]> 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: retry load_imsm_mpb if we suspect mdmon has made modifications
[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_disk_slot(struct imsm_map *map, int idx)
454 {
455 int slot;
456 __u32 ord;
457
458 for (slot = 0; slot < map->num_members; slot++) {
459 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
460 if (ord_to_idx(ord) == idx)
461 return slot;
462 }
463
464 return -1;
465 }
466
467 static int get_imsm_raid_level(struct imsm_map *map)
468 {
469 if (map->raid_level == 1) {
470 if (map->num_members == 2)
471 return 1;
472 else
473 return 10;
474 }
475
476 return map->raid_level;
477 }
478
479 static int cmp_extent(const void *av, const void *bv)
480 {
481 const struct extent *a = av;
482 const struct extent *b = bv;
483 if (a->start < b->start)
484 return -1;
485 if (a->start > b->start)
486 return 1;
487 return 0;
488 }
489
490 static int count_memberships(struct dl *dl, struct intel_super *super)
491 {
492 int memberships = 0;
493 int i;
494
495 for (i = 0; i < super->anchor->num_raid_devs; i++) {
496 struct imsm_dev *dev = get_imsm_dev(super, i);
497 struct imsm_map *map = get_imsm_map(dev, 0);
498
499 if (get_imsm_disk_slot(map, dl->index) >= 0)
500 memberships++;
501 }
502
503 return memberships;
504 }
505
506 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
507 {
508 /* find a list of used extents on the given physical device */
509 struct extent *rv, *e;
510 int i;
511 int memberships = count_memberships(dl, super);
512 __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
513
514 rv = malloc(sizeof(struct extent) * (memberships + 1));
515 if (!rv)
516 return NULL;
517 e = rv;
518
519 for (i = 0; i < super->anchor->num_raid_devs; i++) {
520 struct imsm_dev *dev = get_imsm_dev(super, i);
521 struct imsm_map *map = get_imsm_map(dev, 0);
522
523 if (get_imsm_disk_slot(map, dl->index) >= 0) {
524 e->start = __le32_to_cpu(map->pba_of_lba0);
525 e->size = __le32_to_cpu(map->blocks_per_member);
526 e++;
527 }
528 }
529 qsort(rv, memberships, sizeof(*rv), cmp_extent);
530
531 /* determine the start of the metadata
532 * when no raid devices are defined use the default
533 * ...otherwise allow the metadata to truncate the value
534 * as is the case with older versions of imsm
535 */
536 if (memberships) {
537 struct extent *last = &rv[memberships - 1];
538 __u32 remainder;
539
540 remainder = __le32_to_cpu(dl->disk.total_blocks) -
541 (last->start + last->size);
542 /* round down to 1k block to satisfy precision of the kernel
543 * 'size' interface
544 */
545 remainder &= ~1UL;
546 /* make sure remainder is still sane */
547 if (remainder < ROUND_UP(super->len, 512) >> 9)
548 remainder = ROUND_UP(super->len, 512) >> 9;
549 if (reservation > remainder)
550 reservation = remainder;
551 }
552 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
553 e->size = 0;
554 return rv;
555 }
556
557 /* try to determine how much space is reserved for metadata from
558 * the last get_extents() entry, otherwise fallback to the
559 * default
560 */
561 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
562 {
563 struct extent *e;
564 int i;
565 __u32 rv;
566
567 /* for spares just return a minimal reservation which will grow
568 * once the spare is picked up by an array
569 */
570 if (dl->index == -1)
571 return MPB_SECTOR_CNT;
572
573 e = get_extents(super, dl);
574 if (!e)
575 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
576
577 /* scroll to last entry */
578 for (i = 0; e[i].size; i++)
579 continue;
580
581 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
582
583 free(e);
584
585 return rv;
586 }
587
588 #ifndef MDASSEMBLE
589 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
590 {
591 __u64 sz;
592 int slot;
593 struct imsm_map *map = get_imsm_map(dev, 0);
594 __u32 ord;
595
596 printf("\n");
597 printf("[%.16s]:\n", dev->volume);
598 printf(" UUID : %s\n", uuid);
599 printf(" RAID Level : %d\n", get_imsm_raid_level(map));
600 printf(" Members : %d\n", map->num_members);
601 slot = get_imsm_disk_slot(map, disk_idx);
602 if (slot >= 0) {
603 ord = get_imsm_ord_tbl_ent(dev, slot);
604 printf(" This Slot : %d%s\n", slot,
605 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
606 } else
607 printf(" This Slot : ?\n");
608 sz = __le32_to_cpu(dev->size_high);
609 sz <<= 32;
610 sz += __le32_to_cpu(dev->size_low);
611 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
612 human_size(sz * 512));
613 sz = __le32_to_cpu(map->blocks_per_member);
614 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
615 human_size(sz * 512));
616 printf(" Sector Offset : %u\n",
617 __le32_to_cpu(map->pba_of_lba0));
618 printf(" Num Stripes : %u\n",
619 __le32_to_cpu(map->num_data_stripes));
620 printf(" Chunk Size : %u KiB\n",
621 __le16_to_cpu(map->blocks_per_strip) / 2);
622 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
623 printf(" Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle");
624 if (dev->vol.migr_state)
625 printf(": %s", dev->vol.migr_type ? "rebuilding" : "initializing");
626 printf("\n");
627 printf(" Map State : %s", map_state_str[map->map_state]);
628 if (dev->vol.migr_state) {
629 struct imsm_map *map = get_imsm_map(dev, 1);
630 printf(" <-- %s", map_state_str[map->map_state]);
631 }
632 printf("\n");
633 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
634 }
635
636 static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved)
637 {
638 struct imsm_disk *disk = __get_imsm_disk(mpb, index);
639 char str[MAX_RAID_SERIAL_LEN + 1];
640 __u32 s;
641 __u64 sz;
642
643 if (index < 0)
644 return;
645
646 printf("\n");
647 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
648 printf(" Disk%02d Serial : %s\n", index, str);
649 s = disk->status;
650 printf(" State :%s%s%s%s\n", s&SPARE_DISK ? " spare" : "",
651 s&CONFIGURED_DISK ? " active" : "",
652 s&FAILED_DISK ? " failed" : "",
653 s&USABLE_DISK ? " usable" : "");
654 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
655 sz = __le32_to_cpu(disk->total_blocks) - reserved;
656 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
657 human_size(sz * 512));
658 }
659
660 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info);
661
662 static void examine_super_imsm(struct supertype *st, char *homehost)
663 {
664 struct intel_super *super = st->sb;
665 struct imsm_super *mpb = super->anchor;
666 char str[MAX_SIGNATURE_LENGTH];
667 int i;
668 struct mdinfo info;
669 char nbuf[64];
670 __u32 sum;
671 __u32 reserved = imsm_reserved_sectors(super, super->disks);
672
673
674 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
675 printf(" Magic : %s\n", str);
676 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
677 printf(" Version : %s\n", get_imsm_version(mpb));
678 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
679 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
680 getinfo_super_imsm(st, &info);
681 fname_from_uuid(st, &info, nbuf,'-');
682 printf(" UUID : %s\n", nbuf + 5);
683 sum = __le32_to_cpu(mpb->check_sum);
684 printf(" Checksum : %08x %s\n", sum,
685 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
686 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
687 printf(" Disks : %d\n", mpb->num_disks);
688 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
689 print_imsm_disk(mpb, super->disks->index, reserved);
690 if (super->bbm_log) {
691 struct bbm_log *log = super->bbm_log;
692
693 printf("\n");
694 printf("Bad Block Management Log:\n");
695 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
696 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
697 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
698 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
699 printf(" First Spare : %llx\n", __le64_to_cpu(log->first_spare_lba));
700 }
701 for (i = 0; i < mpb->num_raid_devs; i++) {
702 struct mdinfo info;
703 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
704
705 super->current_vol = i;
706 getinfo_super_imsm(st, &info);
707 fname_from_uuid(st, &info, nbuf, '-');
708 print_imsm_dev(dev, nbuf + 5, super->disks->index);
709 }
710 for (i = 0; i < mpb->num_disks; i++) {
711 if (i == super->disks->index)
712 continue;
713 print_imsm_disk(mpb, i, reserved);
714 }
715 }
716
717 static void brief_examine_super_imsm(struct supertype *st)
718 {
719 /* We just write a generic IMSM ARRAY entry */
720 struct mdinfo info;
721 char nbuf[64];
722 char nbuf1[64];
723 struct intel_super *super = st->sb;
724 int i;
725
726 if (!super->anchor->num_raid_devs)
727 return;
728
729 getinfo_super_imsm(st, &info);
730 fname_from_uuid(st, &info, nbuf,'-');
731 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf + 5);
732 for (i = 0; i < super->anchor->num_raid_devs; i++) {
733 struct imsm_dev *dev = get_imsm_dev(super, i);
734
735 super->current_vol = i;
736 getinfo_super_imsm(st, &info);
737 fname_from_uuid(st, &info, nbuf1,'-');
738 printf("ARRAY /dev/md/%.16s container=%s\n"
739 " member=%d auto=mdp UUID=%s\n",
740 dev->volume, nbuf + 5, i, nbuf1 + 5);
741 }
742 }
743
744 static void detail_super_imsm(struct supertype *st, char *homehost)
745 {
746 struct mdinfo info;
747 char nbuf[64];
748
749 getinfo_super_imsm(st, &info);
750 fname_from_uuid(st, &info, nbuf,'-');
751 printf("\n UUID : %s\n", nbuf + 5);
752 }
753
754 static void brief_detail_super_imsm(struct supertype *st)
755 {
756 struct mdinfo info;
757 char nbuf[64];
758 getinfo_super_imsm(st, &info);
759 fname_from_uuid(st, &info, nbuf,'-');
760 printf(" UUID=%s", nbuf + 5);
761 }
762
763 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
764 static void fd2devname(int fd, char *name);
765
766 static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
767 {
768 /* dump an unsorted list of devices attached to ahci, as well as
769 * non-connected ports
770 */
771 int hba_len = strlen(hba_path) + 1;
772 struct dirent *ent;
773 DIR *dir;
774 char *path = NULL;
775 int err = 0;
776 unsigned long port_mask = (1 << port_count) - 1;
777
778 if (port_count > sizeof(port_mask) * 8) {
779 if (verbose)
780 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
781 return 2;
782 }
783
784 /* scroll through /sys/dev/block looking for devices attached to
785 * this hba
786 */
787 dir = opendir("/sys/dev/block");
788 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
789 int fd;
790 char model[64];
791 char vendor[64];
792 char buf[1024];
793 int major, minor;
794 char *device;
795 char *c;
796 int port;
797 int type;
798
799 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
800 continue;
801 path = devt_to_devpath(makedev(major, minor));
802 if (!path)
803 continue;
804 if (!path_attached_to_hba(path, hba_path)) {
805 free(path);
806 path = NULL;
807 continue;
808 }
809
810 /* retrieve the scsi device type */
811 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
812 if (verbose)
813 fprintf(stderr, Name ": failed to allocate 'device'\n");
814 err = 2;
815 break;
816 }
817 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
818 if (load_sys(device, buf) != 0) {
819 if (verbose)
820 fprintf(stderr, Name ": failed to read device type for %s\n",
821 path);
822 err = 2;
823 free(device);
824 break;
825 }
826 type = strtoul(buf, NULL, 10);
827
828 /* if it's not a disk print the vendor and model */
829 if (!(type == 0 || type == 7 || type == 14)) {
830 vendor[0] = '\0';
831 model[0] = '\0';
832 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
833 if (load_sys(device, buf) == 0) {
834 strncpy(vendor, buf, sizeof(vendor));
835 vendor[sizeof(vendor) - 1] = '\0';
836 c = (char *) &vendor[sizeof(vendor) - 1];
837 while (isspace(*c) || *c == '\0')
838 *c-- = '\0';
839
840 }
841 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
842 if (load_sys(device, buf) == 0) {
843 strncpy(model, buf, sizeof(model));
844 model[sizeof(model) - 1] = '\0';
845 c = (char *) &model[sizeof(model) - 1];
846 while (isspace(*c) || *c == '\0')
847 *c-- = '\0';
848 }
849
850 if (vendor[0] && model[0])
851 sprintf(buf, "%.64s %.64s", vendor, model);
852 else
853 switch (type) { /* numbers from hald/linux/device.c */
854 case 1: sprintf(buf, "tape"); break;
855 case 2: sprintf(buf, "printer"); break;
856 case 3: sprintf(buf, "processor"); break;
857 case 4:
858 case 5: sprintf(buf, "cdrom"); break;
859 case 6: sprintf(buf, "scanner"); break;
860 case 8: sprintf(buf, "media_changer"); break;
861 case 9: sprintf(buf, "comm"); break;
862 case 12: sprintf(buf, "raid"); break;
863 default: sprintf(buf, "unknown");
864 }
865 } else
866 buf[0] = '\0';
867 free(device);
868
869 /* chop device path to 'host%d' and calculate the port number */
870 c = strchr(&path[hba_len], '/');
871 *c = '\0';
872 if (sscanf(&path[hba_len], "host%d", &port) == 1)
873 port -= host_base;
874 else {
875 if (verbose) {
876 *c = '/'; /* repair the full string */
877 fprintf(stderr, Name ": failed to determine port number for %s\n",
878 path);
879 }
880 err = 2;
881 break;
882 }
883
884 /* mark this port as used */
885 port_mask &= ~(1 << port);
886
887 /* print out the device information */
888 if (buf[0]) {
889 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
890 continue;
891 }
892
893 fd = dev_open(ent->d_name, O_RDONLY);
894 if (fd < 0)
895 printf(" Port%d : - disk info unavailable -\n", port);
896 else {
897 fd2devname(fd, buf);
898 printf(" Port%d : %s", port, buf);
899 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
900 printf(" (%s)\n", buf);
901 else
902 printf("()\n");
903 }
904 close(fd);
905 free(path);
906 path = NULL;
907 }
908 if (path)
909 free(path);
910 if (dir)
911 closedir(dir);
912 if (err == 0) {
913 int i;
914
915 for (i = 0; i < port_count; i++)
916 if (port_mask & (1 << i))
917 printf(" Port%d : - no device attached -\n", i);
918 }
919
920 return err;
921 }
922
923 static int detail_platform_imsm(int verbose, int enumerate_only)
924 {
925 /* There are two components to imsm platform support, the ahci SATA
926 * controller and the option-rom. To find the SATA controller we
927 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
928 * controller with the Intel vendor id is present. This approach
929 * allows mdadm to leverage the kernel's ahci detection logic, with the
930 * caveat that if ahci.ko is not loaded mdadm will not be able to
931 * detect platform raid capabilities. The option-rom resides in a
932 * platform "Adapter ROM". We scan for its signature to retrieve the
933 * platform capabilities. If raid support is disabled in the BIOS the
934 * option-rom capability structure will not be available.
935 */
936 const struct imsm_orom *orom;
937 struct sys_dev *list, *hba;
938 DIR *dir;
939 struct dirent *ent;
940 const char *hba_path;
941 int host_base = 0;
942 int port_count = 0;
943
944 if (enumerate_only) {
945 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
946 return 0;
947 return 2;
948 }
949
950 list = find_driver_devices("pci", "ahci");
951 for (hba = list; hba; hba = hba->next)
952 if (devpath_to_vendor(hba->path) == 0x8086)
953 break;
954
955 if (!hba) {
956 if (verbose)
957 fprintf(stderr, Name ": unable to find active ahci controller\n");
958 free_sys_dev(&list);
959 return 2;
960 } else if (verbose)
961 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
962 hba_path = hba->path;
963 hba->path = NULL;
964 free_sys_dev(&list);
965
966 orom = find_imsm_orom();
967 if (!orom) {
968 if (verbose)
969 fprintf(stderr, Name ": imsm option-rom not found\n");
970 return 2;
971 }
972
973 printf(" Platform : Intel(R) Matrix Storage Manager\n");
974 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
975 orom->hotfix_ver, orom->build);
976 printf(" RAID Levels :%s%s%s%s%s\n",
977 imsm_orom_has_raid0(orom) ? " raid0" : "",
978 imsm_orom_has_raid1(orom) ? " raid1" : "",
979 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
980 imsm_orom_has_raid10(orom) ? " raid10" : "",
981 imsm_orom_has_raid5(orom) ? " raid5" : "");
982 printf(" Max Disks : %d\n", orom->tds);
983 printf(" Max Volumes : %d\n", orom->vpa);
984 printf(" I/O Controller : %s\n", hba_path);
985
986 /* find the smallest scsi host number to determine a port number base */
987 dir = opendir(hba_path);
988 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
989 int host;
990
991 if (sscanf(ent->d_name, "host%d", &host) != 1)
992 continue;
993 if (port_count == 0)
994 host_base = host;
995 else if (host < host_base)
996 host_base = host;
997
998 if (host + 1 > port_count + host_base)
999 port_count = host + 1 - host_base;
1000
1001 }
1002 if (dir)
1003 closedir(dir);
1004
1005 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1006 host_base, verbose) != 0) {
1007 if (verbose)
1008 fprintf(stderr, Name ": failed to enumerate ports\n");
1009 return 2;
1010 }
1011
1012 return 0;
1013 }
1014 #endif
1015
1016 static int match_home_imsm(struct supertype *st, char *homehost)
1017 {
1018 /* the imsm metadata format does not specify any host
1019 * identification information. We return -1 since we can never
1020 * confirm nor deny whether a given array is "meant" for this
1021 * host. We rely on compare_super and the 'family_num' field to
1022 * exclude member disks that do not belong, and we rely on
1023 * mdadm.conf to specify the arrays that should be assembled.
1024 * Auto-assembly may still pick up "foreign" arrays.
1025 */
1026
1027 return -1;
1028 }
1029
1030 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1031 {
1032 /* The uuid returned here is used for:
1033 * uuid to put into bitmap file (Create, Grow)
1034 * uuid for backup header when saving critical section (Grow)
1035 * comparing uuids when re-adding a device into an array
1036 * In these cases the uuid required is that of the data-array,
1037 * not the device-set.
1038 * uuid to recognise same set when adding a missing device back
1039 * to an array. This is a uuid for the device-set.
1040 *
1041 * For each of these we can make do with a truncated
1042 * or hashed uuid rather than the original, as long as
1043 * everyone agrees.
1044 * In each case the uuid required is that of the data-array,
1045 * not the device-set.
1046 */
1047 /* imsm does not track uuid's so we synthesis one using sha1 on
1048 * - The signature (Which is constant for all imsm array, but no matter)
1049 * - the family_num of the container
1050 * - the index number of the volume
1051 * - the 'serial' number of the volume.
1052 * Hopefully these are all constant.
1053 */
1054 struct intel_super *super = st->sb;
1055
1056 char buf[20];
1057 struct sha1_ctx ctx;
1058 struct imsm_dev *dev = NULL;
1059
1060 sha1_init_ctx(&ctx);
1061 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1062 sha1_process_bytes(&super->anchor->family_num, sizeof(__u32), &ctx);
1063 if (super->current_vol >= 0)
1064 dev = get_imsm_dev(super, super->current_vol);
1065 if (dev) {
1066 __u32 vol = super->current_vol;
1067 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1068 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1069 }
1070 sha1_finish_ctx(&ctx, buf);
1071 memcpy(uuid, buf, 4*4);
1072 }
1073
1074 #if 0
1075 static void
1076 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1077 {
1078 __u8 *v = get_imsm_version(mpb);
1079 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1080 char major[] = { 0, 0, 0 };
1081 char minor[] = { 0 ,0, 0 };
1082 char patch[] = { 0, 0, 0 };
1083 char *ver_parse[] = { major, minor, patch };
1084 int i, j;
1085
1086 i = j = 0;
1087 while (*v != '\0' && v < end) {
1088 if (*v != '.' && j < 2)
1089 ver_parse[i][j++] = *v;
1090 else {
1091 i++;
1092 j = 0;
1093 }
1094 v++;
1095 }
1096
1097 *m = strtol(minor, NULL, 0);
1098 *p = strtol(patch, NULL, 0);
1099 }
1100 #endif
1101
1102 static int imsm_level_to_layout(int level)
1103 {
1104 switch (level) {
1105 case 0:
1106 case 1:
1107 return 0;
1108 case 5:
1109 case 6:
1110 return ALGORITHM_LEFT_ASYMMETRIC;
1111 case 10:
1112 return 0x102;
1113 }
1114 return UnSet;
1115 }
1116
1117 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
1118 {
1119 struct intel_super *super = st->sb;
1120 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1121 struct imsm_map *map = get_imsm_map(dev, 0);
1122
1123 info->container_member = super->current_vol;
1124 info->array.raid_disks = map->num_members;
1125 info->array.level = get_imsm_raid_level(map);
1126 info->array.layout = imsm_level_to_layout(info->array.level);
1127 info->array.md_minor = -1;
1128 info->array.ctime = 0;
1129 info->array.utime = 0;
1130 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1131 info->array.state = !dev->vol.dirty;
1132
1133 info->disk.major = 0;
1134 info->disk.minor = 0;
1135
1136 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1137 info->component_size = __le32_to_cpu(map->blocks_per_member);
1138 memset(info->uuid, 0, sizeof(info->uuid));
1139
1140 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
1141 info->resync_start = 0;
1142 else if (dev->vol.migr_state)
1143 info->resync_start = __le32_to_cpu(dev->vol.curr_migr_unit);
1144 else
1145 info->resync_start = ~0ULL;
1146
1147 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1148 info->name[MAX_RAID_SERIAL_LEN] = 0;
1149
1150 info->array.major_version = -1;
1151 info->array.minor_version = -2;
1152 sprintf(info->text_version, "/%s/%d",
1153 devnum2devname(st->container_dev),
1154 info->container_member);
1155 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1156 uuid_from_super_imsm(st, info->uuid);
1157 }
1158
1159 /* check the config file to see if we can return a real uuid for this spare */
1160 static void fixup_container_spare_uuid(struct mdinfo *inf)
1161 {
1162 struct mddev_ident_s *array_list;
1163
1164 if (inf->array.level != LEVEL_CONTAINER ||
1165 memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0)
1166 return;
1167
1168 array_list = conf_get_ident(NULL);
1169
1170 for (; array_list; array_list = array_list->next) {
1171 if (array_list->uuid_set) {
1172 struct supertype *_sst; /* spare supertype */
1173 struct supertype *_cst; /* container supertype */
1174
1175 _cst = array_list->st;
1176 _sst = _cst->ss->match_metadata_desc(inf->text_version);
1177 if (_sst) {
1178 memcpy(inf->uuid, array_list->uuid, sizeof(int[4]));
1179 free(_sst);
1180 break;
1181 }
1182 }
1183 }
1184 }
1185
1186 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
1187 {
1188 struct intel_super *super = st->sb;
1189 struct imsm_disk *disk;
1190 __u32 s;
1191
1192 if (super->current_vol >= 0) {
1193 getinfo_super_imsm_volume(st, info);
1194 return;
1195 }
1196
1197 /* Set raid_disks to zero so that Assemble will always pull in valid
1198 * spares
1199 */
1200 info->array.raid_disks = 0;
1201 info->array.level = LEVEL_CONTAINER;
1202 info->array.layout = 0;
1203 info->array.md_minor = -1;
1204 info->array.ctime = 0; /* N/A for imsm */
1205 info->array.utime = 0;
1206 info->array.chunk_size = 0;
1207
1208 info->disk.major = 0;
1209 info->disk.minor = 0;
1210 info->disk.raid_disk = -1;
1211 info->reshape_active = 0;
1212 info->array.major_version = -1;
1213 info->array.minor_version = -2;
1214 strcpy(info->text_version, "imsm");
1215 info->safe_mode_delay = 0;
1216 info->disk.number = -1;
1217 info->disk.state = 0;
1218 info->name[0] = 0;
1219
1220 if (super->disks) {
1221 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1222
1223 disk = &super->disks->disk;
1224 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1225 info->component_size = reserved;
1226 s = disk->status;
1227 info->disk.state = s & CONFIGURED_DISK ? (1 << MD_DISK_ACTIVE) : 0;
1228 /* we don't change info->disk.raid_disk here because
1229 * this state will be finalized in mdmon after we have
1230 * found the 'most fresh' version of the metadata
1231 */
1232 info->disk.state |= s & FAILED_DISK ? (1 << MD_DISK_FAULTY) : 0;
1233 info->disk.state |= s & SPARE_DISK ? 0 : (1 << MD_DISK_SYNC);
1234 }
1235
1236 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1237 * ->compare_super may have updated the 'num_raid_devs' field for spares
1238 */
1239 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1240 uuid_from_super_imsm(st, info->uuid);
1241 else {
1242 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1243 fixup_container_spare_uuid(info);
1244 }
1245 }
1246
1247 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1248 char *update, char *devname, int verbose,
1249 int uuid_set, char *homehost)
1250 {
1251 /* FIXME */
1252
1253 /* For 'assemble' and 'force' we need to return non-zero if any
1254 * change was made. For others, the return value is ignored.
1255 * Update options are:
1256 * force-one : This device looks a bit old but needs to be included,
1257 * update age info appropriately.
1258 * assemble: clear any 'faulty' flag to allow this device to
1259 * be assembled.
1260 * force-array: Array is degraded but being forced, mark it clean
1261 * if that will be needed to assemble it.
1262 *
1263 * newdev: not used ????
1264 * grow: Array has gained a new device - this is currently for
1265 * linear only
1266 * resync: mark as dirty so a resync will happen.
1267 * name: update the name - preserving the homehost
1268 *
1269 * Following are not relevant for this imsm:
1270 * sparc2.2 : update from old dodgey metadata
1271 * super-minor: change the preferred_minor number
1272 * summaries: update redundant counters.
1273 * uuid: Change the uuid of the array to match watch is given
1274 * homehost: update the recorded homehost
1275 * _reshape_progress: record new reshape_progress position.
1276 */
1277 int rv = 0;
1278 //struct intel_super *super = st->sb;
1279 //struct imsm_super *mpb = super->mpb;
1280
1281 if (strcmp(update, "grow") == 0) {
1282 }
1283 if (strcmp(update, "resync") == 0) {
1284 /* dev->vol.dirty = 1; */
1285 }
1286
1287 /* IMSM has no concept of UUID or homehost */
1288
1289 return rv;
1290 }
1291
1292 static size_t disks_to_mpb_size(int disks)
1293 {
1294 size_t size;
1295
1296 size = sizeof(struct imsm_super);
1297 size += (disks - 1) * sizeof(struct imsm_disk);
1298 size += 2 * sizeof(struct imsm_dev);
1299 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1300 size += (4 - 2) * sizeof(struct imsm_map);
1301 /* 4 possible disk_ord_tbl's */
1302 size += 4 * (disks - 1) * sizeof(__u32);
1303
1304 return size;
1305 }
1306
1307 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1308 {
1309 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1310 return 0;
1311
1312 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1313 }
1314
1315 static void free_devlist(struct intel_super *super)
1316 {
1317 struct intel_dev *dv;
1318
1319 while (super->devlist) {
1320 dv = super->devlist->next;
1321 free(super->devlist->dev);
1322 free(super->devlist);
1323 super->devlist = dv;
1324 }
1325 }
1326
1327 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1328 {
1329 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1330 }
1331
1332 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1333 {
1334 /*
1335 * return:
1336 * 0 same, or first was empty, and second was copied
1337 * 1 second had wrong number
1338 * 2 wrong uuid
1339 * 3 wrong other info
1340 */
1341 struct intel_super *first = st->sb;
1342 struct intel_super *sec = tst->sb;
1343
1344 if (!first) {
1345 st->sb = tst->sb;
1346 tst->sb = NULL;
1347 return 0;
1348 }
1349
1350 if (memcmp(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH) != 0)
1351 return 3;
1352
1353 /* if an anchor does not have num_raid_devs set then it is a free
1354 * floating spare
1355 */
1356 if (first->anchor->num_raid_devs > 0 &&
1357 sec->anchor->num_raid_devs > 0) {
1358 if (first->anchor->family_num != sec->anchor->family_num)
1359 return 3;
1360 }
1361
1362 /* if 'first' is a spare promote it to a populated mpb with sec's
1363 * family number
1364 */
1365 if (first->anchor->num_raid_devs == 0 &&
1366 sec->anchor->num_raid_devs > 0) {
1367 int i;
1368 struct intel_dev *dv;
1369 struct imsm_dev *dev;
1370
1371 /* we need to copy raid device info from sec if an allocation
1372 * fails here we don't associate the spare
1373 */
1374 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1375 dv = malloc(sizeof(*dv));
1376 if (!dv)
1377 break;
1378 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1379 if (!dev) {
1380 free(dv);
1381 break;
1382 }
1383 dv->dev = dev;
1384 dv->index = i;
1385 dv->next = first->devlist;
1386 first->devlist = dv;
1387 }
1388 if (i <= sec->anchor->num_raid_devs) {
1389 /* allocation failure */
1390 free_devlist(first);
1391 fprintf(stderr, "imsm: failed to associate spare\n");
1392 return 3;
1393 }
1394 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1395 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1396
1397 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1398 first->anchor->family_num = sec->anchor->family_num;
1399 }
1400
1401 return 0;
1402 }
1403
1404 static void fd2devname(int fd, char *name)
1405 {
1406 struct stat st;
1407 char path[256];
1408 char dname[100];
1409 char *nm;
1410 int rv;
1411
1412 name[0] = '\0';
1413 if (fstat(fd, &st) != 0)
1414 return;
1415 sprintf(path, "/sys/dev/block/%d:%d",
1416 major(st.st_rdev), minor(st.st_rdev));
1417
1418 rv = readlink(path, dname, sizeof(dname));
1419 if (rv <= 0)
1420 return;
1421
1422 dname[rv] = '\0';
1423 nm = strrchr(dname, '/');
1424 nm++;
1425 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1426 }
1427
1428
1429 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1430
1431 static int imsm_read_serial(int fd, char *devname,
1432 __u8 serial[MAX_RAID_SERIAL_LEN])
1433 {
1434 unsigned char scsi_serial[255];
1435 int rv;
1436 int rsp_len;
1437 int len;
1438 char *c, *rsp_buf;
1439
1440 memset(scsi_serial, 0, sizeof(scsi_serial));
1441
1442 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1443
1444 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1445 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1446 fd2devname(fd, (char *) serial);
1447 return 0;
1448 }
1449
1450 if (rv != 0) {
1451 if (devname)
1452 fprintf(stderr,
1453 Name ": Failed to retrieve serial for %s\n",
1454 devname);
1455 return rv;
1456 }
1457
1458 /* trim leading whitespace */
1459 rsp_len = scsi_serial[3];
1460 if (!rsp_len) {
1461 if (devname)
1462 fprintf(stderr,
1463 Name ": Failed to retrieve serial for %s\n",
1464 devname);
1465 return 2;
1466 }
1467 rsp_buf = (char *) &scsi_serial[4];
1468 c = rsp_buf;
1469 while (isspace(*c))
1470 c++;
1471
1472 /* truncate len to the end of rsp_buf if necessary */
1473 if (c + MAX_RAID_SERIAL_LEN > rsp_buf + rsp_len)
1474 len = rsp_len - (c - rsp_buf);
1475 else
1476 len = MAX_RAID_SERIAL_LEN;
1477
1478 /* initialize the buffer and copy rsp_buf characters */
1479 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1480 memcpy(serial, c, len);
1481
1482 /* trim trailing whitespace starting with the last character copied */
1483 c = (char *) &serial[len - 1];
1484 while (isspace(*c) || *c == '\0')
1485 *c-- = '\0';
1486
1487 return 0;
1488 }
1489
1490 static int serialcmp(__u8 *s1, __u8 *s2)
1491 {
1492 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1493 }
1494
1495 static void serialcpy(__u8 *dest, __u8 *src)
1496 {
1497 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1498 }
1499
1500 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1501 {
1502 struct dl *dl;
1503
1504 for (dl = super->disks; dl; dl = dl->next)
1505 if (serialcmp(dl->serial, serial) == 0)
1506 break;
1507
1508 return dl;
1509 }
1510
1511 static int
1512 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1513 {
1514 struct dl *dl;
1515 struct stat stb;
1516 int rv;
1517 int i;
1518 int alloc = 1;
1519 __u8 serial[MAX_RAID_SERIAL_LEN];
1520
1521 rv = imsm_read_serial(fd, devname, serial);
1522
1523 if (rv != 0)
1524 return 2;
1525
1526 /* check if this is a disk we have seen before. it may be a spare in
1527 * super->disks while the current anchor believes it is a raid member,
1528 * check if we need to update dl->index
1529 */
1530 dl = serial_to_dl(serial, super);
1531 if (!dl)
1532 dl = malloc(sizeof(*dl));
1533 else
1534 alloc = 0;
1535
1536 if (!dl) {
1537 if (devname)
1538 fprintf(stderr,
1539 Name ": failed to allocate disk buffer for %s\n",
1540 devname);
1541 return 2;
1542 }
1543
1544 if (alloc) {
1545 fstat(fd, &stb);
1546 dl->major = major(stb.st_rdev);
1547 dl->minor = minor(stb.st_rdev);
1548 dl->next = super->disks;
1549 dl->fd = keep_fd ? fd : -1;
1550 dl->devname = devname ? strdup(devname) : NULL;
1551 serialcpy(dl->serial, serial);
1552 dl->index = -2;
1553 dl->e = NULL;
1554 } else if (keep_fd) {
1555 close(dl->fd);
1556 dl->fd = fd;
1557 }
1558
1559 /* look up this disk's index in the current anchor */
1560 for (i = 0; i < super->anchor->num_disks; i++) {
1561 struct imsm_disk *disk_iter;
1562
1563 disk_iter = __get_imsm_disk(super->anchor, i);
1564
1565 if (serialcmp(disk_iter->serial, dl->serial) == 0) {
1566 dl->disk = *disk_iter;
1567 /* only set index on disks that are a member of a
1568 * populated contianer, i.e. one with raid_devs
1569 */
1570 if (dl->disk.status & FAILED_DISK)
1571 dl->index = -2;
1572 else if (dl->disk.status & SPARE_DISK)
1573 dl->index = -1;
1574 else
1575 dl->index = i;
1576
1577 break;
1578 }
1579 }
1580
1581 /* no match, maybe a stale failed drive */
1582 if (i == super->anchor->num_disks && dl->index >= 0) {
1583 dl->disk = *__get_imsm_disk(super->anchor, dl->index);
1584 if (dl->disk.status & FAILED_DISK)
1585 dl->index = -2;
1586 }
1587
1588 if (alloc)
1589 super->disks = dl;
1590
1591 return 0;
1592 }
1593
1594 #ifndef MDASSEMBLE
1595 /* When migrating map0 contains the 'destination' state while map1
1596 * contains the current state. When not migrating map0 contains the
1597 * current state. This routine assumes that map[0].map_state is set to
1598 * the current array state before being called.
1599 *
1600 * Migration is indicated by one of the following states
1601 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1602 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1603 * map1state=unitialized)
1604 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1605 * map1state=normal)
1606 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1607 * map1state=degraded)
1608 */
1609 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
1610 {
1611 struct imsm_map *dest;
1612 struct imsm_map *src = get_imsm_map(dev, 0);
1613
1614 dev->vol.migr_state = 1;
1615 dev->vol.migr_type = migr_type;
1616 dev->vol.curr_migr_unit = 0;
1617 dest = get_imsm_map(dev, 1);
1618
1619 /* duplicate and then set the target end state in map[0] */
1620 memcpy(dest, src, sizeof_imsm_map(src));
1621 if (migr_type == MIGR_REBUILD) {
1622 __u32 ord;
1623 int i;
1624
1625 for (i = 0; i < src->num_members; i++) {
1626 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
1627 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
1628 }
1629 }
1630
1631 src->map_state = to_state;
1632 }
1633
1634 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1635 {
1636 struct imsm_map *map = get_imsm_map(dev, 0);
1637 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
1638 int i;
1639
1640 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1641 * completed in the last migration.
1642 *
1643 * FIXME add support for online capacity expansion and
1644 * raid-level-migration
1645 */
1646 for (i = 0; i < prev->num_members; i++)
1647 map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i];
1648
1649 dev->vol.migr_state = 0;
1650 dev->vol.curr_migr_unit = 0;
1651 map->map_state = map_state;
1652 }
1653 #endif
1654
1655 static int parse_raid_devices(struct intel_super *super)
1656 {
1657 int i;
1658 struct imsm_dev *dev_new;
1659 size_t len, len_migr;
1660 size_t space_needed = 0;
1661 struct imsm_super *mpb = super->anchor;
1662
1663 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1664 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1665 struct intel_dev *dv;
1666
1667 len = sizeof_imsm_dev(dev_iter, 0);
1668 len_migr = sizeof_imsm_dev(dev_iter, 1);
1669 if (len_migr > len)
1670 space_needed += len_migr - len;
1671
1672 dv = malloc(sizeof(*dv));
1673 if (!dv)
1674 return 1;
1675 dev_new = malloc(len_migr);
1676 if (!dev_new) {
1677 free(dv);
1678 return 1;
1679 }
1680 imsm_copy_dev(dev_new, dev_iter);
1681 dv->dev = dev_new;
1682 dv->index = i;
1683 dv->next = super->devlist;
1684 super->devlist = dv;
1685 }
1686
1687 /* ensure that super->buf is large enough when all raid devices
1688 * are migrating
1689 */
1690 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1691 void *buf;
1692
1693 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1694 if (posix_memalign(&buf, 512, len) != 0)
1695 return 1;
1696
1697 memcpy(buf, super->buf, len);
1698 free(super->buf);
1699 super->buf = buf;
1700 super->len = len;
1701 }
1702
1703 return 0;
1704 }
1705
1706 /* retrieve a pointer to the bbm log which starts after all raid devices */
1707 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1708 {
1709 void *ptr = NULL;
1710
1711 if (__le32_to_cpu(mpb->bbm_log_size)) {
1712 ptr = mpb;
1713 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1714 }
1715
1716 return ptr;
1717 }
1718
1719 static void __free_imsm(struct intel_super *super, int free_disks);
1720
1721 /* load_imsm_mpb - read matrix metadata
1722 * allocates super->mpb to be freed by free_super
1723 */
1724 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1725 {
1726 unsigned long long dsize;
1727 unsigned long long sectors;
1728 struct stat;
1729 struct imsm_super *anchor;
1730 __u32 check_sum;
1731 int rc;
1732
1733 get_dev_size(fd, NULL, &dsize);
1734
1735 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1736 if (devname)
1737 fprintf(stderr,
1738 Name ": Cannot seek to anchor block on %s: %s\n",
1739 devname, strerror(errno));
1740 return 1;
1741 }
1742
1743 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1744 if (devname)
1745 fprintf(stderr,
1746 Name ": Failed to allocate imsm anchor buffer"
1747 " on %s\n", devname);
1748 return 1;
1749 }
1750 if (read(fd, anchor, 512) != 512) {
1751 if (devname)
1752 fprintf(stderr,
1753 Name ": Cannot read anchor block on %s: %s\n",
1754 devname, strerror(errno));
1755 free(anchor);
1756 return 1;
1757 }
1758
1759 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1760 if (devname)
1761 fprintf(stderr,
1762 Name ": no IMSM anchor on %s\n", devname);
1763 free(anchor);
1764 return 2;
1765 }
1766
1767 __free_imsm(super, 0);
1768 super->len = ROUND_UP(anchor->mpb_size, 512);
1769 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1770 if (devname)
1771 fprintf(stderr,
1772 Name ": unable to allocate %zu byte mpb buffer\n",
1773 super->len);
1774 free(anchor);
1775 return 2;
1776 }
1777 memcpy(super->buf, anchor, 512);
1778
1779 sectors = mpb_sectors(anchor) - 1;
1780 free(anchor);
1781 if (!sectors) {
1782 check_sum = __gen_imsm_checksum(super->anchor);
1783 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1784 if (devname)
1785 fprintf(stderr,
1786 Name ": IMSM checksum %x != %x on %s\n",
1787 check_sum,
1788 __le32_to_cpu(super->anchor->check_sum),
1789 devname);
1790 return 2;
1791 }
1792
1793 rc = load_imsm_disk(fd, super, devname, 0);
1794 if (rc == 0)
1795 rc = parse_raid_devices(super);
1796 return rc;
1797 }
1798
1799 /* read the extended mpb */
1800 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1801 if (devname)
1802 fprintf(stderr,
1803 Name ": Cannot seek to extended mpb on %s: %s\n",
1804 devname, strerror(errno));
1805 return 1;
1806 }
1807
1808 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1809 if (devname)
1810 fprintf(stderr,
1811 Name ": Cannot read extended mpb on %s: %s\n",
1812 devname, strerror(errno));
1813 return 2;
1814 }
1815
1816 check_sum = __gen_imsm_checksum(super->anchor);
1817 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1818 if (devname)
1819 fprintf(stderr,
1820 Name ": IMSM checksum %x != %x on %s\n",
1821 check_sum, __le32_to_cpu(super->anchor->check_sum),
1822 devname);
1823 return 3;
1824 }
1825
1826 /* FIXME the BBM log is disk specific so we cannot use this global
1827 * buffer for all disks. Ok for now since we only look at the global
1828 * bbm_log_size parameter to gate assembly
1829 */
1830 super->bbm_log = __get_imsm_bbm_log(super->anchor);
1831
1832 rc = load_imsm_disk(fd, super, devname, 0);
1833 if (rc == 0)
1834 rc = parse_raid_devices(super);
1835
1836 return rc;
1837 }
1838
1839 static void __free_imsm_disk(struct dl *d)
1840 {
1841 if (d->fd >= 0)
1842 close(d->fd);
1843 if (d->devname)
1844 free(d->devname);
1845 if (d->e)
1846 free(d->e);
1847 free(d);
1848
1849 }
1850 static void free_imsm_disks(struct intel_super *super)
1851 {
1852 struct dl *d;
1853
1854 while (super->disks) {
1855 d = super->disks;
1856 super->disks = d->next;
1857 __free_imsm_disk(d);
1858 }
1859 while (super->missing) {
1860 d = super->missing;
1861 super->missing = d->next;
1862 __free_imsm_disk(d);
1863 }
1864
1865 }
1866
1867 /* free all the pieces hanging off of a super pointer */
1868 static void __free_imsm(struct intel_super *super, int free_disks)
1869 {
1870 if (super->buf) {
1871 free(super->buf);
1872 super->buf = NULL;
1873 }
1874 if (free_disks)
1875 free_imsm_disks(super);
1876 free_devlist(super);
1877 if (super->hba) {
1878 free((void *) super->hba);
1879 super->hba = NULL;
1880 }
1881 }
1882
1883 static void free_imsm(struct intel_super *super)
1884 {
1885 __free_imsm(super, 1);
1886 free(super);
1887 }
1888
1889 static void free_super_imsm(struct supertype *st)
1890 {
1891 struct intel_super *super = st->sb;
1892
1893 if (!super)
1894 return;
1895
1896 free_imsm(super);
1897 st->sb = NULL;
1898 }
1899
1900 static struct intel_super *alloc_super(int creating_imsm)
1901 {
1902 struct intel_super *super = malloc(sizeof(*super));
1903
1904 if (super) {
1905 memset(super, 0, sizeof(*super));
1906 super->creating_imsm = creating_imsm;
1907 super->current_vol = -1;
1908 super->create_offset = ~((__u32 ) 0);
1909 if (!check_env("IMSM_NO_PLATFORM"))
1910 super->orom = find_imsm_orom();
1911 if (super->orom && !check_env("IMSM_TEST_OROM")) {
1912 struct sys_dev *list, *ent;
1913
1914 /* find the first intel ahci controller */
1915 list = find_driver_devices("pci", "ahci");
1916 for (ent = list; ent; ent = ent->next)
1917 if (devpath_to_vendor(ent->path) == 0x8086)
1918 break;
1919 if (ent) {
1920 super->hba = ent->path;
1921 ent->path = NULL;
1922 }
1923 free_sys_dev(&list);
1924 }
1925 }
1926
1927 return super;
1928 }
1929
1930 #ifndef MDASSEMBLE
1931 /* find_missing - helper routine for load_super_imsm_all that identifies
1932 * disks that have disappeared from the system. This routine relies on
1933 * the mpb being uptodate, which it is at load time.
1934 */
1935 static int find_missing(struct intel_super *super)
1936 {
1937 int i;
1938 struct imsm_super *mpb = super->anchor;
1939 struct dl *dl;
1940 struct imsm_disk *disk;
1941
1942 for (i = 0; i < mpb->num_disks; i++) {
1943 disk = __get_imsm_disk(mpb, i);
1944 dl = serial_to_dl(disk->serial, super);
1945 if (dl)
1946 continue;
1947
1948 dl = malloc(sizeof(*dl));
1949 if (!dl)
1950 return 1;
1951 dl->major = 0;
1952 dl->minor = 0;
1953 dl->fd = -1;
1954 dl->devname = strdup("missing");
1955 dl->index = i;
1956 serialcpy(dl->serial, disk->serial);
1957 dl->disk = *disk;
1958 dl->e = NULL;
1959 dl->next = super->missing;
1960 super->missing = dl;
1961 }
1962
1963 return 0;
1964 }
1965
1966 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
1967 char *devname, int keep_fd)
1968 {
1969 struct mdinfo *sra;
1970 struct intel_super *super;
1971 struct mdinfo *sd, *best = NULL;
1972 __u32 bestgen = 0;
1973 __u32 gen;
1974 char nm[20];
1975 int dfd;
1976 int rv;
1977 int devnum = fd2devnum(fd);
1978 int retry;
1979
1980 /* check if this disk is a member of an active array */
1981 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
1982 if (!sra)
1983 return 1;
1984
1985 if (sra->array.major_version != -1 ||
1986 sra->array.minor_version != -2 ||
1987 strcmp(sra->text_version, "imsm") != 0)
1988 return 1;
1989
1990 super = alloc_super(0);
1991 if (!super)
1992 return 1;
1993
1994 /* find the most up to date disk in this array, skipping spares */
1995 for (sd = sra->devs; sd; sd = sd->next) {
1996 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
1997 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
1998 if (dfd < 0) {
1999 free_imsm(super);
2000 return 2;
2001 }
2002 rv = load_imsm_mpb(dfd, super, NULL);
2003
2004 /* retry the load if we might have raced against mdmon */
2005 if (rv == 3 && mdmon_running(devnum))
2006 for (retry = 0; retry < 3; retry++) {
2007 usleep(3000);
2008 rv = load_imsm_mpb(dfd, super, NULL);
2009 if (rv != 3)
2010 break;
2011 }
2012 if (!keep_fd)
2013 close(dfd);
2014 if (rv == 0) {
2015 if (super->anchor->num_raid_devs == 0)
2016 gen = 0;
2017 else
2018 gen = __le32_to_cpu(super->anchor->generation_num);
2019 if (!best || gen > bestgen) {
2020 bestgen = gen;
2021 best = sd;
2022 }
2023 } else {
2024 free_imsm(super);
2025 return rv;
2026 }
2027 }
2028
2029 if (!best) {
2030 free_imsm(super);
2031 return 1;
2032 }
2033
2034 /* load the most up to date anchor */
2035 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2036 dfd = dev_open(nm, O_RDONLY);
2037 if (dfd < 0) {
2038 free_imsm(super);
2039 return 1;
2040 }
2041 rv = load_imsm_mpb(dfd, super, NULL);
2042 close(dfd);
2043 if (rv != 0) {
2044 free_imsm(super);
2045 return 2;
2046 }
2047
2048 /* re-parse the disk list with the current anchor */
2049 for (sd = sra->devs ; sd ; sd = sd->next) {
2050 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2051 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2052 if (dfd < 0) {
2053 free_imsm(super);
2054 return 2;
2055 }
2056 load_imsm_disk(dfd, super, NULL, keep_fd);
2057 if (!keep_fd)
2058 close(dfd);
2059 }
2060
2061
2062 if (find_missing(super) != 0) {
2063 free_imsm(super);
2064 return 2;
2065 }
2066
2067 if (st->subarray[0]) {
2068 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2069 super->current_vol = atoi(st->subarray);
2070 else
2071 return 1;
2072 }
2073
2074 *sbp = super;
2075 st->container_dev = devnum;
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 = 1;
2082
2083 return 0;
2084 }
2085 #endif
2086
2087 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2088 {
2089 struct intel_super *super;
2090 int rv;
2091
2092 #ifndef MDASSEMBLE
2093 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2094 return 0;
2095 #endif
2096 if (st->subarray[0])
2097 return 1; /* FIXME */
2098
2099 super = alloc_super(0);
2100 if (!super) {
2101 fprintf(stderr,
2102 Name ": malloc of %zu failed.\n",
2103 sizeof(*super));
2104 return 1;
2105 }
2106
2107 rv = load_imsm_mpb(fd, super, devname);
2108
2109 if (rv) {
2110 if (devname)
2111 fprintf(stderr,
2112 Name ": Failed to load all information "
2113 "sections on %s\n", devname);
2114 free_imsm(super);
2115 return rv;
2116 }
2117
2118 st->sb = super;
2119 if (st->ss == NULL) {
2120 st->ss = &super_imsm;
2121 st->minor_version = 0;
2122 st->max_devs = IMSM_MAX_DEVICES;
2123 }
2124 st->loaded_container = 0;
2125
2126 return 0;
2127 }
2128
2129 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2130 {
2131 if (info->level == 1)
2132 return 128;
2133 return info->chunk_size >> 9;
2134 }
2135
2136 static __u32 info_to_num_data_stripes(mdu_array_info_t *info)
2137 {
2138 __u32 num_stripes;
2139
2140 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2141 if (info->level == 1)
2142 num_stripes /= 2;
2143
2144 return num_stripes;
2145 }
2146
2147 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2148 {
2149 if (info->level == 1)
2150 return info->size * 2;
2151 else
2152 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2153 }
2154
2155 static void imsm_update_version_info(struct intel_super *super)
2156 {
2157 /* update the version and attributes */
2158 struct imsm_super *mpb = super->anchor;
2159 char *version;
2160 struct imsm_dev *dev;
2161 struct imsm_map *map;
2162 int i;
2163
2164 for (i = 0; i < mpb->num_raid_devs; i++) {
2165 dev = get_imsm_dev(super, i);
2166 map = get_imsm_map(dev, 0);
2167 if (__le32_to_cpu(dev->size_high) > 0)
2168 mpb->attributes |= MPB_ATTRIB_2TB;
2169
2170 /* FIXME detect when an array spans a port multiplier */
2171 #if 0
2172 mpb->attributes |= MPB_ATTRIB_PM;
2173 #endif
2174
2175 if (mpb->num_raid_devs > 1 ||
2176 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2177 version = MPB_VERSION_ATTRIBS;
2178 switch (get_imsm_raid_level(map)) {
2179 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2180 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2181 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2182 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2183 }
2184 } else {
2185 if (map->num_members >= 5)
2186 version = MPB_VERSION_5OR6_DISK_ARRAY;
2187 else if (dev->status == DEV_CLONE_N_GO)
2188 version = MPB_VERSION_CNG;
2189 else if (get_imsm_raid_level(map) == 5)
2190 version = MPB_VERSION_RAID5;
2191 else if (map->num_members >= 3)
2192 version = MPB_VERSION_3OR4_DISK_ARRAY;
2193 else if (get_imsm_raid_level(map) == 1)
2194 version = MPB_VERSION_RAID1;
2195 else
2196 version = MPB_VERSION_RAID0;
2197 }
2198 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2199 }
2200 }
2201
2202 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2203 unsigned long long size, char *name,
2204 char *homehost, int *uuid)
2205 {
2206 /* We are creating a volume inside a pre-existing container.
2207 * so st->sb is already set.
2208 */
2209 struct intel_super *super = st->sb;
2210 struct imsm_super *mpb = super->anchor;
2211 struct intel_dev *dv;
2212 struct imsm_dev *dev;
2213 struct imsm_vol *vol;
2214 struct imsm_map *map;
2215 int idx = mpb->num_raid_devs;
2216 int i;
2217 unsigned long long array_blocks;
2218 size_t size_old, size_new;
2219
2220 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2221 fprintf(stderr, Name": This imsm-container already has the "
2222 "maximum of %d volumes\n", super->orom->vpa);
2223 return 0;
2224 }
2225
2226 /* ensure the mpb is large enough for the new data */
2227 size_old = __le32_to_cpu(mpb->mpb_size);
2228 size_new = disks_to_mpb_size(info->nr_disks);
2229 if (size_new > size_old) {
2230 void *mpb_new;
2231 size_t size_round = ROUND_UP(size_new, 512);
2232
2233 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2234 fprintf(stderr, Name": could not allocate new mpb\n");
2235 return 0;
2236 }
2237 memcpy(mpb_new, mpb, size_old);
2238 free(mpb);
2239 mpb = mpb_new;
2240 super->anchor = mpb_new;
2241 mpb->mpb_size = __cpu_to_le32(size_new);
2242 memset(mpb_new + size_old, 0, size_round - size_old);
2243 }
2244 super->current_vol = idx;
2245 /* when creating the first raid device in this container set num_disks
2246 * to zero, i.e. delete this spare and add raid member devices in
2247 * add_to_super_imsm_volume()
2248 */
2249 if (super->current_vol == 0)
2250 mpb->num_disks = 0;
2251
2252 for (i = 0; i < super->current_vol; i++) {
2253 dev = get_imsm_dev(super, i);
2254 if (strncmp((char *) dev->volume, name,
2255 MAX_RAID_SERIAL_LEN) == 0) {
2256 fprintf(stderr, Name": '%s' is already defined for this container\n",
2257 name);
2258 return 0;
2259 }
2260 }
2261
2262 sprintf(st->subarray, "%d", idx);
2263 dv = malloc(sizeof(*dv));
2264 if (!dv) {
2265 fprintf(stderr, Name ": failed to allocate device list entry\n");
2266 return 0;
2267 }
2268 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2269 if (!dev) {
2270 free(dv);
2271 fprintf(stderr, Name": could not allocate raid device\n");
2272 return 0;
2273 }
2274 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2275 if (info->level == 1)
2276 array_blocks = info_to_blocks_per_member(info);
2277 else
2278 array_blocks = calc_array_size(info->level, info->raid_disks,
2279 info->layout, info->chunk_size,
2280 info->size*2);
2281 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2282 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2283 dev->status = __cpu_to_le32(0);
2284 dev->reserved_blocks = __cpu_to_le32(0);
2285 vol = &dev->vol;
2286 vol->migr_state = 0;
2287 vol->migr_type = MIGR_INIT;
2288 vol->dirty = 0;
2289 vol->curr_migr_unit = 0;
2290 map = get_imsm_map(dev, 0);
2291 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2292 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2293 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2294 map->num_data_stripes = __cpu_to_le32(info_to_num_data_stripes(info));
2295 map->failed_disk_num = ~0;
2296 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2297 IMSM_T_STATE_NORMAL;
2298
2299 if (info->level == 1 && info->raid_disks > 2) {
2300 fprintf(stderr, Name": imsm does not support more than 2 disks"
2301 "in a raid1 volume\n");
2302 return 0;
2303 }
2304 if (info->level == 10) {
2305 map->raid_level = 1;
2306 map->num_domains = info->raid_disks / 2;
2307 } else {
2308 map->raid_level = info->level;
2309 map->num_domains = !!map->raid_level;
2310 }
2311
2312 map->num_members = info->raid_disks;
2313 for (i = 0; i < map->num_members; i++) {
2314 /* initialized in add_to_super */
2315 set_imsm_ord_tbl_ent(map, i, 0);
2316 }
2317 mpb->num_raid_devs++;
2318
2319 dv->dev = dev;
2320 dv->index = super->current_vol;
2321 dv->next = super->devlist;
2322 super->devlist = dv;
2323
2324 imsm_update_version_info(super);
2325
2326 return 1;
2327 }
2328
2329 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2330 unsigned long long size, char *name,
2331 char *homehost, int *uuid)
2332 {
2333 /* This is primarily called by Create when creating a new array.
2334 * We will then get add_to_super called for each component, and then
2335 * write_init_super called to write it out to each device.
2336 * For IMSM, Create can create on fresh devices or on a pre-existing
2337 * array.
2338 * To create on a pre-existing array a different method will be called.
2339 * This one is just for fresh drives.
2340 */
2341 struct intel_super *super;
2342 struct imsm_super *mpb;
2343 size_t mpb_size;
2344 char *version;
2345
2346 if (!info) {
2347 st->sb = NULL;
2348 return 0;
2349 }
2350 if (st->sb)
2351 return init_super_imsm_volume(st, info, size, name, homehost,
2352 uuid);
2353
2354 super = alloc_super(1);
2355 if (!super)
2356 return 0;
2357 mpb_size = disks_to_mpb_size(info->nr_disks);
2358 if (posix_memalign(&super->buf, 512, mpb_size) != 0) {
2359 free(super);
2360 return 0;
2361 }
2362 mpb = super->buf;
2363 memset(mpb, 0, mpb_size);
2364
2365 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2366
2367 version = (char *) mpb->sig;
2368 strcpy(version, MPB_SIGNATURE);
2369 version += strlen(MPB_SIGNATURE);
2370 strcpy(version, MPB_VERSION_RAID0);
2371 mpb->mpb_size = mpb_size;
2372
2373 st->sb = super;
2374 return 1;
2375 }
2376
2377 #ifndef MDASSEMBLE
2378 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2379 int fd, char *devname)
2380 {
2381 struct intel_super *super = st->sb;
2382 struct imsm_super *mpb = super->anchor;
2383 struct dl *dl;
2384 struct imsm_dev *dev;
2385 struct imsm_map *map;
2386
2387 dev = get_imsm_dev(super, super->current_vol);
2388 map = get_imsm_map(dev, 0);
2389
2390 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2391 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2392 devname);
2393 return 1;
2394 }
2395
2396 for (dl = super->disks; dl ; dl = dl->next)
2397 if (dl->major == dk->major &&
2398 dl->minor == dk->minor)
2399 break;
2400
2401 if (!dl) {
2402 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2403 return 1;
2404 }
2405
2406 /* add a pristine spare to the metadata */
2407 if (dl->index < 0) {
2408 dl->index = super->anchor->num_disks;
2409 super->anchor->num_disks++;
2410 }
2411 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2412 dl->disk.status = CONFIGURED_DISK | USABLE_DISK;
2413
2414 /* if we are creating the first raid device update the family number */
2415 if (super->current_vol == 0) {
2416 __u32 sum;
2417 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2418 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2419
2420 *_dev = *dev;
2421 *_disk = dl->disk;
2422 sum = __gen_imsm_checksum(mpb);
2423 mpb->family_num = __cpu_to_le32(sum);
2424 }
2425
2426 return 0;
2427 }
2428
2429 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2430 int fd, char *devname)
2431 {
2432 struct intel_super *super = st->sb;
2433 struct dl *dd;
2434 unsigned long long size;
2435 __u32 id;
2436 int rv;
2437 struct stat stb;
2438
2439 /* if we are on an RAID enabled platform check that the disk is
2440 * attached to the raid controller
2441 */
2442 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2443 fprintf(stderr,
2444 Name ": %s is not attached to the raid controller: %s\n",
2445 devname ? : "disk", super->hba);
2446 return 1;
2447 }
2448
2449 if (super->current_vol >= 0)
2450 return add_to_super_imsm_volume(st, dk, fd, devname);
2451
2452 fstat(fd, &stb);
2453 dd = malloc(sizeof(*dd));
2454 if (!dd) {
2455 fprintf(stderr,
2456 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2457 return 1;
2458 }
2459 memset(dd, 0, sizeof(*dd));
2460 dd->major = major(stb.st_rdev);
2461 dd->minor = minor(stb.st_rdev);
2462 dd->index = -1;
2463 dd->devname = devname ? strdup(devname) : NULL;
2464 dd->fd = fd;
2465 dd->e = NULL;
2466 rv = imsm_read_serial(fd, devname, dd->serial);
2467 if (rv) {
2468 fprintf(stderr,
2469 Name ": failed to retrieve scsi serial, aborting\n");
2470 free(dd);
2471 abort();
2472 }
2473
2474 get_dev_size(fd, NULL, &size);
2475 size /= 512;
2476 serialcpy(dd->disk.serial, dd->serial);
2477 dd->disk.total_blocks = __cpu_to_le32(size);
2478 dd->disk.status = USABLE_DISK | SPARE_DISK;
2479 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
2480 dd->disk.scsi_id = __cpu_to_le32(id);
2481 else
2482 dd->disk.scsi_id = __cpu_to_le32(0);
2483
2484 if (st->update_tail) {
2485 dd->next = super->add;
2486 super->add = dd;
2487 } else {
2488 dd->next = super->disks;
2489 super->disks = dd;
2490 }
2491
2492 return 0;
2493 }
2494
2495 static int store_imsm_mpb(int fd, struct intel_super *super);
2496
2497 /* spare records have their own family number and do not have any defined raid
2498 * devices
2499 */
2500 static int write_super_imsm_spares(struct intel_super *super, int doclose)
2501 {
2502 struct imsm_super mpb_save;
2503 struct imsm_super *mpb = super->anchor;
2504 __u32 sum;
2505 struct dl *d;
2506
2507 mpb_save = *mpb;
2508 mpb->num_raid_devs = 0;
2509 mpb->num_disks = 1;
2510 mpb->mpb_size = sizeof(struct imsm_super);
2511 mpb->generation_num = __cpu_to_le32(1UL);
2512
2513 for (d = super->disks; d; d = d->next) {
2514 if (d->index != -1)
2515 continue;
2516
2517 mpb->disk[0] = d->disk;
2518 sum = __gen_imsm_checksum(mpb);
2519 mpb->family_num = __cpu_to_le32(sum);
2520 sum = __gen_imsm_checksum(mpb);
2521 mpb->check_sum = __cpu_to_le32(sum);
2522
2523 if (store_imsm_mpb(d->fd, super)) {
2524 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2525 __func__, d->major, d->minor, strerror(errno));
2526 *mpb = mpb_save;
2527 return 1;
2528 }
2529 if (doclose) {
2530 close(d->fd);
2531 d->fd = -1;
2532 }
2533 }
2534
2535 *mpb = mpb_save;
2536 return 0;
2537 }
2538
2539 static int write_super_imsm(struct intel_super *super, int doclose)
2540 {
2541 struct imsm_super *mpb = super->anchor;
2542 struct dl *d;
2543 __u32 generation;
2544 __u32 sum;
2545 int spares = 0;
2546 int i;
2547 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
2548
2549 /* 'generation' is incremented everytime the metadata is written */
2550 generation = __le32_to_cpu(mpb->generation_num);
2551 generation++;
2552 mpb->generation_num = __cpu_to_le32(generation);
2553
2554 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
2555 for (d = super->disks; d; d = d->next) {
2556 if (d->index == -1)
2557 spares++;
2558 else
2559 mpb->disk[d->index] = d->disk;
2560 }
2561 for (d = super->missing; d; d = d->next)
2562 mpb->disk[d->index] = d->disk;
2563
2564 for (i = 0; i < mpb->num_raid_devs; i++) {
2565 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
2566
2567 imsm_copy_dev(dev, get_imsm_dev(super, i));
2568 mpb_size += sizeof_imsm_dev(dev, 0);
2569 }
2570 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
2571 mpb->mpb_size = __cpu_to_le32(mpb_size);
2572
2573 /* recalculate checksum */
2574 sum = __gen_imsm_checksum(mpb);
2575 mpb->check_sum = __cpu_to_le32(sum);
2576
2577 /* write the mpb for disks that compose raid devices */
2578 for (d = super->disks; d ; d = d->next) {
2579 if (d->index < 0)
2580 continue;
2581 if (store_imsm_mpb(d->fd, super))
2582 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2583 __func__, d->major, d->minor, strerror(errno));
2584 if (doclose) {
2585 close(d->fd);
2586 d->fd = -1;
2587 }
2588 }
2589
2590 if (spares)
2591 return write_super_imsm_spares(super, doclose);
2592
2593 return 0;
2594 }
2595
2596
2597 static int create_array(struct supertype *st)
2598 {
2599 size_t len;
2600 struct imsm_update_create_array *u;
2601 struct intel_super *super = st->sb;
2602 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
2603 struct imsm_map *map = get_imsm_map(dev, 0);
2604 struct disk_info *inf;
2605 struct imsm_disk *disk;
2606 int i;
2607 int idx;
2608
2609 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
2610 sizeof(*inf) * map->num_members;
2611 u = malloc(len);
2612 if (!u) {
2613 fprintf(stderr, "%s: failed to allocate update buffer\n",
2614 __func__);
2615 return 1;
2616 }
2617
2618 u->type = update_create_array;
2619 u->dev_idx = super->current_vol;
2620 imsm_copy_dev(&u->dev, dev);
2621 inf = get_disk_info(u);
2622 for (i = 0; i < map->num_members; i++) {
2623 idx = get_imsm_disk_idx(dev, i);
2624 disk = get_imsm_disk(super, idx);
2625 serialcpy(inf[i].serial, disk->serial);
2626 }
2627 append_metadata_update(st, u, len);
2628
2629 return 0;
2630 }
2631
2632 static int _add_disk(struct supertype *st)
2633 {
2634 struct intel_super *super = st->sb;
2635 size_t len;
2636 struct imsm_update_add_disk *u;
2637
2638 if (!super->add)
2639 return 0;
2640
2641 len = sizeof(*u);
2642 u = malloc(len);
2643 if (!u) {
2644 fprintf(stderr, "%s: failed to allocate update buffer\n",
2645 __func__);
2646 return 1;
2647 }
2648
2649 u->type = update_add_disk;
2650 append_metadata_update(st, u, len);
2651
2652 return 0;
2653 }
2654
2655 static int write_init_super_imsm(struct supertype *st)
2656 {
2657 if (st->update_tail) {
2658 /* queue the recently created array / added disk
2659 * as a metadata update */
2660 struct intel_super *super = st->sb;
2661 struct dl *d;
2662 int rv;
2663
2664 /* determine if we are creating a volume or adding a disk */
2665 if (super->current_vol < 0) {
2666 /* in the add disk case we are running in mdmon
2667 * context, so don't close fd's
2668 */
2669 return _add_disk(st);
2670 } else
2671 rv = create_array(st);
2672
2673 for (d = super->disks; d ; d = d->next) {
2674 close(d->fd);
2675 d->fd = -1;
2676 }
2677
2678 return rv;
2679 } else
2680 return write_super_imsm(st->sb, 1);
2681 }
2682 #endif
2683
2684 static int store_zero_imsm(struct supertype *st, int fd)
2685 {
2686 unsigned long long dsize;
2687 void *buf;
2688
2689 get_dev_size(fd, NULL, &dsize);
2690
2691 /* first block is stored on second to last sector of the disk */
2692 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
2693 return 1;
2694
2695 if (posix_memalign(&buf, 512, 512) != 0)
2696 return 1;
2697
2698 memset(buf, 0, 512);
2699 if (write(fd, buf, 512) != 512)
2700 return 1;
2701 return 0;
2702 }
2703
2704 static int imsm_bbm_log_size(struct imsm_super *mpb)
2705 {
2706 return __le32_to_cpu(mpb->bbm_log_size);
2707 }
2708
2709 #ifndef MDASSEMBLE
2710 static int validate_geometry_imsm_container(struct supertype *st, int level,
2711 int layout, int raiddisks, int chunk,
2712 unsigned long long size, char *dev,
2713 unsigned long long *freesize,
2714 int verbose)
2715 {
2716 int fd;
2717 unsigned long long ldsize;
2718 const struct imsm_orom *orom;
2719
2720 if (level != LEVEL_CONTAINER)
2721 return 0;
2722 if (!dev)
2723 return 1;
2724
2725 if (check_env("IMSM_NO_PLATFORM"))
2726 orom = NULL;
2727 else
2728 orom = find_imsm_orom();
2729 if (orom && raiddisks > orom->tds) {
2730 if (verbose)
2731 fprintf(stderr, Name ": %d exceeds maximum number of"
2732 " platform supported disks: %d\n",
2733 raiddisks, orom->tds);
2734 return 0;
2735 }
2736
2737 fd = open(dev, O_RDONLY|O_EXCL, 0);
2738 if (fd < 0) {
2739 if (verbose)
2740 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
2741 dev, strerror(errno));
2742 return 0;
2743 }
2744 if (!get_dev_size(fd, dev, &ldsize)) {
2745 close(fd);
2746 return 0;
2747 }
2748 close(fd);
2749
2750 *freesize = avail_size_imsm(st, ldsize >> 9);
2751
2752 return 1;
2753 }
2754
2755 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
2756 {
2757 const unsigned long long base_start = e[*idx].start;
2758 unsigned long long end = base_start + e[*idx].size;
2759 int i;
2760
2761 if (base_start == end)
2762 return 0;
2763
2764 *idx = *idx + 1;
2765 for (i = *idx; i < num_extents; i++) {
2766 /* extend overlapping extents */
2767 if (e[i].start >= base_start &&
2768 e[i].start <= end) {
2769 if (e[i].size == 0)
2770 return 0;
2771 if (e[i].start + e[i].size > end)
2772 end = e[i].start + e[i].size;
2773 } else if (e[i].start > end) {
2774 *idx = i;
2775 break;
2776 }
2777 }
2778
2779 return end - base_start;
2780 }
2781
2782 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
2783 {
2784 /* build a composite disk with all known extents and generate a new
2785 * 'maxsize' given the "all disks in an array must share a common start
2786 * offset" constraint
2787 */
2788 struct extent *e = calloc(sum_extents, sizeof(*e));
2789 struct dl *dl;
2790 int i, j;
2791 int start_extent;
2792 unsigned long long pos;
2793 unsigned long long start;
2794 unsigned long long maxsize;
2795 unsigned long reserve;
2796
2797 if (!e)
2798 return ~0ULL; /* error */
2799
2800 /* coalesce and sort all extents. also, check to see if we need to
2801 * reserve space between member arrays
2802 */
2803 j = 0;
2804 for (dl = super->disks; dl; dl = dl->next) {
2805 if (!dl->e)
2806 continue;
2807 for (i = 0; i < dl->extent_cnt; i++)
2808 e[j++] = dl->e[i];
2809 }
2810 qsort(e, sum_extents, sizeof(*e), cmp_extent);
2811
2812 /* merge extents */
2813 i = 0;
2814 j = 0;
2815 while (i < sum_extents) {
2816 e[j].start = e[i].start;
2817 e[j].size = find_size(e, &i, sum_extents);
2818 j++;
2819 if (e[j-1].size == 0)
2820 break;
2821 }
2822
2823 pos = 0;
2824 maxsize = 0;
2825 start_extent = 0;
2826 i = 0;
2827 do {
2828 unsigned long long esize;
2829
2830 esize = e[i].start - pos;
2831 if (esize >= maxsize) {
2832 maxsize = esize;
2833 start = pos;
2834 start_extent = i;
2835 }
2836 pos = e[i].start + e[i].size;
2837 i++;
2838 } while (e[i-1].size);
2839 free(e);
2840
2841 if (start_extent > 0)
2842 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
2843 else
2844 reserve = 0;
2845
2846 if (maxsize < reserve)
2847 return ~0ULL;
2848
2849 super->create_offset = ~((__u32) 0);
2850 if (start + reserve > super->create_offset)
2851 return ~0ULL; /* start overflows create_offset */
2852 super->create_offset = start + reserve;
2853
2854 return maxsize - reserve;
2855 }
2856
2857 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
2858 {
2859 if (level < 0 || level == 6 || level == 4)
2860 return 0;
2861
2862 /* if we have an orom prevent invalid raid levels */
2863 if (orom)
2864 switch (level) {
2865 case 0: return imsm_orom_has_raid0(orom);
2866 case 1:
2867 if (raiddisks > 2)
2868 return imsm_orom_has_raid1e(orom);
2869 return imsm_orom_has_raid1(orom) && raiddisks == 2;
2870 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
2871 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
2872 }
2873 else
2874 return 1; /* not on an Intel RAID platform so anything goes */
2875
2876 return 0;
2877 }
2878
2879 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2880 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2881 * FIX ME add ahci details
2882 */
2883 static int validate_geometry_imsm_volume(struct supertype *st, int level,
2884 int layout, int raiddisks, int chunk,
2885 unsigned long long size, char *dev,
2886 unsigned long long *freesize,
2887 int verbose)
2888 {
2889 struct stat stb;
2890 struct intel_super *super = st->sb;
2891 struct imsm_super *mpb = super->anchor;
2892 struct dl *dl;
2893 unsigned long long pos = 0;
2894 unsigned long long maxsize;
2895 struct extent *e;
2896 int i;
2897
2898 /* We must have the container info already read in. */
2899 if (!super)
2900 return 0;
2901
2902 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
2903 pr_vrb(": platform does not support raid%d with %d disk%s\n",
2904 level, raiddisks, raiddisks > 1 ? "s" : "");
2905 return 0;
2906 }
2907 if (super->orom && level != 1 &&
2908 !imsm_orom_has_chunk(super->orom, chunk)) {
2909 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
2910 return 0;
2911 }
2912 if (layout != imsm_level_to_layout(level)) {
2913 if (level == 5)
2914 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
2915 else if (level == 10)
2916 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
2917 else
2918 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
2919 layout, level);
2920 return 0;
2921 }
2922
2923 if (!dev) {
2924 /* General test: make sure there is space for
2925 * 'raiddisks' device extents of size 'size' at a given
2926 * offset
2927 */
2928 unsigned long long minsize = size;
2929 unsigned long long start_offset = ~0ULL;
2930 int dcnt = 0;
2931 if (minsize == 0)
2932 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
2933 for (dl = super->disks; dl ; dl = dl->next) {
2934 int found = 0;
2935
2936 pos = 0;
2937 i = 0;
2938 e = get_extents(super, dl);
2939 if (!e) continue;
2940 do {
2941 unsigned long long esize;
2942 esize = e[i].start - pos;
2943 if (esize >= minsize)
2944 found = 1;
2945 if (found && start_offset == ~0ULL) {
2946 start_offset = pos;
2947 break;
2948 } else if (found && pos != start_offset) {
2949 found = 0;
2950 break;
2951 }
2952 pos = e[i].start + e[i].size;
2953 i++;
2954 } while (e[i-1].size);
2955 if (found)
2956 dcnt++;
2957 free(e);
2958 }
2959 if (dcnt < raiddisks) {
2960 if (verbose)
2961 fprintf(stderr, Name ": imsm: Not enough "
2962 "devices with space for this array "
2963 "(%d < %d)\n",
2964 dcnt, raiddisks);
2965 return 0;
2966 }
2967 return 1;
2968 }
2969
2970 /* This device must be a member of the set */
2971 if (stat(dev, &stb) < 0)
2972 return 0;
2973 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2974 return 0;
2975 for (dl = super->disks ; dl ; dl = dl->next) {
2976 if (dl->major == major(stb.st_rdev) &&
2977 dl->minor == minor(stb.st_rdev))
2978 break;
2979 }
2980 if (!dl) {
2981 if (verbose)
2982 fprintf(stderr, Name ": %s is not in the "
2983 "same imsm set\n", dev);
2984 return 0;
2985 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
2986 /* If a volume is present then the current creation attempt
2987 * cannot incorporate new spares because the orom may not
2988 * understand this configuration (all member disks must be
2989 * members of each array in the container).
2990 */
2991 fprintf(stderr, Name ": %s is a spare and a volume"
2992 " is already defined for this container\n", dev);
2993 fprintf(stderr, Name ": The option-rom requires all member"
2994 " disks to be a member of all volumes\n");
2995 return 0;
2996 }
2997
2998 /* retrieve the largest free space block */
2999 e = get_extents(super, dl);
3000 maxsize = 0;
3001 i = 0;
3002 if (e) {
3003 do {
3004 unsigned long long esize;
3005
3006 esize = e[i].start - pos;
3007 if (esize >= maxsize)
3008 maxsize = esize;
3009 pos = e[i].start + e[i].size;
3010 i++;
3011 } while (e[i-1].size);
3012 dl->e = e;
3013 dl->extent_cnt = i;
3014 } else {
3015 if (verbose)
3016 fprintf(stderr, Name ": unable to determine free space for: %s\n",
3017 dev);
3018 return 0;
3019 }
3020 if (maxsize < size) {
3021 if (verbose)
3022 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
3023 dev, maxsize, size);
3024 return 0;
3025 }
3026
3027 /* count total number of extents for merge */
3028 i = 0;
3029 for (dl = super->disks; dl; dl = dl->next)
3030 if (dl->e)
3031 i += dl->extent_cnt;
3032
3033 maxsize = merge_extents(super, i);
3034 if (maxsize < size) {
3035 if (verbose)
3036 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
3037 maxsize, size);
3038 return 0;
3039 } else if (maxsize == ~0ULL) {
3040 if (verbose)
3041 fprintf(stderr, Name ": failed to merge %d extents\n", i);
3042 return 0;
3043 }
3044
3045 *freesize = maxsize;
3046
3047 return 1;
3048 }
3049
3050 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
3051 int raiddisks, int chunk, unsigned long long size,
3052 char *dev, unsigned long long *freesize,
3053 int verbose)
3054 {
3055 int fd, cfd;
3056 struct mdinfo *sra;
3057
3058 /* if given unused devices create a container
3059 * if given given devices in a container create a member volume
3060 */
3061 if (level == LEVEL_CONTAINER) {
3062 /* Must be a fresh device to add to a container */
3063 return validate_geometry_imsm_container(st, level, layout,
3064 raiddisks, chunk, size,
3065 dev, freesize,
3066 verbose);
3067 }
3068
3069 if (!dev) {
3070 if (st->sb && freesize) {
3071 /* Should do auto-layout here */
3072 fprintf(stderr, Name ": IMSM does not support auto-layout yet\n");
3073 return 0;
3074 }
3075 return 1;
3076 }
3077 if (st->sb) {
3078 /* creating in a given container */
3079 return validate_geometry_imsm_volume(st, level, layout,
3080 raiddisks, chunk, size,
3081 dev, freesize, verbose);
3082 }
3083
3084 /* limit creation to the following levels */
3085 if (!dev)
3086 switch (level) {
3087 case 0:
3088 case 1:
3089 case 10:
3090 case 5:
3091 break;
3092 default:
3093 return 1;
3094 }
3095
3096 /* This device needs to be a device in an 'imsm' container */
3097 fd = open(dev, O_RDONLY|O_EXCL, 0);
3098 if (fd >= 0) {
3099 if (verbose)
3100 fprintf(stderr,
3101 Name ": Cannot create this array on device %s\n",
3102 dev);
3103 close(fd);
3104 return 0;
3105 }
3106 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3107 if (verbose)
3108 fprintf(stderr, Name ": Cannot open %s: %s\n",
3109 dev, strerror(errno));
3110 return 0;
3111 }
3112 /* Well, it is in use by someone, maybe an 'imsm' container. */
3113 cfd = open_container(fd);
3114 if (cfd < 0) {
3115 close(fd);
3116 if (verbose)
3117 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3118 dev);
3119 return 0;
3120 }
3121 sra = sysfs_read(cfd, 0, GET_VERSION);
3122 close(fd);
3123 if (sra && sra->array.major_version == -1 &&
3124 strcmp(sra->text_version, "imsm") == 0) {
3125 /* This is a member of a imsm container. Load the container
3126 * and try to create a volume
3127 */
3128 struct intel_super *super;
3129
3130 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3131 st->sb = super;
3132 st->container_dev = fd2devnum(cfd);
3133 close(cfd);
3134 return validate_geometry_imsm_volume(st, level, layout,
3135 raiddisks, chunk,
3136 size, dev,
3137 freesize, verbose);
3138 }
3139 close(cfd);
3140 } else /* may belong to another container */
3141 return 0;
3142
3143 return 1;
3144 }
3145 #endif /* MDASSEMBLE */
3146
3147 static struct mdinfo *container_content_imsm(struct supertype *st)
3148 {
3149 /* Given a container loaded by load_super_imsm_all,
3150 * extract information about all the arrays into
3151 * an mdinfo tree.
3152 *
3153 * For each imsm_dev create an mdinfo, fill it in,
3154 * then look for matching devices in super->disks
3155 * and create appropriate device mdinfo.
3156 */
3157 struct intel_super *super = st->sb;
3158 struct imsm_super *mpb = super->anchor;
3159 struct mdinfo *rest = NULL;
3160 int i;
3161
3162 /* do not assemble arrays that might have bad blocks */
3163 if (imsm_bbm_log_size(super->anchor)) {
3164 fprintf(stderr, Name ": BBM log found in metadata. "
3165 "Cannot activate array(s).\n");
3166 return NULL;
3167 }
3168
3169 for (i = 0; i < mpb->num_raid_devs; i++) {
3170 struct imsm_dev *dev = get_imsm_dev(super, i);
3171 struct imsm_map *map = get_imsm_map(dev, 0);
3172 struct mdinfo *this;
3173 int slot;
3174
3175 this = malloc(sizeof(*this));
3176 memset(this, 0, sizeof(*this));
3177 this->next = rest;
3178
3179 super->current_vol = i;
3180 getinfo_super_imsm_volume(st, this);
3181 for (slot = 0 ; slot < map->num_members; slot++) {
3182 struct mdinfo *info_d;
3183 struct dl *d;
3184 int idx;
3185 int skip;
3186 __u32 s;
3187 __u32 ord;
3188
3189 skip = 0;
3190 idx = get_imsm_disk_idx(dev, slot);
3191 ord = get_imsm_ord_tbl_ent(dev, slot);
3192 for (d = super->disks; d ; d = d->next)
3193 if (d->index == idx)
3194 break;
3195
3196 if (d == NULL)
3197 skip = 1;
3198
3199 s = d ? d->disk.status : 0;
3200 if (s & FAILED_DISK)
3201 skip = 1;
3202 if (!(s & USABLE_DISK))
3203 skip = 1;
3204 if (ord & IMSM_ORD_REBUILD)
3205 skip = 1;
3206
3207 /*
3208 * if we skip some disks the array will be assmebled degraded;
3209 * reset resync start to avoid a dirty-degraded situation
3210 *
3211 * FIXME handle dirty degraded
3212 */
3213 if (skip && !dev->vol.dirty)
3214 this->resync_start = ~0ULL;
3215 if (skip)
3216 continue;
3217
3218 info_d = malloc(sizeof(*info_d));
3219 if (!info_d) {
3220 fprintf(stderr, Name ": failed to allocate disk"
3221 " for volume %s\n", (char *) dev->volume);
3222 free(this);
3223 this = rest;
3224 break;
3225 }
3226 memset(info_d, 0, sizeof(*info_d));
3227 info_d->next = this->devs;
3228 this->devs = info_d;
3229
3230 info_d->disk.number = d->index;
3231 info_d->disk.major = d->major;
3232 info_d->disk.minor = d->minor;
3233 info_d->disk.raid_disk = slot;
3234
3235 this->array.working_disks++;
3236
3237 info_d->events = __le32_to_cpu(mpb->generation_num);
3238 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3239 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3240 if (d->devname)
3241 strcpy(info_d->name, d->devname);
3242 }
3243 rest = this;
3244 }
3245
3246 return rest;
3247 }
3248
3249
3250 #ifndef MDASSEMBLE
3251 static int imsm_open_new(struct supertype *c, struct active_array *a,
3252 char *inst)
3253 {
3254 struct intel_super *super = c->sb;
3255 struct imsm_super *mpb = super->anchor;
3256
3257 if (atoi(inst) >= mpb->num_raid_devs) {
3258 fprintf(stderr, "%s: subarry index %d, out of range\n",
3259 __func__, atoi(inst));
3260 return -ENODEV;
3261 }
3262
3263 dprintf("imsm: open_new %s\n", inst);
3264 a->info.container_member = atoi(inst);
3265 return 0;
3266 }
3267
3268 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3269 {
3270 struct imsm_map *map = get_imsm_map(dev, 0);
3271
3272 if (!failed)
3273 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3274 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3275
3276 switch (get_imsm_raid_level(map)) {
3277 case 0:
3278 return IMSM_T_STATE_FAILED;
3279 break;
3280 case 1:
3281 if (failed < map->num_members)
3282 return IMSM_T_STATE_DEGRADED;
3283 else
3284 return IMSM_T_STATE_FAILED;
3285 break;
3286 case 10:
3287 {
3288 /**
3289 * check to see if any mirrors have failed, otherwise we
3290 * are degraded. Even numbered slots are mirrored on
3291 * slot+1
3292 */
3293 int i;
3294 /* gcc -Os complains that this is unused */
3295 int insync = insync;
3296
3297 for (i = 0; i < map->num_members; i++) {
3298 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3299 int idx = ord_to_idx(ord);
3300 struct imsm_disk *disk;
3301
3302 /* reset the potential in-sync count on even-numbered
3303 * slots. num_copies is always 2 for imsm raid10
3304 */
3305 if ((i & 1) == 0)
3306 insync = 2;
3307
3308 disk = get_imsm_disk(super, idx);
3309 if (!disk || disk->status & FAILED_DISK ||
3310 ord & IMSM_ORD_REBUILD)
3311 insync--;
3312
3313 /* no in-sync disks left in this mirror the
3314 * array has failed
3315 */
3316 if (insync == 0)
3317 return IMSM_T_STATE_FAILED;
3318 }
3319
3320 return IMSM_T_STATE_DEGRADED;
3321 }
3322 case 5:
3323 if (failed < 2)
3324 return IMSM_T_STATE_DEGRADED;
3325 else
3326 return IMSM_T_STATE_FAILED;
3327 break;
3328 default:
3329 break;
3330 }
3331
3332 return map->map_state;
3333 }
3334
3335 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3336 {
3337 int i;
3338 int failed = 0;
3339 struct imsm_disk *disk;
3340 struct imsm_map *map = get_imsm_map(dev, 0);
3341 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3342 __u32 ord;
3343 int idx;
3344
3345 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3346 * disks that are being rebuilt. New failures are recorded to
3347 * map[0]. So we look through all the disks we started with and
3348 * see if any failures are still present, or if any new ones
3349 * have arrived
3350 *
3351 * FIXME add support for online capacity expansion and
3352 * raid-level-migration
3353 */
3354 for (i = 0; i < prev->num_members; i++) {
3355 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
3356 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
3357 idx = ord_to_idx(ord);
3358
3359 disk = get_imsm_disk(super, idx);
3360 if (!disk || disk->status & FAILED_DISK ||
3361 ord & IMSM_ORD_REBUILD)
3362 failed++;
3363 }
3364
3365 return failed;
3366 }
3367
3368 static int is_resyncing(struct imsm_dev *dev)
3369 {
3370 struct imsm_map *migr_map;
3371
3372 if (!dev->vol.migr_state)
3373 return 0;
3374
3375 if (dev->vol.migr_type == MIGR_INIT)
3376 return 1;
3377
3378 migr_map = get_imsm_map(dev, 1);
3379
3380 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
3381 return 1;
3382 else
3383 return 0;
3384 }
3385
3386 static int is_rebuilding(struct imsm_dev *dev)
3387 {
3388 struct imsm_map *migr_map;
3389
3390 if (!dev->vol.migr_state)
3391 return 0;
3392
3393 if (dev->vol.migr_type != MIGR_REBUILD)
3394 return 0;
3395
3396 migr_map = get_imsm_map(dev, 1);
3397
3398 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
3399 return 1;
3400 else
3401 return 0;
3402 }
3403
3404 /* return true if we recorded new information */
3405 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
3406 {
3407 __u32 ord;
3408 int slot;
3409 struct imsm_map *map;
3410
3411 /* new failures are always set in map[0] */
3412 map = get_imsm_map(dev, 0);
3413
3414 slot = get_imsm_disk_slot(map, idx);
3415 if (slot < 0)
3416 return 0;
3417
3418 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
3419 if ((disk->status & FAILED_DISK) && (ord & IMSM_ORD_REBUILD))
3420 return 0;
3421
3422 disk->status |= FAILED_DISK;
3423 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
3424 if (map->failed_disk_num == ~0)
3425 map->failed_disk_num = slot;
3426 return 1;
3427 }
3428
3429 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
3430 {
3431 mark_failure(dev, disk, idx);
3432
3433 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
3434 return;
3435
3436 disk->scsi_id = __cpu_to_le32(~(__u32)0);
3437 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
3438 }
3439
3440 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3441 * states are handled in imsm_set_disk() with one exception, when a
3442 * resync is stopped due to a new failure this routine will set the
3443 * 'degraded' state for the array.
3444 */
3445 static int imsm_set_array_state(struct active_array *a, int consistent)
3446 {
3447 int inst = a->info.container_member;
3448 struct intel_super *super = a->container->sb;
3449 struct imsm_dev *dev = get_imsm_dev(super, inst);
3450 struct imsm_map *map = get_imsm_map(dev, 0);
3451 int failed = imsm_count_failed(super, dev);
3452 __u8 map_state = imsm_check_degraded(super, dev, failed);
3453
3454 /* before we activate this array handle any missing disks */
3455 if (consistent == 2 && super->missing) {
3456 struct dl *dl;
3457
3458 dprintf("imsm: mark missing\n");
3459 end_migration(dev, map_state);
3460 for (dl = super->missing; dl; dl = dl->next)
3461 mark_missing(dev, &dl->disk, dl->index);
3462 super->updates_pending++;
3463 }
3464
3465 if (consistent == 2 &&
3466 (!is_resync_complete(a) ||
3467 map_state != IMSM_T_STATE_NORMAL ||
3468 dev->vol.migr_state))
3469 consistent = 0;
3470
3471 if (is_resync_complete(a)) {
3472 /* complete intialization / resync,
3473 * recovery and interrupted recovery is completed in
3474 * ->set_disk
3475 */
3476 if (is_resyncing(dev)) {
3477 dprintf("imsm: mark resync done\n");
3478 end_migration(dev, map_state);
3479 super->updates_pending++;
3480 }
3481 } else if (!is_resyncing(dev) && !failed) {
3482 /* mark the start of the init process if nothing is failed */
3483 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
3484 if (map->map_state == IMSM_T_STATE_NORMAL)
3485 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REBUILD);
3486 else
3487 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
3488 super->updates_pending++;
3489 }
3490
3491 /* check if we can update the migration checkpoint */
3492 if (dev->vol.migr_state &&
3493 __le32_to_cpu(dev->vol.curr_migr_unit) != a->resync_start) {
3494 dprintf("imsm: checkpoint migration (%llu)\n", a->resync_start);
3495 dev->vol.curr_migr_unit = __cpu_to_le32(a->resync_start);
3496 super->updates_pending++;
3497 }
3498
3499 /* mark dirty / clean */
3500 if (dev->vol.dirty != !consistent) {
3501 dprintf("imsm: mark '%s' (%llu)\n",
3502 consistent ? "clean" : "dirty", a->resync_start);
3503 if (consistent)
3504 dev->vol.dirty = 0;
3505 else
3506 dev->vol.dirty = 1;
3507 super->updates_pending++;
3508 }
3509 return consistent;
3510 }
3511
3512 static void imsm_set_disk(struct active_array *a, int n, int state)
3513 {
3514 int inst = a->info.container_member;
3515 struct intel_super *super = a->container->sb;
3516 struct imsm_dev *dev = get_imsm_dev(super, inst);
3517 struct imsm_map *map = get_imsm_map(dev, 0);
3518 struct imsm_disk *disk;
3519 int failed;
3520 __u32 ord;
3521 __u8 map_state;
3522
3523 if (n > map->num_members)
3524 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
3525 n, map->num_members - 1);
3526
3527 if (n < 0)
3528 return;
3529
3530 dprintf("imsm: set_disk %d:%x\n", n, state);
3531
3532 ord = get_imsm_ord_tbl_ent(dev, n);
3533 disk = get_imsm_disk(super, ord_to_idx(ord));
3534
3535 /* check for new failures */
3536 if (state & DS_FAULTY) {
3537 if (mark_failure(dev, disk, ord_to_idx(ord)))
3538 super->updates_pending++;
3539 }
3540
3541 /* check if in_sync */
3542 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
3543 struct imsm_map *migr_map = get_imsm_map(dev, 1);
3544
3545 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
3546 super->updates_pending++;
3547 }
3548
3549 failed = imsm_count_failed(super, dev);
3550 map_state = imsm_check_degraded(super, dev, failed);
3551
3552 /* check if recovery complete, newly degraded, or failed */
3553 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
3554 end_migration(dev, map_state);
3555 map = get_imsm_map(dev, 0);
3556 map->failed_disk_num = ~0;
3557 super->updates_pending++;
3558 } else if (map_state == IMSM_T_STATE_DEGRADED &&
3559 map->map_state != map_state &&
3560 !dev->vol.migr_state) {
3561 dprintf("imsm: mark degraded\n");
3562 map->map_state = map_state;
3563 super->updates_pending++;
3564 } else if (map_state == IMSM_T_STATE_FAILED &&
3565 map->map_state != map_state) {
3566 dprintf("imsm: mark failed\n");
3567 end_migration(dev, map_state);
3568 super->updates_pending++;
3569 }
3570 }
3571
3572 static int store_imsm_mpb(int fd, struct intel_super *super)
3573 {
3574 struct imsm_super *mpb = super->anchor;
3575 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
3576 unsigned long long dsize;
3577 unsigned long long sectors;
3578
3579 get_dev_size(fd, NULL, &dsize);
3580
3581 if (mpb_size > 512) {
3582 /* -1 to account for anchor */
3583 sectors = mpb_sectors(mpb) - 1;
3584
3585 /* write the extended mpb to the sectors preceeding the anchor */
3586 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
3587 return 1;
3588
3589 if (write(fd, super->buf + 512, 512 * sectors) != 512 * sectors)
3590 return 1;
3591 }
3592
3593 /* first block is stored on second to last sector of the disk */
3594 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
3595 return 1;
3596
3597 if (write(fd, super->buf, 512) != 512)
3598 return 1;
3599
3600 return 0;
3601 }
3602
3603 static void imsm_sync_metadata(struct supertype *container)
3604 {
3605 struct intel_super *super = container->sb;
3606
3607 if (!super->updates_pending)
3608 return;
3609
3610 write_super_imsm(super, 0);
3611
3612 super->updates_pending = 0;
3613 }
3614
3615 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
3616 {
3617 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3618 int i = get_imsm_disk_idx(dev, idx);
3619 struct dl *dl;
3620
3621 for (dl = super->disks; dl; dl = dl->next)
3622 if (dl->index == i)
3623 break;
3624
3625 if (dl && dl->disk.status & FAILED_DISK)
3626 dl = NULL;
3627
3628 if (dl)
3629 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
3630
3631 return dl;
3632 }
3633
3634 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
3635 struct active_array *a, int activate_new)
3636 {
3637 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3638 int idx = get_imsm_disk_idx(dev, slot);
3639 struct imsm_super *mpb = super->anchor;
3640 struct imsm_map *map;
3641 unsigned long long esize;
3642 unsigned long long pos;
3643 struct mdinfo *d;
3644 struct extent *ex;
3645 int i, j;
3646 int found;
3647 __u32 array_start;
3648 __u32 blocks;
3649 struct dl *dl;
3650
3651 for (dl = super->disks; dl; dl = dl->next) {
3652 /* If in this array, skip */
3653 for (d = a->info.devs ; d ; d = d->next)
3654 if (d->state_fd >= 0 &&
3655 d->disk.major == dl->major &&
3656 d->disk.minor == dl->minor) {
3657 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3658 break;
3659 }
3660 if (d)
3661 continue;
3662
3663 /* skip in use or failed drives */
3664 if (dl->disk.status & FAILED_DISK || idx == dl->index ||
3665 dl->index == -2) {
3666 dprintf("%x:%x status (failed: %d index: %d)\n",
3667 dl->major, dl->minor,
3668 (dl->disk.status & FAILED_DISK) == FAILED_DISK, idx);
3669 continue;
3670 }
3671
3672 /* skip pure spares when we are looking for partially
3673 * assimilated drives
3674 */
3675 if (dl->index == -1 && !activate_new)
3676 continue;
3677
3678 /* Does this unused device have the requisite free space?
3679 * It needs to be able to cover all member volumes
3680 */
3681 ex = get_extents(super, dl);
3682 if (!ex) {
3683 dprintf("cannot get extents\n");
3684 continue;
3685 }
3686 for (i = 0; i < mpb->num_raid_devs; i++) {
3687 dev = get_imsm_dev(super, i);
3688 map = get_imsm_map(dev, 0);
3689
3690 /* check if this disk is already a member of
3691 * this array
3692 */
3693 if (get_imsm_disk_slot(map, dl->index) >= 0)
3694 continue;
3695
3696 found = 0;
3697 j = 0;
3698 pos = 0;
3699 array_start = __le32_to_cpu(map->pba_of_lba0);
3700 blocks = __le32_to_cpu(map->blocks_per_member);
3701
3702 do {
3703 /* check that we can start at pba_of_lba0 with
3704 * blocks_per_member of space
3705 */
3706 esize = ex[j].start - pos;
3707 if (array_start >= pos &&
3708 array_start + blocks < ex[j].start) {
3709 found = 1;
3710 break;
3711 }
3712 pos = ex[j].start + ex[j].size;
3713 j++;
3714 } while (ex[j-1].size);
3715
3716 if (!found)
3717 break;
3718 }
3719
3720 free(ex);
3721 if (i < mpb->num_raid_devs) {
3722 dprintf("%x:%x does not have %u at %u\n",
3723 dl->major, dl->minor,
3724 blocks, array_start);
3725 /* No room */
3726 continue;
3727 }
3728 return dl;
3729 }
3730
3731 return dl;
3732 }
3733
3734 static struct mdinfo *imsm_activate_spare(struct active_array *a,
3735 struct metadata_update **updates)
3736 {
3737 /**
3738 * Find a device with unused free space and use it to replace a
3739 * failed/vacant region in an array. We replace failed regions one a
3740 * array at a time. The result is that a new spare disk will be added
3741 * to the first failed array and after the monitor has finished
3742 * propagating failures the remainder will be consumed.
3743 *
3744 * FIXME add a capability for mdmon to request spares from another
3745 * container.
3746 */
3747
3748 struct intel_super *super = a->container->sb;
3749 int inst = a->info.container_member;
3750 struct imsm_dev *dev = get_imsm_dev(super, inst);
3751 struct imsm_map *map = get_imsm_map(dev, 0);
3752 int failed = a->info.array.raid_disks;
3753 struct mdinfo *rv = NULL;
3754 struct mdinfo *d;
3755 struct mdinfo *di;
3756 struct metadata_update *mu;
3757 struct dl *dl;
3758 struct imsm_update_activate_spare *u;
3759 int num_spares = 0;
3760 int i;
3761
3762 for (d = a->info.devs ; d ; d = d->next) {
3763 if ((d->curr_state & DS_FAULTY) &&
3764 d->state_fd >= 0)
3765 /* wait for Removal to happen */
3766 return NULL;
3767 if (d->state_fd >= 0)
3768 failed--;
3769 }
3770
3771 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3772 inst, failed, a->info.array.raid_disks, a->info.array.level);
3773 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
3774 return NULL;
3775
3776 /* For each slot, if it is not working, find a spare */
3777 for (i = 0; i < a->info.array.raid_disks; i++) {
3778 for (d = a->info.devs ; d ; d = d->next)
3779 if (d->disk.raid_disk == i)
3780 break;
3781 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3782 if (d && (d->state_fd >= 0))
3783 continue;
3784
3785 /*
3786 * OK, this device needs recovery. Try to re-add the
3787 * previous occupant of this slot, if this fails see if
3788 * we can continue the assimilation of a spare that was
3789 * partially assimilated, finally try to activate a new
3790 * spare.
3791 */
3792 dl = imsm_readd(super, i, a);
3793 if (!dl)
3794 dl = imsm_add_spare(super, i, a, 0);
3795 if (!dl)
3796 dl = imsm_add_spare(super, i, a, 1);
3797 if (!dl)
3798 continue;
3799
3800 /* found a usable disk with enough space */
3801 di = malloc(sizeof(*di));
3802 if (!di)
3803 continue;
3804 memset(di, 0, sizeof(*di));
3805
3806 /* dl->index will be -1 in the case we are activating a
3807 * pristine spare. imsm_process_update() will create a
3808 * new index in this case. Once a disk is found to be
3809 * failed in all member arrays it is kicked from the
3810 * metadata
3811 */
3812 di->disk.number = dl->index;
3813
3814 /* (ab)use di->devs to store a pointer to the device
3815 * we chose
3816 */
3817 di->devs = (struct mdinfo *) dl;
3818
3819 di->disk.raid_disk = i;
3820 di->disk.major = dl->major;
3821 di->disk.minor = dl->minor;
3822 di->disk.state = 0;
3823 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
3824 di->component_size = a->info.component_size;
3825 di->container_member = inst;
3826 di->next = rv;
3827 rv = di;
3828 num_spares++;
3829 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3830 i, di->data_offset);
3831
3832 break;
3833 }
3834
3835 if (!rv)
3836 /* No spares found */
3837 return rv;
3838 /* Now 'rv' has a list of devices to return.
3839 * Create a metadata_update record to update the
3840 * disk_ord_tbl for the array
3841 */
3842 mu = malloc(sizeof(*mu));
3843 if (mu) {
3844 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
3845 if (mu->buf == NULL) {
3846 free(mu);
3847 mu = NULL;
3848 }
3849 }
3850 if (!mu) {
3851 while (rv) {
3852 struct mdinfo *n = rv->next;
3853
3854 free(rv);
3855 rv = n;
3856 }
3857 return NULL;
3858 }
3859
3860 mu->space = NULL;
3861 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
3862 mu->next = *updates;
3863 u = (struct imsm_update_activate_spare *) mu->buf;
3864
3865 for (di = rv ; di ; di = di->next) {
3866 u->type = update_activate_spare;
3867 u->dl = (struct dl *) di->devs;
3868 di->devs = NULL;
3869 u->slot = di->disk.raid_disk;
3870 u->array = inst;
3871 u->next = u + 1;
3872 u++;
3873 }
3874 (u-1)->next = NULL;
3875 *updates = mu;
3876
3877 return rv;
3878 }
3879
3880 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
3881 {
3882 struct imsm_dev *dev = get_imsm_dev(super, idx);
3883 struct imsm_map *map = get_imsm_map(dev, 0);
3884 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
3885 struct disk_info *inf = get_disk_info(u);
3886 struct imsm_disk *disk;
3887 int i;
3888 int j;
3889
3890 for (i = 0; i < map->num_members; i++) {
3891 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
3892 for (j = 0; j < new_map->num_members; j++)
3893 if (serialcmp(disk->serial, inf[j].serial) == 0)
3894 return 1;
3895 }
3896
3897 return 0;
3898 }
3899
3900 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
3901
3902 static void imsm_process_update(struct supertype *st,
3903 struct metadata_update *update)
3904 {
3905 /**
3906 * crack open the metadata_update envelope to find the update record
3907 * update can be one of:
3908 * update_activate_spare - a spare device has replaced a failed
3909 * device in an array, update the disk_ord_tbl. If this disk is
3910 * present in all member arrays then also clear the SPARE_DISK
3911 * flag
3912 */
3913 struct intel_super *super = st->sb;
3914 struct imsm_super *mpb;
3915 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
3916
3917 /* update requires a larger buf but the allocation failed */
3918 if (super->next_len && !super->next_buf) {
3919 super->next_len = 0;
3920 return;
3921 }
3922
3923 if (super->next_buf) {
3924 memcpy(super->next_buf, super->buf, super->len);
3925 free(super->buf);
3926 super->len = super->next_len;
3927 super->buf = super->next_buf;
3928
3929 super->next_len = 0;
3930 super->next_buf = NULL;
3931 }
3932
3933 mpb = super->anchor;
3934
3935 switch (type) {
3936 case update_activate_spare: {
3937 struct imsm_update_activate_spare *u = (void *) update->buf;
3938 struct imsm_dev *dev = get_imsm_dev(super, u->array);
3939 struct imsm_map *map = get_imsm_map(dev, 0);
3940 struct imsm_map *migr_map;
3941 struct active_array *a;
3942 struct imsm_disk *disk;
3943 __u8 to_state;
3944 struct dl *dl;
3945 unsigned int found;
3946 int failed;
3947 int victim = get_imsm_disk_idx(dev, u->slot);
3948 int i;
3949
3950 for (dl = super->disks; dl; dl = dl->next)
3951 if (dl == u->dl)
3952 break;
3953
3954 if (!dl) {
3955 fprintf(stderr, "error: imsm_activate_spare passed "
3956 "an unknown disk (index: %d)\n",
3957 u->dl->index);
3958 return;
3959 }
3960
3961 super->updates_pending++;
3962
3963 /* count failures (excluding rebuilds and the victim)
3964 * to determine map[0] state
3965 */
3966 failed = 0;
3967 for (i = 0; i < map->num_members; i++) {
3968 if (i == u->slot)
3969 continue;
3970 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
3971 if (!disk || disk->status & FAILED_DISK)
3972 failed++;
3973 }
3974
3975 /* adding a pristine spare, assign a new index */
3976 if (dl->index < 0) {
3977 dl->index = super->anchor->num_disks;
3978 super->anchor->num_disks++;
3979 }
3980 disk = &dl->disk;
3981 disk->status |= CONFIGURED_DISK;
3982 disk->status &= ~SPARE_DISK;
3983
3984 /* mark rebuild */
3985 to_state = imsm_check_degraded(super, dev, failed);
3986 map->map_state = IMSM_T_STATE_DEGRADED;
3987 migrate(dev, to_state, MIGR_REBUILD);
3988 migr_map = get_imsm_map(dev, 1);
3989 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
3990 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
3991
3992 /* count arrays using the victim in the metadata */
3993 found = 0;
3994 for (a = st->arrays; a ; a = a->next) {
3995 dev = get_imsm_dev(super, a->info.container_member);
3996 map = get_imsm_map(dev, 0);
3997
3998 if (get_imsm_disk_slot(map, victim) >= 0)
3999 found++;
4000 }
4001
4002 /* delete the victim if it is no longer being
4003 * utilized anywhere
4004 */
4005 if (!found) {
4006 struct dl **dlp;
4007
4008 /* We know that 'manager' isn't touching anything,
4009 * so it is safe to delete
4010 */
4011 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
4012 if ((*dlp)->index == victim)
4013 break;
4014
4015 /* victim may be on the missing list */
4016 if (!*dlp)
4017 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
4018 if ((*dlp)->index == victim)
4019 break;
4020 imsm_delete(super, dlp, victim);
4021 }
4022 break;
4023 }
4024 case update_create_array: {
4025 /* someone wants to create a new array, we need to be aware of
4026 * a few races/collisions:
4027 * 1/ 'Create' called by two separate instances of mdadm
4028 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4029 * devices that have since been assimilated via
4030 * activate_spare.
4031 * In the event this update can not be carried out mdadm will
4032 * (FIX ME) notice that its update did not take hold.
4033 */
4034 struct imsm_update_create_array *u = (void *) update->buf;
4035 struct intel_dev *dv;
4036 struct imsm_dev *dev;
4037 struct imsm_map *map, *new_map;
4038 unsigned long long start, end;
4039 unsigned long long new_start, new_end;
4040 int i;
4041 struct disk_info *inf;
4042 struct dl *dl;
4043
4044 /* handle racing creates: first come first serve */
4045 if (u->dev_idx < mpb->num_raid_devs) {
4046 dprintf("%s: subarray %d already defined\n",
4047 __func__, u->dev_idx);
4048 goto create_error;
4049 }
4050
4051 /* check update is next in sequence */
4052 if (u->dev_idx != mpb->num_raid_devs) {
4053 dprintf("%s: can not create array %d expected index %d\n",
4054 __func__, u->dev_idx, mpb->num_raid_devs);
4055 goto create_error;
4056 }
4057
4058 new_map = get_imsm_map(&u->dev, 0);
4059 new_start = __le32_to_cpu(new_map->pba_of_lba0);
4060 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
4061 inf = get_disk_info(u);
4062
4063 /* handle activate_spare versus create race:
4064 * check to make sure that overlapping arrays do not include
4065 * overalpping disks
4066 */
4067 for (i = 0; i < mpb->num_raid_devs; i++) {
4068 dev = get_imsm_dev(super, i);
4069 map = get_imsm_map(dev, 0);
4070 start = __le32_to_cpu(map->pba_of_lba0);
4071 end = start + __le32_to_cpu(map->blocks_per_member);
4072 if ((new_start >= start && new_start <= end) ||
4073 (start >= new_start && start <= new_end))
4074 /* overlap */;
4075 else
4076 continue;
4077
4078 if (disks_overlap(super, i, u)) {
4079 dprintf("%s: arrays overlap\n", __func__);
4080 goto create_error;
4081 }
4082 }
4083
4084 /* check that prepare update was successful */
4085 if (!update->space) {
4086 dprintf("%s: prepare update failed\n", __func__);
4087 goto create_error;
4088 }
4089
4090 /* check that all disks are still active before committing
4091 * changes. FIXME: could we instead handle this by creating a
4092 * degraded array? That's probably not what the user expects,
4093 * so better to drop this update on the floor.
4094 */
4095 for (i = 0; i < new_map->num_members; i++) {
4096 dl = serial_to_dl(inf[i].serial, super);
4097 if (!dl) {
4098 dprintf("%s: disk disappeared\n", __func__);
4099 goto create_error;
4100 }
4101 }
4102
4103 super->updates_pending++;
4104
4105 /* convert spares to members and fixup ord_tbl */
4106 for (i = 0; i < new_map->num_members; i++) {
4107 dl = serial_to_dl(inf[i].serial, super);
4108 if (dl->index == -1) {
4109 dl->index = mpb->num_disks;
4110 mpb->num_disks++;
4111 dl->disk.status |= CONFIGURED_DISK;
4112 dl->disk.status &= ~SPARE_DISK;
4113 }
4114 set_imsm_ord_tbl_ent(new_map, i, dl->index);
4115 }
4116
4117 dv = update->space;
4118 dev = dv->dev;
4119 update->space = NULL;
4120 imsm_copy_dev(dev, &u->dev);
4121 dv->index = u->dev_idx;
4122 dv->next = super->devlist;
4123 super->devlist = dv;
4124 mpb->num_raid_devs++;
4125
4126 imsm_update_version_info(super);
4127 break;
4128 create_error:
4129 /* mdmon knows how to release update->space, but not
4130 * ((struct intel_dev *) update->space)->dev
4131 */
4132 if (update->space) {
4133 dv = update->space;
4134 free(dv->dev);
4135 }
4136 break;
4137 }
4138 case update_add_disk:
4139
4140 /* we may be able to repair some arrays if disks are
4141 * being added */
4142 if (super->add) {
4143 struct active_array *a;
4144
4145 super->updates_pending++;
4146 for (a = st->arrays; a; a = a->next)
4147 a->check_degraded = 1;
4148 }
4149 /* add some spares to the metadata */
4150 while (super->add) {
4151 struct dl *al;
4152
4153 al = super->add;
4154 super->add = al->next;
4155 al->next = super->disks;
4156 super->disks = al;
4157 dprintf("%s: added %x:%x\n",
4158 __func__, al->major, al->minor);
4159 }
4160
4161 break;
4162 }
4163 }
4164
4165 static void imsm_prepare_update(struct supertype *st,
4166 struct metadata_update *update)
4167 {
4168 /**
4169 * Allocate space to hold new disk entries, raid-device entries or a new
4170 * mpb if necessary. The manager synchronously waits for updates to
4171 * complete in the monitor, so new mpb buffers allocated here can be
4172 * integrated by the monitor thread without worrying about live pointers
4173 * in the manager thread.
4174 */
4175 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4176 struct intel_super *super = st->sb;
4177 struct imsm_super *mpb = super->anchor;
4178 size_t buf_len;
4179 size_t len = 0;
4180
4181 switch (type) {
4182 case update_create_array: {
4183 struct imsm_update_create_array *u = (void *) update->buf;
4184 struct intel_dev *dv;
4185 struct imsm_dev *dev = &u->dev;
4186 struct imsm_map *map = get_imsm_map(dev, 0);
4187 struct dl *dl;
4188 struct disk_info *inf;
4189 int i;
4190 int activate = 0;
4191
4192 inf = get_disk_info(u);
4193 len = sizeof_imsm_dev(dev, 1);
4194 /* allocate a new super->devlist entry */
4195 dv = malloc(sizeof(*dv));
4196 if (dv) {
4197 dv->dev = malloc(len);
4198 if (dv->dev)
4199 update->space = dv;
4200 else {
4201 free(dv);
4202 update->space = NULL;
4203 }
4204 }
4205
4206 /* count how many spares will be converted to members */
4207 for (i = 0; i < map->num_members; i++) {
4208 dl = serial_to_dl(inf[i].serial, super);
4209 if (!dl) {
4210 /* hmm maybe it failed?, nothing we can do about
4211 * it here
4212 */
4213 continue;
4214 }
4215 if (count_memberships(dl, super) == 0)
4216 activate++;
4217 }
4218 len += activate * sizeof(struct imsm_disk);
4219 break;
4220 default:
4221 break;
4222 }
4223 }
4224
4225 /* check if we need a larger metadata buffer */
4226 if (super->next_buf)
4227 buf_len = super->next_len;
4228 else
4229 buf_len = super->len;
4230
4231 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4232 /* ok we need a larger buf than what is currently allocated
4233 * if this allocation fails process_update will notice that
4234 * ->next_len is set and ->next_buf is NULL
4235 */
4236 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4237 if (super->next_buf)
4238 free(super->next_buf);
4239
4240 super->next_len = buf_len;
4241 if (posix_memalign(&super->next_buf, 512, buf_len) != 0)
4242 super->next_buf = NULL;
4243 }
4244 }
4245
4246 /* must be called while manager is quiesced */
4247 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4248 {
4249 struct imsm_super *mpb = super->anchor;
4250 struct dl *iter;
4251 struct imsm_dev *dev;
4252 struct imsm_map *map;
4253 int i, j, num_members;
4254 __u32 ord;
4255
4256 dprintf("%s: deleting device[%d] from imsm_super\n",
4257 __func__, index);
4258
4259 /* shift all indexes down one */
4260 for (iter = super->disks; iter; iter = iter->next)
4261 if (iter->index > index)
4262 iter->index--;
4263 for (iter = super->missing; iter; iter = iter->next)
4264 if (iter->index > index)
4265 iter->index--;
4266
4267 for (i = 0; i < mpb->num_raid_devs; i++) {
4268 dev = get_imsm_dev(super, i);
4269 map = get_imsm_map(dev, 0);
4270 num_members = map->num_members;
4271 for (j = 0; j < num_members; j++) {
4272 /* update ord entries being careful not to propagate
4273 * ord-flags to the first map
4274 */
4275 ord = get_imsm_ord_tbl_ent(dev, j);
4276
4277 if (ord_to_idx(ord) <= index)
4278 continue;
4279
4280 map = get_imsm_map(dev, 0);
4281 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4282 map = get_imsm_map(dev, 1);
4283 if (map)
4284 set_imsm_ord_tbl_ent(map, j, ord - 1);
4285 }
4286 }
4287
4288 mpb->num_disks--;
4289 super->updates_pending++;
4290 if (*dlp) {
4291 struct dl *dl = *dlp;
4292
4293 *dlp = (*dlp)->next;
4294 __free_imsm_disk(dl);
4295 }
4296 }
4297 #endif /* MDASSEMBLE */
4298
4299 struct superswitch super_imsm = {
4300 #ifndef MDASSEMBLE
4301 .examine_super = examine_super_imsm,
4302 .brief_examine_super = brief_examine_super_imsm,
4303 .detail_super = detail_super_imsm,
4304 .brief_detail_super = brief_detail_super_imsm,
4305 .write_init_super = write_init_super_imsm,
4306 .validate_geometry = validate_geometry_imsm,
4307 .add_to_super = add_to_super_imsm,
4308 .detail_platform = detail_platform_imsm,
4309 #endif
4310 .match_home = match_home_imsm,
4311 .uuid_from_super= uuid_from_super_imsm,
4312 .getinfo_super = getinfo_super_imsm,
4313 .update_super = update_super_imsm,
4314
4315 .avail_size = avail_size_imsm,
4316
4317 .compare_super = compare_super_imsm,
4318
4319 .load_super = load_super_imsm,
4320 .init_super = init_super_imsm,
4321 .store_super = store_zero_imsm,
4322 .free_super = free_super_imsm,
4323 .match_metadata_desc = match_metadata_desc_imsm,
4324 .container_content = container_content_imsm,
4325 .default_layout = imsm_level_to_layout,
4326
4327 .external = 1,
4328 .name = "imsm",
4329
4330 #ifndef MDASSEMBLE
4331 /* for mdmon */
4332 .open_new = imsm_open_new,
4333 .load_super = load_super_imsm,
4334 .set_array_state= imsm_set_array_state,
4335 .set_disk = imsm_set_disk,
4336 .sync_metadata = imsm_sync_metadata,
4337 .activate_spare = imsm_activate_spare,
4338 .process_update = imsm_process_update,
4339 .prepare_update = imsm_prepare_update,
4340 #endif /* MDASSEMBLE */
4341 };
Unnamed repository; edit this file 'description' to name the repository.