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