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