]> 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: fix activate spare to ignore foreign disks
[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, int enumerate_only)
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 if (enumerate_only) {
941 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
942 return 0;
943 return 2;
944 }
945
946 list = find_driver_devices("pci", "ahci");
947 for (hba = list; hba; hba = hba->next)
948 if (devpath_to_vendor(hba->path) == 0x8086)
949 break;
950
951 if (!hba) {
952 if (verbose)
953 fprintf(stderr, Name ": unable to find active ahci controller\n");
954 free_sys_dev(&list);
955 return 2;
956 } else if (verbose)
957 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
958 hba_path = hba->path;
959 hba->path = NULL;
960 free_sys_dev(&list);
961
962 orom = find_imsm_orom();
963 if (!orom) {
964 if (verbose)
965 fprintf(stderr, Name ": imsm option-rom not found\n");
966 return 2;
967 }
968
969 printf(" Platform : Intel(R) Matrix Storage Manager\n");
970 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
971 orom->hotfix_ver, orom->build);
972 printf(" RAID Levels :%s%s%s%s%s\n",
973 imsm_orom_has_raid0(orom) ? " raid0" : "",
974 imsm_orom_has_raid1(orom) ? " raid1" : "",
975 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
976 imsm_orom_has_raid10(orom) ? " raid10" : "",
977 imsm_orom_has_raid5(orom) ? " raid5" : "");
978 printf(" Max Disks : %d\n", orom->tds);
979 printf(" Max Volumes : %d\n", orom->vpa);
980 printf(" I/O Controller : %s\n", hba_path);
981
982 /* find the smallest scsi host number to determine a port number base */
983 dir = opendir(hba_path);
984 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
985 int host;
986
987 if (sscanf(ent->d_name, "host%d", &host) != 1)
988 continue;
989 if (port_count == 0)
990 host_base = host;
991 else if (host < host_base)
992 host_base = host;
993
994 if (host + 1 > port_count + host_base)
995 port_count = host + 1 - host_base;
996
997 }
998 if (dir)
999 closedir(dir);
1000
1001 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1002 host_base, verbose) != 0) {
1003 if (verbose)
1004 fprintf(stderr, Name ": failed to enumerate ports\n");
1005 return 2;
1006 }
1007
1008 return 0;
1009 }
1010 #endif
1011
1012 static int match_home_imsm(struct supertype *st, char *homehost)
1013 {
1014 /* the imsm metadata format does not specify any host
1015 * identification information. We return -1 since we can never
1016 * confirm nor deny whether a given array is "meant" for this
1017 * host. We rely on compare_super and the 'family_num' field to
1018 * exclude member disks that do not belong, and we rely on
1019 * mdadm.conf to specify the arrays that should be assembled.
1020 * Auto-assembly may still pick up "foreign" arrays.
1021 */
1022
1023 return -1;
1024 }
1025
1026 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1027 {
1028 /* The uuid returned here is used for:
1029 * uuid to put into bitmap file (Create, Grow)
1030 * uuid for backup header when saving critical section (Grow)
1031 * comparing uuids when re-adding a device into an array
1032 * In these cases the uuid required is that of the data-array,
1033 * not the device-set.
1034 * uuid to recognise same set when adding a missing device back
1035 * to an array. This is a uuid for the device-set.
1036 *
1037 * For each of these we can make do with a truncated
1038 * or hashed uuid rather than the original, as long as
1039 * everyone agrees.
1040 * In each case the uuid required is that of the data-array,
1041 * not the device-set.
1042 */
1043 /* imsm does not track uuid's so we synthesis one using sha1 on
1044 * - The signature (Which is constant for all imsm array, but no matter)
1045 * - the family_num of the container
1046 * - the index number of the volume
1047 * - the 'serial' number of the volume.
1048 * Hopefully these are all constant.
1049 */
1050 struct intel_super *super = st->sb;
1051
1052 char buf[20];
1053 struct sha1_ctx ctx;
1054 struct imsm_dev *dev = NULL;
1055
1056 sha1_init_ctx(&ctx);
1057 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1058 sha1_process_bytes(&super->anchor->family_num, sizeof(__u32), &ctx);
1059 if (super->current_vol >= 0)
1060 dev = get_imsm_dev(super, super->current_vol);
1061 if (dev) {
1062 __u32 vol = super->current_vol;
1063 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1064 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1065 }
1066 sha1_finish_ctx(&ctx, buf);
1067 memcpy(uuid, buf, 4*4);
1068 }
1069
1070 #if 0
1071 static void
1072 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1073 {
1074 __u8 *v = get_imsm_version(mpb);
1075 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1076 char major[] = { 0, 0, 0 };
1077 char minor[] = { 0 ,0, 0 };
1078 char patch[] = { 0, 0, 0 };
1079 char *ver_parse[] = { major, minor, patch };
1080 int i, j;
1081
1082 i = j = 0;
1083 while (*v != '\0' && v < end) {
1084 if (*v != '.' && j < 2)
1085 ver_parse[i][j++] = *v;
1086 else {
1087 i++;
1088 j = 0;
1089 }
1090 v++;
1091 }
1092
1093 *m = strtol(minor, NULL, 0);
1094 *p = strtol(patch, NULL, 0);
1095 }
1096 #endif
1097
1098 static int imsm_level_to_layout(int level)
1099 {
1100 switch (level) {
1101 case 0:
1102 case 1:
1103 return 0;
1104 case 5:
1105 case 6:
1106 return ALGORITHM_LEFT_ASYMMETRIC;
1107 case 10:
1108 return 0x102;
1109 }
1110 return UnSet;
1111 }
1112
1113 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
1114 {
1115 struct intel_super *super = st->sb;
1116 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1117 struct imsm_map *map = get_imsm_map(dev, 0);
1118
1119 info->container_member = super->current_vol;
1120 info->array.raid_disks = map->num_members;
1121 info->array.level = get_imsm_raid_level(map);
1122 info->array.layout = imsm_level_to_layout(info->array.level);
1123 info->array.md_minor = -1;
1124 info->array.ctime = 0;
1125 info->array.utime = 0;
1126 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1127 info->array.state = !dev->vol.dirty;
1128
1129 info->disk.major = 0;
1130 info->disk.minor = 0;
1131
1132 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1133 info->component_size = __le32_to_cpu(map->blocks_per_member);
1134 memset(info->uuid, 0, sizeof(info->uuid));
1135
1136 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
1137 info->resync_start = 0;
1138 else if (dev->vol.migr_state)
1139 info->resync_start = __le32_to_cpu(dev->vol.curr_migr_unit);
1140 else
1141 info->resync_start = ~0ULL;
1142
1143 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1144 info->name[MAX_RAID_SERIAL_LEN] = 0;
1145
1146 info->array.major_version = -1;
1147 info->array.minor_version = -2;
1148 sprintf(info->text_version, "/%s/%d",
1149 devnum2devname(st->container_dev),
1150 info->container_member);
1151 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1152 uuid_from_super_imsm(st, info->uuid);
1153 }
1154
1155 /* check the config file to see if we can return a real uuid for this spare */
1156 static void fixup_container_spare_uuid(struct mdinfo *inf)
1157 {
1158 struct mddev_ident_s *array_list;
1159
1160 if (inf->array.level != LEVEL_CONTAINER ||
1161 memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0)
1162 return;
1163
1164 array_list = conf_get_ident(NULL);
1165
1166 for (; array_list; array_list = array_list->next) {
1167 if (array_list->uuid_set) {
1168 struct supertype *_sst; /* spare supertype */
1169 struct supertype *_cst; /* container supertype */
1170
1171 _cst = array_list->st;
1172 _sst = _cst->ss->match_metadata_desc(inf->text_version);
1173 if (_sst) {
1174 memcpy(inf->uuid, array_list->uuid, sizeof(int[4]));
1175 free(_sst);
1176 break;
1177 }
1178 }
1179 }
1180 }
1181
1182 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
1183 {
1184 struct intel_super *super = st->sb;
1185 struct imsm_disk *disk;
1186 __u32 s;
1187
1188 if (super->current_vol >= 0) {
1189 getinfo_super_imsm_volume(st, info);
1190 return;
1191 }
1192
1193 /* Set raid_disks to zero so that Assemble will always pull in valid
1194 * spares
1195 */
1196 info->array.raid_disks = 0;
1197 info->array.level = LEVEL_CONTAINER;
1198 info->array.layout = 0;
1199 info->array.md_minor = -1;
1200 info->array.ctime = 0; /* N/A for imsm */
1201 info->array.utime = 0;
1202 info->array.chunk_size = 0;
1203
1204 info->disk.major = 0;
1205 info->disk.minor = 0;
1206 info->disk.raid_disk = -1;
1207 info->reshape_active = 0;
1208 info->array.major_version = -1;
1209 info->array.minor_version = -2;
1210 strcpy(info->text_version, "imsm");
1211 info->safe_mode_delay = 0;
1212 info->disk.number = -1;
1213 info->disk.state = 0;
1214 info->name[0] = 0;
1215
1216 if (super->disks) {
1217 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1218
1219 disk = &super->disks->disk;
1220 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1221 info->component_size = reserved;
1222 s = disk->status;
1223 info->disk.state = s & CONFIGURED_DISK ? (1 << MD_DISK_ACTIVE) : 0;
1224 /* we don't change info->disk.raid_disk here because
1225 * this state will be finalized in mdmon after we have
1226 * found the 'most fresh' version of the metadata
1227 */
1228 info->disk.state |= s & FAILED_DISK ? (1 << MD_DISK_FAULTY) : 0;
1229 info->disk.state |= s & SPARE_DISK ? 0 : (1 << MD_DISK_SYNC);
1230 }
1231
1232 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1233 * ->compare_super may have updated the 'num_raid_devs' field for spares
1234 */
1235 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1236 uuid_from_super_imsm(st, info->uuid);
1237 else {
1238 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1239 fixup_container_spare_uuid(info);
1240 }
1241 }
1242
1243 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1244 char *update, char *devname, int verbose,
1245 int uuid_set, char *homehost)
1246 {
1247 /* FIXME */
1248
1249 /* For 'assemble' and 'force' we need to return non-zero if any
1250 * change was made. For others, the return value is ignored.
1251 * Update options are:
1252 * force-one : This device looks a bit old but needs to be included,
1253 * update age info appropriately.
1254 * assemble: clear any 'faulty' flag to allow this device to
1255 * be assembled.
1256 * force-array: Array is degraded but being forced, mark it clean
1257 * if that will be needed to assemble it.
1258 *
1259 * newdev: not used ????
1260 * grow: Array has gained a new device - this is currently for
1261 * linear only
1262 * resync: mark as dirty so a resync will happen.
1263 * name: update the name - preserving the homehost
1264 *
1265 * Following are not relevant for this imsm:
1266 * sparc2.2 : update from old dodgey metadata
1267 * super-minor: change the preferred_minor number
1268 * summaries: update redundant counters.
1269 * uuid: Change the uuid of the array to match watch is given
1270 * homehost: update the recorded homehost
1271 * _reshape_progress: record new reshape_progress position.
1272 */
1273 int rv = 0;
1274 //struct intel_super *super = st->sb;
1275 //struct imsm_super *mpb = super->mpb;
1276
1277 if (strcmp(update, "grow") == 0) {
1278 }
1279 if (strcmp(update, "resync") == 0) {
1280 /* dev->vol.dirty = 1; */
1281 }
1282
1283 /* IMSM has no concept of UUID or homehost */
1284
1285 return rv;
1286 }
1287
1288 static size_t disks_to_mpb_size(int disks)
1289 {
1290 size_t size;
1291
1292 size = sizeof(struct imsm_super);
1293 size += (disks - 1) * sizeof(struct imsm_disk);
1294 size += 2 * sizeof(struct imsm_dev);
1295 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1296 size += (4 - 2) * sizeof(struct imsm_map);
1297 /* 4 possible disk_ord_tbl's */
1298 size += 4 * (disks - 1) * sizeof(__u32);
1299
1300 return size;
1301 }
1302
1303 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1304 {
1305 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1306 return 0;
1307
1308 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1309 }
1310
1311 static void free_devlist(struct intel_super *super)
1312 {
1313 struct intel_dev *dv;
1314
1315 while (super->devlist) {
1316 dv = super->devlist->next;
1317 free(super->devlist->dev);
1318 free(super->devlist);
1319 super->devlist = dv;
1320 }
1321 }
1322
1323 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1324 {
1325 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1326 }
1327
1328 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1329 {
1330 /*
1331 * return:
1332 * 0 same, or first was empty, and second was copied
1333 * 1 second had wrong number
1334 * 2 wrong uuid
1335 * 3 wrong other info
1336 */
1337 struct intel_super *first = st->sb;
1338 struct intel_super *sec = tst->sb;
1339
1340 if (!first) {
1341 st->sb = tst->sb;
1342 tst->sb = NULL;
1343 return 0;
1344 }
1345
1346 if (memcmp(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH) != 0)
1347 return 3;
1348
1349 /* if an anchor does not have num_raid_devs set then it is a free
1350 * floating spare
1351 */
1352 if (first->anchor->num_raid_devs > 0 &&
1353 sec->anchor->num_raid_devs > 0) {
1354 if (first->anchor->family_num != sec->anchor->family_num)
1355 return 3;
1356 }
1357
1358 /* if 'first' is a spare promote it to a populated mpb with sec's
1359 * family number
1360 */
1361 if (first->anchor->num_raid_devs == 0 &&
1362 sec->anchor->num_raid_devs > 0) {
1363 int i;
1364 struct intel_dev *dv;
1365 struct imsm_dev *dev;
1366
1367 /* we need to copy raid device info from sec if an allocation
1368 * fails here we don't associate the spare
1369 */
1370 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1371 dv = malloc(sizeof(*dv));
1372 if (!dv)
1373 break;
1374 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1375 if (!dev) {
1376 free(dv);
1377 break;
1378 }
1379 dv->dev = dev;
1380 dv->index = i;
1381 dv->next = first->devlist;
1382 first->devlist = dv;
1383 }
1384 if (i <= sec->anchor->num_raid_devs) {
1385 /* allocation failure */
1386 free_devlist(first);
1387 fprintf(stderr, "imsm: failed to associate spare\n");
1388 return 3;
1389 }
1390 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1391 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1392
1393 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1394 first->anchor->family_num = sec->anchor->family_num;
1395 }
1396
1397 return 0;
1398 }
1399
1400 static void fd2devname(int fd, char *name)
1401 {
1402 struct stat st;
1403 char path[256];
1404 char dname[100];
1405 char *nm;
1406 int rv;
1407
1408 name[0] = '\0';
1409 if (fstat(fd, &st) != 0)
1410 return;
1411 sprintf(path, "/sys/dev/block/%d:%d",
1412 major(st.st_rdev), minor(st.st_rdev));
1413
1414 rv = readlink(path, dname, sizeof(dname));
1415 if (rv <= 0)
1416 return;
1417
1418 dname[rv] = '\0';
1419 nm = strrchr(dname, '/');
1420 nm++;
1421 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1422 }
1423
1424
1425 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1426
1427 static int imsm_read_serial(int fd, char *devname,
1428 __u8 serial[MAX_RAID_SERIAL_LEN])
1429 {
1430 unsigned char scsi_serial[255];
1431 int rv;
1432 int rsp_len;
1433 int len;
1434 char *c, *rsp_buf;
1435
1436 memset(scsi_serial, 0, sizeof(scsi_serial));
1437
1438 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1439
1440 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1441 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1442 fd2devname(fd, (char *) serial);
1443 return 0;
1444 }
1445
1446 if (rv != 0) {
1447 if (devname)
1448 fprintf(stderr,
1449 Name ": Failed to retrieve serial for %s\n",
1450 devname);
1451 return rv;
1452 }
1453
1454 /* trim leading whitespace */
1455 rsp_len = scsi_serial[3];
1456 if (!rsp_len) {
1457 if (devname)
1458 fprintf(stderr,
1459 Name ": Failed to retrieve serial for %s\n",
1460 devname);
1461 return 2;
1462 }
1463 rsp_buf = (char *) &scsi_serial[4];
1464 c = rsp_buf;
1465 while (isspace(*c))
1466 c++;
1467
1468 /* truncate len to the end of rsp_buf if necessary */
1469 if (c + MAX_RAID_SERIAL_LEN > rsp_buf + rsp_len)
1470 len = rsp_len - (c - rsp_buf);
1471 else
1472 len = MAX_RAID_SERIAL_LEN;
1473
1474 /* initialize the buffer and copy rsp_buf characters */
1475 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1476 memcpy(serial, c, len);
1477
1478 /* trim trailing whitespace starting with the last character copied */
1479 c = (char *) &serial[len - 1];
1480 while (isspace(*c) || *c == '\0')
1481 *c-- = '\0';
1482
1483 return 0;
1484 }
1485
1486 static int serialcmp(__u8 *s1, __u8 *s2)
1487 {
1488 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1489 }
1490
1491 static void serialcpy(__u8 *dest, __u8 *src)
1492 {
1493 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1494 }
1495
1496 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1497 {
1498 struct dl *dl;
1499
1500 for (dl = super->disks; dl; dl = dl->next)
1501 if (serialcmp(dl->serial, serial) == 0)
1502 break;
1503
1504 return dl;
1505 }
1506
1507 static int
1508 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1509 {
1510 struct dl *dl;
1511 struct stat stb;
1512 int rv;
1513 int i;
1514 int alloc = 1;
1515 __u8 serial[MAX_RAID_SERIAL_LEN];
1516
1517 rv = imsm_read_serial(fd, devname, serial);
1518
1519 if (rv != 0)
1520 return 2;
1521
1522 /* check if this is a disk we have seen before. it may be a spare in
1523 * super->disks while the current anchor believes it is a raid member,
1524 * check if we need to update dl->index
1525 */
1526 dl = serial_to_dl(serial, super);
1527 if (!dl)
1528 dl = malloc(sizeof(*dl));
1529 else
1530 alloc = 0;
1531
1532 if (!dl) {
1533 if (devname)
1534 fprintf(stderr,
1535 Name ": failed to allocate disk buffer for %s\n",
1536 devname);
1537 return 2;
1538 }
1539
1540 if (alloc) {
1541 fstat(fd, &stb);
1542 dl->major = major(stb.st_rdev);
1543 dl->minor = minor(stb.st_rdev);
1544 dl->next = super->disks;
1545 dl->fd = keep_fd ? fd : -1;
1546 dl->devname = devname ? strdup(devname) : NULL;
1547 serialcpy(dl->serial, serial);
1548 dl->index = -2;
1549 dl->e = NULL;
1550 } else if (keep_fd) {
1551 close(dl->fd);
1552 dl->fd = fd;
1553 }
1554
1555 /* look up this disk's index in the current anchor */
1556 for (i = 0; i < super->anchor->num_disks; i++) {
1557 struct imsm_disk *disk_iter;
1558
1559 disk_iter = __get_imsm_disk(super->anchor, i);
1560
1561 if (serialcmp(disk_iter->serial, dl->serial) == 0) {
1562 dl->disk = *disk_iter;
1563 /* only set index on disks that are a member of a
1564 * populated contianer, i.e. one with raid_devs
1565 */
1566 if (dl->disk.status & FAILED_DISK)
1567 dl->index = -2;
1568 else if (dl->disk.status & SPARE_DISK)
1569 dl->index = -1;
1570 else
1571 dl->index = i;
1572
1573 break;
1574 }
1575 }
1576
1577 /* no match, maybe a stale failed drive */
1578 if (i == super->anchor->num_disks && dl->index >= 0) {
1579 dl->disk = *__get_imsm_disk(super->anchor, dl->index);
1580 if (dl->disk.status & FAILED_DISK)
1581 dl->index = -2;
1582 }
1583
1584 if (alloc)
1585 super->disks = dl;
1586
1587 return 0;
1588 }
1589
1590 #ifndef MDASSEMBLE
1591 /* When migrating map0 contains the 'destination' state while map1
1592 * contains the current state. When not migrating map0 contains the
1593 * current state. This routine assumes that map[0].map_state is set to
1594 * the current array state before being called.
1595 *
1596 * Migration is indicated by one of the following states
1597 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1598 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1599 * map1state=unitialized)
1600 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1601 * map1state=normal)
1602 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1603 * map1state=degraded)
1604 */
1605 static void migrate(struct imsm_dev *dev, __u8 to_state, int rebuild_resync)
1606 {
1607 struct imsm_map *dest;
1608 struct imsm_map *src = get_imsm_map(dev, 0);
1609
1610 dev->vol.migr_state = 1;
1611 dev->vol.migr_type = rebuild_resync;
1612 dev->vol.curr_migr_unit = 0;
1613 dest = get_imsm_map(dev, 1);
1614
1615 memcpy(dest, src, sizeof_imsm_map(src));
1616 src->map_state = to_state;
1617 }
1618
1619 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1620 {
1621 struct imsm_map *map = get_imsm_map(dev, 0);
1622
1623 dev->vol.migr_state = 0;
1624 dev->vol.curr_migr_unit = 0;
1625 map->map_state = map_state;
1626 }
1627 #endif
1628
1629 static int parse_raid_devices(struct intel_super *super)
1630 {
1631 int i;
1632 struct imsm_dev *dev_new;
1633 size_t len, len_migr;
1634 size_t space_needed = 0;
1635 struct imsm_super *mpb = super->anchor;
1636
1637 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1638 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1639 struct intel_dev *dv;
1640
1641 len = sizeof_imsm_dev(dev_iter, 0);
1642 len_migr = sizeof_imsm_dev(dev_iter, 1);
1643 if (len_migr > len)
1644 space_needed += len_migr - len;
1645
1646 dv = malloc(sizeof(*dv));
1647 if (!dv)
1648 return 1;
1649 dev_new = malloc(len_migr);
1650 if (!dev_new) {
1651 free(dv);
1652 return 1;
1653 }
1654 imsm_copy_dev(dev_new, dev_iter);
1655 dv->dev = dev_new;
1656 dv->index = i;
1657 dv->next = super->devlist;
1658 super->devlist = dv;
1659 }
1660
1661 /* ensure that super->buf is large enough when all raid devices
1662 * are migrating
1663 */
1664 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1665 void *buf;
1666
1667 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1668 if (posix_memalign(&buf, 512, len) != 0)
1669 return 1;
1670
1671 memcpy(buf, super->buf, len);
1672 free(super->buf);
1673 super->buf = buf;
1674 super->len = len;
1675 }
1676
1677 return 0;
1678 }
1679
1680 /* retrieve a pointer to the bbm log which starts after all raid devices */
1681 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1682 {
1683 void *ptr = NULL;
1684
1685 if (__le32_to_cpu(mpb->bbm_log_size)) {
1686 ptr = mpb;
1687 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1688 }
1689
1690 return ptr;
1691 }
1692
1693 static void __free_imsm(struct intel_super *super, int free_disks);
1694
1695 /* load_imsm_mpb - read matrix metadata
1696 * allocates super->mpb to be freed by free_super
1697 */
1698 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1699 {
1700 unsigned long long dsize;
1701 unsigned long long sectors;
1702 struct stat;
1703 struct imsm_super *anchor;
1704 __u32 check_sum;
1705 int rc;
1706
1707 get_dev_size(fd, NULL, &dsize);
1708
1709 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1710 if (devname)
1711 fprintf(stderr,
1712 Name ": Cannot seek to anchor block on %s: %s\n",
1713 devname, strerror(errno));
1714 return 1;
1715 }
1716
1717 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1718 if (devname)
1719 fprintf(stderr,
1720 Name ": Failed to allocate imsm anchor buffer"
1721 " on %s\n", devname);
1722 return 1;
1723 }
1724 if (read(fd, anchor, 512) != 512) {
1725 if (devname)
1726 fprintf(stderr,
1727 Name ": Cannot read anchor block on %s: %s\n",
1728 devname, strerror(errno));
1729 free(anchor);
1730 return 1;
1731 }
1732
1733 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1734 if (devname)
1735 fprintf(stderr,
1736 Name ": no IMSM anchor on %s\n", devname);
1737 free(anchor);
1738 return 2;
1739 }
1740
1741 __free_imsm(super, 0);
1742 super->len = ROUND_UP(anchor->mpb_size, 512);
1743 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1744 if (devname)
1745 fprintf(stderr,
1746 Name ": unable to allocate %zu byte mpb buffer\n",
1747 super->len);
1748 free(anchor);
1749 return 2;
1750 }
1751 memcpy(super->buf, anchor, 512);
1752
1753 sectors = mpb_sectors(anchor) - 1;
1754 free(anchor);
1755 if (!sectors) {
1756 rc = load_imsm_disk(fd, super, devname, 0);
1757 if (rc == 0)
1758 rc = parse_raid_devices(super);
1759 return rc;
1760 }
1761
1762 /* read the extended mpb */
1763 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1764 if (devname)
1765 fprintf(stderr,
1766 Name ": Cannot seek to extended mpb on %s: %s\n",
1767 devname, strerror(errno));
1768 return 1;
1769 }
1770
1771 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1772 if (devname)
1773 fprintf(stderr,
1774 Name ": Cannot read extended mpb on %s: %s\n",
1775 devname, strerror(errno));
1776 return 2;
1777 }
1778
1779 check_sum = __gen_imsm_checksum(super->anchor);
1780 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1781 if (devname)
1782 fprintf(stderr,
1783 Name ": IMSM checksum %x != %x on %s\n",
1784 check_sum, __le32_to_cpu(super->anchor->check_sum),
1785 devname);
1786 return 2;
1787 }
1788
1789 /* FIXME the BBM log is disk specific so we cannot use this global
1790 * buffer for all disks. Ok for now since we only look at the global
1791 * bbm_log_size parameter to gate assembly
1792 */
1793 super->bbm_log = __get_imsm_bbm_log(super->anchor);
1794
1795 rc = load_imsm_disk(fd, super, devname, 0);
1796 if (rc == 0)
1797 rc = parse_raid_devices(super);
1798
1799 return rc;
1800 }
1801
1802 static void __free_imsm_disk(struct dl *d)
1803 {
1804 if (d->fd >= 0)
1805 close(d->fd);
1806 if (d->devname)
1807 free(d->devname);
1808 if (d->e)
1809 free(d->e);
1810 free(d);
1811
1812 }
1813 static void free_imsm_disks(struct intel_super *super)
1814 {
1815 struct dl *d;
1816
1817 while (super->disks) {
1818 d = super->disks;
1819 super->disks = d->next;
1820 __free_imsm_disk(d);
1821 }
1822 while (super->missing) {
1823 d = super->missing;
1824 super->missing = d->next;
1825 __free_imsm_disk(d);
1826 }
1827
1828 }
1829
1830 /* free all the pieces hanging off of a super pointer */
1831 static void __free_imsm(struct intel_super *super, int free_disks)
1832 {
1833 if (super->buf) {
1834 free(super->buf);
1835 super->buf = NULL;
1836 }
1837 if (free_disks)
1838 free_imsm_disks(super);
1839 free_devlist(super);
1840 if (super->hba) {
1841 free((void *) super->hba);
1842 super->hba = NULL;
1843 }
1844 }
1845
1846 static void free_imsm(struct intel_super *super)
1847 {
1848 __free_imsm(super, 1);
1849 free(super);
1850 }
1851
1852 static void free_super_imsm(struct supertype *st)
1853 {
1854 struct intel_super *super = st->sb;
1855
1856 if (!super)
1857 return;
1858
1859 free_imsm(super);
1860 st->sb = NULL;
1861 }
1862
1863 static struct intel_super *alloc_super(int creating_imsm)
1864 {
1865 struct intel_super *super = malloc(sizeof(*super));
1866
1867 if (super) {
1868 memset(super, 0, sizeof(*super));
1869 super->creating_imsm = creating_imsm;
1870 super->current_vol = -1;
1871 super->create_offset = ~((__u32 ) 0);
1872 if (!check_env("IMSM_NO_PLATFORM"))
1873 super->orom = find_imsm_orom();
1874 if (super->orom && !check_env("IMSM_TEST_OROM")) {
1875 struct sys_dev *list, *ent;
1876
1877 /* find the first intel ahci controller */
1878 list = find_driver_devices("pci", "ahci");
1879 for (ent = list; ent; ent = ent->next)
1880 if (devpath_to_vendor(ent->path) == 0x8086)
1881 break;
1882 if (ent) {
1883 super->hba = ent->path;
1884 ent->path = NULL;
1885 }
1886 free_sys_dev(&list);
1887 }
1888 }
1889
1890 return super;
1891 }
1892
1893 #ifndef MDASSEMBLE
1894 /* find_missing - helper routine for load_super_imsm_all that identifies
1895 * disks that have disappeared from the system. This routine relies on
1896 * the mpb being uptodate, which it is at load time.
1897 */
1898 static int find_missing(struct intel_super *super)
1899 {
1900 int i;
1901 struct imsm_super *mpb = super->anchor;
1902 struct dl *dl;
1903 struct imsm_disk *disk;
1904
1905 for (i = 0; i < mpb->num_disks; i++) {
1906 disk = __get_imsm_disk(mpb, i);
1907 dl = serial_to_dl(disk->serial, super);
1908 if (dl)
1909 continue;
1910 /* ok we have a 'disk' without a live entry in
1911 * super->disks
1912 */
1913 if (disk->status & FAILED_DISK || !(disk->status & USABLE_DISK))
1914 continue; /* never mind, already marked */
1915
1916 dl = malloc(sizeof(*dl));
1917 if (!dl)
1918 return 1;
1919 dl->major = 0;
1920 dl->minor = 0;
1921 dl->fd = -1;
1922 dl->devname = strdup("missing");
1923 dl->index = i;
1924 serialcpy(dl->serial, disk->serial);
1925 dl->disk = *disk;
1926 dl->e = NULL;
1927 dl->next = super->missing;
1928 super->missing = dl;
1929 }
1930
1931 return 0;
1932 }
1933
1934 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
1935 char *devname, int keep_fd)
1936 {
1937 struct mdinfo *sra;
1938 struct intel_super *super;
1939 struct mdinfo *sd, *best = NULL;
1940 __u32 bestgen = 0;
1941 __u32 gen;
1942 char nm[20];
1943 int dfd;
1944 int rv;
1945
1946 /* check if this disk is a member of an active array */
1947 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
1948 if (!sra)
1949 return 1;
1950
1951 if (sra->array.major_version != -1 ||
1952 sra->array.minor_version != -2 ||
1953 strcmp(sra->text_version, "imsm") != 0)
1954 return 1;
1955
1956 super = alloc_super(0);
1957 if (!super)
1958 return 1;
1959
1960 /* find the most up to date disk in this array, skipping spares */
1961 for (sd = sra->devs; sd; sd = sd->next) {
1962 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
1963 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
1964 if (dfd < 0) {
1965 free_imsm(super);
1966 return 2;
1967 }
1968 rv = load_imsm_mpb(dfd, super, NULL);
1969 if (!keep_fd)
1970 close(dfd);
1971 if (rv == 0) {
1972 if (super->anchor->num_raid_devs == 0)
1973 gen = 0;
1974 else
1975 gen = __le32_to_cpu(super->anchor->generation_num);
1976 if (!best || gen > bestgen) {
1977 bestgen = gen;
1978 best = sd;
1979 }
1980 } else {
1981 free_imsm(super);
1982 return 2;
1983 }
1984 }
1985
1986 if (!best) {
1987 free_imsm(super);
1988 return 1;
1989 }
1990
1991 /* load the most up to date anchor */
1992 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
1993 dfd = dev_open(nm, O_RDONLY);
1994 if (dfd < 0) {
1995 free_imsm(super);
1996 return 1;
1997 }
1998 rv = load_imsm_mpb(dfd, super, NULL);
1999 close(dfd);
2000 if (rv != 0) {
2001 free_imsm(super);
2002 return 2;
2003 }
2004
2005 /* re-parse the disk list with the current anchor */
2006 for (sd = sra->devs ; sd ; sd = sd->next) {
2007 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2008 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2009 if (dfd < 0) {
2010 free_imsm(super);
2011 return 2;
2012 }
2013 load_imsm_disk(dfd, super, NULL, keep_fd);
2014 if (!keep_fd)
2015 close(dfd);
2016 }
2017
2018
2019 if (find_missing(super) != 0) {
2020 free_imsm(super);
2021 return 2;
2022 }
2023
2024 if (st->subarray[0]) {
2025 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2026 super->current_vol = atoi(st->subarray);
2027 else
2028 return 1;
2029 }
2030
2031 *sbp = super;
2032 st->container_dev = fd2devnum(fd);
2033 if (st->ss == NULL) {
2034 st->ss = &super_imsm;
2035 st->minor_version = 0;
2036 st->max_devs = IMSM_MAX_DEVICES;
2037 }
2038 st->loaded_container = 1;
2039
2040 return 0;
2041 }
2042 #endif
2043
2044 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2045 {
2046 struct intel_super *super;
2047 int rv;
2048
2049 #ifndef MDASSEMBLE
2050 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2051 return 0;
2052 #endif
2053 if (st->subarray[0])
2054 return 1; /* FIXME */
2055
2056 super = alloc_super(0);
2057 if (!super) {
2058 fprintf(stderr,
2059 Name ": malloc of %zu failed.\n",
2060 sizeof(*super));
2061 return 1;
2062 }
2063
2064 rv = load_imsm_mpb(fd, super, devname);
2065
2066 if (rv) {
2067 if (devname)
2068 fprintf(stderr,
2069 Name ": Failed to load all information "
2070 "sections on %s\n", devname);
2071 free_imsm(super);
2072 return rv;
2073 }
2074
2075 st->sb = super;
2076 if (st->ss == NULL) {
2077 st->ss = &super_imsm;
2078 st->minor_version = 0;
2079 st->max_devs = IMSM_MAX_DEVICES;
2080 }
2081 st->loaded_container = 0;
2082
2083 return 0;
2084 }
2085
2086 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2087 {
2088 if (info->level == 1)
2089 return 128;
2090 return info->chunk_size >> 9;
2091 }
2092
2093 static __u32 info_to_num_data_stripes(mdu_array_info_t *info)
2094 {
2095 __u32 num_stripes;
2096
2097 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2098 if (info->level == 1)
2099 num_stripes /= 2;
2100
2101 return num_stripes;
2102 }
2103
2104 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2105 {
2106 if (info->level == 1)
2107 return info->size * 2;
2108 else
2109 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2110 }
2111
2112 static void imsm_update_version_info(struct intel_super *super)
2113 {
2114 /* update the version and attributes */
2115 struct imsm_super *mpb = super->anchor;
2116 char *version;
2117 struct imsm_dev *dev;
2118 struct imsm_map *map;
2119 int i;
2120
2121 for (i = 0; i < mpb->num_raid_devs; i++) {
2122 dev = get_imsm_dev(super, i);
2123 map = get_imsm_map(dev, 0);
2124 if (__le32_to_cpu(dev->size_high) > 0)
2125 mpb->attributes |= MPB_ATTRIB_2TB;
2126
2127 /* FIXME detect when an array spans a port multiplier */
2128 #if 0
2129 mpb->attributes |= MPB_ATTRIB_PM;
2130 #endif
2131
2132 if (mpb->num_raid_devs > 1 ||
2133 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2134 version = MPB_VERSION_ATTRIBS;
2135 switch (get_imsm_raid_level(map)) {
2136 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2137 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2138 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2139 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2140 }
2141 } else {
2142 if (map->num_members >= 5)
2143 version = MPB_VERSION_5OR6_DISK_ARRAY;
2144 else if (dev->status == DEV_CLONE_N_GO)
2145 version = MPB_VERSION_CNG;
2146 else if (get_imsm_raid_level(map) == 5)
2147 version = MPB_VERSION_RAID5;
2148 else if (map->num_members >= 3)
2149 version = MPB_VERSION_3OR4_DISK_ARRAY;
2150 else if (get_imsm_raid_level(map) == 1)
2151 version = MPB_VERSION_RAID1;
2152 else
2153 version = MPB_VERSION_RAID0;
2154 }
2155 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2156 }
2157 }
2158
2159 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2160 unsigned long long size, char *name,
2161 char *homehost, int *uuid)
2162 {
2163 /* We are creating a volume inside a pre-existing container.
2164 * so st->sb is already set.
2165 */
2166 struct intel_super *super = st->sb;
2167 struct imsm_super *mpb = super->anchor;
2168 struct intel_dev *dv;
2169 struct imsm_dev *dev;
2170 struct imsm_vol *vol;
2171 struct imsm_map *map;
2172 int idx = mpb->num_raid_devs;
2173 int i;
2174 unsigned long long array_blocks;
2175 size_t size_old, size_new;
2176
2177 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2178 fprintf(stderr, Name": This imsm-container already has the "
2179 "maximum of %d volumes\n", super->orom->vpa);
2180 return 0;
2181 }
2182
2183 /* ensure the mpb is large enough for the new data */
2184 size_old = __le32_to_cpu(mpb->mpb_size);
2185 size_new = disks_to_mpb_size(info->nr_disks);
2186 if (size_new > size_old) {
2187 void *mpb_new;
2188 size_t size_round = ROUND_UP(size_new, 512);
2189
2190 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2191 fprintf(stderr, Name": could not allocate new mpb\n");
2192 return 0;
2193 }
2194 memcpy(mpb_new, mpb, size_old);
2195 free(mpb);
2196 mpb = mpb_new;
2197 super->anchor = mpb_new;
2198 mpb->mpb_size = __cpu_to_le32(size_new);
2199 memset(mpb_new + size_old, 0, size_round - size_old);
2200 }
2201 super->current_vol = idx;
2202 /* when creating the first raid device in this container set num_disks
2203 * to zero, i.e. delete this spare and add raid member devices in
2204 * add_to_super_imsm_volume()
2205 */
2206 if (super->current_vol == 0)
2207 mpb->num_disks = 0;
2208
2209 for (i = 0; i < super->current_vol; i++) {
2210 dev = get_imsm_dev(super, i);
2211 if (strncmp((char *) dev->volume, name,
2212 MAX_RAID_SERIAL_LEN) == 0) {
2213 fprintf(stderr, Name": '%s' is already defined for this container\n",
2214 name);
2215 return 0;
2216 }
2217 }
2218
2219 sprintf(st->subarray, "%d", idx);
2220 dv = malloc(sizeof(*dv));
2221 if (!dv) {
2222 fprintf(stderr, Name ": failed to allocate device list entry\n");
2223 return 0;
2224 }
2225 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2226 if (!dev) {
2227 free(dv);
2228 fprintf(stderr, Name": could not allocate raid device\n");
2229 return 0;
2230 }
2231 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2232 if (info->level == 1)
2233 array_blocks = info_to_blocks_per_member(info);
2234 else
2235 array_blocks = calc_array_size(info->level, info->raid_disks,
2236 info->layout, info->chunk_size,
2237 info->size*2);
2238 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2239 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2240 dev->status = __cpu_to_le32(0);
2241 dev->reserved_blocks = __cpu_to_le32(0);
2242 vol = &dev->vol;
2243 vol->migr_state = 0;
2244 vol->migr_type = MIGR_INIT;
2245 vol->dirty = 0;
2246 vol->curr_migr_unit = 0;
2247 map = get_imsm_map(dev, 0);
2248 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2249 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2250 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2251 map->num_data_stripes = __cpu_to_le32(info_to_num_data_stripes(info));
2252 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2253 IMSM_T_STATE_NORMAL;
2254
2255 if (info->level == 1 && info->raid_disks > 2) {
2256 fprintf(stderr, Name": imsm does not support more than 2 disks"
2257 "in a raid1 volume\n");
2258 return 0;
2259 }
2260 if (info->level == 10) {
2261 map->raid_level = 1;
2262 map->num_domains = info->raid_disks / 2;
2263 } else {
2264 map->raid_level = info->level;
2265 map->num_domains = !!map->raid_level;
2266 }
2267
2268 map->num_members = info->raid_disks;
2269 for (i = 0; i < map->num_members; i++) {
2270 /* initialized in add_to_super */
2271 set_imsm_ord_tbl_ent(map, i, 0);
2272 }
2273 mpb->num_raid_devs++;
2274
2275 dv->dev = dev;
2276 dv->index = super->current_vol;
2277 dv->next = super->devlist;
2278 super->devlist = dv;
2279
2280 imsm_update_version_info(super);
2281
2282 return 1;
2283 }
2284
2285 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2286 unsigned long long size, char *name,
2287 char *homehost, int *uuid)
2288 {
2289 /* This is primarily called by Create when creating a new array.
2290 * We will then get add_to_super called for each component, and then
2291 * write_init_super called to write it out to each device.
2292 * For IMSM, Create can create on fresh devices or on a pre-existing
2293 * array.
2294 * To create on a pre-existing array a different method will be called.
2295 * This one is just for fresh drives.
2296 */
2297 struct intel_super *super;
2298 struct imsm_super *mpb;
2299 size_t mpb_size;
2300 char *version;
2301
2302 if (!info) {
2303 st->sb = NULL;
2304 return 0;
2305 }
2306 if (st->sb)
2307 return init_super_imsm_volume(st, info, size, name, homehost,
2308 uuid);
2309
2310 super = alloc_super(1);
2311 if (!super)
2312 return 0;
2313 mpb_size = disks_to_mpb_size(info->nr_disks);
2314 if (posix_memalign(&super->buf, 512, mpb_size) != 0) {
2315 free(super);
2316 return 0;
2317 }
2318 mpb = super->buf;
2319 memset(mpb, 0, mpb_size);
2320
2321 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2322
2323 version = (char *) mpb->sig;
2324 strcpy(version, MPB_SIGNATURE);
2325 version += strlen(MPB_SIGNATURE);
2326 strcpy(version, MPB_VERSION_RAID0);
2327 mpb->mpb_size = mpb_size;
2328
2329 st->sb = super;
2330 return 1;
2331 }
2332
2333 #ifndef MDASSEMBLE
2334 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2335 int fd, char *devname)
2336 {
2337 struct intel_super *super = st->sb;
2338 struct imsm_super *mpb = super->anchor;
2339 struct dl *dl;
2340 struct imsm_dev *dev;
2341 struct imsm_map *map;
2342
2343 dev = get_imsm_dev(super, super->current_vol);
2344 map = get_imsm_map(dev, 0);
2345
2346 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2347 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2348 devname);
2349 return 1;
2350 }
2351
2352 for (dl = super->disks; dl ; dl = dl->next)
2353 if (dl->major == dk->major &&
2354 dl->minor == dk->minor)
2355 break;
2356
2357 if (!dl) {
2358 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2359 return 1;
2360 }
2361
2362 /* add a pristine spare to the metadata */
2363 if (dl->index < 0) {
2364 dl->index = super->anchor->num_disks;
2365 super->anchor->num_disks++;
2366 }
2367 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2368 dl->disk.status = CONFIGURED_DISK | USABLE_DISK;
2369
2370 /* if we are creating the first raid device update the family number */
2371 if (super->current_vol == 0) {
2372 __u32 sum;
2373 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2374 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2375
2376 *_dev = *dev;
2377 *_disk = dl->disk;
2378 sum = __gen_imsm_checksum(mpb);
2379 mpb->family_num = __cpu_to_le32(sum);
2380 }
2381
2382 return 0;
2383 }
2384
2385 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2386 int fd, char *devname)
2387 {
2388 struct intel_super *super = st->sb;
2389 struct dl *dd;
2390 unsigned long long size;
2391 __u32 id;
2392 int rv;
2393 struct stat stb;
2394
2395 /* if we are on an RAID enabled platform check that the disk is
2396 * attached to the raid controller
2397 */
2398 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2399 fprintf(stderr,
2400 Name ": %s is not attached to the raid controller: %s\n",
2401 devname ? : "disk", super->hba);
2402 return 1;
2403 }
2404
2405 if (super->current_vol >= 0)
2406 return add_to_super_imsm_volume(st, dk, fd, devname);
2407
2408 fstat(fd, &stb);
2409 dd = malloc(sizeof(*dd));
2410 if (!dd) {
2411 fprintf(stderr,
2412 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2413 return 1;
2414 }
2415 memset(dd, 0, sizeof(*dd));
2416 dd->major = major(stb.st_rdev);
2417 dd->minor = minor(stb.st_rdev);
2418 dd->index = -1;
2419 dd->devname = devname ? strdup(devname) : NULL;
2420 dd->fd = fd;
2421 dd->e = NULL;
2422 rv = imsm_read_serial(fd, devname, dd->serial);
2423 if (rv) {
2424 fprintf(stderr,
2425 Name ": failed to retrieve scsi serial, aborting\n");
2426 free(dd);
2427 abort();
2428 }
2429
2430 get_dev_size(fd, NULL, &size);
2431 size /= 512;
2432 serialcpy(dd->disk.serial, dd->serial);
2433 dd->disk.total_blocks = __cpu_to_le32(size);
2434 dd->disk.status = USABLE_DISK | SPARE_DISK;
2435 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
2436 dd->disk.scsi_id = __cpu_to_le32(id);
2437 else
2438 dd->disk.scsi_id = __cpu_to_le32(0);
2439
2440 if (st->update_tail) {
2441 dd->next = super->add;
2442 super->add = dd;
2443 } else {
2444 dd->next = super->disks;
2445 super->disks = dd;
2446 }
2447
2448 return 0;
2449 }
2450
2451 static int store_imsm_mpb(int fd, struct intel_super *super);
2452
2453 /* spare records have their own family number and do not have any defined raid
2454 * devices
2455 */
2456 static int write_super_imsm_spares(struct intel_super *super, int doclose)
2457 {
2458 struct imsm_super mpb_save;
2459 struct imsm_super *mpb = super->anchor;
2460 __u32 sum;
2461 struct dl *d;
2462
2463 mpb_save = *mpb;
2464 mpb->num_raid_devs = 0;
2465 mpb->num_disks = 1;
2466 mpb->mpb_size = sizeof(struct imsm_super);
2467 mpb->generation_num = __cpu_to_le32(1UL);
2468
2469 for (d = super->disks; d; d = d->next) {
2470 if (d->index != -1)
2471 continue;
2472
2473 mpb->disk[0] = d->disk;
2474 sum = __gen_imsm_checksum(mpb);
2475 mpb->family_num = __cpu_to_le32(sum);
2476 sum = __gen_imsm_checksum(mpb);
2477 mpb->check_sum = __cpu_to_le32(sum);
2478
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 *mpb = mpb_save;
2483 return 1;
2484 }
2485 if (doclose) {
2486 close(d->fd);
2487 d->fd = -1;
2488 }
2489 }
2490
2491 *mpb = mpb_save;
2492 return 0;
2493 }
2494
2495 static int write_super_imsm(struct intel_super *super, int doclose)
2496 {
2497 struct imsm_super *mpb = super->anchor;
2498 struct dl *d;
2499 __u32 generation;
2500 __u32 sum;
2501 int spares = 0;
2502 int i;
2503 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
2504
2505 /* 'generation' is incremented everytime the metadata is written */
2506 generation = __le32_to_cpu(mpb->generation_num);
2507 generation++;
2508 mpb->generation_num = __cpu_to_le32(generation);
2509
2510 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
2511 for (d = super->disks; d; d = d->next) {
2512 if (d->index == -1)
2513 spares++;
2514 else
2515 mpb->disk[d->index] = d->disk;
2516 }
2517 for (d = super->missing; d; d = d->next)
2518 mpb->disk[d->index] = d->disk;
2519
2520 for (i = 0; i < mpb->num_raid_devs; i++) {
2521 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
2522
2523 imsm_copy_dev(dev, get_imsm_dev(super, i));
2524 mpb_size += sizeof_imsm_dev(dev, 0);
2525 }
2526 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
2527 mpb->mpb_size = __cpu_to_le32(mpb_size);
2528
2529 /* recalculate checksum */
2530 sum = __gen_imsm_checksum(mpb);
2531 mpb->check_sum = __cpu_to_le32(sum);
2532
2533 /* write the mpb for disks that compose raid devices */
2534 for (d = super->disks; d ; d = d->next) {
2535 if (d->index < 0)
2536 continue;
2537 if (store_imsm_mpb(d->fd, super))
2538 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2539 __func__, d->major, d->minor, strerror(errno));
2540 if (doclose) {
2541 close(d->fd);
2542 d->fd = -1;
2543 }
2544 }
2545
2546 if (spares)
2547 return write_super_imsm_spares(super, doclose);
2548
2549 return 0;
2550 }
2551
2552
2553 static int create_array(struct supertype *st)
2554 {
2555 size_t len;
2556 struct imsm_update_create_array *u;
2557 struct intel_super *super = st->sb;
2558 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
2559 struct imsm_map *map = get_imsm_map(dev, 0);
2560 struct disk_info *inf;
2561 struct imsm_disk *disk;
2562 int i;
2563 int idx;
2564
2565 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
2566 sizeof(*inf) * map->num_members;
2567 u = malloc(len);
2568 if (!u) {
2569 fprintf(stderr, "%s: failed to allocate update buffer\n",
2570 __func__);
2571 return 1;
2572 }
2573
2574 u->type = update_create_array;
2575 u->dev_idx = super->current_vol;
2576 imsm_copy_dev(&u->dev, dev);
2577 inf = get_disk_info(u);
2578 for (i = 0; i < map->num_members; i++) {
2579 idx = get_imsm_disk_idx(dev, i);
2580 disk = get_imsm_disk(super, idx);
2581 serialcpy(inf[i].serial, disk->serial);
2582 }
2583 append_metadata_update(st, u, len);
2584
2585 return 0;
2586 }
2587
2588 static int _add_disk(struct supertype *st)
2589 {
2590 struct intel_super *super = st->sb;
2591 size_t len;
2592 struct imsm_update_add_disk *u;
2593
2594 if (!super->add)
2595 return 0;
2596
2597 len = sizeof(*u);
2598 u = malloc(len);
2599 if (!u) {
2600 fprintf(stderr, "%s: failed to allocate update buffer\n",
2601 __func__);
2602 return 1;
2603 }
2604
2605 u->type = update_add_disk;
2606 append_metadata_update(st, u, len);
2607
2608 return 0;
2609 }
2610
2611 static int write_init_super_imsm(struct supertype *st)
2612 {
2613 if (st->update_tail) {
2614 /* queue the recently created array / added disk
2615 * as a metadata update */
2616 struct intel_super *super = st->sb;
2617 struct dl *d;
2618 int rv;
2619
2620 /* determine if we are creating a volume or adding a disk */
2621 if (super->current_vol < 0) {
2622 /* in the add disk case we are running in mdmon
2623 * context, so don't close fd's
2624 */
2625 return _add_disk(st);
2626 } else
2627 rv = create_array(st);
2628
2629 for (d = super->disks; d ; d = d->next) {
2630 close(d->fd);
2631 d->fd = -1;
2632 }
2633
2634 return rv;
2635 } else
2636 return write_super_imsm(st->sb, 1);
2637 }
2638 #endif
2639
2640 static int store_zero_imsm(struct supertype *st, int fd)
2641 {
2642 unsigned long long dsize;
2643 void *buf;
2644
2645 get_dev_size(fd, NULL, &dsize);
2646
2647 /* first block is stored on second to last sector of the disk */
2648 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
2649 return 1;
2650
2651 if (posix_memalign(&buf, 512, 512) != 0)
2652 return 1;
2653
2654 memset(buf, 0, 512);
2655 if (write(fd, buf, 512) != 512)
2656 return 1;
2657 return 0;
2658 }
2659
2660 static int imsm_bbm_log_size(struct imsm_super *mpb)
2661 {
2662 return __le32_to_cpu(mpb->bbm_log_size);
2663 }
2664
2665 #ifndef MDASSEMBLE
2666 static int validate_geometry_imsm_container(struct supertype *st, int level,
2667 int layout, int raiddisks, int chunk,
2668 unsigned long long size, char *dev,
2669 unsigned long long *freesize,
2670 int verbose)
2671 {
2672 int fd;
2673 unsigned long long ldsize;
2674 const struct imsm_orom *orom;
2675
2676 if (level != LEVEL_CONTAINER)
2677 return 0;
2678 if (!dev)
2679 return 1;
2680
2681 if (check_env("IMSM_NO_PLATFORM"))
2682 orom = NULL;
2683 else
2684 orom = find_imsm_orom();
2685 if (orom && raiddisks > orom->tds) {
2686 if (verbose)
2687 fprintf(stderr, Name ": %d exceeds maximum number of"
2688 " platform supported disks: %d\n",
2689 raiddisks, orom->tds);
2690 return 0;
2691 }
2692
2693 fd = open(dev, O_RDONLY|O_EXCL, 0);
2694 if (fd < 0) {
2695 if (verbose)
2696 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
2697 dev, strerror(errno));
2698 return 0;
2699 }
2700 if (!get_dev_size(fd, dev, &ldsize)) {
2701 close(fd);
2702 return 0;
2703 }
2704 close(fd);
2705
2706 *freesize = avail_size_imsm(st, ldsize >> 9);
2707
2708 return 1;
2709 }
2710
2711 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
2712 {
2713 const unsigned long long base_start = e[*idx].start;
2714 unsigned long long end = base_start + e[*idx].size;
2715 int i;
2716
2717 if (base_start == end)
2718 return 0;
2719
2720 *idx = *idx + 1;
2721 for (i = *idx; i < num_extents; i++) {
2722 /* extend overlapping extents */
2723 if (e[i].start >= base_start &&
2724 e[i].start <= end) {
2725 if (e[i].size == 0)
2726 return 0;
2727 if (e[i].start + e[i].size > end)
2728 end = e[i].start + e[i].size;
2729 } else if (e[i].start > end) {
2730 *idx = i;
2731 break;
2732 }
2733 }
2734
2735 return end - base_start;
2736 }
2737
2738 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
2739 {
2740 /* build a composite disk with all known extents and generate a new
2741 * 'maxsize' given the "all disks in an array must share a common start
2742 * offset" constraint
2743 */
2744 struct extent *e = calloc(sum_extents, sizeof(*e));
2745 struct dl *dl;
2746 int i, j;
2747 int start_extent;
2748 unsigned long long pos;
2749 unsigned long long start;
2750 unsigned long long maxsize;
2751 unsigned long reserve;
2752
2753 if (!e)
2754 return ~0ULL; /* error */
2755
2756 /* coalesce and sort all extents. also, check to see if we need to
2757 * reserve space between member arrays
2758 */
2759 j = 0;
2760 for (dl = super->disks; dl; dl = dl->next) {
2761 if (!dl->e)
2762 continue;
2763 for (i = 0; i < dl->extent_cnt; i++)
2764 e[j++] = dl->e[i];
2765 }
2766 qsort(e, sum_extents, sizeof(*e), cmp_extent);
2767
2768 /* merge extents */
2769 i = 0;
2770 j = 0;
2771 while (i < sum_extents) {
2772 e[j].start = e[i].start;
2773 e[j].size = find_size(e, &i, sum_extents);
2774 j++;
2775 if (e[j-1].size == 0)
2776 break;
2777 }
2778
2779 pos = 0;
2780 maxsize = 0;
2781 start_extent = 0;
2782 i = 0;
2783 do {
2784 unsigned long long esize;
2785
2786 esize = e[i].start - pos;
2787 if (esize >= maxsize) {
2788 maxsize = esize;
2789 start = pos;
2790 start_extent = i;
2791 }
2792 pos = e[i].start + e[i].size;
2793 i++;
2794 } while (e[i-1].size);
2795 free(e);
2796
2797 if (start_extent > 0)
2798 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
2799 else
2800 reserve = 0;
2801
2802 if (maxsize < reserve)
2803 return ~0ULL;
2804
2805 super->create_offset = ~((__u32) 0);
2806 if (start + reserve > super->create_offset)
2807 return ~0ULL; /* start overflows create_offset */
2808 super->create_offset = start + reserve;
2809
2810 return maxsize - reserve;
2811 }
2812
2813 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
2814 {
2815 if (level < 0 || level == 6 || level == 4)
2816 return 0;
2817
2818 /* if we have an orom prevent invalid raid levels */
2819 if (orom)
2820 switch (level) {
2821 case 0: return imsm_orom_has_raid0(orom);
2822 case 1:
2823 if (raiddisks > 2)
2824 return imsm_orom_has_raid1e(orom);
2825 return imsm_orom_has_raid1(orom) && raiddisks == 2;
2826 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
2827 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
2828 }
2829 else
2830 return 1; /* not on an Intel RAID platform so anything goes */
2831
2832 return 0;
2833 }
2834
2835 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2836 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2837 * FIX ME add ahci details
2838 */
2839 static int validate_geometry_imsm_volume(struct supertype *st, int level,
2840 int layout, int raiddisks, int chunk,
2841 unsigned long long size, char *dev,
2842 unsigned long long *freesize,
2843 int verbose)
2844 {
2845 struct stat stb;
2846 struct intel_super *super = st->sb;
2847 struct imsm_super *mpb = super->anchor;
2848 struct dl *dl;
2849 unsigned long long pos = 0;
2850 unsigned long long maxsize;
2851 struct extent *e;
2852 int i;
2853
2854 /* We must have the container info already read in. */
2855 if (!super)
2856 return 0;
2857
2858 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
2859 pr_vrb(": platform does not support raid%d with %d disk%s\n",
2860 level, raiddisks, raiddisks > 1 ? "s" : "");
2861 return 0;
2862 }
2863 if (super->orom && level != 1 &&
2864 !imsm_orom_has_chunk(super->orom, chunk)) {
2865 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
2866 return 0;
2867 }
2868 if (layout != imsm_level_to_layout(level)) {
2869 if (level == 5)
2870 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
2871 else if (level == 10)
2872 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
2873 else
2874 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
2875 layout, level);
2876 return 0;
2877 }
2878
2879 if (!dev) {
2880 /* General test: make sure there is space for
2881 * 'raiddisks' device extents of size 'size' at a given
2882 * offset
2883 */
2884 unsigned long long minsize = size;
2885 unsigned long long start_offset = ~0ULL;
2886 int dcnt = 0;
2887 if (minsize == 0)
2888 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
2889 for (dl = super->disks; dl ; dl = dl->next) {
2890 int found = 0;
2891
2892 pos = 0;
2893 i = 0;
2894 e = get_extents(super, dl);
2895 if (!e) continue;
2896 do {
2897 unsigned long long esize;
2898 esize = e[i].start - pos;
2899 if (esize >= minsize)
2900 found = 1;
2901 if (found && start_offset == ~0ULL) {
2902 start_offset = pos;
2903 break;
2904 } else if (found && pos != start_offset) {
2905 found = 0;
2906 break;
2907 }
2908 pos = e[i].start + e[i].size;
2909 i++;
2910 } while (e[i-1].size);
2911 if (found)
2912 dcnt++;
2913 free(e);
2914 }
2915 if (dcnt < raiddisks) {
2916 if (verbose)
2917 fprintf(stderr, Name ": imsm: Not enough "
2918 "devices with space for this array "
2919 "(%d < %d)\n",
2920 dcnt, raiddisks);
2921 return 0;
2922 }
2923 return 1;
2924 }
2925
2926 /* This device must be a member of the set */
2927 if (stat(dev, &stb) < 0)
2928 return 0;
2929 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2930 return 0;
2931 for (dl = super->disks ; dl ; dl = dl->next) {
2932 if (dl->major == major(stb.st_rdev) &&
2933 dl->minor == minor(stb.st_rdev))
2934 break;
2935 }
2936 if (!dl) {
2937 if (verbose)
2938 fprintf(stderr, Name ": %s is not in the "
2939 "same imsm set\n", dev);
2940 return 0;
2941 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
2942 /* If a volume is present then the current creation attempt
2943 * cannot incorporate new spares because the orom may not
2944 * understand this configuration (all member disks must be
2945 * members of each array in the container).
2946 */
2947 fprintf(stderr, Name ": %s is a spare and a volume"
2948 " is already defined for this container\n", dev);
2949 fprintf(stderr, Name ": The option-rom requires all member"
2950 " disks to be a member of all volumes\n");
2951 return 0;
2952 }
2953
2954 /* retrieve the largest free space block */
2955 e = get_extents(super, dl);
2956 maxsize = 0;
2957 i = 0;
2958 if (e) {
2959 do {
2960 unsigned long long esize;
2961
2962 esize = e[i].start - pos;
2963 if (esize >= maxsize)
2964 maxsize = esize;
2965 pos = e[i].start + e[i].size;
2966 i++;
2967 } while (e[i-1].size);
2968 dl->e = e;
2969 dl->extent_cnt = i;
2970 } else {
2971 if (verbose)
2972 fprintf(stderr, Name ": unable to determine free space for: %s\n",
2973 dev);
2974 return 0;
2975 }
2976 if (maxsize < size) {
2977 if (verbose)
2978 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
2979 dev, maxsize, size);
2980 return 0;
2981 }
2982
2983 /* count total number of extents for merge */
2984 i = 0;
2985 for (dl = super->disks; dl; dl = dl->next)
2986 if (dl->e)
2987 i += dl->extent_cnt;
2988
2989 maxsize = merge_extents(super, i);
2990 if (maxsize < size) {
2991 if (verbose)
2992 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
2993 maxsize, size);
2994 return 0;
2995 } else if (maxsize == ~0ULL) {
2996 if (verbose)
2997 fprintf(stderr, Name ": failed to merge %d extents\n", i);
2998 return 0;
2999 }
3000
3001 *freesize = maxsize;
3002
3003 return 1;
3004 }
3005
3006 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
3007 int raiddisks, int chunk, unsigned long long size,
3008 char *dev, unsigned long long *freesize,
3009 int verbose)
3010 {
3011 int fd, cfd;
3012 struct mdinfo *sra;
3013
3014 /* if given unused devices create a container
3015 * if given given devices in a container create a member volume
3016 */
3017 if (level == LEVEL_CONTAINER) {
3018 /* Must be a fresh device to add to a container */
3019 return validate_geometry_imsm_container(st, level, layout,
3020 raiddisks, chunk, size,
3021 dev, freesize,
3022 verbose);
3023 }
3024
3025 if (!dev) {
3026 if (st->sb && freesize) {
3027 /* Should do auto-layout here */
3028 fprintf(stderr, Name ": IMSM does not support auto-layout yet\n");
3029 return 0;
3030 }
3031 return 1;
3032 }
3033 if (st->sb) {
3034 /* creating in a given container */
3035 return validate_geometry_imsm_volume(st, level, layout,
3036 raiddisks, chunk, size,
3037 dev, freesize, verbose);
3038 }
3039
3040 /* limit creation to the following levels */
3041 if (!dev)
3042 switch (level) {
3043 case 0:
3044 case 1:
3045 case 10:
3046 case 5:
3047 break;
3048 default:
3049 return 1;
3050 }
3051
3052 /* This device needs to be a device in an 'imsm' container */
3053 fd = open(dev, O_RDONLY|O_EXCL, 0);
3054 if (fd >= 0) {
3055 if (verbose)
3056 fprintf(stderr,
3057 Name ": Cannot create this array on device %s\n",
3058 dev);
3059 close(fd);
3060 return 0;
3061 }
3062 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3063 if (verbose)
3064 fprintf(stderr, Name ": Cannot open %s: %s\n",
3065 dev, strerror(errno));
3066 return 0;
3067 }
3068 /* Well, it is in use by someone, maybe an 'imsm' container. */
3069 cfd = open_container(fd);
3070 if (cfd < 0) {
3071 close(fd);
3072 if (verbose)
3073 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3074 dev);
3075 return 0;
3076 }
3077 sra = sysfs_read(cfd, 0, GET_VERSION);
3078 close(fd);
3079 if (sra && sra->array.major_version == -1 &&
3080 strcmp(sra->text_version, "imsm") == 0) {
3081 /* This is a member of a imsm container. Load the container
3082 * and try to create a volume
3083 */
3084 struct intel_super *super;
3085
3086 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3087 st->sb = super;
3088 st->container_dev = fd2devnum(cfd);
3089 close(cfd);
3090 return validate_geometry_imsm_volume(st, level, layout,
3091 raiddisks, chunk,
3092 size, dev,
3093 freesize, verbose);
3094 }
3095 close(cfd);
3096 } else /* may belong to another container */
3097 return 0;
3098
3099 return 1;
3100 }
3101 #endif /* MDASSEMBLE */
3102
3103 static struct mdinfo *container_content_imsm(struct supertype *st)
3104 {
3105 /* Given a container loaded by load_super_imsm_all,
3106 * extract information about all the arrays into
3107 * an mdinfo tree.
3108 *
3109 * For each imsm_dev create an mdinfo, fill it in,
3110 * then look for matching devices in super->disks
3111 * and create appropriate device mdinfo.
3112 */
3113 struct intel_super *super = st->sb;
3114 struct imsm_super *mpb = super->anchor;
3115 struct mdinfo *rest = NULL;
3116 int i;
3117
3118 /* do not assemble arrays that might have bad blocks */
3119 if (imsm_bbm_log_size(super->anchor)) {
3120 fprintf(stderr, Name ": BBM log found in metadata. "
3121 "Cannot activate array(s).\n");
3122 return NULL;
3123 }
3124
3125 for (i = 0; i < mpb->num_raid_devs; i++) {
3126 struct imsm_dev *dev = get_imsm_dev(super, i);
3127 struct imsm_map *map = get_imsm_map(dev, 0);
3128 struct mdinfo *this;
3129 int slot;
3130
3131 this = malloc(sizeof(*this));
3132 memset(this, 0, sizeof(*this));
3133 this->next = rest;
3134
3135 super->current_vol = i;
3136 getinfo_super_imsm_volume(st, this);
3137 for (slot = 0 ; slot < map->num_members; slot++) {
3138 struct mdinfo *info_d;
3139 struct dl *d;
3140 int idx;
3141 int skip;
3142 __u32 s;
3143 __u32 ord;
3144
3145 skip = 0;
3146 idx = get_imsm_disk_idx(dev, slot);
3147 ord = get_imsm_ord_tbl_ent(dev, slot);
3148 for (d = super->disks; d ; d = d->next)
3149 if (d->index == idx)
3150 break;
3151
3152 if (d == NULL)
3153 skip = 1;
3154
3155 s = d ? d->disk.status : 0;
3156 if (s & FAILED_DISK)
3157 skip = 1;
3158 if (!(s & USABLE_DISK))
3159 skip = 1;
3160 if (ord & IMSM_ORD_REBUILD)
3161 skip = 1;
3162
3163 /*
3164 * if we skip some disks the array will be assmebled degraded;
3165 * reset resync start to avoid a dirty-degraded situation
3166 *
3167 * FIXME handle dirty degraded
3168 */
3169 if (skip && !dev->vol.dirty)
3170 this->resync_start = ~0ULL;
3171 if (skip)
3172 continue;
3173
3174 info_d = malloc(sizeof(*info_d));
3175 if (!info_d) {
3176 fprintf(stderr, Name ": failed to allocate disk"
3177 " for volume %s\n", (char *) dev->volume);
3178 free(this);
3179 this = rest;
3180 break;
3181 }
3182 memset(info_d, 0, sizeof(*info_d));
3183 info_d->next = this->devs;
3184 this->devs = info_d;
3185
3186 info_d->disk.number = d->index;
3187 info_d->disk.major = d->major;
3188 info_d->disk.minor = d->minor;
3189 info_d->disk.raid_disk = slot;
3190
3191 this->array.working_disks++;
3192
3193 info_d->events = __le32_to_cpu(mpb->generation_num);
3194 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3195 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3196 if (d->devname)
3197 strcpy(info_d->name, d->devname);
3198 }
3199 rest = this;
3200 }
3201
3202 return rest;
3203 }
3204
3205
3206 #ifndef MDASSEMBLE
3207 static int imsm_open_new(struct supertype *c, struct active_array *a,
3208 char *inst)
3209 {
3210 struct intel_super *super = c->sb;
3211 struct imsm_super *mpb = super->anchor;
3212
3213 if (atoi(inst) >= mpb->num_raid_devs) {
3214 fprintf(stderr, "%s: subarry index %d, out of range\n",
3215 __func__, atoi(inst));
3216 return -ENODEV;
3217 }
3218
3219 dprintf("imsm: open_new %s\n", inst);
3220 a->info.container_member = atoi(inst);
3221 return 0;
3222 }
3223
3224 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3225 {
3226 struct imsm_map *map = get_imsm_map(dev, 0);
3227
3228 if (!failed)
3229 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3230 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3231
3232 switch (get_imsm_raid_level(map)) {
3233 case 0:
3234 return IMSM_T_STATE_FAILED;
3235 break;
3236 case 1:
3237 if (failed < map->num_members)
3238 return IMSM_T_STATE_DEGRADED;
3239 else
3240 return IMSM_T_STATE_FAILED;
3241 break;
3242 case 10:
3243 {
3244 /**
3245 * check to see if any mirrors have failed, otherwise we
3246 * are degraded. Even numbered slots are mirrored on
3247 * slot+1
3248 */
3249 int i;
3250 /* gcc -Os complains that this is unused */
3251 int insync = insync;
3252
3253 for (i = 0; i < map->num_members; i++) {
3254 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3255 int idx = ord_to_idx(ord);
3256 struct imsm_disk *disk;
3257
3258 /* reset the potential in-sync count on even-numbered
3259 * slots. num_copies is always 2 for imsm raid10
3260 */
3261 if ((i & 1) == 0)
3262 insync = 2;
3263
3264 disk = get_imsm_disk(super, idx);
3265 if (!disk || disk->status & FAILED_DISK ||
3266 ord & IMSM_ORD_REBUILD)
3267 insync--;
3268
3269 /* no in-sync disks left in this mirror the
3270 * array has failed
3271 */
3272 if (insync == 0)
3273 return IMSM_T_STATE_FAILED;
3274 }
3275
3276 return IMSM_T_STATE_DEGRADED;
3277 }
3278 case 5:
3279 if (failed < 2)
3280 return IMSM_T_STATE_DEGRADED;
3281 else
3282 return IMSM_T_STATE_FAILED;
3283 break;
3284 default:
3285 break;
3286 }
3287
3288 return map->map_state;
3289 }
3290
3291 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3292 {
3293 int i;
3294 int failed = 0;
3295 struct imsm_disk *disk;
3296 struct imsm_map *map = get_imsm_map(dev, 0);
3297
3298 for (i = 0; i < map->num_members; i++) {
3299 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3300 int idx = ord_to_idx(ord);
3301
3302 disk = get_imsm_disk(super, idx);
3303 if (!disk || disk->status & FAILED_DISK ||
3304 ord & IMSM_ORD_REBUILD)
3305 failed++;
3306 }
3307
3308 return failed;
3309 }
3310
3311 static int is_resyncing(struct imsm_dev *dev)
3312 {
3313 struct imsm_map *migr_map;
3314
3315 if (!dev->vol.migr_state)
3316 return 0;
3317
3318 if (dev->vol.migr_type == MIGR_INIT)
3319 return 1;
3320
3321 migr_map = get_imsm_map(dev, 1);
3322
3323 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
3324 return 1;
3325 else
3326 return 0;
3327 }
3328
3329 static int is_rebuilding(struct imsm_dev *dev)
3330 {
3331 struct imsm_map *migr_map;
3332
3333 if (!dev->vol.migr_state)
3334 return 0;
3335
3336 if (dev->vol.migr_type != MIGR_REBUILD)
3337 return 0;
3338
3339 migr_map = get_imsm_map(dev, 1);
3340
3341 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
3342 return 1;
3343 else
3344 return 0;
3345 }
3346
3347 static void mark_failure(struct imsm_disk *disk)
3348 {
3349 if (disk->status & FAILED_DISK)
3350 return;
3351 disk->status |= FAILED_DISK;
3352 disk->scsi_id = __cpu_to_le32(~(__u32)0);
3353 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
3354 }
3355
3356 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3357 * states are handled in imsm_set_disk() with one exception, when a
3358 * resync is stopped due to a new failure this routine will set the
3359 * 'degraded' state for the array.
3360 */
3361 static int imsm_set_array_state(struct active_array *a, int consistent)
3362 {
3363 int inst = a->info.container_member;
3364 struct intel_super *super = a->container->sb;
3365 struct imsm_dev *dev = get_imsm_dev(super, inst);
3366 struct imsm_map *map = get_imsm_map(dev, 0);
3367 int failed = imsm_count_failed(super, dev);
3368 __u8 map_state = imsm_check_degraded(super, dev, failed);
3369
3370 /* before we activate this array handle any missing disks */
3371 if (consistent == 2 && super->missing) {
3372 struct dl *dl;
3373
3374 dprintf("imsm: mark missing\n");
3375 end_migration(dev, map_state);
3376 for (dl = super->missing; dl; dl = dl->next)
3377 mark_failure(&dl->disk);
3378 super->updates_pending++;
3379 }
3380
3381 if (consistent == 2 &&
3382 (!is_resync_complete(a) ||
3383 map_state != IMSM_T_STATE_NORMAL ||
3384 dev->vol.migr_state))
3385 consistent = 0;
3386
3387 if (is_resync_complete(a)) {
3388 /* complete intialization / resync,
3389 * recovery is completed in ->set_disk
3390 */
3391 if (is_resyncing(dev)) {
3392 dprintf("imsm: mark resync done\n");
3393 end_migration(dev, map_state);
3394 super->updates_pending++;
3395 }
3396 } else if (!is_resyncing(dev) && !failed) {
3397 /* mark the start of the init process if nothing is failed */
3398 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
3399 if (map->map_state == IMSM_T_STATE_NORMAL)
3400 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REBUILD);
3401 else
3402 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
3403 super->updates_pending++;
3404 }
3405
3406 /* check if we can update the migration checkpoint */
3407 if (dev->vol.migr_state &&
3408 __le32_to_cpu(dev->vol.curr_migr_unit) != a->resync_start) {
3409 dprintf("imsm: checkpoint migration (%llu)\n", a->resync_start);
3410 dev->vol.curr_migr_unit = __cpu_to_le32(a->resync_start);
3411 super->updates_pending++;
3412 }
3413
3414 /* mark dirty / clean */
3415 if (dev->vol.dirty != !consistent) {
3416 dprintf("imsm: mark '%s' (%llu)\n",
3417 consistent ? "clean" : "dirty", a->resync_start);
3418 if (consistent)
3419 dev->vol.dirty = 0;
3420 else
3421 dev->vol.dirty = 1;
3422 super->updates_pending++;
3423 }
3424 return consistent;
3425 }
3426
3427 static void imsm_set_disk(struct active_array *a, int n, int state)
3428 {
3429 int inst = a->info.container_member;
3430 struct intel_super *super = a->container->sb;
3431 struct imsm_dev *dev = get_imsm_dev(super, inst);
3432 struct imsm_map *map = get_imsm_map(dev, 0);
3433 struct imsm_disk *disk;
3434 int failed;
3435 __u32 ord;
3436 __u8 map_state;
3437
3438 if (n > map->num_members)
3439 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
3440 n, map->num_members - 1);
3441
3442 if (n < 0)
3443 return;
3444
3445 dprintf("imsm: set_disk %d:%x\n", n, state);
3446
3447 ord = get_imsm_ord_tbl_ent(dev, n);
3448 disk = get_imsm_disk(super, ord_to_idx(ord));
3449
3450 /* check for new failures */
3451 if ((state & DS_FAULTY) && !(disk->status & FAILED_DISK)) {
3452 mark_failure(disk);
3453 super->updates_pending++;
3454 }
3455
3456 /* check if in_sync */
3457 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD) {
3458 struct imsm_map *migr_map = get_imsm_map(dev, 1);
3459
3460 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
3461 super->updates_pending++;
3462 }
3463
3464 failed = imsm_count_failed(super, dev);
3465 map_state = imsm_check_degraded(super, dev, failed);
3466
3467 /* check if recovery complete, newly degraded, or failed */
3468 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
3469 end_migration(dev, map_state);
3470 super->updates_pending++;
3471 } else if (map_state == IMSM_T_STATE_DEGRADED &&
3472 map->map_state != map_state &&
3473 !dev->vol.migr_state) {
3474 dprintf("imsm: mark degraded\n");
3475 map->map_state = map_state;
3476 super->updates_pending++;
3477 } else if (map_state == IMSM_T_STATE_FAILED &&
3478 map->map_state != map_state) {
3479 dprintf("imsm: mark failed\n");
3480 end_migration(dev, map_state);
3481 super->updates_pending++;
3482 }
3483 }
3484
3485 static int store_imsm_mpb(int fd, struct intel_super *super)
3486 {
3487 struct imsm_super *mpb = super->anchor;
3488 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
3489 unsigned long long dsize;
3490 unsigned long long sectors;
3491
3492 get_dev_size(fd, NULL, &dsize);
3493
3494 if (mpb_size > 512) {
3495 /* -1 to account for anchor */
3496 sectors = mpb_sectors(mpb) - 1;
3497
3498 /* write the extended mpb to the sectors preceeding the anchor */
3499 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
3500 return 1;
3501
3502 if (write(fd, super->buf + 512, 512 * sectors) != 512 * sectors)
3503 return 1;
3504 }
3505
3506 /* first block is stored on second to last sector of the disk */
3507 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
3508 return 1;
3509
3510 if (write(fd, super->buf, 512) != 512)
3511 return 1;
3512
3513 return 0;
3514 }
3515
3516 static void imsm_sync_metadata(struct supertype *container)
3517 {
3518 struct intel_super *super = container->sb;
3519
3520 if (!super->updates_pending)
3521 return;
3522
3523 write_super_imsm(super, 0);
3524
3525 super->updates_pending = 0;
3526 }
3527
3528 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
3529 {
3530 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3531 int i = get_imsm_disk_idx(dev, idx);
3532 struct dl *dl;
3533
3534 for (dl = super->disks; dl; dl = dl->next)
3535 if (dl->index == i)
3536 break;
3537
3538 if (dl && dl->disk.status & FAILED_DISK)
3539 dl = NULL;
3540
3541 if (dl)
3542 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
3543
3544 return dl;
3545 }
3546
3547 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
3548 struct active_array *a, int activate_new)
3549 {
3550 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3551 int idx = get_imsm_disk_idx(dev, slot);
3552 struct imsm_super *mpb = super->anchor;
3553 struct imsm_map *map;
3554 unsigned long long esize;
3555 unsigned long long pos;
3556 struct mdinfo *d;
3557 struct extent *ex;
3558 int i, j;
3559 int found;
3560 __u32 array_start;
3561 __u32 blocks;
3562 struct dl *dl;
3563
3564 for (dl = super->disks; dl; dl = dl->next) {
3565 /* If in this array, skip */
3566 for (d = a->info.devs ; d ; d = d->next)
3567 if (d->state_fd >= 0 &&
3568 d->disk.major == dl->major &&
3569 d->disk.minor == dl->minor) {
3570 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3571 break;
3572 }
3573 if (d)
3574 continue;
3575
3576 /* skip in use or failed drives */
3577 if (dl->disk.status & FAILED_DISK || idx == dl->index ||
3578 dl->index == -2) {
3579 dprintf("%x:%x status (failed: %d index: %d)\n",
3580 dl->major, dl->minor,
3581 (dl->disk.status & FAILED_DISK) == FAILED_DISK, idx);
3582 continue;
3583 }
3584
3585 /* skip pure spares when we are looking for partially
3586 * assimilated drives
3587 */
3588 if (dl->index == -1 && !activate_new)
3589 continue;
3590
3591 /* Does this unused device have the requisite free space?
3592 * It needs to be able to cover all member volumes
3593 */
3594 ex = get_extents(super, dl);
3595 if (!ex) {
3596 dprintf("cannot get extents\n");
3597 continue;
3598 }
3599 for (i = 0; i < mpb->num_raid_devs; i++) {
3600 dev = get_imsm_dev(super, i);
3601 map = get_imsm_map(dev, 0);
3602
3603 /* check if this disk is already a member of
3604 * this array
3605 */
3606 for (j = 0; j < map->num_members; j++)
3607 if (get_imsm_disk_idx(dev, j) == dl->index)
3608 break;
3609 if (j < map->num_members)
3610 continue;
3611
3612 found = 0;
3613 j = 0;
3614 pos = 0;
3615 array_start = __le32_to_cpu(map->pba_of_lba0);
3616 blocks = __le32_to_cpu(map->blocks_per_member);
3617
3618 do {
3619 /* check that we can start at pba_of_lba0 with
3620 * blocks_per_member of space
3621 */
3622 esize = ex[j].start - pos;
3623 if (array_start >= pos &&
3624 array_start + blocks < ex[j].start) {
3625 found = 1;
3626 break;
3627 }
3628 pos = ex[j].start + ex[j].size;
3629 j++;
3630 } while (ex[j-1].size);
3631
3632 if (!found)
3633 break;
3634 }
3635
3636 free(ex);
3637 if (i < mpb->num_raid_devs) {
3638 dprintf("%x:%x does not have %u at %u\n",
3639 dl->major, dl->minor,
3640 blocks, array_start);
3641 /* No room */
3642 continue;
3643 }
3644 return dl;
3645 }
3646
3647 return dl;
3648 }
3649
3650 static struct mdinfo *imsm_activate_spare(struct active_array *a,
3651 struct metadata_update **updates)
3652 {
3653 /**
3654 * Find a device with unused free space and use it to replace a
3655 * failed/vacant region in an array. We replace failed regions one a
3656 * array at a time. The result is that a new spare disk will be added
3657 * to the first failed array and after the monitor has finished
3658 * propagating failures the remainder will be consumed.
3659 *
3660 * FIXME add a capability for mdmon to request spares from another
3661 * container.
3662 */
3663
3664 struct intel_super *super = a->container->sb;
3665 int inst = a->info.container_member;
3666 struct imsm_dev *dev = get_imsm_dev(super, inst);
3667 struct imsm_map *map = get_imsm_map(dev, 0);
3668 int failed = a->info.array.raid_disks;
3669 struct mdinfo *rv = NULL;
3670 struct mdinfo *d;
3671 struct mdinfo *di;
3672 struct metadata_update *mu;
3673 struct dl *dl;
3674 struct imsm_update_activate_spare *u;
3675 int num_spares = 0;
3676 int i;
3677
3678 for (d = a->info.devs ; d ; d = d->next) {
3679 if ((d->curr_state & DS_FAULTY) &&
3680 d->state_fd >= 0)
3681 /* wait for Removal to happen */
3682 return NULL;
3683 if (d->state_fd >= 0)
3684 failed--;
3685 }
3686
3687 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3688 inst, failed, a->info.array.raid_disks, a->info.array.level);
3689 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
3690 return NULL;
3691
3692 /* For each slot, if it is not working, find a spare */
3693 for (i = 0; i < a->info.array.raid_disks; i++) {
3694 for (d = a->info.devs ; d ; d = d->next)
3695 if (d->disk.raid_disk == i)
3696 break;
3697 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3698 if (d && (d->state_fd >= 0))
3699 continue;
3700
3701 /*
3702 * OK, this device needs recovery. Try to re-add the
3703 * previous occupant of this slot, if this fails see if
3704 * we can continue the assimilation of a spare that was
3705 * partially assimilated, finally try to activate a new
3706 * spare.
3707 */
3708 dl = imsm_readd(super, i, a);
3709 if (!dl)
3710 dl = imsm_add_spare(super, i, a, 0);
3711 if (!dl)
3712 dl = imsm_add_spare(super, i, a, 1);
3713 if (!dl)
3714 continue;
3715
3716 /* found a usable disk with enough space */
3717 di = malloc(sizeof(*di));
3718 if (!di)
3719 continue;
3720 memset(di, 0, sizeof(*di));
3721
3722 /* dl->index will be -1 in the case we are activating a
3723 * pristine spare. imsm_process_update() will create a
3724 * new index in this case. Once a disk is found to be
3725 * failed in all member arrays it is kicked from the
3726 * metadata
3727 */
3728 di->disk.number = dl->index;
3729
3730 /* (ab)use di->devs to store a pointer to the device
3731 * we chose
3732 */
3733 di->devs = (struct mdinfo *) dl;
3734
3735 di->disk.raid_disk = i;
3736 di->disk.major = dl->major;
3737 di->disk.minor = dl->minor;
3738 di->disk.state = 0;
3739 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
3740 di->component_size = a->info.component_size;
3741 di->container_member = inst;
3742 di->next = rv;
3743 rv = di;
3744 num_spares++;
3745 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3746 i, di->data_offset);
3747
3748 break;
3749 }
3750
3751 if (!rv)
3752 /* No spares found */
3753 return rv;
3754 /* Now 'rv' has a list of devices to return.
3755 * Create a metadata_update record to update the
3756 * disk_ord_tbl for the array
3757 */
3758 mu = malloc(sizeof(*mu));
3759 if (mu) {
3760 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
3761 if (mu->buf == NULL) {
3762 free(mu);
3763 mu = NULL;
3764 }
3765 }
3766 if (!mu) {
3767 while (rv) {
3768 struct mdinfo *n = rv->next;
3769
3770 free(rv);
3771 rv = n;
3772 }
3773 return NULL;
3774 }
3775
3776 mu->space = NULL;
3777 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
3778 mu->next = *updates;
3779 u = (struct imsm_update_activate_spare *) mu->buf;
3780
3781 for (di = rv ; di ; di = di->next) {
3782 u->type = update_activate_spare;
3783 u->dl = (struct dl *) di->devs;
3784 di->devs = NULL;
3785 u->slot = di->disk.raid_disk;
3786 u->array = inst;
3787 u->next = u + 1;
3788 u++;
3789 }
3790 (u-1)->next = NULL;
3791 *updates = mu;
3792
3793 return rv;
3794 }
3795
3796 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
3797 {
3798 struct imsm_dev *dev = get_imsm_dev(super, idx);
3799 struct imsm_map *map = get_imsm_map(dev, 0);
3800 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
3801 struct disk_info *inf = get_disk_info(u);
3802 struct imsm_disk *disk;
3803 int i;
3804 int j;
3805
3806 for (i = 0; i < map->num_members; i++) {
3807 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
3808 for (j = 0; j < new_map->num_members; j++)
3809 if (serialcmp(disk->serial, inf[j].serial) == 0)
3810 return 1;
3811 }
3812
3813 return 0;
3814 }
3815
3816 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
3817
3818 static void imsm_process_update(struct supertype *st,
3819 struct metadata_update *update)
3820 {
3821 /**
3822 * crack open the metadata_update envelope to find the update record
3823 * update can be one of:
3824 * update_activate_spare - a spare device has replaced a failed
3825 * device in an array, update the disk_ord_tbl. If this disk is
3826 * present in all member arrays then also clear the SPARE_DISK
3827 * flag
3828 */
3829 struct intel_super *super = st->sb;
3830 struct imsm_super *mpb;
3831 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
3832
3833 /* update requires a larger buf but the allocation failed */
3834 if (super->next_len && !super->next_buf) {
3835 super->next_len = 0;
3836 return;
3837 }
3838
3839 if (super->next_buf) {
3840 memcpy(super->next_buf, super->buf, super->len);
3841 free(super->buf);
3842 super->len = super->next_len;
3843 super->buf = super->next_buf;
3844
3845 super->next_len = 0;
3846 super->next_buf = NULL;
3847 }
3848
3849 mpb = super->anchor;
3850
3851 switch (type) {
3852 case update_activate_spare: {
3853 struct imsm_update_activate_spare *u = (void *) update->buf;
3854 struct imsm_dev *dev = get_imsm_dev(super, u->array);
3855 struct imsm_map *map = get_imsm_map(dev, 0);
3856 struct imsm_map *migr_map;
3857 struct active_array *a;
3858 struct imsm_disk *disk;
3859 __u8 to_state;
3860 struct dl *dl;
3861 unsigned int found;
3862 int failed;
3863 int victim = get_imsm_disk_idx(dev, u->slot);
3864 int i;
3865
3866 for (dl = super->disks; dl; dl = dl->next)
3867 if (dl == u->dl)
3868 break;
3869
3870 if (!dl) {
3871 fprintf(stderr, "error: imsm_activate_spare passed "
3872 "an unknown disk (index: %d)\n",
3873 u->dl->index);
3874 return;
3875 }
3876
3877 super->updates_pending++;
3878
3879 /* count failures (excluding rebuilds and the victim)
3880 * to determine map[0] state
3881 */
3882 failed = 0;
3883 for (i = 0; i < map->num_members; i++) {
3884 if (i == u->slot)
3885 continue;
3886 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
3887 if (!disk || disk->status & FAILED_DISK)
3888 failed++;
3889 }
3890
3891 /* adding a pristine spare, assign a new index */
3892 if (dl->index < 0) {
3893 dl->index = super->anchor->num_disks;
3894 super->anchor->num_disks++;
3895 }
3896 disk = &dl->disk;
3897 disk->status |= CONFIGURED_DISK;
3898 disk->status &= ~SPARE_DISK;
3899
3900 /* mark rebuild */
3901 to_state = imsm_check_degraded(super, dev, failed);
3902 map->map_state = IMSM_T_STATE_DEGRADED;
3903 migrate(dev, to_state, MIGR_REBUILD);
3904 migr_map = get_imsm_map(dev, 1);
3905 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
3906 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
3907
3908 /* count arrays using the victim in the metadata */
3909 found = 0;
3910 for (a = st->arrays; a ; a = a->next) {
3911 dev = get_imsm_dev(super, a->info.container_member);
3912 for (i = 0; i < map->num_members; i++)
3913 if (victim == get_imsm_disk_idx(dev, i))
3914 found++;
3915 }
3916
3917 /* delete the victim if it is no longer being
3918 * utilized anywhere
3919 */
3920 if (!found) {
3921 struct dl **dlp;
3922
3923 /* We know that 'manager' isn't touching anything,
3924 * so it is safe to delete
3925 */
3926 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
3927 if ((*dlp)->index == victim)
3928 break;
3929
3930 /* victim may be on the missing list */
3931 if (!*dlp)
3932 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
3933 if ((*dlp)->index == victim)
3934 break;
3935 imsm_delete(super, dlp, victim);
3936 }
3937 break;
3938 }
3939 case update_create_array: {
3940 /* someone wants to create a new array, we need to be aware of
3941 * a few races/collisions:
3942 * 1/ 'Create' called by two separate instances of mdadm
3943 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
3944 * devices that have since been assimilated via
3945 * activate_spare.
3946 * In the event this update can not be carried out mdadm will
3947 * (FIX ME) notice that its update did not take hold.
3948 */
3949 struct imsm_update_create_array *u = (void *) update->buf;
3950 struct intel_dev *dv;
3951 struct imsm_dev *dev;
3952 struct imsm_map *map, *new_map;
3953 unsigned long long start, end;
3954 unsigned long long new_start, new_end;
3955 int i;
3956 struct disk_info *inf;
3957 struct dl *dl;
3958
3959 /* handle racing creates: first come first serve */
3960 if (u->dev_idx < mpb->num_raid_devs) {
3961 dprintf("%s: subarray %d already defined\n",
3962 __func__, u->dev_idx);
3963 goto create_error;
3964 }
3965
3966 /* check update is next in sequence */
3967 if (u->dev_idx != mpb->num_raid_devs) {
3968 dprintf("%s: can not create array %d expected index %d\n",
3969 __func__, u->dev_idx, mpb->num_raid_devs);
3970 goto create_error;
3971 }
3972
3973 new_map = get_imsm_map(&u->dev, 0);
3974 new_start = __le32_to_cpu(new_map->pba_of_lba0);
3975 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
3976 inf = get_disk_info(u);
3977
3978 /* handle activate_spare versus create race:
3979 * check to make sure that overlapping arrays do not include
3980 * overalpping disks
3981 */
3982 for (i = 0; i < mpb->num_raid_devs; i++) {
3983 dev = get_imsm_dev(super, i);
3984 map = get_imsm_map(dev, 0);
3985 start = __le32_to_cpu(map->pba_of_lba0);
3986 end = start + __le32_to_cpu(map->blocks_per_member);
3987 if ((new_start >= start && new_start <= end) ||
3988 (start >= new_start && start <= new_end))
3989 /* overlap */;
3990 else
3991 continue;
3992
3993 if (disks_overlap(super, i, u)) {
3994 dprintf("%s: arrays overlap\n", __func__);
3995 goto create_error;
3996 }
3997 }
3998
3999 /* check that prepare update was successful */
4000 if (!update->space) {
4001 dprintf("%s: prepare update failed\n", __func__);
4002 goto create_error;
4003 }
4004
4005 /* check that all disks are still active before committing
4006 * changes. FIXME: could we instead handle this by creating a
4007 * degraded array? That's probably not what the user expects,
4008 * so better to drop this update on the floor.
4009 */
4010 for (i = 0; i < new_map->num_members; i++) {
4011 dl = serial_to_dl(inf[i].serial, super);
4012 if (!dl) {
4013 dprintf("%s: disk disappeared\n", __func__);
4014 goto create_error;
4015 }
4016 }
4017
4018 super->updates_pending++;
4019
4020 /* convert spares to members and fixup ord_tbl */
4021 for (i = 0; i < new_map->num_members; i++) {
4022 dl = serial_to_dl(inf[i].serial, super);
4023 if (dl->index == -1) {
4024 dl->index = mpb->num_disks;
4025 mpb->num_disks++;
4026 dl->disk.status |= CONFIGURED_DISK;
4027 dl->disk.status &= ~SPARE_DISK;
4028 }
4029 set_imsm_ord_tbl_ent(new_map, i, dl->index);
4030 }
4031
4032 dv = update->space;
4033 dev = dv->dev;
4034 update->space = NULL;
4035 imsm_copy_dev(dev, &u->dev);
4036 dv->index = u->dev_idx;
4037 dv->next = super->devlist;
4038 super->devlist = dv;
4039 mpb->num_raid_devs++;
4040
4041 imsm_update_version_info(super);
4042 break;
4043 create_error:
4044 /* mdmon knows how to release update->space, but not
4045 * ((struct intel_dev *) update->space)->dev
4046 */
4047 if (update->space) {
4048 dv = update->space;
4049 free(dv->dev);
4050 }
4051 break;
4052 }
4053 case update_add_disk:
4054
4055 /* we may be able to repair some arrays if disks are
4056 * being added */
4057 if (super->add) {
4058 struct active_array *a;
4059
4060 super->updates_pending++;
4061 for (a = st->arrays; a; a = a->next)
4062 a->check_degraded = 1;
4063 }
4064 /* add some spares to the metadata */
4065 while (super->add) {
4066 struct dl *al;
4067
4068 al = super->add;
4069 super->add = al->next;
4070 al->next = super->disks;
4071 super->disks = al;
4072 dprintf("%s: added %x:%x\n",
4073 __func__, al->major, al->minor);
4074 }
4075
4076 break;
4077 }
4078 }
4079
4080 static void imsm_prepare_update(struct supertype *st,
4081 struct metadata_update *update)
4082 {
4083 /**
4084 * Allocate space to hold new disk entries, raid-device entries or a new
4085 * mpb if necessary. The manager synchronously waits for updates to
4086 * complete in the monitor, so new mpb buffers allocated here can be
4087 * integrated by the monitor thread without worrying about live pointers
4088 * in the manager thread.
4089 */
4090 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4091 struct intel_super *super = st->sb;
4092 struct imsm_super *mpb = super->anchor;
4093 size_t buf_len;
4094 size_t len = 0;
4095
4096 switch (type) {
4097 case update_create_array: {
4098 struct imsm_update_create_array *u = (void *) update->buf;
4099 struct intel_dev *dv;
4100 struct imsm_dev *dev = &u->dev;
4101 struct imsm_map *map = get_imsm_map(dev, 0);
4102 struct dl *dl;
4103 struct disk_info *inf;
4104 int i;
4105 int activate = 0;
4106
4107 inf = get_disk_info(u);
4108 len = sizeof_imsm_dev(dev, 1);
4109 /* allocate a new super->devlist entry */
4110 dv = malloc(sizeof(*dv));
4111 if (dv) {
4112 dv->dev = malloc(len);
4113 if (dv->dev)
4114 update->space = dv;
4115 else {
4116 free(dv);
4117 update->space = NULL;
4118 }
4119 }
4120
4121 /* count how many spares will be converted to members */
4122 for (i = 0; i < map->num_members; i++) {
4123 dl = serial_to_dl(inf[i].serial, super);
4124 if (!dl) {
4125 /* hmm maybe it failed?, nothing we can do about
4126 * it here
4127 */
4128 continue;
4129 }
4130 if (count_memberships(dl, super) == 0)
4131 activate++;
4132 }
4133 len += activate * sizeof(struct imsm_disk);
4134 break;
4135 default:
4136 break;
4137 }
4138 }
4139
4140 /* check if we need a larger metadata buffer */
4141 if (super->next_buf)
4142 buf_len = super->next_len;
4143 else
4144 buf_len = super->len;
4145
4146 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4147 /* ok we need a larger buf than what is currently allocated
4148 * if this allocation fails process_update will notice that
4149 * ->next_len is set and ->next_buf is NULL
4150 */
4151 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4152 if (super->next_buf)
4153 free(super->next_buf);
4154
4155 super->next_len = buf_len;
4156 if (posix_memalign(&super->next_buf, 512, buf_len) != 0)
4157 super->next_buf = NULL;
4158 }
4159 }
4160
4161 /* must be called while manager is quiesced */
4162 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4163 {
4164 struct imsm_super *mpb = super->anchor;
4165 struct dl *iter;
4166 struct imsm_dev *dev;
4167 struct imsm_map *map;
4168 int i, j, num_members;
4169 __u32 ord;
4170
4171 dprintf("%s: deleting device[%d] from imsm_super\n",
4172 __func__, index);
4173
4174 /* shift all indexes down one */
4175 for (iter = super->disks; iter; iter = iter->next)
4176 if (iter->index > index)
4177 iter->index--;
4178 for (iter = super->missing; iter; iter = iter->next)
4179 if (iter->index > index)
4180 iter->index--;
4181
4182 for (i = 0; i < mpb->num_raid_devs; i++) {
4183 dev = get_imsm_dev(super, i);
4184 map = get_imsm_map(dev, 0);
4185 num_members = map->num_members;
4186 for (j = 0; j < num_members; j++) {
4187 /* update ord entries being careful not to propagate
4188 * ord-flags to the first map
4189 */
4190 ord = get_imsm_ord_tbl_ent(dev, j);
4191
4192 if (ord_to_idx(ord) <= index)
4193 continue;
4194
4195 map = get_imsm_map(dev, 0);
4196 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4197 map = get_imsm_map(dev, 1);
4198 if (map)
4199 set_imsm_ord_tbl_ent(map, j, ord - 1);
4200 }
4201 }
4202
4203 mpb->num_disks--;
4204 super->updates_pending++;
4205 if (*dlp) {
4206 struct dl *dl = *dlp;
4207
4208 *dlp = (*dlp)->next;
4209 __free_imsm_disk(dl);
4210 }
4211 }
4212 #endif /* MDASSEMBLE */
4213
4214 struct superswitch super_imsm = {
4215 #ifndef MDASSEMBLE
4216 .examine_super = examine_super_imsm,
4217 .brief_examine_super = brief_examine_super_imsm,
4218 .detail_super = detail_super_imsm,
4219 .brief_detail_super = brief_detail_super_imsm,
4220 .write_init_super = write_init_super_imsm,
4221 .validate_geometry = validate_geometry_imsm,
4222 .add_to_super = add_to_super_imsm,
4223 .detail_platform = detail_platform_imsm,
4224 #endif
4225 .match_home = match_home_imsm,
4226 .uuid_from_super= uuid_from_super_imsm,
4227 .getinfo_super = getinfo_super_imsm,
4228 .update_super = update_super_imsm,
4229
4230 .avail_size = avail_size_imsm,
4231
4232 .compare_super = compare_super_imsm,
4233
4234 .load_super = load_super_imsm,
4235 .init_super = init_super_imsm,
4236 .store_super = store_zero_imsm,
4237 .free_super = free_super_imsm,
4238 .match_metadata_desc = match_metadata_desc_imsm,
4239 .container_content = container_content_imsm,
4240 .default_layout = imsm_level_to_layout,
4241
4242 .external = 1,
4243 .name = "imsm",
4244
4245 #ifndef MDASSEMBLE
4246 /* for mdmon */
4247 .open_new = imsm_open_new,
4248 .load_super = load_super_imsm,
4249 .set_array_state= imsm_set_array_state,
4250 .set_disk = imsm_set_disk,
4251 .sync_metadata = imsm_sync_metadata,
4252 .activate_spare = imsm_activate_spare,
4253 .process_update = imsm_process_update,
4254 .prepare_update = imsm_prepare_update,
4255 #endif /* MDASSEMBLE */
4256 };
Unnamed repository; edit this file 'description' to name the repository.