]> 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
Switch open_subarray to use the new load_container
[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 unsigned 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 current_vol; /* index of raid device undergoing creation */
247 __u32 create_offset; /* common start for 'current_vol' */
248 __u32 random; /* random data for seeding new family numbers */
249 struct intel_dev *devlist;
250 struct dl {
251 struct dl *next;
252 int index;
253 __u8 serial[MAX_RAID_SERIAL_LEN];
254 int major, minor;
255 char *devname;
256 struct imsm_disk disk;
257 int fd;
258 int extent_cnt;
259 struct extent *e; /* for determining freespace @ create */
260 int raiddisk; /* slot to fill in autolayout */
261 } *disks;
262 struct dl *add; /* list of disks to add while mdmon active */
263 struct dl *missing; /* disks removed while we weren't looking */
264 struct bbm_log *bbm_log;
265 const char *hba; /* device path of the raid controller for this metadata */
266 const struct imsm_orom *orom; /* platform firmware support */
267 struct intel_super *next; /* (temp) list for disambiguating family_num */
268 };
269
270 struct intel_disk {
271 struct imsm_disk disk;
272 #define IMSM_UNKNOWN_OWNER (-1)
273 int owner;
274 struct intel_disk *next;
275 };
276
277 struct extent {
278 unsigned long long start, size;
279 };
280
281 /* definition of messages passed to imsm_process_update */
282 enum imsm_update_type {
283 update_activate_spare,
284 update_create_array,
285 update_kill_array,
286 update_rename_array,
287 update_add_disk,
288 };
289
290 struct imsm_update_activate_spare {
291 enum imsm_update_type type;
292 struct dl *dl;
293 int slot;
294 int array;
295 struct imsm_update_activate_spare *next;
296 };
297
298 struct disk_info {
299 __u8 serial[MAX_RAID_SERIAL_LEN];
300 };
301
302 struct imsm_update_create_array {
303 enum imsm_update_type type;
304 int dev_idx;
305 struct imsm_dev dev;
306 };
307
308 struct imsm_update_kill_array {
309 enum imsm_update_type type;
310 int dev_idx;
311 };
312
313 struct imsm_update_rename_array {
314 enum imsm_update_type type;
315 __u8 name[MAX_RAID_SERIAL_LEN];
316 int dev_idx;
317 };
318
319 struct imsm_update_add_disk {
320 enum imsm_update_type type;
321 };
322
323 static struct supertype *match_metadata_desc_imsm(char *arg)
324 {
325 struct supertype *st;
326
327 if (strcmp(arg, "imsm") != 0 &&
328 strcmp(arg, "default") != 0
329 )
330 return NULL;
331
332 st = malloc(sizeof(*st));
333 if (!st)
334 return NULL;
335 memset(st, 0, sizeof(*st));
336 st->container_dev = NoMdDev;
337 st->ss = &super_imsm;
338 st->max_devs = IMSM_MAX_DEVICES;
339 st->minor_version = 0;
340 st->sb = NULL;
341 return st;
342 }
343
344 #ifndef MDASSEMBLE
345 static __u8 *get_imsm_version(struct imsm_super *mpb)
346 {
347 return &mpb->sig[MPB_SIG_LEN];
348 }
349 #endif
350
351 /* retrieve a disk directly from the anchor when the anchor is known to be
352 * up-to-date, currently only at load time
353 */
354 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
355 {
356 if (index >= mpb->num_disks)
357 return NULL;
358 return &mpb->disk[index];
359 }
360
361 /* retrieve a disk from the parsed metadata */
362 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
363 {
364 struct dl *d;
365
366 for (d = super->disks; d; d = d->next)
367 if (d->index == index)
368 return &d->disk;
369
370 return NULL;
371 }
372
373 /* generate a checksum directly from the anchor when the anchor is known to be
374 * up-to-date, currently only at load or write_super after coalescing
375 */
376 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
377 {
378 __u32 end = mpb->mpb_size / sizeof(end);
379 __u32 *p = (__u32 *) mpb;
380 __u32 sum = 0;
381
382 while (end--) {
383 sum += __le32_to_cpu(*p);
384 p++;
385 }
386
387 return sum - __le32_to_cpu(mpb->check_sum);
388 }
389
390 static size_t sizeof_imsm_map(struct imsm_map *map)
391 {
392 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
393 }
394
395 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
396 {
397 struct imsm_map *map = &dev->vol.map[0];
398
399 if (second_map && !dev->vol.migr_state)
400 return NULL;
401 else if (second_map) {
402 void *ptr = map;
403
404 return ptr + sizeof_imsm_map(map);
405 } else
406 return map;
407
408 }
409
410 /* return the size of the device.
411 * migr_state increases the returned size if map[0] were to be duplicated
412 */
413 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
414 {
415 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
416 sizeof_imsm_map(get_imsm_map(dev, 0));
417
418 /* migrating means an additional map */
419 if (dev->vol.migr_state)
420 size += sizeof_imsm_map(get_imsm_map(dev, 1));
421 else if (migr_state)
422 size += sizeof_imsm_map(get_imsm_map(dev, 0));
423
424 return size;
425 }
426
427 #ifndef MDASSEMBLE
428 /* retrieve disk serial number list from a metadata update */
429 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
430 {
431 void *u = update;
432 struct disk_info *inf;
433
434 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
435 sizeof_imsm_dev(&update->dev, 0);
436
437 return inf;
438 }
439 #endif
440
441 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
442 {
443 int offset;
444 int i;
445 void *_mpb = mpb;
446
447 if (index >= mpb->num_raid_devs)
448 return NULL;
449
450 /* devices start after all disks */
451 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
452
453 for (i = 0; i <= index; i++)
454 if (i == index)
455 return _mpb + offset;
456 else
457 offset += sizeof_imsm_dev(_mpb + offset, 0);
458
459 return NULL;
460 }
461
462 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
463 {
464 struct intel_dev *dv;
465
466 if (index >= super->anchor->num_raid_devs)
467 return NULL;
468 for (dv = super->devlist; dv; dv = dv->next)
469 if (dv->index == index)
470 return dv->dev;
471 return NULL;
472 }
473
474 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, int slot)
475 {
476 struct imsm_map *map;
477
478 if (dev->vol.migr_state)
479 map = get_imsm_map(dev, 1);
480 else
481 map = get_imsm_map(dev, 0);
482
483 /* top byte identifies disk under rebuild */
484 return __le32_to_cpu(map->disk_ord_tbl[slot]);
485 }
486
487 #define ord_to_idx(ord) (((ord) << 8) >> 8)
488 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot)
489 {
490 __u32 ord = get_imsm_ord_tbl_ent(dev, slot);
491
492 return ord_to_idx(ord);
493 }
494
495 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
496 {
497 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
498 }
499
500 static int get_imsm_disk_slot(struct imsm_map *map, unsigned idx)
501 {
502 int slot;
503 __u32 ord;
504
505 for (slot = 0; slot < map->num_members; slot++) {
506 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
507 if (ord_to_idx(ord) == idx)
508 return slot;
509 }
510
511 return -1;
512 }
513
514 static int get_imsm_raid_level(struct imsm_map *map)
515 {
516 if (map->raid_level == 1) {
517 if (map->num_members == 2)
518 return 1;
519 else
520 return 10;
521 }
522
523 return map->raid_level;
524 }
525
526 static int cmp_extent(const void *av, const void *bv)
527 {
528 const struct extent *a = av;
529 const struct extent *b = bv;
530 if (a->start < b->start)
531 return -1;
532 if (a->start > b->start)
533 return 1;
534 return 0;
535 }
536
537 static int count_memberships(struct dl *dl, struct intel_super *super)
538 {
539 int memberships = 0;
540 int i;
541
542 for (i = 0; i < super->anchor->num_raid_devs; i++) {
543 struct imsm_dev *dev = get_imsm_dev(super, i);
544 struct imsm_map *map = get_imsm_map(dev, 0);
545
546 if (get_imsm_disk_slot(map, dl->index) >= 0)
547 memberships++;
548 }
549
550 return memberships;
551 }
552
553 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
554 {
555 /* find a list of used extents on the given physical device */
556 struct extent *rv, *e;
557 int i;
558 int memberships = count_memberships(dl, super);
559 __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
560
561 rv = malloc(sizeof(struct extent) * (memberships + 1));
562 if (!rv)
563 return NULL;
564 e = rv;
565
566 for (i = 0; i < super->anchor->num_raid_devs; i++) {
567 struct imsm_dev *dev = get_imsm_dev(super, i);
568 struct imsm_map *map = get_imsm_map(dev, 0);
569
570 if (get_imsm_disk_slot(map, dl->index) >= 0) {
571 e->start = __le32_to_cpu(map->pba_of_lba0);
572 e->size = __le32_to_cpu(map->blocks_per_member);
573 e++;
574 }
575 }
576 qsort(rv, memberships, sizeof(*rv), cmp_extent);
577
578 /* determine the start of the metadata
579 * when no raid devices are defined use the default
580 * ...otherwise allow the metadata to truncate the value
581 * as is the case with older versions of imsm
582 */
583 if (memberships) {
584 struct extent *last = &rv[memberships - 1];
585 __u32 remainder;
586
587 remainder = __le32_to_cpu(dl->disk.total_blocks) -
588 (last->start + last->size);
589 /* round down to 1k block to satisfy precision of the kernel
590 * 'size' interface
591 */
592 remainder &= ~1UL;
593 /* make sure remainder is still sane */
594 if (remainder < (unsigned)ROUND_UP(super->len, 512) >> 9)
595 remainder = ROUND_UP(super->len, 512) >> 9;
596 if (reservation > remainder)
597 reservation = remainder;
598 }
599 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
600 e->size = 0;
601 return rv;
602 }
603
604 /* try to determine how much space is reserved for metadata from
605 * the last get_extents() entry, otherwise fallback to the
606 * default
607 */
608 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
609 {
610 struct extent *e;
611 int i;
612 __u32 rv;
613
614 /* for spares just return a minimal reservation which will grow
615 * once the spare is picked up by an array
616 */
617 if (dl->index == -1)
618 return MPB_SECTOR_CNT;
619
620 e = get_extents(super, dl);
621 if (!e)
622 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
623
624 /* scroll to last entry */
625 for (i = 0; e[i].size; i++)
626 continue;
627
628 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
629
630 free(e);
631
632 return rv;
633 }
634
635 static int is_spare(struct imsm_disk *disk)
636 {
637 return (disk->status & SPARE_DISK) == SPARE_DISK;
638 }
639
640 static int is_configured(struct imsm_disk *disk)
641 {
642 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
643 }
644
645 static int is_failed(struct imsm_disk *disk)
646 {
647 return (disk->status & FAILED_DISK) == FAILED_DISK;
648 }
649
650 #ifndef MDASSEMBLE
651 static __u64 blocks_per_migr_unit(struct imsm_dev *dev);
652
653 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
654 {
655 __u64 sz;
656 int slot, i;
657 struct imsm_map *map = get_imsm_map(dev, 0);
658 __u32 ord;
659
660 printf("\n");
661 printf("[%.16s]:\n", dev->volume);
662 printf(" UUID : %s\n", uuid);
663 printf(" RAID Level : %d\n", get_imsm_raid_level(map));
664 printf(" Members : %d\n", map->num_members);
665 printf(" Slots : [");
666 for (i = 0; i < map->num_members; i++) {
667 ord = get_imsm_ord_tbl_ent(dev, i);
668 printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
669 }
670 printf("]\n");
671 slot = get_imsm_disk_slot(map, disk_idx);
672 if (slot >= 0) {
673 ord = get_imsm_ord_tbl_ent(dev, slot);
674 printf(" This Slot : %d%s\n", slot,
675 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
676 } else
677 printf(" This Slot : ?\n");
678 sz = __le32_to_cpu(dev->size_high);
679 sz <<= 32;
680 sz += __le32_to_cpu(dev->size_low);
681 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
682 human_size(sz * 512));
683 sz = __le32_to_cpu(map->blocks_per_member);
684 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
685 human_size(sz * 512));
686 printf(" Sector Offset : %u\n",
687 __le32_to_cpu(map->pba_of_lba0));
688 printf(" Num Stripes : %u\n",
689 __le32_to_cpu(map->num_data_stripes));
690 printf(" Chunk Size : %u KiB\n",
691 __le16_to_cpu(map->blocks_per_strip) / 2);
692 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
693 printf(" Migrate State : ");
694 if (dev->vol.migr_state) {
695 if (migr_type(dev) == MIGR_INIT)
696 printf("initialize\n");
697 else if (migr_type(dev) == MIGR_REBUILD)
698 printf("rebuild\n");
699 else if (migr_type(dev) == MIGR_VERIFY)
700 printf("check\n");
701 else if (migr_type(dev) == MIGR_GEN_MIGR)
702 printf("general migration\n");
703 else if (migr_type(dev) == MIGR_STATE_CHANGE)
704 printf("state change\n");
705 else if (migr_type(dev) == MIGR_REPAIR)
706 printf("repair\n");
707 else
708 printf("<unknown:%d>\n", migr_type(dev));
709 } else
710 printf("idle\n");
711 printf(" Map State : %s", map_state_str[map->map_state]);
712 if (dev->vol.migr_state) {
713 struct imsm_map *map = get_imsm_map(dev, 1);
714
715 printf(" <-- %s", map_state_str[map->map_state]);
716 printf("\n Checkpoint : %u (%llu)",
717 __le32_to_cpu(dev->vol.curr_migr_unit),
718 (unsigned long long)blocks_per_migr_unit(dev));
719 }
720 printf("\n");
721 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
722 }
723
724 static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved)
725 {
726 struct imsm_disk *disk = __get_imsm_disk(mpb, index);
727 char str[MAX_RAID_SERIAL_LEN + 1];
728 __u64 sz;
729
730 if (index < 0 || !disk)
731 return;
732
733 printf("\n");
734 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
735 printf(" Disk%02d Serial : %s\n", index, str);
736 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
737 is_configured(disk) ? " active" : "",
738 is_failed(disk) ? " failed" : "");
739 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
740 sz = __le32_to_cpu(disk->total_blocks) - reserved;
741 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
742 human_size(sz * 512));
743 }
744
745 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map);
746
747 static void examine_super_imsm(struct supertype *st, char *homehost)
748 {
749 struct intel_super *super = st->sb;
750 struct imsm_super *mpb = super->anchor;
751 char str[MAX_SIGNATURE_LENGTH];
752 int i;
753 struct mdinfo info;
754 char nbuf[64];
755 __u32 sum;
756 __u32 reserved = imsm_reserved_sectors(super, super->disks);
757
758
759 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
760 printf(" Magic : %s\n", str);
761 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
762 printf(" Version : %s\n", get_imsm_version(mpb));
763 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
764 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
765 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
766 getinfo_super_imsm(st, &info, NULL);
767 fname_from_uuid(st, &info, nbuf, ':');
768 printf(" UUID : %s\n", nbuf + 5);
769 sum = __le32_to_cpu(mpb->check_sum);
770 printf(" Checksum : %08x %s\n", sum,
771 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
772 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
773 printf(" Disks : %d\n", mpb->num_disks);
774 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
775 print_imsm_disk(mpb, super->disks->index, reserved);
776 if (super->bbm_log) {
777 struct bbm_log *log = super->bbm_log;
778
779 printf("\n");
780 printf("Bad Block Management Log:\n");
781 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
782 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
783 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
784 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
785 printf(" First Spare : %llx\n",
786 (unsigned long long) __le64_to_cpu(log->first_spare_lba));
787 }
788 for (i = 0; i < mpb->num_raid_devs; i++) {
789 struct mdinfo info;
790 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
791
792 super->current_vol = i;
793 getinfo_super_imsm(st, &info, NULL);
794 fname_from_uuid(st, &info, nbuf, ':');
795 print_imsm_dev(dev, nbuf + 5, super->disks->index);
796 }
797 for (i = 0; i < mpb->num_disks; i++) {
798 if (i == super->disks->index)
799 continue;
800 print_imsm_disk(mpb, i, reserved);
801 }
802 }
803
804 static void brief_examine_super_imsm(struct supertype *st, int verbose)
805 {
806 /* We just write a generic IMSM ARRAY entry */
807 struct mdinfo info;
808 char nbuf[64];
809 struct intel_super *super = st->sb;
810
811 if (!super->anchor->num_raid_devs) {
812 printf("ARRAY metadata=imsm\n");
813 return;
814 }
815
816 getinfo_super_imsm(st, &info, NULL);
817 fname_from_uuid(st, &info, nbuf, ':');
818 printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
819 }
820
821 static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
822 {
823 /* We just write a generic IMSM ARRAY entry */
824 struct mdinfo info;
825 char nbuf[64];
826 char nbuf1[64];
827 struct intel_super *super = st->sb;
828 int i;
829
830 if (!super->anchor->num_raid_devs)
831 return;
832
833 getinfo_super_imsm(st, &info, NULL);
834 fname_from_uuid(st, &info, nbuf, ':');
835 for (i = 0; i < super->anchor->num_raid_devs; i++) {
836 struct imsm_dev *dev = get_imsm_dev(super, i);
837
838 super->current_vol = i;
839 getinfo_super_imsm(st, &info, NULL);
840 fname_from_uuid(st, &info, nbuf1, ':');
841 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
842 dev->volume, nbuf + 5, i, nbuf1 + 5);
843 }
844 }
845
846 static void export_examine_super_imsm(struct supertype *st)
847 {
848 struct intel_super *super = st->sb;
849 struct imsm_super *mpb = super->anchor;
850 struct mdinfo info;
851 char nbuf[64];
852
853 getinfo_super_imsm(st, &info, NULL);
854 fname_from_uuid(st, &info, nbuf, ':');
855 printf("MD_METADATA=imsm\n");
856 printf("MD_LEVEL=container\n");
857 printf("MD_UUID=%s\n", nbuf+5);
858 printf("MD_DEVICES=%u\n", mpb->num_disks);
859 }
860
861 static void detail_super_imsm(struct supertype *st, char *homehost)
862 {
863 struct mdinfo info;
864 char nbuf[64];
865
866 getinfo_super_imsm(st, &info, NULL);
867 fname_from_uuid(st, &info, nbuf, ':');
868 printf("\n UUID : %s\n", nbuf + 5);
869 }
870
871 static void brief_detail_super_imsm(struct supertype *st)
872 {
873 struct mdinfo info;
874 char nbuf[64];
875 getinfo_super_imsm(st, &info, NULL);
876 fname_from_uuid(st, &info, nbuf, ':');
877 printf(" UUID=%s", nbuf + 5);
878 }
879
880 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
881 static void fd2devname(int fd, char *name);
882
883 static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
884 {
885 /* dump an unsorted list of devices attached to ahci, as well as
886 * non-connected ports
887 */
888 int hba_len = strlen(hba_path) + 1;
889 struct dirent *ent;
890 DIR *dir;
891 char *path = NULL;
892 int err = 0;
893 unsigned long port_mask = (1 << port_count) - 1;
894
895 if (port_count > (int)sizeof(port_mask) * 8) {
896 if (verbose)
897 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
898 return 2;
899 }
900
901 /* scroll through /sys/dev/block looking for devices attached to
902 * this hba
903 */
904 dir = opendir("/sys/dev/block");
905 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
906 int fd;
907 char model[64];
908 char vendor[64];
909 char buf[1024];
910 int major, minor;
911 char *device;
912 char *c;
913 int port;
914 int type;
915
916 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
917 continue;
918 path = devt_to_devpath(makedev(major, minor));
919 if (!path)
920 continue;
921 if (!path_attached_to_hba(path, hba_path)) {
922 free(path);
923 path = NULL;
924 continue;
925 }
926
927 /* retrieve the scsi device type */
928 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
929 if (verbose)
930 fprintf(stderr, Name ": failed to allocate 'device'\n");
931 err = 2;
932 break;
933 }
934 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
935 if (load_sys(device, buf) != 0) {
936 if (verbose)
937 fprintf(stderr, Name ": failed to read device type for %s\n",
938 path);
939 err = 2;
940 free(device);
941 break;
942 }
943 type = strtoul(buf, NULL, 10);
944
945 /* if it's not a disk print the vendor and model */
946 if (!(type == 0 || type == 7 || type == 14)) {
947 vendor[0] = '\0';
948 model[0] = '\0';
949 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
950 if (load_sys(device, buf) == 0) {
951 strncpy(vendor, buf, sizeof(vendor));
952 vendor[sizeof(vendor) - 1] = '\0';
953 c = (char *) &vendor[sizeof(vendor) - 1];
954 while (isspace(*c) || *c == '\0')
955 *c-- = '\0';
956
957 }
958 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
959 if (load_sys(device, buf) == 0) {
960 strncpy(model, buf, sizeof(model));
961 model[sizeof(model) - 1] = '\0';
962 c = (char *) &model[sizeof(model) - 1];
963 while (isspace(*c) || *c == '\0')
964 *c-- = '\0';
965 }
966
967 if (vendor[0] && model[0])
968 sprintf(buf, "%.64s %.64s", vendor, model);
969 else
970 switch (type) { /* numbers from hald/linux/device.c */
971 case 1: sprintf(buf, "tape"); break;
972 case 2: sprintf(buf, "printer"); break;
973 case 3: sprintf(buf, "processor"); break;
974 case 4:
975 case 5: sprintf(buf, "cdrom"); break;
976 case 6: sprintf(buf, "scanner"); break;
977 case 8: sprintf(buf, "media_changer"); break;
978 case 9: sprintf(buf, "comm"); break;
979 case 12: sprintf(buf, "raid"); break;
980 default: sprintf(buf, "unknown");
981 }
982 } else
983 buf[0] = '\0';
984 free(device);
985
986 /* chop device path to 'host%d' and calculate the port number */
987 c = strchr(&path[hba_len], '/');
988 if (!c) {
989 if (verbose)
990 fprintf(stderr, Name ": %s - invalid path name\n", path + hba_len);
991 err = 2;
992 break;
993 }
994 *c = '\0';
995 if (sscanf(&path[hba_len], "host%d", &port) == 1)
996 port -= host_base;
997 else {
998 if (verbose) {
999 *c = '/'; /* repair the full string */
1000 fprintf(stderr, Name ": failed to determine port number for %s\n",
1001 path);
1002 }
1003 err = 2;
1004 break;
1005 }
1006
1007 /* mark this port as used */
1008 port_mask &= ~(1 << port);
1009
1010 /* print out the device information */
1011 if (buf[0]) {
1012 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
1013 continue;
1014 }
1015
1016 fd = dev_open(ent->d_name, O_RDONLY);
1017 if (fd < 0)
1018 printf(" Port%d : - disk info unavailable -\n", port);
1019 else {
1020 fd2devname(fd, buf);
1021 printf(" Port%d : %s", port, buf);
1022 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
1023 printf(" (%s)\n", buf);
1024 else
1025 printf("()\n");
1026 }
1027 close(fd);
1028 free(path);
1029 path = NULL;
1030 }
1031 if (path)
1032 free(path);
1033 if (dir)
1034 closedir(dir);
1035 if (err == 0) {
1036 int i;
1037
1038 for (i = 0; i < port_count; i++)
1039 if (port_mask & (1 << i))
1040 printf(" Port%d : - no device attached -\n", i);
1041 }
1042
1043 return err;
1044 }
1045
1046 static int detail_platform_imsm(int verbose, int enumerate_only)
1047 {
1048 /* There are two components to imsm platform support, the ahci SATA
1049 * controller and the option-rom. To find the SATA controller we
1050 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1051 * controller with the Intel vendor id is present. This approach
1052 * allows mdadm to leverage the kernel's ahci detection logic, with the
1053 * caveat that if ahci.ko is not loaded mdadm will not be able to
1054 * detect platform raid capabilities. The option-rom resides in a
1055 * platform "Adapter ROM". We scan for its signature to retrieve the
1056 * platform capabilities. If raid support is disabled in the BIOS the
1057 * option-rom capability structure will not be available.
1058 */
1059 const struct imsm_orom *orom;
1060 struct sys_dev *list, *hba;
1061 DIR *dir;
1062 struct dirent *ent;
1063 const char *hba_path;
1064 int host_base = 0;
1065 int port_count = 0;
1066
1067 if (enumerate_only) {
1068 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1069 return 0;
1070 return 2;
1071 }
1072
1073 list = find_driver_devices("pci", "ahci");
1074 for (hba = list; hba; hba = hba->next)
1075 if (devpath_to_vendor(hba->path) == 0x8086)
1076 break;
1077
1078 if (!hba) {
1079 if (verbose)
1080 fprintf(stderr, Name ": unable to find active ahci controller\n");
1081 free_sys_dev(&list);
1082 return 2;
1083 } else if (verbose)
1084 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
1085 hba_path = hba->path;
1086 hba->path = NULL;
1087 free_sys_dev(&list);
1088
1089 orom = find_imsm_orom();
1090 if (!orom) {
1091 if (verbose)
1092 fprintf(stderr, Name ": imsm option-rom not found\n");
1093 return 2;
1094 }
1095
1096 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1097 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
1098 orom->hotfix_ver, orom->build);
1099 printf(" RAID Levels :%s%s%s%s%s\n",
1100 imsm_orom_has_raid0(orom) ? " raid0" : "",
1101 imsm_orom_has_raid1(orom) ? " raid1" : "",
1102 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
1103 imsm_orom_has_raid10(orom) ? " raid10" : "",
1104 imsm_orom_has_raid5(orom) ? " raid5" : "");
1105 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1106 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
1107 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
1108 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
1109 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
1110 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
1111 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
1112 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
1113 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
1114 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
1115 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
1116 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
1117 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
1118 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
1119 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
1120 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
1121 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1122 printf(" Max Disks : %d\n", orom->tds);
1123 printf(" Max Volumes : %d\n", orom->vpa);
1124 printf(" I/O Controller : %s\n", hba_path);
1125
1126 /* find the smallest scsi host number to determine a port number base */
1127 dir = opendir(hba_path);
1128 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1129 int host;
1130
1131 if (sscanf(ent->d_name, "host%d", &host) != 1)
1132 continue;
1133 if (port_count == 0)
1134 host_base = host;
1135 else if (host < host_base)
1136 host_base = host;
1137
1138 if (host + 1 > port_count + host_base)
1139 port_count = host + 1 - host_base;
1140
1141 }
1142 if (dir)
1143 closedir(dir);
1144
1145 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1146 host_base, verbose) != 0) {
1147 if (verbose)
1148 fprintf(stderr, Name ": failed to enumerate ports\n");
1149 return 2;
1150 }
1151
1152 return 0;
1153 }
1154 #endif
1155
1156 static int match_home_imsm(struct supertype *st, char *homehost)
1157 {
1158 /* the imsm metadata format does not specify any host
1159 * identification information. We return -1 since we can never
1160 * confirm nor deny whether a given array is "meant" for this
1161 * host. We rely on compare_super and the 'family_num' fields to
1162 * exclude member disks that do not belong, and we rely on
1163 * mdadm.conf to specify the arrays that should be assembled.
1164 * Auto-assembly may still pick up "foreign" arrays.
1165 */
1166
1167 return -1;
1168 }
1169
1170 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1171 {
1172 /* The uuid returned here is used for:
1173 * uuid to put into bitmap file (Create, Grow)
1174 * uuid for backup header when saving critical section (Grow)
1175 * comparing uuids when re-adding a device into an array
1176 * In these cases the uuid required is that of the data-array,
1177 * not the device-set.
1178 * uuid to recognise same set when adding a missing device back
1179 * to an array. This is a uuid for the device-set.
1180 *
1181 * For each of these we can make do with a truncated
1182 * or hashed uuid rather than the original, as long as
1183 * everyone agrees.
1184 * In each case the uuid required is that of the data-array,
1185 * not the device-set.
1186 */
1187 /* imsm does not track uuid's so we synthesis one using sha1 on
1188 * - The signature (Which is constant for all imsm array, but no matter)
1189 * - the orig_family_num of the container
1190 * - the index number of the volume
1191 * - the 'serial' number of the volume.
1192 * Hopefully these are all constant.
1193 */
1194 struct intel_super *super = st->sb;
1195
1196 char buf[20];
1197 struct sha1_ctx ctx;
1198 struct imsm_dev *dev = NULL;
1199 __u32 family_num;
1200
1201 /* some mdadm versions failed to set ->orig_family_num, in which
1202 * case fall back to ->family_num. orig_family_num will be
1203 * fixed up with the first metadata update.
1204 */
1205 family_num = super->anchor->orig_family_num;
1206 if (family_num == 0)
1207 family_num = super->anchor->family_num;
1208 sha1_init_ctx(&ctx);
1209 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1210 sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
1211 if (super->current_vol >= 0)
1212 dev = get_imsm_dev(super, super->current_vol);
1213 if (dev) {
1214 __u32 vol = super->current_vol;
1215 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1216 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1217 }
1218 sha1_finish_ctx(&ctx, buf);
1219 memcpy(uuid, buf, 4*4);
1220 }
1221
1222 #if 0
1223 static void
1224 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1225 {
1226 __u8 *v = get_imsm_version(mpb);
1227 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1228 char major[] = { 0, 0, 0 };
1229 char minor[] = { 0 ,0, 0 };
1230 char patch[] = { 0, 0, 0 };
1231 char *ver_parse[] = { major, minor, patch };
1232 int i, j;
1233
1234 i = j = 0;
1235 while (*v != '\0' && v < end) {
1236 if (*v != '.' && j < 2)
1237 ver_parse[i][j++] = *v;
1238 else {
1239 i++;
1240 j = 0;
1241 }
1242 v++;
1243 }
1244
1245 *m = strtol(minor, NULL, 0);
1246 *p = strtol(patch, NULL, 0);
1247 }
1248 #endif
1249
1250 static __u32 migr_strip_blocks_resync(struct imsm_dev *dev)
1251 {
1252 /* migr_strip_size when repairing or initializing parity */
1253 struct imsm_map *map = get_imsm_map(dev, 0);
1254 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1255
1256 switch (get_imsm_raid_level(map)) {
1257 case 5:
1258 case 10:
1259 return chunk;
1260 default:
1261 return 128*1024 >> 9;
1262 }
1263 }
1264
1265 static __u32 migr_strip_blocks_rebuild(struct imsm_dev *dev)
1266 {
1267 /* migr_strip_size when rebuilding a degraded disk, no idea why
1268 * this is different than migr_strip_size_resync(), but it's good
1269 * to be compatible
1270 */
1271 struct imsm_map *map = get_imsm_map(dev, 1);
1272 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1273
1274 switch (get_imsm_raid_level(map)) {
1275 case 1:
1276 case 10:
1277 if (map->num_members % map->num_domains == 0)
1278 return 128*1024 >> 9;
1279 else
1280 return chunk;
1281 case 5:
1282 return max((__u32) 64*1024 >> 9, chunk);
1283 default:
1284 return 128*1024 >> 9;
1285 }
1286 }
1287
1288 static __u32 num_stripes_per_unit_resync(struct imsm_dev *dev)
1289 {
1290 struct imsm_map *lo = get_imsm_map(dev, 0);
1291 struct imsm_map *hi = get_imsm_map(dev, 1);
1292 __u32 lo_chunk = __le32_to_cpu(lo->blocks_per_strip);
1293 __u32 hi_chunk = __le32_to_cpu(hi->blocks_per_strip);
1294
1295 return max((__u32) 1, hi_chunk / lo_chunk);
1296 }
1297
1298 static __u32 num_stripes_per_unit_rebuild(struct imsm_dev *dev)
1299 {
1300 struct imsm_map *lo = get_imsm_map(dev, 0);
1301 int level = get_imsm_raid_level(lo);
1302
1303 if (level == 1 || level == 10) {
1304 struct imsm_map *hi = get_imsm_map(dev, 1);
1305
1306 return hi->num_domains;
1307 } else
1308 return num_stripes_per_unit_resync(dev);
1309 }
1310
1311 static __u8 imsm_num_data_members(struct imsm_dev *dev)
1312 {
1313 /* named 'imsm_' because raid0, raid1 and raid10
1314 * counter-intuitively have the same number of data disks
1315 */
1316 struct imsm_map *map = get_imsm_map(dev, 0);
1317
1318 switch (get_imsm_raid_level(map)) {
1319 case 0:
1320 case 1:
1321 case 10:
1322 return map->num_members;
1323 case 5:
1324 return map->num_members - 1;
1325 default:
1326 dprintf("%s: unsupported raid level\n", __func__);
1327 return 0;
1328 }
1329 }
1330
1331 static __u32 parity_segment_depth(struct imsm_dev *dev)
1332 {
1333 struct imsm_map *map = get_imsm_map(dev, 0);
1334 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1335
1336 switch(get_imsm_raid_level(map)) {
1337 case 1:
1338 case 10:
1339 return chunk * map->num_domains;
1340 case 5:
1341 return chunk * map->num_members;
1342 default:
1343 return chunk;
1344 }
1345 }
1346
1347 static __u32 map_migr_block(struct imsm_dev *dev, __u32 block)
1348 {
1349 struct imsm_map *map = get_imsm_map(dev, 1);
1350 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1351 __u32 strip = block / chunk;
1352
1353 switch (get_imsm_raid_level(map)) {
1354 case 1:
1355 case 10: {
1356 __u32 vol_strip = (strip * map->num_domains) + 1;
1357 __u32 vol_stripe = vol_strip / map->num_members;
1358
1359 return vol_stripe * chunk + block % chunk;
1360 } case 5: {
1361 __u32 stripe = strip / (map->num_members - 1);
1362
1363 return stripe * chunk + block % chunk;
1364 }
1365 default:
1366 return 0;
1367 }
1368 }
1369
1370 static __u64 blocks_per_migr_unit(struct imsm_dev *dev)
1371 {
1372 /* calculate the conversion factor between per member 'blocks'
1373 * (md/{resync,rebuild}_start) and imsm migration units, return
1374 * 0 for the 'not migrating' and 'unsupported migration' cases
1375 */
1376 if (!dev->vol.migr_state)
1377 return 0;
1378
1379 switch (migr_type(dev)) {
1380 case MIGR_VERIFY:
1381 case MIGR_REPAIR:
1382 case MIGR_INIT: {
1383 struct imsm_map *map = get_imsm_map(dev, 0);
1384 __u32 stripes_per_unit;
1385 __u32 blocks_per_unit;
1386 __u32 parity_depth;
1387 __u32 migr_chunk;
1388 __u32 block_map;
1389 __u32 block_rel;
1390 __u32 segment;
1391 __u32 stripe;
1392 __u8 disks;
1393
1394 /* yes, this is really the translation of migr_units to
1395 * per-member blocks in the 'resync' case
1396 */
1397 stripes_per_unit = num_stripes_per_unit_resync(dev);
1398 migr_chunk = migr_strip_blocks_resync(dev);
1399 disks = imsm_num_data_members(dev);
1400 blocks_per_unit = stripes_per_unit * migr_chunk * disks;
1401 stripe = __le32_to_cpu(map->blocks_per_strip) * disks;
1402 segment = blocks_per_unit / stripe;
1403 block_rel = blocks_per_unit - segment * stripe;
1404 parity_depth = parity_segment_depth(dev);
1405 block_map = map_migr_block(dev, block_rel);
1406 return block_map + parity_depth * segment;
1407 }
1408 case MIGR_REBUILD: {
1409 __u32 stripes_per_unit;
1410 __u32 migr_chunk;
1411
1412 stripes_per_unit = num_stripes_per_unit_rebuild(dev);
1413 migr_chunk = migr_strip_blocks_rebuild(dev);
1414 return migr_chunk * stripes_per_unit;
1415 }
1416 case MIGR_GEN_MIGR:
1417 case MIGR_STATE_CHANGE:
1418 default:
1419 return 0;
1420 }
1421 }
1422
1423 static int imsm_level_to_layout(int level)
1424 {
1425 switch (level) {
1426 case 0:
1427 case 1:
1428 return 0;
1429 case 5:
1430 case 6:
1431 return ALGORITHM_LEFT_ASYMMETRIC;
1432 case 10:
1433 return 0x102;
1434 }
1435 return UnSet;
1436 }
1437
1438 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info, char *dmap)
1439 {
1440 struct intel_super *super = st->sb;
1441 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1442 struct imsm_map *map = get_imsm_map(dev, 0);
1443 struct dl *dl;
1444 char *devname;
1445 int map_disks = info->array.raid_disks;
1446
1447 for (dl = super->disks; dl; dl = dl->next)
1448 if (dl->raiddisk == info->disk.raid_disk)
1449 break;
1450 info->container_member = super->current_vol;
1451 info->array.raid_disks = map->num_members;
1452 info->array.level = get_imsm_raid_level(map);
1453 info->array.layout = imsm_level_to_layout(info->array.level);
1454 info->array.md_minor = -1;
1455 info->array.ctime = 0;
1456 info->array.utime = 0;
1457 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1458 info->array.state = !dev->vol.dirty;
1459 info->custom_array_size = __le32_to_cpu(dev->size_high);
1460 info->custom_array_size <<= 32;
1461 info->custom_array_size |= __le32_to_cpu(dev->size_low);
1462
1463 info->disk.major = 0;
1464 info->disk.minor = 0;
1465 if (dl) {
1466 info->disk.major = dl->major;
1467 info->disk.minor = dl->minor;
1468 }
1469
1470 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1471 info->component_size = __le32_to_cpu(map->blocks_per_member);
1472 memset(info->uuid, 0, sizeof(info->uuid));
1473 info->recovery_start = MaxSector;
1474 info->reshape_active = 0;
1475
1476 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty) {
1477 info->resync_start = 0;
1478 } else if (dev->vol.migr_state) {
1479 switch (migr_type(dev)) {
1480 case MIGR_REPAIR:
1481 case MIGR_INIT: {
1482 __u64 blocks_per_unit = blocks_per_migr_unit(dev);
1483 __u64 units = __le32_to_cpu(dev->vol.curr_migr_unit);
1484
1485 info->resync_start = blocks_per_unit * units;
1486 break;
1487 }
1488 case MIGR_VERIFY:
1489 /* we could emulate the checkpointing of
1490 * 'sync_action=check' migrations, but for now
1491 * we just immediately complete them
1492 */
1493 case MIGR_REBUILD:
1494 /* this is handled by container_content_imsm() */
1495 case MIGR_GEN_MIGR:
1496 case MIGR_STATE_CHANGE:
1497 /* FIXME handle other migrations */
1498 default:
1499 /* we are not dirty, so... */
1500 info->resync_start = MaxSector;
1501 }
1502 } else
1503 info->resync_start = MaxSector;
1504
1505 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1506 info->name[MAX_RAID_SERIAL_LEN] = 0;
1507
1508 info->array.major_version = -1;
1509 info->array.minor_version = -2;
1510 devname = devnum2devname(st->container_dev);
1511 *info->text_version = '\0';
1512 if (devname)
1513 sprintf(info->text_version, "/%s/%d", devname, info->container_member);
1514 free(devname);
1515 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1516 uuid_from_super_imsm(st, info->uuid);
1517
1518 if (dmap) {
1519 int i, j;
1520 for (i=0; i<map_disks; i++) {
1521 dmap[i] = 0;
1522 if (i < info->array.raid_disks) {
1523 struct imsm_disk *dsk;
1524 j = get_imsm_disk_idx(dev, i);
1525 dsk = get_imsm_disk(super, j);
1526 if (dsk && (dsk->status & CONFIGURED_DISK))
1527 dmap[i] = 1;
1528 }
1529 }
1530 }
1531 }
1532
1533 /* check the config file to see if we can return a real uuid for this spare */
1534 static void fixup_container_spare_uuid(struct mdinfo *inf)
1535 {
1536 struct mddev_ident_s *array_list;
1537
1538 if (inf->array.level != LEVEL_CONTAINER ||
1539 memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0)
1540 return;
1541
1542 array_list = conf_get_ident(NULL);
1543
1544 for (; array_list; array_list = array_list->next) {
1545 if (array_list->uuid_set) {
1546 struct supertype *_sst; /* spare supertype */
1547 struct supertype *_cst; /* container supertype */
1548
1549 _cst = array_list->st;
1550 if (_cst)
1551 _sst = _cst->ss->match_metadata_desc(inf->text_version);
1552 else
1553 _sst = NULL;
1554
1555 if (_sst) {
1556 memcpy(inf->uuid, array_list->uuid, sizeof(int[4]));
1557 free(_sst);
1558 break;
1559 }
1560 }
1561 }
1562 }
1563
1564
1565 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed);
1566 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev);
1567
1568 static struct imsm_disk *get_imsm_missing(struct intel_super *super, __u8 index)
1569 {
1570 struct dl *d;
1571
1572 for (d = super->missing; d; d = d->next)
1573 if (d->index == index)
1574 return &d->disk;
1575 return NULL;
1576 }
1577
1578 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map)
1579 {
1580 struct intel_super *super = st->sb;
1581 struct imsm_disk *disk;
1582 int map_disks = info->array.raid_disks;
1583
1584 if (super->current_vol >= 0) {
1585 getinfo_super_imsm_volume(st, info, map);
1586 return;
1587 }
1588
1589 /* Set raid_disks to zero so that Assemble will always pull in valid
1590 * spares
1591 */
1592 info->array.raid_disks = 0;
1593 info->array.level = LEVEL_CONTAINER;
1594 info->array.layout = 0;
1595 info->array.md_minor = -1;
1596 info->array.ctime = 0; /* N/A for imsm */
1597 info->array.utime = 0;
1598 info->array.chunk_size = 0;
1599
1600 info->disk.major = 0;
1601 info->disk.minor = 0;
1602 info->disk.raid_disk = -1;
1603 info->reshape_active = 0;
1604 info->array.major_version = -1;
1605 info->array.minor_version = -2;
1606 strcpy(info->text_version, "imsm");
1607 info->safe_mode_delay = 0;
1608 info->disk.number = -1;
1609 info->disk.state = 0;
1610 info->name[0] = 0;
1611 info->recovery_start = MaxSector;
1612
1613 /* do we have the all the insync disks that we expect? */
1614 if (st->loaded_container) {
1615 struct imsm_super *mpb = super->anchor;
1616 int max_enough = -1, i;
1617
1618 for (i = 0; i < mpb->num_raid_devs; i++) {
1619 struct imsm_dev *dev = get_imsm_dev(super, i);
1620 int failed, enough, j, missing = 0;
1621 struct imsm_map *map;
1622 __u8 state;
1623
1624 failed = imsm_count_failed(super, dev);
1625 state = imsm_check_degraded(super, dev, failed);
1626 map = get_imsm_map(dev, dev->vol.migr_state);
1627
1628 /* any newly missing disks?
1629 * (catches single-degraded vs double-degraded)
1630 */
1631 for (j = 0; j < map->num_members; j++) {
1632 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
1633 __u32 idx = ord_to_idx(ord);
1634
1635 if (!(ord & IMSM_ORD_REBUILD) &&
1636 get_imsm_missing(super, idx)) {
1637 missing = 1;
1638 break;
1639 }
1640 }
1641
1642 if (state == IMSM_T_STATE_FAILED)
1643 enough = -1;
1644 else if (state == IMSM_T_STATE_DEGRADED &&
1645 (state != map->map_state || missing))
1646 enough = 0;
1647 else /* we're normal, or already degraded */
1648 enough = 1;
1649
1650 /* in the missing/failed disk case check to see
1651 * if at least one array is runnable
1652 */
1653 max_enough = max(max_enough, enough);
1654 }
1655 dprintf("%s: enough: %d\n", __func__, max_enough);
1656 info->container_enough = max_enough;
1657 } else
1658 info->container_enough = -1;
1659
1660 if (super->disks) {
1661 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1662
1663 disk = &super->disks->disk;
1664 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1665 info->component_size = reserved;
1666 info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
1667 /* we don't change info->disk.raid_disk here because
1668 * this state will be finalized in mdmon after we have
1669 * found the 'most fresh' version of the metadata
1670 */
1671 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
1672 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
1673 }
1674
1675 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1676 * ->compare_super may have updated the 'num_raid_devs' field for spares
1677 */
1678 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1679 uuid_from_super_imsm(st, info->uuid);
1680 else {
1681 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1682 fixup_container_spare_uuid(info);
1683 }
1684
1685 /* I don't know how to compute 'map' on imsm, so use safe default */
1686 if (map) {
1687 int i;
1688 for (i = 0; i < map_disks; i++)
1689 map[i] = 1;
1690 }
1691
1692 }
1693
1694 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1695 char *update, char *devname, int verbose,
1696 int uuid_set, char *homehost)
1697 {
1698 /* For 'assemble' and 'force' we need to return non-zero if any
1699 * change was made. For others, the return value is ignored.
1700 * Update options are:
1701 * force-one : This device looks a bit old but needs to be included,
1702 * update age info appropriately.
1703 * assemble: clear any 'faulty' flag to allow this device to
1704 * be assembled.
1705 * force-array: Array is degraded but being forced, mark it clean
1706 * if that will be needed to assemble it.
1707 *
1708 * newdev: not used ????
1709 * grow: Array has gained a new device - this is currently for
1710 * linear only
1711 * resync: mark as dirty so a resync will happen.
1712 * name: update the name - preserving the homehost
1713 * uuid: Change the uuid of the array to match watch is given
1714 *
1715 * Following are not relevant for this imsm:
1716 * sparc2.2 : update from old dodgey metadata
1717 * super-minor: change the preferred_minor number
1718 * summaries: update redundant counters.
1719 * homehost: update the recorded homehost
1720 * _reshape_progress: record new reshape_progress position.
1721 */
1722 int rv = 1;
1723 struct intel_super *super = st->sb;
1724 struct imsm_super *mpb;
1725
1726 /* we can only update container info */
1727 if (!super || super->current_vol >= 0 || !super->anchor)
1728 return 1;
1729
1730 mpb = super->anchor;
1731
1732 if (strcmp(update, "uuid") == 0 && uuid_set && !info->update_private)
1733 rv = -1;
1734 else if (strcmp(update, "uuid") == 0 && uuid_set && info->update_private) {
1735 mpb->orig_family_num = *((__u32 *) info->update_private);
1736 rv = 0;
1737 } else if (strcmp(update, "uuid") == 0) {
1738 __u32 *new_family = malloc(sizeof(*new_family));
1739
1740 /* update orig_family_number with the incoming random
1741 * data, report the new effective uuid, and store the
1742 * new orig_family_num for future updates.
1743 */
1744 if (new_family) {
1745 memcpy(&mpb->orig_family_num, info->uuid, sizeof(__u32));
1746 uuid_from_super_imsm(st, info->uuid);
1747 *new_family = mpb->orig_family_num;
1748 info->update_private = new_family;
1749 rv = 0;
1750 }
1751 } else if (strcmp(update, "assemble") == 0)
1752 rv = 0;
1753 else
1754 rv = -1;
1755
1756 /* successful update? recompute checksum */
1757 if (rv == 0)
1758 mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb));
1759
1760 return rv;
1761 }
1762
1763 static size_t disks_to_mpb_size(int disks)
1764 {
1765 size_t size;
1766
1767 size = sizeof(struct imsm_super);
1768 size += (disks - 1) * sizeof(struct imsm_disk);
1769 size += 2 * sizeof(struct imsm_dev);
1770 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1771 size += (4 - 2) * sizeof(struct imsm_map);
1772 /* 4 possible disk_ord_tbl's */
1773 size += 4 * (disks - 1) * sizeof(__u32);
1774
1775 return size;
1776 }
1777
1778 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1779 {
1780 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1781 return 0;
1782
1783 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1784 }
1785
1786 static void free_devlist(struct intel_super *super)
1787 {
1788 struct intel_dev *dv;
1789
1790 while (super->devlist) {
1791 dv = super->devlist->next;
1792 free(super->devlist->dev);
1793 free(super->devlist);
1794 super->devlist = dv;
1795 }
1796 }
1797
1798 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1799 {
1800 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1801 }
1802
1803 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1804 {
1805 /*
1806 * return:
1807 * 0 same, or first was empty, and second was copied
1808 * 1 second had wrong number
1809 * 2 wrong uuid
1810 * 3 wrong other info
1811 */
1812 struct intel_super *first = st->sb;
1813 struct intel_super *sec = tst->sb;
1814
1815 if (!first) {
1816 st->sb = tst->sb;
1817 tst->sb = NULL;
1818 return 0;
1819 }
1820
1821 /* if an anchor does not have num_raid_devs set then it is a free
1822 * floating spare
1823 */
1824 if (first->anchor->num_raid_devs > 0 &&
1825 sec->anchor->num_raid_devs > 0) {
1826 /* Determine if these disks might ever have been
1827 * related. Further disambiguation can only take place
1828 * in load_super_imsm_all
1829 */
1830 __u32 first_family = first->anchor->orig_family_num;
1831 __u32 sec_family = sec->anchor->orig_family_num;
1832
1833 if (memcmp(first->anchor->sig, sec->anchor->sig,
1834 MAX_SIGNATURE_LENGTH) != 0)
1835 return 3;
1836
1837 if (first_family == 0)
1838 first_family = first->anchor->family_num;
1839 if (sec_family == 0)
1840 sec_family = sec->anchor->family_num;
1841
1842 if (first_family != sec_family)
1843 return 3;
1844
1845 }
1846
1847
1848 /* if 'first' is a spare promote it to a populated mpb with sec's
1849 * family number
1850 */
1851 if (first->anchor->num_raid_devs == 0 &&
1852 sec->anchor->num_raid_devs > 0) {
1853 int i;
1854 struct intel_dev *dv;
1855 struct imsm_dev *dev;
1856
1857 /* we need to copy raid device info from sec if an allocation
1858 * fails here we don't associate the spare
1859 */
1860 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1861 dv = malloc(sizeof(*dv));
1862 if (!dv)
1863 break;
1864 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1865 if (!dev) {
1866 free(dv);
1867 break;
1868 }
1869 dv->dev = dev;
1870 dv->index = i;
1871 dv->next = first->devlist;
1872 first->devlist = dv;
1873 }
1874 if (i < sec->anchor->num_raid_devs) {
1875 /* allocation failure */
1876 free_devlist(first);
1877 fprintf(stderr, "imsm: failed to associate spare\n");
1878 return 3;
1879 }
1880 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1881 first->anchor->orig_family_num = sec->anchor->orig_family_num;
1882 first->anchor->family_num = sec->anchor->family_num;
1883 memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH);
1884 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1885 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1886 }
1887
1888 return 0;
1889 }
1890
1891 static void fd2devname(int fd, char *name)
1892 {
1893 struct stat st;
1894 char path[256];
1895 char dname[PATH_MAX];
1896 char *nm;
1897 int rv;
1898
1899 name[0] = '\0';
1900 if (fstat(fd, &st) != 0)
1901 return;
1902 sprintf(path, "/sys/dev/block/%d:%d",
1903 major(st.st_rdev), minor(st.st_rdev));
1904
1905 rv = readlink(path, dname, sizeof(dname));
1906 if (rv <= 0)
1907 return;
1908
1909 dname[rv] = '\0';
1910 nm = strrchr(dname, '/');
1911 nm++;
1912 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1913 }
1914
1915 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1916
1917 static int imsm_read_serial(int fd, char *devname,
1918 __u8 serial[MAX_RAID_SERIAL_LEN])
1919 {
1920 unsigned char scsi_serial[255];
1921 int rv;
1922 int rsp_len;
1923 int len;
1924 char *dest;
1925 char *src;
1926 char *rsp_buf;
1927 int i;
1928
1929 memset(scsi_serial, 0, sizeof(scsi_serial));
1930
1931 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1932
1933 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1934 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1935 fd2devname(fd, (char *) serial);
1936 return 0;
1937 }
1938
1939 if (rv != 0) {
1940 if (devname)
1941 fprintf(stderr,
1942 Name ": Failed to retrieve serial for %s\n",
1943 devname);
1944 return rv;
1945 }
1946
1947 rsp_len = scsi_serial[3];
1948 if (!rsp_len) {
1949 if (devname)
1950 fprintf(stderr,
1951 Name ": Failed to retrieve serial for %s\n",
1952 devname);
1953 return 2;
1954 }
1955 rsp_buf = (char *) &scsi_serial[4];
1956
1957 /* trim all whitespace and non-printable characters and convert
1958 * ':' to ';'
1959 */
1960 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
1961 src = &rsp_buf[i];
1962 if (*src > 0x20) {
1963 /* ':' is reserved for use in placeholder serial
1964 * numbers for missing disks
1965 */
1966 if (*src == ':')
1967 *dest++ = ';';
1968 else
1969 *dest++ = *src;
1970 }
1971 }
1972 len = dest - rsp_buf;
1973 dest = rsp_buf;
1974
1975 /* truncate leading characters */
1976 if (len > MAX_RAID_SERIAL_LEN) {
1977 dest += len - MAX_RAID_SERIAL_LEN;
1978 len = MAX_RAID_SERIAL_LEN;
1979 }
1980
1981 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1982 memcpy(serial, dest, len);
1983
1984 return 0;
1985 }
1986
1987 static int serialcmp(__u8 *s1, __u8 *s2)
1988 {
1989 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1990 }
1991
1992 static void serialcpy(__u8 *dest, __u8 *src)
1993 {
1994 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1995 }
1996
1997 #ifndef MDASSEMBLE
1998 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1999 {
2000 struct dl *dl;
2001
2002 for (dl = super->disks; dl; dl = dl->next)
2003 if (serialcmp(dl->serial, serial) == 0)
2004 break;
2005
2006 return dl;
2007 }
2008 #endif
2009
2010 static struct imsm_disk *
2011 __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
2012 {
2013 int i;
2014
2015 for (i = 0; i < mpb->num_disks; i++) {
2016 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2017
2018 if (serialcmp(disk->serial, serial) == 0) {
2019 if (idx)
2020 *idx = i;
2021 return disk;
2022 }
2023 }
2024
2025 return NULL;
2026 }
2027
2028 static int
2029 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
2030 {
2031 struct imsm_disk *disk;
2032 struct dl *dl;
2033 struct stat stb;
2034 int rv;
2035 char name[40];
2036 __u8 serial[MAX_RAID_SERIAL_LEN];
2037
2038 rv = imsm_read_serial(fd, devname, serial);
2039
2040 if (rv != 0)
2041 return 2;
2042
2043 dl = calloc(1, sizeof(*dl));
2044 if (!dl) {
2045 if (devname)
2046 fprintf(stderr,
2047 Name ": failed to allocate disk buffer for %s\n",
2048 devname);
2049 return 2;
2050 }
2051
2052 fstat(fd, &stb);
2053 dl->major = major(stb.st_rdev);
2054 dl->minor = minor(stb.st_rdev);
2055 dl->next = super->disks;
2056 dl->fd = keep_fd ? fd : -1;
2057 assert(super->disks == NULL);
2058 super->disks = dl;
2059 serialcpy(dl->serial, serial);
2060 dl->index = -2;
2061 dl->e = NULL;
2062 fd2devname(fd, name);
2063 if (devname)
2064 dl->devname = strdup(devname);
2065 else
2066 dl->devname = strdup(name);
2067
2068 /* look up this disk's index in the current anchor */
2069 disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
2070 if (disk) {
2071 dl->disk = *disk;
2072 /* only set index on disks that are a member of a
2073 * populated contianer, i.e. one with raid_devs
2074 */
2075 if (is_failed(&dl->disk))
2076 dl->index = -2;
2077 else if (is_spare(&dl->disk))
2078 dl->index = -1;
2079 }
2080
2081 return 0;
2082 }
2083
2084 #ifndef MDASSEMBLE
2085 /* When migrating map0 contains the 'destination' state while map1
2086 * contains the current state. When not migrating map0 contains the
2087 * current state. This routine assumes that map[0].map_state is set to
2088 * the current array state before being called.
2089 *
2090 * Migration is indicated by one of the following states
2091 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2092 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2093 * map1state=unitialized)
2094 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2095 * map1state=normal)
2096 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2097 * map1state=degraded)
2098 */
2099 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
2100 {
2101 struct imsm_map *dest;
2102 struct imsm_map *src = get_imsm_map(dev, 0);
2103
2104 dev->vol.migr_state = 1;
2105 set_migr_type(dev, migr_type);
2106 dev->vol.curr_migr_unit = 0;
2107 dest = get_imsm_map(dev, 1);
2108
2109 /* duplicate and then set the target end state in map[0] */
2110 memcpy(dest, src, sizeof_imsm_map(src));
2111 if (migr_type == MIGR_REBUILD) {
2112 __u32 ord;
2113 int i;
2114
2115 for (i = 0; i < src->num_members; i++) {
2116 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
2117 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
2118 }
2119 }
2120
2121 src->map_state = to_state;
2122 }
2123
2124 static void end_migration(struct imsm_dev *dev, __u8 map_state)
2125 {
2126 struct imsm_map *map = get_imsm_map(dev, 0);
2127 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
2128 int i;
2129
2130 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2131 * completed in the last migration.
2132 *
2133 * FIXME add support for online capacity expansion and
2134 * raid-level-migration
2135 */
2136 for (i = 0; i < prev->num_members; i++)
2137 map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i];
2138
2139 dev->vol.migr_state = 0;
2140 dev->vol.curr_migr_unit = 0;
2141 map->map_state = map_state;
2142 }
2143 #endif
2144
2145 static int parse_raid_devices(struct intel_super *super)
2146 {
2147 int i;
2148 struct imsm_dev *dev_new;
2149 size_t len, len_migr;
2150 size_t space_needed = 0;
2151 struct imsm_super *mpb = super->anchor;
2152
2153 for (i = 0; i < super->anchor->num_raid_devs; i++) {
2154 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
2155 struct intel_dev *dv;
2156
2157 len = sizeof_imsm_dev(dev_iter, 0);
2158 len_migr = sizeof_imsm_dev(dev_iter, 1);
2159 if (len_migr > len)
2160 space_needed += len_migr - len;
2161
2162 dv = malloc(sizeof(*dv));
2163 if (!dv)
2164 return 1;
2165 dev_new = malloc(len_migr);
2166 if (!dev_new) {
2167 free(dv);
2168 return 1;
2169 }
2170 imsm_copy_dev(dev_new, dev_iter);
2171 dv->dev = dev_new;
2172 dv->index = i;
2173 dv->next = super->devlist;
2174 super->devlist = dv;
2175 }
2176
2177 /* ensure that super->buf is large enough when all raid devices
2178 * are migrating
2179 */
2180 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
2181 void *buf;
2182
2183 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
2184 if (posix_memalign(&buf, 512, len) != 0)
2185 return 1;
2186
2187 memcpy(buf, super->buf, super->len);
2188 memset(buf + super->len, 0, len - super->len);
2189 free(super->buf);
2190 super->buf = buf;
2191 super->len = len;
2192 }
2193
2194 return 0;
2195 }
2196
2197 /* retrieve a pointer to the bbm log which starts after all raid devices */
2198 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
2199 {
2200 void *ptr = NULL;
2201
2202 if (__le32_to_cpu(mpb->bbm_log_size)) {
2203 ptr = mpb;
2204 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
2205 }
2206
2207 return ptr;
2208 }
2209
2210 static void __free_imsm(struct intel_super *super, int free_disks);
2211
2212 /* load_imsm_mpb - read matrix metadata
2213 * allocates super->mpb to be freed by free_super
2214 */
2215 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
2216 {
2217 unsigned long long dsize;
2218 unsigned long long sectors;
2219 struct stat;
2220 struct imsm_super *anchor;
2221 __u32 check_sum;
2222
2223 get_dev_size(fd, NULL, &dsize);
2224 if (dsize < 1024) {
2225 if (devname)
2226 fprintf(stderr,
2227 Name ": %s: device to small for imsm\n",
2228 devname);
2229 return 1;
2230 }
2231
2232 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
2233 if (devname)
2234 fprintf(stderr,
2235 Name ": Cannot seek to anchor block on %s: %s\n",
2236 devname, strerror(errno));
2237 return 1;
2238 }
2239
2240 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
2241 if (devname)
2242 fprintf(stderr,
2243 Name ": Failed to allocate imsm anchor buffer"
2244 " on %s\n", devname);
2245 return 1;
2246 }
2247 if (read(fd, anchor, 512) != 512) {
2248 if (devname)
2249 fprintf(stderr,
2250 Name ": Cannot read anchor block on %s: %s\n",
2251 devname, strerror(errno));
2252 free(anchor);
2253 return 1;
2254 }
2255
2256 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
2257 if (devname)
2258 fprintf(stderr,
2259 Name ": no IMSM anchor on %s\n", devname);
2260 free(anchor);
2261 return 2;
2262 }
2263
2264 __free_imsm(super, 0);
2265 super->len = ROUND_UP(anchor->mpb_size, 512);
2266 if (posix_memalign(&super->buf, 512, super->len) != 0) {
2267 if (devname)
2268 fprintf(stderr,
2269 Name ": unable to allocate %zu byte mpb buffer\n",
2270 super->len);
2271 free(anchor);
2272 return 2;
2273 }
2274 memcpy(super->buf, anchor, 512);
2275
2276 sectors = mpb_sectors(anchor) - 1;
2277 free(anchor);
2278 if (!sectors) {
2279 check_sum = __gen_imsm_checksum(super->anchor);
2280 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
2281 if (devname)
2282 fprintf(stderr,
2283 Name ": IMSM checksum %x != %x on %s\n",
2284 check_sum,
2285 __le32_to_cpu(super->anchor->check_sum),
2286 devname);
2287 return 2;
2288 }
2289
2290 return 0;
2291 }
2292
2293 /* read the extended mpb */
2294 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
2295 if (devname)
2296 fprintf(stderr,
2297 Name ": Cannot seek to extended mpb on %s: %s\n",
2298 devname, strerror(errno));
2299 return 1;
2300 }
2301
2302 if ((unsigned)read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
2303 if (devname)
2304 fprintf(stderr,
2305 Name ": Cannot read extended mpb on %s: %s\n",
2306 devname, strerror(errno));
2307 return 2;
2308 }
2309
2310 check_sum = __gen_imsm_checksum(super->anchor);
2311 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
2312 if (devname)
2313 fprintf(stderr,
2314 Name ": IMSM checksum %x != %x on %s\n",
2315 check_sum, __le32_to_cpu(super->anchor->check_sum),
2316 devname);
2317 return 3;
2318 }
2319
2320 /* FIXME the BBM log is disk specific so we cannot use this global
2321 * buffer for all disks. Ok for now since we only look at the global
2322 * bbm_log_size parameter to gate assembly
2323 */
2324 super->bbm_log = __get_imsm_bbm_log(super->anchor);
2325
2326 return 0;
2327 }
2328
2329 static int
2330 load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
2331 {
2332 int err;
2333
2334 err = load_imsm_mpb(fd, super, devname);
2335 if (err)
2336 return err;
2337 err = load_imsm_disk(fd, super, devname, keep_fd);
2338 if (err)
2339 return err;
2340 err = parse_raid_devices(super);
2341
2342 return err;
2343 }
2344
2345 static void __free_imsm_disk(struct dl *d)
2346 {
2347 if (d->fd >= 0)
2348 close(d->fd);
2349 if (d->devname)
2350 free(d->devname);
2351 if (d->e)
2352 free(d->e);
2353 free(d);
2354
2355 }
2356 static void free_imsm_disks(struct intel_super *super)
2357 {
2358 struct dl *d;
2359
2360 while (super->disks) {
2361 d = super->disks;
2362 super->disks = d->next;
2363 __free_imsm_disk(d);
2364 }
2365 while (super->missing) {
2366 d = super->missing;
2367 super->missing = d->next;
2368 __free_imsm_disk(d);
2369 }
2370
2371 }
2372
2373 /* free all the pieces hanging off of a super pointer */
2374 static void __free_imsm(struct intel_super *super, int free_disks)
2375 {
2376 if (super->buf) {
2377 free(super->buf);
2378 super->buf = NULL;
2379 }
2380 if (free_disks)
2381 free_imsm_disks(super);
2382 free_devlist(super);
2383 if (super->hba) {
2384 free((void *) super->hba);
2385 super->hba = NULL;
2386 }
2387 }
2388
2389 static void free_imsm(struct intel_super *super)
2390 {
2391 __free_imsm(super, 1);
2392 free(super);
2393 }
2394
2395 static void free_super_imsm(struct supertype *st)
2396 {
2397 struct intel_super *super = st->sb;
2398
2399 if (!super)
2400 return;
2401
2402 free_imsm(super);
2403 st->sb = NULL;
2404 }
2405
2406 static struct intel_super *alloc_super(void)
2407 {
2408 struct intel_super *super = malloc(sizeof(*super));
2409
2410 if (super) {
2411 memset(super, 0, sizeof(*super));
2412 super->current_vol = -1;
2413 super->create_offset = ~((__u32 ) 0);
2414 if (!check_env("IMSM_NO_PLATFORM"))
2415 super->orom = find_imsm_orom();
2416 if (super->orom && !check_env("IMSM_TEST_OROM")) {
2417 struct sys_dev *list, *ent;
2418
2419 /* find the first intel ahci controller */
2420 list = find_driver_devices("pci", "ahci");
2421 for (ent = list; ent; ent = ent->next)
2422 if (devpath_to_vendor(ent->path) == 0x8086)
2423 break;
2424 if (ent) {
2425 super->hba = ent->path;
2426 ent->path = NULL;
2427 }
2428 free_sys_dev(&list);
2429 }
2430 }
2431
2432 return super;
2433 }
2434
2435 #ifndef MDASSEMBLE
2436 /* find_missing - helper routine for load_super_imsm_all that identifies
2437 * disks that have disappeared from the system. This routine relies on
2438 * the mpb being uptodate, which it is at load time.
2439 */
2440 static int find_missing(struct intel_super *super)
2441 {
2442 int i;
2443 struct imsm_super *mpb = super->anchor;
2444 struct dl *dl;
2445 struct imsm_disk *disk;
2446
2447 for (i = 0; i < mpb->num_disks; i++) {
2448 disk = __get_imsm_disk(mpb, i);
2449 dl = serial_to_dl(disk->serial, super);
2450 if (dl)
2451 continue;
2452
2453 dl = malloc(sizeof(*dl));
2454 if (!dl)
2455 return 1;
2456 dl->major = 0;
2457 dl->minor = 0;
2458 dl->fd = -1;
2459 dl->devname = strdup("missing");
2460 dl->index = i;
2461 serialcpy(dl->serial, disk->serial);
2462 dl->disk = *disk;
2463 dl->e = NULL;
2464 dl->next = super->missing;
2465 super->missing = dl;
2466 }
2467
2468 return 0;
2469 }
2470
2471 static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
2472 {
2473 struct intel_disk *idisk = disk_list;
2474
2475 while (idisk) {
2476 if (serialcmp(idisk->disk.serial, serial) == 0)
2477 break;
2478 idisk = idisk->next;
2479 }
2480
2481 return idisk;
2482 }
2483
2484 static int __prep_thunderdome(struct intel_super **table, int tbl_size,
2485 struct intel_super *super,
2486 struct intel_disk **disk_list)
2487 {
2488 struct imsm_disk *d = &super->disks->disk;
2489 struct imsm_super *mpb = super->anchor;
2490 int i, j;
2491
2492 for (i = 0; i < tbl_size; i++) {
2493 struct imsm_super *tbl_mpb = table[i]->anchor;
2494 struct imsm_disk *tbl_d = &table[i]->disks->disk;
2495
2496 if (tbl_mpb->family_num == mpb->family_num) {
2497 if (tbl_mpb->check_sum == mpb->check_sum) {
2498 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2499 __func__, super->disks->major,
2500 super->disks->minor,
2501 table[i]->disks->major,
2502 table[i]->disks->minor);
2503 break;
2504 }
2505
2506 if (((is_configured(d) && !is_configured(tbl_d)) ||
2507 is_configured(d) == is_configured(tbl_d)) &&
2508 tbl_mpb->generation_num < mpb->generation_num) {
2509 /* current version of the mpb is a
2510 * better candidate than the one in
2511 * super_table, but copy over "cross
2512 * generational" status
2513 */
2514 struct intel_disk *idisk;
2515
2516 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2517 __func__, super->disks->major,
2518 super->disks->minor,
2519 table[i]->disks->major,
2520 table[i]->disks->minor);
2521
2522 idisk = disk_list_get(tbl_d->serial, *disk_list);
2523 if (idisk && is_failed(&idisk->disk))
2524 tbl_d->status |= FAILED_DISK;
2525 break;
2526 } else {
2527 struct intel_disk *idisk;
2528 struct imsm_disk *disk;
2529
2530 /* tbl_mpb is more up to date, but copy
2531 * over cross generational status before
2532 * returning
2533 */
2534 disk = __serial_to_disk(d->serial, mpb, NULL);
2535 if (disk && is_failed(disk))
2536 d->status |= FAILED_DISK;
2537
2538 idisk = disk_list_get(d->serial, *disk_list);
2539 if (idisk) {
2540 idisk->owner = i;
2541 if (disk && is_configured(disk))
2542 idisk->disk.status |= CONFIGURED_DISK;
2543 }
2544
2545 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2546 __func__, super->disks->major,
2547 super->disks->minor,
2548 table[i]->disks->major,
2549 table[i]->disks->minor);
2550
2551 return tbl_size;
2552 }
2553 }
2554 }
2555
2556 if (i >= tbl_size)
2557 table[tbl_size++] = super;
2558 else
2559 table[i] = super;
2560
2561 /* update/extend the merged list of imsm_disk records */
2562 for (j = 0; j < mpb->num_disks; j++) {
2563 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
2564 struct intel_disk *idisk;
2565
2566 idisk = disk_list_get(disk->serial, *disk_list);
2567 if (idisk) {
2568 idisk->disk.status |= disk->status;
2569 if (is_configured(&idisk->disk) ||
2570 is_failed(&idisk->disk))
2571 idisk->disk.status &= ~(SPARE_DISK);
2572 } else {
2573 idisk = calloc(1, sizeof(*idisk));
2574 if (!idisk)
2575 return -1;
2576 idisk->owner = IMSM_UNKNOWN_OWNER;
2577 idisk->disk = *disk;
2578 idisk->next = *disk_list;
2579 *disk_list = idisk;
2580 }
2581
2582 if (serialcmp(idisk->disk.serial, d->serial) == 0)
2583 idisk->owner = i;
2584 }
2585
2586 return tbl_size;
2587 }
2588
2589 static struct intel_super *
2590 validate_members(struct intel_super *super, struct intel_disk *disk_list,
2591 const int owner)
2592 {
2593 struct imsm_super *mpb = super->anchor;
2594 int ok_count = 0;
2595 int i;
2596
2597 for (i = 0; i < mpb->num_disks; i++) {
2598 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2599 struct intel_disk *idisk;
2600
2601 idisk = disk_list_get(disk->serial, disk_list);
2602 if (idisk) {
2603 if (idisk->owner == owner ||
2604 idisk->owner == IMSM_UNKNOWN_OWNER)
2605 ok_count++;
2606 else
2607 dprintf("%s: '%.16s' owner %d != %d\n",
2608 __func__, disk->serial, idisk->owner,
2609 owner);
2610 } else {
2611 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2612 __func__, __le32_to_cpu(mpb->family_num), i,
2613 disk->serial);
2614 break;
2615 }
2616 }
2617
2618 if (ok_count == mpb->num_disks)
2619 return super;
2620 return NULL;
2621 }
2622
2623 static void show_conflicts(__u32 family_num, struct intel_super *super_list)
2624 {
2625 struct intel_super *s;
2626
2627 for (s = super_list; s; s = s->next) {
2628 if (family_num != s->anchor->family_num)
2629 continue;
2630 fprintf(stderr, "Conflict, offlining family %#x on '%s'\n",
2631 __le32_to_cpu(family_num), s->disks->devname);
2632 }
2633 }
2634
2635 static struct intel_super *
2636 imsm_thunderdome(struct intel_super **super_list, int len)
2637 {
2638 struct intel_super *super_table[len];
2639 struct intel_disk *disk_list = NULL;
2640 struct intel_super *champion, *spare;
2641 struct intel_super *s, **del;
2642 int tbl_size = 0;
2643 int conflict;
2644 int i;
2645
2646 memset(super_table, 0, sizeof(super_table));
2647 for (s = *super_list; s; s = s->next)
2648 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
2649
2650 for (i = 0; i < tbl_size; i++) {
2651 struct imsm_disk *d;
2652 struct intel_disk *idisk;
2653 struct imsm_super *mpb = super_table[i]->anchor;
2654
2655 s = super_table[i];
2656 d = &s->disks->disk;
2657
2658 /* 'd' must appear in merged disk list for its
2659 * configuration to be valid
2660 */
2661 idisk = disk_list_get(d->serial, disk_list);
2662 if (idisk && idisk->owner == i)
2663 s = validate_members(s, disk_list, i);
2664 else
2665 s = NULL;
2666
2667 if (!s)
2668 dprintf("%s: marking family: %#x from %d:%d offline\n",
2669 __func__, mpb->family_num,
2670 super_table[i]->disks->major,
2671 super_table[i]->disks->minor);
2672 super_table[i] = s;
2673 }
2674
2675 /* This is where the mdadm implementation differs from the Windows
2676 * driver which has no strict concept of a container. We can only
2677 * assemble one family from a container, so when returning a prodigal
2678 * array member to this system the code will not be able to disambiguate
2679 * the container contents that should be assembled ("foreign" versus
2680 * "local"). It requires user intervention to set the orig_family_num
2681 * to a new value to establish a new container. The Windows driver in
2682 * this situation fixes up the volume name in place and manages the
2683 * foreign array as an independent entity.
2684 */
2685 s = NULL;
2686 spare = NULL;
2687 conflict = 0;
2688 for (i = 0; i < tbl_size; i++) {
2689 struct intel_super *tbl_ent = super_table[i];
2690 int is_spare = 0;
2691
2692 if (!tbl_ent)
2693 continue;
2694
2695 if (tbl_ent->anchor->num_raid_devs == 0) {
2696 spare = tbl_ent;
2697 is_spare = 1;
2698 }
2699
2700 if (s && !is_spare) {
2701 show_conflicts(tbl_ent->anchor->family_num, *super_list);
2702 conflict++;
2703 } else if (!s && !is_spare)
2704 s = tbl_ent;
2705 }
2706
2707 if (!s)
2708 s = spare;
2709 if (!s) {
2710 champion = NULL;
2711 goto out;
2712 }
2713 champion = s;
2714
2715 if (conflict)
2716 fprintf(stderr, "Chose family %#x on '%s', "
2717 "assemble conflicts to new container with '--update=uuid'\n",
2718 __le32_to_cpu(s->anchor->family_num), s->disks->devname);
2719
2720 /* collect all dl's onto 'champion', and update them to
2721 * champion's version of the status
2722 */
2723 for (s = *super_list; s; s = s->next) {
2724 struct imsm_super *mpb = champion->anchor;
2725 struct dl *dl = s->disks;
2726
2727 if (s == champion)
2728 continue;
2729
2730 for (i = 0; i < mpb->num_disks; i++) {
2731 struct imsm_disk *disk;
2732
2733 disk = __serial_to_disk(dl->serial, mpb, &dl->index);
2734 if (disk) {
2735 dl->disk = *disk;
2736 /* only set index on disks that are a member of
2737 * a populated contianer, i.e. one with
2738 * raid_devs
2739 */
2740 if (is_failed(&dl->disk))
2741 dl->index = -2;
2742 else if (is_spare(&dl->disk))
2743 dl->index = -1;
2744 break;
2745 }
2746 }
2747
2748 if (i >= mpb->num_disks) {
2749 struct intel_disk *idisk;
2750
2751 idisk = disk_list_get(dl->serial, disk_list);
2752 if (idisk && is_spare(&idisk->disk) &&
2753 !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
2754 dl->index = -1;
2755 else {
2756 dl->index = -2;
2757 continue;
2758 }
2759 }
2760
2761 dl->next = champion->disks;
2762 champion->disks = dl;
2763 s->disks = NULL;
2764 }
2765
2766 /* delete 'champion' from super_list */
2767 for (del = super_list; *del; ) {
2768 if (*del == champion) {
2769 *del = (*del)->next;
2770 break;
2771 } else
2772 del = &(*del)->next;
2773 }
2774 champion->next = NULL;
2775
2776 out:
2777 while (disk_list) {
2778 struct intel_disk *idisk = disk_list;
2779
2780 disk_list = disk_list->next;
2781 free(idisk);
2782 }
2783
2784 return champion;
2785 }
2786
2787 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
2788 char *devname)
2789 {
2790 struct mdinfo *sra;
2791 struct intel_super *super_list = NULL;
2792 struct intel_super *super = NULL;
2793 int devnum = fd2devnum(fd);
2794 struct mdinfo *sd;
2795 int retry;
2796 int err = 0;
2797 int i;
2798
2799 /* check if 'fd' an opened container */
2800 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2801 if (!sra)
2802 return 1;
2803
2804 if (sra->array.major_version != -1 ||
2805 sra->array.minor_version != -2 ||
2806 strcmp(sra->text_version, "imsm") != 0) {
2807 err = 1;
2808 goto error;
2809 }
2810 /* load all mpbs */
2811 for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
2812 struct intel_super *s = alloc_super();
2813 char nm[32];
2814 int dfd;
2815
2816 err = 1;
2817 if (!s)
2818 goto error;
2819 s->next = super_list;
2820 super_list = s;
2821
2822 err = 2;
2823 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2824 dfd = dev_open(nm, O_RDWR);
2825 if (dfd < 0)
2826 goto error;
2827
2828 err = load_and_parse_mpb(dfd, s, NULL, 1);
2829
2830 /* retry the load if we might have raced against mdmon */
2831 if (err == 3 && mdmon_running(devnum))
2832 for (retry = 0; retry < 3; retry++) {
2833 usleep(3000);
2834 err = load_and_parse_mpb(dfd, s, NULL, 1);
2835 if (err != 3)
2836 break;
2837 }
2838 if (err)
2839 goto error;
2840 }
2841
2842 /* all mpbs enter, maybe one leaves */
2843 super = imsm_thunderdome(&super_list, i);
2844 if (!super) {
2845 err = 1;
2846 goto error;
2847 }
2848
2849 if (find_missing(super) != 0) {
2850 free_imsm(super);
2851 err = 2;
2852 goto error;
2853 }
2854 err = 0;
2855
2856 error:
2857 while (super_list) {
2858 struct intel_super *s = super_list;
2859
2860 super_list = super_list->next;
2861 free_imsm(s);
2862 }
2863 sysfs_free(sra);
2864
2865 if (err)
2866 return err;
2867
2868 *sbp = super;
2869 st->container_dev = devnum;
2870 if (err == 0 && st->ss == NULL) {
2871 st->ss = &super_imsm;
2872 st->minor_version = 0;
2873 st->max_devs = IMSM_MAX_DEVICES;
2874 }
2875 st->loaded_container = 1;
2876
2877 return 0;
2878 }
2879
2880 static int load_container_imsm(struct supertype *st, int fd, char *devname)
2881 {
2882 return load_super_imsm_all(st, fd, &st->sb, devname);
2883 }
2884 #endif
2885
2886 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2887 {
2888 struct intel_super *super;
2889 int rv;
2890
2891 #ifndef MDASSEMBLE
2892 if (load_super_imsm_all(st, fd, &st->sb, devname) == 0)
2893 return 0;
2894 #endif
2895
2896 if (test_partition(fd))
2897 /* IMSM not allowed on partitions */
2898 return 1;
2899
2900 free_super_imsm(st);
2901
2902 super = alloc_super();
2903 if (!super) {
2904 fprintf(stderr,
2905 Name ": malloc of %zu failed.\n",
2906 sizeof(*super));
2907 return 1;
2908 }
2909
2910 rv = load_and_parse_mpb(fd, super, devname, 0);
2911
2912 if (rv) {
2913 if (devname)
2914 fprintf(stderr,
2915 Name ": Failed to load all information "
2916 "sections on %s\n", devname);
2917 free_imsm(super);
2918 return rv;
2919 }
2920
2921 st->sb = super;
2922 if (st->ss == NULL) {
2923 st->ss = &super_imsm;
2924 st->minor_version = 0;
2925 st->max_devs = IMSM_MAX_DEVICES;
2926 }
2927 st->loaded_container = 0;
2928
2929 return 0;
2930 }
2931
2932 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2933 {
2934 if (info->level == 1)
2935 return 128;
2936 return info->chunk_size >> 9;
2937 }
2938
2939 static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
2940 {
2941 __u32 num_stripes;
2942
2943 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2944 num_stripes /= num_domains;
2945
2946 return num_stripes;
2947 }
2948
2949 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2950 {
2951 if (info->level == 1)
2952 return info->size * 2;
2953 else
2954 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2955 }
2956
2957 static void imsm_update_version_info(struct intel_super *super)
2958 {
2959 /* update the version and attributes */
2960 struct imsm_super *mpb = super->anchor;
2961 char *version;
2962 struct imsm_dev *dev;
2963 struct imsm_map *map;
2964 int i;
2965
2966 for (i = 0; i < mpb->num_raid_devs; i++) {
2967 dev = get_imsm_dev(super, i);
2968 map = get_imsm_map(dev, 0);
2969 if (__le32_to_cpu(dev->size_high) > 0)
2970 mpb->attributes |= MPB_ATTRIB_2TB;
2971
2972 /* FIXME detect when an array spans a port multiplier */
2973 #if 0
2974 mpb->attributes |= MPB_ATTRIB_PM;
2975 #endif
2976
2977 if (mpb->num_raid_devs > 1 ||
2978 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2979 version = MPB_VERSION_ATTRIBS;
2980 switch (get_imsm_raid_level(map)) {
2981 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2982 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2983 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2984 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2985 }
2986 } else {
2987 if (map->num_members >= 5)
2988 version = MPB_VERSION_5OR6_DISK_ARRAY;
2989 else if (dev->status == DEV_CLONE_N_GO)
2990 version = MPB_VERSION_CNG;
2991 else if (get_imsm_raid_level(map) == 5)
2992 version = MPB_VERSION_RAID5;
2993 else if (map->num_members >= 3)
2994 version = MPB_VERSION_3OR4_DISK_ARRAY;
2995 else if (get_imsm_raid_level(map) == 1)
2996 version = MPB_VERSION_RAID1;
2997 else
2998 version = MPB_VERSION_RAID0;
2999 }
3000 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
3001 }
3002 }
3003
3004 static int check_name(struct intel_super *super, char *name, int quiet)
3005 {
3006 struct imsm_super *mpb = super->anchor;
3007 char *reason = NULL;
3008 int i;
3009
3010 if (strlen(name) > MAX_RAID_SERIAL_LEN)
3011 reason = "must be 16 characters or less";
3012
3013 for (i = 0; i < mpb->num_raid_devs; i++) {
3014 struct imsm_dev *dev = get_imsm_dev(super, i);
3015
3016 if (strncmp((char *) dev->volume, name, MAX_RAID_SERIAL_LEN) == 0) {
3017 reason = "already exists";
3018 break;
3019 }
3020 }
3021
3022 if (reason && !quiet)
3023 fprintf(stderr, Name ": imsm volume name %s\n", reason);
3024
3025 return !reason;
3026 }
3027
3028 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
3029 unsigned long long size, char *name,
3030 char *homehost, int *uuid)
3031 {
3032 /* We are creating a volume inside a pre-existing container.
3033 * so st->sb is already set.
3034 */
3035 struct intel_super *super = st->sb;
3036 struct imsm_super *mpb = super->anchor;
3037 struct intel_dev *dv;
3038 struct imsm_dev *dev;
3039 struct imsm_vol *vol;
3040 struct imsm_map *map;
3041 int idx = mpb->num_raid_devs;
3042 int i;
3043 unsigned long long array_blocks;
3044 size_t size_old, size_new;
3045 __u32 num_data_stripes;
3046
3047 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
3048 fprintf(stderr, Name": This imsm-container already has the "
3049 "maximum of %d volumes\n", super->orom->vpa);
3050 return 0;
3051 }
3052
3053 /* ensure the mpb is large enough for the new data */
3054 size_old = __le32_to_cpu(mpb->mpb_size);
3055 size_new = disks_to_mpb_size(info->nr_disks);
3056 if (size_new > size_old) {
3057 void *mpb_new;
3058 size_t size_round = ROUND_UP(size_new, 512);
3059
3060 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
3061 fprintf(stderr, Name": could not allocate new mpb\n");
3062 return 0;
3063 }
3064 memcpy(mpb_new, mpb, size_old);
3065 free(mpb);
3066 mpb = mpb_new;
3067 super->anchor = mpb_new;
3068 mpb->mpb_size = __cpu_to_le32(size_new);
3069 memset(mpb_new + size_old, 0, size_round - size_old);
3070 }
3071 super->current_vol = idx;
3072 /* when creating the first raid device in this container set num_disks
3073 * to zero, i.e. delete this spare and add raid member devices in
3074 * add_to_super_imsm_volume()
3075 */
3076 if (super->current_vol == 0)
3077 mpb->num_disks = 0;
3078
3079 if (!check_name(super, name, 0))
3080 return 0;
3081 dv = malloc(sizeof(*dv));
3082 if (!dv) {
3083 fprintf(stderr, Name ": failed to allocate device list entry\n");
3084 return 0;
3085 }
3086 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
3087 if (!dev) {
3088 free(dv);
3089 fprintf(stderr, Name": could not allocate raid device\n");
3090 return 0;
3091 }
3092 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
3093 if (info->level == 1)
3094 array_blocks = info_to_blocks_per_member(info);
3095 else
3096 array_blocks = calc_array_size(info->level, info->raid_disks,
3097 info->layout, info->chunk_size,
3098 info->size*2);
3099 /* round array size down to closest MB */
3100 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
3101
3102 dev->size_low = __cpu_to_le32((__u32) array_blocks);
3103 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
3104 dev->status = __cpu_to_le32(0);
3105 dev->reserved_blocks = __cpu_to_le32(0);
3106 vol = &dev->vol;
3107 vol->migr_state = 0;
3108 set_migr_type(dev, MIGR_INIT);
3109 vol->dirty = 0;
3110 vol->curr_migr_unit = 0;
3111 map = get_imsm_map(dev, 0);
3112 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
3113 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
3114 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
3115 map->failed_disk_num = ~0;
3116 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
3117 IMSM_T_STATE_NORMAL;
3118 map->ddf = 1;
3119
3120 if (info->level == 1 && info->raid_disks > 2) {
3121 free(dev);
3122 free(dv);
3123 fprintf(stderr, Name": imsm does not support more than 2 disks"
3124 "in a raid1 volume\n");
3125 return 0;
3126 }
3127
3128 map->raid_level = info->level;
3129 if (info->level == 10) {
3130 map->raid_level = 1;
3131 map->num_domains = info->raid_disks / 2;
3132 } else if (info->level == 1)
3133 map->num_domains = info->raid_disks;
3134 else
3135 map->num_domains = 1;
3136
3137 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
3138 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
3139
3140 map->num_members = info->raid_disks;
3141 for (i = 0; i < map->num_members; i++) {
3142 /* initialized in add_to_super */
3143 set_imsm_ord_tbl_ent(map, i, IMSM_ORD_REBUILD);
3144 }
3145 mpb->num_raid_devs++;
3146
3147 dv->dev = dev;
3148 dv->index = super->current_vol;
3149 dv->next = super->devlist;
3150 super->devlist = dv;
3151
3152 imsm_update_version_info(super);
3153
3154 return 1;
3155 }
3156
3157 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
3158 unsigned long long size, char *name,
3159 char *homehost, int *uuid)
3160 {
3161 /* This is primarily called by Create when creating a new array.
3162 * We will then get add_to_super called for each component, and then
3163 * write_init_super called to write it out to each device.
3164 * For IMSM, Create can create on fresh devices or on a pre-existing
3165 * array.
3166 * To create on a pre-existing array a different method will be called.
3167 * This one is just for fresh drives.
3168 */
3169 struct intel_super *super;
3170 struct imsm_super *mpb;
3171 size_t mpb_size;
3172 char *version;
3173
3174 if (st->sb)
3175 return init_super_imsm_volume(st, info, size, name, homehost, uuid);
3176
3177 if (info)
3178 mpb_size = disks_to_mpb_size(info->nr_disks);
3179 else
3180 mpb_size = 512;
3181
3182 super = alloc_super();
3183 if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) {
3184 free(super);
3185 super = NULL;
3186 }
3187 if (!super) {
3188 fprintf(stderr, Name
3189 ": %s could not allocate superblock\n", __func__);
3190 return 0;
3191 }
3192 memset(super->buf, 0, mpb_size);
3193 mpb = super->buf;
3194 mpb->mpb_size = __cpu_to_le32(mpb_size);
3195 st->sb = super;
3196
3197 if (info == NULL) {
3198 /* zeroing superblock */
3199 return 0;
3200 }
3201
3202 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
3203
3204 version = (char *) mpb->sig;
3205 strcpy(version, MPB_SIGNATURE);
3206 version += strlen(MPB_SIGNATURE);
3207 strcpy(version, MPB_VERSION_RAID0);
3208
3209 return 1;
3210 }
3211
3212 #ifndef MDASSEMBLE
3213 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
3214 int fd, char *devname)
3215 {
3216 struct intel_super *super = st->sb;
3217 struct imsm_super *mpb = super->anchor;
3218 struct dl *dl;
3219 struct imsm_dev *dev;
3220 struct imsm_map *map;
3221 int slot;
3222
3223 dev = get_imsm_dev(super, super->current_vol);
3224 map = get_imsm_map(dev, 0);
3225
3226 if (! (dk->state & (1<<MD_DISK_SYNC))) {
3227 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
3228 devname);
3229 return 1;
3230 }
3231
3232 if (fd == -1) {
3233 /* we're doing autolayout so grab the pre-marked (in
3234 * validate_geometry) raid_disk
3235 */
3236 for (dl = super->disks; dl; dl = dl->next)
3237 if (dl->raiddisk == dk->raid_disk)
3238 break;
3239 } else {
3240 for (dl = super->disks; dl ; dl = dl->next)
3241 if (dl->major == dk->major &&
3242 dl->minor == dk->minor)
3243 break;
3244 }
3245
3246 if (!dl) {
3247 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
3248 return 1;
3249 }
3250
3251 /* add a pristine spare to the metadata */
3252 if (dl->index < 0) {
3253 dl->index = super->anchor->num_disks;
3254 super->anchor->num_disks++;
3255 }
3256 /* Check the device has not already been added */
3257 slot = get_imsm_disk_slot(map, dl->index);
3258 if (slot >= 0 &&
3259 (get_imsm_ord_tbl_ent(dev, slot) & IMSM_ORD_REBUILD) == 0) {
3260 fprintf(stderr, Name ": %s has been included in this array twice\n",
3261 devname);
3262 return 1;
3263 }
3264 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
3265 dl->disk.status = CONFIGURED_DISK;
3266
3267 /* if we are creating the first raid device update the family number */
3268 if (super->current_vol == 0) {
3269 __u32 sum;
3270 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
3271 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
3272
3273 if (!_dev || !_disk) {
3274 fprintf(stderr, Name ": BUG mpb setup error\n");
3275 return 1;
3276 }
3277 *_dev = *dev;
3278 *_disk = dl->disk;
3279 sum = random32();
3280 sum += __gen_imsm_checksum(mpb);
3281 mpb->family_num = __cpu_to_le32(sum);
3282 mpb->orig_family_num = mpb->family_num;
3283 }
3284
3285 return 0;
3286 }
3287
3288 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
3289 int fd, char *devname)
3290 {
3291 struct intel_super *super = st->sb;
3292 struct dl *dd;
3293 unsigned long long size;
3294 __u32 id;
3295 int rv;
3296 struct stat stb;
3297
3298 /* if we are on an RAID enabled platform check that the disk is
3299 * attached to the raid controller
3300 */
3301 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
3302 fprintf(stderr,
3303 Name ": %s is not attached to the raid controller: %s\n",
3304 devname ? : "disk", super->hba);
3305 return 1;
3306 }
3307
3308 if (super->current_vol >= 0)
3309 return add_to_super_imsm_volume(st, dk, fd, devname);
3310
3311 fstat(fd, &stb);
3312 dd = malloc(sizeof(*dd));
3313 if (!dd) {
3314 fprintf(stderr,
3315 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
3316 return 1;
3317 }
3318 memset(dd, 0, sizeof(*dd));
3319 dd->major = major(stb.st_rdev);
3320 dd->minor = minor(stb.st_rdev);
3321 dd->index = -1;
3322 dd->devname = devname ? strdup(devname) : NULL;
3323 dd->fd = fd;
3324 dd->e = NULL;
3325 rv = imsm_read_serial(fd, devname, dd->serial);
3326 if (rv) {
3327 fprintf(stderr,
3328 Name ": failed to retrieve scsi serial, aborting\n");
3329 free(dd);
3330 abort();
3331 }
3332
3333 get_dev_size(fd, NULL, &size);
3334 size /= 512;
3335 serialcpy(dd->disk.serial, dd->serial);
3336 dd->disk.total_blocks = __cpu_to_le32(size);
3337 dd->disk.status = SPARE_DISK;
3338 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
3339 dd->disk.scsi_id = __cpu_to_le32(id);
3340 else
3341 dd->disk.scsi_id = __cpu_to_le32(0);
3342
3343 if (st->update_tail) {
3344 dd->next = super->add;
3345 super->add = dd;
3346 } else {
3347 dd->next = super->disks;
3348 super->disks = dd;
3349 }
3350
3351 return 0;
3352 }
3353
3354 static int store_imsm_mpb(int fd, struct imsm_super *mpb);
3355
3356 static union {
3357 char buf[512];
3358 struct imsm_super anchor;
3359 } spare_record __attribute__ ((aligned(512)));
3360
3361 /* spare records have their own family number and do not have any defined raid
3362 * devices
3363 */
3364 static int write_super_imsm_spares(struct intel_super *super, int doclose)
3365 {
3366 struct imsm_super *mpb = super->anchor;
3367 struct imsm_super *spare = &spare_record.anchor;
3368 __u32 sum;
3369 struct dl *d;
3370
3371 spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)),
3372 spare->generation_num = __cpu_to_le32(1UL),
3373 spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
3374 spare->num_disks = 1,
3375 spare->num_raid_devs = 0,
3376 spare->cache_size = mpb->cache_size,
3377 spare->pwr_cycle_count = __cpu_to_le32(1),
3378
3379 snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
3380 MPB_SIGNATURE MPB_VERSION_RAID0);
3381
3382 for (d = super->disks; d; d = d->next) {
3383 if (d->index != -1)
3384 continue;
3385
3386 spare->disk[0] = d->disk;
3387 sum = __gen_imsm_checksum(spare);
3388 spare->family_num = __cpu_to_le32(sum);
3389 spare->orig_family_num = 0;
3390 sum = __gen_imsm_checksum(spare);
3391 spare->check_sum = __cpu_to_le32(sum);
3392
3393 if (store_imsm_mpb(d->fd, spare)) {
3394 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3395 __func__, d->major, d->minor, strerror(errno));
3396 return 1;
3397 }
3398 if (doclose) {
3399 close(d->fd);
3400 d->fd = -1;
3401 }
3402 }
3403
3404 return 0;
3405 }
3406
3407 static int write_super_imsm(struct intel_super *super, int doclose)
3408 {
3409 struct imsm_super *mpb = super->anchor;
3410 struct dl *d;
3411 __u32 generation;
3412 __u32 sum;
3413 int spares = 0;
3414 int i;
3415 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
3416
3417 /* 'generation' is incremented everytime the metadata is written */
3418 generation = __le32_to_cpu(mpb->generation_num);
3419 generation++;
3420 mpb->generation_num = __cpu_to_le32(generation);
3421
3422 /* fix up cases where previous mdadm releases failed to set
3423 * orig_family_num
3424 */
3425 if (mpb->orig_family_num == 0)
3426 mpb->orig_family_num = mpb->family_num;
3427
3428 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
3429 for (d = super->disks; d; d = d->next) {
3430 if (d->index == -1)
3431 spares++;
3432 else
3433 mpb->disk[d->index] = d->disk;
3434 }
3435 for (d = super->missing; d; d = d->next)
3436 mpb->disk[d->index] = d->disk;
3437
3438 for (i = 0; i < mpb->num_raid_devs; i++) {
3439 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
3440
3441 imsm_copy_dev(dev, get_imsm_dev(super, i));
3442 mpb_size += sizeof_imsm_dev(dev, 0);
3443 }
3444 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
3445 mpb->mpb_size = __cpu_to_le32(mpb_size);
3446
3447 /* recalculate checksum */
3448 sum = __gen_imsm_checksum(mpb);
3449 mpb->check_sum = __cpu_to_le32(sum);
3450
3451 /* write the mpb for disks that compose raid devices */
3452 for (d = super->disks; d ; d = d->next) {
3453 if (d->index < 0)
3454 continue;
3455 if (store_imsm_mpb(d->fd, mpb))
3456 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3457 __func__, d->major, d->minor, strerror(errno));
3458 if (doclose) {
3459 close(d->fd);
3460 d->fd = -1;
3461 }
3462 }
3463
3464 if (spares)
3465 return write_super_imsm_spares(super, doclose);
3466
3467 return 0;
3468 }
3469
3470
3471 static int create_array(struct supertype *st, int dev_idx)
3472 {
3473 size_t len;
3474 struct imsm_update_create_array *u;
3475 struct intel_super *super = st->sb;
3476 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
3477 struct imsm_map *map = get_imsm_map(dev, 0);
3478 struct disk_info *inf;
3479 struct imsm_disk *disk;
3480 int i;
3481
3482 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
3483 sizeof(*inf) * map->num_members;
3484 u = malloc(len);
3485 if (!u) {
3486 fprintf(stderr, "%s: failed to allocate update buffer\n",
3487 __func__);
3488 return 1;
3489 }
3490
3491 u->type = update_create_array;
3492 u->dev_idx = dev_idx;
3493 imsm_copy_dev(&u->dev, dev);
3494 inf = get_disk_info(u);
3495 for (i = 0; i < map->num_members; i++) {
3496 int idx = get_imsm_disk_idx(dev, i);
3497
3498 disk = get_imsm_disk(super, idx);
3499 serialcpy(inf[i].serial, disk->serial);
3500 }
3501 append_metadata_update(st, u, len);
3502
3503 return 0;
3504 }
3505
3506 static int _add_disk(struct supertype *st)
3507 {
3508 struct intel_super *super = st->sb;
3509 size_t len;
3510 struct imsm_update_add_disk *u;
3511
3512 if (!super->add)
3513 return 0;
3514
3515 len = sizeof(*u);
3516 u = malloc(len);
3517 if (!u) {
3518 fprintf(stderr, "%s: failed to allocate update buffer\n",
3519 __func__);
3520 return 1;
3521 }
3522
3523 u->type = update_add_disk;
3524 append_metadata_update(st, u, len);
3525
3526 return 0;
3527 }
3528
3529 static int write_init_super_imsm(struct supertype *st)
3530 {
3531 struct intel_super *super = st->sb;
3532 int current_vol = super->current_vol;
3533
3534 /* we are done with current_vol reset it to point st at the container */
3535 super->current_vol = -1;
3536
3537 if (st->update_tail) {
3538 /* queue the recently created array / added disk
3539 * as a metadata update */
3540 struct dl *d;
3541 int rv;
3542
3543 /* determine if we are creating a volume or adding a disk */
3544 if (current_vol < 0) {
3545 /* in the add disk case we are running in mdmon
3546 * context, so don't close fd's
3547 */
3548 return _add_disk(st);
3549 } else
3550 rv = create_array(st, current_vol);
3551
3552 for (d = super->disks; d ; d = d->next) {
3553 close(d->fd);
3554 d->fd = -1;
3555 }
3556
3557 return rv;
3558 } else {
3559 struct dl *d;
3560 for (d = super->disks; d; d = d->next)
3561 Kill(d->devname, NULL, 0, 1, 1);
3562 return write_super_imsm(st->sb, 1);
3563 }
3564 }
3565 #endif
3566
3567 static int store_super_imsm(struct supertype *st, int fd)
3568 {
3569 struct intel_super *super = st->sb;
3570 struct imsm_super *mpb = super ? super->anchor : NULL;
3571
3572 if (!mpb)
3573 return 1;
3574
3575 #ifndef MDASSEMBLE
3576 return store_imsm_mpb(fd, mpb);
3577 #else
3578 return 1;
3579 #endif
3580 }
3581
3582 static int imsm_bbm_log_size(struct imsm_super *mpb)
3583 {
3584 return __le32_to_cpu(mpb->bbm_log_size);
3585 }
3586
3587 #ifndef MDASSEMBLE
3588 static int validate_geometry_imsm_container(struct supertype *st, int level,
3589 int layout, int raiddisks, int chunk,
3590 unsigned long long size, char *dev,
3591 unsigned long long *freesize,
3592 int verbose)
3593 {
3594 int fd;
3595 unsigned long long ldsize;
3596 const struct imsm_orom *orom;
3597
3598 if (level != LEVEL_CONTAINER)
3599 return 0;
3600 if (!dev)
3601 return 1;
3602
3603 if (check_env("IMSM_NO_PLATFORM"))
3604 orom = NULL;
3605 else
3606 orom = find_imsm_orom();
3607 if (orom && raiddisks > orom->tds) {
3608 if (verbose)
3609 fprintf(stderr, Name ": %d exceeds maximum number of"
3610 " platform supported disks: %d\n",
3611 raiddisks, orom->tds);
3612 return 0;
3613 }
3614
3615 fd = open(dev, O_RDONLY|O_EXCL, 0);
3616 if (fd < 0) {
3617 if (verbose)
3618 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
3619 dev, strerror(errno));
3620 return 0;
3621 }
3622 if (!get_dev_size(fd, dev, &ldsize)) {
3623 close(fd);
3624 return 0;
3625 }
3626 close(fd);
3627
3628 *freesize = avail_size_imsm(st, ldsize >> 9);
3629
3630 return 1;
3631 }
3632
3633 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
3634 {
3635 const unsigned long long base_start = e[*idx].start;
3636 unsigned long long end = base_start + e[*idx].size;
3637 int i;
3638
3639 if (base_start == end)
3640 return 0;
3641
3642 *idx = *idx + 1;
3643 for (i = *idx; i < num_extents; i++) {
3644 /* extend overlapping extents */
3645 if (e[i].start >= base_start &&
3646 e[i].start <= end) {
3647 if (e[i].size == 0)
3648 return 0;
3649 if (e[i].start + e[i].size > end)
3650 end = e[i].start + e[i].size;
3651 } else if (e[i].start > end) {
3652 *idx = i;
3653 break;
3654 }
3655 }
3656
3657 return end - base_start;
3658 }
3659
3660 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
3661 {
3662 /* build a composite disk with all known extents and generate a new
3663 * 'maxsize' given the "all disks in an array must share a common start
3664 * offset" constraint
3665 */
3666 struct extent *e = calloc(sum_extents, sizeof(*e));
3667 struct dl *dl;
3668 int i, j;
3669 int start_extent;
3670 unsigned long long pos;
3671 unsigned long long start = 0;
3672 unsigned long long maxsize;
3673 unsigned long reserve;
3674
3675 if (!e)
3676 return 0;
3677
3678 /* coalesce and sort all extents. also, check to see if we need to
3679 * reserve space between member arrays
3680 */
3681 j = 0;
3682 for (dl = super->disks; dl; dl = dl->next) {
3683 if (!dl->e)
3684 continue;
3685 for (i = 0; i < dl->extent_cnt; i++)
3686 e[j++] = dl->e[i];
3687 }
3688 qsort(e, sum_extents, sizeof(*e), cmp_extent);
3689
3690 /* merge extents */
3691 i = 0;
3692 j = 0;
3693 while (i < sum_extents) {
3694 e[j].start = e[i].start;
3695 e[j].size = find_size(e, &i, sum_extents);
3696 j++;
3697 if (e[j-1].size == 0)
3698 break;
3699 }
3700
3701 pos = 0;
3702 maxsize = 0;
3703 start_extent = 0;
3704 i = 0;
3705 do {
3706 unsigned long long esize;
3707
3708 esize = e[i].start - pos;
3709 if (esize >= maxsize) {
3710 maxsize = esize;
3711 start = pos;
3712 start_extent = i;
3713 }
3714 pos = e[i].start + e[i].size;
3715 i++;
3716 } while (e[i-1].size);
3717 free(e);
3718
3719 if (maxsize == 0)
3720 return 0;
3721
3722 /* FIXME assumes volume at offset 0 is the first volume in a
3723 * container
3724 */
3725 if (start_extent > 0)
3726 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
3727 else
3728 reserve = 0;
3729
3730 if (maxsize < reserve)
3731 return 0;
3732
3733 super->create_offset = ~((__u32) 0);
3734 if (start + reserve > super->create_offset)
3735 return 0; /* start overflows create_offset */
3736 super->create_offset = start + reserve;
3737
3738 return maxsize - reserve;
3739 }
3740
3741 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
3742 {
3743 if (level < 0 || level == 6 || level == 4)
3744 return 0;
3745
3746 /* if we have an orom prevent invalid raid levels */
3747 if (orom)
3748 switch (level) {
3749 case 0: return imsm_orom_has_raid0(orom);
3750 case 1:
3751 if (raiddisks > 2)
3752 return imsm_orom_has_raid1e(orom);
3753 return imsm_orom_has_raid1(orom) && raiddisks == 2;
3754 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
3755 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
3756 }
3757 else
3758 return 1; /* not on an Intel RAID platform so anything goes */
3759
3760 return 0;
3761 }
3762
3763 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3764 static int
3765 validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
3766 int raiddisks, int chunk, int verbose)
3767 {
3768 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
3769 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3770 level, raiddisks, raiddisks > 1 ? "s" : "");
3771 return 0;
3772 }
3773 if (super->orom && level != 1 &&
3774 !imsm_orom_has_chunk(super->orom, chunk)) {
3775 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
3776 return 0;
3777 }
3778 if (layout != imsm_level_to_layout(level)) {
3779 if (level == 5)
3780 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3781 else if (level == 10)
3782 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3783 else
3784 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3785 layout, level);
3786 return 0;
3787 }
3788
3789 return 1;
3790 }
3791
3792 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3793 * FIX ME add ahci details
3794 */
3795 static int validate_geometry_imsm_volume(struct supertype *st, int level,
3796 int layout, int raiddisks, int chunk,
3797 unsigned long long size, char *dev,
3798 unsigned long long *freesize,
3799 int verbose)
3800 {
3801 struct stat stb;
3802 struct intel_super *super = st->sb;
3803 struct imsm_super *mpb = super->anchor;
3804 struct dl *dl;
3805 unsigned long long pos = 0;
3806 unsigned long long maxsize;
3807 struct extent *e;
3808 int i;
3809
3810 /* We must have the container info already read in. */
3811 if (!super)
3812 return 0;
3813
3814 if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose))
3815 return 0;
3816
3817 if (!dev) {
3818 /* General test: make sure there is space for
3819 * 'raiddisks' device extents of size 'size' at a given
3820 * offset
3821 */
3822 unsigned long long minsize = size;
3823 unsigned long long start_offset = MaxSector;
3824 int dcnt = 0;
3825 if (minsize == 0)
3826 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
3827 for (dl = super->disks; dl ; dl = dl->next) {
3828 int found = 0;
3829
3830 pos = 0;
3831 i = 0;
3832 e = get_extents(super, dl);
3833 if (!e) continue;
3834 do {
3835 unsigned long long esize;
3836 esize = e[i].start - pos;
3837 if (esize >= minsize)
3838 found = 1;
3839 if (found && start_offset == MaxSector) {
3840 start_offset = pos;
3841 break;
3842 } else if (found && pos != start_offset) {
3843 found = 0;
3844 break;
3845 }
3846 pos = e[i].start + e[i].size;
3847 i++;
3848 } while (e[i-1].size);
3849 if (found)
3850 dcnt++;
3851 free(e);
3852 }
3853 if (dcnt < raiddisks) {
3854 if (verbose)
3855 fprintf(stderr, Name ": imsm: Not enough "
3856 "devices with space for this array "
3857 "(%d < %d)\n",
3858 dcnt, raiddisks);
3859 return 0;
3860 }
3861 return 1;
3862 }
3863
3864 /* This device must be a member of the set */
3865 if (stat(dev, &stb) < 0)
3866 return 0;
3867 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3868 return 0;
3869 for (dl = super->disks ; dl ; dl = dl->next) {
3870 if (dl->major == (int)major(stb.st_rdev) &&
3871 dl->minor == (int)minor(stb.st_rdev))
3872 break;
3873 }
3874 if (!dl) {
3875 if (verbose)
3876 fprintf(stderr, Name ": %s is not in the "
3877 "same imsm set\n", dev);
3878 return 0;
3879 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
3880 /* If a volume is present then the current creation attempt
3881 * cannot incorporate new spares because the orom may not
3882 * understand this configuration (all member disks must be
3883 * members of each array in the container).
3884 */
3885 fprintf(stderr, Name ": %s is a spare and a volume"
3886 " is already defined for this container\n", dev);
3887 fprintf(stderr, Name ": The option-rom requires all member"
3888 " disks to be a member of all volumes\n");
3889 return 0;
3890 }
3891
3892 /* retrieve the largest free space block */
3893 e = get_extents(super, dl);
3894 maxsize = 0;
3895 i = 0;
3896 if (e) {
3897 do {
3898 unsigned long long esize;
3899
3900 esize = e[i].start - pos;
3901 if (esize >= maxsize)
3902 maxsize = esize;
3903 pos = e[i].start + e[i].size;
3904 i++;
3905 } while (e[i-1].size);
3906 dl->e = e;
3907 dl->extent_cnt = i;
3908 } else {
3909 if (verbose)
3910 fprintf(stderr, Name ": unable to determine free space for: %s\n",
3911 dev);
3912 return 0;
3913 }
3914 if (maxsize < size) {
3915 if (verbose)
3916 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
3917 dev, maxsize, size);
3918 return 0;
3919 }
3920
3921 /* count total number of extents for merge */
3922 i = 0;
3923 for (dl = super->disks; dl; dl = dl->next)
3924 if (dl->e)
3925 i += dl->extent_cnt;
3926
3927 maxsize = merge_extents(super, i);
3928 if (maxsize < size || maxsize == 0) {
3929 if (verbose)
3930 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
3931 maxsize, size);
3932 return 0;
3933 }
3934
3935 *freesize = maxsize;
3936
3937 return 1;
3938 }
3939
3940 static int reserve_space(struct supertype *st, int raiddisks,
3941 unsigned long long size, int chunk,
3942 unsigned long long *freesize)
3943 {
3944 struct intel_super *super = st->sb;
3945 struct imsm_super *mpb = super->anchor;
3946 struct dl *dl;
3947 int i;
3948 int extent_cnt;
3949 struct extent *e;
3950 unsigned long long maxsize;
3951 unsigned long long minsize;
3952 int cnt;
3953 int used;
3954
3955 /* find the largest common start free region of the possible disks */
3956 used = 0;
3957 extent_cnt = 0;
3958 cnt = 0;
3959 for (dl = super->disks; dl; dl = dl->next) {
3960 dl->raiddisk = -1;
3961
3962 if (dl->index >= 0)
3963 used++;
3964
3965 /* don't activate new spares if we are orom constrained
3966 * and there is already a volume active in the container
3967 */
3968 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
3969 continue;
3970
3971 e = get_extents(super, dl);
3972 if (!e)
3973 continue;
3974 for (i = 1; e[i-1].size; i++)
3975 ;
3976 dl->e = e;
3977 dl->extent_cnt = i;
3978 extent_cnt += i;
3979 cnt++;
3980 }
3981
3982 maxsize = merge_extents(super, extent_cnt);
3983 minsize = size;
3984 if (size == 0)
3985 minsize = chunk;
3986
3987 if (cnt < raiddisks ||
3988 (super->orom && used && used != raiddisks) ||
3989 maxsize < minsize ||
3990 maxsize == 0) {
3991 fprintf(stderr, Name ": not enough devices with space to create array.\n");
3992 return 0; /* No enough free spaces large enough */
3993 }
3994
3995 if (size == 0) {
3996 size = maxsize;
3997 if (chunk) {
3998 size /= chunk;
3999 size *= chunk;
4000 }
4001 }
4002
4003 cnt = 0;
4004 for (dl = super->disks; dl; dl = dl->next)
4005 if (dl->e)
4006 dl->raiddisk = cnt++;
4007
4008 *freesize = size;
4009
4010 return 1;
4011 }
4012
4013 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
4014 int raiddisks, int chunk, unsigned long long size,
4015 char *dev, unsigned long long *freesize,
4016 int verbose)
4017 {
4018 int fd, cfd;
4019 struct mdinfo *sra;
4020 int is_member = 0;
4021
4022 /* if given unused devices create a container
4023 * if given given devices in a container create a member volume
4024 */
4025 if (level == LEVEL_CONTAINER) {
4026 /* Must be a fresh device to add to a container */
4027 return validate_geometry_imsm_container(st, level, layout,
4028 raiddisks, chunk, size,
4029 dev, freesize,
4030 verbose);
4031 }
4032
4033 if (!dev) {
4034 if (st->sb && freesize) {
4035 /* we are being asked to automatically layout a
4036 * new volume based on the current contents of
4037 * the container. If the the parameters can be
4038 * satisfied reserve_space will record the disks,
4039 * start offset, and size of the volume to be
4040 * created. add_to_super and getinfo_super
4041 * detect when autolayout is in progress.
4042 */
4043 if (!validate_geometry_imsm_orom(st->sb, level, layout,
4044 raiddisks, chunk,
4045 verbose))
4046 return 0;
4047 return reserve_space(st, raiddisks, size, chunk, freesize);
4048 }
4049 return 1;
4050 }
4051 if (st->sb) {
4052 /* creating in a given container */
4053 return validate_geometry_imsm_volume(st, level, layout,
4054 raiddisks, chunk, size,
4055 dev, freesize, verbose);
4056 }
4057
4058 /* This device needs to be a device in an 'imsm' container */
4059 fd = open(dev, O_RDONLY|O_EXCL, 0);
4060 if (fd >= 0) {
4061 if (verbose)
4062 fprintf(stderr,
4063 Name ": Cannot create this array on device %s\n",
4064 dev);
4065 close(fd);
4066 return 0;
4067 }
4068 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
4069 if (verbose)
4070 fprintf(stderr, Name ": Cannot open %s: %s\n",
4071 dev, strerror(errno));
4072 return 0;
4073 }
4074 /* Well, it is in use by someone, maybe an 'imsm' container. */
4075 cfd = open_container(fd);
4076 close(fd);
4077 if (cfd < 0) {
4078 if (verbose)
4079 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
4080 dev);
4081 return 0;
4082 }
4083 sra = sysfs_read(cfd, 0, GET_VERSION);
4084 if (sra && sra->array.major_version == -1 &&
4085 strcmp(sra->text_version, "imsm") == 0)
4086 is_member = 1;
4087 sysfs_free(sra);
4088 if (is_member) {
4089 /* This is a member of a imsm container. Load the container
4090 * and try to create a volume
4091 */
4092 struct intel_super *super;
4093
4094 if (load_super_imsm_all(st, cfd, (void **) &super, NULL) == 0) {
4095 st->sb = super;
4096 st->container_dev = fd2devnum(cfd);
4097 close(cfd);
4098 return validate_geometry_imsm_volume(st, level, layout,
4099 raiddisks, chunk,
4100 size, dev,
4101 freesize, verbose);
4102 }
4103 }
4104
4105 if (verbose)
4106 fprintf(stderr, Name ": failed container membership check\n");
4107
4108 close(cfd);
4109 return 0;
4110 }
4111
4112 static int default_chunk_imsm(struct supertype *st)
4113 {
4114 struct intel_super *super = st->sb;
4115
4116 if (!super->orom)
4117 return 0;
4118
4119 return imsm_orom_default_chunk(super->orom);
4120 }
4121
4122 static void handle_missing(struct intel_super *super, struct imsm_dev *dev);
4123
4124 static int kill_subarray_imsm(struct supertype *st)
4125 {
4126 /* remove the subarray currently referenced by ->current_vol */
4127 __u8 i;
4128 struct intel_dev **dp;
4129 struct intel_super *super = st->sb;
4130 __u8 current_vol = super->current_vol;
4131 struct imsm_super *mpb = super->anchor;
4132
4133 if (super->current_vol < 0)
4134 return 2;
4135 super->current_vol = -1; /* invalidate subarray cursor */
4136
4137 /* block deletions that would change the uuid of active subarrays
4138 *
4139 * FIXME when immutable ids are available, but note that we'll
4140 * also need to fixup the invalidated/active subarray indexes in
4141 * mdstat
4142 */
4143 for (i = 0; i < mpb->num_raid_devs; i++) {
4144 char subarray[4];
4145
4146 if (i < current_vol)
4147 continue;
4148 sprintf(subarray, "%u", i);
4149 if (is_subarray_active(subarray, st->devname)) {
4150 fprintf(stderr,
4151 Name ": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4152 current_vol, i);
4153
4154 return 2;
4155 }
4156 }
4157
4158 if (st->update_tail) {
4159 struct imsm_update_kill_array *u = malloc(sizeof(*u));
4160
4161 if (!u)
4162 return 2;
4163 u->type = update_kill_array;
4164 u->dev_idx = current_vol;
4165 append_metadata_update(st, u, sizeof(*u));
4166
4167 return 0;
4168 }
4169
4170 for (dp = &super->devlist; *dp;)
4171 if ((*dp)->index == current_vol) {
4172 *dp = (*dp)->next;
4173 } else {
4174 handle_missing(super, (*dp)->dev);
4175 if ((*dp)->index > current_vol)
4176 (*dp)->index--;
4177 dp = &(*dp)->next;
4178 }
4179
4180 /* no more raid devices, all active components are now spares,
4181 * but of course failed are still failed
4182 */
4183 if (--mpb->num_raid_devs == 0) {
4184 struct dl *d;
4185
4186 for (d = super->disks; d; d = d->next)
4187 if (d->index > -2) {
4188 d->index = -1;
4189 d->disk.status = SPARE_DISK;
4190 }
4191 }
4192
4193 super->updates_pending++;
4194
4195 return 0;
4196 }
4197
4198 static int update_subarray_imsm(struct supertype *st, char *subarray,
4199 char *update, mddev_ident_t ident)
4200 {
4201 /* update the subarray currently referenced by ->current_vol */
4202 struct intel_super *super = st->sb;
4203 struct imsm_super *mpb = super->anchor;
4204
4205 if (strcmp(update, "name") == 0) {
4206 char *name = ident->name;
4207 char *ep;
4208 int vol;
4209
4210 if (is_subarray_active(subarray, st->devname)) {
4211 fprintf(stderr,
4212 Name ": Unable to update name of active subarray\n");
4213 return 2;
4214 }
4215
4216 if (!check_name(super, name, 0))
4217 return 2;
4218
4219 vol = strtoul(subarray, &ep, 10);
4220 if (*ep != '\0' || vol >= super->anchor->num_raid_devs)
4221 return 2;
4222
4223 if (st->update_tail) {
4224 struct imsm_update_rename_array *u = malloc(sizeof(*u));
4225
4226 if (!u)
4227 return 2;
4228 u->type = update_rename_array;
4229 u->dev_idx = vol;
4230 snprintf((char *) u->name, MAX_RAID_SERIAL_LEN, "%s", name);
4231 append_metadata_update(st, u, sizeof(*u));
4232 } else {
4233 struct imsm_dev *dev;
4234 int i;
4235
4236 dev = get_imsm_dev(super, vol);
4237 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
4238 for (i = 0; i < mpb->num_raid_devs; i++) {
4239 dev = get_imsm_dev(super, i);
4240 handle_missing(super, dev);
4241 }
4242 super->updates_pending++;
4243 }
4244 } else
4245 return 2;
4246
4247 return 0;
4248 }
4249 #endif /* MDASSEMBLE */
4250
4251 static int is_rebuilding(struct imsm_dev *dev)
4252 {
4253 struct imsm_map *migr_map;
4254
4255 if (!dev->vol.migr_state)
4256 return 0;
4257
4258 if (migr_type(dev) != MIGR_REBUILD)
4259 return 0;
4260
4261 migr_map = get_imsm_map(dev, 1);
4262
4263 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
4264 return 1;
4265 else
4266 return 0;
4267 }
4268
4269 static void update_recovery_start(struct imsm_dev *dev, struct mdinfo *array)
4270 {
4271 struct mdinfo *rebuild = NULL;
4272 struct mdinfo *d;
4273 __u32 units;
4274
4275 if (!is_rebuilding(dev))
4276 return;
4277
4278 /* Find the rebuild target, but punt on the dual rebuild case */
4279 for (d = array->devs; d; d = d->next)
4280 if (d->recovery_start == 0) {
4281 if (rebuild)
4282 return;
4283 rebuild = d;
4284 }
4285
4286 if (!rebuild) {
4287 /* (?) none of the disks are marked with
4288 * IMSM_ORD_REBUILD, so assume they are missing and the
4289 * disk_ord_tbl was not correctly updated
4290 */
4291 dprintf("%s: failed to locate out-of-sync disk\n", __func__);
4292 return;
4293 }
4294
4295 units = __le32_to_cpu(dev->vol.curr_migr_unit);
4296 rebuild->recovery_start = units * blocks_per_migr_unit(dev);
4297 }
4298
4299
4300 static struct mdinfo *container_content_imsm(struct supertype *st, char *subarray)
4301 {
4302 /* Given a container loaded by load_super_imsm_all,
4303 * extract information about all the arrays into
4304 * an mdinfo tree.
4305 * If 'subarray' is given, just extract info about that array.
4306 *
4307 * For each imsm_dev create an mdinfo, fill it in,
4308 * then look for matching devices in super->disks
4309 * and create appropriate device mdinfo.
4310 */
4311 struct intel_super *super = st->sb;
4312 struct imsm_super *mpb = super->anchor;
4313 struct mdinfo *rest = NULL;
4314 unsigned int i;
4315
4316 /* do not assemble arrays that might have bad blocks */
4317 if (imsm_bbm_log_size(super->anchor)) {
4318 fprintf(stderr, Name ": BBM log found in metadata. "
4319 "Cannot activate array(s).\n");
4320 return NULL;
4321 }
4322
4323 for (i = 0; i < mpb->num_raid_devs; i++) {
4324 struct imsm_dev *dev;
4325 struct imsm_map *map;
4326 struct mdinfo *this;
4327 int slot;
4328 char *ep;
4329
4330 if (subarray &&
4331 (i != strtoul(subarray, &ep, 10) || *ep != '\0'))
4332 continue;
4333
4334 dev = get_imsm_dev(super, i);
4335 map = get_imsm_map(dev, 0);
4336
4337 /* do not publish arrays that are in the middle of an
4338 * unsupported migration
4339 */
4340 if (dev->vol.migr_state &&
4341 (migr_type(dev) == MIGR_GEN_MIGR ||
4342 migr_type(dev) == MIGR_STATE_CHANGE)) {
4343 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
4344 " unsupported migration in progress\n",
4345 dev->volume);
4346 continue;
4347 }
4348
4349 this = malloc(sizeof(*this));
4350 if (!this) {
4351 fprintf(stderr, Name ": failed to allocate %zu bytes\n",
4352 sizeof(*this));
4353 break;
4354 }
4355 memset(this, 0, sizeof(*this));
4356 this->next = rest;
4357
4358 super->current_vol = i;
4359 getinfo_super_imsm_volume(st, this, NULL);
4360 for (slot = 0 ; slot < map->num_members; slot++) {
4361 unsigned long long recovery_start;
4362 struct mdinfo *info_d;
4363 struct dl *d;
4364 int idx;
4365 int skip;
4366 __u32 ord;
4367
4368 skip = 0;
4369 idx = get_imsm_disk_idx(dev, slot);
4370 ord = get_imsm_ord_tbl_ent(dev, slot);
4371 for (d = super->disks; d ; d = d->next)
4372 if (d->index == idx)
4373 break;
4374
4375 recovery_start = MaxSector;
4376 if (d == NULL)
4377 skip = 1;
4378 if (d && is_failed(&d->disk))
4379 skip = 1;
4380 if (ord & IMSM_ORD_REBUILD)
4381 recovery_start = 0;
4382
4383 /*
4384 * if we skip some disks the array will be assmebled degraded;
4385 * reset resync start to avoid a dirty-degraded
4386 * situation when performing the intial sync
4387 *
4388 * FIXME handle dirty degraded
4389 */
4390 if ((skip || recovery_start == 0) && !dev->vol.dirty)
4391 this->resync_start = MaxSector;
4392 if (skip)
4393 continue;
4394
4395 info_d = calloc(1, sizeof(*info_d));
4396 if (!info_d) {
4397 fprintf(stderr, Name ": failed to allocate disk"
4398 " for volume %.16s\n", dev->volume);
4399 info_d = this->devs;
4400 while (info_d) {
4401 struct mdinfo *d = info_d->next;
4402
4403 free(info_d);
4404 info_d = d;
4405 }
4406 free(this);
4407 this = rest;
4408 break;
4409 }
4410 info_d->next = this->devs;
4411 this->devs = info_d;
4412
4413 info_d->disk.number = d->index;
4414 info_d->disk.major = d->major;
4415 info_d->disk.minor = d->minor;
4416 info_d->disk.raid_disk = slot;
4417 info_d->recovery_start = recovery_start;
4418
4419 if (info_d->recovery_start == MaxSector)
4420 this->array.working_disks++;
4421
4422 info_d->events = __le32_to_cpu(mpb->generation_num);
4423 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
4424 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
4425 }
4426 /* now that the disk list is up-to-date fixup recovery_start */
4427 update_recovery_start(dev, this);
4428 rest = this;
4429 }
4430
4431 return rest;
4432 }
4433
4434
4435 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
4436 {
4437 struct imsm_map *map = get_imsm_map(dev, 0);
4438
4439 if (!failed)
4440 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
4441 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
4442
4443 switch (get_imsm_raid_level(map)) {
4444 case 0:
4445 return IMSM_T_STATE_FAILED;
4446 break;
4447 case 1:
4448 if (failed < map->num_members)
4449 return IMSM_T_STATE_DEGRADED;
4450 else
4451 return IMSM_T_STATE_FAILED;
4452 break;
4453 case 10:
4454 {
4455 /**
4456 * check to see if any mirrors have failed, otherwise we
4457 * are degraded. Even numbered slots are mirrored on
4458 * slot+1
4459 */
4460 int i;
4461 /* gcc -Os complains that this is unused */
4462 int insync = insync;
4463
4464 for (i = 0; i < map->num_members; i++) {
4465 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
4466 int idx = ord_to_idx(ord);
4467 struct imsm_disk *disk;
4468
4469 /* reset the potential in-sync count on even-numbered
4470 * slots. num_copies is always 2 for imsm raid10
4471 */
4472 if ((i & 1) == 0)
4473 insync = 2;
4474
4475 disk = get_imsm_disk(super, idx);
4476 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
4477 insync--;
4478
4479 /* no in-sync disks left in this mirror the
4480 * array has failed
4481 */
4482 if (insync == 0)
4483 return IMSM_T_STATE_FAILED;
4484 }
4485
4486 return IMSM_T_STATE_DEGRADED;
4487 }
4488 case 5:
4489 if (failed < 2)
4490 return IMSM_T_STATE_DEGRADED;
4491 else
4492 return IMSM_T_STATE_FAILED;
4493 break;
4494 default:
4495 break;
4496 }
4497
4498 return map->map_state;
4499 }
4500
4501 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
4502 {
4503 int i;
4504 int failed = 0;
4505 struct imsm_disk *disk;
4506 struct imsm_map *map = get_imsm_map(dev, 0);
4507 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
4508 __u32 ord;
4509 int idx;
4510
4511 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4512 * disks that are being rebuilt. New failures are recorded to
4513 * map[0]. So we look through all the disks we started with and
4514 * see if any failures are still present, or if any new ones
4515 * have arrived
4516 *
4517 * FIXME add support for online capacity expansion and
4518 * raid-level-migration
4519 */
4520 for (i = 0; i < prev->num_members; i++) {
4521 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
4522 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
4523 idx = ord_to_idx(ord);
4524
4525 disk = get_imsm_disk(super, idx);
4526 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
4527 failed++;
4528 }
4529
4530 return failed;
4531 }
4532
4533 #ifndef MDASSEMBLE
4534 static int imsm_open_new(struct supertype *c, struct active_array *a,
4535 char *inst)
4536 {
4537 struct intel_super *super = c->sb;
4538 struct imsm_super *mpb = super->anchor;
4539
4540 if (atoi(inst) >= mpb->num_raid_devs) {
4541 fprintf(stderr, "%s: subarry index %d, out of range\n",
4542 __func__, atoi(inst));
4543 return -ENODEV;
4544 }
4545
4546 dprintf("imsm: open_new %s\n", inst);
4547 a->info.container_member = atoi(inst);
4548 return 0;
4549 }
4550
4551 static int is_resyncing(struct imsm_dev *dev)
4552 {
4553 struct imsm_map *migr_map;
4554
4555 if (!dev->vol.migr_state)
4556 return 0;
4557
4558 if (migr_type(dev) == MIGR_INIT ||
4559 migr_type(dev) == MIGR_REPAIR)
4560 return 1;
4561
4562 migr_map = get_imsm_map(dev, 1);
4563
4564 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
4565 return 1;
4566 else
4567 return 0;
4568 }
4569
4570 /* return true if we recorded new information */
4571 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4572 {
4573 __u32 ord;
4574 int slot;
4575 struct imsm_map *map;
4576
4577 /* new failures are always set in map[0] */
4578 map = get_imsm_map(dev, 0);
4579
4580 slot = get_imsm_disk_slot(map, idx);
4581 if (slot < 0)
4582 return 0;
4583
4584 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
4585 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
4586 return 0;
4587
4588 disk->status |= FAILED_DISK;
4589 disk->status &= ~CONFIGURED_DISK;
4590 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
4591 if (map->failed_disk_num == 0xff)
4592 map->failed_disk_num = slot;
4593 return 1;
4594 }
4595
4596 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4597 {
4598 mark_failure(dev, disk, idx);
4599
4600 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
4601 return;
4602
4603 disk->scsi_id = __cpu_to_le32(~(__u32)0);
4604 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
4605 }
4606
4607 static void handle_missing(struct intel_super *super, struct imsm_dev *dev)
4608 {
4609 __u8 map_state;
4610 struct dl *dl;
4611 int failed;
4612
4613 if (!super->missing)
4614 return;
4615 failed = imsm_count_failed(super, dev);
4616 map_state = imsm_check_degraded(super, dev, failed);
4617
4618 dprintf("imsm: mark missing\n");
4619 end_migration(dev, map_state);
4620 for (dl = super->missing; dl; dl = dl->next)
4621 mark_missing(dev, &dl->disk, dl->index);
4622 super->updates_pending++;
4623 }
4624
4625 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4626 * states are handled in imsm_set_disk() with one exception, when a
4627 * resync is stopped due to a new failure this routine will set the
4628 * 'degraded' state for the array.
4629 */
4630 static int imsm_set_array_state(struct active_array *a, int consistent)
4631 {
4632 int inst = a->info.container_member;
4633 struct intel_super *super = a->container->sb;
4634 struct imsm_dev *dev = get_imsm_dev(super, inst);
4635 struct imsm_map *map = get_imsm_map(dev, 0);
4636 int failed = imsm_count_failed(super, dev);
4637 __u8 map_state = imsm_check_degraded(super, dev, failed);
4638 __u32 blocks_per_unit;
4639
4640 /* before we activate this array handle any missing disks */
4641 if (consistent == 2)
4642 handle_missing(super, dev);
4643
4644 if (consistent == 2 &&
4645 (!is_resync_complete(&a->info) ||
4646 map_state != IMSM_T_STATE_NORMAL ||
4647 dev->vol.migr_state))
4648 consistent = 0;
4649
4650 if (is_resync_complete(&a->info)) {
4651 /* complete intialization / resync,
4652 * recovery and interrupted recovery is completed in
4653 * ->set_disk
4654 */
4655 if (is_resyncing(dev)) {
4656 dprintf("imsm: mark resync done\n");
4657 end_migration(dev, map_state);
4658 super->updates_pending++;
4659 a->last_checkpoint = 0;
4660 }
4661 } else if (!is_resyncing(dev) && !failed) {
4662 /* mark the start of the init process if nothing is failed */
4663 dprintf("imsm: mark resync start\n");
4664 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
4665 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
4666 else
4667 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
4668 super->updates_pending++;
4669 }
4670
4671 /* check if we can update curr_migr_unit from resync_start, recovery_start */
4672 blocks_per_unit = blocks_per_migr_unit(dev);
4673 if (blocks_per_unit) {
4674 __u32 units32;
4675 __u64 units;
4676
4677 units = a->last_checkpoint / blocks_per_unit;
4678 units32 = units;
4679
4680 /* check that we did not overflow 32-bits, and that
4681 * curr_migr_unit needs updating
4682 */
4683 if (units32 == units &&
4684 __le32_to_cpu(dev->vol.curr_migr_unit) != units32) {
4685 dprintf("imsm: mark checkpoint (%u)\n", units32);
4686 dev->vol.curr_migr_unit = __cpu_to_le32(units32);
4687 super->updates_pending++;
4688 }
4689 }
4690
4691 /* mark dirty / clean */
4692 if (dev->vol.dirty != !consistent) {
4693 dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty");
4694 if (consistent)
4695 dev->vol.dirty = 0;
4696 else
4697 dev->vol.dirty = 1;
4698 super->updates_pending++;
4699 }
4700 return consistent;
4701 }
4702
4703 static void imsm_set_disk(struct active_array *a, int n, int state)
4704 {
4705 int inst = a->info.container_member;
4706 struct intel_super *super = a->container->sb;
4707 struct imsm_dev *dev = get_imsm_dev(super, inst);
4708 struct imsm_map *map = get_imsm_map(dev, 0);
4709 struct imsm_disk *disk;
4710 int failed;
4711 __u32 ord;
4712 __u8 map_state;
4713
4714 if (n > map->num_members)
4715 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
4716 n, map->num_members - 1);
4717
4718 if (n < 0)
4719 return;
4720
4721 dprintf("imsm: set_disk %d:%x\n", n, state);
4722
4723 ord = get_imsm_ord_tbl_ent(dev, n);
4724 disk = get_imsm_disk(super, ord_to_idx(ord));
4725
4726 /* check for new failures */
4727 if (state & DS_FAULTY) {
4728 if (mark_failure(dev, disk, ord_to_idx(ord)))
4729 super->updates_pending++;
4730 }
4731
4732 /* check if in_sync */
4733 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
4734 struct imsm_map *migr_map = get_imsm_map(dev, 1);
4735
4736 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
4737 super->updates_pending++;
4738 }
4739
4740 failed = imsm_count_failed(super, dev);
4741 map_state = imsm_check_degraded(super, dev, failed);
4742
4743 /* check if recovery complete, newly degraded, or failed */
4744 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
4745 end_migration(dev, map_state);
4746 map = get_imsm_map(dev, 0);
4747 map->failed_disk_num = ~0;
4748 super->updates_pending++;
4749 a->last_checkpoint = 0;
4750 } else if (map_state == IMSM_T_STATE_DEGRADED &&
4751 map->map_state != map_state &&
4752 !dev->vol.migr_state) {
4753 dprintf("imsm: mark degraded\n");
4754 map->map_state = map_state;
4755 super->updates_pending++;
4756 a->last_checkpoint = 0;
4757 } else if (map_state == IMSM_T_STATE_FAILED &&
4758 map->map_state != map_state) {
4759 dprintf("imsm: mark failed\n");
4760 end_migration(dev, map_state);
4761 super->updates_pending++;
4762 a->last_checkpoint = 0;
4763 }
4764 }
4765
4766 static int store_imsm_mpb(int fd, struct imsm_super *mpb)
4767 {
4768 void *buf = mpb;
4769 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
4770 unsigned long long dsize;
4771 unsigned long long sectors;
4772
4773 get_dev_size(fd, NULL, &dsize);
4774
4775 if (mpb_size > 512) {
4776 /* -1 to account for anchor */
4777 sectors = mpb_sectors(mpb) - 1;
4778
4779 /* write the extended mpb to the sectors preceeding the anchor */
4780 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
4781 return 1;
4782
4783 if ((unsigned long long)write(fd, buf + 512, 512 * sectors)
4784 != 512 * sectors)
4785 return 1;
4786 }
4787
4788 /* first block is stored on second to last sector of the disk */
4789 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
4790 return 1;
4791
4792 if (write(fd, buf, 512) != 512)
4793 return 1;
4794
4795 return 0;
4796 }
4797
4798 static void imsm_sync_metadata(struct supertype *container)
4799 {
4800 struct intel_super *super = container->sb;
4801
4802 if (!super->updates_pending)
4803 return;
4804
4805 write_super_imsm(super, 0);
4806
4807 super->updates_pending = 0;
4808 }
4809
4810 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
4811 {
4812 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4813 int i = get_imsm_disk_idx(dev, idx);
4814 struct dl *dl;
4815
4816 for (dl = super->disks; dl; dl = dl->next)
4817 if (dl->index == i)
4818 break;
4819
4820 if (dl && is_failed(&dl->disk))
4821 dl = NULL;
4822
4823 if (dl)
4824 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
4825
4826 return dl;
4827 }
4828
4829 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
4830 struct active_array *a, int activate_new)
4831 {
4832 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4833 int idx = get_imsm_disk_idx(dev, slot);
4834 struct imsm_super *mpb = super->anchor;
4835 struct imsm_map *map;
4836 unsigned long long pos;
4837 struct mdinfo *d;
4838 struct extent *ex;
4839 int i, j;
4840 int found;
4841 __u32 array_start = 0;
4842 __u32 array_end = 0;
4843 struct dl *dl;
4844
4845 for (dl = super->disks; dl; dl = dl->next) {
4846 /* If in this array, skip */
4847 for (d = a->info.devs ; d ; d = d->next)
4848 if (d->state_fd >= 0 &&
4849 d->disk.major == dl->major &&
4850 d->disk.minor == dl->minor) {
4851 dprintf("%x:%x already in array\n", dl->major, dl->minor);
4852 break;
4853 }
4854 if (d)
4855 continue;
4856
4857 /* skip in use or failed drives */
4858 if (is_failed(&dl->disk) || idx == dl->index ||
4859 dl->index == -2) {
4860 dprintf("%x:%x status (failed: %d index: %d)\n",
4861 dl->major, dl->minor, is_failed(&dl->disk), idx);
4862 continue;
4863 }
4864
4865 /* skip pure spares when we are looking for partially
4866 * assimilated drives
4867 */
4868 if (dl->index == -1 && !activate_new)
4869 continue;
4870
4871 /* Does this unused device have the requisite free space?
4872 * It needs to be able to cover all member volumes
4873 */
4874 ex = get_extents(super, dl);
4875 if (!ex) {
4876 dprintf("cannot get extents\n");
4877 continue;
4878 }
4879 for (i = 0; i < mpb->num_raid_devs; i++) {
4880 dev = get_imsm_dev(super, i);
4881 map = get_imsm_map(dev, 0);
4882
4883 /* check if this disk is already a member of
4884 * this array
4885 */
4886 if (get_imsm_disk_slot(map, dl->index) >= 0)
4887 continue;
4888
4889 found = 0;
4890 j = 0;
4891 pos = 0;
4892 array_start = __le32_to_cpu(map->pba_of_lba0);
4893 array_end = array_start +
4894 __le32_to_cpu(map->blocks_per_member) - 1;
4895
4896 do {
4897 /* check that we can start at pba_of_lba0 with
4898 * blocks_per_member of space
4899 */
4900 if (array_start >= pos && array_end < ex[j].start) {
4901 found = 1;
4902 break;
4903 }
4904 pos = ex[j].start + ex[j].size;
4905 j++;
4906 } while (ex[j-1].size);
4907
4908 if (!found)
4909 break;
4910 }
4911
4912 free(ex);
4913 if (i < mpb->num_raid_devs) {
4914 dprintf("%x:%x does not have %u to %u available\n",
4915 dl->major, dl->minor, array_start, array_end);
4916 /* No room */
4917 continue;
4918 }
4919 return dl;
4920 }
4921
4922 return dl;
4923 }
4924
4925 static struct mdinfo *imsm_activate_spare(struct active_array *a,
4926 struct metadata_update **updates)
4927 {
4928 /**
4929 * Find a device with unused free space and use it to replace a
4930 * failed/vacant region in an array. We replace failed regions one a
4931 * array at a time. The result is that a new spare disk will be added
4932 * to the first failed array and after the monitor has finished
4933 * propagating failures the remainder will be consumed.
4934 *
4935 * FIXME add a capability for mdmon to request spares from another
4936 * container.
4937 */
4938
4939 struct intel_super *super = a->container->sb;
4940 int inst = a->info.container_member;
4941 struct imsm_dev *dev = get_imsm_dev(super, inst);
4942 struct imsm_map *map = get_imsm_map(dev, 0);
4943 int failed = a->info.array.raid_disks;
4944 struct mdinfo *rv = NULL;
4945 struct mdinfo *d;
4946 struct mdinfo *di;
4947 struct metadata_update *mu;
4948 struct dl *dl;
4949 struct imsm_update_activate_spare *u;
4950 int num_spares = 0;
4951 int i;
4952
4953 for (d = a->info.devs ; d ; d = d->next) {
4954 if ((d->curr_state & DS_FAULTY) &&
4955 d->state_fd >= 0)
4956 /* wait for Removal to happen */
4957 return NULL;
4958 if (d->state_fd >= 0)
4959 failed--;
4960 }
4961
4962 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4963 inst, failed, a->info.array.raid_disks, a->info.array.level);
4964 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
4965 return NULL;
4966
4967 /* For each slot, if it is not working, find a spare */
4968 for (i = 0; i < a->info.array.raid_disks; i++) {
4969 for (d = a->info.devs ; d ; d = d->next)
4970 if (d->disk.raid_disk == i)
4971 break;
4972 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
4973 if (d && (d->state_fd >= 0))
4974 continue;
4975
4976 /*
4977 * OK, this device needs recovery. Try to re-add the
4978 * previous occupant of this slot, if this fails see if
4979 * we can continue the assimilation of a spare that was
4980 * partially assimilated, finally try to activate a new
4981 * spare.
4982 */
4983 dl = imsm_readd(super, i, a);
4984 if (!dl)
4985 dl = imsm_add_spare(super, i, a, 0);
4986 if (!dl)
4987 dl = imsm_add_spare(super, i, a, 1);
4988 if (!dl)
4989 continue;
4990
4991 /* found a usable disk with enough space */
4992 di = malloc(sizeof(*di));
4993 if (!di)
4994 continue;
4995 memset(di, 0, sizeof(*di));
4996
4997 /* dl->index will be -1 in the case we are activating a
4998 * pristine spare. imsm_process_update() will create a
4999 * new index in this case. Once a disk is found to be
5000 * failed in all member arrays it is kicked from the
5001 * metadata
5002 */
5003 di->disk.number = dl->index;
5004
5005 /* (ab)use di->devs to store a pointer to the device
5006 * we chose
5007 */
5008 di->devs = (struct mdinfo *) dl;
5009
5010 di->disk.raid_disk = i;
5011 di->disk.major = dl->major;
5012 di->disk.minor = dl->minor;
5013 di->disk.state = 0;
5014 di->recovery_start = 0;
5015 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
5016 di->component_size = a->info.component_size;
5017 di->container_member = inst;
5018 super->random = random32();
5019 di->next = rv;
5020 rv = di;
5021 num_spares++;
5022 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
5023 i, di->data_offset);
5024
5025 break;
5026 }
5027
5028 if (!rv)
5029 /* No spares found */
5030 return rv;
5031 /* Now 'rv' has a list of devices to return.
5032 * Create a metadata_update record to update the
5033 * disk_ord_tbl for the array
5034 */
5035 mu = malloc(sizeof(*mu));
5036 if (mu) {
5037 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
5038 if (mu->buf == NULL) {
5039 free(mu);
5040 mu = NULL;
5041 }
5042 }
5043 if (!mu) {
5044 while (rv) {
5045 struct mdinfo *n = rv->next;
5046
5047 free(rv);
5048 rv = n;
5049 }
5050 return NULL;
5051 }
5052
5053 mu->space = NULL;
5054 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
5055 mu->next = *updates;
5056 u = (struct imsm_update_activate_spare *) mu->buf;
5057
5058 for (di = rv ; di ; di = di->next) {
5059 u->type = update_activate_spare;
5060 u->dl = (struct dl *) di->devs;
5061 di->devs = NULL;
5062 u->slot = di->disk.raid_disk;
5063 u->array = inst;
5064 u->next = u + 1;
5065 u++;
5066 }
5067 (u-1)->next = NULL;
5068 *updates = mu;
5069
5070 return rv;
5071 }
5072
5073 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
5074 {
5075 struct imsm_dev *dev = get_imsm_dev(super, idx);
5076 struct imsm_map *map = get_imsm_map(dev, 0);
5077 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
5078 struct disk_info *inf = get_disk_info(u);
5079 struct imsm_disk *disk;
5080 int i;
5081 int j;
5082
5083 for (i = 0; i < map->num_members; i++) {
5084 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
5085 for (j = 0; j < new_map->num_members; j++)
5086 if (serialcmp(disk->serial, inf[j].serial) == 0)
5087 return 1;
5088 }
5089
5090 return 0;
5091 }
5092
5093 static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index);
5094
5095 static void imsm_process_update(struct supertype *st,
5096 struct metadata_update *update)
5097 {
5098 /**
5099 * crack open the metadata_update envelope to find the update record
5100 * update can be one of:
5101 * update_activate_spare - a spare device has replaced a failed
5102 * device in an array, update the disk_ord_tbl. If this disk is
5103 * present in all member arrays then also clear the SPARE_DISK
5104 * flag
5105 */
5106 struct intel_super *super = st->sb;
5107 struct imsm_super *mpb;
5108 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
5109
5110 /* update requires a larger buf but the allocation failed */
5111 if (super->next_len && !super->next_buf) {
5112 super->next_len = 0;
5113 return;
5114 }
5115
5116 if (super->next_buf) {
5117 memcpy(super->next_buf, super->buf, super->len);
5118 free(super->buf);
5119 super->len = super->next_len;
5120 super->buf = super->next_buf;
5121
5122 super->next_len = 0;
5123 super->next_buf = NULL;
5124 }
5125
5126 mpb = super->anchor;
5127
5128 switch (type) {
5129 case update_activate_spare: {
5130 struct imsm_update_activate_spare *u = (void *) update->buf;
5131 struct imsm_dev *dev = get_imsm_dev(super, u->array);
5132 struct imsm_map *map = get_imsm_map(dev, 0);
5133 struct imsm_map *migr_map;
5134 struct active_array *a;
5135 struct imsm_disk *disk;
5136 __u8 to_state;
5137 struct dl *dl;
5138 unsigned int found;
5139 int failed;
5140 int victim = get_imsm_disk_idx(dev, u->slot);
5141 int i;
5142
5143 for (dl = super->disks; dl; dl = dl->next)
5144 if (dl == u->dl)
5145 break;
5146
5147 if (!dl) {
5148 fprintf(stderr, "error: imsm_activate_spare passed "
5149 "an unknown disk (index: %d)\n",
5150 u->dl->index);
5151 return;
5152 }
5153
5154 super->updates_pending++;
5155
5156 /* count failures (excluding rebuilds and the victim)
5157 * to determine map[0] state
5158 */
5159 failed = 0;
5160 for (i = 0; i < map->num_members; i++) {
5161 if (i == u->slot)
5162 continue;
5163 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
5164 if (!disk || is_failed(disk))
5165 failed++;
5166 }
5167
5168 /* adding a pristine spare, assign a new index */
5169 if (dl->index < 0) {
5170 dl->index = super->anchor->num_disks;
5171 super->anchor->num_disks++;
5172 }
5173 disk = &dl->disk;
5174 disk->status |= CONFIGURED_DISK;
5175 disk->status &= ~SPARE_DISK;
5176
5177 /* mark rebuild */
5178 to_state = imsm_check_degraded(super, dev, failed);
5179 map->map_state = IMSM_T_STATE_DEGRADED;
5180 migrate(dev, to_state, MIGR_REBUILD);
5181 migr_map = get_imsm_map(dev, 1);
5182 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
5183 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
5184
5185 /* update the family_num to mark a new container
5186 * generation, being careful to record the existing
5187 * family_num in orig_family_num to clean up after
5188 * earlier mdadm versions that neglected to set it.
5189 */
5190 if (mpb->orig_family_num == 0)
5191 mpb->orig_family_num = mpb->family_num;
5192 mpb->family_num += super->random;
5193
5194 /* count arrays using the victim in the metadata */
5195 found = 0;
5196 for (a = st->arrays; a ; a = a->next) {
5197 dev = get_imsm_dev(super, a->info.container_member);
5198 map = get_imsm_map(dev, 0);
5199
5200 if (get_imsm_disk_slot(map, victim) >= 0)
5201 found++;
5202 }
5203
5204 /* delete the victim if it is no longer being
5205 * utilized anywhere
5206 */
5207 if (!found) {
5208 struct dl **dlp;
5209
5210 /* We know that 'manager' isn't touching anything,
5211 * so it is safe to delete
5212 */
5213 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
5214 if ((*dlp)->index == victim)
5215 break;
5216
5217 /* victim may be on the missing list */
5218 if (!*dlp)
5219 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
5220 if ((*dlp)->index == victim)
5221 break;
5222 imsm_delete(super, dlp, victim);
5223 }
5224 break;
5225 }
5226 case update_create_array: {
5227 /* someone wants to create a new array, we need to be aware of
5228 * a few races/collisions:
5229 * 1/ 'Create' called by two separate instances of mdadm
5230 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5231 * devices that have since been assimilated via
5232 * activate_spare.
5233 * In the event this update can not be carried out mdadm will
5234 * (FIX ME) notice that its update did not take hold.
5235 */
5236 struct imsm_update_create_array *u = (void *) update->buf;
5237 struct intel_dev *dv;
5238 struct imsm_dev *dev;
5239 struct imsm_map *map, *new_map;
5240 unsigned long long start, end;
5241 unsigned long long new_start, new_end;
5242 int i;
5243 struct disk_info *inf;
5244 struct dl *dl;
5245
5246 /* handle racing creates: first come first serve */
5247 if (u->dev_idx < mpb->num_raid_devs) {
5248 dprintf("%s: subarray %d already defined\n",
5249 __func__, u->dev_idx);
5250 goto create_error;
5251 }
5252
5253 /* check update is next in sequence */
5254 if (u->dev_idx != mpb->num_raid_devs) {
5255 dprintf("%s: can not create array %d expected index %d\n",
5256 __func__, u->dev_idx, mpb->num_raid_devs);
5257 goto create_error;
5258 }
5259
5260 new_map = get_imsm_map(&u->dev, 0);
5261 new_start = __le32_to_cpu(new_map->pba_of_lba0);
5262 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
5263 inf = get_disk_info(u);
5264
5265 /* handle activate_spare versus create race:
5266 * check to make sure that overlapping arrays do not include
5267 * overalpping disks
5268 */
5269 for (i = 0; i < mpb->num_raid_devs; i++) {
5270 dev = get_imsm_dev(super, i);
5271 map = get_imsm_map(dev, 0);
5272 start = __le32_to_cpu(map->pba_of_lba0);
5273 end = start + __le32_to_cpu(map->blocks_per_member);
5274 if ((new_start >= start && new_start <= end) ||
5275 (start >= new_start && start <= new_end))
5276 /* overlap */;
5277 else
5278 continue;
5279
5280 if (disks_overlap(super, i, u)) {
5281 dprintf("%s: arrays overlap\n", __func__);
5282 goto create_error;
5283 }
5284 }
5285
5286 /* check that prepare update was successful */
5287 if (!update->space) {
5288 dprintf("%s: prepare update failed\n", __func__);
5289 goto create_error;
5290 }
5291
5292 /* check that all disks are still active before committing
5293 * changes. FIXME: could we instead handle this by creating a
5294 * degraded array? That's probably not what the user expects,
5295 * so better to drop this update on the floor.
5296 */
5297 for (i = 0; i < new_map->num_members; i++) {
5298 dl = serial_to_dl(inf[i].serial, super);
5299 if (!dl) {
5300 dprintf("%s: disk disappeared\n", __func__);
5301 goto create_error;
5302 }
5303 }
5304
5305 super->updates_pending++;
5306
5307 /* convert spares to members and fixup ord_tbl */
5308 for (i = 0; i < new_map->num_members; i++) {
5309 dl = serial_to_dl(inf[i].serial, super);
5310 if (dl->index == -1) {
5311 dl->index = mpb->num_disks;
5312 mpb->num_disks++;
5313 dl->disk.status |= CONFIGURED_DISK;
5314 dl->disk.status &= ~SPARE_DISK;
5315 }
5316 set_imsm_ord_tbl_ent(new_map, i, dl->index);
5317 }
5318
5319 dv = update->space;
5320 dev = dv->dev;
5321 update->space = NULL;
5322 imsm_copy_dev(dev, &u->dev);
5323 dv->index = u->dev_idx;
5324 dv->next = super->devlist;
5325 super->devlist = dv;
5326 mpb->num_raid_devs++;
5327
5328 imsm_update_version_info(super);
5329 break;
5330 create_error:
5331 /* mdmon knows how to release update->space, but not
5332 * ((struct intel_dev *) update->space)->dev
5333 */
5334 if (update->space) {
5335 dv = update->space;
5336 free(dv->dev);
5337 }
5338 break;
5339 }
5340 case update_kill_array: {
5341 struct imsm_update_kill_array *u = (void *) update->buf;
5342 int victim = u->dev_idx;
5343 struct active_array *a;
5344 struct intel_dev **dp;
5345 struct imsm_dev *dev;
5346
5347 /* sanity check that we are not affecting the uuid of
5348 * active arrays, or deleting an active array
5349 *
5350 * FIXME when immutable ids are available, but note that
5351 * we'll also need to fixup the invalidated/active
5352 * subarray indexes in mdstat
5353 */
5354 for (a = st->arrays; a; a = a->next)
5355 if (a->info.container_member >= victim)
5356 break;
5357 /* by definition if mdmon is running at least one array
5358 * is active in the container, so checking
5359 * mpb->num_raid_devs is just extra paranoia
5360 */
5361 dev = get_imsm_dev(super, victim);
5362 if (a || !dev || mpb->num_raid_devs == 1) {
5363 dprintf("failed to delete subarray-%d\n", victim);
5364 break;
5365 }
5366
5367 for (dp = &super->devlist; *dp;)
5368 if ((*dp)->index == (unsigned)super->current_vol) {
5369 *dp = (*dp)->next;
5370 } else {
5371 if ((*dp)->index > (unsigned)victim)
5372 (*dp)->index--;
5373 dp = &(*dp)->next;
5374 }
5375 mpb->num_raid_devs--;
5376 super->updates_pending++;
5377 break;
5378 }
5379 case update_rename_array: {
5380 struct imsm_update_rename_array *u = (void *) update->buf;
5381 char name[MAX_RAID_SERIAL_LEN+1];
5382 int target = u->dev_idx;
5383 struct active_array *a;
5384 struct imsm_dev *dev;
5385
5386 /* sanity check that we are not affecting the uuid of
5387 * an active array
5388 */
5389 snprintf(name, MAX_RAID_SERIAL_LEN, "%s", (char *) u->name);
5390 name[MAX_RAID_SERIAL_LEN] = '\0';
5391 for (a = st->arrays; a; a = a->next)
5392 if (a->info.container_member == target)
5393 break;
5394 dev = get_imsm_dev(super, u->dev_idx);
5395 if (a || !dev || !check_name(super, name, 1)) {
5396 dprintf("failed to rename subarray-%d\n", target);
5397 break;
5398 }
5399
5400 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
5401 super->updates_pending++;
5402 break;
5403 }
5404 case update_add_disk:
5405
5406 /* we may be able to repair some arrays if disks are
5407 * being added */
5408 if (super->add) {
5409 struct active_array *a;
5410
5411 super->updates_pending++;
5412 for (a = st->arrays; a; a = a->next)
5413 a->check_degraded = 1;
5414 }
5415 /* add some spares to the metadata */
5416 while (super->add) {
5417 struct dl *al;
5418
5419 al = super->add;
5420 super->add = al->next;
5421 al->next = super->disks;
5422 super->disks = al;
5423 dprintf("%s: added %x:%x\n",
5424 __func__, al->major, al->minor);
5425 }
5426
5427 break;
5428 }
5429 }
5430
5431 static void imsm_prepare_update(struct supertype *st,
5432 struct metadata_update *update)
5433 {
5434 /**
5435 * Allocate space to hold new disk entries, raid-device entries or a new
5436 * mpb if necessary. The manager synchronously waits for updates to
5437 * complete in the monitor, so new mpb buffers allocated here can be
5438 * integrated by the monitor thread without worrying about live pointers
5439 * in the manager thread.
5440 */
5441 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
5442 struct intel_super *super = st->sb;
5443 struct imsm_super *mpb = super->anchor;
5444 size_t buf_len;
5445 size_t len = 0;
5446
5447 switch (type) {
5448 case update_create_array: {
5449 struct imsm_update_create_array *u = (void *) update->buf;
5450 struct intel_dev *dv;
5451 struct imsm_dev *dev = &u->dev;
5452 struct imsm_map *map = get_imsm_map(dev, 0);
5453 struct dl *dl;
5454 struct disk_info *inf;
5455 int i;
5456 int activate = 0;
5457
5458 inf = get_disk_info(u);
5459 len = sizeof_imsm_dev(dev, 1);
5460 /* allocate a new super->devlist entry */
5461 dv = malloc(sizeof(*dv));
5462 if (dv) {
5463 dv->dev = malloc(len);
5464 if (dv->dev)
5465 update->space = dv;
5466 else {
5467 free(dv);
5468 update->space = NULL;
5469 }
5470 }
5471
5472 /* count how many spares will be converted to members */
5473 for (i = 0; i < map->num_members; i++) {
5474 dl = serial_to_dl(inf[i].serial, super);
5475 if (!dl) {
5476 /* hmm maybe it failed?, nothing we can do about
5477 * it here
5478 */
5479 continue;
5480 }
5481 if (count_memberships(dl, super) == 0)
5482 activate++;
5483 }
5484 len += activate * sizeof(struct imsm_disk);
5485 break;
5486 default:
5487 break;
5488 }
5489 }
5490
5491 /* check if we need a larger metadata buffer */
5492 if (super->next_buf)
5493 buf_len = super->next_len;
5494 else
5495 buf_len = super->len;
5496
5497 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
5498 /* ok we need a larger buf than what is currently allocated
5499 * if this allocation fails process_update will notice that
5500 * ->next_len is set and ->next_buf is NULL
5501 */
5502 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
5503 if (super->next_buf)
5504 free(super->next_buf);
5505
5506 super->next_len = buf_len;
5507 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
5508 memset(super->next_buf, 0, buf_len);
5509 else
5510 super->next_buf = NULL;
5511 }
5512 }
5513
5514 /* must be called while manager is quiesced */
5515 static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index)
5516 {
5517 struct imsm_super *mpb = super->anchor;
5518 struct dl *iter;
5519 struct imsm_dev *dev;
5520 struct imsm_map *map;
5521 int i, j, num_members;
5522 __u32 ord;
5523
5524 dprintf("%s: deleting device[%d] from imsm_super\n",
5525 __func__, index);
5526
5527 /* shift all indexes down one */
5528 for (iter = super->disks; iter; iter = iter->next)
5529 if (iter->index > (int)index)
5530 iter->index--;
5531 for (iter = super->missing; iter; iter = iter->next)
5532 if (iter->index > (int)index)
5533 iter->index--;
5534
5535 for (i = 0; i < mpb->num_raid_devs; i++) {
5536 dev = get_imsm_dev(super, i);
5537 map = get_imsm_map(dev, 0);
5538 num_members = map->num_members;
5539 for (j = 0; j < num_members; j++) {
5540 /* update ord entries being careful not to propagate
5541 * ord-flags to the first map
5542 */
5543 ord = get_imsm_ord_tbl_ent(dev, j);
5544
5545 if (ord_to_idx(ord) <= index)
5546 continue;
5547
5548 map = get_imsm_map(dev, 0);
5549 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
5550 map = get_imsm_map(dev, 1);
5551 if (map)
5552 set_imsm_ord_tbl_ent(map, j, ord - 1);
5553 }
5554 }
5555
5556 mpb->num_disks--;
5557 super->updates_pending++;
5558 if (*dlp) {
5559 struct dl *dl = *dlp;
5560
5561 *dlp = (*dlp)->next;
5562 __free_imsm_disk(dl);
5563 }
5564 }
5565 #endif /* MDASSEMBLE */
5566
5567 struct superswitch super_imsm = {
5568 #ifndef MDASSEMBLE
5569 .examine_super = examine_super_imsm,
5570 .brief_examine_super = brief_examine_super_imsm,
5571 .brief_examine_subarrays = brief_examine_subarrays_imsm,
5572 .export_examine_super = export_examine_super_imsm,
5573 .detail_super = detail_super_imsm,
5574 .brief_detail_super = brief_detail_super_imsm,
5575 .write_init_super = write_init_super_imsm,
5576 .validate_geometry = validate_geometry_imsm,
5577 .default_chunk = default_chunk_imsm,
5578 .add_to_super = add_to_super_imsm,
5579 .detail_platform = detail_platform_imsm,
5580 .kill_subarray = kill_subarray_imsm,
5581 .update_subarray = update_subarray_imsm,
5582 .load_container = load_container_imsm,
5583 #endif
5584 .match_home = match_home_imsm,
5585 .uuid_from_super= uuid_from_super_imsm,
5586 .getinfo_super = getinfo_super_imsm,
5587 .update_super = update_super_imsm,
5588
5589 .avail_size = avail_size_imsm,
5590
5591 .compare_super = compare_super_imsm,
5592
5593 .load_super = load_super_imsm,
5594 .init_super = init_super_imsm,
5595 .store_super = store_super_imsm,
5596 .free_super = free_super_imsm,
5597 .match_metadata_desc = match_metadata_desc_imsm,
5598 .container_content = container_content_imsm,
5599 .default_layout = imsm_level_to_layout,
5600
5601 .external = 1,
5602 .name = "imsm",
5603
5604 #ifndef MDASSEMBLE
5605 /* for mdmon */
5606 .open_new = imsm_open_new,
5607 .set_array_state= imsm_set_array_state,
5608 .set_disk = imsm_set_disk,
5609 .sync_metadata = imsm_sync_metadata,
5610 .activate_spare = imsm_activate_spare,
5611 .process_update = imsm_process_update,
5612 .prepare_update = imsm_prepare_update,
5613 #endif /* MDASSEMBLE */
5614 };
Unnamed repository; edit this file 'description' to name the repository.