memcpy(inf->uuid, array_list->uuid, sizeof(int[4]));
1311 free(_sst);
1312 break;
1313 }
1314 }
1315 }
1316 }
1317
1318 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
1319 {
1320 struct intel_super *super = st->sb;
1321 struct imsm_disk *disk;
1322
1323 if (super->current_vol >= 0) {
1324 getinfo_super_imsm_volume(st, info);
1325 return;
1326 }
1327
1328 /* Set raid_disks to zero so that Assemble will always pull in valid
1329 * spares
1330 */
1331 info->array.raid_disks = 0;
1332 info->array.level = LEVEL_CONTAINER;
1333 info->array.layout = 0;
1334 info->array.md_minor = -1;
1335 info->array.ctime = 0; /* N/A for imsm */
1336 info->array.utime = 0;
1337 info->array.chunk_size = 0;
1338
1339 info->disk.major = 0;
1340 info->disk.minor = 0;
1341 info->disk.raid_disk = -1;
1342 info->reshape_active = 0;
1343 info->array.major_version = -1;
1344 info->array.minor_version = -2;
1345 strcpy(info->text_version, "imsm");
1346 info->safe_mode_delay = 0;
1347 info->disk.number = -1;
1348 info->disk.state = 0;
1349 info->name[0] = 0;
1350
1351 if (super->disks) {
1352 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1353
1354 disk = &super->disks->disk;
1355 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1356 info->component_size = reserved;
1357 info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
1358 /* we don't change info->disk.raid_disk here because
1359 * this state will be finalized in mdmon after we have
1360 * found the 'most fresh' version of the metadata
1361 */
1362 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
1363 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
1364 }
1365
1366 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1367 * ->compare_super may have updated the 'num_raid_devs' field for spares
1368 */
1369 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1370 uuid_from_super_imsm(st, info->uuid);
1371 else {
1372 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1373 fixup_container_spare_uuid(info);
1374 }
1375 }
1376
1377 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1378 char *update, char *devname, int verbose,
1379 int uuid_set, char *homehost)
1380 {
1381 /* FIXME */
1382
1383 /* For 'assemble' and 'force' we need to return non-zero if any
1384 * change was made. For others, the return value is ignored.
1385 * Update options are:
1386 * force-one : This device looks a bit old but needs to be included,
1387 * update age info appropriately.
1388 * assemble: clear any 'faulty' flag to allow this device to
1389 * be assembled.
1390 * force-array: Array is degraded but being forced, mark it clean
1391 * if that will be needed to assemble it.
1392 *
1393 * newdev: not used ????
1394 * grow: Array has gained a new device - this is currently for
1395 * linear only
1396 * resync: mark as dirty so a resync will happen.
1397 * name: update the name - preserving the homehost
1398 *
1399 * Following are not relevant for this imsm:
1400 * sparc2.2 : update from old dodgey metadata
1401 * super-minor: change the preferred_minor number
1402 * summaries: update redundant counters.
1403 * uuid: Change the uuid of the array to match watch is given
1404 * homehost: update the recorded homehost
1405 * _reshape_progress: record new reshape_progress position.
1406 */
1407 int rv = 0;
1408 //struct intel_super *super = st->sb;
1409 //struct imsm_super *mpb = super->mpb;
1410
1411 if (strcmp(update, "grow") == 0) {
1412 }
1413 if (strcmp(update, "resync") == 0) {
1414 /* dev->vol.dirty = 1; */
1415 }
1416
1417 /* IMSM has no concept of UUID or homehost */
1418
1419 return rv;
1420 }
1421
1422 static size_t disks_to_mpb_size(int disks)
1423 {
1424 size_t size;
1425
1426 size = sizeof(struct imsm_super);
1427 size += (disks - 1) * sizeof(struct imsm_disk);
1428 size += 2 * sizeof(struct imsm_dev);
1429 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1430 size += (4 - 2) * sizeof(struct imsm_map);
1431 /* 4 possible disk_ord_tbl's */
1432 size += 4 * (disks - 1) * sizeof(__u32);
1433
1434 return size;
1435 }
1436
1437 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1438 {
1439 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1440 return 0;
1441
1442 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1443 }
1444
1445 static void free_devlist(struct intel_super *super)
1446 {
1447 struct intel_dev *dv;
1448
1449 while (super->devlist) {
1450 dv = super->devlist->next;
1451 free(super->devlist->dev);
1452 free(super->devlist);
1453 super->devlist = dv;
1454 }
1455 }
1456
1457 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1458 {
1459 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1460 }
1461
1462 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1463 {
1464 /*
1465 * return:
1466 * 0 same, or first was empty, and second was copied
1467 * 1 second had wrong number
1468 * 2 wrong uuid
1469 * 3 wrong other info
1470 */
1471 struct intel_super *first = st->sb;
1472 struct intel_super *sec = tst->sb;
1473
1474 if (!first) {
1475 st->sb = tst->sb;
1476 tst->sb = NULL;
1477 return 0;
1478 }
1479
1480 /* if an anchor does not have num_raid_devs set then it is a free
1481 * floating spare
1482 */
1483 if (first->anchor->num_raid_devs > 0 &&
1484 sec->anchor->num_raid_devs > 0) {
1485 /* Determine if these disks might ever have been
1486 * related. Further disambiguation can only take place
1487 * in load_super_imsm_all
1488 */
1489 __u32 first_family = first->anchor->orig_family_num;
1490 __u32 sec_family = sec->anchor->orig_family_num;
1491
1492 if (memcmp(first->anchor->sig, sec->anchor->sig,
1493 MAX_SIGNATURE_LENGTH) != 0)
1494 return 3;
1495
1496 if (first_family == 0)
1497 first_family = first->anchor->family_num;
1498 if (sec_family == 0)
1499 sec_family = sec->anchor->family_num;
1500
1501 if (first_family != sec_family)
1502 return 3;
1503
1504 }
1505
1506
1507 /* if 'first' is a spare promote it to a populated mpb with sec's
1508 * family number
1509 */
1510 if (first->anchor->num_raid_devs == 0 &&
1511 sec->anchor->num_raid_devs > 0) {
1512 int i;
1513 struct intel_dev *dv;
1514 struct imsm_dev *dev;
1515
1516 /* we need to copy raid device info from sec if an allocation
1517 * fails here we don't associate the spare
1518 */
1519 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1520 dv = malloc(sizeof(*dv));
1521 if (!dv)
1522 break;
1523 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1524 if (!dev) {
1525 free(dv);
1526 break;
1527 }
1528 dv->dev = dev;
1529 dv->index = i;
1530 dv->next = first->devlist;
1531 first->devlist = dv;
1532 }
1533 if (i < sec->anchor->num_raid_devs) {
1534 /* allocation failure */
1535 free_devlist(first);
1536 fprintf(stderr, "imsm: failed to associate spare\n");
1537 return 3;
1538 }
1539 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1540 first->anchor->orig_family_num = sec->anchor->orig_family_num;
1541 first->anchor->family_num = sec->anchor->family_num;
1542 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1543 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1544 }
1545
1546 return 0;
1547 }
1548
1549 static void fd2devname(int fd, char *name)
1550 {
1551 struct stat st;
1552 char path[256];
1553 char dname[100];
1554 char *nm;
1555 int rv;
1556
1557 name[0] = '\0';
1558 if (fstat(fd, &st) != 0)
1559 return;
1560 sprintf(path, "/sys/dev/block/%d:%d",
1561 major(st.st_rdev), minor(st.st_rdev));
1562
1563 rv = readlink(path, dname, sizeof(dname));
1564 if (rv <= 0)
1565 return;
1566
1567 dname[rv] = '\0';
1568 nm = strrchr(dname, '/');
1569 nm++;
1570 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1571 }
1572
1573 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1574
1575 static int imsm_read_serial(int fd, char *devname,
1576 __u8 serial[MAX_RAID_SERIAL_LEN])
1577 {
1578 unsigned char scsi_serial[255];
1579 int rv;
1580 int rsp_len;
1581 int len;
1582 char *dest;
1583 char *src;
1584 char *rsp_buf;
1585 int i;
1586
1587 memset(scsi_serial, 0, sizeof(scsi_serial));
1588
1589 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1590
1591 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1592 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1593 fd2devname(fd, (char *) serial);
1594 return 0;
1595 }
1596
1597 if (rv != 0) {
1598 if (devname)
1599 fprintf(stderr,
1600 Name ": Failed to retrieve serial for %s\n",
1601 devname);
1602 return rv;
1603 }
1604
1605 rsp_len = scsi_serial[3];
1606 if (!rsp_len) {
1607 if (devname)
1608 fprintf(stderr,
1609 Name ": Failed to retrieve serial for %s\n",
1610 devname);
1611 return 2;
1612 }
1613 rsp_buf = (char *) &scsi_serial[4];
1614
1615 /* trim all whitespace and non-printable characters and convert
1616 * ':' to ';'
1617 */
1618 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
1619 src = &rsp_buf[i];
1620 if (*src > 0x20) {
1621 /* ':' is reserved for use in placeholder serial
1622 * numbers for missing disks
1623 */
1624 if (*src == ':')
1625 *dest++ = ';';
1626 else
1627 *dest++ = *src;
1628 }
1629 }
1630 len = dest - rsp_buf;
1631 dest = rsp_buf;
1632
1633 /* truncate leading characters */
1634 if (len > MAX_RAID_SERIAL_LEN) {
1635 dest += len - MAX_RAID_SERIAL_LEN;
1636 len = MAX_RAID_SERIAL_LEN;
1637 }
1638
1639 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1640 memcpy(serial, dest, len);
1641
1642 return 0;
1643 }
1644
1645 static int serialcmp(__u8 *s1, __u8 *s2)
1646 {
1647 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1648 }
1649
1650 static void serialcpy(__u8 *dest, __u8 *src)
1651 {
1652 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1653 }
1654
1655 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1656 {
1657 struct dl *dl;
1658
1659 for (dl = super->disks; dl; dl = dl->next)
1660 if (serialcmp(dl->serial, serial) == 0)
1661 break;
1662
1663 return dl;
1664 }
1665
1666 static struct imsm_disk *
1667 __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
1668 {
1669 int i;
1670
1671 for (i = 0; i < mpb->num_disks; i++) {
1672 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
1673
1674 if (serialcmp(disk->serial, serial) == 0) {
1675 if (idx)
1676 *idx = i;
1677 return disk;
1678 }
1679 }
1680
1681 return NULL;
1682 }
1683
1684 static int
1685 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1686 {
1687 struct imsm_disk *disk;
1688 struct dl *dl;
1689 struct stat stb;
1690 int rv;
1691 char name[40];
1692 __u8 serial[MAX_RAID_SERIAL_LEN];
1693
1694 rv = imsm_read_serial(fd, devname, serial);
1695
1696 if (rv != 0)
1697 return 2;
1698
1699 dl = calloc(1, sizeof(*dl));
1700 if (!dl) {
1701 if (devname)
1702 fprintf(stderr,
1703 Name ": failed to allocate disk buffer for %s\n",
1704 devname);
1705 return 2;
1706 }
1707
1708 fstat(fd, &stb);
1709 dl->major = major(stb.st_rdev);
1710 dl->minor = minor(stb.st_rdev);
1711 dl->next = super->disks;
1712 dl->fd = keep_fd ? fd : -1;
1713 assert(super->disks == NULL);
1714 super->disks = dl;
1715 serialcpy(dl->serial, serial);
1716 dl->index = -2;
1717 dl->e = NULL;
1718 fd2devname(fd, name);
1719 if (devname)
1720 dl->devname = strdup(devname);
1721 else
1722 dl->devname = strdup(name);
1723
1724 /* look up this disk's index in the current anchor */
1725 disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
1726 if (disk) {
1727 dl->disk = *disk;
1728 /* only set index on disks that are a member of a
1729 * populated contianer, i.e. one with raid_devs
1730 */
1731 if (is_failed(&dl->disk))
1732 dl->index = -2;
1733 else if (is_spare(&dl->disk))
1734 dl->index = -1;
1735 }
1736
1737 return 0;
1738 }
1739
1740 #ifndef MDASSEMBLE
1741 /* When migrating map0 contains the 'destination' state while map1
1742 * contains the current state. When not migrating map0 contains the
1743 * current state. This routine assumes that map[0].map_state is set to
1744 * the current array state before being called.
1745 *
1746 * Migration is indicated by one of the following states
1747 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1748 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1749 * map1state=unitialized)
1750 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1751 * map1state=normal)
1752 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1753 * map1state=degraded)
1754 */
1755 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
1756 {
1757 struct imsm_map *dest;
1758 struct imsm_map *src = get_imsm_map(dev, 0);
1759
1760 dev->vol.migr_state = 1;
1761 set_migr_type(dev, migr_type);
1762 dev->vol.curr_migr_unit = 0;
1763 dest = get_imsm_map(dev, 1);
1764
1765 /* duplicate and then set the target end state in map[0] */
1766 memcpy(dest, src, sizeof_imsm_map(src));
1767 if (migr_type == MIGR_REBUILD) {
1768 __u32 ord;
1769 int i;
1770
1771 for (i = 0; i < src->num_members; i++) {
1772 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
1773 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
1774 }
1775 }
1776
1777 src->map_state = to_state;
1778 }
1779
1780 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1781 {
1782 struct imsm_map *map = get_imsm_map(dev, 0);
1783 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
1784 int i;
1785
1786 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1787 * completed in the last migration.
1788 *
1789 * FIXME add support for online capacity expansion and
1790 * raid-level-migration
1791 */
1792 for (i = 0; i < prev->num_members; i++)
1793 map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i];
1794
1795 dev->vol.migr_state = 0;
1796 dev->vol.curr_migr_unit = 0;
1797 map->map_state = map_state;
1798 }
1799 #endif
1800
1801 static int parse_raid_devices(struct intel_super *super)
1802 {
1803 int i;
1804 struct imsm_dev *dev_new;
1805 size_t len, len_migr;
1806 size_t space_needed = 0;
1807 struct imsm_super *mpb = super->anchor;
1808
1809 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1810 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1811 struct intel_dev *dv;
1812
1813 len = sizeof_imsm_dev(dev_iter, 0);
1814 len_migr = sizeof_imsm_dev(dev_iter, 1);
1815 if (len_migr > len)
1816 space_needed += len_migr - len;
1817
1818 dv = malloc(sizeof(*dv));
1819 if (!dv)
1820 return 1;
1821 dev_new = malloc(len_migr);
1822 if (!dev_new) {
1823 free(dv);
1824 return 1;
1825 }
1826 imsm_copy_dev(dev_new, dev_iter);
1827 dv->dev = dev_new;
1828 dv->index = i;
1829 dv->next = super->devlist;
1830 super->devlist = dv;
1831 }
1832
1833 /* ensure that super->buf is large enough when all raid devices
1834 * are migrating
1835 */
1836 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1837 void *buf;
1838
1839 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1840 if (posix_memalign(&buf, 512, len) != 0)
1841 return 1;
1842
1843 memcpy(buf, super->buf, super->len);
1844 memset(buf + super->len, 0, len - super->len);
1845 free(super->buf);
1846 super->buf = buf;
1847 super->len = len;
1848 }
1849
1850 return 0;
1851 }
1852
1853 /* retrieve a pointer to the bbm log which starts after all raid devices */
1854 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1855 {
1856 void *ptr = NULL;
1857
1858 if (__le32_to_cpu(mpb->bbm_log_size)) {
1859 ptr = mpb;
1860 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1861 }
1862
1863 return ptr;
1864 }
1865
1866 static void __free_imsm(struct intel_super *super, int free_disks);
1867
1868 /* load_imsm_mpb - read matrix metadata
1869 * allocates super->mpb to be freed by free_super
1870 */
1871 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1872 {
1873 unsigned long long dsize;
1874 unsigned long long sectors;
1875 struct stat;
1876 struct imsm_super *anchor;
1877 __u32 check_sum;
1878
1879 get_dev_size(fd, NULL, &dsize);
1880
1881 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1882 if (devname)
1883 fprintf(stderr,
1884 Name ": Cannot seek to anchor block on %s: %s\n",
1885 devname, strerror(errno));
1886 return 1;
1887 }
1888
1889 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1890 if (devname)
1891 fprintf(stderr,
1892 Name ": Failed to allocate imsm anchor buffer"
1893 " on %s\n", devname);
1894 return 1;
1895 }
1896 if (read(fd, anchor, 512) != 512) {
1897 if (devname)
1898 fprintf(stderr,
1899 Name ": Cannot read anchor block on %s: %s\n",
1900 devname, strerror(errno));
1901 free(anchor);
1902 return 1;
1903 }
1904
1905 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1906 if (devname)
1907 fprintf(stderr,
1908 Name ": no IMSM anchor on %s\n", devname);
1909 free(anchor);
1910 return 2;
1911 }
1912
1913 __free_imsm(super, 0);
1914 super->len = ROUND_UP(anchor->mpb_size, 512);
1915 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1916 if (devname)
1917 fprintf(stderr,
1918 Name ": unable to allocate %zu byte mpb buffer\n",
1919 super->len);
1920 free(anchor);
1921 return 2;
1922 }
1923 memcpy(super->buf, anchor, 512);
1924
1925 sectors = mpb_sectors(anchor) - 1;
1926 free(anchor);
1927 if (!sectors) {
1928 check_sum = __gen_imsm_checksum(super->anchor);
1929 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1930 if (devname)
1931 fprintf(stderr,
1932 Name ": IMSM checksum %x != %x on %s\n",
1933 check_sum,
1934 __le32_to_cpu(super->anchor->check_sum),
1935 devname);
1936 return 2;
1937 }
1938
1939 return 0;
1940 }
1941
1942 /* read the extended mpb */
1943 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1944 if (devname)
1945 fprintf(stderr,
1946 Name ": Cannot seek to extended mpb on %s: %s\n",
1947 devname, strerror(errno));
1948 return 1;
1949 }
1950
1951 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1952 if (devname)
1953 fprintf(stderr,
1954 Name ": Cannot read extended mpb on %s: %s\n",
1955 devname, strerror(errno));
1956 return 2;
1957 }
1958
1959 check_sum = __gen_imsm_checksum(super->anchor);
1960 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1961 if (devname)
1962 fprintf(stderr,
1963 Name ": IMSM checksum %x != %x on %s\n",
1964 check_sum, __le32_to_cpu(super->anchor->check_sum),
1965 devname);
1966 return 3;
1967 }
1968
1969 /* FIXME the BBM log is disk specific so we cannot use this global
1970 * buffer for all disks. Ok for now since we only look at the global
1971 * bbm_log_size parameter to gate assembly
1972 */
1973 super->bbm_log = __get_imsm_bbm_log(super->anchor);
1974
1975 return 0;
1976 }
1977
1978 static int
1979 load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
1980 {
1981 int err;
1982
1983 err = load_imsm_mpb(fd, super, devname);
1984 if (err)
1985 return err;
1986 err = load_imsm_disk(fd, super, devname, keep_fd);
1987 if (err)
1988 return err;
1989 err = parse_raid_devices(super);
1990
1991 return err;
1992 }
1993
1994 static void __free_imsm_disk(struct dl *d)
1995 {
1996 if (d->fd >= 0)
1997 close(d->fd);
1998 if (d->devname)
1999 free(d->devname);
2000 if (d->e)
2001 free(d->e);
2002 free(d);
2003
2004 }
2005 static void free_imsm_disks(struct intel_super *super)
2006 {
2007 struct dl *d;
2008
2009 while (super->disks) {
2010 d = super->disks;
2011 super->disks = d->next;
2012 __free_imsm_disk(d);
2013 }
2014 while (super->missing) {
2015 d = super->missing;
2016 super->missing = d->next;
2017 __free_imsm_disk(d);
2018 }
2019
2020 }
2021
2022 /* free all the pieces hanging off of a super pointer */
2023 static void __free_imsm(struct intel_super *super, int free_disks)
2024 {
2025 if (super->buf) {
2026 free(super->buf);
2027 super->buf = NULL;
2028 }
2029 if (free_disks)
2030 free_imsm_disks(super);
2031 free_devlist(super);
2032 if (super->hba) {
2033 free((void *) super->hba);
2034 super->hba = NULL;
2035 }
2036 }
2037
2038 static void free_imsm(struct intel_super *super)
2039 {
2040 __free_imsm(super, 1);
2041 free(super);
2042 }
2043
2044 static void free_super_imsm(struct supertype *st)
2045 {
2046 struct intel_super *super = st->sb;
2047
2048 if (!super)
2049 return;
2050
2051 free_imsm(super);
2052 st->sb = NULL;
2053 }
2054
2055 static struct intel_super *alloc_super(int creating_imsm)
2056 {
2057 struct intel_super *super = malloc(sizeof(*super));
2058
2059 if (super) {
2060 memset(super, 0, sizeof(*super));
2061 super->creating_imsm = creating_imsm;
2062 super->current_vol = -1;
2063 super->create_offset = ~((__u32 ) 0);
2064 if (!check_env("IMSM_NO_PLATFORM"))
2065 super->orom = find_imsm_orom();
2066 if (super->orom && !check_env("IMSM_TEST_OROM")) {
2067 struct sys_dev *list, *ent;
2068
2069 /* find the first intel ahci controller */
2070 list = find_driver_devices("pci", "ahci");
2071 for (ent = list; ent; ent = ent->next)
2072 if (devpath_to_vendor(ent->path) == 0x8086)
2073 break;
2074 if (ent) {
2075 super->hba = ent->path;
2076 ent->path = NULL;
2077 }
2078 free_sys_dev(&list);
2079 }
2080 }
2081
2082 return super;
2083 }
2084
2085 #ifndef MDASSEMBLE
2086 /* find_missing - helper routine for load_super_imsm_all that identifies
2087 * disks that have disappeared from the system. This routine relies on
2088 * the mpb being uptodate, which it is at load time.
2089 */
2090 static int find_missing(struct intel_super *super)
2091 {
2092 int i;
2093 struct imsm_super *mpb = super->anchor;
2094 struct dl *dl;
2095 struct imsm_disk *disk;
2096
2097 for (i = 0; i < mpb->num_disks; i++) {
2098 disk = __get_imsm_disk(mpb, i);
2099 dl = serial_to_dl(disk->serial, super);
2100 if (dl)
2101 continue;
2102
2103 dl = malloc(sizeof(*dl));
2104 if (!dl)
2105 return 1;
2106 dl->major = 0;
2107 dl->minor = 0;
2108 dl->fd = -1;
2109 dl->devname = strdup("missing");
2110 dl->index = i;
2111 serialcpy(dl->serial, disk->serial);
2112 dl->disk = *disk;
2113 dl->e = NULL;
2114 dl->next = super->missing;
2115 super->missing = dl;
2116 }
2117
2118 return 0;
2119 }
2120
2121 static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
2122 {
2123 struct intel_disk *idisk = disk_list;
2124
2125 while (idisk) {
2126 if (serialcmp(idisk->disk.serial, serial) == 0)
2127 break;
2128 idisk = idisk->next;
2129 }
2130
2131 return idisk;
2132 }
2133
2134 static int __prep_thunderdome(struct intel_super **table, int tbl_size,
2135 struct intel_super *super,
2136 struct intel_disk **disk_list)
2137 {
2138 struct imsm_disk *d = &super->disks->disk;
2139 struct imsm_super *mpb = super->anchor;
2140 int i, j;
2141
2142 for (i = 0; i < tbl_size; i++) {
2143 struct imsm_super *tbl_mpb = table[i]->anchor;
2144 struct imsm_disk *tbl_d = &table[i]->disks->disk;
2145
2146 if (tbl_mpb->family_num == mpb->family_num) {
2147 if (tbl_mpb->check_sum == mpb->check_sum) {
2148 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2149 __func__, super->disks->major,
2150 super->disks->minor,
2151 table[i]->disks->major,
2152 table[i]->disks->minor);
2153 break;
2154 }
2155
2156 if (((is_configured(d) && !is_configured(tbl_d)) ||
2157 is_configured(d) == is_configured(tbl_d)) &&
2158 tbl_mpb->generation_num < mpb->generation_num) {
2159 /* current version of the mpb is a
2160 * better candidate than the one in
2161 * super_table, but copy over "cross
2162 * generational" status
2163 */
2164 struct intel_disk *idisk;
2165
2166 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2167 __func__, super->disks->major,
2168 super->disks->minor,
2169 table[i]->disks->major,
2170 table[i]->disks->minor);
2171
2172 idisk = disk_list_get(tbl_d->serial, *disk_list);
2173 if (idisk && is_failed(&idisk->disk))
2174 tbl_d->status |= FAILED_DISK;
2175 break;
2176 } else {
2177 struct intel_disk *idisk;
2178 struct imsm_disk *disk;
2179
2180 /* tbl_mpb is more up to date, but copy
2181 * over cross generational status before
2182 * returning
2183 */
2184 disk = __serial_to_disk(d->serial, mpb, NULL);
2185 if (disk && is_failed(disk))
2186 d->status |= FAILED_DISK;
2187
2188 idisk = disk_list_get(d->serial, *disk_list);
2189 if (idisk) {
2190 idisk->owner = i;
2191 if (disk && is_configured(disk))
2192 idisk->disk.status |= CONFIGURED_DISK;
2193 }
2194
2195 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2196 __func__, super->disks->major,
2197 super->disks->minor,
2198 table[i]->disks->major,
2199 table[i]->disks->minor);
2200
2201 return tbl_size;
2202 }
2203 }
2204 }
2205
2206 if (i >= tbl_size)
2207 table[tbl_size++] = super;
2208 else
2209 table[i] = super;
2210
2211 /* update/extend the merged list of imsm_disk records */
2212 for (j = 0; j < mpb->num_disks; j++) {
2213 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
2214 struct intel_disk *idisk;
2215
2216 idisk = disk_list_get(disk->serial, *disk_list);
2217 if (idisk) {
2218 idisk->disk.status |= disk->status;
2219 if (is_configured(&idisk->disk) ||
2220 is_failed(&idisk->disk))
2221 idisk->disk.status &= ~(SPARE_DISK);
2222 } else {
2223 idisk = calloc(1, sizeof(*idisk));
2224 if (!idisk)
2225 return -1;
2226 idisk->owner = IMSM_UNKNOWN_OWNER;
2227 idisk->disk = *disk;
2228 idisk->next = *disk_list;
2229 *disk_list = idisk;
2230 }
2231
2232 if (serialcmp(idisk->disk.serial, d->serial) == 0)
2233 idisk->owner = i;
2234 }
2235
2236 return tbl_size;
2237 }
2238
2239 static struct intel_super *
2240 validate_members(struct intel_super *super, struct intel_disk *disk_list,
2241 const int owner)
2242 {
2243 struct imsm_super *mpb = super->anchor;
2244 int ok_count = 0;
2245 int i;
2246
2247 for (i = 0; i < mpb->num_disks; i++) {
2248 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2249 struct intel_disk *idisk;
2250
2251 idisk = disk_list_get(disk->serial, disk_list);
2252 if (idisk) {
2253 if (idisk->owner == owner ||
2254 idisk->owner == IMSM_UNKNOWN_OWNER)
2255 ok_count++;
2256 else
2257 dprintf("%s: '%.16s' owner %d != %d\n",
2258 __func__, disk->serial, idisk->owner,
2259 owner);
2260 } else {
2261 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2262 __func__, __le32_to_cpu(mpb->family_num), i,
2263 disk->serial);
2264 break;
2265 }
2266 }
2267
2268 if (ok_count == mpb->num_disks)
2269 return super;
2270 return NULL;
2271 }
2272
2273 static void show_conflicts(__u32 family_num, struct intel_super *super_list)
2274 {
2275 struct intel_super *s;
2276
2277 for (s = super_list; s; s = s->next) {
2278 if (family_num != s->anchor->family_num)
2279 continue;
2280 fprintf(stderr, "Conflict, offlining family %#x on '%s'\n",
2281 __le32_to_cpu(family_num), s->disks->devname);
2282 }
2283 }
2284
2285 static struct intel_super *
2286 imsm_thunderdome(struct intel_super **super_list, int len)
2287 {
2288 struct intel_super *super_table[len];
2289 struct intel_disk *disk_list = NULL;
2290 struct intel_super *champion, *spare;
2291 struct intel_super *s, **del;
2292 int tbl_size = 0;
2293 int conflict;
2294 int i;
2295
2296 memset(super_table, 0, sizeof(super_table));
2297 for (s = *super_list; s; s = s->next)
2298 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
2299
2300 for (i = 0; i < tbl_size; i++) {
2301 struct imsm_disk *d;
2302 struct intel_disk *idisk;
2303 struct imsm_super *mpb = super_table[i]->anchor;
2304
2305 s = super_table[i];
2306 d = &s->disks->disk;
2307
2308 /* 'd' must appear in merged disk list for its
2309 * configuration to be valid
2310 */
2311 idisk = disk_list_get(d->serial, disk_list);
2312 if (idisk && idisk->owner == i)
2313 s = validate_members(s, disk_list, i);
2314 else
2315 s = NULL;
2316
2317 if (!s)
2318 dprintf("%s: marking family: %#x from %d:%d offline\n",
2319 __func__, mpb->family_num,
2320 super_table[i]->disks->major,
2321 super_table[i]->disks->minor);
2322 super_table[i] = s;
2323 }
2324
2325 /* This is where the mdadm implementation differs from the Windows
2326 * driver which has no strict concept of a container. We can only
2327 * assemble one family from a container, so when returning a prodigal
2328 * array member to this system the code will not be able to disambiguate
2329 * the container contents that should be assembled ("foreign" versus
2330 * "local"). It requires user intervention to set the orig_family_num
2331 * to a new value to establish a new container. The Windows driver in
2332 * this situation fixes up the volume name in place and manages the
2333 * foreign array as an independent entity.
2334 */
2335 s = NULL;
2336 spare = NULL;
2337 conflict = 0;
2338 for (i = 0; i < tbl_size; i++) {
2339 struct intel_super *tbl_ent = super_table[i];
2340 int is_spare = 0;
2341
2342 if (!tbl_ent)
2343 continue;
2344
2345 if (tbl_ent->anchor->num_raid_devs == 0) {
2346 spare = tbl_ent;
2347 is_spare = 1;
2348 }
2349
2350 if (s && !is_spare) {
2351 show_conflicts(tbl_ent->anchor->family_num, *super_list);
2352 conflict++;
2353 } else if (!s && !is_spare)
2354 s = tbl_ent;
2355 }
2356
2357 if (!s)
2358 s = spare;
2359 if (!s) {
2360 champion = NULL;
2361 goto out;
2362 }
2363 champion = s;
2364
2365 if (conflict)
2366 fprintf(stderr, "Chose family %#x on '%s', "
2367 "assemble conflicts to new container with '--update=uuid'\n",
2368 __le32_to_cpu(s->anchor->family_num), s->disks->devname);
2369
2370 /* collect all dl's onto 'champion', and update them to
2371 * champion's version of the status
2372 */
2373 for (s = *super_list; s; s = s->next) {
2374 struct imsm_super *mpb = champion->anchor;
2375 struct dl *dl = s->disks;
2376
2377 if (s == champion)
2378 continue;
2379
2380 for (i = 0; i < mpb->num_disks; i++) {
2381 struct imsm_disk *disk;
2382
2383 disk = __serial_to_disk(dl->serial, mpb, &dl->index);
2384 if (disk) {
2385 dl->disk = *disk;
2386 /* only set index on disks that are a member of
2387 * a populated contianer, i.e. one with
2388 * raid_devs
2389 */
2390 if (is_failed(&dl->disk))
2391 dl->index = -2;
2392 else if (is_spare(&dl->disk))
2393 dl->index = -1;
2394 break;
2395 }
2396 }
2397
2398 if (i >= mpb->num_disks) {
2399 struct intel_disk *idisk;
2400
2401 idisk = disk_list_get(dl->serial, disk_list);
2402 if (is_spare(&idisk->disk) &&
2403 !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
2404 dl->index = -1;
2405 else {
2406 dl->index = -2;
2407 continue;
2408 }
2409 }
2410
2411 dl->next = champion->disks;
2412 champion->disks = dl;
2413 s->disks = NULL;
2414 }
2415
2416 /* delete 'champion' from super_list */
2417 for (del = super_list; *del; ) {
2418 if (*del == champion) {
2419 *del = (*del)->next;
2420 break;
2421 } else
2422 del = &(*del)->next;
2423 }
2424 champion->next = NULL;
2425
2426 out:
2427 while (disk_list) {
2428 struct intel_disk *idisk = disk_list;
2429
2430 disk_list = disk_list->next;
2431 free(idisk);
2432 }
2433
2434 return champion;
2435 }
2436
2437 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
2438 char *devname, int keep_fd)
2439 {
2440 struct mdinfo *sra;
2441 struct intel_super *super_list = NULL;
2442 struct intel_super *super = NULL;
2443 int devnum = fd2devnum(fd);
2444 struct mdinfo *sd;
2445 int retry;
2446 int err = 0;
2447 int i;
2448 enum sysfs_read_flags flags;
2449
2450 flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2451 if (mdmon_running(devnum))
2452 flags |= SKIP_GONE_DEVS;
2453
2454 /* check if 'fd' an opened container */
2455 sra = sysfs_read(fd, 0, flags);
2456 if (!sra)
2457 return 1;
2458
2459 if (sra->array.major_version != -1 ||
2460 sra->array.minor_version != -2 ||
2461 strcmp(sra->text_version, "imsm") != 0)
2462 return 1;
2463
2464 /* load all mpbs */
2465 for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
2466 struct intel_super *s = alloc_super(0);
2467 char nm[20];
2468 int dfd;
2469
2470 err = 1;
2471 if (!s)
2472 goto error;
2473 s->next = super_list;
2474 super_list = s;
2475
2476 err = 2;
2477 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2478 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
2479 if (dfd < 0)
2480 goto error;
2481
2482 err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
2483
2484 /* retry the load if we might have raced against mdmon */
2485 if (err == 3 && mdmon_running(devnum))
2486 for (retry = 0; retry < 3; retry++) {
2487 usleep(3000);
2488 err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
2489 if (err != 3)
2490 break;
2491 }
2492 if (!keep_fd)
2493 close(dfd);
2494 if (err)
2495 goto error;
2496 }
2497
2498 /* all mpbs enter, maybe one leaves */
2499 super = imsm_thunderdome(&super_list, i);
2500 if (!super) {
2501 err = 1;
2502 goto error;
2503 }
2504
2505 if (find_missing(super) != 0) {
2506 free_imsm(super);
2507 err = 2;
2508 goto error;
2509 }
2510
2511 if (st->subarray[0]) {
2512 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2513 super->current_vol = atoi(st->subarray);
2514 else {
2515 free_imsm(super);
2516 err = 1;
2517 goto error;
2518 }
2519 }
2520 err = 0;
2521
2522 error:
2523 while (super_list) {
2524 struct intel_super *s = super_list;
2525
2526 super_list = super_list->next;
2527 free_imsm(s);
2528 }
2529
2530 if (err)
2531 return err;
2532
2533 *sbp = super;
2534 st->container_dev = devnum;
2535 if (err == 0 && st->ss == NULL) {
2536 st->ss = &super_imsm;
2537 st->minor_version = 0;
2538 st->max_devs = IMSM_MAX_DEVICES;
2539 }
2540 st->loaded_container = 1;
2541
2542 return 0;
2543 }
2544 #endif
2545
2546 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2547 {
2548 struct intel_super *super;
2549 int rv;
2550
2551 #ifndef MDASSEMBLE
2552 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2553 return 0;
2554 #endif
2555
2556 free_super_imsm(st);
2557
2558 super = alloc_super(0);
2559 if (!super) {
2560 fprintf(stderr,
2561 Name ": malloc of %zu failed.\n",
2562 sizeof(*super));
2563 return 1;
2564 }
2565
2566 rv = load_and_parse_mpb(fd, super, devname, 0);
2567
2568 if (rv) {
2569 if (devname)
2570 fprintf(stderr,
2571 Name ": Failed to load all information "
2572 "sections on %s\n", devname);
2573 free_imsm(super);
2574 return rv;
2575 }
2576
2577 if (st->subarray[0]) {
2578 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2579 super->current_vol = atoi(st->subarray);
2580 else {
2581 free_imsm(super);
2582 return 1;
2583 }
2584 }
2585
2586 st->sb = super;
2587 if (st->ss == NULL) {
2588 st->ss = &super_imsm;
2589 st->minor_version = 0;
2590 st->max_devs = IMSM_MAX_DEVICES;
2591 }
2592 st->loaded_container = 0;
2593
2594 return 0;
2595 }
2596
2597 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2598 {
2599 if (info->level == 1)
2600 return 128;
2601 return info->chunk_size >> 9;
2602 }
2603
2604 static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
2605 {
2606 __u32 num_stripes;
2607
2608 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2609 num_stripes /= num_domains;
2610
2611 return num_stripes;
2612 }
2613
2614 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2615 {
2616 if (info->level == 1)
2617 return info->size * 2;
2618 else
2619 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2620 }
2621
2622 static void imsm_update_version_info(struct intel_super *super)
2623 {
2624 /* update the version and attributes */
2625 struct imsm_super *mpb = super->anchor;
2626 char *version;
2627 struct imsm_dev *dev;
2628 struct imsm_map *map;
2629 int i;
2630
2631 for (i = 0; i < mpb->num_raid_devs; i++) {
2632 dev = get_imsm_dev(super, i);
2633 map = get_imsm_map(dev, 0);
2634 if (__le32_to_cpu(dev->size_high) > 0)
2635 mpb->attributes |= MPB_ATTRIB_2TB;
2636
2637 /* FIXME detect when an array spans a port multiplier */
2638 #if 0
2639 mpb->attributes |= MPB_ATTRIB_PM;
2640 #endif
2641
2642 if (mpb->num_raid_devs > 1 ||
2643 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2644 version = MPB_VERSION_ATTRIBS;
2645 switch (get_imsm_raid_level(map)) {
2646 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2647 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2648 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2649 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2650 }
2651 } else {
2652 if (map->num_members >= 5)
2653 version = MPB_VERSION_5OR6_DISK_ARRAY;
2654 else if (dev->status == DEV_CLONE_N_GO)
2655 version = MPB_VERSION_CNG;
2656 else if (get_imsm_raid_level(map) == 5)
2657 version = MPB_VERSION_RAID5;
2658 else if (map->num_members >= 3)
2659 version = MPB_VERSION_3OR4_DISK_ARRAY;
2660 else if (get_imsm_raid_level(map) == 1)
2661 version = MPB_VERSION_RAID1;
2662 else
2663 version = MPB_VERSION_RAID0;
2664 }
2665 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2666 }
2667 }
2668
2669 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2670 unsigned long long size, char *name,
2671 char *homehost, int *uuid)
2672 {
2673 /* We are creating a volume inside a pre-existing container.
2674 * so st->sb is already set.
2675 */
2676 struct intel_super *super = st->sb;
2677 struct imsm_super *mpb = super->anchor;
2678 struct intel_dev *dv;
2679 struct imsm_dev *dev;
2680 struct imsm_vol *vol;
2681 struct imsm_map *map;
2682 int idx = mpb->num_raid_devs;
2683 int i;
2684 unsigned long long array_blocks;
2685 size_t size_old, size_new;
2686 __u32 num_data_stripes;
2687
2688 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2689 fprintf(stderr, Name": This imsm-container already has the "
2690 "maximum of %d volumes\n", super->orom->vpa);
2691 return 0;
2692 }
2693
2694 /* ensure the mpb is large enough for the new data */
2695 size_old = __le32_to_cpu(mpb->mpb_size);
2696 size_new = disks_to_mpb_size(info->nr_disks);
2697 if (size_new > size_old) {
2698 void *mpb_new;
2699 size_t size_round = ROUND_UP(size_new, 512);
2700
2701 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2702 fprintf(stderr, Name": could not allocate new mpb\n");
2703 return 0;
2704 }
2705 memcpy(mpb_new, mpb, size_old);
2706 free(mpb);
2707 mpb = mpb_new;
2708 super->anchor = mpb_new;
2709 mpb->mpb_size = __cpu_to_le32(size_new);
2710 memset(mpb_new + size_old, 0, size_round - size_old);
2711 }
2712 super->current_vol = idx;
2713 /* when creating the first raid device in this container set num_disks
2714 * to zero, i.e. delete this spare and add raid member devices in
2715 * add_to_super_imsm_volume()
2716 */
2717 if (super->current_vol == 0)
2718 mpb->num_disks = 0;
2719
2720 for (i = 0; i < super->current_vol; i++) {
2721 dev = get_imsm_dev(super, i);
2722 if (strncmp((char *) dev->volume, name,
2723 MAX_RAID_SERIAL_LEN) == 0) {
2724 fprintf(stderr, Name": '%s' is already defined for this container\n",
2725 name);
2726 return 0;
2727 }
2728 }
2729
2730 sprintf(st->subarray, "%d", idx);
2731 dv = malloc(sizeof(*dv));
2732 if (!dv) {
2733 fprintf(stderr, Name ": failed to allocate device list entry\n");
2734 return 0;
2735 }
2736 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2737 if (!dev) {
2738 free(dv);
2739 fprintf(stderr, Name": could not allocate raid device\n");
2740 return 0;
2741 }
2742 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2743 if (info->level == 1)
2744 array_blocks = info_to_blocks_per_member(info);
2745 else
2746 array_blocks = calc_array_size(info->level, info->raid_disks,
2747 info->layout, info->chunk_size,
2748 info->size*2);
2749 /* round array size down to closest MB */
2750 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
2751
2752 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2753 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2754 dev->status = __cpu_to_le32(0);
2755 dev->reserved_blocks = __cpu_to_le32(0);
2756 vol = &dev->vol;
2757 vol->migr_state = 0;
2758 set_migr_type(dev, MIGR_INIT);
2759 vol->dirty = 0;
2760 vol->curr_migr_unit = 0;
2761 map = get_imsm_map(dev, 0);
2762 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2763 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2764 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2765 map->failed_disk_num = ~0;
2766 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2767 IMSM_T_STATE_NORMAL;
2768 map->ddf = 1;
2769
2770 if (info->level == 1 && info->raid_disks > 2) {
2771 fprintf(stderr, Name": imsm does not support more than 2 disks"
2772 "in a raid1 volume\n");
2773 return 0;
2774 }
2775
2776 map->raid_level = info->level;
2777 if (info->level == 10) {
2778 map->raid_level = 1;
2779 map->num_domains = info->raid_disks / 2;
2780 } else if (info->level == 1)
2781 map->num_domains = info->raid_disks;
2782 else
2783 map->num_domains = 1;
2784
2785 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
2786 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
2787
2788 map->num_members = info->raid_disks;
2789 for (i = 0; i < map->num_members; i++) {
2790 /* initialized in add_to_super */
2791 set_imsm_ord_tbl_ent(map, i, 0);
2792 }
2793 mpb->num_raid_devs++;
2794
2795 dv->dev = dev;
2796 dv->index = super->current_vol;
2797 dv->next = super->devlist;
2798 super->devlist = dv;
2799
2800 imsm_update_version_info(super);
2801
2802 return 1;
2803 }
2804
2805 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2806 unsigned long long size, char *name,
2807 char *homehost, int *uuid)
2808 {
2809 /* This is primarily called by Create when creating a new array.
2810 * We will then get add_to_super called for each component, and then
2811 * write_init_super called to write it out to each device.
2812 * For IMSM, Create can create on fresh devices or on a pre-existing
2813 * array.
2814 * To create on a pre-existing array a different method will be called.
2815 * This one is just for fresh drives.
2816 */
2817 struct intel_super *super;
2818 struct imsm_super *mpb;
2819 size_t mpb_size;
2820 char *version;
2821
2822 if (!info) {
2823 st->sb = NULL;
2824 return 0;
2825 }
2826 if (st->sb)
2827 return init_super_imsm_volume(st, info, size, name, homehost,
2828 uuid);
2829
2830 super = alloc_super(1);
2831 if (!super)
2832 return 0;
2833 mpb_size = disks_to_mpb_size(info->nr_disks);
2834 if (posix_memalign(&super->buf, 512, mpb_size) != 0) {
2835 free(super);
2836 return 0;
2837 }
2838 mpb = super->buf;
2839 memset(mpb, 0, mpb_size);
2840
2841 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2842
2843 version = (char *) mpb->sig;
2844 strcpy(version, MPB_SIGNATURE);
2845 version += strlen(MPB_SIGNATURE);
2846 strcpy(version, MPB_VERSION_RAID0);
2847 mpb->mpb_size = mpb_size;
2848
2849 st->sb = super;
2850 return 1;
2851 }
2852
2853 #ifndef MDASSEMBLE
2854 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2855 int fd, char *devname)
2856 {
2857 struct intel_super *super = st->sb;
2858 struct imsm_super *mpb = super->anchor;
2859 struct dl *dl;
2860 struct imsm_dev *dev;
2861 struct imsm_map *map;
2862
2863 dev = get_imsm_dev(super, super->current_vol);
2864 map = get_imsm_map(dev, 0);
2865
2866 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2867 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2868 devname);
2869 return 1;
2870 }
2871
2872 if (fd == -1) {
2873 /* we're doing autolayout so grab the pre-marked (in
2874 * validate_geometry) raid_disk
2875 */
2876 for (dl = super->disks; dl; dl = dl->next)
2877 if (dl->raiddisk == dk->raid_disk)
2878 break;
2879 } else {
2880 for (dl = super->disks; dl ; dl = dl->next)
2881 if (dl->major == dk->major &&
2882 dl->minor == dk->minor)
2883 break;
2884 }
2885
2886 if (!dl) {
2887 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2888 return 1;
2889 }
2890
2891 /* add a pristine spare to the metadata */
2892 if (dl->index < 0) {
2893 dl->index = super->anchor->num_disks;
2894 super->anchor->num_disks++;
2895 }
2896 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2897 dl->disk.status = CONFIGURED_DISK;
2898
2899 /* if we are creating the first raid device update the family number */
2900 if (super->current_vol == 0) {
2901 __u32 sum;
2902 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2903 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2904
2905 *_dev = *dev;
2906 *_disk = dl->disk;
2907 sum = random32();
2908 sum += __gen_imsm_checksum(mpb);
2909 mpb->family_num = __cpu_to_le32(sum);
2910 mpb->orig_family_num = mpb->family_num;
2911 }
2912
2913 return 0;
2914 }
2915
2916 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2917 int fd, char *devname)
2918 {
2919 struct intel_super *super = st->sb;
2920 struct dl *dd;
2921 unsigned long long size;
2922 __u32 id;
2923 int rv;
2924 struct stat stb;
2925
2926 /* if we are on an RAID enabled platform check that the disk is
2927 * attached to the raid controller
2928 */
2929 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2930 fprintf(stderr,
2931 Name ": %s is not attached to the raid controller: %s\n",
2932 devname ? : "disk", super->hba);
2933 return 1;
2934 }
2935
2936 if (super->current_vol >= 0)
2937 return add_to_super_imsm_volume(st, dk, fd, devname);
2938
2939 fstat(fd, &stb);
2940 dd = malloc(sizeof(*dd));
2941 if (!dd) {
2942 fprintf(stderr,
2943 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2944 return 1;
2945 }
2946 memset(dd, 0, sizeof(*dd));
2947 dd->major = major(stb.st_rdev);
2948 dd->minor = minor(stb.st_rdev);
2949 dd->index = -1;
2950 dd->devname = devname ? strdup(devname) : NULL;
2951 dd->fd = fd;
2952 dd->e = NULL;
2953 rv = imsm_read_serial(fd, devname, dd->serial);
2954 if (rv) {
2955 fprintf(stderr,
2956 Name ": failed to retrieve scsi serial, aborting\n");
2957 free(dd);
2958 abort();
2959 }
2960
2961 get_dev_size(fd, NULL, &size);
2962 size /= 512;
2963 serialcpy(dd->disk.serial, dd->serial);
2964 dd->disk.total_blocks = __cpu_to_le32(size);
2965 dd->disk.status = SPARE_DISK;
2966 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
2967 dd->disk.scsi_id = __cpu_to_le32(id);
2968 else
2969 dd->disk.scsi_id = __cpu_to_le32(0);
2970
2971 if (st->update_tail) {
2972 dd->next = super->add;
2973 super->add = dd;
2974 } else {
2975 dd->next = super->disks;
2976 super->disks = dd;
2977 }
2978
2979 return 0;
2980 }
2981
2982 static int store_imsm_mpb(int fd, struct imsm_super *mpb);
2983
2984 static union {
2985 char buf[512];
2986 struct imsm_super anchor;
2987 } spare_record __attribute__ ((aligned(512)));
2988
2989 /* spare records have their own family number and do not have any defined raid
2990 * devices
2991 */
2992 static int write_super_imsm_spares(struct intel_super *super, int doclose)
2993 {
2994 struct imsm_super *mpb = super->anchor;
2995 struct imsm_super *spare = &spare_record.anchor;
2996 __u32 sum;
2997 struct dl *d;
2998
2999 spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)),
3000 spare->generation_num = __cpu_to_le32(1UL),
3001 spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
3002 spare->num_disks = 1,
3003 spare->num_raid_devs = 0,
3004 spare->cache_size = mpb->cache_size,
3005 spare->pwr_cycle_count = __cpu_to_le32(1),
3006
3007 snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
3008 MPB_SIGNATURE MPB_VERSION_RAID0);
3009
3010 for (d = super->disks; d; d = d->next) {
3011 if (d->index != -1)
3012 continue;
3013
3014 spare->disk[0] = d->disk;
3015 sum = __gen_imsm_checksum(spare);
3016 spare->family_num = __cpu_to_le32(sum);
3017 spare->orig_family_num = 0;
3018 sum = __gen_imsm_checksum(spare);
3019 spare->check_sum = __cpu_to_le32(sum);
3020
3021 if (store_imsm_mpb(d->fd, spare)) {
3022 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3023 __func__, d->major, d->minor, strerror(errno));
3024 return 1;
3025 }
3026 if (doclose) {
3027 close(d->fd);
3028 d->fd = -1;
3029 }
3030 }
3031
3032 return 0;
3033 }
3034
3035 static int write_super_imsm(struct intel_super *super, int doclose)
3036 {
3037 struct imsm_super *mpb = super->anchor;
3038 struct dl *d;
3039 __u32 generation;
3040 __u32 sum;
3041 int spares = 0;
3042 int i;
3043 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
3044
3045 /* 'generation' is incremented everytime the metadata is written */
3046 generation = __le32_to_cpu(mpb->generation_num);
3047 generation++;
3048 mpb->generation_num = __cpu_to_le32(generation);
3049
3050 /* fix up cases where previous mdadm releases failed to set
3051 * orig_family_num
3052 */
3053 if (mpb->orig_family_num == 0)
3054 mpb->orig_family_num = mpb->family_num;
3055
3056 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
3057 for (d = super->disks; d; d = d->next) {
3058 if (d->index == -1)
3059 spares++;
3060 else
3061 mpb->disk[d->index] = d->disk;
3062 }
3063 for (d = super->missing; d; d = d->next)
3064 mpb->disk[d->index] = d->disk;
3065
3066 for (i = 0; i < mpb->num_raid_devs; i++) {
3067 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
3068
3069 imsm_copy_dev(dev, get_imsm_dev(super, i));
3070 mpb_size += sizeof_imsm_dev(dev, 0);
3071 }
3072 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
3073 mpb->mpb_size = __cpu_to_le32(mpb_size);
3074
3075 /* recalculate checksum */
3076 sum = __gen_imsm_checksum(mpb);
3077 mpb->check_sum = __cpu_to_le32(sum);
3078
3079 /* write the mpb for disks that compose raid devices */
3080 for (d = super->disks; d ; d = d->next) {
3081 if (d->index < 0)
3082 continue;
3083 if (store_imsm_mpb(d->fd, mpb))
3084 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3085 __func__, d->major, d->minor, strerror(errno));
3086 if (doclose) {
3087 close(d->fd);
3088 d->fd = -1;
3089 }
3090 }
3091
3092 if (spares)
3093 return write_super_imsm_spares(super, doclose);
3094
3095 return 0;
3096 }
3097
3098
3099 static int create_array(struct supertype *st, int dev_idx)
3100 {
3101 size_t len;
3102 struct imsm_update_create_array *u;
3103 struct intel_super *super = st->sb;
3104 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
3105 struct imsm_map *map = get_imsm_map(dev, 0);
3106 struct disk_info *inf;
3107 struct imsm_disk *disk;
3108 int i;
3109
3110 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
3111 sizeof(*inf) * map->num_members;
3112 u = malloc(len);
3113 if (!u) {
3114 fprintf(stderr, "%s: failed to allocate update buffer\n",
3115 __func__);
3116 return 1;
3117 }
3118
3119 u->type = update_create_array;
3120 u->dev_idx = dev_idx;
3121 imsm_copy_dev(&u->dev, dev);
3122 inf = get_disk_info(u);
3123 for (i = 0; i < map->num_members; i++) {
3124 int idx = get_imsm_disk_idx(dev, i);
3125
3126 disk = get_imsm_disk(super, idx);
3127 serialcpy(inf[i].serial, disk->serial);
3128 }
3129 append_metadata_update(st, u, len);
3130
3131 return 0;
3132 }
3133
3134 static int _add_disk(struct supertype *st)
3135 {
3136 struct intel_super *super = st->sb;
3137 size_t len;
3138 struct imsm_update_add_disk *u;
3139
3140 if (!super->add)
3141 return 0;
3142
3143 len = sizeof(*u);
3144 u = malloc(len);
3145 if (!u) {
3146 fprintf(stderr, "%s: failed to allocate update buffer\n",
3147 __func__);
3148 return 1;
3149 }
3150
3151 u->type = update_add_disk;
3152 append_metadata_update(st, u, len);
3153
3154 return 0;
3155 }
3156
3157 static int write_init_super_imsm(struct supertype *st)
3158 {
3159 struct intel_super *super = st->sb;
3160 int current_vol = super->current_vol;
3161
3162 /* we are done with current_vol reset it to point st at the container */
3163 super->current_vol = -1;
3164
3165 if (st->update_tail) {
3166 /* queue the recently created array / added disk
3167 * as a metadata update */
3168 struct dl *d;
3169 int rv;
3170
3171 /* determine if we are creating a volume or adding a disk */
3172 if (current_vol < 0) {
3173 /* in the add disk case we are running in mdmon
3174 * context, so don't close fd's
3175 */
3176 return _add_disk(st);
3177 } else
3178 rv = create_array(st, current_vol);
3179
3180 for (d = super->disks; d ; d = d->next) {
3181 close(d->fd);
3182 d->fd = -1;
3183 }
3184
3185 return rv;
3186 } else
3187 return write_super_imsm(st->sb, 1);
3188 }
3189 #endif
3190
3191 static int store_zero_imsm(struct supertype *st, int fd)
3192 {
3193 unsigned long long dsize;
3194 void *buf;
3195
3196 get_dev_size(fd, NULL, &dsize);
3197
3198 /* first block is stored on second to last sector of the disk */
3199 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
3200 return 1;
3201
3202 if (posix_memalign(&buf, 512, 512) != 0)
3203 return 1;
3204
3205 memset(buf, 0, 512);
3206 if (write(fd, buf, 512) != 512)
3207 return 1;
3208 return 0;
3209 }
3210
3211 static int imsm_bbm_log_size(struct imsm_super *mpb)
3212 {
3213 return __le32_to_cpu(mpb->bbm_log_size);
3214 }
3215
3216 #ifndef MDASSEMBLE
3217 static int validate_geometry_imsm_container(struct supertype *st, int level,
3218 int layout, int raiddisks, int chunk,
3219 unsigned long long size, char *dev,
3220 unsigned long long *freesize,
3221 int verbose)
3222 {
3223 int fd;
3224 unsigned long long ldsize;
3225 const struct imsm_orom *orom;
3226
3227 if (level != LEVEL_CONTAINER)
3228 return 0;
3229 if (!dev)
3230 return 1;
3231
3232 if (check_env("IMSM_NO_PLATFORM"))
3233 orom = NULL;
3234 else
3235 orom = find_imsm_orom();
3236 if (orom && raiddisks > orom->tds) {
3237 if (verbose)
3238 fprintf(stderr, Name ": %d exceeds maximum number of"
3239 " platform supported disks: %d\n",
3240 raiddisks, orom->tds);
3241 return 0;
3242 }
3243
3244 fd = open(dev, O_RDONLY|O_EXCL, 0);
3245 if (fd < 0) {
3246 if (verbose)
3247 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
3248 dev, strerror(errno));
3249 return 0;
3250 }
3251 if (!get_dev_size(fd, dev, &ldsize)) {
3252 close(fd);
3253 return 0;
3254 }
3255 close(fd);
3256
3257 *freesize = avail_size_imsm(st, ldsize >> 9);
3258
3259 return 1;
3260 }
3261
3262 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
3263 {
3264 const unsigned long long base_start = e[*idx].start;
3265 unsigned long long end = base_start + e[*idx].size;
3266 int i;
3267
3268 if (base_start == end)
3269 return 0;
3270
3271 *idx = *idx + 1;
3272 for (i = *idx; i < num_extents; i++) {
3273 /* extend overlapping extents */
3274 if (e[i].start >= base_start &&
3275 e[i].start <= end) {
3276 if (e[i].size == 0)
3277 return 0;
3278 if (e[i].start + e[i].size > end)
3279 end = e[i].start + e[i].size;
3280 } else if (e[i].start > end) {
3281 *idx = i;
3282 break;
3283 }
3284 }
3285
3286 return end - base_start;
3287 }
3288
3289 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
3290 {
3291 /* build a composite disk with all known extents and generate a new
3292 * 'maxsize' given the "all disks in an array must share a common start
3293 * offset" constraint
3294 */
3295 struct extent *e = calloc(sum_extents, sizeof(*e));
3296 struct dl *dl;
3297 int i, j;
3298 int start_extent;
3299 unsigned long long pos;
3300 unsigned long long start = 0;
3301 unsigned long long maxsize;
3302 unsigned long reserve;
3303
3304 if (!e)
3305 return ~0ULL; /* error */
3306
3307 /* coalesce and sort all extents. also, check to see if we need to
3308 * reserve space between member arrays
3309 */
3310 j = 0;
3311 for (dl = super->disks; dl; dl = dl->next) {
3312 if (!dl->e)
3313 continue;
3314 for (i = 0; i < dl->extent_cnt; i++)
3315 e[j++] = dl->e[i];
3316 }
3317 qsort(e, sum_extents, sizeof(*e), cmp_extent);
3318
3319 /* merge extents */
3320 i = 0;
3321 j = 0;
3322 while (i < sum_extents) {
3323 e[j].start = e[i].start;
3324 e[j].size = find_size(e, &i, sum_extents);
3325 j++;
3326 if (e[j-1].size == 0)
3327 break;
3328 }
3329
3330 pos = 0;
3331 maxsize = 0;
3332 start_extent = 0;
3333 i = 0;
3334 do {
3335 unsigned long long esize;
3336
3337 esize = e[i].start - pos;
3338 if (esize >= maxsize) {
3339 maxsize = esize;
3340 start = pos;
3341 start_extent = i;
3342 }
3343 pos = e[i].start + e[i].size;
3344 i++;
3345 } while (e[i-1].size);
3346 free(e);
3347
3348 if (start_extent > 0)
3349 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
3350 else
3351 reserve = 0;
3352
3353 if (maxsize < reserve)
3354 return ~0ULL;
3355
3356 super->create_offset = ~((__u32) 0);
3357 if (start + reserve > super->create_offset)
3358 return ~0ULL; /* start overflows create_offset */
3359 super->create_offset = start + reserve;
3360
3361 return maxsize - reserve;
3362 }
3363
3364 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
3365 {
3366 if (level < 0 || level == 6 || level == 4)
3367 return 0;
3368
3369 /* if we have an orom prevent invalid raid levels */
3370 if (orom)
3371 switch (level) {
3372 case 0: return imsm_orom_has_raid0(orom);
3373 case 1:
3374 if (raiddisks > 2)
3375 return imsm_orom_has_raid1e(orom);
3376 return imsm_orom_has_raid1(orom) && raiddisks == 2;
3377 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
3378 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
3379 }
3380 else
3381 return 1; /* not on an Intel RAID platform so anything goes */
3382
3383 return 0;
3384 }
3385
3386 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3387 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3388 * FIX ME add ahci details
3389 */
3390 static int validate_geometry_imsm_volume(struct supertype *st, int level,
3391 int layout, int raiddisks, int chunk,
3392 unsigned long long size, char *dev,
3393 unsigned long long *freesize,
3394 int verbose)
3395 {
3396 struct stat stb;
3397 struct intel_super *super = st->sb;
3398 struct imsm_super *mpb = super->anchor;
3399 struct dl *dl;
3400 unsigned long long pos = 0;
3401 unsigned long long maxsize;
3402 struct extent *e;
3403 int i;
3404
3405 /* We must have the container info already read in. */
3406 if (!super)
3407 return 0;
3408
3409 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
3410 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3411 level, raiddisks, raiddisks > 1 ? "s" : "");
3412 return 0;
3413 }
3414 if (super->orom && level != 1 &&
3415 !imsm_orom_has_chunk(super->orom, chunk)) {
3416 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
3417 return 0;
3418 }
3419 if (layout != imsm_level_to_layout(level)) {
3420 if (level == 5)
3421 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3422 else if (level == 10)
3423 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3424 else
3425 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3426 layout, level);
3427 return 0;
3428 }
3429
3430 if (!dev) {
3431 /* General test: make sure there is space for
3432 * 'raiddisks' device extents of size 'size' at a given
3433 * offset
3434 */
3435 unsigned long long minsize = size;
3436 unsigned long long start_offset = ~0ULL;
3437 int dcnt = 0;
3438 if (minsize == 0)
3439 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
3440 for (dl = super->disks; dl ; dl = dl->next) {
3441 int found = 0;
3442
3443 pos = 0;
3444 i = 0;
3445 e = get_extents(super, dl);
3446 if (!e) continue;
3447 do {
3448 unsigned long long esize;
3449 esize = e[i].start - pos;
3450 if (esize >= minsize)
3451 found = 1;
3452 if (found && start_offset == ~0ULL) {
3453 start_offset = pos;
3454 break;
3455 } else if (found && pos != start_offset) {
3456 found = 0;
3457 break;
3458 }
3459 pos = e[i].start + e[i].size;
3460 i++;
3461 } while (e[i-1].size);
3462 if (found)
3463 dcnt++;
3464 free(e);
3465 }
3466 if (dcnt < raiddisks) {
3467 if (verbose)
3468 fprintf(stderr, Name ": imsm: Not enough "
3469 "devices with space for this array "
3470 "(%d < %d)\n",
3471 dcnt, raiddisks);
3472 return 0;
3473 }
3474 return 1;
3475 }
3476
3477 /* This device must be a member of the set */
3478 if (stat(dev, &stb) < 0)
3479 return 0;
3480 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3481 return 0;
3482 for (dl = super->disks ; dl ; dl = dl->next) {
3483 if (dl->major == major(stb.st_rdev) &&
3484 dl->minor == minor(stb.st_rdev))
3485 break;
3486 }
3487 if (!dl) {
3488 if (verbose)
3489 fprintf(stderr, Name ": %s is not in the "
3490 "same imsm set\n", dev);
3491 return 0;
3492 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
3493 /* If a volume is present then the current creation attempt
3494 * cannot incorporate new spares because the orom may not
3495 * understand this configuration (all member disks must be
3496 * members of each array in the container).
3497 */
3498 fprintf(stderr, Name ": %s is a spare and a volume"
3499 " is already defined for this container\n", dev);
3500 fprintf(stderr, Name ": The option-rom requires all member"
3501 " disks to be a member of all volumes\n");
3502 return 0;
3503 }
3504
3505 /* retrieve the largest free space block */
3506 e = get_extents(super, dl);
3507 maxsize = 0;
3508 i = 0;
3509 if (e) {
3510 do {
3511 unsigned long long esize;
3512
3513 esize = e[i].start - pos;
3514 if (esize >= maxsize)
3515 maxsize = esize;
3516 pos = e[i].start + e[i].size;
3517 i++;
3518 } while (e[i-1].size);
3519 dl->e = e;
3520 dl->extent_cnt = i;
3521 } else {
3522 if (verbose)
3523 fprintf(stderr, Name ": unable to determine free space for: %s\n",
3524 dev);
3525 return 0;
3526 }
3527 if (maxsize < size) {
3528 if (verbose)
3529 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
3530 dev, maxsize, size);
3531 return 0;
3532 }
3533
3534 /* count total number of extents for merge */
3535 i = 0;
3536 for (dl = super->disks; dl; dl = dl->next)
3537 if (dl->e)
3538 i += dl->extent_cnt;
3539
3540 maxsize = merge_extents(super, i);
3541 if (maxsize < size) {
3542 if (verbose)
3543 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
3544 maxsize, size);
3545 return 0;
3546 } else if (maxsize == ~0ULL) {
3547 if (verbose)
3548 fprintf(stderr, Name ": failed to merge %d extents\n", i);
3549 return 0;
3550 }
3551
3552 *freesize = maxsize;
3553
3554 return 1;
3555 }
3556
3557 static int reserve_space(struct supertype *st, int raiddisks,
3558 unsigned long long size, int chunk,
3559 unsigned long long *freesize)
3560 {
3561 struct intel_super *super = st->sb;
3562 struct imsm_super *mpb = super->anchor;
3563 struct dl *dl;
3564 int i;
3565 int extent_cnt;
3566 struct extent *e;
3567 unsigned long long maxsize;
3568 unsigned long long minsize;
3569 int cnt;
3570 int used;
3571
3572 /* find the largest common start free region of the possible disks */
3573 used = 0;
3574 extent_cnt = 0;
3575 cnt = 0;
3576 for (dl = super->disks; dl; dl = dl->next) {
3577 dl->raiddisk = -1;
3578
3579 if (dl->index >= 0)
3580 used++;
3581
3582 /* don't activate new spares if we are orom constrained
3583 * and there is already a volume active in the container
3584 */
3585 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
3586 continue;
3587
3588 e = get_extents(super, dl);
3589 if (!e)
3590 continue;
3591 for (i = 1; e[i-1].size; i++)
3592 ;
3593 dl->e = e;
3594 dl->extent_cnt = i;
3595 extent_cnt += i;
3596 cnt++;
3597 }
3598
3599 maxsize = merge_extents(super, extent_cnt);
3600 minsize = size;
3601 if (size == 0)
3602 minsize = chunk;
3603
3604 if (cnt < raiddisks ||
3605 (super->orom && used && used != raiddisks) ||
3606 maxsize < minsize) {
3607 fprintf(stderr, Name ": not enough devices with space to create array.\n");
3608 return 0; /* No enough free spaces large enough */
3609 }
3610
3611 if (size == 0) {
3612 size = maxsize;
3613 if (chunk) {
3614 size /= chunk;
3615 size *= chunk;
3616 }
3617 }
3618
3619 cnt = 0;
3620 for (dl = super->disks; dl; dl = dl->next)
3621 if (dl->e)
3622 dl->raiddisk = cnt++;
3623
3624 *freesize = size;
3625
3626 return 1;
3627 }
3628
3629 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
3630 int raiddisks, int chunk, unsigned long long size,
3631 char *dev, unsigned long long *freesize,
3632 int verbose)
3633 {
3634 int fd, cfd;
3635 struct mdinfo *sra;
3636
3637 /* if given unused devices create a container
3638 * if given given devices in a container create a member volume
3639 */
3640 if (level == LEVEL_CONTAINER) {
3641 /* Must be a fresh device to add to a container */
3642 return validate_geometry_imsm_container(st, level, layout,
3643 raiddisks, chunk, size,
3644 dev, freesize,
3645 verbose);
3646 }
3647
3648 if (!dev) {
3649 if (st->sb && freesize) {
3650 /* we are being asked to automatically layout a
3651 * new volume based on the current contents of
3652 * the container. If the the parameters can be
3653 * satisfied reserve_space will record the disks,
3654 * start offset, and size of the volume to be
3655 * created. add_to_super and getinfo_super
3656 * detect when autolayout is in progress.
3657 */
3658 return reserve_space(st, raiddisks, size, chunk, freesize);
3659 }
3660 return 1;
3661 }
3662 if (st->sb) {
3663 /* creating in a given container */
3664 return validate_geometry_imsm_volume(st, level, layout,
3665 raiddisks, chunk, size,
3666 dev, freesize, verbose);
3667 }
3668
3669 /* limit creation to the following levels */
3670 if (!dev)
3671 switch (level) {
3672 case 0:
3673 case 1:
3674 case 10:
3675 case 5:
3676 break;
3677 default:
3678 return 1;
3679 }
3680
3681 /* This device needs to be a device in an 'imsm' container */
3682 fd = open(dev, O_RDONLY|O_EXCL, 0);
3683 if (fd >= 0) {
3684 if (verbose)
3685 fprintf(stderr,
3686 Name ": Cannot create this array on device %s\n",
3687 dev);
3688 close(fd);
3689 return 0;
3690 }
3691 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3692 if (verbose)
3693 fprintf(stderr, Name ": Cannot open %s: %s\n",
3694 dev, strerror(errno));
3695 return 0;
3696 }
3697 /* Well, it is in use by someone, maybe an 'imsm' container. */
3698 cfd = open_container(fd);
3699 if (cfd < 0) {
3700 close(fd);
3701 if (verbose)
3702 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3703 dev);
3704 return 0;
3705 }
3706 sra = sysfs_read(cfd, 0, GET_VERSION);
3707 close(fd);
3708 if (sra && sra->array.major_version == -1 &&
3709 strcmp(sra->text_version, "imsm") == 0) {
3710 /* This is a member of a imsm container. Load the container
3711 * and try to create a volume
3712 */
3713 struct intel_super *super;
3714
3715 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3716 st->sb = super;
3717 st->container_dev = fd2devnum(cfd);
3718 close(cfd);
3719 return validate_geometry_imsm_volume(st, level, layout,
3720 raiddisks, chunk,
3721 size, dev,
3722 freesize, verbose);
3723 }
3724 close(cfd);
3725 } else /* may belong to another container */
3726 return 0;
3727
3728 return 1;
3729 }
3730 #endif /* MDASSEMBLE */
3731
3732 static struct mdinfo *container_content_imsm(struct supertype *st)
3733 {
3734 /* Given a container loaded by load_super_imsm_all,
3735 * extract information about all the arrays into
3736 * an mdinfo tree.
3737 *
3738 * For each imsm_dev create an mdinfo, fill it in,
3739 * then look for matching devices in super->disks
3740 * and create appropriate device mdinfo.
3741 */
3742 struct intel_super *super = st->sb;
3743 struct imsm_super *mpb = super->anchor;
3744 struct mdinfo *rest = NULL;
3745 int i;
3746
3747 /* do not assemble arrays that might have bad blocks */
3748 if (imsm_bbm_log_size(super->anchor)) {
3749 fprintf(stderr, Name ": BBM log found in metadata. "
3750 "Cannot activate array(s).\n");
3751 return NULL;
3752 }
3753
3754 for (i = 0; i < mpb->num_raid_devs; i++) {
3755 struct imsm_dev *dev = get_imsm_dev(super, i);
3756 struct imsm_map *map = get_imsm_map(dev, 0);
3757 struct mdinfo *this;
3758 int slot;
3759
3760 /* do not publish arrays that are in the middle of an
3761 * unsupported migration
3762 */
3763 if (dev->vol.migr_state &&
3764 (migr_type(dev) == MIGR_GEN_MIGR ||
3765 migr_type(dev) == MIGR_STATE_CHANGE)) {
3766 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
3767 " unsupported migration in progress\n",
3768 dev->volume);
3769 continue;
3770 }
3771
3772 this = malloc(sizeof(*this));
3773 memset(this, 0, sizeof(*this));
3774 this->next = rest;
3775
3776 super->current_vol = i;
3777 getinfo_super_imsm_volume(st, this);
3778 for (slot = 0 ; slot < map->num_members; slot++) {
3779 struct mdinfo *info_d;
3780 struct dl *d;
3781 int idx;
3782 int skip;
3783 __u32 ord;
3784
3785 skip = 0;
3786 idx = get_imsm_disk_idx(dev, slot);
3787 ord = get_imsm_ord_tbl_ent(dev, slot);
3788 for (d = super->disks; d ; d = d->next)
3789 if (d->index == idx)
3790 break;
3791
3792 if (d == NULL)
3793 skip = 1;
3794 if (d && is_failed(&d->disk))
3795 skip = 1;
3796 if (ord & IMSM_ORD_REBUILD)
3797 skip = 1;
3798
3799 /*
3800 * if we skip some disks the array will be assmebled degraded;
3801 * reset resync start to avoid a dirty-degraded situation
3802 *
3803 * FIXME handle dirty degraded
3804 */
3805 if (skip && !dev->vol.dirty)
3806 this->resync_start = ~0ULL;
3807 if (skip)
3808 continue;
3809
3810 info_d = malloc(sizeof(*info_d));
3811 if (!info_d) {
3812 fprintf(stderr, Name ": failed to allocate disk"
3813 " for volume %.16s\n", dev->volume);
3814 free(this);
3815 this = rest;
3816 break;
3817 }
3818 memset(info_d, 0, sizeof(*info_d));
3819 info_d->next = this->devs;
3820 this->devs = info_d;
3821
3822 info_d->disk.number = d->index;
3823 info_d->disk.major = d->major;
3824 info_d->disk.minor = d->minor;
3825 info_d->disk.raid_disk = slot;
3826
3827 this->array.working_disks++;
3828
3829 info_d->events = __le32_to_cpu(mpb->generation_num);
3830 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3831 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3832 if (d->devname)
3833 strcpy(info_d->name, d->devname);
3834 }
3835 rest = this;
3836 }
3837
3838 return rest;
3839 }
3840
3841
3842 #ifndef MDASSEMBLE
3843 static int imsm_open_new(struct supertype *c, struct active_array *a,
3844 char *inst)
3845 {
3846 struct intel_super *super = c->sb;
3847 struct imsm_super *mpb = super->anchor;
3848
3849 if (atoi(inst) >= mpb->num_raid_devs) {
3850 fprintf(stderr, "%s: subarry index %d, out of range\n",
3851 __func__, atoi(inst));
3852 return -ENODEV;
3853 }
3854
3855 dprintf("imsm: open_new %s\n", inst);
3856 a->info.container_member = atoi(inst);
3857 return 0;
3858 }
3859
3860 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3861 {
3862 struct imsm_map *map = get_imsm_map(dev, 0);
3863
3864 if (!failed)
3865 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3866 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3867
3868 switch (get_imsm_raid_level(map)) {
3869 case 0:
3870 return IMSM_T_STATE_FAILED;
3871 break;
3872 case 1:
3873 if (failed < map->num_members)
3874 return IMSM_T_STATE_DEGRADED;
3875 else
3876 return IMSM_T_STATE_FAILED;
3877 break;
3878 case 10:
3879 {
3880 /**
3881 * check to see if any mirrors have failed, otherwise we
3882 * are degraded. Even numbered slots are mirrored on
3883 * slot+1
3884 */
3885 int i;
3886 /* gcc -Os complains that this is unused */
3887 int insync = insync;
3888
3889 for (i = 0; i < map->num_members; i++) {
3890 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3891 int idx = ord_to_idx(ord);
3892 struct imsm_disk *disk;
3893
3894 /* reset the potential in-sync count on even-numbered
3895 * slots. num_copies is always 2 for imsm raid10
3896 */
3897 if ((i & 1) == 0)
3898 insync = 2;
3899
3900 disk = get_imsm_disk(super, idx);
3901 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
3902 insync--;
3903
3904 /* no in-sync disks left in this mirror the
3905 * array has failed
3906 */
3907 if (insync == 0)
3908 return IMSM_T_STATE_FAILED;
3909 }
3910
3911 return IMSM_T_STATE_DEGRADED;
3912 }
3913 case 5:
3914 if (failed < 2)
3915 return IMSM_T_STATE_DEGRADED;
3916 else
3917 return IMSM_T_STATE_FAILED;
3918 break;
3919 default:
3920 break;
3921 }
3922
3923 return map->map_state;
3924 }
3925
3926 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3927 {
3928 int i;
3929 int failed = 0;
3930 struct imsm_disk *disk;
3931 struct imsm_map *map = get_imsm_map(dev, 0);
3932 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3933 __u32 ord;
3934 int idx;
3935
3936 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3937 * disks that are being rebuilt. New failures are recorded to
3938 * map[0]. So we look through all the disks we started with and
3939 * see if any failures are still present, or if any new ones
3940 * have arrived
3941 *
3942 * FIXME add support for online capacity expansion and
3943 * raid-level-migration
3944 */
3945 for (i = 0; i < prev->num_members; i++) {
3946 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
3947 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
3948 idx = ord_to_idx(ord);
3949
3950 disk = get_imsm_disk(super, idx);
3951 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
3952 failed++;
3953 }
3954
3955 return failed;
3956 }
3957
3958 static int is_resyncing(struct imsm_dev *dev)
3959 {
3960 struct imsm_map *migr_map;
3961
3962 if (!dev->vol.migr_state)
3963 return 0;
3964
3965 if (migr_type(dev) == MIGR_INIT ||
3966 migr_type(dev) == MIGR_REPAIR)
3967 return 1;
3968
3969 migr_map = get_imsm_map(dev, 1);
3970
3971 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
3972 return 1;
3973 else
3974 return 0;
3975 }
3976
3977 static int is_rebuilding(struct imsm_dev *dev)
3978 {
3979 struct imsm_map *migr_map;
3980
3981 if (!dev->vol.migr_state)
3982 return 0;
3983
3984 if (migr_type(dev) != MIGR_REBUILD)
3985 return 0;
3986
3987 migr_map = get_imsm_map(dev, 1);
3988
3989 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
3990 return 1;
3991 else
3992 return 0;
3993 }
3994
3995 /* return true if we recorded new information */
3996 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
3997 {
3998 __u32 ord;
3999 int slot;
4000 struct imsm_map *map;
4001
4002 /* new failures are always set in map[0] */
4003 map = get_imsm_map(dev, 0);
4004
4005 slot = get_imsm_disk_slot(map, idx);
4006 if (slot < 0)
4007 return 0;
4008
4009 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
4010 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
4011 return 0;
4012
4013 disk->status |= FAILED_DISK;
4014 disk->status &= ~CONFIGURED_DISK;
4015 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
4016 if (~map->failed_disk_num == 0)
4017 map->failed_disk_num = slot;
4018 return 1;
4019 }
4020
4021 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4022 {
4023 mark_failure(dev, disk, idx);
4024
4025 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
4026 return;
4027
4028 disk->scsi_id = __cpu_to_le32(~(__u32)0);
4029 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
4030 }
4031
4032 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4033 * states are handled in imsm_set_disk() with one exception, when a
4034 * resync is stopped due to a new failure this routine will set the
4035 * 'degraded' state for the array.
4036 */
4037 static int imsm_set_array_state(struct active_array *a, int consistent)
4038 {
4039 int inst = a->info.container_member;
4040 struct intel_super *super = a->container->sb;
4041 struct imsm_dev *dev = get_imsm_dev(super, inst);
4042 struct imsm_map *map = get_imsm_map(dev, 0);
4043 int failed = imsm_count_failed(super, dev);
4044 __u8 map_state = imsm_check_degraded(super, dev, failed);
4045
4046 /* before we activate this array handle any missing disks */
4047 if (consistent == 2 && super->missing) {
4048 struct dl *dl;
4049
4050 dprintf("imsm: mark missing\n");
4051 end_migration(dev, map_state);
4052 for (dl = super->missing; dl; dl = dl->next)
4053 mark_missing(dev, &dl->disk, dl->index);
4054 super->updates_pending++;
4055 }
4056
4057 if (consistent == 2 &&
4058 (!is_resync_complete(a) ||
4059 map_state != IMSM_T_STATE_NORMAL ||
4060 dev->vol.migr_state))
4061 consistent = 0;
4062
4063 if (is_resync_complete(a)) {
4064 /* complete intialization / resync,
4065 * recovery and interrupted recovery is completed in
4066 * ->set_disk
4067 */
4068 if (is_resyncing(dev)) {
4069 dprintf("imsm: mark resync done\n");
4070 end_migration(dev, map_state);
4071 super->updates_pending++;
4072 }
4073 } else if (!is_resyncing(dev) && !failed) {
4074 /* mark the start of the init process if nothing is failed */
4075 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
4076 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
4077 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
4078 else
4079 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
4080 super->updates_pending++;
4081 }
4082
4083 /* FIXME check if we can update curr_migr_unit from resync_start */
4084
4085 /* mark dirty / clean */
4086 if (dev->vol.dirty != !consistent) {
4087 dprintf("imsm: mark '%s' (%llu)\n",
4088 consistent ? "clean" : "dirty", a->resync_start);
4089 if (consistent)
4090 dev->vol.dirty = 0;
4091 else
4092 dev->vol.dirty = 1;
4093 super->updates_pending++;
4094 }
4095 return consistent;
4096 }
4097
4098 static void imsm_set_disk(struct active_array *a, int n, int state)
4099 {
4100 int inst = a->info.container_member;
4101 struct intel_super *super = a->container->sb;
4102 struct imsm_dev *dev = get_imsm_dev(super, inst);
4103 struct imsm_map *map = get_imsm_map(dev, 0);
4104 struct imsm_disk *disk;
4105 int failed;
4106 __u32 ord;
4107 __u8 map_state;
4108
4109 if (n > map->num_members)
4110 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
4111 n, map->num_members - 1);
4112
4113 if (n < 0)
4114 return;
4115
4116 dprintf("imsm: set_disk %d:%x\n", n, state);
4117
4118 ord = get_imsm_ord_tbl_ent(dev, n);
4119 disk = get_imsm_disk(super, ord_to_idx(ord));
4120
4121 /* check for new failures */
4122 if (state & DS_FAULTY) {
4123 if (mark_failure(dev, disk, ord_to_idx(ord)))
4124 super->updates_pending++;
4125 }
4126
4127 /* check if in_sync */
4128 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
4129 struct imsm_map *migr_map = get_imsm_map(dev, 1);
4130
4131 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
4132 super->updates_pending++;
4133 }
4134
4135 failed = imsm_count_failed(super, dev);
4136 map_state = imsm_check_degraded(super, dev, failed);
4137
4138 /* check if recovery complete, newly degraded, or failed */
4139 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
4140 end_migration(dev, map_state);
4141 map = get_imsm_map(dev, 0);
4142 map->failed_disk_num = ~0;
4143 super->updates_pending++;
4144 } else if (map_state == IMSM_T_STATE_DEGRADED &&
4145 map->map_state != map_state &&
4146 !dev->vol.migr_state) {
4147 dprintf("imsm: mark degraded\n");
4148 map->map_state = map_state;
4149 super->updates_pending++;
4150 } else if (map_state == IMSM_T_STATE_FAILED &&
4151 map->map_state != map_state) {
4152 dprintf("imsm: mark failed\n");
4153 end_migration(dev, map_state);
4154 super->updates_pending++;
4155 }
4156 }
4157
4158 static int store_imsm_mpb(int fd, struct imsm_super *mpb)
4159 {
4160 void *buf = mpb;
4161 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
4162 unsigned long long dsize;
4163 unsigned long long sectors;
4164
4165 get_dev_size(fd, NULL, &dsize);
4166
4167 if (mpb_size > 512) {
4168 /* -1 to account for anchor */
4169 sectors = mpb_sectors(mpb) - 1;
4170
4171 /* write the extended mpb to the sectors preceeding the anchor */
4172 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
4173 return 1;
4174
4175 if (write(fd, buf + 512, 512 * sectors) != 512 * sectors)
4176 return 1;
4177 }
4178
4179 /* first block is stored on second to last sector of the disk */
4180 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
4181 return 1;
4182
4183 if (write(fd, buf, 512) != 512)
4184 return 1;
4185
4186 return 0;
4187 }
4188
4189 static void imsm_sync_metadata(struct supertype *container)
4190 {
4191 struct intel_super *super = container->sb;
4192
4193 if (!super->updates_pending)
4194 return;
4195
4196 write_super_imsm(super, 0);
4197
4198 super->updates_pending = 0;
4199 }
4200
4201 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
4202 {
4203 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4204 int i = get_imsm_disk_idx(dev, idx);
4205 struct dl *dl;
4206
4207 for (dl = super->disks; dl; dl = dl->next)
4208 if (dl->index == i)
4209 break;
4210
4211 if (dl && is_failed(&dl->disk))
4212 dl = NULL;
4213
4214 if (dl)
4215 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
4216
4217 return dl;
4218 }
4219
4220 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
4221 struct active_array *a, int activate_new)
4222 {
4223 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4224 int idx = get_imsm_disk_idx(dev, slot);
4225 struct imsm_super *mpb = super->anchor;
4226 struct imsm_map *map;
4227 unsigned long long pos;
4228 struct mdinfo *d;
4229 struct extent *ex;
4230 int i, j;
4231 int found;
4232 __u32 array_start;
4233 __u32 array_end;
4234 struct dl *dl;
4235
4236 for (dl = super->disks; dl; dl = dl->next) {
4237 /* If in this array, skip */
4238 for (d = a->info.devs ; d ; d = d->next)
4239 if (d->state_fd >= 0 &&
4240 d->disk.major == dl->major &&
4241 d->disk.minor == dl->minor) {
4242 dprintf("%x:%x already in array\n", dl->major, dl->minor);
4243 break;
4244 }
4245 if (d)
4246 continue;
4247
4248 /* skip in use or failed drives */
4249 if (is_failed(&dl->disk) || idx == dl->index ||
4250 dl->index == -2) {
4251 dprintf("%x:%x status (failed: %d index: %d)\n",
4252 dl->major, dl->minor, is_failed(&dl->disk), idx);
4253 continue;
4254 }
4255
4256 /* skip pure spares when we are looking for partially
4257 * assimilated drives
4258 */
4259 if (dl->index == -1 && !activate_new)
4260 continue;
4261
4262 /* Does this unused device have the requisite free space?
4263 * It needs to be able to cover all member volumes
4264 */
4265 ex = get_extents(super, dl);
4266 if (!ex) {
4267 dprintf("cannot get extents\n");
4268 continue;
4269 }
4270 for (i = 0; i < mpb->num_raid_devs; i++) {
4271 dev = get_imsm_dev(super, i);
4272 map = get_imsm_map(dev, 0);
4273
4274 /* check if this disk is already a member of
4275 * this array
4276 */
4277 if (get_imsm_disk_slot(map, dl->index) >= 0)
4278 continue;
4279
4280 found = 0;
4281 j = 0;
4282 pos = 0;
4283 array_start = __le32_to_cpu(map->pba_of_lba0);
4284 array_end = array_start +
4285 __le32_to_cpu(map->blocks_per_member) - 1;
4286
4287 do {
4288 /* check that we can start at pba_of_lba0 with
4289 * blocks_per_member of space
4290 */
4291 if (array_start >= pos && array_end < ex[j].start) {
4292 found = 1;
4293 break;
4294 }
4295 pos = ex[j].start + ex[j].size;
4296 j++;
4297 } while (ex[j-1].size);
4298
4299 if (!found)
4300 break;
4301 }
4302
4303 free(ex);
4304 if (i < mpb->num_raid_devs) {
4305 dprintf("%x:%x does not have %u to %u available\n",
4306 dl->major, dl->minor, array_start, array_end);
4307 /* No room */
4308 continue;
4309 }
4310 return dl;
4311 }
4312
4313 return dl;
4314 }
4315
4316 static struct mdinfo *imsm_activate_spare(struct active_array *a,
4317 struct metadata_update **updates)
4318 {
4319 /**
4320 * Find a device with unused free space and use it to replace a
4321 * failed/vacant region in an array. We replace failed regions one a
4322 * array at a time. The result is that a new spare disk will be added
4323 * to the first failed array and after the monitor has finished
4324 * propagating failures the remainder will be consumed.
4325 *
4326 * FIXME add a capability for mdmon to request spares from another
4327 * container.
4328 */
4329
4330 struct intel_super *super = a->container->sb;
4331 int inst = a->info.container_member;
4332 struct imsm_dev *dev = get_imsm_dev(super, inst);
4333 struct imsm_map *map = get_imsm_map(dev, 0);
4334 int failed = a->info.array.raid_disks;
4335 struct mdinfo *rv = NULL;
4336 struct mdinfo *d;
4337 struct mdinfo *di;
4338 struct metadata_update *mu;
4339 struct dl *dl;
4340 struct imsm_update_activate_spare *u;
4341 int num_spares = 0;
4342 int i;
4343
4344 for (d = a->info.devs ; d ; d = d->next) {
4345 if ((d->curr_state & DS_FAULTY) &&
4346 d->state_fd >= 0)
4347 /* wait for Removal to happen */
4348 return NULL;
4349 if (d->state_fd >= 0)
4350 failed--;
4351 }
4352
4353 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4354 inst, failed, a->info.array.raid_disks, a->info.array.level);
4355 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
4356 return NULL;
4357
4358 /* For each slot, if it is not working, find a spare */
4359 for (i = 0; i < a->info.array.raid_disks; i++) {
4360 for (d = a->info.devs ; d ; d = d->next)
4361 if (d->disk.raid_disk == i)
4362 break;
4363 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
4364 if (d && (d->state_fd >= 0))
4365 continue;
4366
4367 /*
4368 * OK, this device needs recovery. Try to re-add the
4369 * previous occupant of this slot, if this fails see if
4370 * we can continue the assimilation of a spare that was
4371 * partially assimilated, finally try to activate a new
4372 * spare.
4373 */
4374 dl = imsm_readd(super, i, a);
4375 if (!dl)
4376 dl = imsm_add_spare(super, i, a, 0);
4377 if (!dl)
4378 dl = imsm_add_spare(super, i, a, 1);
4379 if (!dl)
4380 continue;
4381
4382 /* found a usable disk with enough space */
4383 di = malloc(sizeof(*di));
4384 if (!di)
4385 continue;
4386 memset(di, 0, sizeof(*di));
4387
4388 /* dl->index will be -1 in the case we are activating a
4389 * pristine spare. imsm_process_update() will create a
4390 * new index in this case. Once a disk is found to be
4391 * failed in all member arrays it is kicked from the
4392 * metadata
4393 */
4394 di->disk.number = dl->index;
4395
4396 /* (ab)use di->devs to store a pointer to the device
4397 * we chose
4398 */
4399 di->devs = (struct mdinfo *) dl;
4400
4401 di->disk.raid_disk = i;
4402 di->disk.major = dl->major;
4403 di->disk.minor = dl->minor;
4404 di->disk.state = 0;
4405 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
4406 di->component_size = a->info.component_size;
4407 di->container_member = inst;
4408 super->random = random32();
4409 di->next = rv;
4410 rv = di;
4411 num_spares++;
4412 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
4413 i, di->data_offset);
4414
4415 break;
4416 }
4417
4418 if (!rv)
4419 /* No spares found */
4420 return rv;
4421 /* Now 'rv' has a list of devices to return.
4422 * Create a metadata_update record to update the
4423 * disk_ord_tbl for the array
4424 */
4425 mu = malloc(sizeof(*mu));
4426 if (mu) {
4427 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
4428 if (mu->buf == NULL) {
4429 free(mu);
4430 mu = NULL;
4431 }
4432 }
4433 if (!mu) {
4434 while (rv) {
4435 struct mdinfo *n = rv->next;
4436
4437 free(rv);
4438 rv = n;
4439 }
4440 return NULL;
4441 }
4442
4443 mu->space = NULL;
4444 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
4445 mu->next = *updates;
4446 u = (struct imsm_update_activate_spare *) mu->buf;
4447
4448 for (di = rv ; di ; di = di->next) {
4449 u->type = update_activate_spare;
4450 u->dl = (struct dl *) di->devs;
4451 di->devs = NULL;
4452 u->slot = di->disk.raid_disk;
4453 u->array = inst;
4454 u->next = u + 1;
4455 u++;
4456 }
4457 (u-1)->next = NULL;
4458 *updates = mu;
4459
4460 return rv;
4461 }
4462
4463 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
4464 {
4465 struct imsm_dev *dev = get_imsm_dev(super, idx);
4466 struct imsm_map *map = get_imsm_map(dev, 0);
4467 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
4468 struct disk_info *inf = get_disk_info(u);
4469 struct imsm_disk *disk;
4470 int i;
4471 int j;
4472
4473 for (i = 0; i < map->num_members; i++) {
4474 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4475 for (j = 0; j < new_map->num_members; j++)
4476 if (serialcmp(disk->serial, inf[j].serial) == 0)
4477 return 1;
4478 }
4479
4480 return 0;
4481 }
4482
4483 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
4484
4485 static void imsm_process_update(struct supertype *st,
4486 struct metadata_update *update)
4487 {
4488 /**
4489 * crack open the metadata_update envelope to find the update record
4490 * update can be one of:
4491 * update_activate_spare - a spare device has replaced a failed
4492 * device in an array, update the disk_ord_tbl. If this disk is
4493 * present in all member arrays then also clear the SPARE_DISK
4494 * flag
4495 */
4496 struct intel_super *super = st->sb;
4497 struct imsm_super *mpb;
4498 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4499
4500 /* update requires a larger buf but the allocation failed */
4501 if (super->next_len && !super->next_buf) {
4502 super->next_len = 0;
4503 return;
4504 }
4505
4506 if (super->next_buf) {
4507 memcpy(super->next_buf, super->buf, super->len);
4508 free(super->buf);
4509 super->len = super->next_len;
4510 super->buf = super->next_buf;
4511
4512 super->next_len = 0;
4513 super->next_buf = NULL;
4514 }
4515
4516 mpb = super->anchor;
4517
4518 switch (type) {
4519 case update_activate_spare: {
4520 struct imsm_update_activate_spare *u = (void *) update->buf;
4521 struct imsm_dev *dev = get_imsm_dev(super, u->array);
4522 struct imsm_map *map = get_imsm_map(dev, 0);
4523 struct imsm_map *migr_map;
4524 struct active_array *a;
4525 struct imsm_disk *disk;
4526 __u8 to_state;
4527 struct dl *dl;
4528 unsigned int found;
4529 int failed;
4530 int victim = get_imsm_disk_idx(dev, u->slot);
4531 int i;
4532
4533 for (dl = super->disks; dl; dl = dl->next)
4534 if (dl == u->dl)
4535 break;
4536
4537 if (!dl) {
4538 fprintf(stderr, "error: imsm_activate_spare passed "
4539 "an unknown disk (index: %d)\n",
4540 u->dl->index);
4541 return;
4542 }
4543
4544 super->updates_pending++;
4545
4546 /* count failures (excluding rebuilds and the victim)
4547 * to determine map[0] state
4548 */
4549 failed = 0;
4550 for (i = 0; i < map->num_members; i++) {
4551 if (i == u->slot)
4552 continue;
4553 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4554 if (!disk || is_failed(disk))
4555 failed++;
4556 }
4557
4558 /* adding a pristine spare, assign a new index */
4559 if (dl->index < 0) {
4560 dl->index = super->anchor->num_disks;
4561 super->anchor->num_disks++;
4562 }
4563 disk = &dl->disk;
4564 disk->status |= CONFIGURED_DISK;
4565 disk->status &= ~SPARE_DISK;
4566
4567 /* mark rebuild */
4568 to_state = imsm_check_degraded(super, dev, failed);
4569 map->map_state = IMSM_T_STATE_DEGRADED;
4570 migrate(dev, to_state, MIGR_REBUILD);
4571 migr_map = get_imsm_map(dev, 1);
4572 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
4573 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
4574
4575 /* update the family_num to mark a new container
4576 * generation, being careful to record the existing
4577 * family_num in orig_family_num to clean up after
4578 * earlier mdadm versions that neglected to set it.
4579 */
4580 if (mpb->orig_family_num == 0)
4581 mpb->orig_family_num = mpb->family_num;
4582 mpb->family_num += super->random;
4583
4584 /* count arrays using the victim in the metadata */
4585 found = 0;
4586 for (a = st->arrays; a ; a = a->next) {
4587 dev = get_imsm_dev(super, a->info.container_member);
4588 map = get_imsm_map(dev, 0);
4589
4590 if (get_imsm_disk_slot(map, victim) >= 0)
4591 found++;
4592 }
4593
4594 /* delete the victim if it is no longer being
4595 * utilized anywhere
4596 */
4597 if (!found) {
4598 struct dl **dlp;
4599
4600 /* We know that 'manager' isn't touching anything,
4601 * so it is safe to delete
4602 */
4603 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
4604 if ((*dlp)->index == victim)
4605 break;
4606
4607 /* victim may be on the missing list */
4608 if (!*dlp)
4609 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
4610 if ((*dlp)->index == victim)
4611 break;
4612 imsm_delete(super, dlp, victim);
4613 }
4614 break;
4615 }
4616 case update_create_array: {
4617 /* someone wants to create a new array, we need to be aware of
4618 * a few races/collisions:
4619 * 1/ 'Create' called by two separate instances of mdadm
4620 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4621 * devices that have since been assimilated via
4622 * activate_spare.
4623 * In the event this update can not be carried out mdadm will
4624 * (FIX ME) notice that its update did not take hold.
4625 */
4626 struct imsm_update_create_array *u = (void *) update->buf;
4627 struct intel_dev *dv;
4628 struct imsm_dev *dev;
4629 struct imsm_map *map, *new_map;
4630 unsigned long long start, end;
4631 unsigned long long new_start, new_end;
4632 int i;
4633 struct disk_info *inf;
4634 struct dl *dl;
4635
4636 /* handle racing creates: first come first serve */
4637 if (u->dev_idx < mpb->num_raid_devs) {
4638 dprintf("%s: subarray %d already defined\n",
4639 __func__, u->dev_idx);
4640 goto create_error;
4641 }
4642
4643 /* check update is next in sequence */
4644 if (u->dev_idx != mpb->num_raid_devs) {
4645 dprintf("%s: can not create array %d expected index %d\n",
4646 __func__, u->dev_idx, mpb->num_raid_devs);
4647 goto create_error;
4648 }
4649
4650 new_map = get_imsm_map(&u->dev, 0);
4651 new_start = __le32_to_cpu(new_map->pba_of_lba0);
4652 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
4653 inf = get_disk_info(u);
4654
4655 /* handle activate_spare versus create race:
4656 * check to make sure that overlapping arrays do not include
4657 * overalpping disks
4658 */
4659 for (i = 0; i < mpb->num_raid_devs; i++) {
4660 dev = get_imsm_dev(super, i);
4661 map = get_imsm_map(dev, 0);
4662 start = __le32_to_cpu(map->pba_of_lba0);
4663 end = start + __le32_to_cpu(map->blocks_per_member);
4664 if ((new_start >= start && new_start <= end) ||
4665 (start >= new_start && start <= new_end))
4666 /* overlap */;
4667 else
4668 continue;
4669
4670 if (disks_overlap(super, i, u)) {
4671 dprintf("%s: arrays overlap\n", __func__);
4672 goto create_error;
4673 }
4674 }
4675
4676 /* check that prepare update was successful */
4677 if (!update->space) {
4678 dprintf("%s: prepare update failed\n", __func__);
4679 goto create_error;
4680 }
4681
4682 /* check that all disks are still active before committing
4683 * changes. FIXME: could we instead handle this by creating a
4684 * degraded array? That's probably not what the user expects,
4685 * so better to drop this update on the floor.
4686 */
4687 for (i = 0; i < new_map->num_members; i++) {
4688 dl = serial_to_dl(inf[i].serial, super);
4689 if (!dl) {
4690 dprintf("%s: disk disappeared\n", __func__);
4691 goto create_error;
4692 }
4693 }
4694
4695 super->updates_pending++;
4696
4697 /* convert spares to members and fixup ord_tbl */
4698 for (i = 0; i < new_map->num_members; i++) {
4699 dl = serial_to_dl(inf[i].serial, super);
4700 if (dl->index == -1) {
4701 dl->index = mpb->num_disks;
4702 mpb->num_disks++;
4703 dl->disk.status |= CONFIGURED_DISK;
4704 dl->disk.status &= ~SPARE_DISK;
4705 }
4706 set_imsm_ord_tbl_ent(new_map, i, dl->index);
4707 }
4708
4709 dv = update->space;
4710 dev = dv->dev;
4711 update->space = NULL;
4712 imsm_copy_dev(dev, &u->dev);
4713 dv->index = u->dev_idx;
4714 dv->next = super->devlist;
4715 super->devlist = dv;
4716 mpb->num_raid_devs++;
4717
4718 imsm_update_version_info(super);
4719 break;
4720 create_error:
4721 /* mdmon knows how to release update->space, but not
4722 * ((struct intel_dev *) update->space)->dev
4723 */
4724 if (update->space) {
4725 dv = update->space;
4726 free(dv->dev);
4727 }
4728 break;
4729 }
4730 case update_add_disk:
4731
4732 /* we may be able to repair some arrays if disks are
4733 * being added */
4734 if (super->add) {
4735 struct active_array *a;
4736
4737 super->updates_pending++;
4738 for (a = st->arrays; a; a = a->next)
4739 a->check_degraded = 1;
4740 }
4741 /* add some spares to the metadata */
4742 while (super->add) {
4743 struct dl *al;
4744
4745 al = super->add;
4746 super->add = al->next;
4747 al->next = super->disks;
4748 super->disks = al;
4749 dprintf("%s: added %x:%x\n",
4750 __func__, al->major, al->minor);
4751 }
4752
4753 break;
4754 }
4755 }
4756
4757 static void imsm_prepare_update(struct supertype *st,
4758 struct metadata_update *update)
4759 {
4760 /**
4761 * Allocate space to hold new disk entries, raid-device entries or a new
4762 * mpb if necessary. The manager synchronously waits for updates to
4763 * complete in the monitor, so new mpb buffers allocated here can be
4764 * integrated by the monitor thread without worrying about live pointers
4765 * in the manager thread.
4766 */
4767 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4768 struct intel_super *super = st->sb;
4769 struct imsm_super *mpb = super->anchor;
4770 size_t buf_len;
4771 size_t len = 0;
4772
4773 switch (type) {
4774 case update_create_array: {
4775 struct imsm_update_create_array *u = (void *) update->buf;
4776 struct intel_dev *dv;
4777 struct imsm_dev *dev = &u->dev;
4778 struct imsm_map *map = get_imsm_map(dev, 0);
4779 struct dl *dl;
4780 struct disk_info *inf;
4781 int i;
4782 int activate = 0;
4783
4784 inf = get_disk_info(u);
4785 len = sizeof_imsm_dev(dev, 1);
4786 /* allocate a new super->devlist entry */
4787 dv = malloc(sizeof(*dv));
4788 if (dv) {
4789 dv->dev = malloc(len);
4790 if (dv->dev)
4791 update->space = dv;
4792 else {
4793 free(dv);
4794 update->space = NULL;
4795 }
4796 }
4797
4798 /* count how many spares will be converted to members */
4799 for (i = 0; i < map->num_members; i++) {
4800 dl = serial_to_dl(inf[i].serial, super);
4801 if (!dl) {
4802 /* hmm maybe it failed?, nothing we can do about
4803 * it here
4804 */
4805 continue;
4806 }
4807 if (count_memberships(dl, super) == 0)
4808 activate++;
4809 }
4810 len += activate * sizeof(struct imsm_disk);
4811 break;
4812 default:
4813 break;
4814 }
4815 }
4816
4817 /* check if we need a larger metadata buffer */
4818 if (super->next_buf)
4819 buf_len = super->next_len;
4820 else
4821 buf_len = super->len;
4822
4823 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4824 /* ok we need a larger buf than what is currently allocated
4825 * if this allocation fails process_update will notice that
4826 * ->next_len is set and ->next_buf is NULL
4827 */
4828 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4829 if (super->next_buf)
4830 free(super->next_buf);
4831
4832 super->next_len = buf_len;
4833 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
4834 memset(super->next_buf, 0, buf_len);
4835 else
4836 super->next_buf = NULL;
4837 }
4838 }
4839
4840 /* must be called while manager is quiesced */
4841 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4842 {
4843 struct imsm_super *mpb = super->anchor;
4844 struct dl *iter;
4845 struct imsm_dev *dev;
4846 struct imsm_map *map;
4847 int i, j, num_members;
4848 __u32 ord;
4849
4850 dprintf("%s: deleting device[%d] from imsm_super\n",
4851 __func__, index);
4852
4853 /* shift all indexes down one */
4854 for (iter = super->disks; iter; iter = iter->next)
4855 if (iter->index > index)
4856 iter->index--;
4857 for (iter = super->missing; iter; iter = iter->next)
4858 if (iter->index > index)
4859 iter->index--;
4860
4861 for (i = 0; i < mpb->num_raid_devs; i++) {
4862 dev = get_imsm_dev(super, i);
4863 map = get_imsm_map(dev, 0);
4864 num_members = map->num_members;
4865 for (j = 0; j < num_members; j++) {
4866 /* update ord entries being careful not to propagate
4867 * ord-flags to the first map
4868 */
4869 ord = get_imsm_ord_tbl_ent(dev, j);
4870
4871 if (ord_to_idx(ord) <= index)
4872 continue;
4873
4874 map = get_imsm_map(dev, 0);
4875 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4876 map = get_imsm_map(dev, 1);
4877 if (map)
4878 set_imsm_ord_tbl_ent(map, j, ord - 1);
4879 }
4880 }
4881
4882 mpb->num_disks--;
4883 super->updates_pending++;
4884 if (*dlp) {
4885 struct dl *dl = *dlp;
4886
4887 *dlp = (*dlp)->next;
4888 __free_imsm_disk(dl);
4889 }
4890 }
4891 #endif /* MDASSEMBLE */
4892
4893 struct superswitch super_imsm = {
4894 #ifndef MDASSEMBLE
4895 .examine_super = examine_super_imsm,
4896 .brief_examine_super = brief_examine_super_imsm,
4897 .brief_examine_subarrays = brief_examine_subarrays_imsm,
4898 .export_examine_super = export_examine_super_imsm,
4899 .detail_super = detail_super_imsm,
4900 .brief_detail_super = brief_detail_super_imsm,
4901 .write_init_super = write_init_super_imsm,
4902 .validate_geometry = validate_geometry_imsm,
4903 .add_to_super = add_to_super_imsm,
4904 .detail_platform = detail_platform_imsm,
4905 #endif
4906 .match_home = match_home_imsm,
4907 .uuid_from_super= uuid_from_super_imsm,
4908 .getinfo_super = getinfo_super_imsm,
4909 .update_super = update_super_imsm,
4910
4911 .avail_size = avail_size_imsm,
4912
4913 .compare_super = compare_super_imsm,
4914
4915 .load_super = load_super_imsm,
4916 .init_super = init_super_imsm,
4917 .store_super = store_zero_imsm,
4918 .free_super = free_super_imsm,
4919 .match_metadata_desc = match_metadata_desc_imsm,
4920 .container_content = container_content_imsm,
4921 .default_layout = imsm_level_to_layout,
4922
4923 .external = 1,
4924 .name = "imsm",
4925
4926 #ifndef MDASSEMBLE
4927 /* for mdmon */
4928 .open_new = imsm_open_new,
4929 .load_super = load_super_imsm,
4930 .set_array_state= imsm_set_array_state,
4931 .set_disk = imsm_set_disk,
4932 .sync_metadata = imsm_sync_metadata,
4933 .activate_spare = imsm_activate_spare,
4934 .process_update = imsm_process_update,
4935 .prepare_update = imsm_prepare_update,
4936 #endif /* MDASSEMBLE */
4937 };
Unnamed repository; edit this file 'description' to name the repository.