]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-intel.c
projects / thirdparty / mdadm.git / blob
? search:


fbde7e69ccd0a2af125fa71f6bbba6f771c5cf7d
[thirdparty/mdadm.git] / super-intel.c
1 /*
2 * mdadm - Intel(R) Matrix Storage Manager Support
3 *
4 * Copyright (C) 2002-2008 Intel Corporation
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define HAVE_STDINT_H 1
21 #include "mdadm.h"
22 #include "mdmon.h"
23 #include "sha1.h"
24 #include "platform-intel.h"
25 #include <values.h>
26 #include <scsi/sg.h>
27 #include <ctype.h>
28 #include <dirent.h>
29
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
43
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
53
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56 #define SECT_PER_MB_SHIFT 11
57
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
60 struct imsm_disk {
61 __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 __u32 status; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
71 };
72
73 /* RAID map configuration infos. */
74 struct imsm_map {
75 __u32 pba_of_lba0; /* start address of partition */
76 __u32 blocks_per_member;/* blocks per member */
77 __u32 num_data_stripes; /* number of data stripes */
78 __u16 blocks_per_strip;
79 __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
84 __u8 raid_level;
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members; /* number of member disks */
89 __u8 num_domains; /* number of parity domains */
90 __u8 failed_disk_num; /* valid only when state is degraded */
91 __u8 ddf;
92 __u32 filler[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
96 */
97 } __attribute__ ((packed));
98
99 struct imsm_vol {
100 __u32 curr_migr_unit;
101 __u32 checkpoint_id; /* id to access curr_migr_unit */
102 __u8 migr_state; /* Normal or Migrating */
103 #define MIGR_INIT 0
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type; /* Initializing, Rebuilding, ... */
110 __u8 dirty;
111 __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors; /* number of mismatches */
113 __u16 bad_blocks; /* number of bad blocks during verify */
114 __u32 filler[4];
115 struct imsm_map map[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed));
118
119 struct imsm_dev {
120 __u8 volume[MAX_RAID_SERIAL_LEN];
121 __u32 size_low;
122 __u32 size_high;
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status; /* Persistent RaidDev status */
137 __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
138 __u8 migr_priority;
139 __u8 num_sub_vols;
140 __u8 tid;
141 __u8 cng_master_disk;
142 __u16 cache_policy;
143 __u8 cng_state;
144 __u8 cng_sub_state;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler[IMSM_DEV_FILLERS];
147 struct imsm_vol vol;
148 } __attribute__ ((packed));
149
150 struct imsm_super {
151 __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
152 __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size; /* 0x30 - 0x33 in bytes */
157 __u32 attributes; /* 0x34 - 0x37 */
158 __u8 num_disks; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos; /* 0x3A */
161 __u8 fill[1]; /* 0x3B */
162 __u32 cache_size; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed));
172
173 #define BBM_LOG_MAX_ENTRIES 254
174
175 struct bbm_log_entry {
176 __u64 defective_block_start;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset;
179 __u16 remapped_marked_count;
180 __u16 disk_ordinal;
181 } __attribute__ ((__packed__));
182
183 struct bbm_log {
184 __u32 signature; /* 0xABADB10C */
185 __u32 entry_count;
186 __u32 reserved_spare_block_count; /* 0 */
187 __u32 reserved; /* 0xFFFF */
188 __u64 first_spare_lba;
189 struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
190 } __attribute__ ((__packed__));
191
192
193 #ifndef MDASSEMBLE
194 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
195 #endif
196
197 static __u8 migr_type(struct imsm_dev *dev)
198 {
199 if (dev->vol.migr_type == MIGR_VERIFY &&
200 dev->status & DEV_VERIFY_AND_FIX)
201 return MIGR_REPAIR;
202 else
203 return dev->vol.migr_type;
204 }
205
206 static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
207 {
208 /* for compatibility with older oroms convert MIGR_REPAIR, into
209 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
210 */
211 if (migr_type == MIGR_REPAIR) {
212 dev->vol.migr_type = MIGR_VERIFY;
213 dev->status |= DEV_VERIFY_AND_FIX;
214 } else {
215 dev->vol.migr_type = migr_type;
216 dev->status &= ~DEV_VERIFY_AND_FIX;
217 }
218 }
219
220 static unsigned int sector_count(__u32 bytes)
221 {
222 return ((bytes + (512-1)) & (~(512-1))) / 512;
223 }
224
225 static unsigned int mpb_sectors(struct imsm_super *mpb)
226 {
227 return sector_count(__le32_to_cpu(mpb->mpb_size));
228 }
229
230 struct intel_dev {
231 struct imsm_dev *dev;
232 struct intel_dev *next;
233 int index;
234 };
235
236 /* internal representation of IMSM metadata */
237 struct intel_super {
238 union {
239 void *buf; /* O_DIRECT buffer for reading/writing metadata */
240 struct imsm_super *anchor; /* immovable parameters */
241 };
242 size_t len; /* size of the 'buf' allocation */
243 void *next_buf; /* for realloc'ing buf from the manager */
244 size_t next_len;
245 int updates_pending; /* count of pending updates for mdmon */
246 int creating_imsm; /* flag to indicate container creation */
247 int current_vol; /* index of raid device undergoing creation */
248 __u32 create_offset; /* common start for 'current_vol' */
249 __u32 random; /* random data for seeding new family numbers */
250 struct intel_dev *devlist;
251 struct dl {
252 struct dl *next;
253 int index;
254 __u8 serial[MAX_RAID_SERIAL_LEN];
255 int major, minor;
256 char *devname;
257 struct imsm_disk disk;
258 int fd;
259 int extent_cnt;
260 struct extent *e; /* for determining freespace @ create */
261 int raiddisk; /* slot to fill in autolayout */
262 } *disks;
263 struct dl *add; /* list of disks to add while mdmon active */
264 struct dl *missing; /* disks removed while we weren't looking */
265 struct bbm_log *bbm_log;
266 const char *hba; /* device path of the raid controller for this metadata */
267 const struct imsm_orom *orom; /* platform firmware support */
268 struct intel_super *next; /* (temp) list for disambiguating family_num */
269 };
270
271 struct intel_disk {
272 struct imsm_disk disk;
273 #define IMSM_UNKNOWN_OWNER (-1)
274 int owner;
275 struct intel_disk *next;
276 };
277
278 struct extent {
279 unsigned long long start, size;
280 };
281
282 /* definition of messages passed to imsm_process_update */
283 enum imsm_update_type {
284 update_activate_spare,
285 update_create_array,
286 update_add_disk,
287 };
288
289 struct imsm_update_activate_spare {
290 enum imsm_update_type type;
291 struct dl *dl;
292 int slot;
293 int array;
294 struct imsm_update_activate_spare *next;
295 };
296
297 struct disk_info {
298 __u8 serial[MAX_RAID_SERIAL_LEN];
299 };
300
301 struct imsm_update_create_array {
302 enum imsm_update_type type;
303 int dev_idx;
304 struct imsm_dev dev;
305 };
306
307 struct imsm_update_add_disk {
308 enum imsm_update_type type;
309 };
310
311 static struct supertype *match_metadata_desc_imsm(char *arg)
312 {
313 struct supertype *st;
314
315 if (strcmp(arg, "imsm") != 0 &&
316 strcmp(arg, "default") != 0
317 )
318 return NULL;
319
320 st = malloc(sizeof(*st));
321 if (!st)
322 return NULL;
323 memset(st, 0, sizeof(*st));
324 st->ss = &super_imsm;
325 st->max_devs = IMSM_MAX_DEVICES;
326 st->minor_version = 0;
327 st->sb = NULL;
328 return st;
329 }
330
331 #ifndef MDASSEMBLE
332 static __u8 *get_imsm_version(struct imsm_super *mpb)
333 {
334 return &mpb->sig[MPB_SIG_LEN];
335 }
336 #endif
337
338 /* retrieve a disk directly from the anchor when the anchor is known to be
339 * up-to-date, currently only at load time
340 */
341 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
342 {
343 if (index >= mpb->num_disks)
344 return NULL;
345 return &mpb->disk[index];
346 }
347
348 #ifndef MDASSEMBLE
349 /* retrieve a disk from the parsed metadata */
350 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
351 {
352 struct dl *d;
353
354 for (d = super->disks; d; d = d->next)
355 if (d->index == index)
356 return &d->disk;
357
358 return NULL;
359 }
360 #endif
361
362 /* generate a checksum directly from the anchor when the anchor is known to be
363 * up-to-date, currently only at load or write_super after coalescing
364 */
365 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
366 {
367 __u32 end = mpb->mpb_size / sizeof(end);
368 __u32 *p = (__u32 *) mpb;
369 __u32 sum = 0;
370
371 while (end--) {
372 sum += __le32_to_cpu(*p);
373 p++;
374 }
375
376 return sum - __le32_to_cpu(mpb->check_sum);
377 }
378
379 static size_t sizeof_imsm_map(struct imsm_map *map)
380 {
381 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
382 }
383
384 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
385 {
386 struct imsm_map *map = &dev->vol.map[0];
387
388 if (second_map && !dev->vol.migr_state)
389 return NULL;
390 else if (second_map) {
391 void *ptr = map;
392
393 return ptr + sizeof_imsm_map(map);
394 } else
395 return map;
396
397 }
398
399 /* return the size of the device.
400 * migr_state increases the returned size if map[0] were to be duplicated
401 */
402 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
403 {
404 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
405 sizeof_imsm_map(get_imsm_map(dev, 0));
406
407 /* migrating means an additional map */
408 if (dev->vol.migr_state)
409 size += sizeof_imsm_map(get_imsm_map(dev, 1));
410 else if (migr_state)
411 size += sizeof_imsm_map(get_imsm_map(dev, 0));
412
413 return size;
414 }
415
416 #ifndef MDASSEMBLE
417 /* retrieve disk serial number list from a metadata update */
418 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
419 {
420 void *u = update;
421 struct disk_info *inf;
422
423 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
424 sizeof_imsm_dev(&update->dev, 0);
425
426 return inf;
427 }
428 #endif
429
430 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
431 {
432 int offset;
433 int i;
434 void *_mpb = mpb;
435
436 if (index >= mpb->num_raid_devs)
437 return NULL;
438
439 /* devices start after all disks */
440 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
441
442 for (i = 0; i <= index; i++)
443 if (i == index)
444 return _mpb + offset;
445 else
446 offset += sizeof_imsm_dev(_mpb + offset, 0);
447
448 return NULL;
449 }
450
451 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
452 {
453 struct intel_dev *dv;
454
455 if (index >= super->anchor->num_raid_devs)
456 return NULL;
457 for (dv = super->devlist; dv; dv = dv->next)
458 if (dv->index == index)
459 return dv->dev;
460 return NULL;
461 }
462
463 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, int slot)
464 {
465 struct imsm_map *map;
466
467 if (dev->vol.migr_state)
468 map = get_imsm_map(dev, 1);
469 else
470 map = get_imsm_map(dev, 0);
471
472 /* top byte identifies disk under rebuild */
473 return __le32_to_cpu(map->disk_ord_tbl[slot]);
474 }
475
476 #define ord_to_idx(ord) (((ord) << 8) >> 8)
477 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot)
478 {
479 __u32 ord = get_imsm_ord_tbl_ent(dev, slot);
480
481 return ord_to_idx(ord);
482 }
483
484 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
485 {
486 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
487 }
488
489 static int get_imsm_disk_slot(struct imsm_map *map, int idx)
490 {
491 int slot;
492 __u32 ord;
493
494 for (slot = 0; slot < map->num_members; slot++) {
495 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
496 if (ord_to_idx(ord) == idx)
497 return slot;
498 }
499
500 return -1;
501 }
502
503 static int get_imsm_raid_level(struct imsm_map *map)
504 {
505 if (map->raid_level == 1) {
506 if (map->num_members == 2)
507 return 1;
508 else
509 return 10;
510 }
511
512 return map->raid_level;
513 }
514
515 static int cmp_extent(const void *av, const void *bv)
516 {
517 const struct extent *a = av;
518 const struct extent *b = bv;
519 if (a->start < b->start)
520 return -1;
521 if (a->start > b->start)
522 return 1;
523 return 0;
524 }
525
526 static int count_memberships(struct dl *dl, struct intel_super *super)
527 {
528 int memberships = 0;
529 int i;
530
531 for (i = 0; i < super->anchor->num_raid_devs; i++) {
532 struct imsm_dev *dev = get_imsm_dev(super, i);
533 struct imsm_map *map = get_imsm_map(dev, 0);
534
535 if (get_imsm_disk_slot(map, dl->index) >= 0)
536 memberships++;
537 }
538
539 return memberships;
540 }
541
542 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
543 {
544 /* find a list of used extents on the given physical device */
545 struct extent *rv, *e;
546 int i;
547 int memberships = count_memberships(dl, super);
548 __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
549
550 rv = malloc(sizeof(struct extent) * (memberships + 1));
551 if (!rv)
552 return NULL;
553 e = rv;
554
555 for (i = 0; i < super->anchor->num_raid_devs; i++) {
556 struct imsm_dev *dev = get_imsm_dev(super, i);
557 struct imsm_map *map = get_imsm_map(dev, 0);
558
559 if (get_imsm_disk_slot(map, dl->index) >= 0) {
560 e->start = __le32_to_cpu(map->pba_of_lba0);
561 e->size = __le32_to_cpu(map->blocks_per_member);
562 e++;
563 }
564 }
565 qsort(rv, memberships, sizeof(*rv), cmp_extent);
566
567 /* determine the start of the metadata
568 * when no raid devices are defined use the default
569 * ...otherwise allow the metadata to truncate the value
570 * as is the case with older versions of imsm
571 */
572 if (memberships) {
573 struct extent *last = &rv[memberships - 1];
574 __u32 remainder;
575
576 remainder = __le32_to_cpu(dl->disk.total_blocks) -
577 (last->start + last->size);
578 /* round down to 1k block to satisfy precision of the kernel
579 * 'size' interface
580 */
581 remainder &= ~1UL;
582 /* make sure remainder is still sane */
583 if (remainder < ROUND_UP(super->len, 512) >> 9)
584 remainder = ROUND_UP(super->len, 512) >> 9;
585 if (reservation > remainder)
586 reservation = remainder;
587 }
588 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
589 e->size = 0;
590 return rv;
591 }
592
593 /* try to determine how much space is reserved for metadata from
594 * the last get_extents() entry, otherwise fallback to the
595 * default
596 */
597 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
598 {
599 struct extent *e;
600 int i;
601 __u32 rv;
602
603 /* for spares just return a minimal reservation which will grow
604 * once the spare is picked up by an array
605 */
606 if (dl->index == -1)
607 return MPB_SECTOR_CNT;
608
609 e = get_extents(super, dl);
610 if (!e)
611 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
612
613 /* scroll to last entry */
614 for (i = 0; e[i].size; i++)
615 continue;
616
617 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
618
619 free(e);
620
621 return rv;
622 }
623
624 static int is_spare(struct imsm_disk *disk)
625 {
626 return (disk->status & SPARE_DISK) == SPARE_DISK;
627 }
628
629 static int is_configured(struct imsm_disk *disk)
630 {
631 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
632 }
633
634 static int is_failed(struct imsm_disk *disk)
635 {
636 return (disk->status & FAILED_DISK) == FAILED_DISK;
637 }
638
639 #ifndef MDASSEMBLE
640 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
641 {
642 __u64 sz;
643 int slot;
644 struct imsm_map *map = get_imsm_map(dev, 0);
645 __u32 ord;
646
647 printf("\n");
648 printf("[%.16s]:\n", dev->volume);
649 printf(" UUID : %s\n", uuid);
650 printf(" RAID Level : %d\n", get_imsm_raid_level(map));
651 printf(" Members : %d\n", map->num_members);
652 slot = get_imsm_disk_slot(map, disk_idx);
653 if (slot >= 0) {
654 ord = get_imsm_ord_tbl_ent(dev, slot);
655 printf(" This Slot : %d%s\n", slot,
656 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
657 } else
658 printf(" This Slot : ?\n");
659 sz = __le32_to_cpu(dev->size_high);
660 sz <<= 32;
661 sz += __le32_to_cpu(dev->size_low);
662 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
663 human_size(sz * 512));
664 sz = __le32_to_cpu(map->blocks_per_member);
665 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
666 human_size(sz * 512));
667 printf(" Sector Offset : %u\n",
668 __le32_to_cpu(map->pba_of_lba0));
669 printf(" Num Stripes : %u\n",
670 __le32_to_cpu(map->num_data_stripes));
671 printf(" Chunk Size : %u KiB\n",
672 __le16_to_cpu(map->blocks_per_strip) / 2);
673 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
674 printf(" Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle\n");
675 if (dev->vol.migr_state) {
676 if (migr_type(dev) == MIGR_INIT)
677 printf(": initializing\n");
678 else if (migr_type(dev) == MIGR_REBUILD)
679 printf(": rebuilding\n");
680 else if (migr_type(dev) == MIGR_VERIFY)
681 printf(": check\n");
682 else if (migr_type(dev) == MIGR_GEN_MIGR)
683 printf(": general migration\n");
684 else if (migr_type(dev) == MIGR_STATE_CHANGE)
685 printf(": state change\n");
686 else if (migr_type(dev) == MIGR_REPAIR)
687 printf(": repair\n");
688 else
689 printf(": <unknown:%d>\n", migr_type(dev));
690 }
691 printf(" Map State : %s", map_state_str[map->map_state]);
692 if (dev->vol.migr_state) {
693 struct imsm_map *map = get_imsm_map(dev, 1);
694 printf(" <-- %s", map_state_str[map->map_state]);
695 }
696 printf("\n");
697 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
698 }
699
700 static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved)
701 {
702 struct imsm_disk *disk = __get_imsm_disk(mpb, index);
703 char str[MAX_RAID_SERIAL_LEN + 1];
704 __u64 sz;
705
706 if (index < 0 || !disk)
707 return;
708
709 printf("\n");
710 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
711 printf(" Disk%02d Serial : %s\n", index, str);
712 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
713 is_configured(disk) ? " active" : "",
714 is_failed(disk) ? " failed" : "");
715 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
716 sz = __le32_to_cpu(disk->total_blocks) - reserved;
717 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
718 human_size(sz * 512));
719 }
720
721 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info);
722
723 static void examine_super_imsm(struct supertype *st, char *homehost)
724 {
725 struct intel_super *super = st->sb;
726 struct imsm_super *mpb = super->anchor;
727 char str[MAX_SIGNATURE_LENGTH];
728 int i;
729 struct mdinfo info;
730 char nbuf[64];
731 __u32 sum;
732 __u32 reserved = imsm_reserved_sectors(super, super->disks);
733
734
735 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
736 printf(" Magic : %s\n", str);
737 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
738 printf(" Version : %s\n", get_imsm_version(mpb));
739 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
740 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
741 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
742 getinfo_super_imsm(st, &info);
743 fname_from_uuid(st, &info, nbuf, ':');
744 printf(" UUID : %s\n", nbuf + 5);
745 sum = __le32_to_cpu(mpb->check_sum);
746 printf(" Checksum : %08x %s\n", sum,
747 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
748 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
749 printf(" Disks : %d\n", mpb->num_disks);
750 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
751 print_imsm_disk(mpb, super->disks->index, reserved);
752 if (super->bbm_log) {
753 struct bbm_log *log = super->bbm_log;
754
755 printf("\n");
756 printf("Bad Block Management Log:\n");
757 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
758 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
759 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
760 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
761 printf(" First Spare : %llx\n",
762 (unsigned long long) __le64_to_cpu(log->first_spare_lba));
763 }
764 for (i = 0; i < mpb->num_raid_devs; i++) {
765 struct mdinfo info;
766 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
767
768 super->current_vol = i;
769 getinfo_super_imsm(st, &info);
770 fname_from_uuid(st, &info, nbuf, ':');
771 print_imsm_dev(dev, nbuf + 5, super->disks->index);
772 }
773 for (i = 0; i < mpb->num_disks; i++) {
774 if (i == super->disks->index)
775 continue;
776 print_imsm_disk(mpb, i, reserved);
777 }
778 }
779
780 static void brief_examine_super_imsm(struct supertype *st, int verbose)
781 {
782 /* We just write a generic IMSM ARRAY entry */
783 struct mdinfo info;
784 char nbuf[64];
785 struct intel_super *super = st->sb;
786
787 if (!super->anchor->num_raid_devs) {
788 printf("ARRAY metadata=imsm\n");
789 return;
790 }
791
792 getinfo_super_imsm(st, &info);
793 fname_from_uuid(st, &info, nbuf, ':');
794 printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
795 }
796
797 static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
798 {
799 /* We just write a generic IMSM ARRAY entry */
800 struct mdinfo info;
801 char nbuf[64];
802 char nbuf1[64];
803 struct intel_super *super = st->sb;
804 int i;
805
806 if (!super->anchor->num_raid_devs)
807 return;
808
809 getinfo_super_imsm(st, &info);
810 fname_from_uuid(st, &info, nbuf, ':');
811 for (i = 0; i < super->anchor->num_raid_devs; i++) {
812 struct imsm_dev *dev = get_imsm_dev(super, i);
813
814 super->current_vol = i;
815 getinfo_super_imsm(st, &info);
816 fname_from_uuid(st, &info, nbuf1, ':');
817 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
818 dev->volume, nbuf + 5, i, nbuf1 + 5);
819 }
820 }
821
822 static void export_examine_super_imsm(struct supertype *st)
823 {
824 struct intel_super *super = st->sb;
825 struct imsm_super *mpb = super->anchor;
826 struct mdinfo info;
827 char nbuf[64];
828
829 getinfo_super_imsm(st, &info);
830 fname_from_uuid(st, &info, nbuf, ':');
831 printf("MD_METADATA=imsm\n");
832 printf("MD_LEVEL=container\n");
833 printf("MD_UUID=%s\n", nbuf+5);
834 printf("MD_DEVICES=%u\n", mpb->num_disks);
835 }
836
837 static void detail_super_imsm(struct supertype *st, char *homehost)
838 {
839 struct mdinfo info;
840 char nbuf[64];
841
842 getinfo_super_imsm(st, &info);
843 fname_from_uuid(st, &info, nbuf, ':');
844 printf("\n UUID : %s\n", nbuf + 5);
845 }
846
847 static void brief_detail_super_imsm(struct supertype *st)
848 {
849 struct mdinfo info;
850 char nbuf[64];
851 getinfo_super_imsm(st, &info);
852 fname_from_uuid(st, &info, nbuf, ':');
853 printf(" UUID=%s", nbuf + 5);
854 }
855
856 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
857 static void fd2devname(int fd, char *name);
858
859 static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
860 {
861 /* dump an unsorted list of devices attached to ahci, as well as
862 * non-connected ports
863 */
864 int hba_len = strlen(hba_path) + 1;
865 struct dirent *ent;
866 DIR *dir;
867 char *path = NULL;
868 int err = 0;
869 unsigned long port_mask = (1 << port_count) - 1;
870
871 if (port_count > sizeof(port_mask) * 8) {
872 if (verbose)
873 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
874 return 2;
875 }
876
877 /* scroll through /sys/dev/block looking for devices attached to
878 * this hba
879 */
880 dir = opendir("/sys/dev/block");
881 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
882 int fd;
883 char model[64];
884 char vendor[64];
885 char buf[1024];
886 int major, minor;
887 char *device;
888 char *c;
889 int port;
890 int type;
891
892 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
893 continue;
894 path = devt_to_devpath(makedev(major, minor));
895 if (!path)
896 continue;
897 if (!path_attached_to_hba(path, hba_path)) {
898 free(path);
899 path = NULL;
900 continue;
901 }
902
903 /* retrieve the scsi device type */
904 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
905 if (verbose)
906 fprintf(stderr, Name ": failed to allocate 'device'\n");
907 err = 2;
908 break;
909 }
910 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
911 if (load_sys(device, buf) != 0) {
912 if (verbose)
913 fprintf(stderr, Name ": failed to read device type for %s\n",
914 path);
915 err = 2;
916 free(device);
917 break;
918 }
919 type = strtoul(buf, NULL, 10);
920
921 /* if it's not a disk print the vendor and model */
922 if (!(type == 0 || type == 7 || type == 14)) {
923 vendor[0] = '\0';
924 model[0] = '\0';
925 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
926 if (load_sys(device, buf) == 0) {
927 strncpy(vendor, buf, sizeof(vendor));
928 vendor[sizeof(vendor) - 1] = '\0';
929 c = (char *) &vendor[sizeof(vendor) - 1];
930 while (isspace(*c) || *c == '\0')
931 *c-- = '\0';
932
933 }
934 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
935 if (load_sys(device, buf) == 0) {
936 strncpy(model, buf, sizeof(model));
937 model[sizeof(model) - 1] = '\0';
938 c = (char *) &model[sizeof(model) - 1];
939 while (isspace(*c) || *c == '\0')
940 *c-- = '\0';
941 }
942
943 if (vendor[0] && model[0])
944 sprintf(buf, "%.64s %.64s", vendor, model);
945 else
946 switch (type) { /* numbers from hald/linux/device.c */
947 case 1: sprintf(buf, "tape"); break;
948 case 2: sprintf(buf, "printer"); break;
949 case 3: sprintf(buf, "processor"); break;
950 case 4:
951 case 5: sprintf(buf, "cdrom"); break;
952 case 6: sprintf(buf, "scanner"); break;
953 case 8: sprintf(buf, "media_changer"); break;
954 case 9: sprintf(buf, "comm"); break;
955 case 12: sprintf(buf, "raid"); break;
956 default: sprintf(buf, "unknown");
957 }
958 } else
959 buf[0] = '\0';
960 free(device);
961
962 /* chop device path to 'host%d' and calculate the port number */
963 c = strchr(&path[hba_len], '/');
964 if (!c) {
965 if (verbose)
966 fprintf(stderr, Name ": %s - invalid path name\n", path + hba_len);
967 err = 2;
968 break;
969 }
970 *c = '\0';
971 if (sscanf(&path[hba_len], "host%d", &port) == 1)
972 port -= host_base;
973 else {
974 if (verbose) {
975 *c = '/'; /* repair the full string */
976 fprintf(stderr, Name ": failed to determine port number for %s\n",
977 path);
978 }
979 err = 2;
980 break;
981 }
982
983 /* mark this port as used */
984 port_mask &= ~(1 << port);
985
986 /* print out the device information */
987 if (buf[0]) {
988 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
989 continue;
990 }
991
992 fd = dev_open(ent->d_name, O_RDONLY);
993 if (fd < 0)
994 printf(" Port%d : - disk info unavailable -\n", port);
995 else {
996 fd2devname(fd, buf);
997 printf(" Port%d : %s", port, buf);
998 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
999 printf(" (%s)\n", buf);
1000 else
1001 printf("()\n");
1002 }
1003 close(fd);
1004 free(path);
1005 path = NULL;
1006 }
1007 if (path)
1008 free(path);
1009 if (dir)
1010 closedir(dir);
1011 if (err == 0) {
1012 int i;
1013
1014 for (i = 0; i < port_count; i++)
1015 if (port_mask & (1 << i))
1016 printf(" Port%d : - no device attached -\n", i);
1017 }
1018
1019 return err;
1020 }
1021
1022 static int detail_platform_imsm(int verbose, int enumerate_only)
1023 {
1024 /* There are two components to imsm platform support, the ahci SATA
1025 * controller and the option-rom. To find the SATA controller we
1026 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1027 * controller with the Intel vendor id is present. This approach
1028 * allows mdadm to leverage the kernel's ahci detection logic, with the
1029 * caveat that if ahci.ko is not loaded mdadm will not be able to
1030 * detect platform raid capabilities. The option-rom resides in a
1031 * platform "Adapter ROM". We scan for its signature to retrieve the
1032 * platform capabilities. If raid support is disabled in the BIOS the
1033 * option-rom capability structure will not be available.
1034 */
1035 const struct imsm_orom *orom;
1036 struct sys_dev *list, *hba;
1037 DIR *dir;
1038 struct dirent *ent;
1039 const char *hba_path;
1040 int host_base = 0;
1041 int port_count = 0;
1042
1043 if (enumerate_only) {
1044 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1045 return 0;
1046 return 2;
1047 }
1048
1049 list = find_driver_devices("pci", "ahci");
1050 for (hba = list; hba; hba = hba->next)
1051 if (devpath_to_vendor(hba->path) == 0x8086)
1052 break;
1053
1054 if (!hba) {
1055 if (verbose)
1056 fprintf(stderr, Name ": unable to find active ahci controller\n");
1057 free_sys_dev(&list);
1058 return 2;
1059 } else if (verbose)
1060 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
1061 hba_path = hba->path;
1062 hba->path = NULL;
1063 free_sys_dev(&list);
1064
1065 orom = find_imsm_orom();
1066 if (!orom) {
1067 if (verbose)
1068 fprintf(stderr, Name ": imsm option-rom not found\n");
1069 return 2;
1070 }
1071
1072 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1073 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
1074 orom->hotfix_ver, orom->build);
1075 printf(" RAID Levels :%s%s%s%s%s\n",
1076 imsm_orom_has_raid0(orom) ? " raid0" : "",
1077 imsm_orom_has_raid1(orom) ? " raid1" : "",
1078 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
1079 imsm_orom_has_raid10(orom) ? " raid10" : "",
1080 imsm_orom_has_raid5(orom) ? " raid5" : "");
1081 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1082 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
1083 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
1084 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
1085 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
1086 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
1087 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
1088 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
1089 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
1090 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
1091 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
1092 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
1093 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
1094 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
1095 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
1096 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
1097 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1098 printf(" Max Disks : %d\n", orom->tds);
1099 printf(" Max Volumes : %d\n", orom->vpa);
1100 printf(" I/O Controller : %s\n", hba_path);
1101
1102 /* find the smallest scsi host number to determine a port number base */
1103 dir = opendir(hba_path);
1104 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1105 int host;
1106
1107 if (sscanf(ent->d_name, "host%d", &host) != 1)
1108 continue;
1109 if (port_count == 0)
1110 host_base = host;
1111 else if (host < host_base)
1112 host_base = host;
1113
1114 if (host + 1 > port_count + host_base)
1115 port_count = host + 1 - host_base;
1116
1117 }
1118 if (dir)
1119 closedir(dir);
1120
1121 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1122 host_base, verbose) != 0) {
1123 if (verbose)
1124 fprintf(stderr, Name ": failed to enumerate ports\n");
1125 return 2;
1126 }
1127
1128 return 0;
1129 }
1130 #endif
1131
1132 static int match_home_imsm(struct supertype *st, char *homehost)
1133 {
1134 /* the imsm metadata format does not specify any host
1135 * identification information. We return -1 since we can never
1136 * confirm nor deny whether a given array is "meant" for this
1137 * host. We rely on compare_super and the 'family_num' fields to
1138 * exclude member disks that do not belong, and we rely on
1139 * mdadm.conf to specify the arrays that should be assembled.
1140 * Auto-assembly may still pick up "foreign" arrays.
1141 */
1142
1143 return -1;
1144 }
1145
1146 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1147 {
1148 /* The uuid returned here is used for:
1149 * uuid to put into bitmap file (Create, Grow)
1150 * uuid for backup header when saving critical section (Grow)
1151 * comparing uuids when re-adding a device into an array
1152 * In these cases the uuid required is that of the data-array,
1153 * not the device-set.
1154 * uuid to recognise same set when adding a missing device back
1155 * to an array. This is a uuid for the device-set.
1156 *
1157 * For each of these we can make do with a truncated
1158 * or hashed uuid rather than the original, as long as
1159 * everyone agrees.
1160 * In each case the uuid required is that of the data-array,
1161 * not the device-set.
1162 */
1163 /* imsm does not track uuid's so we synthesis one using sha1 on
1164 * - The signature (Which is constant for all imsm array, but no matter)
1165 * - the orig_family_num of the container
1166 * - the index number of the volume
1167 * - the 'serial' number of the volume.
1168 * Hopefully these are all constant.
1169 */
1170 struct intel_super *super = st->sb;
1171
1172 char buf[20];
1173 struct sha1_ctx ctx;
1174 struct imsm_dev *dev = NULL;
1175 __u32 family_num;
1176
1177 /* some mdadm versions failed to set ->orig_family_num, in which
1178 * case fall back to ->family_num. orig_family_num will be
1179 * fixed up with the first metadata update.
1180 */
1181 family_num = super->anchor->orig_family_num;
1182 if (family_num == 0)
1183 family_num = super->anchor->family_num;
1184 sha1_init_ctx(&ctx);
1185 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1186 sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
1187 if (super->current_vol >= 0)
1188 dev = get_imsm_dev(super, super->current_vol);
1189 if (dev) {
1190 __u32 vol = super->current_vol;
1191 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1192 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1193 }
1194 sha1_finish_ctx(&ctx, buf);
1195 memcpy(uuid, buf, 4*4);
1196 }
1197
1198 #if 0
1199 static void
1200 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1201 {
1202 __u8 *v = get_imsm_version(mpb);
1203 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1204 char major[] = { 0, 0, 0 };
1205 char minor[] = { 0 ,0, 0 };
1206 char patch[] = { 0, 0, 0 };
1207 char *ver_parse[] = { major, minor, patch };
1208 int i, j;
1209
1210 i = j = 0;
1211 while (*v != '\0' && v < end) {
1212 if (*v != '.' && j < 2)
1213 ver_parse[i][j++] = *v;
1214 else {
1215 i++;
1216 j = 0;
1217 }
1218 v++;
1219 }
1220
1221 *m = strtol(minor, NULL, 0);
1222 *p = strtol(patch, NULL, 0);
1223 }
1224 #endif
1225
1226 static int imsm_level_to_layout(int level)
1227 {
1228 switch (level) {
1229 case 0:
1230 case 1:
1231 return 0;
1232 case 5:
1233 case 6:
1234 return ALGORITHM_LEFT_ASYMMETRIC;
1235 case 10:
1236 return 0x102;
1237 }
1238 return UnSet;
1239 }
1240
1241 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
1242 {
1243 struct intel_super *super = st->sb;
1244 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1245 struct imsm_map *map = get_imsm_map(dev, 0);
1246 struct dl *dl;
1247 char *devname;
1248
1249 for (dl = super->disks; dl; dl = dl->next)
1250 if (dl->raiddisk == info->disk.raid_disk)
1251 break;
1252 info->container_member = super->current_vol;
1253 info->array.raid_disks = map->num_members;
1254 info->array.level = get_imsm_raid_level(map);
1255 info->array.layout = imsm_level_to_layout(info->array.level);
1256 info->array.md_minor = -1;
1257 info->array.ctime = 0;
1258 info->array.utime = 0;
1259 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1260 info->array.state = !dev->vol.dirty;
1261 info->custom_array_size = __le32_to_cpu(dev->size_high);
1262 info->custom_array_size <<= 32;
1263 info->custom_array_size |= __le32_to_cpu(dev->size_low);
1264
1265 info->disk.major = 0;
1266 info->disk.minor = 0;
1267 if (dl) {
1268 info->disk.major = dl->major;
1269 info->disk.minor = dl->minor;
1270 }
1271
1272 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1273 info->component_size = __le32_to_cpu(map->blocks_per_member);
1274 memset(info->uuid, 0, sizeof(info->uuid));
1275
1276 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
1277 info->resync_start = 0;
1278 else if (dev->vol.migr_state)
1279 /* FIXME add curr_migr_unit to resync_start conversion */
1280 info->resync_start = 0;
1281 else
1282 info->resync_start = ~0ULL;
1283
1284 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1285 info->name[MAX_RAID_SERIAL_LEN] = 0;
1286
1287 info->array.major_version = -1;
1288 info->array.minor_version = -2;
1289 devname = devnum2devname(st->container_dev);
1290 *info->text_version = '\0';
1291 if (devname)
1292 sprintf(info->text_version, "/%s/%d", devname, info->container_member);
1293 free(devname);
1294 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1295 uuid_from_super_imsm(st, info->uuid);
1296 }
1297
1298 /* check the config file to see if we can return a real uuid for this spare */
1299 static void fixup_container_spare_uuid(struct mdinfo *inf)
1300 {
1301 struct mddev_ident_s *array_list;
1302
1303 if (inf->array.level != LEVEL_CONTAINER ||
1304 memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0)
1305 return;
1306
1307 array_list = conf_get_ident(NULL);
1308
1309 for (; array_list; array_list = array_list->next) {
1310 if (array_list->uuid_set) {
1311 struct supertype *_sst; /* spare supertype */
1312 struct supertype *_cst; /* container supertype */
1313
1314 _cst = array_list->st;
1315 if (_cst)
1316 _sst = _cst->ss->match_metadata_desc(inf->text_version);
1317 else
1318 _sst = NULL;
1319
1320 if (_sst) {
1321 memcpy(inf->uuid, array_list->uuid, sizeof(int[4]));
1322 free(_sst);
1323 break;
1324 }
1325 }
1326 }
1327 }
1328
1329 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
1330 {
1331 struct intel_super *super = st->sb;
1332 struct imsm_disk *disk;
1333
1334 if (super->current_vol >= 0) {
1335 getinfo_super_imsm_volume(st, info);
1336 return;
1337 }
1338
1339 /* Set raid_disks to zero so that Assemble will always pull in valid
1340 * spares
1341 */
1342 info->array.raid_disks = 0;
1343 info->array.level = LEVEL_CONTAINER;
1344 info->array.layout = 0;
1345 info->array.md_minor = -1;
1346 info->array.ctime = 0; /* N/A for imsm */
1347 info->array.utime = 0;
1348 info->array.chunk_size = 0;
1349
1350 info->disk.major = 0;
1351 info->disk.minor = 0;
1352 info->disk.raid_disk = -1;
1353 info->reshape_active = 0;
1354 info->array.major_version = -1;
1355 info->array.minor_version = -2;
1356 strcpy(info->text_version, "imsm");
1357 info->safe_mode_delay = 0;
1358 info->disk.number = -1;
1359 info->disk.state = 0;
1360 info->name[0] = 0;
1361
1362 if (super->disks) {
1363 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1364
1365 disk = &super->disks->disk;
1366 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1367 info->component_size = reserved;
1368 info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
1369 /* we don't change info->disk.raid_disk here because
1370 * this state will be finalized in mdmon after we have
1371 * found the 'most fresh' version of the metadata
1372 */
1373 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
1374 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
1375 }
1376
1377 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1378 * ->compare_super may have updated the 'num_raid_devs' field for spares
1379 */
1380 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1381 uuid_from_super_imsm(st, info->uuid);
1382 else {
1383 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1384 fixup_container_spare_uuid(info);
1385 }
1386 }
1387
1388 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1389 char *update, char *devname, int verbose,
1390 int uuid_set, char *homehost)
1391 {
1392 /* For 'assemble' and 'force' we need to return non-zero if any
1393 * change was made. For others, the return value is ignored.
1394 * Update options are:
1395 * force-one : This device looks a bit old but needs to be included,
1396 * update age info appropriately.
1397 * assemble: clear any 'faulty' flag to allow this device to
1398 * be assembled.
1399 * force-array: Array is degraded but being forced, mark it clean
1400 * if that will be needed to assemble it.
1401 *
1402 * newdev: not used ????
1403 * grow: Array has gained a new device - this is currently for
1404 * linear only
1405 * resync: mark as dirty so a resync will happen.
1406 * name: update the name - preserving the homehost
1407 * uuid: Change the uuid of the array to match watch is given
1408 *
1409 * Following are not relevant for this imsm:
1410 * sparc2.2 : update from old dodgey metadata
1411 * super-minor: change the preferred_minor number
1412 * summaries: update redundant counters.
1413 * homehost: update the recorded homehost
1414 * _reshape_progress: record new reshape_progress position.
1415 */
1416 int rv = 1;
1417 struct intel_super *super = st->sb;
1418 struct imsm_super *mpb;
1419
1420 /* we can only update container info */
1421 if (!super || super->current_vol >= 0 || !super->anchor)
1422 return 1;
1423
1424 mpb = super->anchor;
1425
1426 if (strcmp(update, "uuid") == 0 && uuid_set && !info->update_private)
1427 fprintf(stderr,
1428 Name ": '--uuid' not supported for imsm metadata\n");
1429 else if (strcmp(update, "uuid") == 0 && uuid_set && info->update_private) {
1430 mpb->orig_family_num = *((__u32 *) info->update_private);
1431 rv = 0;
1432 } else if (strcmp(update, "uuid") == 0) {
1433 __u32 *new_family = malloc(sizeof(*new_family));
1434
1435 /* update orig_family_number with the incoming random
1436 * data, report the new effective uuid, and store the
1437 * new orig_family_num for future updates.
1438 */
1439 if (new_family) {
1440 memcpy(&mpb->orig_family_num, info->uuid, sizeof(__u32));
1441 uuid_from_super_imsm(st, info->uuid);
1442 *new_family = mpb->orig_family_num;
1443 info->update_private = new_family;
1444 rv = 0;
1445 }
1446 } else if (strcmp(update, "assemble") == 0)
1447 rv = 0;
1448 else
1449 fprintf(stderr,
1450 Name ": '--update=%s' not supported for imsm metadata\n",
1451 update);
1452
1453 /* successful update? recompute checksum */
1454 if (rv == 0)
1455 mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb));
1456
1457 return rv;
1458 }
1459
1460 static size_t disks_to_mpb_size(int disks)
1461 {
1462 size_t size;
1463
1464 size = sizeof(struct imsm_super);
1465 size += (disks - 1) * sizeof(struct imsm_disk);
1466 size += 2 * sizeof(struct imsm_dev);
1467 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1468 size += (4 - 2) * sizeof(struct imsm_map);
1469 /* 4 possible disk_ord_tbl's */
1470 size += 4 * (disks - 1) * sizeof(__u32);
1471
1472 return size;
1473 }
1474
1475 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1476 {
1477 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1478 return 0;
1479
1480 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1481 }
1482
1483 static void free_devlist(struct intel_super *super)
1484 {
1485 struct intel_dev *dv;
1486
1487 while (super->devlist) {
1488 dv = super->devlist->next;
1489 free(super->devlist->dev);
1490 free(super->devlist);
1491 super->devlist = dv;
1492 }
1493 }
1494
1495 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1496 {
1497 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1498 }
1499
1500 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1501 {
1502 /*
1503 * return:
1504 * 0 same, or first was empty, and second was copied
1505 * 1 second had wrong number
1506 * 2 wrong uuid
1507 * 3 wrong other info
1508 */
1509 struct intel_super *first = st->sb;
1510 struct intel_super *sec = tst->sb;
1511
1512 if (!first) {
1513 st->sb = tst->sb;
1514 tst->sb = NULL;
1515 return 0;
1516 }
1517
1518 /* if an anchor does not have num_raid_devs set then it is a free
1519 * floating spare
1520 */
1521 if (first->anchor->num_raid_devs > 0 &&
1522 sec->anchor->num_raid_devs > 0) {
1523 /* Determine if these disks might ever have been
1524 * related. Further disambiguation can only take place
1525 * in load_super_imsm_all
1526 */
1527 __u32 first_family = first->anchor->orig_family_num;
1528 __u32 sec_family = sec->anchor->orig_family_num;
1529
1530 if (memcmp(first->anchor->sig, sec->anchor->sig,
1531 MAX_SIGNATURE_LENGTH) != 0)
1532 return 3;
1533
1534 if (first_family == 0)
1535 first_family = first->anchor->family_num;
1536 if (sec_family == 0)
1537 sec_family = sec->anchor->family_num;
1538
1539 if (first_family != sec_family)
1540 return 3;
1541
1542 }
1543
1544
1545 /* if 'first' is a spare promote it to a populated mpb with sec's
1546 * family number
1547 */
1548 if (first->anchor->num_raid_devs == 0 &&
1549 sec->anchor->num_raid_devs > 0) {
1550 int i;
1551 struct intel_dev *dv;
1552 struct imsm_dev *dev;
1553
1554 /* we need to copy raid device info from sec if an allocation
1555 * fails here we don't associate the spare
1556 */
1557 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1558 dv = malloc(sizeof(*dv));
1559 if (!dv)
1560 break;
1561 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1562 if (!dev) {
1563 free(dv);
1564 break;
1565 }
1566 dv->dev = dev;
1567 dv->index = i;
1568 dv->next = first->devlist;
1569 first->devlist = dv;
1570 }
1571 if (i < sec->anchor->num_raid_devs) {
1572 /* allocation failure */
1573 free_devlist(first);
1574 fprintf(stderr, "imsm: failed to associate spare\n");
1575 return 3;
1576 }
1577 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1578 first->anchor->orig_family_num = sec->anchor->orig_family_num;
1579 first->anchor->family_num = sec->anchor->family_num;
1580 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1581 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1582 }
1583
1584 return 0;
1585 }
1586
1587 static void fd2devname(int fd, char *name)
1588 {
1589 struct stat st;
1590 char path[256];
1591 char dname[100];
1592 char *nm;
1593 int rv;
1594
1595 name[0] = '\0';
1596 if (fstat(fd, &st) != 0)
1597 return;
1598 sprintf(path, "/sys/dev/block/%d:%d",
1599 major(st.st_rdev), minor(st.st_rdev));
1600
1601 rv = readlink(path, dname, sizeof(dname));
1602 if (rv <= 0)
1603 return;
1604
1605 dname[rv] = '\0';
1606 nm = strrchr(dname, '/');
1607 nm++;
1608 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1609 }
1610
1611 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1612
1613 static int imsm_read_serial(int fd, char *devname,
1614 __u8 serial[MAX_RAID_SERIAL_LEN])
1615 {
1616 unsigned char scsi_serial[255];
1617 int rv;
1618 int rsp_len;
1619 int len;
1620 char *dest;
1621 char *src;
1622 char *rsp_buf;
1623 int i;
1624
1625 memset(scsi_serial, 0, sizeof(scsi_serial));
1626
1627 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1628
1629 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1630 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1631 fd2devname(fd, (char *) serial);
1632 return 0;
1633 }
1634
1635 if (rv != 0) {
1636 if (devname)
1637 fprintf(stderr,
1638 Name ": Failed to retrieve serial for %s\n",
1639 devname);
1640 return rv;
1641 }
1642
1643 rsp_len = scsi_serial[3];
1644 if (!rsp_len) {
1645 if (devname)
1646 fprintf(stderr,
1647 Name ": Failed to retrieve serial for %s\n",
1648 devname);
1649 return 2;
1650 }
1651 rsp_buf = (char *) &scsi_serial[4];
1652
1653 /* trim all whitespace and non-printable characters and convert
1654 * ':' to ';'
1655 */
1656 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
1657 src = &rsp_buf[i];
1658 if (*src > 0x20) {
1659 /* ':' is reserved for use in placeholder serial
1660 * numbers for missing disks
1661 */
1662 if (*src == ':')
1663 *dest++ = ';';
1664 else
1665 *dest++ = *src;
1666 }
1667 }
1668 len = dest - rsp_buf;
1669 dest = rsp_buf;
1670
1671 /* truncate leading characters */
1672 if (len > MAX_RAID_SERIAL_LEN) {
1673 dest += len - MAX_RAID_SERIAL_LEN;
1674 len = MAX_RAID_SERIAL_LEN;
1675 }
1676
1677 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1678 memcpy(serial, dest, len);
1679
1680 return 0;
1681 }
1682
1683 static int serialcmp(__u8 *s1, __u8 *s2)
1684 {
1685 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1686 }
1687
1688 static void serialcpy(__u8 *dest, __u8 *src)
1689 {
1690 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1691 }
1692
1693 #ifndef MDASSEMBLE
1694 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1695 {
1696 struct dl *dl;
1697
1698 for (dl = super->disks; dl; dl = dl->next)
1699 if (serialcmp(dl->serial, serial) == 0)
1700 break;
1701
1702 return dl;
1703 }
1704 #endif
1705
1706 static struct imsm_disk *
1707 __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
1708 {
1709 int i;
1710
1711 for (i = 0; i < mpb->num_disks; i++) {
1712 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
1713
1714 if (serialcmp(disk->serial, serial) == 0) {
1715 if (idx)
1716 *idx = i;
1717 return disk;
1718 }
1719 }
1720
1721 return NULL;
1722 }
1723
1724 static int
1725 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1726 {
1727 struct imsm_disk *disk;
1728 struct dl *dl;
1729 struct stat stb;
1730 int rv;
1731 char name[40];
1732 __u8 serial[MAX_RAID_SERIAL_LEN];
1733
1734 rv = imsm_read_serial(fd, devname, serial);
1735
1736 if (rv != 0)
1737 return 2;
1738
1739 dl = calloc(1, sizeof(*dl));
1740 if (!dl) {
1741 if (devname)
1742 fprintf(stderr,
1743 Name ": failed to allocate disk buffer for %s\n",
1744 devname);
1745 return 2;
1746 }
1747
1748 fstat(fd, &stb);
1749 dl->major = major(stb.st_rdev);
1750 dl->minor = minor(stb.st_rdev);
1751 dl->next = super->disks;
1752 dl->fd = keep_fd ? fd : -1;
1753 assert(super->disks == NULL);
1754 super->disks = dl;
1755 serialcpy(dl->serial, serial);
1756 dl->index = -2;
1757 dl->e = NULL;
1758 fd2devname(fd, name);
1759 if (devname)
1760 dl->devname = strdup(devname);
1761 else
1762 dl->devname = strdup(name);
1763
1764 /* look up this disk's index in the current anchor */
1765 disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
1766 if (disk) {
1767 dl->disk = *disk;
1768 /* only set index on disks that are a member of a
1769 * populated contianer, i.e. one with raid_devs
1770 */
1771 if (is_failed(&dl->disk))
1772 dl->index = -2;
1773 else if (is_spare(&dl->disk))
1774 dl->index = -1;
1775 }
1776
1777 return 0;
1778 }
1779
1780 #ifndef MDASSEMBLE
1781 /* When migrating map0 contains the 'destination' state while map1
1782 * contains the current state. When not migrating map0 contains the
1783 * current state. This routine assumes that map[0].map_state is set to
1784 * the current array state before being called.
1785 *
1786 * Migration is indicated by one of the following states
1787 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1788 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1789 * map1state=unitialized)
1790 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1791 * map1state=normal)
1792 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1793 * map1state=degraded)
1794 */
1795 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
1796 {
1797 struct imsm_map *dest;
1798 struct imsm_map *src = get_imsm_map(dev, 0);
1799
1800 dev->vol.migr_state = 1;
1801 set_migr_type(dev, migr_type);
1802 dev->vol.curr_migr_unit = 0;
1803 dest = get_imsm_map(dev, 1);
1804
1805 /* duplicate and then set the target end state in map[0] */
1806 memcpy(dest, src, sizeof_imsm_map(src));
1807 if (migr_type == MIGR_REBUILD) {
1808 __u32 ord;
1809 int i;
1810
1811 for (i = 0; i < src->num_members; i++) {
1812 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
1813 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
1814 }
1815 }
1816
1817 src->map_state = to_state;
1818 }
1819
1820 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1821 {
1822 struct imsm_map *map = get_imsm_map(dev, 0);
1823 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
1824 int i;
1825
1826 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1827 * completed in the last migration.
1828 *
1829 * FIXME add support for online capacity expansion and
1830 * raid-level-migration
1831 */
1832 for (i = 0; i < prev->num_members; i++)
1833 map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i];
1834
1835 dev->vol.migr_state = 0;
1836 dev->vol.curr_migr_unit = 0;
1837 map->map_state = map_state;
1838 }
1839 #endif
1840
1841 static int parse_raid_devices(struct intel_super *super)
1842 {
1843 int i;
1844 struct imsm_dev *dev_new;
1845 size_t len, len_migr;
1846 size_t space_needed = 0;
1847 struct imsm_super *mpb = super->anchor;
1848
1849 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1850 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1851 struct intel_dev *dv;
1852
1853 len = sizeof_imsm_dev(dev_iter, 0);
1854 len_migr = sizeof_imsm_dev(dev_iter, 1);
1855 if (len_migr > len)
1856 space_needed += len_migr - len;
1857
1858 dv = malloc(sizeof(*dv));
1859 if (!dv)
1860 return 1;
1861 dev_new = malloc(len_migr);
1862 if (!dev_new) {
1863 free(dv);
1864 return 1;
1865 }
1866 imsm_copy_dev(dev_new, dev_iter);
1867 dv->dev = dev_new;
1868 dv->index = i;
1869 dv->next = super->devlist;
1870 super->devlist = dv;
1871 }
1872
1873 /* ensure that super->buf is large enough when all raid devices
1874 * are migrating
1875 */
1876 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1877 void *buf;
1878
1879 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1880 if (posix_memalign(&buf, 512, len) != 0)
1881 return 1;
1882
1883 memcpy(buf, super->buf, super->len);
1884 memset(buf + super->len, 0, len - super->len);
1885 free(super->buf);
1886 super->buf = buf;
1887 super->len = len;
1888 }
1889
1890 return 0;
1891 }
1892
1893 /* retrieve a pointer to the bbm log which starts after all raid devices */
1894 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1895 {
1896 void *ptr = NULL;
1897
1898 if (__le32_to_cpu(mpb->bbm_log_size)) {
1899 ptr = mpb;
1900 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1901 }
1902
1903 return ptr;
1904 }
1905
1906 static void __free_imsm(struct intel_super *super, int free_disks);
1907
1908 /* load_imsm_mpb - read matrix metadata
1909 * allocates super->mpb to be freed by free_super
1910 */
1911 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1912 {
1913 unsigned long long dsize;
1914 unsigned long long sectors;
1915 struct stat;
1916 struct imsm_super *anchor;
1917 __u32 check_sum;
1918
1919 get_dev_size(fd, NULL, &dsize);
1920
1921 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1922 if (devname)
1923 fprintf(stderr,
1924 Name ": Cannot seek to anchor block on %s: %s\n",
1925 devname, strerror(errno));
1926 return 1;
1927 }
1928
1929 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1930 if (devname)
1931 fprintf(stderr,
1932 Name ": Failed to allocate imsm anchor buffer"
1933 " on %s\n", devname);
1934 return 1;
1935 }
1936 if (read(fd, anchor, 512) != 512) {
1937 if (devname)
1938 fprintf(stderr,
1939 Name ": Cannot read anchor block on %s: %s\n",
1940 devname, strerror(errno));
1941 free(anchor);
1942 return 1;
1943 }
1944
1945 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1946 if (devname)
1947 fprintf(stderr,
1948 Name ": no IMSM anchor on %s\n", devname);
1949 free(anchor);
1950 return 2;
1951 }
1952
1953 __free_imsm(super, 0);
1954 super->len = ROUND_UP(anchor->mpb_size, 512);
1955 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1956 if (devname)
1957 fprintf(stderr,
1958 Name ": unable to allocate %zu byte mpb buffer\n",
1959 super->len);
1960 free(anchor);
1961 return 2;
1962 }
1963 memcpy(super->buf, anchor, 512);
1964
1965 sectors = mpb_sectors(anchor) - 1;
1966 free(anchor);
1967 if (!sectors) {
1968 check_sum = __gen_imsm_checksum(super->anchor);
1969 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1970 if (devname)
1971 fprintf(stderr,
1972 Name ": IMSM checksum %x != %x on %s\n",
1973 check_sum,
1974 __le32_to_cpu(super->anchor->check_sum),
1975 devname);
1976 return 2;
1977 }
1978
1979 return 0;
1980 }
1981
1982 /* read the extended mpb */
1983 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1984 if (devname)
1985 fprintf(stderr,
1986 Name ": Cannot seek to extended mpb on %s: %s\n",
1987 devname, strerror(errno));
1988 return 1;
1989 }
1990
1991 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1992 if (devname)
1993 fprintf(stderr,
1994 Name ": Cannot read extended mpb on %s: %s\n",
1995 devname, strerror(errno));
1996 return 2;
1997 }
1998
1999 check_sum = __gen_imsm_checksum(super->anchor);
2000 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
2001 if (devname)
2002 fprintf(stderr,
2003 Name ": IMSM checksum %x != %x on %s\n",
2004 check_sum, __le32_to_cpu(super->anchor->check_sum),
2005 devname);
2006 return 3;
2007 }
2008
2009 /* FIXME the BBM log is disk specific so we cannot use this global
2010 * buffer for all disks. Ok for now since we only look at the global
2011 * bbm_log_size parameter to gate assembly
2012 */
2013 super->bbm_log = __get_imsm_bbm_log(super->anchor);
2014
2015 return 0;
2016 }
2017
2018 static int
2019 load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
2020 {
2021 int err;
2022
2023 err = load_imsm_mpb(fd, super, devname);
2024 if (err)
2025 return err;
2026 err = load_imsm_disk(fd, super, devname, keep_fd);
2027 if (err)
2028 return err;
2029 err = parse_raid_devices(super);
2030
2031 return err;
2032 }
2033
2034 static void __free_imsm_disk(struct dl *d)
2035 {
2036 if (d->fd >= 0)
2037 close(d->fd);
2038 if (d->devname)
2039 free(d->devname);
2040 if (d->e)
2041 free(d->e);
2042 free(d);
2043
2044 }
2045 static void free_imsm_disks(struct intel_super *super)
2046 {
2047 struct dl *d;
2048
2049 while (super->disks) {
2050 d = super->disks;
2051 super->disks = d->next;
2052 __free_imsm_disk(d);
2053 }
2054 while (super->missing) {
2055 d = super->missing;
2056 super->missing = d->next;
2057 __free_imsm_disk(d);
2058 }
2059
2060 }
2061
2062 /* free all the pieces hanging off of a super pointer */
2063 static void __free_imsm(struct intel_super *super, int free_disks)
2064 {
2065 if (super->buf) {
2066 free(super->buf);
2067 super->buf = NULL;
2068 }
2069 if (free_disks)
2070 free_imsm_disks(super);
2071 free_devlist(super);
2072 if (super->hba) {
2073 free((void *) super->hba);
2074 super->hba = NULL;
2075 }
2076 }
2077
2078 static void free_imsm(struct intel_super *super)
2079 {
2080 __free_imsm(super, 1);
2081 free(super);
2082 }
2083
2084 static void free_super_imsm(struct supertype *st)
2085 {
2086 struct intel_super *super = st->sb;
2087
2088 if (!super)
2089 return;
2090
2091 free_imsm(super);
2092 st->sb = NULL;
2093 }
2094
2095 static struct intel_super *alloc_super(int creating_imsm)
2096 {
2097 struct intel_super *super = malloc(sizeof(*super));
2098
2099 if (super) {
2100 memset(super, 0, sizeof(*super));
2101 super->creating_imsm = creating_imsm;
2102 super->current_vol = -1;
2103 super->create_offset = ~((__u32 ) 0);
2104 if (!check_env("IMSM_NO_PLATFORM"))
2105 super->orom = find_imsm_orom();
2106 if (super->orom && !check_env("IMSM_TEST_OROM")) {
2107 struct sys_dev *list, *ent;
2108
2109 /* find the first intel ahci controller */
2110 list = find_driver_devices("pci", "ahci");
2111 for (ent = list; ent; ent = ent->next)
2112 if (devpath_to_vendor(ent->path) == 0x8086)
2113 break;
2114 if (ent) {
2115 super->hba = ent->path;
2116 ent->path = NULL;
2117 }
2118 free_sys_dev(&list);
2119 }
2120 }
2121
2122 return super;
2123 }
2124
2125 #ifndef MDASSEMBLE
2126 /* find_missing - helper routine for load_super_imsm_all that identifies
2127 * disks that have disappeared from the system. This routine relies on
2128 * the mpb being uptodate, which it is at load time.
2129 */
2130 static int find_missing(struct intel_super *super)
2131 {
2132 int i;
2133 struct imsm_super *mpb = super->anchor;
2134 struct dl *dl;
2135 struct imsm_disk *disk;
2136
2137 for (i = 0; i < mpb->num_disks; i++) {
2138 disk = __get_imsm_disk(mpb, i);
2139 dl = serial_to_dl(disk->serial, super);
2140 if (dl)
2141 continue;
2142
2143 dl = malloc(sizeof(*dl));
2144 if (!dl)
2145 return 1;
2146 dl->major = 0;
2147 dl->minor = 0;
2148 dl->fd = -1;
2149 dl->devname = strdup("missing");
2150 dl->index = i;
2151 serialcpy(dl->serial, disk->serial);
2152 dl->disk = *disk;
2153 dl->e = NULL;
2154 dl->next = super->missing;
2155 super->missing = dl;
2156 }
2157
2158 return 0;
2159 }
2160
2161 static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
2162 {
2163 struct intel_disk *idisk = disk_list;
2164
2165 while (idisk) {
2166 if (serialcmp(idisk->disk.serial, serial) == 0)
2167 break;
2168 idisk = idisk->next;
2169 }
2170
2171 return idisk;
2172 }
2173
2174 static int __prep_thunderdome(struct intel_super **table, int tbl_size,
2175 struct intel_super *super,
2176 struct intel_disk **disk_list)
2177 {
2178 struct imsm_disk *d = &super->disks->disk;
2179 struct imsm_super *mpb = super->anchor;
2180 int i, j;
2181
2182 for (i = 0; i < tbl_size; i++) {
2183 struct imsm_super *tbl_mpb = table[i]->anchor;
2184 struct imsm_disk *tbl_d = &table[i]->disks->disk;
2185
2186 if (tbl_mpb->family_num == mpb->family_num) {
2187 if (tbl_mpb->check_sum == mpb->check_sum) {
2188 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2189 __func__, super->disks->major,
2190 super->disks->minor,
2191 table[i]->disks->major,
2192 table[i]->disks->minor);
2193 break;
2194 }
2195
2196 if (((is_configured(d) && !is_configured(tbl_d)) ||
2197 is_configured(d) == is_configured(tbl_d)) &&
2198 tbl_mpb->generation_num < mpb->generation_num) {
2199 /* current version of the mpb is a
2200 * better candidate than the one in
2201 * super_table, but copy over "cross
2202 * generational" status
2203 */
2204 struct intel_disk *idisk;
2205
2206 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2207 __func__, super->disks->major,
2208 super->disks->minor,
2209 table[i]->disks->major,
2210 table[i]->disks->minor);
2211
2212 idisk = disk_list_get(tbl_d->serial, *disk_list);
2213 if (idisk && is_failed(&idisk->disk))
2214 tbl_d->status |= FAILED_DISK;
2215 break;
2216 } else {
2217 struct intel_disk *idisk;
2218 struct imsm_disk *disk;
2219
2220 /* tbl_mpb is more up to date, but copy
2221 * over cross generational status before
2222 * returning
2223 */
2224 disk = __serial_to_disk(d->serial, mpb, NULL);
2225 if (disk && is_failed(disk))
2226 d->status |= FAILED_DISK;
2227
2228 idisk = disk_list_get(d->serial, *disk_list);
2229 if (idisk) {
2230 idisk->owner = i;
2231 if (disk && is_configured(disk))
2232 idisk->disk.status |= CONFIGURED_DISK;
2233 }
2234
2235 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2236 __func__, super->disks->major,
2237 super->disks->minor,
2238 table[i]->disks->major,
2239 table[i]->disks->minor);
2240
2241 return tbl_size;
2242 }
2243 }
2244 }
2245
2246 if (i >= tbl_size)
2247 table[tbl_size++] = super;
2248 else
2249 table[i] = super;
2250
2251 /* update/extend the merged list of imsm_disk records */
2252 for (j = 0; j < mpb->num_disks; j++) {
2253 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
2254 struct intel_disk *idisk;
2255
2256 idisk = disk_list_get(disk->serial, *disk_list);
2257 if (idisk) {
2258 idisk->disk.status |= disk->status;
2259 if (is_configured(&idisk->disk) ||
2260 is_failed(&idisk->disk))
2261 idisk->disk.status &= ~(SPARE_DISK);
2262 } else {
2263 idisk = calloc(1, sizeof(*idisk));
2264 if (!idisk)
2265 return -1;
2266 idisk->owner = IMSM_UNKNOWN_OWNER;
2267 idisk->disk = *disk;
2268 idisk->next = *disk_list;
2269 *disk_list = idisk;
2270 }
2271
2272 if (serialcmp(idisk->disk.serial, d->serial) == 0)
2273 idisk->owner = i;
2274 }
2275
2276 return tbl_size;
2277 }
2278
2279 static struct intel_super *
2280 validate_members(struct intel_super *super, struct intel_disk *disk_list,
2281 const int owner)
2282 {
2283 struct imsm_super *mpb = super->anchor;
2284 int ok_count = 0;
2285 int i;
2286
2287 for (i = 0; i < mpb->num_disks; i++) {
2288 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2289 struct intel_disk *idisk;
2290
2291 idisk = disk_list_get(disk->serial, disk_list);
2292 if (idisk) {
2293 if (idisk->owner == owner ||
2294 idisk->owner == IMSM_UNKNOWN_OWNER)
2295 ok_count++;
2296 else
2297 dprintf("%s: '%.16s' owner %d != %d\n",
2298 __func__, disk->serial, idisk->owner,
2299 owner);
2300 } else {
2301 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2302 __func__, __le32_to_cpu(mpb->family_num), i,
2303 disk->serial);
2304 break;
2305 }
2306 }
2307
2308 if (ok_count == mpb->num_disks)
2309 return super;
2310 return NULL;
2311 }
2312
2313 static void show_conflicts(__u32 family_num, struct intel_super *super_list)
2314 {
2315 struct intel_super *s;
2316
2317 for (s = super_list; s; s = s->next) {
2318 if (family_num != s->anchor->family_num)
2319 continue;
2320 fprintf(stderr, "Conflict, offlining family %#x on '%s'\n",
2321 __le32_to_cpu(family_num), s->disks->devname);
2322 }
2323 }
2324
2325 static struct intel_super *
2326 imsm_thunderdome(struct intel_super **super_list, int len)
2327 {
2328 struct intel_super *super_table[len];
2329 struct intel_disk *disk_list = NULL;
2330 struct intel_super *champion, *spare;
2331 struct intel_super *s, **del;
2332 int tbl_size = 0;
2333 int conflict;
2334 int i;
2335
2336 memset(super_table, 0, sizeof(super_table));
2337 for (s = *super_list; s; s = s->next)
2338 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
2339
2340 for (i = 0; i < tbl_size; i++) {
2341 struct imsm_disk *d;
2342 struct intel_disk *idisk;
2343 struct imsm_super *mpb = super_table[i]->anchor;
2344
2345 s = super_table[i];
2346 d = &s->disks->disk;
2347
2348 /* 'd' must appear in merged disk list for its
2349 * configuration to be valid
2350 */
2351 idisk = disk_list_get(d->serial, disk_list);
2352 if (idisk && idisk->owner == i)
2353 s = validate_members(s, disk_list, i);
2354 else
2355 s = NULL;
2356
2357 if (!s)
2358 dprintf("%s: marking family: %#x from %d:%d offline\n",
2359 __func__, mpb->family_num,
2360 super_table[i]->disks->major,
2361 super_table[i]->disks->minor);
2362 super_table[i] = s;
2363 }
2364
2365 /* This is where the mdadm implementation differs from the Windows
2366 * driver which has no strict concept of a container. We can only
2367 * assemble one family from a container, so when returning a prodigal
2368 * array member to this system the code will not be able to disambiguate
2369 * the container contents that should be assembled ("foreign" versus
2370 * "local"). It requires user intervention to set the orig_family_num
2371 * to a new value to establish a new container. The Windows driver in
2372 * this situation fixes up the volume name in place and manages the
2373 * foreign array as an independent entity.
2374 */
2375 s = NULL;
2376 spare = NULL;
2377 conflict = 0;
2378 for (i = 0; i < tbl_size; i++) {
2379 struct intel_super *tbl_ent = super_table[i];
2380 int is_spare = 0;
2381
2382 if (!tbl_ent)
2383 continue;
2384
2385 if (tbl_ent->anchor->num_raid_devs == 0) {
2386 spare = tbl_ent;
2387 is_spare = 1;
2388 }
2389
2390 if (s && !is_spare) {
2391 show_conflicts(tbl_ent->anchor->family_num, *super_list);
2392 conflict++;
2393 } else if (!s && !is_spare)
2394 s = tbl_ent;
2395 }
2396
2397 if (!s)
2398 s = spare;
2399 if (!s) {
2400 champion = NULL;
2401 goto out;
2402 }
2403 champion = s;
2404
2405 if (conflict)
2406 fprintf(stderr, "Chose family %#x on '%s', "
2407 "assemble conflicts to new container with '--update=uuid'\n",
2408 __le32_to_cpu(s->anchor->family_num), s->disks->devname);
2409
2410 /* collect all dl's onto 'champion', and update them to
2411 * champion's version of the status
2412 */
2413 for (s = *super_list; s; s = s->next) {
2414 struct imsm_super *mpb = champion->anchor;
2415 struct dl *dl = s->disks;
2416
2417 if (s == champion)
2418 continue;
2419
2420 for (i = 0; i < mpb->num_disks; i++) {
2421 struct imsm_disk *disk;
2422
2423 disk = __serial_to_disk(dl->serial, mpb, &dl->index);
2424 if (disk) {
2425 dl->disk = *disk;
2426 /* only set index on disks that are a member of
2427 * a populated contianer, i.e. one with
2428 * raid_devs
2429 */
2430 if (is_failed(&dl->disk))
2431 dl->index = -2;
2432 else if (is_spare(&dl->disk))
2433 dl->index = -1;
2434 break;
2435 }
2436 }
2437
2438 if (i >= mpb->num_disks) {
2439 struct intel_disk *idisk;
2440
2441 idisk = disk_list_get(dl->serial, disk_list);
2442 if (is_spare(&idisk->disk) &&
2443 !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
2444 dl->index = -1;
2445 else {
2446 dl->index = -2;
2447 continue;
2448 }
2449 }
2450
2451 dl->next = champion->disks;
2452 champion->disks = dl;
2453 s->disks = NULL;
2454 }
2455
2456 /* delete 'champion' from super_list */
2457 for (del = super_list; *del; ) {
2458 if (*del == champion) {
2459 *del = (*del)->next;
2460 break;
2461 } else
2462 del = &(*del)->next;
2463 }
2464 champion->next = NULL;
2465
2466 out:
2467 while (disk_list) {
2468 struct intel_disk *idisk = disk_list;
2469
2470 disk_list = disk_list->next;
2471 free(idisk);
2472 }
2473
2474 return champion;
2475 }
2476
2477 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
2478 char *devname, int keep_fd)
2479 {
2480 struct mdinfo *sra;
2481 struct intel_super *super_list = NULL;
2482 struct intel_super *super = NULL;
2483 int devnum = fd2devnum(fd);
2484 struct mdinfo *sd;
2485 int retry;
2486 int err = 0;
2487 int i;
2488 enum sysfs_read_flags flags;
2489
2490 flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2491 if (mdmon_running(devnum))
2492 flags |= SKIP_GONE_DEVS;
2493
2494 /* check if 'fd' an opened container */
2495 sra = sysfs_read(fd, 0, flags);
2496 if (!sra)
2497 return 1;
2498
2499 if (sra->array.major_version != -1 ||
2500 sra->array.minor_version != -2 ||
2501 strcmp(sra->text_version, "imsm") != 0)
2502 return 1;
2503
2504 /* load all mpbs */
2505 for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
2506 struct intel_super *s = alloc_super(0);
2507 char nm[32];
2508 int dfd;
2509
2510 err = 1;
2511 if (!s)
2512 goto error;
2513 s->next = super_list;
2514 super_list = s;
2515
2516 err = 2;
2517 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2518 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
2519 if (dfd < 0)
2520 goto error;
2521
2522 err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
2523
2524 /* retry the load if we might have raced against mdmon */
2525 if (err == 3 && mdmon_running(devnum))
2526 for (retry = 0; retry < 3; retry++) {
2527 usleep(3000);
2528 err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
2529 if (err != 3)
2530 break;
2531 }
2532 if (!keep_fd)
2533 close(dfd);
2534 if (err)
2535 goto error;
2536 }
2537
2538 /* all mpbs enter, maybe one leaves */
2539 super = imsm_thunderdome(&super_list, i);
2540 if (!super) {
2541 err = 1;
2542 goto error;
2543 }
2544
2545 if (find_missing(super) != 0) {
2546 free_imsm(super);
2547 err = 2;
2548 goto error;
2549 }
2550
2551 if (st->subarray[0]) {
2552 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2553 super->current_vol = atoi(st->subarray);
2554 else {
2555 free_imsm(super);
2556 err = 1;
2557 goto error;
2558 }
2559 }
2560 err = 0;
2561
2562 error:
2563 while (super_list) {
2564 struct intel_super *s = super_list;
2565
2566 super_list = super_list->next;
2567 free_imsm(s);
2568 }
2569
2570 if (err)
2571 return err;
2572
2573 *sbp = super;
2574 st->container_dev = devnum;
2575 if (err == 0 && st->ss == NULL) {
2576 st->ss = &super_imsm;
2577 st->minor_version = 0;
2578 st->max_devs = IMSM_MAX_DEVICES;
2579 }
2580 st->loaded_container = 1;
2581
2582 return 0;
2583 }
2584 #endif
2585
2586 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2587 {
2588 struct intel_super *super;
2589 int rv;
2590
2591 #ifndef MDASSEMBLE
2592 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2593 return 0;
2594 #endif
2595
2596 free_super_imsm(st);
2597
2598 super = alloc_super(0);
2599 if (!super) {
2600 fprintf(stderr,
2601 Name ": malloc of %zu failed.\n",
2602 sizeof(*super));
2603 return 1;
2604 }
2605
2606 rv = load_and_parse_mpb(fd, super, devname, 0);
2607
2608 if (rv) {
2609 if (devname)
2610 fprintf(stderr,
2611 Name ": Failed to load all information "
2612 "sections on %s\n", devname);
2613 free_imsm(super);
2614 return rv;
2615 }
2616
2617 if (st->subarray[0]) {
2618 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2619 super->current_vol = atoi(st->subarray);
2620 else {
2621 free_imsm(super);
2622 return 1;
2623 }
2624 }
2625
2626 st->sb = super;
2627 if (st->ss == NULL) {
2628 st->ss = &super_imsm;
2629 st->minor_version = 0;
2630 st->max_devs = IMSM_MAX_DEVICES;
2631 }
2632 st->loaded_container = 0;
2633
2634 return 0;
2635 }
2636
2637 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2638 {
2639 if (info->level == 1)
2640 return 128;
2641 return info->chunk_size >> 9;
2642 }
2643
2644 static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
2645 {
2646 __u32 num_stripes;
2647
2648 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2649 num_stripes /= num_domains;
2650
2651 return num_stripes;
2652 }
2653
2654 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2655 {
2656 if (info->level == 1)
2657 return info->size * 2;
2658 else
2659 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2660 }
2661
2662 static void imsm_update_version_info(struct intel_super *super)
2663 {
2664 /* update the version and attributes */
2665 struct imsm_super *mpb = super->anchor;
2666 char *version;
2667 struct imsm_dev *dev;
2668 struct imsm_map *map;
2669 int i;
2670
2671 for (i = 0; i < mpb->num_raid_devs; i++) {
2672 dev = get_imsm_dev(super, i);
2673 map = get_imsm_map(dev, 0);
2674 if (__le32_to_cpu(dev->size_high) > 0)
2675 mpb->attributes |= MPB_ATTRIB_2TB;
2676
2677 /* FIXME detect when an array spans a port multiplier */
2678 #if 0
2679 mpb->attributes |= MPB_ATTRIB_PM;
2680 #endif
2681
2682 if (mpb->num_raid_devs > 1 ||
2683 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2684 version = MPB_VERSION_ATTRIBS;
2685 switch (get_imsm_raid_level(map)) {
2686 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2687 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2688 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2689 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2690 }
2691 } else {
2692 if (map->num_members >= 5)
2693 version = MPB_VERSION_5OR6_DISK_ARRAY;
2694 else if (dev->status == DEV_CLONE_N_GO)
2695 version = MPB_VERSION_CNG;
2696 else if (get_imsm_raid_level(map) == 5)
2697 version = MPB_VERSION_RAID5;
2698 else if (map->num_members >= 3)
2699 version = MPB_VERSION_3OR4_DISK_ARRAY;
2700 else if (get_imsm_raid_level(map) == 1)
2701 version = MPB_VERSION_RAID1;
2702 else
2703 version = MPB_VERSION_RAID0;
2704 }
2705 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2706 }
2707 }
2708
2709 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2710 unsigned long long size, char *name,
2711 char *homehost, int *uuid)
2712 {
2713 /* We are creating a volume inside a pre-existing container.
2714 * so st->sb is already set.
2715 */
2716 struct intel_super *super = st->sb;
2717 struct imsm_super *mpb = super->anchor;
2718 struct intel_dev *dv;
2719 struct imsm_dev *dev;
2720 struct imsm_vol *vol;
2721 struct imsm_map *map;
2722 int idx = mpb->num_raid_devs;
2723 int i;
2724 unsigned long long array_blocks;
2725 size_t size_old, size_new;
2726 __u32 num_data_stripes;
2727
2728 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2729 fprintf(stderr, Name": This imsm-container already has the "
2730 "maximum of %d volumes\n", super->orom->vpa);
2731 return 0;
2732 }
2733
2734 /* ensure the mpb is large enough for the new data */
2735 size_old = __le32_to_cpu(mpb->mpb_size);
2736 size_new = disks_to_mpb_size(info->nr_disks);
2737 if (size_new > size_old) {
2738 void *mpb_new;
2739 size_t size_round = ROUND_UP(size_new, 512);
2740
2741 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2742 fprintf(stderr, Name": could not allocate new mpb\n");
2743 return 0;
2744 }
2745 memcpy(mpb_new, mpb, size_old);
2746 free(mpb);
2747 mpb = mpb_new;
2748 super->anchor = mpb_new;
2749 mpb->mpb_size = __cpu_to_le32(size_new);
2750 memset(mpb_new + size_old, 0, size_round - size_old);
2751 }
2752 super->current_vol = idx;
2753 /* when creating the first raid device in this container set num_disks
2754 * to zero, i.e. delete this spare and add raid member devices in
2755 * add_to_super_imsm_volume()
2756 */
2757 if (super->current_vol == 0)
2758 mpb->num_disks = 0;
2759
2760 for (i = 0; i < super->current_vol; i++) {
2761 dev = get_imsm_dev(super, i);
2762 if (strncmp((char *) dev->volume, name,
2763 MAX_RAID_SERIAL_LEN) == 0) {
2764 fprintf(stderr, Name": '%s' is already defined for this container\n",
2765 name);
2766 return 0;
2767 }
2768 }
2769
2770 sprintf(st->subarray, "%d", idx);
2771 dv = malloc(sizeof(*dv));
2772 if (!dv) {
2773 fprintf(stderr, Name ": failed to allocate device list entry\n");
2774 return 0;
2775 }
2776 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2777 if (!dev) {
2778 free(dv);
2779 fprintf(stderr, Name": could not allocate raid device\n");
2780 return 0;
2781 }
2782 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2783 if (info->level == 1)
2784 array_blocks = info_to_blocks_per_member(info);
2785 else
2786 array_blocks = calc_array_size(info->level, info->raid_disks,
2787 info->layout, info->chunk_size,
2788 info->size*2);
2789 /* round array size down to closest MB */
2790 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
2791
2792 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2793 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2794 dev->status = __cpu_to_le32(0);
2795 dev->reserved_blocks = __cpu_to_le32(0);
2796 vol = &dev->vol;
2797 vol->migr_state = 0;
2798 set_migr_type(dev, MIGR_INIT);
2799 vol->dirty = 0;
2800 vol->curr_migr_unit = 0;
2801 map = get_imsm_map(dev, 0);
2802 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2803 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2804 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2805 map->failed_disk_num = ~0;
2806 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2807 IMSM_T_STATE_NORMAL;
2808 map->ddf = 1;
2809
2810 if (info->level == 1 && info->raid_disks > 2) {
2811 fprintf(stderr, Name": imsm does not support more than 2 disks"
2812 "in a raid1 volume\n");
2813 return 0;
2814 }
2815
2816 map->raid_level = info->level;
2817 if (info->level == 10) {
2818 map->raid_level = 1;
2819 map->num_domains = info->raid_disks / 2;
2820 } else if (info->level == 1)
2821 map->num_domains = info->raid_disks;
2822 else
2823 map->num_domains = 1;
2824
2825 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
2826 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
2827
2828 map->num_members = info->raid_disks;
2829 for (i = 0; i < map->num_members; i++) {
2830 /* initialized in add_to_super */
2831 set_imsm_ord_tbl_ent(map, i, 0);
2832 }
2833 mpb->num_raid_devs++;
2834
2835 dv->dev = dev;
2836 dv->index = super->current_vol;
2837 dv->next = super->devlist;
2838 super->devlist = dv;
2839
2840 imsm_update_version_info(super);
2841
2842 return 1;
2843 }
2844
2845 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2846 unsigned long long size, char *name,
2847 char *homehost, int *uuid)
2848 {
2849 /* This is primarily called by Create when creating a new array.
2850 * We will then get add_to_super called for each component, and then
2851 * write_init_super called to write it out to each device.
2852 * For IMSM, Create can create on fresh devices or on a pre-existing
2853 * array.
2854 * To create on a pre-existing array a different method will be called.
2855 * This one is just for fresh drives.
2856 */
2857 struct intel_super *super;
2858 struct imsm_super *mpb;
2859 size_t mpb_size;
2860 char *version;
2861
2862 if (st->sb)
2863 return init_super_imsm_volume(st, info, size, name, homehost, uuid);
2864
2865 if (info)
2866 mpb_size = disks_to_mpb_size(info->nr_disks);
2867 else
2868 mpb_size = 512;
2869
2870 super = alloc_super(1);
2871 if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) {
2872 free(super);
2873 super = NULL;
2874 }
2875 if (!super) {
2876 fprintf(stderr, Name
2877 ": %s could not allocate superblock\n", __func__);
2878 return 0;
2879 }
2880 memset(super->buf, 0, mpb_size);
2881 mpb = super->buf;
2882 mpb->mpb_size = __cpu_to_le32(mpb_size);
2883 st->sb = super;
2884
2885 if (info == NULL) {
2886 /* zeroing superblock */
2887 return 0;
2888 }
2889
2890 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2891
2892 version = (char *) mpb->sig;
2893 strcpy(version, MPB_SIGNATURE);
2894 version += strlen(MPB_SIGNATURE);
2895 strcpy(version, MPB_VERSION_RAID0);
2896
2897 return 1;
2898 }
2899
2900 #ifndef MDASSEMBLE
2901 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2902 int fd, char *devname)
2903 {
2904 struct intel_super *super = st->sb;
2905 struct imsm_super *mpb = super->anchor;
2906 struct dl *dl;
2907 struct imsm_dev *dev;
2908 struct imsm_map *map;
2909
2910 dev = get_imsm_dev(super, super->current_vol);
2911 map = get_imsm_map(dev, 0);
2912
2913 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2914 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2915 devname);
2916 return 1;
2917 }
2918
2919 if (fd == -1) {
2920 /* we're doing autolayout so grab the pre-marked (in
2921 * validate_geometry) raid_disk
2922 */
2923 for (dl = super->disks; dl; dl = dl->next)
2924 if (dl->raiddisk == dk->raid_disk)
2925 break;
2926 } else {
2927 for (dl = super->disks; dl ; dl = dl->next)
2928 if (dl->major == dk->major &&
2929 dl->minor == dk->minor)
2930 break;
2931 }
2932
2933 if (!dl) {
2934 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2935 return 1;
2936 }
2937
2938 /* add a pristine spare to the metadata */
2939 if (dl->index < 0) {
2940 dl->index = super->anchor->num_disks;
2941 super->anchor->num_disks++;
2942 }
2943 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2944 dl->disk.status = CONFIGURED_DISK;
2945
2946 /* if we are creating the first raid device update the family number */
2947 if (super->current_vol == 0) {
2948 __u32 sum;
2949 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2950 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2951
2952 if (!_dev || !_disk) {
2953 fprintf(stderr, Name ": BUG mpb setup error\n");
2954 return 1;
2955 }
2956 *_dev = *dev;
2957 *_disk = dl->disk;
2958 sum = random32();
2959 sum += __gen_imsm_checksum(mpb);
2960 mpb->family_num = __cpu_to_le32(sum);
2961 mpb->orig_family_num = mpb->family_num;
2962 }
2963
2964 return 0;
2965 }
2966
2967 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2968 int fd, char *devname)
2969 {
2970 struct intel_super *super = st->sb;
2971 struct dl *dd;
2972 unsigned long long size;
2973 __u32 id;
2974 int rv;
2975 struct stat stb;
2976
2977 /* if we are on an RAID enabled platform check that the disk is
2978 * attached to the raid controller
2979 */
2980 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2981 fprintf(stderr,
2982 Name ": %s is not attached to the raid controller: %s\n",
2983 devname ? : "disk", super->hba);
2984 return 1;
2985 }
2986
2987 if (super->current_vol >= 0)
2988 return add_to_super_imsm_volume(st, dk, fd, devname);
2989
2990 fstat(fd, &stb);
2991 dd = malloc(sizeof(*dd));
2992 if (!dd) {
2993 fprintf(stderr,
2994 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2995 return 1;
2996 }
2997 memset(dd, 0, sizeof(*dd));
2998 dd->major = major(stb.st_rdev);
2999 dd->minor = minor(stb.st_rdev);
3000 dd->index = -1;
3001 dd->devname = devname ? strdup(devname) : NULL;
3002 dd->fd = fd;
3003 dd->e = NULL;
3004 rv = imsm_read_serial(fd, devname, dd->serial);
3005 if (rv) {
3006 fprintf(stderr,
3007 Name ": failed to retrieve scsi serial, aborting\n");
3008 free(dd);
3009 abort();
3010 }
3011
3012 get_dev_size(fd, NULL, &size);
3013 size /= 512;
3014 serialcpy(dd->disk.serial, dd->serial);
3015 dd->disk.total_blocks = __cpu_to_le32(size);
3016 dd->disk.status = SPARE_DISK;
3017 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
3018 dd->disk.scsi_id = __cpu_to_le32(id);
3019 else
3020 dd->disk.scsi_id = __cpu_to_le32(0);
3021
3022 if (st->update_tail) {
3023 dd->next = super->add;
3024 super->add = dd;
3025 } else {
3026 dd->next = super->disks;
3027 super->disks = dd;
3028 }
3029
3030 return 0;
3031 }
3032
3033 static int store_imsm_mpb(int fd, struct imsm_super *mpb);
3034
3035 static union {
3036 char buf[512];
3037 struct imsm_super anchor;
3038 } spare_record __attribute__ ((aligned(512)));
3039
3040 /* spare records have their own family number and do not have any defined raid
3041 * devices
3042 */
3043 static int write_super_imsm_spares(struct intel_super *super, int doclose)
3044 {
3045 struct imsm_super *mpb = super->anchor;
3046 struct imsm_super *spare = &spare_record.anchor;
3047 __u32 sum;
3048 struct dl *d;
3049
3050 spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)),
3051 spare->generation_num = __cpu_to_le32(1UL),
3052 spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
3053 spare->num_disks = 1,
3054 spare->num_raid_devs = 0,
3055 spare->cache_size = mpb->cache_size,
3056 spare->pwr_cycle_count = __cpu_to_le32(1),
3057
3058 snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
3059 MPB_SIGNATURE MPB_VERSION_RAID0);
3060
3061 for (d = super->disks; d; d = d->next) {
3062 if (d->index != -1)
3063 continue;
3064
3065 spare->disk[0] = d->disk;
3066 sum = __gen_imsm_checksum(spare);
3067 spare->family_num = __cpu_to_le32(sum);
3068 spare->orig_family_num = 0;
3069 sum = __gen_imsm_checksum(spare);
3070 spare->check_sum = __cpu_to_le32(sum);
3071
3072 if (store_imsm_mpb(d->fd, spare)) {
3073 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3074 __func__, d->major, d->minor, strerror(errno));
3075 return 1;
3076 }
3077 if (doclose) {
3078 close(d->fd);
3079 d->fd = -1;
3080 }
3081 }
3082
3083 return 0;
3084 }
3085
3086 static int write_super_imsm(struct intel_super *super, int doclose)
3087 {
3088 struct imsm_super *mpb = super->anchor;
3089 struct dl *d;
3090 __u32 generation;
3091 __u32 sum;
3092 int spares = 0;
3093 int i;
3094 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
3095
3096 /* 'generation' is incremented everytime the metadata is written */
3097 generation = __le32_to_cpu(mpb->generation_num);
3098 generation++;
3099 mpb->generation_num = __cpu_to_le32(generation);
3100
3101 /* fix up cases where previous mdadm releases failed to set
3102 * orig_family_num
3103 */
3104 if (mpb->orig_family_num == 0)
3105 mpb->orig_family_num = mpb->family_num;
3106
3107 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
3108 for (d = super->disks; d; d = d->next) {
3109 if (d->index == -1)
3110 spares++;
3111 else
3112 mpb->disk[d->index] = d->disk;
3113 }
3114 for (d = super->missing; d; d = d->next)
3115 mpb->disk[d->index] = d->disk;
3116
3117 for (i = 0; i < mpb->num_raid_devs; i++) {
3118 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
3119
3120 imsm_copy_dev(dev, get_imsm_dev(super, i));
3121 mpb_size += sizeof_imsm_dev(dev, 0);
3122 }
3123 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
3124 mpb->mpb_size = __cpu_to_le32(mpb_size);
3125
3126 /* recalculate checksum */
3127 sum = __gen_imsm_checksum(mpb);
3128 mpb->check_sum = __cpu_to_le32(sum);
3129
3130 /* write the mpb for disks that compose raid devices */
3131 for (d = super->disks; d ; d = d->next) {
3132 if (d->index < 0)
3133 continue;
3134 if (store_imsm_mpb(d->fd, mpb))
3135 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3136 __func__, d->major, d->minor, strerror(errno));
3137 if (doclose) {
3138 close(d->fd);
3139 d->fd = -1;
3140 }
3141 }
3142
3143 if (spares)
3144 return write_super_imsm_spares(super, doclose);
3145
3146 return 0;
3147 }
3148
3149
3150 static int create_array(struct supertype *st, int dev_idx)
3151 {
3152 size_t len;
3153 struct imsm_update_create_array *u;
3154 struct intel_super *super = st->sb;
3155 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
3156 struct imsm_map *map = get_imsm_map(dev, 0);
3157 struct disk_info *inf;
3158 struct imsm_disk *disk;
3159 int i;
3160
3161 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
3162 sizeof(*inf) * map->num_members;
3163 u = malloc(len);
3164 if (!u) {
3165 fprintf(stderr, "%s: failed to allocate update buffer\n",
3166 __func__);
3167 return 1;
3168 }
3169
3170 u->type = update_create_array;
3171 u->dev_idx = dev_idx;
3172 imsm_copy_dev(&u->dev, dev);
3173 inf = get_disk_info(u);
3174 for (i = 0; i < map->num_members; i++) {
3175 int idx = get_imsm_disk_idx(dev, i);
3176
3177 disk = get_imsm_disk(super, idx);
3178 serialcpy(inf[i].serial, disk->serial);
3179 }
3180 append_metadata_update(st, u, len);
3181
3182 return 0;
3183 }
3184
3185 static int _add_disk(struct supertype *st)
3186 {
3187 struct intel_super *super = st->sb;
3188 size_t len;
3189 struct imsm_update_add_disk *u;
3190
3191 if (!super->add)
3192 return 0;
3193
3194 len = sizeof(*u);
3195 u = malloc(len);
3196 if (!u) {
3197 fprintf(stderr, "%s: failed to allocate update buffer\n",
3198 __func__);
3199 return 1;
3200 }
3201
3202 u->type = update_add_disk;
3203 append_metadata_update(st, u, len);
3204
3205 return 0;
3206 }
3207
3208 static int write_init_super_imsm(struct supertype *st)
3209 {
3210 struct intel_super *super = st->sb;
3211 int current_vol = super->current_vol;
3212
3213 /* we are done with current_vol reset it to point st at the container */
3214 super->current_vol = -1;
3215
3216 if (st->update_tail) {
3217 /* queue the recently created array / added disk
3218 * as a metadata update */
3219 struct dl *d;
3220 int rv;
3221
3222 /* determine if we are creating a volume or adding a disk */
3223 if (current_vol < 0) {
3224 /* in the add disk case we are running in mdmon
3225 * context, so don't close fd's
3226 */
3227 return _add_disk(st);
3228 } else
3229 rv = create_array(st, current_vol);
3230
3231 for (d = super->disks; d ; d = d->next) {
3232 close(d->fd);
3233 d->fd = -1;
3234 }
3235
3236 return rv;
3237 } else
3238 return write_super_imsm(st->sb, 1);
3239 }
3240 #endif
3241
3242 static int store_super_imsm(struct supertype *st, int fd)
3243 {
3244 struct intel_super *super = st->sb;
3245 struct imsm_super *mpb = super ? super->anchor : NULL;
3246
3247 if (!mpb)
3248 return 1;
3249
3250 #ifndef MDASSEMBLE
3251 return store_imsm_mpb(fd, mpb);
3252 #else
3253 return 1;
3254 #endif
3255 }
3256
3257 static int imsm_bbm_log_size(struct imsm_super *mpb)
3258 {
3259 return __le32_to_cpu(mpb->bbm_log_size);
3260 }
3261
3262 #ifndef MDASSEMBLE
3263 static int validate_geometry_imsm_container(struct supertype *st, int level,
3264 int layout, int raiddisks, int chunk,
3265 unsigned long long size, char *dev,
3266 unsigned long long *freesize,
3267 int verbose)
3268 {
3269 int fd;
3270 unsigned long long ldsize;
3271 const struct imsm_orom *orom;
3272
3273 if (level != LEVEL_CONTAINER)
3274 return 0;
3275 if (!dev)
3276 return 1;
3277
3278 if (check_env("IMSM_NO_PLATFORM"))
3279 orom = NULL;
3280 else
3281 orom = find_imsm_orom();
3282 if (orom && raiddisks > orom->tds) {
3283 if (verbose)
3284 fprintf(stderr, Name ": %d exceeds maximum number of"
3285 " platform supported disks: %d\n",
3286 raiddisks, orom->tds);
3287 return 0;
3288 }
3289
3290 fd = open(dev, O_RDONLY|O_EXCL, 0);
3291 if (fd < 0) {
3292 if (verbose)
3293 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
3294 dev, strerror(errno));
3295 return 0;
3296 }
3297 if (!get_dev_size(fd, dev, &ldsize)) {
3298 close(fd);
3299 return 0;
3300 }
3301 close(fd);
3302
3303 *freesize = avail_size_imsm(st, ldsize >> 9);
3304
3305 return 1;
3306 }
3307
3308 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
3309 {
3310 const unsigned long long base_start = e[*idx].start;
3311 unsigned long long end = base_start + e[*idx].size;
3312 int i;
3313
3314 if (base_start == end)
3315 return 0;
3316
3317 *idx = *idx + 1;
3318 for (i = *idx; i < num_extents; i++) {
3319 /* extend overlapping extents */
3320 if (e[i].start >= base_start &&
3321 e[i].start <= end) {
3322 if (e[i].size == 0)
3323 return 0;
3324 if (e[i].start + e[i].size > end)
3325 end = e[i].start + e[i].size;
3326 } else if (e[i].start > end) {
3327 *idx = i;
3328 break;
3329 }
3330 }
3331
3332 return end - base_start;
3333 }
3334
3335 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
3336 {
3337 /* build a composite disk with all known extents and generate a new
3338 * 'maxsize' given the "all disks in an array must share a common start
3339 * offset" constraint
3340 */
3341 struct extent *e = calloc(sum_extents, sizeof(*e));
3342 struct dl *dl;
3343 int i, j;
3344 int start_extent;
3345 unsigned long long pos;
3346 unsigned long long start = 0;
3347 unsigned long long maxsize;
3348 unsigned long reserve;
3349
3350 if (!e)
3351 return ~0ULL; /* error */
3352
3353 /* coalesce and sort all extents. also, check to see if we need to
3354 * reserve space between member arrays
3355 */
3356 j = 0;
3357 for (dl = super->disks; dl; dl = dl->next) {
3358 if (!dl->e)
3359 continue;
3360 for (i = 0; i < dl->extent_cnt; i++)
3361 e[j++] = dl->e[i];
3362 }
3363 qsort(e, sum_extents, sizeof(*e), cmp_extent);
3364
3365 /* merge extents */
3366 i = 0;
3367 j = 0;
3368 while (i < sum_extents) {
3369 e[j].start = e[i].start;
3370 e[j].size = find_size(e, &i, sum_extents);
3371 j++;
3372 if (e[j-1].size == 0)
3373 break;
3374 }
3375
3376 pos = 0;
3377 maxsize = 0;
3378 start_extent = 0;
3379 i = 0;
3380 do {
3381 unsigned long long esize;
3382
3383 esize = e[i].start - pos;
3384 if (esize >= maxsize) {
3385 maxsize = esize;
3386 start = pos;
3387 start_extent = i;
3388 }
3389 pos = e[i].start + e[i].size;
3390 i++;
3391 } while (e[i-1].size);
3392 free(e);
3393
3394 if (start_extent > 0)
3395 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
3396 else
3397 reserve = 0;
3398
3399 if (maxsize < reserve)
3400 return ~0ULL;
3401
3402 super->create_offset = ~((__u32) 0);
3403 if (start + reserve > super->create_offset)
3404 return ~0ULL; /* start overflows create_offset */
3405 super->create_offset = start + reserve;
3406
3407 return maxsize - reserve;
3408 }
3409
3410 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
3411 {
3412 if (level < 0 || level == 6 || level == 4)
3413 return 0;
3414
3415 /* if we have an orom prevent invalid raid levels */
3416 if (orom)
3417 switch (level) {
3418 case 0: return imsm_orom_has_raid0(orom);
3419 case 1:
3420 if (raiddisks > 2)
3421 return imsm_orom_has_raid1e(orom);
3422 return imsm_orom_has_raid1(orom) && raiddisks == 2;
3423 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
3424 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
3425 }
3426 else
3427 return 1; /* not on an Intel RAID platform so anything goes */
3428
3429 return 0;
3430 }
3431
3432 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3433 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3434 * FIX ME add ahci details
3435 */
3436 static int validate_geometry_imsm_volume(struct supertype *st, int level,
3437 int layout, int raiddisks, int chunk,
3438 unsigned long long size, char *dev,
3439 unsigned long long *freesize,
3440 int verbose)
3441 {
3442 struct stat stb;
3443 struct intel_super *super = st->sb;
3444 struct imsm_super *mpb;
3445 struct dl *dl;
3446 unsigned long long pos = 0;
3447 unsigned long long maxsize;
3448 struct extent *e;
3449 int i;
3450
3451 /* We must have the container info already read in. */
3452 if (!super)
3453 return 0;
3454 mpb = super->anchor;
3455
3456 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
3457 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3458 level, raiddisks, raiddisks > 1 ? "s" : "");
3459 return 0;
3460 }
3461 if (super->orom && level != 1 &&
3462 !imsm_orom_has_chunk(super->orom, chunk)) {
3463 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
3464 return 0;
3465 }
3466 if (layout != imsm_level_to_layout(level)) {
3467 if (level == 5)
3468 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3469 else if (level == 10)
3470 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3471 else
3472 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3473 layout, level);
3474 return 0;
3475 }
3476
3477 if (!dev) {
3478 /* General test: make sure there is space for
3479 * 'raiddisks' device extents of size 'size' at a given
3480 * offset
3481 */
3482 unsigned long long minsize = size;
3483 unsigned long long start_offset = ~0ULL;
3484 int dcnt = 0;
3485 if (minsize == 0)
3486 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
3487 for (dl = super->disks; dl ; dl = dl->next) {
3488 int found = 0;
3489
3490 pos = 0;
3491 i = 0;
3492 e = get_extents(super, dl);
3493 if (!e) continue;
3494 do {
3495 unsigned long long esize;
3496 esize = e[i].start - pos;
3497 if (esize >= minsize)
3498 found = 1;
3499 if (found && start_offset == ~0ULL) {
3500 start_offset = pos;
3501 break;
3502 } else if (found && pos != start_offset) {
3503 found = 0;
3504 break;
3505 }
3506 pos = e[i].start + e[i].size;
3507 i++;
3508 } while (e[i-1].size);
3509 if (found)
3510 dcnt++;
3511 free(e);
3512 }
3513 if (dcnt < raiddisks) {
3514 if (verbose)
3515 fprintf(stderr, Name ": imsm: Not enough "
3516 "devices with space for this array "
3517 "(%d < %d)\n",
3518 dcnt, raiddisks);
3519 return 0;
3520 }
3521 return 1;
3522 }
3523
3524 /* This device must be a member of the set */
3525 if (stat(dev, &stb) < 0)
3526 return 0;
3527 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3528 return 0;
3529 for (dl = super->disks ; dl ; dl = dl->next) {
3530 if (dl->major == major(stb.st_rdev) &&
3531 dl->minor == minor(stb.st_rdev))
3532 break;
3533 }
3534 if (!dl) {
3535 if (verbose)
3536 fprintf(stderr, Name ": %s is not in the "
3537 "same imsm set\n", dev);
3538 return 0;
3539 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
3540 /* If a volume is present then the current creation attempt
3541 * cannot incorporate new spares because the orom may not
3542 * understand this configuration (all member disks must be
3543 * members of each array in the container).
3544 */
3545 fprintf(stderr, Name ": %s is a spare and a volume"
3546 " is already defined for this container\n", dev);
3547 fprintf(stderr, Name ": The option-rom requires all member"
3548 " disks to be a member of all volumes\n");
3549 return 0;
3550 }
3551
3552 /* retrieve the largest free space block */
3553 e = get_extents(super, dl);
3554 maxsize = 0;
3555 i = 0;
3556 if (e) {
3557 do {
3558 unsigned long long esize;
3559
3560 esize = e[i].start - pos;
3561 if (esize >= maxsize)
3562 maxsize = esize;
3563 pos = e[i].start + e[i].size;
3564 i++;
3565 } while (e[i-1].size);
3566 dl->e = e;
3567 dl->extent_cnt = i;
3568 } else {
3569 if (verbose)
3570 fprintf(stderr, Name ": unable to determine free space for: %s\n",
3571 dev);
3572 return 0;
3573 }
3574 if (maxsize < size) {
3575 if (verbose)
3576 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
3577 dev, maxsize, size);
3578 return 0;
3579 }
3580
3581 /* count total number of extents for merge */
3582 i = 0;
3583 for (dl = super->disks; dl; dl = dl->next)
3584 if (dl->e)
3585 i += dl->extent_cnt;
3586
3587 maxsize = merge_extents(super, i);
3588 if (maxsize < size) {
3589 if (verbose)
3590 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
3591 maxsize, size);
3592 return 0;
3593 } else if (maxsize == ~0ULL) {
3594 if (verbose)
3595 fprintf(stderr, Name ": failed to merge %d extents\n", i);
3596 return 0;
3597 }
3598
3599 *freesize = maxsize;
3600
3601 return 1;
3602 }
3603
3604 static int reserve_space(struct supertype *st, int raiddisks,
3605 unsigned long long size, int chunk,
3606 unsigned long long *freesize)
3607 {
3608 struct intel_super *super = st->sb;
3609 struct imsm_super *mpb = super->anchor;
3610 struct dl *dl;
3611 int i;
3612 int extent_cnt;
3613 struct extent *e;
3614 unsigned long long maxsize;
3615 unsigned long long minsize;
3616 int cnt;
3617 int used;
3618
3619 /* find the largest common start free region of the possible disks */
3620 used = 0;
3621 extent_cnt = 0;
3622 cnt = 0;
3623 for (dl = super->disks; dl; dl = dl->next) {
3624 dl->raiddisk = -1;
3625
3626 if (dl->index >= 0)
3627 used++;
3628
3629 /* don't activate new spares if we are orom constrained
3630 * and there is already a volume active in the container
3631 */
3632 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
3633 continue;
3634
3635 e = get_extents(super, dl);
3636 if (!e)
3637 continue;
3638 for (i = 1; e[i-1].size; i++)
3639 ;
3640 dl->e = e;
3641 dl->extent_cnt = i;
3642 extent_cnt += i;
3643 cnt++;
3644 }
3645
3646 maxsize = merge_extents(super, extent_cnt);
3647 minsize = size;
3648 if (size == 0)
3649 minsize = chunk;
3650
3651 if (cnt < raiddisks ||
3652 (super->orom && used && used != raiddisks) ||
3653 maxsize < minsize) {
3654 fprintf(stderr, Name ": not enough devices with space to create array.\n");
3655 return 0; /* No enough free spaces large enough */
3656 }
3657
3658 if (size == 0) {
3659 size = maxsize;
3660 if (chunk) {
3661 size /= chunk;
3662 size *= chunk;
3663 }
3664 }
3665
3666 cnt = 0;
3667 for (dl = super->disks; dl; dl = dl->next)
3668 if (dl->e)
3669 dl->raiddisk = cnt++;
3670
3671 *freesize = size;
3672
3673 return 1;
3674 }
3675
3676 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
3677 int raiddisks, int chunk, unsigned long long size,
3678 char *dev, unsigned long long *freesize,
3679 int verbose)
3680 {
3681 int fd, cfd;
3682 struct mdinfo *sra;
3683 int is_member = 0;
3684
3685 /* if given unused devices create a container
3686 * if given given devices in a container create a member volume
3687 */
3688 if (level == LEVEL_CONTAINER) {
3689 /* Must be a fresh device to add to a container */
3690 return validate_geometry_imsm_container(st, level, layout,
3691 raiddisks, chunk, size,
3692 dev, freesize,
3693 verbose);
3694 }
3695
3696 if (!dev) {
3697 if (st->sb && freesize) {
3698 /* we are being asked to automatically layout a
3699 * new volume based on the current contents of
3700 * the container. If the the parameters can be
3701 * satisfied reserve_space will record the disks,
3702 * start offset, and size of the volume to be
3703 * created. add_to_super and getinfo_super
3704 * detect when autolayout is in progress.
3705 */
3706 return reserve_space(st, raiddisks, size, chunk, freesize);
3707 }
3708 return 1;
3709 }
3710 if (st->sb) {
3711 /* creating in a given container */
3712 return validate_geometry_imsm_volume(st, level, layout,
3713 raiddisks, chunk, size,
3714 dev, freesize, verbose);
3715 }
3716
3717 /* limit creation to the following levels */
3718 if (!dev)
3719 switch (level) {
3720 case 0:
3721 case 1:
3722 case 10:
3723 case 5:
3724 return 0;
3725 default:
3726 if (verbose)
3727 fprintf(stderr, Name
3728 ": IMSM only supports levels 0,1,5,10\n");
3729 return 1;
3730 }
3731
3732 /* This device needs to be a device in an 'imsm' container */
3733 fd = open(dev, O_RDONLY|O_EXCL, 0);
3734 if (fd >= 0) {
3735 if (verbose)
3736 fprintf(stderr,
3737 Name ": Cannot create this array on device %s\n",
3738 dev);
3739 close(fd);
3740 return 0;
3741 }
3742 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3743 if (verbose)
3744 fprintf(stderr, Name ": Cannot open %s: %s\n",
3745 dev, strerror(errno));
3746 return 0;
3747 }
3748 /* Well, it is in use by someone, maybe an 'imsm' container. */
3749 cfd = open_container(fd);
3750 close(fd);
3751 if (cfd < 0) {
3752 if (verbose)
3753 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3754 dev);
3755 return 0;
3756 }
3757 sra = sysfs_read(cfd, 0, GET_VERSION);
3758 if (sra && sra->array.major_version == -1 &&
3759 strcmp(sra->text_version, "imsm") == 0)
3760 is_member = 1;
3761 sysfs_free(sra);
3762 if (is_member) {
3763 /* This is a member of a imsm container. Load the container
3764 * and try to create a volume
3765 */
3766 struct intel_super *super;
3767
3768 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3769 st->sb = super;
3770 st->container_dev = fd2devnum(cfd);
3771 close(cfd);
3772 return validate_geometry_imsm_volume(st, level, layout,
3773 raiddisks, chunk,
3774 size, dev,
3775 freesize, verbose);
3776 }
3777 }
3778
3779 if (verbose)
3780 fprintf(stderr, Name ": failed container membership check\n");
3781
3782 close(cfd);
3783 return 0;
3784 }
3785 #endif /* MDASSEMBLE */
3786
3787 static struct mdinfo *container_content_imsm(struct supertype *st)
3788 {
3789 /* Given a container loaded by load_super_imsm_all,
3790 * extract information about all the arrays into
3791 * an mdinfo tree.
3792 *
3793 * For each imsm_dev create an mdinfo, fill it in,
3794 * then look for matching devices in super->disks
3795 * and create appropriate device mdinfo.
3796 */
3797 struct intel_super *super = st->sb;
3798 struct imsm_super *mpb = super->anchor;
3799 struct mdinfo *rest = NULL;
3800 int i;
3801
3802 /* do not assemble arrays that might have bad blocks */
3803 if (imsm_bbm_log_size(super->anchor)) {
3804 fprintf(stderr, Name ": BBM log found in metadata. "
3805 "Cannot activate array(s).\n");
3806 return NULL;
3807 }
3808
3809 for (i = 0; i < mpb->num_raid_devs; i++) {
3810 struct imsm_dev *dev = get_imsm_dev(super, i);
3811 struct imsm_map *map = get_imsm_map(dev, 0);
3812 struct mdinfo *this;
3813 int slot;
3814
3815 /* do not publish arrays that are in the middle of an
3816 * unsupported migration
3817 */
3818 if (dev->vol.migr_state &&
3819 (migr_type(dev) == MIGR_GEN_MIGR ||
3820 migr_type(dev) == MIGR_STATE_CHANGE)) {
3821 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
3822 " unsupported migration in progress\n",
3823 dev->volume);
3824 continue;
3825 }
3826
3827 this = malloc(sizeof(*this));
3828 if (!this) {
3829 fprintf(stderr, Name ": failed to allocate %lu bytes\n",
3830 sizeof(*this));
3831 break;
3832 }
3833 memset(this, 0, sizeof(*this));
3834 this->next = rest;
3835
3836 super->current_vol = i;
3837 getinfo_super_imsm_volume(st, this);
3838 for (slot = 0 ; slot < map->num_members; slot++) {
3839 struct mdinfo *info_d;
3840 struct dl *d;
3841 int idx;
3842 int skip;
3843 __u32 ord;
3844
3845 skip = 0;
3846 idx = get_imsm_disk_idx(dev, slot);
3847 ord = get_imsm_ord_tbl_ent(dev, slot);
3848 for (d = super->disks; d ; d = d->next)
3849 if (d->index == idx)
3850 break;
3851
3852 if (d == NULL)
3853 skip = 1;
3854 if (d && is_failed(&d->disk))
3855 skip = 1;
3856 if (ord & IMSM_ORD_REBUILD)
3857 skip = 1;
3858
3859 /*
3860 * if we skip some disks the array will be assmebled degraded;
3861 * reset resync start to avoid a dirty-degraded situation
3862 *
3863 * FIXME handle dirty degraded
3864 */
3865 if (skip && !dev->vol.dirty)
3866 this->resync_start = ~0ULL;
3867 if (skip)
3868 continue;
3869
3870 info_d = malloc(sizeof(*info_d));
3871 if (!info_d) {
3872 fprintf(stderr, Name ": failed to allocate disk"
3873 " for volume %.16s\n", dev->volume);
3874 free(this);
3875 this = rest;
3876 break;
3877 }
3878 memset(info_d, 0, sizeof(*info_d));
3879 info_d->next = this->devs;
3880 this->devs = info_d;
3881
3882 info_d->disk.number = d->index;
3883 info_d->disk.major = d->major;
3884 info_d->disk.minor = d->minor;
3885 info_d->disk.raid_disk = slot;
3886
3887 this->array.working_disks++;
3888
3889 info_d->events = __le32_to_cpu(mpb->generation_num);
3890 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3891 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3892 }
3893 rest = this;
3894 }
3895
3896 return rest;
3897 }
3898
3899
3900 #ifndef MDASSEMBLE
3901 static int imsm_open_new(struct supertype *c, struct active_array *a,
3902 char *inst)
3903 {
3904 struct intel_super *super = c->sb;
3905 struct imsm_super *mpb = super->anchor;
3906
3907 if (atoi(inst) >= mpb->num_raid_devs) {
3908 fprintf(stderr, "%s: subarry index %d, out of range\n",
3909 __func__, atoi(inst));
3910 return -ENODEV;
3911 }
3912
3913 dprintf("imsm: open_new %s\n", inst);
3914 a->info.container_member = atoi(inst);
3915 return 0;
3916 }
3917
3918 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3919 {
3920 struct imsm_map *map = get_imsm_map(dev, 0);
3921
3922 if (!failed)
3923 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3924 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3925
3926 switch (get_imsm_raid_level(map)) {
3927 case 0:
3928 return IMSM_T_STATE_FAILED;
3929 break;
3930 case 1:
3931 if (failed < map->num_members)
3932 return IMSM_T_STATE_DEGRADED;
3933 else
3934 return IMSM_T_STATE_FAILED;
3935 break;
3936 case 10:
3937 {
3938 /**
3939 * check to see if any mirrors have failed, otherwise we
3940 * are degraded. Even numbered slots are mirrored on
3941 * slot+1
3942 */
3943 int i;
3944 /* gcc -Os complains that this is unused */
3945 int insync = insync;
3946
3947 for (i = 0; i < map->num_members; i++) {
3948 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3949 int idx = ord_to_idx(ord);
3950 struct imsm_disk *disk;
3951
3952 /* reset the potential in-sync count on even-numbered
3953 * slots. num_copies is always 2 for imsm raid10
3954 */
3955 if ((i & 1) == 0)
3956 insync = 2;
3957
3958 disk = get_imsm_disk(super, idx);
3959 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
3960 insync--;
3961
3962 /* no in-sync disks left in this mirror the
3963 * array has failed
3964 */
3965 if (insync == 0)
3966 return IMSM_T_STATE_FAILED;
3967 }
3968
3969 return IMSM_T_STATE_DEGRADED;
3970 }
3971 case 5:
3972 if (failed < 2)
3973 return IMSM_T_STATE_DEGRADED;
3974 else
3975 return IMSM_T_STATE_FAILED;
3976 break;
3977 default:
3978 break;
3979 }
3980
3981 return map->map_state;
3982 }
3983
3984 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3985 {
3986 int i;
3987 int failed = 0;
3988 struct imsm_disk *disk;
3989 struct imsm_map *map = get_imsm_map(dev, 0);
3990 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3991 __u32 ord;
3992 int idx;
3993
3994 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3995 * disks that are being rebuilt. New failures are recorded to
3996 * map[0]. So we look through all the disks we started with and
3997 * see if any failures are still present, or if any new ones
3998 * have arrived
3999 *
4000 * FIXME add support for online capacity expansion and
4001 * raid-level-migration
4002 */
4003 for (i = 0; i < prev->num_members; i++) {
4004 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
4005 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
4006 idx = ord_to_idx(ord);
4007
4008 disk = get_imsm_disk(super, idx);
4009 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
4010 failed++;
4011 }
4012
4013 return failed;
4014 }
4015
4016 static int is_resyncing(struct imsm_dev *dev)
4017 {
4018 struct imsm_map *migr_map;
4019
4020 if (!dev->vol.migr_state)
4021 return 0;
4022
4023 if (migr_type(dev) == MIGR_INIT ||
4024 migr_type(dev) == MIGR_REPAIR)
4025 return 1;
4026
4027 migr_map = get_imsm_map(dev, 1);
4028
4029 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
4030 return 1;
4031 else
4032 return 0;
4033 }
4034
4035 static int is_rebuilding(struct imsm_dev *dev)
4036 {
4037 struct imsm_map *migr_map;
4038
4039 if (!dev->vol.migr_state)
4040 return 0;
4041
4042 if (migr_type(dev) != MIGR_REBUILD)
4043 return 0;
4044
4045 migr_map = get_imsm_map(dev, 1);
4046
4047 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
4048 return 1;
4049 else
4050 return 0;
4051 }
4052
4053 /* return true if we recorded new information */
4054 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4055 {
4056 __u32 ord;
4057 int slot;
4058 struct imsm_map *map;
4059
4060 /* new failures are always set in map[0] */
4061 map = get_imsm_map(dev, 0);
4062
4063 slot = get_imsm_disk_slot(map, idx);
4064 if (slot < 0)
4065 return 0;
4066
4067 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
4068 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
4069 return 0;
4070
4071 disk->status |= FAILED_DISK;
4072 disk->status &= ~CONFIGURED_DISK;
4073 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
4074 if (~map->failed_disk_num == 0)
4075 map->failed_disk_num = slot;
4076 return 1;
4077 }
4078
4079 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4080 {
4081 mark_failure(dev, disk, idx);
4082
4083 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
4084 return;
4085
4086 disk->scsi_id = __cpu_to_le32(~(__u32)0);
4087 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
4088 }
4089
4090 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4091 * states are handled in imsm_set_disk() with one exception, when a
4092 * resync is stopped due to a new failure this routine will set the
4093 * 'degraded' state for the array.
4094 */
4095 static int imsm_set_array_state(struct active_array *a, int consistent)
4096 {
4097 int inst = a->info.container_member;
4098 struct intel_super *super = a->container->sb;
4099 struct imsm_dev *dev = get_imsm_dev(super, inst);
4100 struct imsm_map *map = get_imsm_map(dev, 0);
4101 int failed = imsm_count_failed(super, dev);
4102 __u8 map_state = imsm_check_degraded(super, dev, failed);
4103
4104 /* before we activate this array handle any missing disks */
4105 if (consistent == 2 && super->missing) {
4106 struct dl *dl;
4107
4108 dprintf("imsm: mark missing\n");
4109 end_migration(dev, map_state);
4110 for (dl = super->missing; dl; dl = dl->next)
4111 mark_missing(dev, &dl->disk, dl->index);
4112 super->updates_pending++;
4113 }
4114
4115 if (consistent == 2 &&
4116 (!is_resync_complete(a) ||
4117 map_state != IMSM_T_STATE_NORMAL ||
4118 dev->vol.migr_state))
4119 consistent = 0;
4120
4121 if (is_resync_complete(a)) {
4122 /* complete intialization / resync,
4123 * recovery and interrupted recovery is completed in
4124 * ->set_disk
4125 */
4126 if (is_resyncing(dev)) {
4127 dprintf("imsm: mark resync done\n");
4128 end_migration(dev, map_state);
4129 super->updates_pending++;
4130 }
4131 } else if (!is_resyncing(dev) && !failed) {
4132 /* mark the start of the init process if nothing is failed */
4133 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
4134 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
4135 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
4136 else
4137 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
4138 super->updates_pending++;
4139 }
4140
4141 /* FIXME check if we can update curr_migr_unit from resync_start */
4142
4143 /* mark dirty / clean */
4144 if (dev->vol.dirty != !consistent) {
4145 dprintf("imsm: mark '%s' (%llu)\n",
4146 consistent ? "clean" : "dirty", a->resync_start);
4147 if (consistent)
4148 dev->vol.dirty = 0;
4149 else
4150 dev->vol.dirty = 1;
4151 super->updates_pending++;
4152 }
4153 return consistent;
4154 }
4155
4156 static void imsm_set_disk(struct active_array *a, int n, int state)
4157 {
4158 int inst = a->info.container_member;
4159 struct intel_super *super = a->container->sb;
4160 struct imsm_dev *dev = get_imsm_dev(super, inst);
4161 struct imsm_map *map = get_imsm_map(dev, 0);
4162 struct imsm_disk *disk;
4163 int failed;
4164 __u32 ord;
4165 __u8 map_state;
4166
4167 if (n > map->num_members)
4168 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
4169 n, map->num_members - 1);
4170
4171 if (n < 0)
4172 return;
4173
4174 dprintf("imsm: set_disk %d:%x\n", n, state);
4175
4176 ord = get_imsm_ord_tbl_ent(dev, n);
4177 disk = get_imsm_disk(super, ord_to_idx(ord));
4178
4179 /* check for new failures */
4180 if (state & DS_FAULTY) {
4181 if (mark_failure(dev, disk, ord_to_idx(ord)))
4182 super->updates_pending++;
4183 }
4184
4185 /* check if in_sync */
4186 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
4187 struct imsm_map *migr_map = get_imsm_map(dev, 1);
4188
4189 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
4190 super->updates_pending++;
4191 }
4192
4193 failed = imsm_count_failed(super, dev);
4194 map_state = imsm_check_degraded(super, dev, failed);
4195
4196 /* check if recovery complete, newly degraded, or failed */
4197 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
4198 end_migration(dev, map_state);
4199 map = get_imsm_map(dev, 0);
4200 map->failed_disk_num = ~0;
4201 super->updates_pending++;
4202 } else if (map_state == IMSM_T_STATE_DEGRADED &&
4203 map->map_state != map_state &&
4204 !dev->vol.migr_state) {
4205 dprintf("imsm: mark degraded\n");
4206 map->map_state = map_state;
4207 super->updates_pending++;
4208 } else if (map_state == IMSM_T_STATE_FAILED &&
4209 map->map_state != map_state) {
4210 dprintf("imsm: mark failed\n");
4211 end_migration(dev, map_state);
4212 super->updates_pending++;
4213 }
4214 }
4215
4216 static int store_imsm_mpb(int fd, struct imsm_super *mpb)
4217 {
4218 void *buf = mpb;
4219 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
4220 unsigned long long dsize;
4221 unsigned long long sectors;
4222
4223 get_dev_size(fd, NULL, &dsize);
4224
4225 if (mpb_size > 512) {
4226 /* -1 to account for anchor */
4227 sectors = mpb_sectors(mpb) - 1;
4228
4229 /* write the extended mpb to the sectors preceeding the anchor */
4230 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
4231 return 1;
4232
4233 if (write(fd, buf + 512, 512 * sectors) != 512 * sectors)
4234 return 1;
4235 }
4236
4237 /* first block is stored on second to last sector of the disk */
4238 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
4239 return 1;
4240
4241 if (write(fd, buf, 512) != 512)
4242 return 1;
4243
4244 return 0;
4245 }
4246
4247 static void imsm_sync_metadata(struct supertype *container)
4248 {
4249 struct intel_super *super = container->sb;
4250
4251 if (!super->updates_pending)
4252 return;
4253
4254 write_super_imsm(super, 0);
4255
4256 super->updates_pending = 0;
4257 }
4258
4259 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
4260 {
4261 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4262 int i = get_imsm_disk_idx(dev, idx);
4263 struct dl *dl;
4264
4265 for (dl = super->disks; dl; dl = dl->next)
4266 if (dl->index == i)
4267 break;
4268
4269 if (dl && is_failed(&dl->disk))
4270 dl = NULL;
4271
4272 if (dl)
4273 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
4274
4275 return dl;
4276 }
4277
4278 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
4279 struct active_array *a, int activate_new)
4280 {
4281 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4282 int idx = get_imsm_disk_idx(dev, slot);
4283 struct imsm_super *mpb = super->anchor;
4284 struct imsm_map *map;
4285 unsigned long long pos;
4286 struct mdinfo *d;
4287 struct extent *ex;
4288 int i, j;
4289 int found;
4290 __u32 array_start;
4291 __u32 array_end;
4292 struct dl *dl;
4293
4294 for (dl = super->disks; dl; dl = dl->next) {
4295 /* If in this array, skip */
4296 for (d = a->info.devs ; d ; d = d->next)
4297 if (d->state_fd >= 0 &&
4298 d->disk.major == dl->major &&
4299 d->disk.minor == dl->minor) {
4300 dprintf("%x:%x already in array\n", dl->major, dl->minor);
4301 break;
4302 }
4303 if (d)
4304 continue;
4305
4306 /* skip in use or failed drives */
4307 if (is_failed(&dl->disk) || idx == dl->index ||
4308 dl->index == -2) {
4309 dprintf("%x:%x status (failed: %d index: %d)\n",
4310 dl->major, dl->minor, is_failed(&dl->disk), idx);
4311 continue;
4312 }
4313
4314 /* skip pure spares when we are looking for partially
4315 * assimilated drives
4316 */
4317 if (dl->index == -1 && !activate_new)
4318 continue;
4319
4320 /* Does this unused device have the requisite free space?
4321 * It needs to be able to cover all member volumes
4322 */
4323 ex = get_extents(super, dl);
4324 if (!ex) {
4325 dprintf("cannot get extents\n");
4326 continue;
4327 }
4328 for (i = 0; i < mpb->num_raid_devs; i++) {
4329 dev = get_imsm_dev(super, i);
4330 map = get_imsm_map(dev, 0);
4331
4332 /* check if this disk is already a member of
4333 * this array
4334 */
4335 if (get_imsm_disk_slot(map, dl->index) >= 0)
4336 continue;
4337
4338 found = 0;
4339 j = 0;
4340 pos = 0;
4341 array_start = __le32_to_cpu(map->pba_of_lba0);
4342 array_end = array_start +
4343 __le32_to_cpu(map->blocks_per_member) - 1;
4344
4345 do {
4346 /* check that we can start at pba_of_lba0 with
4347 * blocks_per_member of space
4348 */
4349 if (array_start >= pos && array_end < ex[j].start) {
4350 found = 1;
4351 break;
4352 }
4353 pos = ex[j].start + ex[j].size;
4354 j++;
4355 } while (ex[j-1].size);
4356
4357 if (!found)
4358 break;
4359 }
4360
4361 free(ex);
4362 if (i < mpb->num_raid_devs) {
4363 dprintf("%x:%x does not have %u to %u available\n",
4364 dl->major, dl->minor, array_start, array_end);
4365 /* No room */
4366 continue;
4367 }
4368 return dl;
4369 }
4370
4371 return dl;
4372 }
4373
4374 static struct mdinfo *imsm_activate_spare(struct active_array *a,
4375 struct metadata_update **updates)
4376 {
4377 /**
4378 * Find a device with unused free space and use it to replace a
4379 * failed/vacant region in an array. We replace failed regions one a
4380 * array at a time. The result is that a new spare disk will be added
4381 * to the first failed array and after the monitor has finished
4382 * propagating failures the remainder will be consumed.
4383 *
4384 * FIXME add a capability for mdmon to request spares from another
4385 * container.
4386 */
4387
4388 struct intel_super *super = a->container->sb;
4389 int inst = a->info.container_member;
4390 struct imsm_dev *dev = get_imsm_dev(super, inst);
4391 struct imsm_map *map = get_imsm_map(dev, 0);
4392 int failed = a->info.array.raid_disks;
4393 struct mdinfo *rv = NULL;
4394 struct mdinfo *d;
4395 struct mdinfo *di;
4396 struct metadata_update *mu;
4397 struct dl *dl;
4398 struct imsm_update_activate_spare *u;
4399 int num_spares = 0;
4400 int i;
4401
4402 for (d = a->info.devs ; d ; d = d->next) {
4403 if ((d->curr_state & DS_FAULTY) &&
4404 d->state_fd >= 0)
4405 /* wait for Removal to happen */
4406 return NULL;
4407 if (d->state_fd >= 0)
4408 failed--;
4409 }
4410
4411 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4412 inst, failed, a->info.array.raid_disks, a->info.array.level);
4413 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
4414 return NULL;
4415
4416 /* For each slot, if it is not working, find a spare */
4417 for (i = 0; i < a->info.array.raid_disks; i++) {
4418 for (d = a->info.devs ; d ; d = d->next)
4419 if (d->disk.raid_disk == i)
4420 break;
4421 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
4422 if (d && (d->state_fd >= 0))
4423 continue;
4424
4425 /*
4426 * OK, this device needs recovery. Try to re-add the
4427 * previous occupant of this slot, if this fails see if
4428 * we can continue the assimilation of a spare that was
4429 * partially assimilated, finally try to activate a new
4430 * spare.
4431 */
4432 dl = imsm_readd(super, i, a);
4433 if (!dl)
4434 dl = imsm_add_spare(super, i, a, 0);
4435 if (!dl)
4436 dl = imsm_add_spare(super, i, a, 1);
4437 if (!dl)
4438 continue;
4439
4440 /* found a usable disk with enough space */
4441 di = malloc(sizeof(*di));
4442 if (!di)
4443 continue;
4444 memset(di, 0, sizeof(*di));
4445
4446 /* dl->index will be -1 in the case we are activating a
4447 * pristine spare. imsm_process_update() will create a
4448 * new index in this case. Once a disk is found to be
4449 * failed in all member arrays it is kicked from the
4450 * metadata
4451 */
4452 di->disk.number = dl->index;
4453
4454 /* (ab)use di->devs to store a pointer to the device
4455 * we chose
4456 */
4457 di->devs = (struct mdinfo *) dl;
4458
4459 di->disk.raid_disk = i;
4460 di->disk.major = dl->major;
4461 di->disk.minor = dl->minor;
4462 di->disk.state = 0;
4463 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
4464 di->component_size = a->info.component_size;
4465 di->container_member = inst;
4466 super->random = random32();
4467 di->next = rv;
4468 rv = di;
4469 num_spares++;
4470 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
4471 i, di->data_offset);
4472
4473 break;
4474 }
4475
4476 if (!rv)
4477 /* No spares found */
4478 return rv;
4479 /* Now 'rv' has a list of devices to return.
4480 * Create a metadata_update record to update the
4481 * disk_ord_tbl for the array
4482 */
4483 mu = malloc(sizeof(*mu));
4484 if (mu) {
4485 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
4486 if (mu->buf == NULL) {
4487 free(mu);
4488 mu = NULL;
4489 }
4490 }
4491 if (!mu) {
4492 while (rv) {
4493 struct mdinfo *n = rv->next;
4494
4495 free(rv);
4496 rv = n;
4497 }
4498 return NULL;
4499 }
4500
4501 mu->space = NULL;
4502 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
4503 mu->next = *updates;
4504 u = (struct imsm_update_activate_spare *) mu->buf;
4505
4506 for (di = rv ; di ; di = di->next) {
4507 u->type = update_activate_spare;
4508 u->dl = (struct dl *) di->devs;
4509 di->devs = NULL;
4510 u->slot = di->disk.raid_disk;
4511 u->array = inst;
4512 u->next = u + 1;
4513 u++;
4514 }
4515 (u-1)->next = NULL;
4516 *updates = mu;
4517
4518 return rv;
4519 }
4520
4521 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
4522 {
4523 struct imsm_dev *dev = get_imsm_dev(super, idx);
4524 struct imsm_map *map = get_imsm_map(dev, 0);
4525 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
4526 struct disk_info *inf = get_disk_info(u);
4527 struct imsm_disk *disk;
4528 int i;
4529 int j;
4530
4531 for (i = 0; i < map->num_members; i++) {
4532 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4533 for (j = 0; j < new_map->num_members; j++)
4534 if (serialcmp(disk->serial, inf[j].serial) == 0)
4535 return 1;
4536 }
4537
4538 return 0;
4539 }
4540
4541 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
4542
4543 static void imsm_process_update(struct supertype *st,
4544 struct metadata_update *update)
4545 {
4546 /**
4547 * crack open the metadata_update envelope to find the update record
4548 * update can be one of:
4549 * update_activate_spare - a spare device has replaced a failed
4550 * device in an array, update the disk_ord_tbl. If this disk is
4551 * present in all member arrays then also clear the SPARE_DISK
4552 * flag
4553 */
4554 struct intel_super *super = st->sb;
4555 struct imsm_super *mpb;
4556 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4557
4558 /* update requires a larger buf but the allocation failed */
4559 if (super->next_len && !super->next_buf) {
4560 super->next_len = 0;
4561 return;
4562 }
4563
4564 if (super->next_buf) {
4565 memcpy(super->next_buf, super->buf, super->len);
4566 free(super->buf);
4567 super->len = super->next_len;
4568 super->buf = super->next_buf;
4569
4570 super->next_len = 0;
4571 super->next_buf = NULL;
4572 }
4573
4574 mpb = super->anchor;
4575
4576 switch (type) {
4577 case update_activate_spare: {
4578 struct imsm_update_activate_spare *u = (void *) update->buf;
4579 struct imsm_dev *dev = get_imsm_dev(super, u->array);
4580 struct imsm_map *map = get_imsm_map(dev, 0);
4581 struct imsm_map *migr_map;
4582 struct active_array *a;
4583 struct imsm_disk *disk;
4584 __u8 to_state;
4585 struct dl *dl;
4586 unsigned int found;
4587 int failed;
4588 int victim = get_imsm_disk_idx(dev, u->slot);
4589 int i;
4590
4591 for (dl = super->disks; dl; dl = dl->next)
4592 if (dl == u->dl)
4593 break;
4594
4595 if (!dl) {
4596 fprintf(stderr, "error: imsm_activate_spare passed "
4597 "an unknown disk (index: %d)\n",
4598 u->dl->index);
4599 return;
4600 }
4601
4602 super->updates_pending++;
4603
4604 /* count failures (excluding rebuilds and the victim)
4605 * to determine map[0] state
4606 */
4607 failed = 0;
4608 for (i = 0; i < map->num_members; i++) {
4609 if (i == u->slot)
4610 continue;
4611 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4612 if (!disk || is_failed(disk))
4613 failed++;
4614 }
4615
4616 /* adding a pristine spare, assign a new index */
4617 if (dl->index < 0) {
4618 dl->index = super->anchor->num_disks;
4619 super->anchor->num_disks++;
4620 }
4621 disk = &dl->disk;
4622 disk->status |= CONFIGURED_DISK;
4623 disk->status &= ~SPARE_DISK;
4624
4625 /* mark rebuild */
4626 to_state = imsm_check_degraded(super, dev, failed);
4627 map->map_state = IMSM_T_STATE_DEGRADED;
4628 migrate(dev, to_state, MIGR_REBUILD);
4629 migr_map = get_imsm_map(dev, 1);
4630 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
4631 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
4632
4633 /* update the family_num to mark a new container
4634 * generation, being careful to record the existing
4635 * family_num in orig_family_num to clean up after
4636 * earlier mdadm versions that neglected to set it.
4637 */
4638 if (mpb->orig_family_num == 0)
4639 mpb->orig_family_num = mpb->family_num;
4640 mpb->family_num += super->random;
4641
4642 /* count arrays using the victim in the metadata */
4643 found = 0;
4644 for (a = st->arrays; a ; a = a->next) {
4645 dev = get_imsm_dev(super, a->info.container_member);
4646 map = get_imsm_map(dev, 0);
4647
4648 if (get_imsm_disk_slot(map, victim) >= 0)
4649 found++;
4650 }
4651
4652 /* delete the victim if it is no longer being
4653 * utilized anywhere
4654 */
4655 if (!found) {
4656 struct dl **dlp;
4657
4658 /* We know that 'manager' isn't touching anything,
4659 * so it is safe to delete
4660 */
4661 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
4662 if ((*dlp)->index == victim)
4663 break;
4664
4665 /* victim may be on the missing list */
4666 if (!*dlp)
4667 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
4668 if ((*dlp)->index == victim)
4669 break;
4670 imsm_delete(super, dlp, victim);
4671 }
4672 break;
4673 }
4674 case update_create_array: {
4675 /* someone wants to create a new array, we need to be aware of
4676 * a few races/collisions:
4677 * 1/ 'Create' called by two separate instances of mdadm
4678 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4679 * devices that have since been assimilated via
4680 * activate_spare.
4681 * In the event this update can not be carried out mdadm will
4682 * (FIX ME) notice that its update did not take hold.
4683 */
4684 struct imsm_update_create_array *u = (void *) update->buf;
4685 struct intel_dev *dv;
4686 struct imsm_dev *dev;
4687 struct imsm_map *map, *new_map;
4688 unsigned long long start, end;
4689 unsigned long long new_start, new_end;
4690 int i;
4691 struct disk_info *inf;
4692 struct dl *dl;
4693
4694 /* handle racing creates: first come first serve */
4695 if (u->dev_idx < mpb->num_raid_devs) {
4696 dprintf("%s: subarray %d already defined\n",
4697 __func__, u->dev_idx);
4698 goto create_error;
4699 }
4700
4701 /* check update is next in sequence */
4702 if (u->dev_idx != mpb->num_raid_devs) {
4703 dprintf("%s: can not create array %d expected index %d\n",
4704 __func__, u->dev_idx, mpb->num_raid_devs);
4705 goto create_error;
4706 }
4707
4708 new_map = get_imsm_map(&u->dev, 0);
4709 new_start = __le32_to_cpu(new_map->pba_of_lba0);
4710 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
4711 inf = get_disk_info(u);
4712
4713 /* handle activate_spare versus create race:
4714 * check to make sure that overlapping arrays do not include
4715 * overalpping disks
4716 */
4717 for (i = 0; i < mpb->num_raid_devs; i++) {
4718 dev = get_imsm_dev(super, i);
4719 map = get_imsm_map(dev, 0);
4720 start = __le32_to_cpu(map->pba_of_lba0);
4721 end = start + __le32_to_cpu(map->blocks_per_member);
4722 if ((new_start >= start && new_start <= end) ||
4723 (start >= new_start && start <= new_end))
4724 /* overlap */;
4725 else
4726 continue;
4727
4728 if (disks_overlap(super, i, u)) {
4729 dprintf("%s: arrays overlap\n", __func__);
4730 goto create_error;
4731 }
4732 }
4733
4734 /* check that prepare update was successful */
4735 if (!update->space) {
4736 dprintf("%s: prepare update failed\n", __func__);
4737 goto create_error;
4738 }
4739
4740 /* check that all disks are still active before committing
4741 * changes. FIXME: could we instead handle this by creating a
4742 * degraded array? That's probably not what the user expects,
4743 * so better to drop this update on the floor.
4744 */
4745 for (i = 0; i < new_map->num_members; i++) {
4746 dl = serial_to_dl(inf[i].serial, super);
4747 if (!dl) {
4748 dprintf("%s: disk disappeared\n", __func__);
4749 goto create_error;
4750 }
4751 }
4752
4753 super->updates_pending++;
4754
4755 /* convert spares to members and fixup ord_tbl */
4756 for (i = 0; i < new_map->num_members; i++) {
4757 dl = serial_to_dl(inf[i].serial, super);
4758 if (dl->index == -1) {
4759 dl->index = mpb->num_disks;
4760 mpb->num_disks++;
4761 dl->disk.status |= CONFIGURED_DISK;
4762 dl->disk.status &= ~SPARE_DISK;
4763 }
4764 set_imsm_ord_tbl_ent(new_map, i, dl->index);
4765 }
4766
4767 dv = update->space;
4768 dev = dv->dev;
4769 update->space = NULL;
4770 imsm_copy_dev(dev, &u->dev);
4771 dv->index = u->dev_idx;
4772 dv->next = super->devlist;
4773 super->devlist = dv;
4774 mpb->num_raid_devs++;
4775
4776 imsm_update_version_info(super);
4777 break;
4778 create_error:
4779 /* mdmon knows how to release update->space, but not
4780 * ((struct intel_dev *) update->space)->dev
4781 */
4782 if (update->space) {
4783 dv = update->space;
4784 free(dv->dev);
4785 }
4786 break;
4787 }
4788 case update_add_disk:
4789
4790 /* we may be able to repair some arrays if disks are
4791 * being added */
4792 if (super->add) {
4793 struct active_array *a;
4794
4795 super->updates_pending++;
4796 for (a = st->arrays; a; a = a->next)
4797 a->check_degraded = 1;
4798 }
4799 /* add some spares to the metadata */
4800 while (super->add) {
4801 struct dl *al;
4802
4803 al = super->add;
4804 super->add = al->next;
4805 al->next = super->disks;
4806 super->disks = al;
4807 dprintf("%s: added %x:%x\n",
4808 __func__, al->major, al->minor);
4809 }
4810
4811 break;
4812 }
4813 }
4814
4815 static void imsm_prepare_update(struct supertype *st,
4816 struct metadata_update *update)
4817 {
4818 /**
4819 * Allocate space to hold new disk entries, raid-device entries or a new
4820 * mpb if necessary. The manager synchronously waits for updates to
4821 * complete in the monitor, so new mpb buffers allocated here can be
4822 * integrated by the monitor thread without worrying about live pointers
4823 * in the manager thread.
4824 */
4825 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4826 struct intel_super *super = st->sb;
4827 struct imsm_super *mpb = super->anchor;
4828 size_t buf_len;
4829 size_t len = 0;
4830
4831 switch (type) {
4832 case update_create_array: {
4833 struct imsm_update_create_array *u = (void *) update->buf;
4834 struct intel_dev *dv;
4835 struct imsm_dev *dev = &u->dev;
4836 struct imsm_map *map = get_imsm_map(dev, 0);
4837 struct dl *dl;
4838 struct disk_info *inf;
4839 int i;
4840 int activate = 0;
4841
4842 inf = get_disk_info(u);
4843 len = sizeof_imsm_dev(dev, 1);
4844 /* allocate a new super->devlist entry */
4845 dv = malloc(sizeof(*dv));
4846 if (dv) {
4847 dv->dev = malloc(len);
4848 if (dv->dev)
4849 update->space = dv;
4850 else {
4851 free(dv);
4852 update->space = NULL;
4853 }
4854 }
4855
4856 /* count how many spares will be converted to members */
4857 for (i = 0; i < map->num_members; i++) {
4858 dl = serial_to_dl(inf[i].serial, super);
4859 if (!dl) {
4860 /* hmm maybe it failed?, nothing we can do about
4861 * it here
4862 */
4863 continue;
4864 }
4865 if (count_memberships(dl, super) == 0)
4866 activate++;
4867 }
4868 len += activate * sizeof(struct imsm_disk);
4869 break;
4870 default:
4871 break;
4872 }
4873 }
4874
4875 /* check if we need a larger metadata buffer */
4876 if (super->next_buf)
4877 buf_len = super->next_len;
4878 else
4879 buf_len = super->len;
4880
4881 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4882 /* ok we need a larger buf than what is currently allocated
4883 * if this allocation fails process_update will notice that
4884 * ->next_len is set and ->next_buf is NULL
4885 */
4886 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4887 if (super->next_buf)
4888 free(super->next_buf);
4889
4890 super->next_len = buf_len;
4891 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
4892 memset(super->next_buf, 0, buf_len);
4893 else
4894 super->next_buf = NULL;
4895 }
4896 }
4897
4898 /* must be called while manager is quiesced */
4899 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4900 {
4901 struct imsm_super *mpb = super->anchor;
4902 struct dl *iter;
4903 struct imsm_dev *dev;
4904 struct imsm_map *map;
4905 int i, j, num_members;
4906 __u32 ord;
4907
4908 dprintf("%s: deleting device[%d] from imsm_super\n",
4909 __func__, index);
4910
4911 /* shift all indexes down one */
4912 for (iter = super->disks; iter; iter = iter->next)
4913 if (iter->index > index)
4914 iter->index--;
4915 for (iter = super->missing; iter; iter = iter->next)
4916 if (iter->index > index)
4917 iter->index--;
4918
4919 for (i = 0; i < mpb->num_raid_devs; i++) {
4920 dev = get_imsm_dev(super, i);
4921 map = get_imsm_map(dev, 0);
4922 num_members = map->num_members;
4923 for (j = 0; j < num_members; j++) {
4924 /* update ord entries being careful not to propagate
4925 * ord-flags to the first map
4926 */
4927 ord = get_imsm_ord_tbl_ent(dev, j);
4928
4929 if (ord_to_idx(ord) <= index)
4930 continue;
4931
4932 map = get_imsm_map(dev, 0);
4933 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4934 map = get_imsm_map(dev, 1);
4935 if (map)
4936 set_imsm_ord_tbl_ent(map, j, ord - 1);
4937 }
4938 }
4939
4940 mpb->num_disks--;
4941 super->updates_pending++;
4942 if (*dlp) {
4943 struct dl *dl = *dlp;
4944
4945 *dlp = (*dlp)->next;
4946 __free_imsm_disk(dl);
4947 }
4948 }
4949 #endif /* MDASSEMBLE */
4950
4951 struct superswitch super_imsm = {
4952 #ifndef MDASSEMBLE
4953 .examine_super = examine_super_imsm,
4954 .brief_examine_super = brief_examine_super_imsm,
4955 .brief_examine_subarrays = brief_examine_subarrays_imsm,
4956 .export_examine_super = export_examine_super_imsm,
4957 .detail_super = detail_super_imsm,
4958 .brief_detail_super = brief_detail_super_imsm,
4959 .write_init_super = write_init_super_imsm,
4960 .validate_geometry = validate_geometry_imsm,
4961 .add_to_super = add_to_super_imsm,
4962 .detail_platform = detail_platform_imsm,
4963 #endif
4964 .match_home = match_home_imsm,
4965 .uuid_from_super= uuid_from_super_imsm,
4966 .getinfo_super = getinfo_super_imsm,
4967 .update_super = update_super_imsm,
4968
4969 .avail_size = avail_size_imsm,
4970
4971 .compare_super = compare_super_imsm,
4972
4973 .load_super = load_super_imsm,
4974 .init_super = init_super_imsm,
4975 .store_super = store_super_imsm,
4976 .free_super = free_super_imsm,
4977 .match_metadata_desc = match_metadata_desc_imsm,
4978 .container_content = container_content_imsm,
4979 .default_layout = imsm_level_to_layout,
4980
4981 .external = 1,
4982 .name = "imsm",
4983
4984 #ifndef MDASSEMBLE
4985 /* for mdmon */
4986 .open_new = imsm_open_new,
4987 .load_super = load_super_imsm,
4988 .set_array_state= imsm_set_array_state,
4989 .set_disk = imsm_set_disk,
4990 .sync_metadata = imsm_sync_metadata,
4991 .activate_spare = imsm_activate_spare,
4992 .process_update = imsm_process_update,
4993 .prepare_update = imsm_prepare_update,
4994 #endif /* MDASSEMBLE */
4995 };
Unnamed repository; edit this file 'description' to name the repository.