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