]> 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: honor orom constraints for auto-layout
[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 static int is_spare(struct imsm_disk *disk)
623 {
624 return (disk->status & SPARE_DISK) == SPARE_DISK;
625 }
626
627 static int is_configured(struct imsm_disk *disk)
628 {
629 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
630 }
631
632 static int is_failed(struct imsm_disk *disk)
633 {
634 return (disk->status & FAILED_DISK) == FAILED_DISK;
635 }
636
637 #ifndef MDASSEMBLE
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 #ifndef MDASSEMBLE
1683 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1684 {
1685 struct dl *dl;
1686
1687 for (dl = super->disks; dl; dl = dl->next)
1688 if (serialcmp(dl->serial, serial) == 0)
1689 break;
1690
1691 return dl;
1692 }
1693 #endif
1694
1695 static struct imsm_disk *
1696 __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
1697 {
1698 int i;
1699
1700 for (i = 0; i < mpb->num_disks; i++) {
1701 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
1702
1703 if (serialcmp(disk->serial, serial) == 0) {
1704 if (idx)
1705 *idx = i;
1706 return disk;
1707 }
1708 }
1709
1710 return NULL;
1711 }
1712
1713 static int
1714 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1715 {
1716 struct imsm_disk *disk;
1717 struct dl *dl;
1718 struct stat stb;
1719 int rv;
1720 char name[40];
1721 __u8 serial[MAX_RAID_SERIAL_LEN];
1722
1723 rv = imsm_read_serial(fd, devname, serial);
1724
1725 if (rv != 0)
1726 return 2;
1727
1728 dl = calloc(1, sizeof(*dl));
1729 if (!dl) {
1730 if (devname)
1731 fprintf(stderr,
1732 Name ": failed to allocate disk buffer for %s\n",
1733 devname);
1734 return 2;
1735 }
1736
1737 fstat(fd, &stb);
1738 dl->major = major(stb.st_rdev);
1739 dl->minor = minor(stb.st_rdev);
1740 dl->next = super->disks;
1741 dl->fd = keep_fd ? fd : -1;
1742 assert(super->disks == NULL);
1743 super->disks = dl;
1744 serialcpy(dl->serial, serial);
1745 dl->index = -2;
1746 dl->e = NULL;
1747 fd2devname(fd, name);
1748 if (devname)
1749 dl->devname = strdup(devname);
1750 else
1751 dl->devname = strdup(name);
1752
1753 /* look up this disk's index in the current anchor */
1754 disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
1755 if (disk) {
1756 dl->disk = *disk;
1757 /* only set index on disks that are a member of a
1758 * populated contianer, i.e. one with raid_devs
1759 */
1760 if (is_failed(&dl->disk))
1761 dl->index = -2;
1762 else if (is_spare(&dl->disk))
1763 dl->index = -1;
1764 }
1765
1766 return 0;
1767 }
1768
1769 #ifndef MDASSEMBLE
1770 /* When migrating map0 contains the 'destination' state while map1
1771 * contains the current state. When not migrating map0 contains the
1772 * current state. This routine assumes that map[0].map_state is set to
1773 * the current array state before being called.
1774 *
1775 * Migration is indicated by one of the following states
1776 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1777 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1778 * map1state=unitialized)
1779 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1780 * map1state=normal)
1781 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1782 * map1state=degraded)
1783 */
1784 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
1785 {
1786 struct imsm_map *dest;
1787 struct imsm_map *src = get_imsm_map(dev, 0);
1788
1789 dev->vol.migr_state = 1;
1790 set_migr_type(dev, migr_type);
1791 dev->vol.curr_migr_unit = 0;
1792 dest = get_imsm_map(dev, 1);
1793
1794 /* duplicate and then set the target end state in map[0] */
1795 memcpy(dest, src, sizeof_imsm_map(src));
1796 if (migr_type == MIGR_REBUILD) {
1797 __u32 ord;
1798 int i;
1799
1800 for (i = 0; i < src->num_members; i++) {
1801 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
1802 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
1803 }
1804 }
1805
1806 src->map_state = to_state;
1807 }
1808
1809 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1810 {
1811 struct imsm_map *map = get_imsm_map(dev, 0);
1812 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
1813 int i;
1814
1815 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1816 * completed in the last migration.
1817 *
1818 * FIXME add support for online capacity expansion and
1819 * raid-level-migration
1820 */
1821 for (i = 0; i < prev->num_members; i++)
1822 map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i];
1823
1824 dev->vol.migr_state = 0;
1825 dev->vol.curr_migr_unit = 0;
1826 map->map_state = map_state;
1827 }
1828 #endif
1829
1830 static int parse_raid_devices(struct intel_super *super)
1831 {
1832 int i;
1833 struct imsm_dev *dev_new;
1834 size_t len, len_migr;
1835 size_t space_needed = 0;
1836 struct imsm_super *mpb = super->anchor;
1837
1838 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1839 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1840 struct intel_dev *dv;
1841
1842 len = sizeof_imsm_dev(dev_iter, 0);
1843 len_migr = sizeof_imsm_dev(dev_iter, 1);
1844 if (len_migr > len)
1845 space_needed += len_migr - len;
1846
1847 dv = malloc(sizeof(*dv));
1848 if (!dv)
1849 return 1;
1850 dev_new = malloc(len_migr);
1851 if (!dev_new) {
1852 free(dv);
1853 return 1;
1854 }
1855 imsm_copy_dev(dev_new, dev_iter);
1856 dv->dev = dev_new;
1857 dv->index = i;
1858 dv->next = super->devlist;
1859 super->devlist = dv;
1860 }
1861
1862 /* ensure that super->buf is large enough when all raid devices
1863 * are migrating
1864 */
1865 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1866 void *buf;
1867
1868 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1869 if (posix_memalign(&buf, 512, len) != 0)
1870 return 1;
1871
1872 memcpy(buf, super->buf, super->len);
1873 memset(buf + super->len, 0, len - super->len);
1874 free(super->buf);
1875 super->buf = buf;
1876 super->len = len;
1877 }
1878
1879 return 0;
1880 }
1881
1882 /* retrieve a pointer to the bbm log which starts after all raid devices */
1883 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1884 {
1885 void *ptr = NULL;
1886
1887 if (__le32_to_cpu(mpb->bbm_log_size)) {
1888 ptr = mpb;
1889 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1890 }
1891
1892 return ptr;
1893 }
1894
1895 static void __free_imsm(struct intel_super *super, int free_disks);
1896
1897 /* load_imsm_mpb - read matrix metadata
1898 * allocates super->mpb to be freed by free_super
1899 */
1900 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1901 {
1902 unsigned long long dsize;
1903 unsigned long long sectors;
1904 struct stat;
1905 struct imsm_super *anchor;
1906 __u32 check_sum;
1907
1908 get_dev_size(fd, NULL, &dsize);
1909
1910 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1911 if (devname)
1912 fprintf(stderr,
1913 Name ": Cannot seek to anchor block on %s: %s\n",
1914 devname, strerror(errno));
1915 return 1;
1916 }
1917
1918 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1919 if (devname)
1920 fprintf(stderr,
1921 Name ": Failed to allocate imsm anchor buffer"
1922 " on %s\n", devname);
1923 return 1;
1924 }
1925 if (read(fd, anchor, 512) != 512) {
1926 if (devname)
1927 fprintf(stderr,
1928 Name ": Cannot read anchor block on %s: %s\n",
1929 devname, strerror(errno));
1930 free(anchor);
1931 return 1;
1932 }
1933
1934 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1935 if (devname)
1936 fprintf(stderr,
1937 Name ": no IMSM anchor on %s\n", devname);
1938 free(anchor);
1939 return 2;
1940 }
1941
1942 __free_imsm(super, 0);
1943 super->len = ROUND_UP(anchor->mpb_size, 512);
1944 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1945 if (devname)
1946 fprintf(stderr,
1947 Name ": unable to allocate %zu byte mpb buffer\n",
1948 super->len);
1949 free(anchor);
1950 return 2;
1951 }
1952 memcpy(super->buf, anchor, 512);
1953
1954 sectors = mpb_sectors(anchor) - 1;
1955 free(anchor);
1956 if (!sectors) {
1957 check_sum = __gen_imsm_checksum(super->anchor);
1958 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1959 if (devname)
1960 fprintf(stderr,
1961 Name ": IMSM checksum %x != %x on %s\n",
1962 check_sum,
1963 __le32_to_cpu(super->anchor->check_sum),
1964 devname);
1965 return 2;
1966 }
1967
1968 return 0;
1969 }
1970
1971 /* read the extended mpb */
1972 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1973 if (devname)
1974 fprintf(stderr,
1975 Name ": Cannot seek to extended mpb on %s: %s\n",
1976 devname, strerror(errno));
1977 return 1;
1978 }
1979
1980 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1981 if (devname)
1982 fprintf(stderr,
1983 Name ": Cannot read extended mpb on %s: %s\n",
1984 devname, strerror(errno));
1985 return 2;
1986 }
1987
1988 check_sum = __gen_imsm_checksum(super->anchor);
1989 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1990 if (devname)
1991 fprintf(stderr,
1992 Name ": IMSM checksum %x != %x on %s\n",
1993 check_sum, __le32_to_cpu(super->anchor->check_sum),
1994 devname);
1995 return 3;
1996 }
1997
1998 /* FIXME the BBM log is disk specific so we cannot use this global
1999 * buffer for all disks. Ok for now since we only look at the global
2000 * bbm_log_size parameter to gate assembly
2001 */
2002 super->bbm_log = __get_imsm_bbm_log(super->anchor);
2003
2004 return 0;
2005 }
2006
2007 static int
2008 load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
2009 {
2010 int err;
2011
2012 err = load_imsm_mpb(fd, super, devname);
2013 if (err)
2014 return err;
2015 err = load_imsm_disk(fd, super, devname, keep_fd);
2016 if (err)
2017 return err;
2018 err = parse_raid_devices(super);
2019
2020 return err;
2021 }
2022
2023 static void __free_imsm_disk(struct dl *d)
2024 {
2025 if (d->fd >= 0)
2026 close(d->fd);
2027 if (d->devname)
2028 free(d->devname);
2029 if (d->e)
2030 free(d->e);
2031 free(d);
2032
2033 }
2034 static void free_imsm_disks(struct intel_super *super)
2035 {
2036 struct dl *d;
2037
2038 while (super->disks) {
2039 d = super->disks;
2040 super->disks = d->next;
2041 __free_imsm_disk(d);
2042 }
2043 while (super->missing) {
2044 d = super->missing;
2045 super->missing = d->next;
2046 __free_imsm_disk(d);
2047 }
2048
2049 }
2050
2051 /* free all the pieces hanging off of a super pointer */
2052 static void __free_imsm(struct intel_super *super, int free_disks)
2053 {
2054 if (super->buf) {
2055 free(super->buf);
2056 super->buf = NULL;
2057 }
2058 if (free_disks)
2059 free_imsm_disks(super);
2060 free_devlist(super);
2061 if (super->hba) {
2062 free((void *) super->hba);
2063 super->hba = NULL;
2064 }
2065 }
2066
2067 static void free_imsm(struct intel_super *super)
2068 {
2069 __free_imsm(super, 1);
2070 free(super);
2071 }
2072
2073 static void free_super_imsm(struct supertype *st)
2074 {
2075 struct intel_super *super = st->sb;
2076
2077 if (!super)
2078 return;
2079
2080 free_imsm(super);
2081 st->sb = NULL;
2082 }
2083
2084 static struct intel_super *alloc_super(int creating_imsm)
2085 {
2086 struct intel_super *super = malloc(sizeof(*super));
2087
2088 if (super) {
2089 memset(super, 0, sizeof(*super));
2090 super->creating_imsm = creating_imsm;
2091 super->current_vol = -1;
2092 super->create_offset = ~((__u32 ) 0);
2093 if (!check_env("IMSM_NO_PLATFORM"))
2094 super->orom = find_imsm_orom();
2095 if (super->orom && !check_env("IMSM_TEST_OROM")) {
2096 struct sys_dev *list, *ent;
2097
2098 /* find the first intel ahci controller */
2099 list = find_driver_devices("pci", "ahci");
2100 for (ent = list; ent; ent = ent->next)
2101 if (devpath_to_vendor(ent->path) == 0x8086)
2102 break;
2103 if (ent) {
2104 super->hba = ent->path;
2105 ent->path = NULL;
2106 }
2107 free_sys_dev(&list);
2108 }
2109 }
2110
2111 return super;
2112 }
2113
2114 #ifndef MDASSEMBLE
2115 /* find_missing - helper routine for load_super_imsm_all that identifies
2116 * disks that have disappeared from the system. This routine relies on
2117 * the mpb being uptodate, which it is at load time.
2118 */
2119 static int find_missing(struct intel_super *super)
2120 {
2121 int i;
2122 struct imsm_super *mpb = super->anchor;
2123 struct dl *dl;
2124 struct imsm_disk *disk;
2125
2126 for (i = 0; i < mpb->num_disks; i++) {
2127 disk = __get_imsm_disk(mpb, i);
2128 dl = serial_to_dl(disk->serial, super);
2129 if (dl)
2130 continue;
2131
2132 dl = malloc(sizeof(*dl));
2133 if (!dl)
2134 return 1;
2135 dl->major = 0;
2136 dl->minor = 0;
2137 dl->fd = -1;
2138 dl->devname = strdup("missing");
2139 dl->index = i;
2140 serialcpy(dl->serial, disk->serial);
2141 dl->disk = *disk;
2142 dl->e = NULL;
2143 dl->next = super->missing;
2144 super->missing = dl;
2145 }
2146
2147 return 0;
2148 }
2149
2150 static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
2151 {
2152 struct intel_disk *idisk = disk_list;
2153
2154 while (idisk) {
2155 if (serialcmp(idisk->disk.serial, serial) == 0)
2156 break;
2157 idisk = idisk->next;
2158 }
2159
2160 return idisk;
2161 }
2162
2163 static int __prep_thunderdome(struct intel_super **table, int tbl_size,
2164 struct intel_super *super,
2165 struct intel_disk **disk_list)
2166 {
2167 struct imsm_disk *d = &super->disks->disk;
2168 struct imsm_super *mpb = super->anchor;
2169 int i, j;
2170
2171 for (i = 0; i < tbl_size; i++) {
2172 struct imsm_super *tbl_mpb = table[i]->anchor;
2173 struct imsm_disk *tbl_d = &table[i]->disks->disk;
2174
2175 if (tbl_mpb->family_num == mpb->family_num) {
2176 if (tbl_mpb->check_sum == mpb->check_sum) {
2177 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2178 __func__, super->disks->major,
2179 super->disks->minor,
2180 table[i]->disks->major,
2181 table[i]->disks->minor);
2182 break;
2183 }
2184
2185 if (((is_configured(d) && !is_configured(tbl_d)) ||
2186 is_configured(d) == is_configured(tbl_d)) &&
2187 tbl_mpb->generation_num < mpb->generation_num) {
2188 /* current version of the mpb is a
2189 * better candidate than the one in
2190 * super_table, but copy over "cross
2191 * generational" status
2192 */
2193 struct intel_disk *idisk;
2194
2195 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2196 __func__, super->disks->major,
2197 super->disks->minor,
2198 table[i]->disks->major,
2199 table[i]->disks->minor);
2200
2201 idisk = disk_list_get(tbl_d->serial, *disk_list);
2202 if (idisk && is_failed(&idisk->disk))
2203 tbl_d->status |= FAILED_DISK;
2204 break;
2205 } else {
2206 struct intel_disk *idisk;
2207 struct imsm_disk *disk;
2208
2209 /* tbl_mpb is more up to date, but copy
2210 * over cross generational status before
2211 * returning
2212 */
2213 disk = __serial_to_disk(d->serial, mpb, NULL);
2214 if (disk && is_failed(disk))
2215 d->status |= FAILED_DISK;
2216
2217 idisk = disk_list_get(d->serial, *disk_list);
2218 if (idisk) {
2219 idisk->owner = i;
2220 if (disk && is_configured(disk))
2221 idisk->disk.status |= CONFIGURED_DISK;
2222 }
2223
2224 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2225 __func__, super->disks->major,
2226 super->disks->minor,
2227 table[i]->disks->major,
2228 table[i]->disks->minor);
2229
2230 return tbl_size;
2231 }
2232 }
2233 }
2234
2235 if (i >= tbl_size)
2236 table[tbl_size++] = super;
2237 else
2238 table[i] = super;
2239
2240 /* update/extend the merged list of imsm_disk records */
2241 for (j = 0; j < mpb->num_disks; j++) {
2242 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
2243 struct intel_disk *idisk;
2244
2245 idisk = disk_list_get(disk->serial, *disk_list);
2246 if (idisk) {
2247 idisk->disk.status |= disk->status;
2248 if (is_configured(&idisk->disk) ||
2249 is_failed(&idisk->disk))
2250 idisk->disk.status &= ~(SPARE_DISK);
2251 } else {
2252 idisk = calloc(1, sizeof(*idisk));
2253 if (!idisk)
2254 return -1;
2255 idisk->owner = IMSM_UNKNOWN_OWNER;
2256 idisk->disk = *disk;
2257 idisk->next = *disk_list;
2258 *disk_list = idisk;
2259 }
2260
2261 if (serialcmp(idisk->disk.serial, d->serial) == 0)
2262 idisk->owner = i;
2263 }
2264
2265 return tbl_size;
2266 }
2267
2268 static struct intel_super *
2269 validate_members(struct intel_super *super, struct intel_disk *disk_list,
2270 const int owner)
2271 {
2272 struct imsm_super *mpb = super->anchor;
2273 int ok_count = 0;
2274 int i;
2275
2276 for (i = 0; i < mpb->num_disks; i++) {
2277 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2278 struct intel_disk *idisk;
2279
2280 idisk = disk_list_get(disk->serial, disk_list);
2281 if (idisk) {
2282 if (idisk->owner == owner ||
2283 idisk->owner == IMSM_UNKNOWN_OWNER)
2284 ok_count++;
2285 else
2286 dprintf("%s: '%.16s' owner %d != %d\n",
2287 __func__, disk->serial, idisk->owner,
2288 owner);
2289 } else {
2290 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2291 __func__, __le32_to_cpu(mpb->family_num), i,
2292 disk->serial);
2293 break;
2294 }
2295 }
2296
2297 if (ok_count == mpb->num_disks)
2298 return super;
2299 return NULL;
2300 }
2301
2302 static void show_conflicts(__u32 family_num, struct intel_super *super_list)
2303 {
2304 struct intel_super *s;
2305
2306 for (s = super_list; s; s = s->next) {
2307 if (family_num != s->anchor->family_num)
2308 continue;
2309 fprintf(stderr, "Conflict, offlining family %#x on '%s'\n",
2310 __le32_to_cpu(family_num), s->disks->devname);
2311 }
2312 }
2313
2314 static struct intel_super *
2315 imsm_thunderdome(struct intel_super **super_list, int len)
2316 {
2317 struct intel_super *super_table[len];
2318 struct intel_disk *disk_list = NULL;
2319 struct intel_super *champion, *spare;
2320 struct intel_super *s, **del;
2321 int tbl_size = 0;
2322 int conflict;
2323 int i;
2324
2325 memset(super_table, 0, sizeof(super_table));
2326 for (s = *super_list; s; s = s->next)
2327 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
2328
2329 for (i = 0; i < tbl_size; i++) {
2330 struct imsm_disk *d;
2331 struct intel_disk *idisk;
2332 struct imsm_super *mpb = super_table[i]->anchor;
2333
2334 s = super_table[i];
2335 d = &s->disks->disk;
2336
2337 /* 'd' must appear in merged disk list for its
2338 * configuration to be valid
2339 */
2340 idisk = disk_list_get(d->serial, disk_list);
2341 if (idisk && idisk->owner == i)
2342 s = validate_members(s, disk_list, i);
2343 else
2344 s = NULL;
2345
2346 if (!s)
2347 dprintf("%s: marking family: %#x from %d:%d offline\n",
2348 __func__, mpb->family_num,
2349 super_table[i]->disks->major,
2350 super_table[i]->disks->minor);
2351 super_table[i] = s;
2352 }
2353
2354 /* This is where the mdadm implementation differs from the Windows
2355 * driver which has no strict concept of a container. We can only
2356 * assemble one family from a container, so when returning a prodigal
2357 * array member to this system the code will not be able to disambiguate
2358 * the container contents that should be assembled ("foreign" versus
2359 * "local"). It requires user intervention to set the orig_family_num
2360 * to a new value to establish a new container. The Windows driver in
2361 * this situation fixes up the volume name in place and manages the
2362 * foreign array as an independent entity.
2363 */
2364 s = NULL;
2365 spare = NULL;
2366 conflict = 0;
2367 for (i = 0; i < tbl_size; i++) {
2368 struct intel_super *tbl_ent = super_table[i];
2369 int is_spare = 0;
2370
2371 if (!tbl_ent)
2372 continue;
2373
2374 if (tbl_ent->anchor->num_raid_devs == 0) {
2375 spare = tbl_ent;
2376 is_spare = 1;
2377 }
2378
2379 if (s && !is_spare) {
2380 show_conflicts(tbl_ent->anchor->family_num, *super_list);
2381 conflict++;
2382 } else if (!s && !is_spare)
2383 s = tbl_ent;
2384 }
2385
2386 if (!s)
2387 s = spare;
2388 if (!s) {
2389 champion = NULL;
2390 goto out;
2391 }
2392 champion = s;
2393
2394 if (conflict)
2395 fprintf(stderr, "Chose family %#x on '%s', "
2396 "assemble conflicts to new container with '--update=uuid'\n",
2397 __le32_to_cpu(s->anchor->family_num), s->disks->devname);
2398
2399 /* collect all dl's onto 'champion', and update them to
2400 * champion's version of the status
2401 */
2402 for (s = *super_list; s; s = s->next) {
2403 struct imsm_super *mpb = champion->anchor;
2404 struct dl *dl = s->disks;
2405
2406 if (s == champion)
2407 continue;
2408
2409 for (i = 0; i < mpb->num_disks; i++) {
2410 struct imsm_disk *disk;
2411
2412 disk = __serial_to_disk(dl->serial, mpb, &dl->index);
2413 if (disk) {
2414 dl->disk = *disk;
2415 /* only set index on disks that are a member of
2416 * a populated contianer, i.e. one with
2417 * raid_devs
2418 */
2419 if (is_failed(&dl->disk))
2420 dl->index = -2;
2421 else if (is_spare(&dl->disk))
2422 dl->index = -1;
2423 break;
2424 }
2425 }
2426
2427 if (i >= mpb->num_disks) {
2428 struct intel_disk *idisk;
2429
2430 idisk = disk_list_get(dl->serial, disk_list);
2431 if (is_spare(&idisk->disk) &&
2432 !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
2433 dl->index = -1;
2434 else {
2435 dl->index = -2;
2436 continue;
2437 }
2438 }
2439
2440 dl->next = champion->disks;
2441 champion->disks = dl;
2442 s->disks = NULL;
2443 }
2444
2445 /* delete 'champion' from super_list */
2446 for (del = super_list; *del; ) {
2447 if (*del == champion) {
2448 *del = (*del)->next;
2449 break;
2450 } else
2451 del = &(*del)->next;
2452 }
2453 champion->next = NULL;
2454
2455 out:
2456 while (disk_list) {
2457 struct intel_disk *idisk = disk_list;
2458
2459 disk_list = disk_list->next;
2460 free(idisk);
2461 }
2462
2463 return champion;
2464 }
2465
2466 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
2467 char *devname, int keep_fd)
2468 {
2469 struct mdinfo *sra;
2470 struct intel_super *super_list = NULL;
2471 struct intel_super *super = NULL;
2472 int devnum = fd2devnum(fd);
2473 struct mdinfo *sd;
2474 int retry;
2475 int err = 0;
2476 int i;
2477 enum sysfs_read_flags flags;
2478
2479 flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2480 if (mdmon_running(devnum))
2481 flags |= SKIP_GONE_DEVS;
2482
2483 /* check if 'fd' an opened container */
2484 sra = sysfs_read(fd, 0, flags);
2485 if (!sra)
2486 return 1;
2487
2488 if (sra->array.major_version != -1 ||
2489 sra->array.minor_version != -2 ||
2490 strcmp(sra->text_version, "imsm") != 0)
2491 return 1;
2492
2493 /* load all mpbs */
2494 for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
2495 struct intel_super *s = alloc_super(0);
2496 char nm[20];
2497 int dfd;
2498
2499 err = 1;
2500 if (!s)
2501 goto error;
2502 s->next = super_list;
2503 super_list = s;
2504
2505 err = 2;
2506 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2507 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
2508 if (dfd < 0)
2509 goto error;
2510
2511 err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
2512
2513 /* retry the load if we might have raced against mdmon */
2514 if (err == 3 && mdmon_running(devnum))
2515 for (retry = 0; retry < 3; retry++) {
2516 usleep(3000);
2517 err = load_and_parse_mpb(dfd, s, NULL, keep_fd);
2518 if (err != 3)
2519 break;
2520 }
2521 if (!keep_fd)
2522 close(dfd);
2523 if (err)
2524 goto error;
2525 }
2526
2527 /* all mpbs enter, maybe one leaves */
2528 super = imsm_thunderdome(&super_list, i);
2529 if (!super) {
2530 err = 1;
2531 goto error;
2532 }
2533
2534 if (find_missing(super) != 0) {
2535 free_imsm(super);
2536 err = 2;
2537 goto error;
2538 }
2539
2540 if (st->subarray[0]) {
2541 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2542 super->current_vol = atoi(st->subarray);
2543 else {
2544 free_imsm(super);
2545 err = 1;
2546 goto error;
2547 }
2548 }
2549 err = 0;
2550
2551 error:
2552 while (super_list) {
2553 struct intel_super *s = super_list;
2554
2555 super_list = super_list->next;
2556 free_imsm(s);
2557 }
2558
2559 if (err)
2560 return err;
2561
2562 *sbp = super;
2563 st->container_dev = devnum;
2564 if (err == 0 && st->ss == NULL) {
2565 st->ss = &super_imsm;
2566 st->minor_version = 0;
2567 st->max_devs = IMSM_MAX_DEVICES;
2568 }
2569 st->loaded_container = 1;
2570
2571 return 0;
2572 }
2573 #endif
2574
2575 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2576 {
2577 struct intel_super *super;
2578 int rv;
2579
2580 #ifndef MDASSEMBLE
2581 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2582 return 0;
2583 #endif
2584
2585 free_super_imsm(st);
2586
2587 super = alloc_super(0);
2588 if (!super) {
2589 fprintf(stderr,
2590 Name ": malloc of %zu failed.\n",
2591 sizeof(*super));
2592 return 1;
2593 }
2594
2595 rv = load_and_parse_mpb(fd, super, devname, 0);
2596
2597 if (rv) {
2598 if (devname)
2599 fprintf(stderr,
2600 Name ": Failed to load all information "
2601 "sections on %s\n", devname);
2602 free_imsm(super);
2603 return rv;
2604 }
2605
2606 if (st->subarray[0]) {
2607 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2608 super->current_vol = atoi(st->subarray);
2609 else {
2610 free_imsm(super);
2611 return 1;
2612 }
2613 }
2614
2615 st->sb = super;
2616 if (st->ss == NULL) {
2617 st->ss = &super_imsm;
2618 st->minor_version = 0;
2619 st->max_devs = IMSM_MAX_DEVICES;
2620 }
2621 st->loaded_container = 0;
2622
2623 return 0;
2624 }
2625
2626 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2627 {
2628 if (info->level == 1)
2629 return 128;
2630 return info->chunk_size >> 9;
2631 }
2632
2633 static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
2634 {
2635 __u32 num_stripes;
2636
2637 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2638 num_stripes /= num_domains;
2639
2640 return num_stripes;
2641 }
2642
2643 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2644 {
2645 if (info->level == 1)
2646 return info->size * 2;
2647 else
2648 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2649 }
2650
2651 static void imsm_update_version_info(struct intel_super *super)
2652 {
2653 /* update the version and attributes */
2654 struct imsm_super *mpb = super->anchor;
2655 char *version;
2656 struct imsm_dev *dev;
2657 struct imsm_map *map;
2658 int i;
2659
2660 for (i = 0; i < mpb->num_raid_devs; i++) {
2661 dev = get_imsm_dev(super, i);
2662 map = get_imsm_map(dev, 0);
2663 if (__le32_to_cpu(dev->size_high) > 0)
2664 mpb->attributes |= MPB_ATTRIB_2TB;
2665
2666 /* FIXME detect when an array spans a port multiplier */
2667 #if 0
2668 mpb->attributes |= MPB_ATTRIB_PM;
2669 #endif
2670
2671 if (mpb->num_raid_devs > 1 ||
2672 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2673 version = MPB_VERSION_ATTRIBS;
2674 switch (get_imsm_raid_level(map)) {
2675 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2676 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2677 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2678 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2679 }
2680 } else {
2681 if (map->num_members >= 5)
2682 version = MPB_VERSION_5OR6_DISK_ARRAY;
2683 else if (dev->status == DEV_CLONE_N_GO)
2684 version = MPB_VERSION_CNG;
2685 else if (get_imsm_raid_level(map) == 5)
2686 version = MPB_VERSION_RAID5;
2687 else if (map->num_members >= 3)
2688 version = MPB_VERSION_3OR4_DISK_ARRAY;
2689 else if (get_imsm_raid_level(map) == 1)
2690 version = MPB_VERSION_RAID1;
2691 else
2692 version = MPB_VERSION_RAID0;
2693 }
2694 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2695 }
2696 }
2697
2698 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2699 unsigned long long size, char *name,
2700 char *homehost, int *uuid)
2701 {
2702 /* We are creating a volume inside a pre-existing container.
2703 * so st->sb is already set.
2704 */
2705 struct intel_super *super = st->sb;
2706 struct imsm_super *mpb = super->anchor;
2707 struct intel_dev *dv;
2708 struct imsm_dev *dev;
2709 struct imsm_vol *vol;
2710 struct imsm_map *map;
2711 int idx = mpb->num_raid_devs;
2712 int i;
2713 unsigned long long array_blocks;
2714 size_t size_old, size_new;
2715 __u32 num_data_stripes;
2716
2717 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2718 fprintf(stderr, Name": This imsm-container already has the "
2719 "maximum of %d volumes\n", super->orom->vpa);
2720 return 0;
2721 }
2722
2723 /* ensure the mpb is large enough for the new data */
2724 size_old = __le32_to_cpu(mpb->mpb_size);
2725 size_new = disks_to_mpb_size(info->nr_disks);
2726 if (size_new > size_old) {
2727 void *mpb_new;
2728 size_t size_round = ROUND_UP(size_new, 512);
2729
2730 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2731 fprintf(stderr, Name": could not allocate new mpb\n");
2732 return 0;
2733 }
2734 memcpy(mpb_new, mpb, size_old);
2735 free(mpb);
2736 mpb = mpb_new;
2737 super->anchor = mpb_new;
2738 mpb->mpb_size = __cpu_to_le32(size_new);
2739 memset(mpb_new + size_old, 0, size_round - size_old);
2740 }
2741 super->current_vol = idx;
2742 /* when creating the first raid device in this container set num_disks
2743 * to zero, i.e. delete this spare and add raid member devices in
2744 * add_to_super_imsm_volume()
2745 */
2746 if (super->current_vol == 0)
2747 mpb->num_disks = 0;
2748
2749 for (i = 0; i < super->current_vol; i++) {
2750 dev = get_imsm_dev(super, i);
2751 if (strncmp((char *) dev->volume, name,
2752 MAX_RAID_SERIAL_LEN) == 0) {
2753 fprintf(stderr, Name": '%s' is already defined for this container\n",
2754 name);
2755 return 0;
2756 }
2757 }
2758
2759 sprintf(st->subarray, "%d", idx);
2760 dv = malloc(sizeof(*dv));
2761 if (!dv) {
2762 fprintf(stderr, Name ": failed to allocate device list entry\n");
2763 return 0;
2764 }
2765 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2766 if (!dev) {
2767 free(dv);
2768 fprintf(stderr, Name": could not allocate raid device\n");
2769 return 0;
2770 }
2771 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2772 if (info->level == 1)
2773 array_blocks = info_to_blocks_per_member(info);
2774 else
2775 array_blocks = calc_array_size(info->level, info->raid_disks,
2776 info->layout, info->chunk_size,
2777 info->size*2);
2778 /* round array size down to closest MB */
2779 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
2780
2781 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2782 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2783 dev->status = __cpu_to_le32(0);
2784 dev->reserved_blocks = __cpu_to_le32(0);
2785 vol = &dev->vol;
2786 vol->migr_state = 0;
2787 set_migr_type(dev, MIGR_INIT);
2788 vol->dirty = 0;
2789 vol->curr_migr_unit = 0;
2790 map = get_imsm_map(dev, 0);
2791 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2792 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2793 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2794 map->failed_disk_num = ~0;
2795 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2796 IMSM_T_STATE_NORMAL;
2797 map->ddf = 1;
2798
2799 if (info->level == 1 && info->raid_disks > 2) {
2800 fprintf(stderr, Name": imsm does not support more than 2 disks"
2801 "in a raid1 volume\n");
2802 return 0;
2803 }
2804
2805 map->raid_level = info->level;
2806 if (info->level == 10) {
2807 map->raid_level = 1;
2808 map->num_domains = info->raid_disks / 2;
2809 } else if (info->level == 1)
2810 map->num_domains = info->raid_disks;
2811 else
2812 map->num_domains = 1;
2813
2814 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
2815 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
2816
2817 map->num_members = info->raid_disks;
2818 for (i = 0; i < map->num_members; i++) {
2819 /* initialized in add_to_super */
2820 set_imsm_ord_tbl_ent(map, i, 0);
2821 }
2822 mpb->num_raid_devs++;
2823
2824 dv->dev = dev;
2825 dv->index = super->current_vol;
2826 dv->next = super->devlist;
2827 super->devlist = dv;
2828
2829 imsm_update_version_info(super);
2830
2831 return 1;
2832 }
2833
2834 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2835 unsigned long long size, char *name,
2836 char *homehost, int *uuid)
2837 {
2838 /* This is primarily called by Create when creating a new array.
2839 * We will then get add_to_super called for each component, and then
2840 * write_init_super called to write it out to each device.
2841 * For IMSM, Create can create on fresh devices or on a pre-existing
2842 * array.
2843 * To create on a pre-existing array a different method will be called.
2844 * This one is just for fresh drives.
2845 */
2846 struct intel_super *super;
2847 struct imsm_super *mpb;
2848 size_t mpb_size;
2849 char *version;
2850
2851 if (st->sb)
2852 return init_super_imsm_volume(st, info, size, name, homehost, uuid);
2853
2854 if (info)
2855 mpb_size = disks_to_mpb_size(info->nr_disks);
2856 else
2857 mpb_size = 512;
2858
2859 super = alloc_super(1);
2860 if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) {
2861 free(super);
2862 super = NULL;
2863 }
2864 if (!super) {
2865 fprintf(stderr, Name
2866 ": %s could not allocate superblock\n", __func__);
2867 return 0;
2868 }
2869 memset(super->buf, 0, mpb_size);
2870 mpb = super->buf;
2871 mpb->mpb_size = __cpu_to_le32(mpb_size);
2872 st->sb = super;
2873
2874 if (info == NULL) {
2875 /* zeroing superblock */
2876 return 0;
2877 }
2878
2879 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2880
2881 version = (char *) mpb->sig;
2882 strcpy(version, MPB_SIGNATURE);
2883 version += strlen(MPB_SIGNATURE);
2884 strcpy(version, MPB_VERSION_RAID0);
2885
2886 return 1;
2887 }
2888
2889 #ifndef MDASSEMBLE
2890 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2891 int fd, char *devname)
2892 {
2893 struct intel_super *super = st->sb;
2894 struct imsm_super *mpb = super->anchor;
2895 struct dl *dl;
2896 struct imsm_dev *dev;
2897 struct imsm_map *map;
2898
2899 dev = get_imsm_dev(super, super->current_vol);
2900 map = get_imsm_map(dev, 0);
2901
2902 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2903 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2904 devname);
2905 return 1;
2906 }
2907
2908 if (fd == -1) {
2909 /* we're doing autolayout so grab the pre-marked (in
2910 * validate_geometry) raid_disk
2911 */
2912 for (dl = super->disks; dl; dl = dl->next)
2913 if (dl->raiddisk == dk->raid_disk)
2914 break;
2915 } else {
2916 for (dl = super->disks; dl ; dl = dl->next)
2917 if (dl->major == dk->major &&
2918 dl->minor == dk->minor)
2919 break;
2920 }
2921
2922 if (!dl) {
2923 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2924 return 1;
2925 }
2926
2927 /* add a pristine spare to the metadata */
2928 if (dl->index < 0) {
2929 dl->index = super->anchor->num_disks;
2930 super->anchor->num_disks++;
2931 }
2932 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2933 dl->disk.status = CONFIGURED_DISK;
2934
2935 /* if we are creating the first raid device update the family number */
2936 if (super->current_vol == 0) {
2937 __u32 sum;
2938 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2939 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2940
2941 *_dev = *dev;
2942 *_disk = dl->disk;
2943 sum = random32();
2944 sum += __gen_imsm_checksum(mpb);
2945 mpb->family_num = __cpu_to_le32(sum);
2946 mpb->orig_family_num = mpb->family_num;
2947 }
2948
2949 return 0;
2950 }
2951
2952 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2953 int fd, char *devname)
2954 {
2955 struct intel_super *super = st->sb;
2956 struct dl *dd;
2957 unsigned long long size;
2958 __u32 id;
2959 int rv;
2960 struct stat stb;
2961
2962 /* if we are on an RAID enabled platform check that the disk is
2963 * attached to the raid controller
2964 */
2965 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2966 fprintf(stderr,
2967 Name ": %s is not attached to the raid controller: %s\n",
2968 devname ? : "disk", super->hba);
2969 return 1;
2970 }
2971
2972 if (super->current_vol >= 0)
2973 return add_to_super_imsm_volume(st, dk, fd, devname);
2974
2975 fstat(fd, &stb);
2976 dd = malloc(sizeof(*dd));
2977 if (!dd) {
2978 fprintf(stderr,
2979 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2980 return 1;
2981 }
2982 memset(dd, 0, sizeof(*dd));
2983 dd->major = major(stb.st_rdev);
2984 dd->minor = minor(stb.st_rdev);
2985 dd->index = -1;
2986 dd->devname = devname ? strdup(devname) : NULL;
2987 dd->fd = fd;
2988 dd->e = NULL;
2989 rv = imsm_read_serial(fd, devname, dd->serial);
2990 if (rv) {
2991 fprintf(stderr,
2992 Name ": failed to retrieve scsi serial, aborting\n");
2993 free(dd);
2994 abort();
2995 }
2996
2997 get_dev_size(fd, NULL, &size);
2998 size /= 512;
2999 serialcpy(dd->disk.serial, dd->serial);
3000 dd->disk.total_blocks = __cpu_to_le32(size);
3001 dd->disk.status = SPARE_DISK;
3002 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
3003 dd->disk.scsi_id = __cpu_to_le32(id);
3004 else
3005 dd->disk.scsi_id = __cpu_to_le32(0);
3006
3007 if (st->update_tail) {
3008 dd->next = super->add;
3009 super->add = dd;
3010 } else {
3011 dd->next = super->disks;
3012 super->disks = dd;
3013 }
3014
3015 return 0;
3016 }
3017
3018 static int store_imsm_mpb(int fd, struct imsm_super *mpb);
3019
3020 static union {
3021 char buf[512];
3022 struct imsm_super anchor;
3023 } spare_record __attribute__ ((aligned(512)));
3024
3025 /* spare records have their own family number and do not have any defined raid
3026 * devices
3027 */
3028 static int write_super_imsm_spares(struct intel_super *super, int doclose)
3029 {
3030 struct imsm_super *mpb = super->anchor;
3031 struct imsm_super *spare = &spare_record.anchor;
3032 __u32 sum;
3033 struct dl *d;
3034
3035 spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)),
3036 spare->generation_num = __cpu_to_le32(1UL),
3037 spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
3038 spare->num_disks = 1,
3039 spare->num_raid_devs = 0,
3040 spare->cache_size = mpb->cache_size,
3041 spare->pwr_cycle_count = __cpu_to_le32(1),
3042
3043 snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
3044 MPB_SIGNATURE MPB_VERSION_RAID0);
3045
3046 for (d = super->disks; d; d = d->next) {
3047 if (d->index != -1)
3048 continue;
3049
3050 spare->disk[0] = d->disk;
3051 sum = __gen_imsm_checksum(spare);
3052 spare->family_num = __cpu_to_le32(sum);
3053 spare->orig_family_num = 0;
3054 sum = __gen_imsm_checksum(spare);
3055 spare->check_sum = __cpu_to_le32(sum);
3056
3057 if (store_imsm_mpb(d->fd, spare)) {
3058 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3059 __func__, d->major, d->minor, strerror(errno));
3060 return 1;
3061 }
3062 if (doclose) {
3063 close(d->fd);
3064 d->fd = -1;
3065 }
3066 }
3067
3068 return 0;
3069 }
3070
3071 static int write_super_imsm(struct intel_super *super, int doclose)
3072 {
3073 struct imsm_super *mpb = super->anchor;
3074 struct dl *d;
3075 __u32 generation;
3076 __u32 sum;
3077 int spares = 0;
3078 int i;
3079 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
3080
3081 /* 'generation' is incremented everytime the metadata is written */
3082 generation = __le32_to_cpu(mpb->generation_num);
3083 generation++;
3084 mpb->generation_num = __cpu_to_le32(generation);
3085
3086 /* fix up cases where previous mdadm releases failed to set
3087 * orig_family_num
3088 */
3089 if (mpb->orig_family_num == 0)
3090 mpb->orig_family_num = mpb->family_num;
3091
3092 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
3093 for (d = super->disks; d; d = d->next) {
3094 if (d->index == -1)
3095 spares++;
3096 else
3097 mpb->disk[d->index] = d->disk;
3098 }
3099 for (d = super->missing; d; d = d->next)
3100 mpb->disk[d->index] = d->disk;
3101
3102 for (i = 0; i < mpb->num_raid_devs; i++) {
3103 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
3104
3105 imsm_copy_dev(dev, get_imsm_dev(super, i));
3106 mpb_size += sizeof_imsm_dev(dev, 0);
3107 }
3108 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
3109 mpb->mpb_size = __cpu_to_le32(mpb_size);
3110
3111 /* recalculate checksum */
3112 sum = __gen_imsm_checksum(mpb);
3113 mpb->check_sum = __cpu_to_le32(sum);
3114
3115 /* write the mpb for disks that compose raid devices */
3116 for (d = super->disks; d ; d = d->next) {
3117 if (d->index < 0)
3118 continue;
3119 if (store_imsm_mpb(d->fd, mpb))
3120 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3121 __func__, d->major, d->minor, strerror(errno));
3122 if (doclose) {
3123 close(d->fd);
3124 d->fd = -1;
3125 }
3126 }
3127
3128 if (spares)
3129 return write_super_imsm_spares(super, doclose);
3130
3131 return 0;
3132 }
3133
3134
3135 static int create_array(struct supertype *st, int dev_idx)
3136 {
3137 size_t len;
3138 struct imsm_update_create_array *u;
3139 struct intel_super *super = st->sb;
3140 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
3141 struct imsm_map *map = get_imsm_map(dev, 0);
3142 struct disk_info *inf;
3143 struct imsm_disk *disk;
3144 int i;
3145
3146 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
3147 sizeof(*inf) * map->num_members;
3148 u = malloc(len);
3149 if (!u) {
3150 fprintf(stderr, "%s: failed to allocate update buffer\n",
3151 __func__);
3152 return 1;
3153 }
3154
3155 u->type = update_create_array;
3156 u->dev_idx = dev_idx;
3157 imsm_copy_dev(&u->dev, dev);
3158 inf = get_disk_info(u);
3159 for (i = 0; i < map->num_members; i++) {
3160 int idx = get_imsm_disk_idx(dev, i);
3161
3162 disk = get_imsm_disk(super, idx);
3163 serialcpy(inf[i].serial, disk->serial);
3164 }
3165 append_metadata_update(st, u, len);
3166
3167 return 0;
3168 }
3169
3170 static int _add_disk(struct supertype *st)
3171 {
3172 struct intel_super *super = st->sb;
3173 size_t len;
3174 struct imsm_update_add_disk *u;
3175
3176 if (!super->add)
3177 return 0;
3178
3179 len = sizeof(*u);
3180 u = malloc(len);
3181 if (!u) {
3182 fprintf(stderr, "%s: failed to allocate update buffer\n",
3183 __func__);
3184 return 1;
3185 }
3186
3187 u->type = update_add_disk;
3188 append_metadata_update(st, u, len);
3189
3190 return 0;
3191 }
3192
3193 static int write_init_super_imsm(struct supertype *st)
3194 {
3195 struct intel_super *super = st->sb;
3196 int current_vol = super->current_vol;
3197
3198 /* we are done with current_vol reset it to point st at the container */
3199 super->current_vol = -1;
3200
3201 if (st->update_tail) {
3202 /* queue the recently created array / added disk
3203 * as a metadata update */
3204 struct dl *d;
3205 int rv;
3206
3207 /* determine if we are creating a volume or adding a disk */
3208 if (current_vol < 0) {
3209 /* in the add disk case we are running in mdmon
3210 * context, so don't close fd's
3211 */
3212 return _add_disk(st);
3213 } else
3214 rv = create_array(st, current_vol);
3215
3216 for (d = super->disks; d ; d = d->next) {
3217 close(d->fd);
3218 d->fd = -1;
3219 }
3220
3221 return rv;
3222 } else
3223 return write_super_imsm(st->sb, 1);
3224 }
3225 #endif
3226
3227 static int store_super_imsm(struct supertype *st, int fd)
3228 {
3229 struct intel_super *super = st->sb;
3230 struct imsm_super *mpb = super ? super->anchor : NULL;
3231
3232 if (!mpb)
3233 return 1;
3234
3235 #ifndef MDASSEMBLE
3236 return store_imsm_mpb(fd, mpb);
3237 #else
3238 return 1;
3239 #endif
3240 }
3241
3242 static int imsm_bbm_log_size(struct imsm_super *mpb)
3243 {
3244 return __le32_to_cpu(mpb->bbm_log_size);
3245 }
3246
3247 #ifndef MDASSEMBLE
3248 static int validate_geometry_imsm_container(struct supertype *st, int level,
3249 int layout, int raiddisks, int chunk,
3250 unsigned long long size, char *dev,
3251 unsigned long long *freesize,
3252 int verbose)
3253 {
3254 int fd;
3255 unsigned long long ldsize;
3256 const struct imsm_orom *orom;
3257
3258 if (level != LEVEL_CONTAINER)
3259 return 0;
3260 if (!dev)
3261 return 1;
3262
3263 if (check_env("IMSM_NO_PLATFORM"))
3264 orom = NULL;
3265 else
3266 orom = find_imsm_orom();
3267 if (orom && raiddisks > orom->tds) {
3268 if (verbose)
3269 fprintf(stderr, Name ": %d exceeds maximum number of"
3270 " platform supported disks: %d\n",
3271 raiddisks, orom->tds);
3272 return 0;
3273 }
3274
3275 fd = open(dev, O_RDONLY|O_EXCL, 0);
3276 if (fd < 0) {
3277 if (verbose)
3278 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
3279 dev, strerror(errno));
3280 return 0;
3281 }
3282 if (!get_dev_size(fd, dev, &ldsize)) {
3283 close(fd);
3284 return 0;
3285 }
3286 close(fd);
3287
3288 *freesize = avail_size_imsm(st, ldsize >> 9);
3289
3290 return 1;
3291 }
3292
3293 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
3294 {
3295 const unsigned long long base_start = e[*idx].start;
3296 unsigned long long end = base_start + e[*idx].size;
3297 int i;
3298
3299 if (base_start == end)
3300 return 0;
3301
3302 *idx = *idx + 1;
3303 for (i = *idx; i < num_extents; i++) {
3304 /* extend overlapping extents */
3305 if (e[i].start >= base_start &&
3306 e[i].start <= end) {
3307 if (e[i].size == 0)
3308 return 0;
3309 if (e[i].start + e[i].size > end)
3310 end = e[i].start + e[i].size;
3311 } else if (e[i].start > end) {
3312 *idx = i;
3313 break;
3314 }
3315 }
3316
3317 return end - base_start;
3318 }
3319
3320 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
3321 {
3322 /* build a composite disk with all known extents and generate a new
3323 * 'maxsize' given the "all disks in an array must share a common start
3324 * offset" constraint
3325 */
3326 struct extent *e = calloc(sum_extents, sizeof(*e));
3327 struct dl *dl;
3328 int i, j;
3329 int start_extent;
3330 unsigned long long pos;
3331 unsigned long long start = 0;
3332 unsigned long long maxsize;
3333 unsigned long reserve;
3334
3335 if (!e)
3336 return 0;
3337
3338 /* coalesce and sort all extents. also, check to see if we need to
3339 * reserve space between member arrays
3340 */
3341 j = 0;
3342 for (dl = super->disks; dl; dl = dl->next) {
3343 if (!dl->e)
3344 continue;
3345 for (i = 0; i < dl->extent_cnt; i++)
3346 e[j++] = dl->e[i];
3347 }
3348 qsort(e, sum_extents, sizeof(*e), cmp_extent);
3349
3350 /* merge extents */
3351 i = 0;
3352 j = 0;
3353 while (i < sum_extents) {
3354 e[j].start = e[i].start;
3355 e[j].size = find_size(e, &i, sum_extents);
3356 j++;
3357 if (e[j-1].size == 0)
3358 break;
3359 }
3360
3361 pos = 0;
3362 maxsize = 0;
3363 start_extent = 0;
3364 i = 0;
3365 do {
3366 unsigned long long esize;
3367
3368 esize = e[i].start - pos;
3369 if (esize >= maxsize) {
3370 maxsize = esize;
3371 start = pos;
3372 start_extent = i;
3373 }
3374 pos = e[i].start + e[i].size;
3375 i++;
3376 } while (e[i-1].size);
3377 free(e);
3378
3379 if (maxsize == 0)
3380 return 0;
3381
3382 /* FIXME assumes volume at offset 0 is the first volume in a
3383 * container
3384 */
3385 if (start_extent > 0)
3386 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
3387 else
3388 reserve = 0;
3389
3390 if (maxsize < reserve)
3391 return 0;
3392
3393 super->create_offset = ~((__u32) 0);
3394 if (start + reserve > super->create_offset)
3395 return 0; /* start overflows create_offset */
3396 super->create_offset = start + reserve;
3397
3398 return maxsize - reserve;
3399 }
3400
3401 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
3402 {
3403 if (level < 0 || level == 6 || level == 4)
3404 return 0;
3405
3406 /* if we have an orom prevent invalid raid levels */
3407 if (orom)
3408 switch (level) {
3409 case 0: return imsm_orom_has_raid0(orom);
3410 case 1:
3411 if (raiddisks > 2)
3412 return imsm_orom_has_raid1e(orom);
3413 return imsm_orom_has_raid1(orom) && raiddisks == 2;
3414 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
3415 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
3416 }
3417 else
3418 return 1; /* not on an Intel RAID platform so anything goes */
3419
3420 return 0;
3421 }
3422
3423 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3424 static int
3425 validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
3426 int raiddisks, int chunk, int verbose)
3427 {
3428 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
3429 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3430 level, raiddisks, raiddisks > 1 ? "s" : "");
3431 return 0;
3432 }
3433 if (super->orom && level != 1 &&
3434 !imsm_orom_has_chunk(super->orom, chunk)) {
3435 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
3436 return 0;
3437 }
3438 if (layout != imsm_level_to_layout(level)) {
3439 if (level == 5)
3440 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3441 else if (level == 10)
3442 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3443 else
3444 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3445 layout, level);
3446 return 0;
3447 }
3448
3449 return 1;
3450 }
3451
3452 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3453 * FIX ME add ahci details
3454 */
3455 static int validate_geometry_imsm_volume(struct supertype *st, int level,
3456 int layout, int raiddisks, int chunk,
3457 unsigned long long size, char *dev,
3458 unsigned long long *freesize,
3459 int verbose)
3460 {
3461 struct stat stb;
3462 struct intel_super *super = st->sb;
3463 struct imsm_super *mpb = super->anchor;
3464 struct dl *dl;
3465 unsigned long long pos = 0;
3466 unsigned long long maxsize;
3467 struct extent *e;
3468 int i;
3469
3470 /* We must have the container info already read in. */
3471 if (!super)
3472 return 0;
3473
3474 if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose))
3475 return 0;
3476
3477 if (!dev) {
3478 /* General test: make sure there is space for
3479 * 'raiddisks' device extents of size 'size' at a given
3480 * offset
3481 */
3482 unsigned long long minsize = size;
3483 unsigned long long start_offset = ~0ULL;
3484 int dcnt = 0;
3485 if (minsize == 0)
3486 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
3487 for (dl = super->disks; dl ; dl = dl->next) {
3488 int found = 0;
3489
3490 pos = 0;
3491 i = 0;
3492 e = get_extents(super, dl);
3493 if (!e) continue;
3494 do {
3495 unsigned long long esize;
3496 esize = e[i].start - pos;
3497 if (esize >= minsize)
3498 found = 1;
3499 if (found && start_offset == ~0ULL) {
3500 start_offset = pos;
3501 break;
3502 } else if (found && pos != start_offset) {
3503 found = 0;
3504 break;
3505 }
3506 pos = e[i].start + e[i].size;
3507 i++;
3508 } while (e[i-1].size);
3509 if (found)
3510 dcnt++;
3511 free(e);
3512 }
3513 if (dcnt < raiddisks) {
3514 if (verbose)
3515 fprintf(stderr, Name ": imsm: Not enough "
3516 "devices with space for this array "
3517 "(%d < %d)\n",
3518 dcnt, raiddisks);
3519 return 0;
3520 }
3521 return 1;
3522 }
3523
3524 /* This device must be a member of the set */
3525 if (stat(dev, &stb) < 0)
3526 return 0;
3527 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3528 return 0;
3529 for (dl = super->disks ; dl ; dl = dl->next) {
3530 if (dl->major == major(stb.st_rdev) &&
3531 dl->minor == minor(stb.st_rdev))
3532 break;
3533 }
3534 if (!dl) {
3535 if (verbose)
3536 fprintf(stderr, Name ": %s is not in the "
3537 "same imsm set\n", dev);
3538 return 0;
3539 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
3540 /* If a volume is present then the current creation attempt
3541 * cannot incorporate new spares because the orom may not
3542 * understand this configuration (all member disks must be
3543 * members of each array in the container).
3544 */
3545 fprintf(stderr, Name ": %s is a spare and a volume"
3546 " is already defined for this container\n", dev);
3547 fprintf(stderr, Name ": The option-rom requires all member"
3548 " disks to be a member of all volumes\n");
3549 return 0;
3550 }
3551
3552 /* retrieve the largest free space block */
3553 e = get_extents(super, dl);
3554 maxsize = 0;
3555 i = 0;
3556 if (e) {
3557 do {
3558 unsigned long long esize;
3559
3560 esize = e[i].start - pos;
3561 if (esize >= maxsize)
3562 maxsize = esize;
3563 pos = e[i].start + e[i].size;
3564 i++;
3565 } while (e[i-1].size);
3566 dl->e = e;
3567 dl->extent_cnt = i;
3568 } else {
3569 if (verbose)
3570 fprintf(stderr, Name ": unable to determine free space for: %s\n",
3571 dev);
3572 return 0;
3573 }
3574 if (maxsize < size) {
3575 if (verbose)
3576 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
3577 dev, maxsize, size);
3578 return 0;
3579 }
3580
3581 /* count total number of extents for merge */
3582 i = 0;
3583 for (dl = super->disks; dl; dl = dl->next)
3584 if (dl->e)
3585 i += dl->extent_cnt;
3586
3587 maxsize = merge_extents(super, i);
3588 if (maxsize < size || maxsize == 0) {
3589 if (verbose)
3590 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
3591 maxsize, size);
3592 return 0;
3593 }
3594
3595 *freesize = maxsize;
3596
3597 return 1;
3598 }
3599
3600 static int reserve_space(struct supertype *st, int raiddisks,
3601 unsigned long long size, int chunk,
3602 unsigned long long *freesize)
3603 {
3604 struct intel_super *super = st->sb;
3605 struct imsm_super *mpb = super->anchor;
3606 struct dl *dl;
3607 int i;
3608 int extent_cnt;
3609 struct extent *e;
3610 unsigned long long maxsize;
3611 unsigned long long minsize;
3612 int cnt;
3613 int used;
3614
3615 /* find the largest common start free region of the possible disks */
3616 used = 0;
3617 extent_cnt = 0;
3618 cnt = 0;
3619 for (dl = super->disks; dl; dl = dl->next) {
3620 dl->raiddisk = -1;
3621
3622 if (dl->index >= 0)
3623 used++;
3624
3625 /* don't activate new spares if we are orom constrained
3626 * and there is already a volume active in the container
3627 */
3628 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
3629 continue;
3630
3631 e = get_extents(super, dl);
3632 if (!e)
3633 continue;
3634 for (i = 1; e[i-1].size; i++)
3635 ;
3636 dl->e = e;
3637 dl->extent_cnt = i;
3638 extent_cnt += i;
3639 cnt++;
3640 }
3641
3642 maxsize = merge_extents(super, extent_cnt);
3643 minsize = size;
3644 if (size == 0)
3645 minsize = chunk;
3646
3647 if (cnt < raiddisks ||
3648 (super->orom && used && used != raiddisks) ||
3649 maxsize < minsize ||
3650 maxsize == 0) {
3651 fprintf(stderr, Name ": not enough devices with space to create array.\n");
3652 return 0; /* No enough free spaces large enough */
3653 }
3654
3655 if (size == 0) {
3656 size = maxsize;
3657 if (chunk) {
3658 size /= chunk;
3659 size *= chunk;
3660 }
3661 }
3662
3663 cnt = 0;
3664 for (dl = super->disks; dl; dl = dl->next)
3665 if (dl->e)
3666 dl->raiddisk = cnt++;
3667
3668 *freesize = size;
3669
3670 return 1;
3671 }
3672
3673 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
3674 int raiddisks, int chunk, unsigned long long size,
3675 char *dev, unsigned long long *freesize,
3676 int verbose)
3677 {
3678 int fd, cfd;
3679 struct mdinfo *sra;
3680
3681 /* if given unused devices create a container
3682 * if given given devices in a container create a member volume
3683 */
3684 if (level == LEVEL_CONTAINER) {
3685 /* Must be a fresh device to add to a container */
3686 return validate_geometry_imsm_container(st, level, layout,
3687 raiddisks, chunk, size,
3688 dev, freesize,
3689 verbose);
3690 }
3691
3692 if (!dev) {
3693 if (st->sb && freesize) {
3694 /* we are being asked to automatically layout a
3695 * new volume based on the current contents of
3696 * the container. If the the parameters can be
3697 * satisfied reserve_space will record the disks,
3698 * start offset, and size of the volume to be
3699 * created. add_to_super and getinfo_super
3700 * detect when autolayout is in progress.
3701 */
3702 if (!validate_geometry_imsm_orom(st->sb, level, layout,
3703 raiddisks, chunk,
3704 verbose))
3705 return 0;
3706 return reserve_space(st, raiddisks, size, chunk, freesize);
3707 }
3708 return 1;
3709 }
3710 if (st->sb) {
3711 /* creating in a given container */
3712 return validate_geometry_imsm_volume(st, level, layout,
3713 raiddisks, chunk, size,
3714 dev, freesize, verbose);
3715 }
3716
3717 /* limit creation to the following levels */
3718 if (!dev)
3719 switch (level) {
3720 case 0:
3721 case 1:
3722 case 10:
3723 case 5:
3724 return 0;
3725 default:
3726 if (verbose)
3727 fprintf(stderr, Name
3728 ": IMSM only supports levels 0,1,5,10\n");
3729 return 1;
3730 }
3731
3732 /* This device needs to be a device in an 'imsm' container */
3733 fd = open(dev, O_RDONLY|O_EXCL, 0);
3734 if (fd >= 0) {
3735 if (verbose)
3736 fprintf(stderr,
3737 Name ": Cannot create this array on device %s\n",
3738 dev);
3739 close(fd);
3740 return 0;
3741 }
3742 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3743 if (verbose)
3744 fprintf(stderr, Name ": Cannot open %s: %s\n",
3745 dev, strerror(errno));
3746 return 0;
3747 }
3748 /* Well, it is in use by someone, maybe an 'imsm' container. */
3749 cfd = open_container(fd);
3750 if (cfd < 0) {
3751 close(fd);
3752 if (verbose)
3753 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3754 dev);
3755 return 0;
3756 }
3757 sra = sysfs_read(cfd, 0, GET_VERSION);
3758 close(fd);
3759 if (sra && sra->array.major_version == -1 &&
3760 strcmp(sra->text_version, "imsm") == 0) {
3761 /* This is a member of a imsm container. Load the container
3762 * and try to create a volume
3763 */
3764 struct intel_super *super;
3765
3766 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3767 st->sb = super;
3768 st->container_dev = fd2devnum(cfd);
3769 close(cfd);
3770 return validate_geometry_imsm_volume(st, level, layout,
3771 raiddisks, chunk,
3772 size, dev,
3773 freesize, verbose);
3774 }
3775 close(cfd);
3776 } else /* may belong to another container */
3777 return 0;
3778
3779 return 1;
3780 }
3781 #endif /* MDASSEMBLE */
3782
3783 static struct mdinfo *container_content_imsm(struct supertype *st)
3784 {
3785 /* Given a container loaded by load_super_imsm_all,
3786 * extract information about all the arrays into
3787 * an mdinfo tree.
3788 *
3789 * For each imsm_dev create an mdinfo, fill it in,
3790 * then look for matching devices in super->disks
3791 * and create appropriate device mdinfo.
3792 */
3793 struct intel_super *super = st->sb;
3794 struct imsm_super *mpb = super->anchor;
3795 struct mdinfo *rest = NULL;
3796 int i;
3797
3798 /* do not assemble arrays that might have bad blocks */
3799 if (imsm_bbm_log_size(super->anchor)) {
3800 fprintf(stderr, Name ": BBM log found in metadata. "
3801 "Cannot activate array(s).\n");
3802 return NULL;
3803 }
3804
3805 for (i = 0; i < mpb->num_raid_devs; i++) {
3806 struct imsm_dev *dev = get_imsm_dev(super, i);
3807 struct imsm_map *map = get_imsm_map(dev, 0);
3808 struct mdinfo *this;
3809 int slot;
3810
3811 /* do not publish arrays that are in the middle of an
3812 * unsupported migration
3813 */
3814 if (dev->vol.migr_state &&
3815 (migr_type(dev) == MIGR_GEN_MIGR ||
3816 migr_type(dev) == MIGR_STATE_CHANGE)) {
3817 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
3818 " unsupported migration in progress\n",
3819 dev->volume);
3820 continue;
3821 }
3822
3823 this = malloc(sizeof(*this));
3824 memset(this, 0, sizeof(*this));
3825 this->next = rest;
3826
3827 super->current_vol = i;
3828 getinfo_super_imsm_volume(st, this);
3829 for (slot = 0 ; slot < map->num_members; slot++) {
3830 struct mdinfo *info_d;
3831 struct dl *d;
3832 int idx;
3833 int skip;
3834 __u32 ord;
3835
3836 skip = 0;
3837 idx = get_imsm_disk_idx(dev, slot);
3838 ord = get_imsm_ord_tbl_ent(dev, slot);
3839 for (d = super->disks; d ; d = d->next)
3840 if (d->index == idx)
3841 break;
3842
3843 if (d == NULL)
3844 skip = 1;
3845 if (d && is_failed(&d->disk))
3846 skip = 1;
3847 if (ord & IMSM_ORD_REBUILD)
3848 skip = 1;
3849
3850 /*
3851 * if we skip some disks the array will be assmebled degraded;
3852 * reset resync start to avoid a dirty-degraded situation
3853 *
3854 * FIXME handle dirty degraded
3855 */
3856 if (skip && !dev->vol.dirty)
3857 this->resync_start = ~0ULL;
3858 if (skip)
3859 continue;
3860
3861 info_d = malloc(sizeof(*info_d));
3862 if (!info_d) {
3863 fprintf(stderr, Name ": failed to allocate disk"
3864 " for volume %.16s\n", dev->volume);
3865 free(this);
3866 this = rest;
3867 break;
3868 }
3869 memset(info_d, 0, sizeof(*info_d));
3870 info_d->next = this->devs;
3871 this->devs = info_d;
3872
3873 info_d->disk.number = d->index;
3874 info_d->disk.major = d->major;
3875 info_d->disk.minor = d->minor;
3876 info_d->disk.raid_disk = slot;
3877
3878 this->array.working_disks++;
3879
3880 info_d->events = __le32_to_cpu(mpb->generation_num);
3881 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3882 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3883 if (d->devname)
3884 strcpy(info_d->name, d->devname);
3885 }
3886 rest = this;
3887 }
3888
3889 return rest;
3890 }
3891
3892
3893 #ifndef MDASSEMBLE
3894 static int imsm_open_new(struct supertype *c, struct active_array *a,
3895 char *inst)
3896 {
3897 struct intel_super *super = c->sb;
3898 struct imsm_super *mpb = super->anchor;
3899
3900 if (atoi(inst) >= mpb->num_raid_devs) {
3901 fprintf(stderr, "%s: subarry index %d, out of range\n",
3902 __func__, atoi(inst));
3903 return -ENODEV;
3904 }
3905
3906 dprintf("imsm: open_new %s\n", inst);
3907 a->info.container_member = atoi(inst);
3908 return 0;
3909 }
3910
3911 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3912 {
3913 struct imsm_map *map = get_imsm_map(dev, 0);
3914
3915 if (!failed)
3916 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3917 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3918
3919 switch (get_imsm_raid_level(map)) {
3920 case 0:
3921 return IMSM_T_STATE_FAILED;
3922 break;
3923 case 1:
3924 if (failed < map->num_members)
3925 return IMSM_T_STATE_DEGRADED;
3926 else
3927 return IMSM_T_STATE_FAILED;
3928 break;
3929 case 10:
3930 {
3931 /**
3932 * check to see if any mirrors have failed, otherwise we
3933 * are degraded. Even numbered slots are mirrored on
3934 * slot+1
3935 */
3936 int i;
3937 /* gcc -Os complains that this is unused */
3938 int insync = insync;
3939
3940 for (i = 0; i < map->num_members; i++) {
3941 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3942 int idx = ord_to_idx(ord);
3943 struct imsm_disk *disk;
3944
3945 /* reset the potential in-sync count on even-numbered
3946 * slots. num_copies is always 2 for imsm raid10
3947 */
3948 if ((i & 1) == 0)
3949 insync = 2;
3950
3951 disk = get_imsm_disk(super, idx);
3952 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
3953 insync--;
3954
3955 /* no in-sync disks left in this mirror the
3956 * array has failed
3957 */
3958 if (insync == 0)
3959 return IMSM_T_STATE_FAILED;
3960 }
3961
3962 return IMSM_T_STATE_DEGRADED;
3963 }
3964 case 5:
3965 if (failed < 2)
3966 return IMSM_T_STATE_DEGRADED;
3967 else
3968 return IMSM_T_STATE_FAILED;
3969 break;
3970 default:
3971 break;
3972 }
3973
3974 return map->map_state;
3975 }
3976
3977 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3978 {
3979 int i;
3980 int failed = 0;
3981 struct imsm_disk *disk;
3982 struct imsm_map *map = get_imsm_map(dev, 0);
3983 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3984 __u32 ord;
3985 int idx;
3986
3987 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3988 * disks that are being rebuilt. New failures are recorded to
3989 * map[0]. So we look through all the disks we started with and
3990 * see if any failures are still present, or if any new ones
3991 * have arrived
3992 *
3993 * FIXME add support for online capacity expansion and
3994 * raid-level-migration
3995 */
3996 for (i = 0; i < prev->num_members; i++) {
3997 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
3998 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
3999 idx = ord_to_idx(ord);
4000
4001 disk = get_imsm_disk(super, idx);
4002 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
4003 failed++;
4004 }
4005
4006 return failed;
4007 }
4008
4009 static int is_resyncing(struct imsm_dev *dev)
4010 {
4011 struct imsm_map *migr_map;
4012
4013 if (!dev->vol.migr_state)
4014 return 0;
4015
4016 if (migr_type(dev) == MIGR_INIT ||
4017 migr_type(dev) == MIGR_REPAIR)
4018 return 1;
4019
4020 migr_map = get_imsm_map(dev, 1);
4021
4022 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
4023 return 1;
4024 else
4025 return 0;
4026 }
4027
4028 static int is_rebuilding(struct imsm_dev *dev)
4029 {
4030 struct imsm_map *migr_map;
4031
4032 if (!dev->vol.migr_state)
4033 return 0;
4034
4035 if (migr_type(dev) != MIGR_REBUILD)
4036 return 0;
4037
4038 migr_map = get_imsm_map(dev, 1);
4039
4040 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
4041 return 1;
4042 else
4043 return 0;
4044 }
4045
4046 /* return true if we recorded new information */
4047 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4048 {
4049 __u32 ord;
4050 int slot;
4051 struct imsm_map *map;
4052
4053 /* new failures are always set in map[0] */
4054 map = get_imsm_map(dev, 0);
4055
4056 slot = get_imsm_disk_slot(map, idx);
4057 if (slot < 0)
4058 return 0;
4059
4060 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
4061 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
4062 return 0;
4063
4064 disk->status |= FAILED_DISK;
4065 disk->status &= ~CONFIGURED_DISK;
4066 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
4067 if (~map->failed_disk_num == 0)
4068 map->failed_disk_num = slot;
4069 return 1;
4070 }
4071
4072 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4073 {
4074 mark_failure(dev, disk, idx);
4075
4076 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
4077 return;
4078
4079 disk->scsi_id = __cpu_to_le32(~(__u32)0);
4080 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
4081 }
4082
4083 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
4084 * states are handled in imsm_set_disk() with one exception, when a
4085 * resync is stopped due to a new failure this routine will set the
4086 * 'degraded' state for the array.
4087 */
4088 static int imsm_set_array_state(struct active_array *a, int consistent)
4089 {
4090 int inst = a->info.container_member;
4091 struct intel_super *super = a->container->sb;
4092 struct imsm_dev *dev = get_imsm_dev(super, inst);
4093 struct imsm_map *map = get_imsm_map(dev, 0);
4094 int failed = imsm_count_failed(super, dev);
4095 __u8 map_state = imsm_check_degraded(super, dev, failed);
4096
4097 /* before we activate this array handle any missing disks */
4098 if (consistent == 2 && super->missing) {
4099 struct dl *dl;
4100
4101 dprintf("imsm: mark missing\n");
4102 end_migration(dev, map_state);
4103 for (dl = super->missing; dl; dl = dl->next)
4104 mark_missing(dev, &dl->disk, dl->index);
4105 super->updates_pending++;
4106 }
4107
4108 if (consistent == 2 &&
4109 (!is_resync_complete(a) ||
4110 map_state != IMSM_T_STATE_NORMAL ||
4111 dev->vol.migr_state))
4112 consistent = 0;
4113
4114 if (is_resync_complete(a)) {
4115 /* complete intialization / resync,
4116 * recovery and interrupted recovery is completed in
4117 * ->set_disk
4118 */
4119 if (is_resyncing(dev)) {
4120 dprintf("imsm: mark resync done\n");
4121 end_migration(dev, map_state);
4122 super->updates_pending++;
4123 }
4124 } else if (!is_resyncing(dev) && !failed) {
4125 /* mark the start of the init process if nothing is failed */
4126 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
4127 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
4128 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
4129 else
4130 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
4131 super->updates_pending++;
4132 }
4133
4134 /* FIXME check if we can update curr_migr_unit from resync_start */
4135
4136 /* mark dirty / clean */
4137 if (dev->vol.dirty != !consistent) {
4138 dprintf("imsm: mark '%s' (%llu)\n",
4139 consistent ? "clean" : "dirty", a->resync_start);
4140 if (consistent)
4141 dev->vol.dirty = 0;
4142 else
4143 dev->vol.dirty = 1;
4144 super->updates_pending++;
4145 }
4146 return consistent;
4147 }
4148
4149 static void imsm_set_disk(struct active_array *a, int n, int state)
4150 {
4151 int inst = a->info.container_member;
4152 struct intel_super *super = a->container->sb;
4153 struct imsm_dev *dev = get_imsm_dev(super, inst);
4154 struct imsm_map *map = get_imsm_map(dev, 0);
4155 struct imsm_disk *disk;
4156 int failed;
4157 __u32 ord;
4158 __u8 map_state;
4159
4160 if (n > map->num_members)
4161 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
4162 n, map->num_members - 1);
4163
4164 if (n < 0)
4165 return;
4166
4167 dprintf("imsm: set_disk %d:%x\n", n, state);
4168
4169 ord = get_imsm_ord_tbl_ent(dev, n);
4170 disk = get_imsm_disk(super, ord_to_idx(ord));
4171
4172 /* check for new failures */
4173 if (state & DS_FAULTY) {
4174 if (mark_failure(dev, disk, ord_to_idx(ord)))
4175 super->updates_pending++;
4176 }
4177
4178 /* check if in_sync */
4179 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
4180 struct imsm_map *migr_map = get_imsm_map(dev, 1);
4181
4182 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
4183 super->updates_pending++;
4184 }
4185
4186 failed = imsm_count_failed(super, dev);
4187 map_state = imsm_check_degraded(super, dev, failed);
4188
4189 /* check if recovery complete, newly degraded, or failed */
4190 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
4191 end_migration(dev, map_state);
4192 map = get_imsm_map(dev, 0);
4193 map->failed_disk_num = ~0;
4194 super->updates_pending++;
4195 } else if (map_state == IMSM_T_STATE_DEGRADED &&
4196 map->map_state != map_state &&
4197 !dev->vol.migr_state) {
4198 dprintf("imsm: mark degraded\n");
4199 map->map_state = map_state;
4200 super->updates_pending++;
4201 } else if (map_state == IMSM_T_STATE_FAILED &&
4202 map->map_state != map_state) {
4203 dprintf("imsm: mark failed\n");
4204 end_migration(dev, map_state);
4205 super->updates_pending++;
4206 }
4207 }
4208
4209 static int store_imsm_mpb(int fd, struct imsm_super *mpb)
4210 {
4211 void *buf = mpb;
4212 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
4213 unsigned long long dsize;
4214 unsigned long long sectors;
4215
4216 get_dev_size(fd, NULL, &dsize);
4217
4218 if (mpb_size > 512) {
4219 /* -1 to account for anchor */
4220 sectors = mpb_sectors(mpb) - 1;
4221
4222 /* write the extended mpb to the sectors preceeding the anchor */
4223 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
4224 return 1;
4225
4226 if (write(fd, buf + 512, 512 * sectors) != 512 * sectors)
4227 return 1;
4228 }
4229
4230 /* first block is stored on second to last sector of the disk */
4231 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
4232 return 1;
4233
4234 if (write(fd, buf, 512) != 512)
4235 return 1;
4236
4237 return 0;
4238 }
4239
4240 static void imsm_sync_metadata(struct supertype *container)
4241 {
4242 struct intel_super *super = container->sb;
4243
4244 if (!super->updates_pending)
4245 return;
4246
4247 write_super_imsm(super, 0);
4248
4249 super->updates_pending = 0;
4250 }
4251
4252 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
4253 {
4254 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4255 int i = get_imsm_disk_idx(dev, idx);
4256 struct dl *dl;
4257
4258 for (dl = super->disks; dl; dl = dl->next)
4259 if (dl->index == i)
4260 break;
4261
4262 if (dl && is_failed(&dl->disk))
4263 dl = NULL;
4264
4265 if (dl)
4266 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
4267
4268 return dl;
4269 }
4270
4271 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
4272 struct active_array *a, int activate_new)
4273 {
4274 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
4275 int idx = get_imsm_disk_idx(dev, slot);
4276 struct imsm_super *mpb = super->anchor;
4277 struct imsm_map *map;
4278 unsigned long long pos;
4279 struct mdinfo *d;
4280 struct extent *ex;
4281 int i, j;
4282 int found;
4283 __u32 array_start;
4284 __u32 array_end;
4285 struct dl *dl;
4286
4287 for (dl = super->disks; dl; dl = dl->next) {
4288 /* If in this array, skip */
4289 for (d = a->info.devs ; d ; d = d->next)
4290 if (d->state_fd >= 0 &&
4291 d->disk.major == dl->major &&
4292 d->disk.minor == dl->minor) {
4293 dprintf("%x:%x already in array\n", dl->major, dl->minor);
4294 break;
4295 }
4296 if (d)
4297 continue;
4298
4299 /* skip in use or failed drives */
4300 if (is_failed(&dl->disk) || idx == dl->index ||
4301 dl->index == -2) {
4302 dprintf("%x:%x status (failed: %d index: %d)\n",
4303 dl->major, dl->minor, is_failed(&dl->disk), idx);
4304 continue;
4305 }
4306
4307 /* skip pure spares when we are looking for partially
4308 * assimilated drives
4309 */
4310 if (dl->index == -1 && !activate_new)
4311 continue;
4312
4313 /* Does this unused device have the requisite free space?
4314 * It needs to be able to cover all member volumes
4315 */
4316 ex = get_extents(super, dl);
4317 if (!ex) {
4318 dprintf("cannot get extents\n");
4319 continue;
4320 }
4321 for (i = 0; i < mpb->num_raid_devs; i++) {
4322 dev = get_imsm_dev(super, i);
4323 map = get_imsm_map(dev, 0);
4324
4325 /* check if this disk is already a member of
4326 * this array
4327 */
4328 if (get_imsm_disk_slot(map, dl->index) >= 0)
4329 continue;
4330
4331 found = 0;
4332 j = 0;
4333 pos = 0;
4334 array_start = __le32_to_cpu(map->pba_of_lba0);
4335 array_end = array_start +
4336 __le32_to_cpu(map->blocks_per_member) - 1;
4337
4338 do {
4339 /* check that we can start at pba_of_lba0 with
4340 * blocks_per_member of space
4341 */
4342 if (array_start >= pos && array_end < ex[j].start) {
4343 found = 1;
4344 break;
4345 }
4346 pos = ex[j].start + ex[j].size;
4347 j++;
4348 } while (ex[j-1].size);
4349
4350 if (!found)
4351 break;
4352 }
4353
4354 free(ex);
4355 if (i < mpb->num_raid_devs) {
4356 dprintf("%x:%x does not have %u to %u available\n",
4357 dl->major, dl->minor, array_start, array_end);
4358 /* No room */
4359 continue;
4360 }
4361 return dl;
4362 }
4363
4364 return dl;
4365 }
4366
4367 static struct mdinfo *imsm_activate_spare(struct active_array *a,
4368 struct metadata_update **updates)
4369 {
4370 /**
4371 * Find a device with unused free space and use it to replace a
4372 * failed/vacant region in an array. We replace failed regions one a
4373 * array at a time. The result is that a new spare disk will be added
4374 * to the first failed array and after the monitor has finished
4375 * propagating failures the remainder will be consumed.
4376 *
4377 * FIXME add a capability for mdmon to request spares from another
4378 * container.
4379 */
4380
4381 struct intel_super *super = a->container->sb;
4382 int inst = a->info.container_member;
4383 struct imsm_dev *dev = get_imsm_dev(super, inst);
4384 struct imsm_map *map = get_imsm_map(dev, 0);
4385 int failed = a->info.array.raid_disks;
4386 struct mdinfo *rv = NULL;
4387 struct mdinfo *d;
4388 struct mdinfo *di;
4389 struct metadata_update *mu;
4390 struct dl *dl;
4391 struct imsm_update_activate_spare *u;
4392 int num_spares = 0;
4393 int i;
4394
4395 for (d = a->info.devs ; d ; d = d->next) {
4396 if ((d->curr_state & DS_FAULTY) &&
4397 d->state_fd >= 0)
4398 /* wait for Removal to happen */
4399 return NULL;
4400 if (d->state_fd >= 0)
4401 failed--;
4402 }
4403
4404 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4405 inst, failed, a->info.array.raid_disks, a->info.array.level);
4406 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
4407 return NULL;
4408
4409 /* For each slot, if it is not working, find a spare */
4410 for (i = 0; i < a->info.array.raid_disks; i++) {
4411 for (d = a->info.devs ; d ; d = d->next)
4412 if (d->disk.raid_disk == i)
4413 break;
4414 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
4415 if (d && (d->state_fd >= 0))
4416 continue;
4417
4418 /*
4419 * OK, this device needs recovery. Try to re-add the
4420 * previous occupant of this slot, if this fails see if
4421 * we can continue the assimilation of a spare that was
4422 * partially assimilated, finally try to activate a new
4423 * spare.
4424 */
4425 dl = imsm_readd(super, i, a);
4426 if (!dl)
4427 dl = imsm_add_spare(super, i, a, 0);
4428 if (!dl)
4429 dl = imsm_add_spare(super, i, a, 1);
4430 if (!dl)
4431 continue;
4432
4433 /* found a usable disk with enough space */
4434 di = malloc(sizeof(*di));
4435 if (!di)
4436 continue;
4437 memset(di, 0, sizeof(*di));
4438
4439 /* dl->index will be -1 in the case we are activating a
4440 * pristine spare. imsm_process_update() will create a
4441 * new index in this case. Once a disk is found to be
4442 * failed in all member arrays it is kicked from the
4443 * metadata
4444 */
4445 di->disk.number = dl->index;
4446
4447 /* (ab)use di->devs to store a pointer to the device
4448 * we chose
4449 */
4450 di->devs = (struct mdinfo *) dl;
4451
4452 di->disk.raid_disk = i;
4453 di->disk.major = dl->major;
4454 di->disk.minor = dl->minor;
4455 di->disk.state = 0;
4456 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
4457 di->component_size = a->info.component_size;
4458 di->container_member = inst;
4459 super->random = random32();
4460 di->next = rv;
4461 rv = di;
4462 num_spares++;
4463 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
4464 i, di->data_offset);
4465
4466 break;
4467 }
4468
4469 if (!rv)
4470 /* No spares found */
4471 return rv;
4472 /* Now 'rv' has a list of devices to return.
4473 * Create a metadata_update record to update the
4474 * disk_ord_tbl for the array
4475 */
4476 mu = malloc(sizeof(*mu));
4477 if (mu) {
4478 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
4479 if (mu->buf == NULL) {
4480 free(mu);
4481 mu = NULL;
4482 }
4483 }
4484 if (!mu) {
4485 while (rv) {
4486 struct mdinfo *n = rv->next;
4487
4488 free(rv);
4489 rv = n;
4490 }
4491 return NULL;
4492 }
4493
4494 mu->space = NULL;
4495 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
4496 mu->next = *updates;
4497 u = (struct imsm_update_activate_spare *) mu->buf;
4498
4499 for (di = rv ; di ; di = di->next) {
4500 u->type = update_activate_spare;
4501 u->dl = (struct dl *) di->devs;
4502 di->devs = NULL;
4503 u->slot = di->disk.raid_disk;
4504 u->array = inst;
4505 u->next = u + 1;
4506 u++;
4507 }
4508 (u-1)->next = NULL;
4509 *updates = mu;
4510
4511 return rv;
4512 }
4513
4514 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
4515 {
4516 struct imsm_dev *dev = get_imsm_dev(super, idx);
4517 struct imsm_map *map = get_imsm_map(dev, 0);
4518 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
4519 struct disk_info *inf = get_disk_info(u);
4520 struct imsm_disk *disk;
4521 int i;
4522 int j;
4523
4524 for (i = 0; i < map->num_members; i++) {
4525 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4526 for (j = 0; j < new_map->num_members; j++)
4527 if (serialcmp(disk->serial, inf[j].serial) == 0)
4528 return 1;
4529 }
4530
4531 return 0;
4532 }
4533
4534 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
4535
4536 static void imsm_process_update(struct supertype *st,
4537 struct metadata_update *update)
4538 {
4539 /**
4540 * crack open the metadata_update envelope to find the update record
4541 * update can be one of:
4542 * update_activate_spare - a spare device has replaced a failed
4543 * device in an array, update the disk_ord_tbl. If this disk is
4544 * present in all member arrays then also clear the SPARE_DISK
4545 * flag
4546 */
4547 struct intel_super *super = st->sb;
4548 struct imsm_super *mpb;
4549 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4550
4551 /* update requires a larger buf but the allocation failed */
4552 if (super->next_len && !super->next_buf) {
4553 super->next_len = 0;
4554 return;
4555 }
4556
4557 if (super->next_buf) {
4558 memcpy(super->next_buf, super->buf, super->len);
4559 free(super->buf);
4560 super->len = super->next_len;
4561 super->buf = super->next_buf;
4562
4563 super->next_len = 0;
4564 super->next_buf = NULL;
4565 }
4566
4567 mpb = super->anchor;
4568
4569 switch (type) {
4570 case update_activate_spare: {
4571 struct imsm_update_activate_spare *u = (void *) update->buf;
4572 struct imsm_dev *dev = get_imsm_dev(super, u->array);
4573 struct imsm_map *map = get_imsm_map(dev, 0);
4574 struct imsm_map *migr_map;
4575 struct active_array *a;
4576 struct imsm_disk *disk;
4577 __u8 to_state;
4578 struct dl *dl;
4579 unsigned int found;
4580 int failed;
4581 int victim = get_imsm_disk_idx(dev, u->slot);
4582 int i;
4583
4584 for (dl = super->disks; dl; dl = dl->next)
4585 if (dl == u->dl)
4586 break;
4587
4588 if (!dl) {
4589 fprintf(stderr, "error: imsm_activate_spare passed "
4590 "an unknown disk (index: %d)\n",
4591 u->dl->index);
4592 return;
4593 }
4594
4595 super->updates_pending++;
4596
4597 /* count failures (excluding rebuilds and the victim)
4598 * to determine map[0] state
4599 */
4600 failed = 0;
4601 for (i = 0; i < map->num_members; i++) {
4602 if (i == u->slot)
4603 continue;
4604 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4605 if (!disk || is_failed(disk))
4606 failed++;
4607 }
4608
4609 /* adding a pristine spare, assign a new index */
4610 if (dl->index < 0) {
4611 dl->index = super->anchor->num_disks;
4612 super->anchor->num_disks++;
4613 }
4614 disk = &dl->disk;
4615 disk->status |= CONFIGURED_DISK;
4616 disk->status &= ~SPARE_DISK;
4617
4618 /* mark rebuild */
4619 to_state = imsm_check_degraded(super, dev, failed);
4620 map->map_state = IMSM_T_STATE_DEGRADED;
4621 migrate(dev, to_state, MIGR_REBUILD);
4622 migr_map = get_imsm_map(dev, 1);
4623 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
4624 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
4625
4626 /* update the family_num to mark a new container
4627 * generation, being careful to record the existing
4628 * family_num in orig_family_num to clean up after
4629 * earlier mdadm versions that neglected to set it.
4630 */
4631 if (mpb->orig_family_num == 0)
4632 mpb->orig_family_num = mpb->family_num;
4633 mpb->family_num += super->random;
4634
4635 /* count arrays using the victim in the metadata */
4636 found = 0;
4637 for (a = st->arrays; a ; a = a->next) {
4638 dev = get_imsm_dev(super, a->info.container_member);
4639 map = get_imsm_map(dev, 0);
4640
4641 if (get_imsm_disk_slot(map, victim) >= 0)
4642 found++;
4643 }
4644
4645 /* delete the victim if it is no longer being
4646 * utilized anywhere
4647 */
4648 if (!found) {
4649 struct dl **dlp;
4650
4651 /* We know that 'manager' isn't touching anything,
4652 * so it is safe to delete
4653 */
4654 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
4655 if ((*dlp)->index == victim)
4656 break;
4657
4658 /* victim may be on the missing list */
4659 if (!*dlp)
4660 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
4661 if ((*dlp)->index == victim)
4662 break;
4663 imsm_delete(super, dlp, victim);
4664 }
4665 break;
4666 }
4667 case update_create_array: {
4668 /* someone wants to create a new array, we need to be aware of
4669 * a few races/collisions:
4670 * 1/ 'Create' called by two separate instances of mdadm
4671 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4672 * devices that have since been assimilated via
4673 * activate_spare.
4674 * In the event this update can not be carried out mdadm will
4675 * (FIX ME) notice that its update did not take hold.
4676 */
4677 struct imsm_update_create_array *u = (void *) update->buf;
4678 struct intel_dev *dv;
4679 struct imsm_dev *dev;
4680 struct imsm_map *map, *new_map;
4681 unsigned long long start, end;
4682 unsigned long long new_start, new_end;
4683 int i;
4684 struct disk_info *inf;
4685 struct dl *dl;
4686
4687 /* handle racing creates: first come first serve */
4688 if (u->dev_idx < mpb->num_raid_devs) {
4689 dprintf("%s: subarray %d already defined\n",
4690 __func__, u->dev_idx);
4691 goto create_error;
4692 }
4693
4694 /* check update is next in sequence */
4695 if (u->dev_idx != mpb->num_raid_devs) {
4696 dprintf("%s: can not create array %d expected index %d\n",
4697 __func__, u->dev_idx, mpb->num_raid_devs);
4698 goto create_error;
4699 }
4700
4701 new_map = get_imsm_map(&u->dev, 0);
4702 new_start = __le32_to_cpu(new_map->pba_of_lba0);
4703 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
4704 inf = get_disk_info(u);
4705
4706 /* handle activate_spare versus create race:
4707 * check to make sure that overlapping arrays do not include
4708 * overalpping disks
4709 */
4710 for (i = 0; i < mpb->num_raid_devs; i++) {
4711 dev = get_imsm_dev(super, i);
4712 map = get_imsm_map(dev, 0);
4713 start = __le32_to_cpu(map->pba_of_lba0);
4714 end = start + __le32_to_cpu(map->blocks_per_member);
4715 if ((new_start >= start && new_start <= end) ||
4716 (start >= new_start && start <= new_end))
4717 /* overlap */;
4718 else
4719 continue;
4720
4721 if (disks_overlap(super, i, u)) {
4722 dprintf("%s: arrays overlap\n", __func__);
4723 goto create_error;
4724 }
4725 }
4726
4727 /* check that prepare update was successful */
4728 if (!update->space) {
4729 dprintf("%s: prepare update failed\n", __func__);
4730 goto create_error;
4731 }
4732
4733 /* check that all disks are still active before committing
4734 * changes. FIXME: could we instead handle this by creating a
4735 * degraded array? That's probably not what the user expects,
4736 * so better to drop this update on the floor.
4737 */
4738 for (i = 0; i < new_map->num_members; i++) {
4739 dl = serial_to_dl(inf[i].serial, super);
4740 if (!dl) {
4741 dprintf("%s: disk disappeared\n", __func__);
4742 goto create_error;
4743 }
4744 }
4745
4746 super->updates_pending++;
4747
4748 /* convert spares to members and fixup ord_tbl */
4749 for (i = 0; i < new_map->num_members; i++) {
4750 dl = serial_to_dl(inf[i].serial, super);
4751 if (dl->index == -1) {
4752 dl->index = mpb->num_disks;
4753 mpb->num_disks++;
4754 dl->disk.status |= CONFIGURED_DISK;
4755 dl->disk.status &= ~SPARE_DISK;
4756 }
4757 set_imsm_ord_tbl_ent(new_map, i, dl->index);
4758 }
4759
4760 dv = update->space;
4761 dev = dv->dev;
4762 update->space = NULL;
4763 imsm_copy_dev(dev, &u->dev);
4764 dv->index = u->dev_idx;
4765 dv->next = super->devlist;
4766 super->devlist = dv;
4767 mpb->num_raid_devs++;
4768
4769 imsm_update_version_info(super);
4770 break;
4771 create_error:
4772 /* mdmon knows how to release update->space, but not
4773 * ((struct intel_dev *) update->space)->dev
4774 */
4775 if (update->space) {
4776 dv = update->space;
4777 free(dv->dev);
4778 }
4779 break;
4780 }
4781 case update_add_disk:
4782
4783 /* we may be able to repair some arrays if disks are
4784 * being added */
4785 if (super->add) {
4786 struct active_array *a;
4787
4788 super->updates_pending++;
4789 for (a = st->arrays; a; a = a->next)
4790 a->check_degraded = 1;
4791 }
4792 /* add some spares to the metadata */
4793 while (super->add) {
4794 struct dl *al;
4795
4796 al = super->add;
4797 super->add = al->next;
4798 al->next = super->disks;
4799 super->disks = al;
4800 dprintf("%s: added %x:%x\n",
4801 __func__, al->major, al->minor);
4802 }
4803
4804 break;
4805 }
4806 }
4807
4808 static void imsm_prepare_update(struct supertype *st,
4809 struct metadata_update *update)
4810 {
4811 /**
4812 * Allocate space to hold new disk entries, raid-device entries or a new
4813 * mpb if necessary. The manager synchronously waits for updates to
4814 * complete in the monitor, so new mpb buffers allocated here can be
4815 * integrated by the monitor thread without worrying about live pointers
4816 * in the manager thread.
4817 */
4818 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4819 struct intel_super *super = st->sb;
4820 struct imsm_super *mpb = super->anchor;
4821 size_t buf_len;
4822 size_t len = 0;
4823
4824 switch (type) {
4825 case update_create_array: {
4826 struct imsm_update_create_array *u = (void *) update->buf;
4827 struct intel_dev *dv;
4828 struct imsm_dev *dev = &u->dev;
4829 struct imsm_map *map = get_imsm_map(dev, 0);
4830 struct dl *dl;
4831 struct disk_info *inf;
4832 int i;
4833 int activate = 0;
4834
4835 inf = get_disk_info(u);
4836 len = sizeof_imsm_dev(dev, 1);
4837 /* allocate a new super->devlist entry */
4838 dv = malloc(sizeof(*dv));
4839 if (dv) {
4840 dv->dev = malloc(len);
4841 if (dv->dev)
4842 update->space = dv;
4843 else {
4844 free(dv);
4845 update->space = NULL;
4846 }
4847 }
4848
4849 /* count how many spares will be converted to members */
4850 for (i = 0; i < map->num_members; i++) {
4851 dl = serial_to_dl(inf[i].serial, super);
4852 if (!dl) {
4853 /* hmm maybe it failed?, nothing we can do about
4854 * it here
4855 */
4856 continue;
4857 }
4858 if (count_memberships(dl, super) == 0)
4859 activate++;
4860 }
4861 len += activate * sizeof(struct imsm_disk);
4862 break;
4863 default:
4864 break;
4865 }
4866 }
4867
4868 /* check if we need a larger metadata buffer */
4869 if (super->next_buf)
4870 buf_len = super->next_len;
4871 else
4872 buf_len = super->len;
4873
4874 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4875 /* ok we need a larger buf than what is currently allocated
4876 * if this allocation fails process_update will notice that
4877 * ->next_len is set and ->next_buf is NULL
4878 */
4879 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4880 if (super->next_buf)
4881 free(super->next_buf);
4882
4883 super->next_len = buf_len;
4884 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
4885 memset(super->next_buf, 0, buf_len);
4886 else
4887 super->next_buf = NULL;
4888 }
4889 }
4890
4891 /* must be called while manager is quiesced */
4892 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4893 {
4894 struct imsm_super *mpb = super->anchor;
4895 struct dl *iter;
4896 struct imsm_dev *dev;
4897 struct imsm_map *map;
4898 int i, j, num_members;
4899 __u32 ord;
4900
4901 dprintf("%s: deleting device[%d] from imsm_super\n",
4902 __func__, index);
4903
4904 /* shift all indexes down one */
4905 for (iter = super->disks; iter; iter = iter->next)
4906 if (iter->index > index)
4907 iter->index--;
4908 for (iter = super->missing; iter; iter = iter->next)
4909 if (iter->index > index)
4910 iter->index--;
4911
4912 for (i = 0; i < mpb->num_raid_devs; i++) {
4913 dev = get_imsm_dev(super, i);
4914 map = get_imsm_map(dev, 0);
4915 num_members = map->num_members;
4916 for (j = 0; j < num_members; j++) {
4917 /* update ord entries being careful not to propagate
4918 * ord-flags to the first map
4919 */
4920 ord = get_imsm_ord_tbl_ent(dev, j);
4921
4922 if (ord_to_idx(ord) <= index)
4923 continue;
4924
4925 map = get_imsm_map(dev, 0);
4926 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4927 map = get_imsm_map(dev, 1);
4928 if (map)
4929 set_imsm_ord_tbl_ent(map, j, ord - 1);
4930 }
4931 }
4932
4933 mpb->num_disks--;
4934 super->updates_pending++;
4935 if (*dlp) {
4936 struct dl *dl = *dlp;
4937
4938 *dlp = (*dlp)->next;
4939 __free_imsm_disk(dl);
4940 }
4941 }
4942 #endif /* MDASSEMBLE */
4943
4944 struct superswitch super_imsm = {
4945 #ifndef MDASSEMBLE
4946 .examine_super = examine_super_imsm,
4947 .brief_examine_super = brief_examine_super_imsm,
4948 .brief_examine_subarrays = brief_examine_subarrays_imsm,
4949 .export_examine_super = export_examine_super_imsm,
4950 .detail_super = detail_super_imsm,
4951 .brief_detail_super = brief_detail_super_imsm,
4952 .write_init_super = write_init_super_imsm,
4953 .validate_geometry = validate_geometry_imsm,
4954 .add_to_super = add_to_super_imsm,
4955 .detail_platform = detail_platform_imsm,
4956 #endif
4957 .match_home = match_home_imsm,
4958 .uuid_from_super= uuid_from_super_imsm,
4959 .getinfo_super = getinfo_super_imsm,
4960 .update_super = update_super_imsm,
4961
4962 .avail_size = avail_size_imsm,
4963
4964 .compare_super = compare_super_imsm,
4965
4966 .load_super = load_super_imsm,
4967 .init_super = init_super_imsm,
4968 .store_super = store_super_imsm,
4969 .free_super = free_super_imsm,
4970 .match_metadata_desc = match_metadata_desc_imsm,
4971 .container_content = container_content_imsm,
4972 .default_layout = imsm_level_to_layout,
4973
4974 .external = 1,
4975 .name = "imsm",
4976
4977 #ifndef MDASSEMBLE
4978 /* for mdmon */
4979 .open_new = imsm_open_new,
4980 .load_super = load_super_imsm,
4981 .set_array_state= imsm_set_array_state,
4982 .set_disk = imsm_set_disk,
4983 .sync_metadata = imsm_sync_metadata,
4984 .activate_spare = imsm_activate_spare,
4985 .process_update = imsm_process_update,
4986 .prepare_update = imsm_prepare_update,
4987 #endif /* MDASSEMBLE */
4988 };
Unnamed repository; edit this file 'description' to name the repository.