]> 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: cleanup disk status tests
[thirdparty/mdadm.git] / super-intel.c
1 /*
2 * mdadm - Intel(R) Matrix Storage Manager Support
3 *
4 * Copyright (C) 2002-2008 Intel Corporation
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define HAVE_STDINT_H 1
21 #include "mdadm.h"
22 #include "mdmon.h"
23 #include "sha1.h"
24 #include "platform-intel.h"
25 #include <values.h>
26 #include <scsi/sg.h>
27 #include <ctype.h>
28 #include <dirent.h>
29
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
43
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
53
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56 #define SECT_PER_MB_SHIFT 11
57
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
60 struct imsm_disk {
61 __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 __u32 status; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
71 };
72
73 /* RAID map configuration infos. */
74 struct imsm_map {
75 __u32 pba_of_lba0; /* start address of partition */
76 __u32 blocks_per_member;/* blocks per member */
77 __u32 num_data_stripes; /* number of data stripes */
78 __u16 blocks_per_strip;
79 __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
84 __u8 raid_level;
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members; /* number of member disks */
89 __u8 num_domains; /* number of parity domains */
90 __u8 failed_disk_num; /* valid only when state is degraded */
91 __u8 ddf;
92 __u32 filler[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
96 */
97 } __attribute__ ((packed));
98
99 struct imsm_vol {
100 __u32 curr_migr_unit;
101 __u32 checkpoint_id; /* id to access curr_migr_unit */
102 __u8 migr_state; /* Normal or Migrating */
103 #define MIGR_INIT 0
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type; /* Initializing, Rebuilding, ... */
110 __u8 dirty;
111 __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors; /* number of mismatches */
113 __u16 bad_blocks; /* number of bad blocks during verify */
114 __u32 filler[4];
115 struct imsm_map map[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed));
118
119 struct imsm_dev {
120 __u8 volume[MAX_RAID_SERIAL_LEN];
121 __u32 size_low;
122 __u32 size_high;
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status; /* Persistent RaidDev status */
137 __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
138 __u8 migr_priority;
139 __u8 num_sub_vols;
140 __u8 tid;
141 __u8 cng_master_disk;
142 __u16 cache_policy;
143 __u8 cng_state;
144 __u8 cng_sub_state;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler[IMSM_DEV_FILLERS];
147 struct imsm_vol vol;
148 } __attribute__ ((packed));
149
150 struct imsm_super {
151 __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
152 __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size; /* 0x30 - 0x33 in bytes */
157 __u32 attributes; /* 0x34 - 0x37 */
158 __u8 num_disks; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos; /* 0x3A */
161 __u8 fill[1]; /* 0x3B */
162 __u32 cache_size; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed));
172
173 #define BBM_LOG_MAX_ENTRIES 254
174
175 struct bbm_log_entry {
176 __u64 defective_block_start;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset;
179 __u16 remapped_marked_count;
180 __u16 disk_ordinal;
181 } __attribute__ ((__packed__));
182
183 struct bbm_log {
184 __u32 signature; /* 0xABADB10C */
185 __u32 entry_count;
186 __u32 reserved_spare_block_count; /* 0 */
187 __u32 reserved; /* 0xFFFF */
188 __u64 first_spare_lba;
189 struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
190 } __attribute__ ((__packed__));
191
192
193 #ifndef MDASSEMBLE
194 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
195 #endif
196
197 static __u8 migr_type(struct imsm_dev *dev)
198 {
199 if (dev->vol.migr_type == MIGR_VERIFY &&
200 dev->status & DEV_VERIFY_AND_FIX)
201 return MIGR_REPAIR;
202 else
203 return dev->vol.migr_type;
204 }
205
206 static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
207 {
208 /* for compatibility with older oroms convert MIGR_REPAIR, into
209 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
210 */
211 if (migr_type == MIGR_REPAIR) {
212 dev->vol.migr_type = MIGR_VERIFY;
213 dev->status |= DEV_VERIFY_AND_FIX;
214 } else {
215 dev->vol.migr_type = migr_type;
216 dev->status &= ~DEV_VERIFY_AND_FIX;
217 }
218 }
219
220 static unsigned int sector_count(__u32 bytes)
221 {
222 return ((bytes + (512-1)) & (~(512-1))) / 512;
223 }
224
225 static unsigned int mpb_sectors(struct imsm_super *mpb)
226 {
227 return sector_count(__le32_to_cpu(mpb->mpb_size));
228 }
229
230 struct intel_dev {
231 struct imsm_dev *dev;
232 struct intel_dev *next;
233 int index;
234 };
235
236 /* internal representation of IMSM metadata */
237 struct intel_super {
238 union {
239 void *buf; /* O_DIRECT buffer for reading/writing metadata */
240 struct imsm_super *anchor; /* immovable parameters */
241 };
242 size_t len; /* size of the 'buf' allocation */
243 void *next_buf; /* for realloc'ing buf from the manager */
244 size_t next_len;
245 int updates_pending; /* count of pending updates for mdmon */
246 int creating_imsm; /* flag to indicate container creation */
247 int current_vol; /* index of raid device undergoing creation */
248 __u32 create_offset; /* common start for 'current_vol' */
249 __u32 random; /* random data for seeding new family numbers */
250 struct intel_dev *devlist;
251 struct dl {
252 struct dl *next;
253 int index;
254 __u8 serial[MAX_RAID_SERIAL_LEN];
255 int major, minor;
256 char *devname;
257 struct imsm_disk disk;
258 int fd;
259 int extent_cnt;
260 struct extent *e; /* for determining freespace @ create */
261 int raiddisk; /* slot to fill in autolayout */
262 } *disks;
263 struct dl *add; /* list of disks to add while mdmon active */
264 struct dl *missing; /* disks removed while we weren't looking */
265 struct bbm_log *bbm_log;
266 const char *hba; /* device path of the raid controller for this metadata */
267 const struct imsm_orom *orom; /* platform firmware support */
268 };
269
270 struct extent {
271 unsigned long long start, size;
272 };
273
274 /* definition of messages passed to imsm_process_update */
275 enum imsm_update_type {
276 update_activate_spare,
277 update_create_array,
278 update_add_disk,
279 };
280
281 struct imsm_update_activate_spare {
282 enum imsm_update_type type;
283 struct dl *dl;
284 int slot;
285 int array;
286 struct imsm_update_activate_spare *next;
287 };
288
289 struct disk_info {
290 __u8 serial[MAX_RAID_SERIAL_LEN];
291 };
292
293 struct imsm_update_create_array {
294 enum imsm_update_type type;
295 int dev_idx;
296 struct imsm_dev dev;
297 };
298
299 struct imsm_update_add_disk {
300 enum imsm_update_type type;
301 };
302
303 static struct supertype *match_metadata_desc_imsm(char *arg)
304 {
305 struct supertype *st;
306
307 if (strcmp(arg, "imsm") != 0 &&
308 strcmp(arg, "default") != 0
309 )
310 return NULL;
311
312 st = malloc(sizeof(*st));
313 memset(st, 0, sizeof(*st));
314 st->ss = &super_imsm;
315 st->max_devs = IMSM_MAX_DEVICES;
316 st->minor_version = 0;
317 st->sb = NULL;
318 return st;
319 }
320
321 #ifndef MDASSEMBLE
322 static __u8 *get_imsm_version(struct imsm_super *mpb)
323 {
324 return &mpb->sig[MPB_SIG_LEN];
325 }
326 #endif
327
328 /* retrieve a disk directly from the anchor when the anchor is known to be
329 * up-to-date, currently only at load time
330 */
331 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
332 {
333 if (index >= mpb->num_disks)
334 return NULL;
335 return &mpb->disk[index];
336 }
337
338 #ifndef MDASSEMBLE
339 /* retrieve a disk from the parsed metadata */
340 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
341 {
342 struct dl *d;
343
344 for (d = super->disks; d; d = d->next)
345 if (d->index == index)
346 return &d->disk;
347
348 return NULL;
349 }
350 #endif
351
352 /* generate a checksum directly from the anchor when the anchor is known to be
353 * up-to-date, currently only at load or write_super after coalescing
354 */
355 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
356 {
357 __u32 end = mpb->mpb_size / sizeof(end);
358 __u32 *p = (__u32 *) mpb;
359 __u32 sum = 0;
360
361 while (end--) {
362 sum += __le32_to_cpu(*p);
363 p++;
364 }
365
366 return sum - __le32_to_cpu(mpb->check_sum);
367 }
368
369 static size_t sizeof_imsm_map(struct imsm_map *map)
370 {
371 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
372 }
373
374 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
375 {
376 struct imsm_map *map = &dev->vol.map[0];
377
378 if (second_map && !dev->vol.migr_state)
379 return NULL;
380 else if (second_map) {
381 void *ptr = map;
382
383 return ptr + sizeof_imsm_map(map);
384 } else
385 return map;
386
387 }
388
389 /* return the size of the device.
390 * migr_state increases the returned size if map[0] were to be duplicated
391 */
392 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
393 {
394 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
395 sizeof_imsm_map(get_imsm_map(dev, 0));
396
397 /* migrating means an additional map */
398 if (dev->vol.migr_state)
399 size += sizeof_imsm_map(get_imsm_map(dev, 1));
400 else if (migr_state)
401 size += sizeof_imsm_map(get_imsm_map(dev, 0));
402
403 return size;
404 }
405
406 #ifndef MDASSEMBLE
407 /* retrieve disk serial number list from a metadata update */
408 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
409 {
410 void *u = update;
411 struct disk_info *inf;
412
413 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
414 sizeof_imsm_dev(&update->dev, 0);
415
416 return inf;
417 }
418 #endif
419
420 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
421 {
422 int offset;
423 int i;
424 void *_mpb = mpb;
425
426 if (index >= mpb->num_raid_devs)
427 return NULL;
428
429 /* devices start after all disks */
430 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
431
432 for (i = 0; i <= index; i++)
433 if (i == index)
434 return _mpb + offset;
435 else
436 offset += sizeof_imsm_dev(_mpb + offset, 0);
437
438 return NULL;
439 }
440
441 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
442 {
443 struct intel_dev *dv;
444
445 if (index >= super->anchor->num_raid_devs)
446 return NULL;
447 for (dv = super->devlist; dv; dv = dv->next)
448 if (dv->index == index)
449 return dv->dev;
450 return NULL;
451 }
452
453 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, int slot)
454 {
455 struct imsm_map *map;
456
457 if (dev->vol.migr_state)
458 map = get_imsm_map(dev, 1);
459 else
460 map = get_imsm_map(dev, 0);
461
462 /* top byte identifies disk under rebuild */
463 return __le32_to_cpu(map->disk_ord_tbl[slot]);
464 }
465
466 #define ord_to_idx(ord) (((ord) << 8) >> 8)
467 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot)
468 {
469 __u32 ord = get_imsm_ord_tbl_ent(dev, slot);
470
471 return ord_to_idx(ord);
472 }
473
474 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
475 {
476 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
477 }
478
479 static int get_imsm_disk_slot(struct imsm_map *map, int idx)
480 {
481 int slot;
482 __u32 ord;
483
484 for (slot = 0; slot < map->num_members; slot++) {
485 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
486 if (ord_to_idx(ord) == idx)
487 return slot;
488 }
489
490 return -1;
491 }
492
493 static int get_imsm_raid_level(struct imsm_map *map)
494 {
495 if (map->raid_level == 1) {
496 if (map->num_members == 2)
497 return 1;
498 else
499 return 10;
500 }
501
502 return map->raid_level;
503 }
504
505 static int cmp_extent(const void *av, const void *bv)
506 {
507 const struct extent *a = av;
508 const struct extent *b = bv;
509 if (a->start < b->start)
510 return -1;
511 if (a->start > b->start)
512 return 1;
513 return 0;
514 }
515
516 static int count_memberships(struct dl *dl, struct intel_super *super)
517 {
518 int memberships = 0;
519 int i;
520
521 for (i = 0; i < super->anchor->num_raid_devs; i++) {
522 struct imsm_dev *dev = get_imsm_dev(super, i);
523 struct imsm_map *map = get_imsm_map(dev, 0);
524
525 if (get_imsm_disk_slot(map, dl->index) >= 0)
526 memberships++;
527 }
528
529 return memberships;
530 }
531
532 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
533 {
534 /* find a list of used extents on the given physical device */
535 struct extent *rv, *e;
536 int i;
537 int memberships = count_memberships(dl, super);
538 __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
539
540 rv = malloc(sizeof(struct extent) * (memberships + 1));
541 if (!rv)
542 return NULL;
543 e = rv;
544
545 for (i = 0; i < super->anchor->num_raid_devs; i++) {
546 struct imsm_dev *dev = get_imsm_dev(super, i);
547 struct imsm_map *map = get_imsm_map(dev, 0);
548
549 if (get_imsm_disk_slot(map, dl->index) >= 0) {
550 e->start = __le32_to_cpu(map->pba_of_lba0);
551 e->size = __le32_to_cpu(map->blocks_per_member);
552 e++;
553 }
554 }
555 qsort(rv, memberships, sizeof(*rv), cmp_extent);
556
557 /* determine the start of the metadata
558 * when no raid devices are defined use the default
559 * ...otherwise allow the metadata to truncate the value
560 * as is the case with older versions of imsm
561 */
562 if (memberships) {
563 struct extent *last = &rv[memberships - 1];
564 __u32 remainder;
565
566 remainder = __le32_to_cpu(dl->disk.total_blocks) -
567 (last->start + last->size);
568 /* round down to 1k block to satisfy precision of the kernel
569 * 'size' interface
570 */
571 remainder &= ~1UL;
572 /* make sure remainder is still sane */
573 if (remainder < ROUND_UP(super->len, 512) >> 9)
574 remainder = ROUND_UP(super->len, 512) >> 9;
575 if (reservation > remainder)
576 reservation = remainder;
577 }
578 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
579 e->size = 0;
580 return rv;
581 }
582
583 /* try to determine how much space is reserved for metadata from
584 * the last get_extents() entry, otherwise fallback to the
585 * default
586 */
587 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
588 {
589 struct extent *e;
590 int i;
591 __u32 rv;
592
593 /* for spares just return a minimal reservation which will grow
594 * once the spare is picked up by an array
595 */
596 if (dl->index == -1)
597 return MPB_SECTOR_CNT;
598
599 e = get_extents(super, dl);
600 if (!e)
601 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
602
603 /* scroll to last entry */
604 for (i = 0; e[i].size; i++)
605 continue;
606
607 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
608
609 free(e);
610
611 return rv;
612 }
613
614 #ifndef MDASSEMBLE
615 static int is_spare(struct imsm_disk *disk)
616 {
617 return (disk->status & SPARE_DISK) == SPARE_DISK;
618 }
619
620 static int is_configured(struct imsm_disk *disk)
621 {
622 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
623 }
624
625 static int is_failed(struct imsm_disk *disk)
626 {
627 return (disk->status & FAILED_DISK) == FAILED_DISK;
628 }
629
630 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
631 {
632 __u64 sz;
633 int slot;
634 struct imsm_map *map = get_imsm_map(dev, 0);
635 __u32 ord;
636
637 printf("\n");
638 printf("[%.16s]:\n", dev->volume);
639 printf(" UUID : %s\n", uuid);
640 printf(" RAID Level : %d\n", get_imsm_raid_level(map));
641 printf(" Members : %d\n", map->num_members);
642 slot = get_imsm_disk_slot(map, disk_idx);
643 if (slot >= 0) {
644 ord = get_imsm_ord_tbl_ent(dev, slot);
645 printf(" This Slot : %d%s\n", slot,
646 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
647 } else
648 printf(" This Slot : ?\n");
649 sz = __le32_to_cpu(dev->size_high);
650 sz <<= 32;
651 sz += __le32_to_cpu(dev->size_low);
652 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
653 human_size(sz * 512));
654 sz = __le32_to_cpu(map->blocks_per_member);
655 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
656 human_size(sz * 512));
657 printf(" Sector Offset : %u\n",
658 __le32_to_cpu(map->pba_of_lba0));
659 printf(" Num Stripes : %u\n",
660 __le32_to_cpu(map->num_data_stripes));
661 printf(" Chunk Size : %u KiB\n",
662 __le16_to_cpu(map->blocks_per_strip) / 2);
663 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
664 printf(" Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle\n");
665 if (dev->vol.migr_state) {
666 if (migr_type(dev) == MIGR_INIT)
667 printf(": initializing\n");
668 else if (migr_type(dev) == MIGR_REBUILD)
669 printf(": rebuilding\n");
670 else if (migr_type(dev) == MIGR_VERIFY)
671 printf(": check\n");
672 else if (migr_type(dev) == MIGR_GEN_MIGR)
673 printf(": general migration\n");
674 else if (migr_type(dev) == MIGR_STATE_CHANGE)
675 printf(": state change\n");
676 else if (migr_type(dev) == MIGR_REPAIR)
677 printf(": repair\n");
678 else
679 printf(": <unknown:%d>\n", migr_type(dev));
680 }
681 printf(" Map State : %s", map_state_str[map->map_state]);
682 if (dev->vol.migr_state) {
683 struct imsm_map *map = get_imsm_map(dev, 1);
684 printf(" <-- %s", map_state_str[map->map_state]);
685 }
686 printf("\n");
687 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
688 }
689
690 static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved)
691 {
692 struct imsm_disk *disk = __get_imsm_disk(mpb, index);
693 char str[MAX_RAID_SERIAL_LEN + 1];
694 __u64 sz;
695
696 if (index < 0)
697 return;
698
699 printf("\n");
700 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
701 printf(" Disk%02d Serial : %s\n", index, str);
702 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
703 is_configured(disk) ? " active" : "",
704 is_failed(disk) ? " failed" : "");
705 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
706 sz = __le32_to_cpu(disk->total_blocks) - reserved;
707 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
708 human_size(sz * 512));
709 }
710
711 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info);
712
713 static void examine_super_imsm(struct supertype *st, char *homehost)
714 {
715 struct intel_super *super = st->sb;
716 struct imsm_super *mpb = super->anchor;
717 char str[MAX_SIGNATURE_LENGTH];
718 int i;
719 struct mdinfo info;
720 char nbuf[64];
721 __u32 sum;
722 __u32 reserved = imsm_reserved_sectors(super, super->disks);
723
724
725 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
726 printf(" Magic : %s\n", str);
727 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
728 printf(" Version : %s\n", get_imsm_version(mpb));
729 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
730 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
731 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
732 getinfo_super_imsm(st, &info);
733 fname_from_uuid(st, &info, nbuf, ':');
734 printf(" UUID : %s\n", nbuf + 5);
735 sum = __le32_to_cpu(mpb->check_sum);
736 printf(" Checksum : %08x %s\n", sum,
737 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
738 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
739 printf(" Disks : %d\n", mpb->num_disks);
740 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
741 print_imsm_disk(mpb, super->disks->index, reserved);
742 if (super->bbm_log) {
743 struct bbm_log *log = super->bbm_log;
744
745 printf("\n");
746 printf("Bad Block Management Log:\n");
747 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
748 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
749 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
750 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
751 printf(" First Spare : %llx\n",
752 (unsigned long long) __le64_to_cpu(log->first_spare_lba));
753 }
754 for (i = 0; i < mpb->num_raid_devs; i++) {
755 struct mdinfo info;
756 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
757
758 super->current_vol = i;
759 getinfo_super_imsm(st, &info);
760 fname_from_uuid(st, &info, nbuf, ':');
761 print_imsm_dev(dev, nbuf + 5, super->disks->index);
762 }
763 for (i = 0; i < mpb->num_disks; i++) {
764 if (i == super->disks->index)
765 continue;
766 print_imsm_disk(mpb, i, reserved);
767 }
768 }
769
770 static void brief_examine_super_imsm(struct supertype *st, int verbose)
771 {
772 /* We just write a generic IMSM ARRAY entry */
773 struct mdinfo info;
774 char nbuf[64];
775 struct intel_super *super = st->sb;
776
777 if (!super->anchor->num_raid_devs) {
778 printf("ARRAY metadata=imsm\n");
779 return;
780 }
781
782 getinfo_super_imsm(st, &info);
783 fname_from_uuid(st, &info, nbuf, ':');
784 printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
785 }
786
787 static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
788 {
789 /* We just write a generic IMSM ARRAY entry */
790 struct mdinfo info;
791 char nbuf[64];
792 char nbuf1[64];
793 struct intel_super *super = st->sb;
794 int i;
795
796 if (!super->anchor->num_raid_devs)
797 return;
798
799 getinfo_super_imsm(st, &info);
800 fname_from_uuid(st, &info, nbuf, ':');
801 for (i = 0; i < super->anchor->num_raid_devs; i++) {
802 struct imsm_dev *dev = get_imsm_dev(super, i);
803
804 super->current_vol = i;
805 getinfo_super_imsm(st, &info);
806 fname_from_uuid(st, &info, nbuf1, ':');
807 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
808 dev->volume, nbuf + 5, i, nbuf1 + 5);
809 }
810 }
811
812 static void export_examine_super_imsm(struct supertype *st)
813 {
814 struct intel_super *super = st->sb;
815 struct imsm_super *mpb = super->anchor;
816 struct mdinfo info;
817 char nbuf[64];
818
819 getinfo_super_imsm(st, &info);
820 fname_from_uuid(st, &info, nbuf, ':');
821 printf("MD_METADATA=imsm\n");
822 printf("MD_LEVEL=container\n");
823 printf("MD_UUID=%s\n", nbuf+5);
824 printf("MD_DEVICES=%u\n", mpb->num_disks);
825 }
826
827 static void detail_super_imsm(struct supertype *st, char *homehost)
828 {
829 struct mdinfo info;
830 char nbuf[64];
831
832 getinfo_super_imsm(st, &info);
833 fname_from_uuid(st, &info, nbuf, ':');
834 printf("\n UUID : %s\n", nbuf + 5);
835 }
836
837 static void brief_detail_super_imsm(struct supertype *st)
838 {
839 struct mdinfo info;
840 char nbuf[64];
841 getinfo_super_imsm(st, &info);
842 fname_from_uuid(st, &info, nbuf, ':');
843 printf(" UUID=%s", nbuf + 5);
844 }
845
846 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
847 static void fd2devname(int fd, char *name);
848
849 static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
850 {
851 /* dump an unsorted list of devices attached to ahci, as well as
852 * non-connected ports
853 */
854 int hba_len = strlen(hba_path) + 1;
855 struct dirent *ent;
856 DIR *dir;
857 char *path = NULL;
858 int err = 0;
859 unsigned long port_mask = (1 << port_count) - 1;
860
861 if (port_count > sizeof(port_mask) * 8) {
862 if (verbose)
863 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
864 return 2;
865 }
866
867 /* scroll through /sys/dev/block looking for devices attached to
868 * this hba
869 */
870 dir = opendir("/sys/dev/block");
871 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
872 int fd;
873 char model[64];
874 char vendor[64];
875 char buf[1024];
876 int major, minor;
877 char *device;
878 char *c;
879 int port;
880 int type;
881
882 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
883 continue;
884 path = devt_to_devpath(makedev(major, minor));
885 if (!path)
886 continue;
887 if (!path_attached_to_hba(path, hba_path)) {
888 free(path);
889 path = NULL;
890 continue;
891 }
892
893 /* retrieve the scsi device type */
894 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
895 if (verbose)
896 fprintf(stderr, Name ": failed to allocate 'device'\n");
897 err = 2;
898 break;
899 }
900 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
901 if (load_sys(device, buf) != 0) {
902 if (verbose)
903 fprintf(stderr, Name ": failed to read device type for %s\n",
904 path);
905 err = 2;
906 free(device);
907 break;
908 }
909 type = strtoul(buf, NULL, 10);
910
911 /* if it's not a disk print the vendor and model */
912 if (!(type == 0 || type == 7 || type == 14)) {
913 vendor[0] = '\0';
914 model[0] = '\0';
915 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
916 if (load_sys(device, buf) == 0) {
917 strncpy(vendor, buf, sizeof(vendor));
918 vendor[sizeof(vendor) - 1] = '\0';
919 c = (char *) &vendor[sizeof(vendor) - 1];
920 while (isspace(*c) || *c == '\0')
921 *c-- = '\0';
922
923 }
924 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
925 if (load_sys(device, buf) == 0) {
926 strncpy(model, buf, sizeof(model));
927 model[sizeof(model) - 1] = '\0';
928 c = (char *) &model[sizeof(model) - 1];
929 while (isspace(*c) || *c == '\0')
930 *c-- = '\0';
931 }
932
933 if (vendor[0] && model[0])
934 sprintf(buf, "%.64s %.64s", vendor, model);
935 else
936 switch (type) { /* numbers from hald/linux/device.c */
937 case 1: sprintf(buf, "tape"); break;
938 case 2: sprintf(buf, "printer"); break;
939 case 3: sprintf(buf, "processor"); break;
940 case 4:
941 case 5: sprintf(buf, "cdrom"); break;
942 case 6: sprintf(buf, "scanner"); break;
943 case 8: sprintf(buf, "media_changer"); break;
944 case 9: sprintf(buf, "comm"); break;
945 case 12: sprintf(buf, "raid"); break;
946 default: sprintf(buf, "unknown");
947 }
948 } else
949 buf[0] = '\0';
950 free(device);
951
952 /* chop device path to 'host%d' and calculate the port number */
953 c = strchr(&path[hba_len], '/');
954 *c = '\0';
955 if (sscanf(&path[hba_len], "host%d", &port) == 1)
956 port -= host_base;
957 else {
958 if (verbose) {
959 *c = '/'; /* repair the full string */
960 fprintf(stderr, Name ": failed to determine port number for %s\n",
961 path);
962 }
963 err = 2;
964 break;
965 }
966
967 /* mark this port as used */
968 port_mask &= ~(1 << port);
969
970 /* print out the device information */
971 if (buf[0]) {
972 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
973 continue;
974 }
975
976 fd = dev_open(ent->d_name, O_RDONLY);
977 if (fd < 0)
978 printf(" Port%d : - disk info unavailable -\n", port);
979 else {
980 fd2devname(fd, buf);
981 printf(" Port%d : %s", port, buf);
982 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
983 printf(" (%s)\n", buf);
984 else
985 printf("()\n");
986 }
987 close(fd);
988 free(path);
989 path = NULL;
990 }
991 if (path)
992 free(path);
993 if (dir)
994 closedir(dir);
995 if (err == 0) {
996 int i;
997
998 for (i = 0; i < port_count; i++)
999 if (port_mask & (1 << i))
1000 printf(" Port%d : - no device attached -\n", i);
1001 }
1002
1003 return err;
1004 }
1005
1006 static int detail_platform_imsm(int verbose, int enumerate_only)
1007 {
1008 /* There are two components to imsm platform support, the ahci SATA
1009 * controller and the option-rom. To find the SATA controller we
1010 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1011 * controller with the Intel vendor id is present. This approach
1012 * allows mdadm to leverage the kernel's ahci detection logic, with the
1013 * caveat that if ahci.ko is not loaded mdadm will not be able to
1014 * detect platform raid capabilities. The option-rom resides in a
1015 * platform "Adapter ROM". We scan for its signature to retrieve the
1016 * platform capabilities. If raid support is disabled in the BIOS the
1017 * option-rom capability structure will not be available.
1018 */
1019 const struct imsm_orom *orom;
1020 struct sys_dev *list, *hba;
1021 DIR *dir;
1022 struct dirent *ent;
1023 const char *hba_path;
1024 int host_base = 0;
1025 int port_count = 0;
1026
1027 if (enumerate_only) {
1028 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1029 return 0;
1030 return 2;
1031 }
1032
1033 list = find_driver_devices("pci", "ahci");
1034 for (hba = list; hba; hba = hba->next)
1035 if (devpath_to_vendor(hba->path) == 0x8086)
1036 break;
1037
1038 if (!hba) {
1039 if (verbose)
1040 fprintf(stderr, Name ": unable to find active ahci controller\n");
1041 free_sys_dev(&list);
1042 return 2;
1043 } else if (verbose)
1044 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
1045 hba_path = hba->path;
1046 hba->path = NULL;
1047 free_sys_dev(&list);
1048
1049 orom = find_imsm_orom();
1050 if (!orom) {
1051 if (verbose)
1052 fprintf(stderr, Name ": imsm option-rom not found\n");
1053 return 2;
1054 }
1055
1056 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1057 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
1058 orom->hotfix_ver, orom->build);
1059 printf(" RAID Levels :%s%s%s%s%s\n",
1060 imsm_orom_has_raid0(orom) ? " raid0" : "",
1061 imsm_orom_has_raid1(orom) ? " raid1" : "",
1062 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
1063 imsm_orom_has_raid10(orom) ? " raid10" : "",
1064 imsm_orom_has_raid5(orom) ? " raid5" : "");
1065 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1066 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
1067 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
1068 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
1069 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
1070 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
1071 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
1072 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
1073 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
1074 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
1075 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
1076 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
1077 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
1078 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
1079 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
1080 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
1081 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1082 printf(" Max Disks : %d\n", orom->tds);
1083 printf(" Max Volumes : %d\n", orom->vpa);
1084 printf(" I/O Controller : %s\n", hba_path);
1085
1086 /* find the smallest scsi host number to determine a port number base */
1087 dir = opendir(hba_path);
1088 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1089 int host;
1090
1091 if (sscanf(ent->d_name, "host%d", &host) != 1)
1092 continue;
1093 if (port_count == 0)
1094 host_base = host;
1095 else if (host < host_base)
1096 host_base = host;
1097
1098 if (host + 1 > port_count + host_base)
1099 port_count = host + 1 - host_base;
1100
1101 }
1102 if (dir)
1103 closedir(dir);
1104
1105 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1106 host_base, verbose) != 0) {
1107 if (verbose)
1108 fprintf(stderr, Name ": failed to enumerate ports\n");
1109 return 2;
1110 }
1111
1112 return 0;
1113 }
1114 #endif
1115
1116 static int match_home_imsm(struct supertype *st, char *homehost)
1117 {
1118 /* the imsm metadata format does not specify any host
1119 * identification information. We return -1 since we can never
1120 * confirm nor deny whether a given array is "meant" for this
1121 * host. We rely on compare_super and the 'family_num' fields to
1122 * exclude member disks that do not belong, and we rely on
1123 * mdadm.conf to specify the arrays that should be assembled.
1124 * Auto-assembly may still pick up "foreign" arrays.
1125 */
1126
1127 return -1;
1128 }
1129
1130 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1131 {
1132 /* The uuid returned here is used for:
1133 * uuid to put into bitmap file (Create, Grow)
1134 * uuid for backup header when saving critical section (Grow)
1135 * comparing uuids when re-adding a device into an array
1136 * In these cases the uuid required is that of the data-array,
1137 * not the device-set.
1138 * uuid to recognise same set when adding a missing device back
1139 * to an array. This is a uuid for the device-set.
1140 *
1141 * For each of these we can make do with a truncated
1142 * or hashed uuid rather than the original, as long as
1143 * everyone agrees.
1144 * In each case the uuid required is that of the data-array,
1145 * not the device-set.
1146 */
1147 /* imsm does not track uuid's so we synthesis one using sha1 on
1148 * - The signature (Which is constant for all imsm array, but no matter)
1149 * - the orig_family_num of the container
1150 * - the index number of the volume
1151 * - the 'serial' number of the volume.
1152 * Hopefully these are all constant.
1153 */
1154 struct intel_super *super = st->sb;
1155
1156 char buf[20];
1157 struct sha1_ctx ctx;
1158 struct imsm_dev *dev = NULL;
1159 __u32 family_num;
1160
1161 /* some mdadm versions failed to set ->orig_family_num, in which
1162 * case fall back to ->family_num. orig_family_num will be
1163 * fixed up with the first metadata update.
1164 */
1165 family_num = super->anchor->orig_family_num;
1166 if (family_num == 0)
1167 family_num = super->anchor->family_num;
1168 sha1_init_ctx(&ctx);
1169 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1170 sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
1171 if (super->current_vol >= 0)
1172 dev = get_imsm_dev(super, super->current_vol);
1173 if (dev) {
1174 __u32 vol = super->current_vol;
1175 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1176 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1177 }
1178 sha1_finish_ctx(&ctx, buf);
1179 memcpy(uuid, buf, 4*4);
1180 }
1181
1182 #if 0
1183 static void
1184 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1185 {
1186 __u8 *v = get_imsm_version(mpb);
1187 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1188 char major[] = { 0, 0, 0 };
1189 char minor[] = { 0 ,0, 0 };
1190 char patch[] = { 0, 0, 0 };
1191 char *ver_parse[] = { major, minor, patch };
1192 int i, j;
1193
1194 i = j = 0;
1195 while (*v != '\0' && v < end) {
1196 if (*v != '.' && j < 2)
1197 ver_parse[i][j++] = *v;
1198 else {
1199 i++;
1200 j = 0;
1201 }
1202 v++;
1203 }
1204
1205 *m = strtol(minor, NULL, 0);
1206 *p = strtol(patch, NULL, 0);
1207 }
1208 #endif
1209
1210 static int imsm_level_to_layout(int level)
1211 {
1212 switch (level) {
1213 case 0:
1214 case 1:
1215 return 0;
1216 case 5:
1217 case 6:
1218 return ALGORITHM_LEFT_ASYMMETRIC;
1219 case 10:
1220 return 0x102;
1221 }
1222 return UnSet;
1223 }
1224
1225 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
1226 {
1227 struct intel_super *super = st->sb;
1228 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1229 struct imsm_map *map = get_imsm_map(dev, 0);
1230 struct dl *dl;
1231
1232 for (dl = super->disks; dl; dl = dl->next)
1233 if (dl->raiddisk == info->disk.raid_disk)
1234 break;
1235 info->container_member = super->current_vol;
1236 info->array.raid_disks = map->num_members;
1237 info->array.level = get_imsm_raid_level(map);
1238 info->array.layout = imsm_level_to_layout(info->array.level);
1239 info->array.md_minor = -1;
1240 info->array.ctime = 0;
1241 info->array.utime = 0;
1242 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1243 info->array.state = !dev->vol.dirty;
1244 info->custom_array_size = __le32_to_cpu(dev->size_high);
1245 info->custom_array_size <<= 32;
1246 info->custom_array_size |= __le32_to_cpu(dev->size_low);
1247
1248 info->disk.major = 0;
1249 info->disk.minor = 0;
1250 if (dl) {
1251 info->disk.major = dl->major;
1252 info->disk.minor = dl->minor;
1253 }
1254
1255 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1256 info->component_size = __le32_to_cpu(map->blocks_per_member);
1257 memset(info->uuid, 0, sizeof(info->uuid));
1258
1259 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
1260 info->resync_start = 0;
1261 else if (dev->vol.migr_state)
1262 /* FIXME add curr_migr_unit to resync_start conversion */
1263 info->resync_start = 0;
1264 else
1265 info->resync_start = ~0ULL;
1266
1267 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1268 info->name[MAX_RAID_SERIAL_LEN] = 0;
1269
1270 info->array.major_version = -1;
1271 info->array.minor_version = -2;
1272 sprintf(info->text_version, "/%s/%d",
1273 devnum2devname(st->container_dev),
1274 info->container_member);
1275 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1276 uuid_from_super_imsm(st, info->uuid);
1277 }
1278
1279 /* check the config file to see if we can return a real uuid for this spare */
1280 static void fixup_container_spare_uuid(struct mdinfo *inf)
1281 {
1282 struct mddev_ident_s *array_list;
1283
1284 if (inf->array.level != LEVEL_CONTAINER ||
1285 memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0)
1286 return;
1287
1288 array_list = conf_get_ident(NULL);
1289
1290 for (; array_list; array_list = array_list->next) {
1291 if (array_list->uuid_set) {
1292 struct supertype *_sst; /* spare supertype */
1293 struct supertype *_cst; /* container supertype */
1294
1295 _cst = array_list->st;
1296 if (_cst)
1297 _sst = _cst->ss->match_metadata_desc(inf->text_version);
1298 else
1299 _sst = NULL;
1300
1301 if (_sst) {
1302 memcpy(inf->uuid, array_list->uuid, sizeof(int[4]));
1303 free(_sst);
1304 break;
1305 }
1306 }
1307 }
1308 }
1309
1310 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
1311 {
1312 struct intel_super *super = st->sb;
1313 struct imsm_disk *disk;
1314
1315 if (super->current_vol >= 0) {
1316 getinfo_super_imsm_volume(st, info);
1317 return;
1318 }
1319
1320 /* Set raid_disks to zero so that Assemble will always pull in valid
1321 * spares
1322 */
1323 info->array.raid_disks = 0;
1324 info->array.level = LEVEL_CONTAINER;
1325 info->array.layout = 0;
1326 info->array.md_minor = -1;
1327 info->array.ctime = 0; /* N/A for imsm */
1328 info->array.utime = 0;
1329 info->array.chunk_size = 0;
1330
1331 info->disk.major = 0;
1332 info->disk.minor = 0;
1333 info->disk.raid_disk = -1;
1334 info->reshape_active = 0;
1335 info->array.major_version = -1;
1336 info->array.minor_version = -2;
1337 strcpy(info->text_version, "imsm");
1338 info->safe_mode_delay = 0;
1339 info->disk.number = -1;
1340 info->disk.state = 0;
1341 info->name[0] = 0;
1342
1343 if (super->disks) {
1344 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1345
1346 disk = &super->disks->disk;
1347 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1348 info->component_size = reserved;
1349 info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
1350 /* we don't change info->disk.raid_disk here because
1351 * this state will be finalized in mdmon after we have
1352 * found the 'most fresh' version of the metadata
1353 */
1354 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
1355 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
1356 }
1357
1358 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1359 * ->compare_super may have updated the 'num_raid_devs' field for spares
1360 */
1361 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1362 uuid_from_super_imsm(st, info->uuid);
1363 else {
1364 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1365 fixup_container_spare_uuid(info);
1366 }
1367 }
1368
1369 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1370 char *update, char *devname, int verbose,
1371 int uuid_set, char *homehost)
1372 {
1373 /* FIXME */
1374
1375 /* For 'assemble' and 'force' we need to return non-zero if any
1376 * change was made. For others, the return value is ignored.
1377 * Update options are:
1378 * force-one : This device looks a bit old but needs to be included,
1379 * update age info appropriately.
1380 * assemble: clear any 'faulty' flag to allow this device to
1381 * be assembled.
1382 * force-array: Array is degraded but being forced, mark it clean
1383 * if that will be needed to assemble it.
1384 *
1385 * newdev: not used ????
1386 * grow: Array has gained a new device - this is currently for
1387 * linear only
1388 * resync: mark as dirty so a resync will happen.
1389 * name: update the name - preserving the homehost
1390 *
1391 * Following are not relevant for this imsm:
1392 * sparc2.2 : update from old dodgey metadata
1393 * super-minor: change the preferred_minor number
1394 * summaries: update redundant counters.
1395 * uuid: Change the uuid of the array to match watch is given
1396 * homehost: update the recorded homehost
1397 * _reshape_progress: record new reshape_progress position.
1398 */
1399 int rv = 0;
1400 //struct intel_super *super = st->sb;
1401 //struct imsm_super *mpb = super->mpb;
1402
1403 if (strcmp(update, "grow") == 0) {
1404 }
1405 if (strcmp(update, "resync") == 0) {
1406 /* dev->vol.dirty = 1; */
1407 }
1408
1409 /* IMSM has no concept of UUID or homehost */
1410
1411 return rv;
1412 }
1413
1414 static size_t disks_to_mpb_size(int disks)
1415 {
1416 size_t size;
1417
1418 size = sizeof(struct imsm_super);
1419 size += (disks - 1) * sizeof(struct imsm_disk);
1420 size += 2 * sizeof(struct imsm_dev);
1421 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1422 size += (4 - 2) * sizeof(struct imsm_map);
1423 /* 4 possible disk_ord_tbl's */
1424 size += 4 * (disks - 1) * sizeof(__u32);
1425
1426 return size;
1427 }
1428
1429 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1430 {
1431 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1432 return 0;
1433
1434 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1435 }
1436
1437 static void free_devlist(struct intel_super *super)
1438 {
1439 struct intel_dev *dv;
1440
1441 while (super->devlist) {
1442 dv = super->devlist->next;
1443 free(super->devlist->dev);
1444 free(super->devlist);
1445 super->devlist = dv;
1446 }
1447 }
1448
1449 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1450 {
1451 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1452 }
1453
1454 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1455 {
1456 /*
1457 * return:
1458 * 0 same, or first was empty, and second was copied
1459 * 1 second had wrong number
1460 * 2 wrong uuid
1461 * 3 wrong other info
1462 */
1463 struct intel_super *first = st->sb;
1464 struct intel_super *sec = tst->sb;
1465
1466 if (!first) {
1467 st->sb = tst->sb;
1468 tst->sb = NULL;
1469 return 0;
1470 }
1471
1472 if (memcmp(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH) != 0)
1473 return 3;
1474
1475 /* if an anchor does not have num_raid_devs set then it is a free
1476 * floating spare
1477 */
1478 if (first->anchor->num_raid_devs > 0 &&
1479 sec->anchor->num_raid_devs > 0) {
1480 if (first->anchor->orig_family_num != sec->anchor->orig_family_num ||
1481 first->anchor->family_num != sec->anchor->family_num)
1482 return 3;
1483 }
1484
1485 /* if 'first' is a spare promote it to a populated mpb with sec's
1486 * family number
1487 */
1488 if (first->anchor->num_raid_devs == 0 &&
1489 sec->anchor->num_raid_devs > 0) {
1490 int i;
1491 struct intel_dev *dv;
1492 struct imsm_dev *dev;
1493
1494 /* we need to copy raid device info from sec if an allocation
1495 * fails here we don't associate the spare
1496 */
1497 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1498 dv = malloc(sizeof(*dv));
1499 if (!dv)
1500 break;
1501 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1502 if (!dev) {
1503 free(dv);
1504 break;
1505 }
1506 dv->dev = dev;
1507 dv->index = i;
1508 dv->next = first->devlist;
1509 first->devlist = dv;
1510 }
1511 if (i < sec->anchor->num_raid_devs) {
1512 /* allocation failure */
1513 free_devlist(first);
1514 fprintf(stderr, "imsm: failed to associate spare\n");
1515 return 3;
1516 }
1517 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1518 first->anchor->orig_family_num = sec->anchor->orig_family_num;
1519 first->anchor->family_num = sec->anchor->family_num;
1520 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1521 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1522 }
1523
1524 return 0;
1525 }
1526
1527 static void fd2devname(int fd, char *name)
1528 {
1529 struct stat st;
1530 char path[256];
1531 char dname[100];
1532 char *nm;
1533 int rv;
1534
1535 name[0] = '\0';
1536 if (fstat(fd, &st) != 0)
1537 return;
1538 sprintf(path, "/sys/dev/block/%d:%d",
1539 major(st.st_rdev), minor(st.st_rdev));
1540
1541 rv = readlink(path, dname, sizeof(dname));
1542 if (rv <= 0)
1543 return;
1544
1545 dname[rv] = '\0';
1546 nm = strrchr(dname, '/');
1547 nm++;
1548 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1549 }
1550
1551
1552 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1553
1554 static int imsm_read_serial(int fd, char *devname,
1555 __u8 serial[MAX_RAID_SERIAL_LEN])
1556 {
1557 unsigned char scsi_serial[255];
1558 int rv;
1559 int rsp_len;
1560 int len;
1561 char *dest;
1562 char *src;
1563 char *rsp_buf;
1564 int i;
1565
1566 memset(scsi_serial, 0, sizeof(scsi_serial));
1567
1568 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1569
1570 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1571 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1572 fd2devname(fd, (char *) serial);
1573 return 0;
1574 }
1575
1576 if (rv != 0) {
1577 if (devname)
1578 fprintf(stderr,
1579 Name ": Failed to retrieve serial for %s\n",
1580 devname);
1581 return rv;
1582 }
1583
1584 rsp_len = scsi_serial[3];
1585 if (!rsp_len) {
1586 if (devname)
1587 fprintf(stderr,
1588 Name ": Failed to retrieve serial for %s\n",
1589 devname);
1590 return 2;
1591 }
1592 rsp_buf = (char *) &scsi_serial[4];
1593
1594 /* trim all whitespace and non-printable characters and convert
1595 * ':' to ';'
1596 */
1597 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
1598 src = &rsp_buf[i];
1599 if (*src > 0x20) {
1600 /* ':' is reserved for use in placeholder serial
1601 * numbers for missing disks
1602 */
1603 if (*src == ':')
1604 *dest++ = ';';
1605 else
1606 *dest++ = *src;
1607 }
1608 }
1609 len = dest - rsp_buf;
1610 dest = rsp_buf;
1611
1612 /* truncate leading characters */
1613 if (len > MAX_RAID_SERIAL_LEN) {
1614 dest += len - MAX_RAID_SERIAL_LEN;
1615 len = MAX_RAID_SERIAL_LEN;
1616 }
1617
1618 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1619 memcpy(serial, dest, len);
1620
1621 return 0;
1622 }
1623
1624 static int serialcmp(__u8 *s1, __u8 *s2)
1625 {
1626 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1627 }
1628
1629 static void serialcpy(__u8 *dest, __u8 *src)
1630 {
1631 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1632 }
1633
1634 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1635 {
1636 struct dl *dl;
1637
1638 for (dl = super->disks; dl; dl = dl->next)
1639 if (serialcmp(dl->serial, serial) == 0)
1640 break;
1641
1642 return dl;
1643 }
1644
1645 static int
1646 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1647 {
1648 struct dl *dl;
1649 struct stat stb;
1650 int rv;
1651 int i;
1652 int alloc = 1;
1653 __u8 serial[MAX_RAID_SERIAL_LEN];
1654
1655 rv = imsm_read_serial(fd, devname, serial);
1656
1657 if (rv != 0)
1658 return 2;
1659
1660 /* check if this is a disk we have seen before. it may be a spare in
1661 * super->disks while the current anchor believes it is a raid member,
1662 * check if we need to update dl->index
1663 */
1664 dl = serial_to_dl(serial, super);
1665 if (!dl)
1666 dl = malloc(sizeof(*dl));
1667 else
1668 alloc = 0;
1669
1670 if (!dl) {
1671 if (devname)
1672 fprintf(stderr,
1673 Name ": failed to allocate disk buffer for %s\n",
1674 devname);
1675 return 2;
1676 }
1677
1678 if (alloc) {
1679 fstat(fd, &stb);
1680 dl->major = major(stb.st_rdev);
1681 dl->minor = minor(stb.st_rdev);
1682 dl->next = super->disks;
1683 dl->fd = keep_fd ? fd : -1;
1684 dl->devname = devname ? strdup(devname) : NULL;
1685 serialcpy(dl->serial, serial);
1686 dl->index = -2;
1687 dl->e = NULL;
1688 } else if (keep_fd) {
1689 close(dl->fd);
1690 dl->fd = fd;
1691 }
1692
1693 /* look up this disk's index in the current anchor */
1694 for (i = 0; i < super->anchor->num_disks; i++) {
1695 struct imsm_disk *disk_iter;
1696
1697 disk_iter = __get_imsm_disk(super->anchor, i);
1698
1699 if (serialcmp(disk_iter->serial, dl->serial) == 0) {
1700 dl->disk = *disk_iter;
1701 /* only set index on disks that are a member of a
1702 * populated contianer, i.e. one with raid_devs
1703 */
1704 if (dl->disk.status & FAILED_DISK)
1705 dl->index = -2;
1706 else if (dl->disk.status & SPARE_DISK)
1707 dl->index = -1;
1708 else
1709 dl->index = i;
1710
1711 break;
1712 }
1713 }
1714
1715 /* no match, maybe a stale failed drive */
1716 if (i == super->anchor->num_disks && dl->index >= 0) {
1717 dl->disk = *__get_imsm_disk(super->anchor, dl->index);
1718 if (dl->disk.status & FAILED_DISK)
1719 dl->index = -2;
1720 }
1721
1722 if (alloc)
1723 super->disks = dl;
1724
1725 return 0;
1726 }
1727
1728 #ifndef MDASSEMBLE
1729 /* When migrating map0 contains the 'destination' state while map1
1730 * contains the current state. When not migrating map0 contains the
1731 * current state. This routine assumes that map[0].map_state is set to
1732 * the current array state before being called.
1733 *
1734 * Migration is indicated by one of the following states
1735 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1736 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1737 * map1state=unitialized)
1738 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
1739 * map1state=normal)
1740 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1741 * map1state=degraded)
1742 */
1743 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
1744 {
1745 struct imsm_map *dest;
1746 struct imsm_map *src = get_imsm_map(dev, 0);
1747
1748 dev->vol.migr_state = 1;
1749 set_migr_type(dev, migr_type);
1750 dev->vol.curr_migr_unit = 0;
1751 dest = get_imsm_map(dev, 1);
1752
1753 /* duplicate and then set the target end state in map[0] */
1754 memcpy(dest, src, sizeof_imsm_map(src));
1755 if (migr_type == MIGR_REBUILD) {
1756 __u32 ord;
1757 int i;
1758
1759 for (i = 0; i < src->num_members; i++) {
1760 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
1761 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
1762 }
1763 }
1764
1765 src->map_state = to_state;
1766 }
1767
1768 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1769 {
1770 struct imsm_map *map = get_imsm_map(dev, 0);
1771 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
1772 int i;
1773
1774 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1775 * completed in the last migration.
1776 *
1777 * FIXME add support for online capacity expansion and
1778 * raid-level-migration
1779 */
1780 for (i = 0; i < prev->num_members; i++)
1781 map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i];
1782
1783 dev->vol.migr_state = 0;
1784 dev->vol.curr_migr_unit = 0;
1785 map->map_state = map_state;
1786 }
1787 #endif
1788
1789 static int parse_raid_devices(struct intel_super *super)
1790 {
1791 int i;
1792 struct imsm_dev *dev_new;
1793 size_t len, len_migr;
1794 size_t space_needed = 0;
1795 struct imsm_super *mpb = super->anchor;
1796
1797 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1798 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1799 struct intel_dev *dv;
1800
1801 len = sizeof_imsm_dev(dev_iter, 0);
1802 len_migr = sizeof_imsm_dev(dev_iter, 1);
1803 if (len_migr > len)
1804 space_needed += len_migr - len;
1805
1806 dv = malloc(sizeof(*dv));
1807 if (!dv)
1808 return 1;
1809 dev_new = malloc(len_migr);
1810 if (!dev_new) {
1811 free(dv);
1812 return 1;
1813 }
1814 imsm_copy_dev(dev_new, dev_iter);
1815 dv->dev = dev_new;
1816 dv->index = i;
1817 dv->next = super->devlist;
1818 super->devlist = dv;
1819 }
1820
1821 /* ensure that super->buf is large enough when all raid devices
1822 * are migrating
1823 */
1824 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1825 void *buf;
1826
1827 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1828 if (posix_memalign(&buf, 512, len) != 0)
1829 return 1;
1830
1831 memcpy(buf, super->buf, super->len);
1832 memset(buf + super->len, 0, len - super->len);
1833 free(super->buf);
1834 super->buf = buf;
1835 super->len = len;
1836 }
1837
1838 return 0;
1839 }
1840
1841 /* retrieve a pointer to the bbm log which starts after all raid devices */
1842 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1843 {
1844 void *ptr = NULL;
1845
1846 if (__le32_to_cpu(mpb->bbm_log_size)) {
1847 ptr = mpb;
1848 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1849 }
1850
1851 return ptr;
1852 }
1853
1854 static void __free_imsm(struct intel_super *super, int free_disks);
1855
1856 /* load_imsm_mpb - read matrix metadata
1857 * allocates super->mpb to be freed by free_super
1858 */
1859 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1860 {
1861 unsigned long long dsize;
1862 unsigned long long sectors;
1863 struct stat;
1864 struct imsm_super *anchor;
1865 __u32 check_sum;
1866 int rc;
1867
1868 get_dev_size(fd, NULL, &dsize);
1869
1870 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1871 if (devname)
1872 fprintf(stderr,
1873 Name ": Cannot seek to anchor block on %s: %s\n",
1874 devname, strerror(errno));
1875 return 1;
1876 }
1877
1878 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1879 if (devname)
1880 fprintf(stderr,
1881 Name ": Failed to allocate imsm anchor buffer"
1882 " on %s\n", devname);
1883 return 1;
1884 }
1885 if (read(fd, anchor, 512) != 512) {
1886 if (devname)
1887 fprintf(stderr,
1888 Name ": Cannot read anchor block on %s: %s\n",
1889 devname, strerror(errno));
1890 free(anchor);
1891 return 1;
1892 }
1893
1894 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1895 if (devname)
1896 fprintf(stderr,
1897 Name ": no IMSM anchor on %s\n", devname);
1898 free(anchor);
1899 return 2;
1900 }
1901
1902 __free_imsm(super, 0);
1903 super->len = ROUND_UP(anchor->mpb_size, 512);
1904 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1905 if (devname)
1906 fprintf(stderr,
1907 Name ": unable to allocate %zu byte mpb buffer\n",
1908 super->len);
1909 free(anchor);
1910 return 2;
1911 }
1912 memcpy(super->buf, anchor, 512);
1913
1914 sectors = mpb_sectors(anchor) - 1;
1915 free(anchor);
1916 if (!sectors) {
1917 check_sum = __gen_imsm_checksum(super->anchor);
1918 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1919 if (devname)
1920 fprintf(stderr,
1921 Name ": IMSM checksum %x != %x on %s\n",
1922 check_sum,
1923 __le32_to_cpu(super->anchor->check_sum),
1924 devname);
1925 return 2;
1926 }
1927
1928 rc = load_imsm_disk(fd, super, devname, 0);
1929 if (rc == 0)
1930 rc = parse_raid_devices(super);
1931 return rc;
1932 }
1933
1934 /* read the extended mpb */
1935 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1936 if (devname)
1937 fprintf(stderr,
1938 Name ": Cannot seek to extended mpb on %s: %s\n",
1939 devname, strerror(errno));
1940 return 1;
1941 }
1942
1943 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1944 if (devname)
1945 fprintf(stderr,
1946 Name ": Cannot read extended mpb on %s: %s\n",
1947 devname, strerror(errno));
1948 return 2;
1949 }
1950
1951 check_sum = __gen_imsm_checksum(super->anchor);
1952 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1953 if (devname)
1954 fprintf(stderr,
1955 Name ": IMSM checksum %x != %x on %s\n",
1956 check_sum, __le32_to_cpu(super->anchor->check_sum),
1957 devname);
1958 return 3;
1959 }
1960
1961 /* FIXME the BBM log is disk specific so we cannot use this global
1962 * buffer for all disks. Ok for now since we only look at the global
1963 * bbm_log_size parameter to gate assembly
1964 */
1965 super->bbm_log = __get_imsm_bbm_log(super->anchor);
1966
1967 rc = load_imsm_disk(fd, super, devname, 0);
1968 if (rc == 0)
1969 rc = parse_raid_devices(super);
1970
1971 return rc;
1972 }
1973
1974 static void __free_imsm_disk(struct dl *d)
1975 {
1976 if (d->fd >= 0)
1977 close(d->fd);
1978 if (d->devname)
1979 free(d->devname);
1980 if (d->e)
1981 free(d->e);
1982 free(d);
1983
1984 }
1985 static void free_imsm_disks(struct intel_super *super)
1986 {
1987 struct dl *d;
1988
1989 while (super->disks) {
1990 d = super->disks;
1991 super->disks = d->next;
1992 __free_imsm_disk(d);
1993 }
1994 while (super->missing) {
1995 d = super->missing;
1996 super->missing = d->next;
1997 __free_imsm_disk(d);
1998 }
1999
2000 }
2001
2002 /* free all the pieces hanging off of a super pointer */
2003 static void __free_imsm(struct intel_super *super, int free_disks)
2004 {
2005 if (super->buf) {
2006 free(super->buf);
2007 super->buf = NULL;
2008 }
2009 if (free_disks)
2010 free_imsm_disks(super);
2011 free_devlist(super);
2012 if (super->hba) {
2013 free((void *) super->hba);
2014 super->hba = NULL;
2015 }
2016 }
2017
2018 static void free_imsm(struct intel_super *super)
2019 {
2020 __free_imsm(super, 1);
2021 free(super);
2022 }
2023
2024 static void free_super_imsm(struct supertype *st)
2025 {
2026 struct intel_super *super = st->sb;
2027
2028 if (!super)
2029 return;
2030
2031 free_imsm(super);
2032 st->sb = NULL;
2033 }
2034
2035 static struct intel_super *alloc_super(int creating_imsm)
2036 {
2037 struct intel_super *super = malloc(sizeof(*super));
2038
2039 if (super) {
2040 memset(super, 0, sizeof(*super));
2041 super->creating_imsm = creating_imsm;
2042 super->current_vol = -1;
2043 super->create_offset = ~((__u32 ) 0);
2044 if (!check_env("IMSM_NO_PLATFORM"))
2045 super->orom = find_imsm_orom();
2046 if (super->orom && !check_env("IMSM_TEST_OROM")) {
2047 struct sys_dev *list, *ent;
2048
2049 /* find the first intel ahci controller */
2050 list = find_driver_devices("pci", "ahci");
2051 for (ent = list; ent; ent = ent->next)
2052 if (devpath_to_vendor(ent->path) == 0x8086)
2053 break;
2054 if (ent) {
2055 super->hba = ent->path;
2056 ent->path = NULL;
2057 }
2058 free_sys_dev(&list);
2059 }
2060 }
2061
2062 return super;
2063 }
2064
2065 #ifndef MDASSEMBLE
2066 /* find_missing - helper routine for load_super_imsm_all that identifies
2067 * disks that have disappeared from the system. This routine relies on
2068 * the mpb being uptodate, which it is at load time.
2069 */
2070 static int find_missing(struct intel_super *super)
2071 {
2072 int i;
2073 struct imsm_super *mpb = super->anchor;
2074 struct dl *dl;
2075 struct imsm_disk *disk;
2076
2077 for (i = 0; i < mpb->num_disks; i++) {
2078 disk = __get_imsm_disk(mpb, i);
2079 dl = serial_to_dl(disk->serial, super);
2080 if (dl)
2081 continue;
2082
2083 dl = malloc(sizeof(*dl));
2084 if (!dl)
2085 return 1;
2086 dl->major = 0;
2087 dl->minor = 0;
2088 dl->fd = -1;
2089 dl->devname = strdup("missing");
2090 dl->index = i;
2091 serialcpy(dl->serial, disk->serial);
2092 dl->disk = *disk;
2093 dl->e = NULL;
2094 dl->next = super->missing;
2095 super->missing = dl;
2096 }
2097
2098 return 0;
2099 }
2100
2101 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
2102 char *devname, int keep_fd)
2103 {
2104 struct mdinfo *sra;
2105 struct intel_super *super;
2106 struct mdinfo *sd, *best = NULL;
2107 __u32 bestgen = 0;
2108 __u32 gen;
2109 char nm[20];
2110 int dfd;
2111 int rv;
2112 int devnum = fd2devnum(fd);
2113 int retry;
2114 enum sysfs_read_flags flags;
2115
2116 flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2117 if (mdmon_running(devnum))
2118 flags |= SKIP_GONE_DEVS;
2119
2120 /* check if 'fd' an opened container */
2121 sra = sysfs_read(fd, 0, flags);
2122 if (!sra)
2123 return 1;
2124
2125 if (sra->array.major_version != -1 ||
2126 sra->array.minor_version != -2 ||
2127 strcmp(sra->text_version, "imsm") != 0)
2128 return 1;
2129
2130 super = alloc_super(0);
2131 if (!super)
2132 return 1;
2133
2134 /* find the most up to date disk in this array, skipping spares */
2135 for (sd = sra->devs; sd; sd = sd->next) {
2136 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2137 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
2138 if (dfd < 0) {
2139 free_imsm(super);
2140 return 2;
2141 }
2142 rv = load_imsm_mpb(dfd, super, NULL);
2143
2144 /* retry the load if we might have raced against mdmon */
2145 if (rv == 3 && mdmon_running(devnum))
2146 for (retry = 0; retry < 3; retry++) {
2147 usleep(3000);
2148 rv = load_imsm_mpb(dfd, super, NULL);
2149 if (rv != 3)
2150 break;
2151 }
2152 if (!keep_fd)
2153 close(dfd);
2154 if (rv == 0) {
2155 if (super->anchor->num_raid_devs == 0)
2156 gen = 0;
2157 else
2158 gen = __le32_to_cpu(super->anchor->generation_num);
2159 if (!best || gen > bestgen) {
2160 bestgen = gen;
2161 best = sd;
2162 }
2163 } else {
2164 free_imsm(super);
2165 return rv;
2166 }
2167 }
2168
2169 if (!best) {
2170 free_imsm(super);
2171 return 1;
2172 }
2173
2174 /* load the most up to date anchor */
2175 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2176 dfd = dev_open(nm, O_RDONLY);
2177 if (dfd < 0) {
2178 free_imsm(super);
2179 return 1;
2180 }
2181 rv = load_imsm_mpb(dfd, super, NULL);
2182 close(dfd);
2183 if (rv != 0) {
2184 free_imsm(super);
2185 return 2;
2186 }
2187
2188 /* re-parse the disk list with the current anchor */
2189 for (sd = sra->devs ; sd ; sd = sd->next) {
2190 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2191 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2192 if (dfd < 0) {
2193 free_imsm(super);
2194 return 2;
2195 }
2196 load_imsm_disk(dfd, super, NULL, keep_fd);
2197 if (!keep_fd)
2198 close(dfd);
2199 }
2200
2201
2202 if (find_missing(super) != 0) {
2203 free_imsm(super);
2204 return 2;
2205 }
2206
2207 if (st->subarray[0]) {
2208 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2209 super->current_vol = atoi(st->subarray);
2210 else {
2211 free_imsm(super);
2212 return 1;
2213 }
2214 }
2215
2216 *sbp = super;
2217 st->container_dev = devnum;
2218 if (st->ss == NULL) {
2219 st->ss = &super_imsm;
2220 st->minor_version = 0;
2221 st->max_devs = IMSM_MAX_DEVICES;
2222 }
2223 st->loaded_container = 1;
2224
2225 return 0;
2226 }
2227 #endif
2228
2229 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2230 {
2231 struct intel_super *super;
2232 int rv;
2233
2234 #ifndef MDASSEMBLE
2235 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2236 return 0;
2237 #endif
2238
2239 free_super_imsm(st);
2240
2241 super = alloc_super(0);
2242 if (!super) {
2243 fprintf(stderr,
2244 Name ": malloc of %zu failed.\n",
2245 sizeof(*super));
2246 return 1;
2247 }
2248
2249 rv = load_imsm_mpb(fd, super, devname);
2250
2251 if (rv) {
2252 if (devname)
2253 fprintf(stderr,
2254 Name ": Failed to load all information "
2255 "sections on %s\n", devname);
2256 free_imsm(super);
2257 return rv;
2258 }
2259
2260 if (st->subarray[0]) {
2261 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2262 super->current_vol = atoi(st->subarray);
2263 else {
2264 free_imsm(super);
2265 return 1;
2266 }
2267 }
2268
2269 st->sb = super;
2270 if (st->ss == NULL) {
2271 st->ss = &super_imsm;
2272 st->minor_version = 0;
2273 st->max_devs = IMSM_MAX_DEVICES;
2274 }
2275 st->loaded_container = 0;
2276
2277 return 0;
2278 }
2279
2280 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2281 {
2282 if (info->level == 1)
2283 return 128;
2284 return info->chunk_size >> 9;
2285 }
2286
2287 static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
2288 {
2289 __u32 num_stripes;
2290
2291 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2292 num_stripes /= num_domains;
2293
2294 return num_stripes;
2295 }
2296
2297 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2298 {
2299 if (info->level == 1)
2300 return info->size * 2;
2301 else
2302 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2303 }
2304
2305 static void imsm_update_version_info(struct intel_super *super)
2306 {
2307 /* update the version and attributes */
2308 struct imsm_super *mpb = super->anchor;
2309 char *version;
2310 struct imsm_dev *dev;
2311 struct imsm_map *map;
2312 int i;
2313
2314 for (i = 0; i < mpb->num_raid_devs; i++) {
2315 dev = get_imsm_dev(super, i);
2316 map = get_imsm_map(dev, 0);
2317 if (__le32_to_cpu(dev->size_high) > 0)
2318 mpb->attributes |= MPB_ATTRIB_2TB;
2319
2320 /* FIXME detect when an array spans a port multiplier */
2321 #if 0
2322 mpb->attributes |= MPB_ATTRIB_PM;
2323 #endif
2324
2325 if (mpb->num_raid_devs > 1 ||
2326 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2327 version = MPB_VERSION_ATTRIBS;
2328 switch (get_imsm_raid_level(map)) {
2329 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2330 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2331 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2332 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2333 }
2334 } else {
2335 if (map->num_members >= 5)
2336 version = MPB_VERSION_5OR6_DISK_ARRAY;
2337 else if (dev->status == DEV_CLONE_N_GO)
2338 version = MPB_VERSION_CNG;
2339 else if (get_imsm_raid_level(map) == 5)
2340 version = MPB_VERSION_RAID5;
2341 else if (map->num_members >= 3)
2342 version = MPB_VERSION_3OR4_DISK_ARRAY;
2343 else if (get_imsm_raid_level(map) == 1)
2344 version = MPB_VERSION_RAID1;
2345 else
2346 version = MPB_VERSION_RAID0;
2347 }
2348 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2349 }
2350 }
2351
2352 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2353 unsigned long long size, char *name,
2354 char *homehost, int *uuid)
2355 {
2356 /* We are creating a volume inside a pre-existing container.
2357 * so st->sb is already set.
2358 */
2359 struct intel_super *super = st->sb;
2360 struct imsm_super *mpb = super->anchor;
2361 struct intel_dev *dv;
2362 struct imsm_dev *dev;
2363 struct imsm_vol *vol;
2364 struct imsm_map *map;
2365 int idx = mpb->num_raid_devs;
2366 int i;
2367 unsigned long long array_blocks;
2368 size_t size_old, size_new;
2369 __u32 num_data_stripes;
2370
2371 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2372 fprintf(stderr, Name": This imsm-container already has the "
2373 "maximum of %d volumes\n", super->orom->vpa);
2374 return 0;
2375 }
2376
2377 /* ensure the mpb is large enough for the new data */
2378 size_old = __le32_to_cpu(mpb->mpb_size);
2379 size_new = disks_to_mpb_size(info->nr_disks);
2380 if (size_new > size_old) {
2381 void *mpb_new;
2382 size_t size_round = ROUND_UP(size_new, 512);
2383
2384 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2385 fprintf(stderr, Name": could not allocate new mpb\n");
2386 return 0;
2387 }
2388 memcpy(mpb_new, mpb, size_old);
2389 free(mpb);
2390 mpb = mpb_new;
2391 super->anchor = mpb_new;
2392 mpb->mpb_size = __cpu_to_le32(size_new);
2393 memset(mpb_new + size_old, 0, size_round - size_old);
2394 }
2395 super->current_vol = idx;
2396 /* when creating the first raid device in this container set num_disks
2397 * to zero, i.e. delete this spare and add raid member devices in
2398 * add_to_super_imsm_volume()
2399 */
2400 if (super->current_vol == 0)
2401 mpb->num_disks = 0;
2402
2403 for (i = 0; i < super->current_vol; i++) {
2404 dev = get_imsm_dev(super, i);
2405 if (strncmp((char *) dev->volume, name,
2406 MAX_RAID_SERIAL_LEN) == 0) {
2407 fprintf(stderr, Name": '%s' is already defined for this container\n",
2408 name);
2409 return 0;
2410 }
2411 }
2412
2413 sprintf(st->subarray, "%d", idx);
2414 dv = malloc(sizeof(*dv));
2415 if (!dv) {
2416 fprintf(stderr, Name ": failed to allocate device list entry\n");
2417 return 0;
2418 }
2419 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2420 if (!dev) {
2421 free(dv);
2422 fprintf(stderr, Name": could not allocate raid device\n");
2423 return 0;
2424 }
2425 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2426 if (info->level == 1)
2427 array_blocks = info_to_blocks_per_member(info);
2428 else
2429 array_blocks = calc_array_size(info->level, info->raid_disks,
2430 info->layout, info->chunk_size,
2431 info->size*2);
2432 /* round array size down to closest MB */
2433 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
2434
2435 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2436 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2437 dev->status = __cpu_to_le32(0);
2438 dev->reserved_blocks = __cpu_to_le32(0);
2439 vol = &dev->vol;
2440 vol->migr_state = 0;
2441 set_migr_type(dev, MIGR_INIT);
2442 vol->dirty = 0;
2443 vol->curr_migr_unit = 0;
2444 map = get_imsm_map(dev, 0);
2445 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2446 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2447 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2448 map->failed_disk_num = ~0;
2449 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2450 IMSM_T_STATE_NORMAL;
2451 map->ddf = 1;
2452
2453 if (info->level == 1 && info->raid_disks > 2) {
2454 fprintf(stderr, Name": imsm does not support more than 2 disks"
2455 "in a raid1 volume\n");
2456 return 0;
2457 }
2458
2459 map->raid_level = info->level;
2460 if (info->level == 10) {
2461 map->raid_level = 1;
2462 map->num_domains = info->raid_disks / 2;
2463 } else if (info->level == 1)
2464 map->num_domains = info->raid_disks;
2465 else
2466 map->num_domains = 1;
2467
2468 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
2469 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
2470
2471 map->num_members = info->raid_disks;
2472 for (i = 0; i < map->num_members; i++) {
2473 /* initialized in add_to_super */
2474 set_imsm_ord_tbl_ent(map, i, 0);
2475 }
2476 mpb->num_raid_devs++;
2477
2478 dv->dev = dev;
2479 dv->index = super->current_vol;
2480 dv->next = super->devlist;
2481 super->devlist = dv;
2482
2483 imsm_update_version_info(super);
2484
2485 return 1;
2486 }
2487
2488 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2489 unsigned long long size, char *name,
2490 char *homehost, int *uuid)
2491 {
2492 /* This is primarily called by Create when creating a new array.
2493 * We will then get add_to_super called for each component, and then
2494 * write_init_super called to write it out to each device.
2495 * For IMSM, Create can create on fresh devices or on a pre-existing
2496 * array.
2497 * To create on a pre-existing array a different method will be called.
2498 * This one is just for fresh drives.
2499 */
2500 struct intel_super *super;
2501 struct imsm_super *mpb;
2502 size_t mpb_size;
2503 char *version;
2504
2505 if (!info) {
2506 st->sb = NULL;
2507 return 0;
2508 }
2509 if (st->sb)
2510 return init_super_imsm_volume(st, info, size, name, homehost,
2511 uuid);
2512
2513 super = alloc_super(1);
2514 if (!super)
2515 return 0;
2516 mpb_size = disks_to_mpb_size(info->nr_disks);
2517 if (posix_memalign(&super->buf, 512, mpb_size) != 0) {
2518 free(super);
2519 return 0;
2520 }
2521 mpb = super->buf;
2522 memset(mpb, 0, mpb_size);
2523
2524 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2525
2526 version = (char *) mpb->sig;
2527 strcpy(version, MPB_SIGNATURE);
2528 version += strlen(MPB_SIGNATURE);
2529 strcpy(version, MPB_VERSION_RAID0);
2530 mpb->mpb_size = mpb_size;
2531
2532 st->sb = super;
2533 return 1;
2534 }
2535
2536 #ifndef MDASSEMBLE
2537 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2538 int fd, char *devname)
2539 {
2540 struct intel_super *super = st->sb;
2541 struct imsm_super *mpb = super->anchor;
2542 struct dl *dl;
2543 struct imsm_dev *dev;
2544 struct imsm_map *map;
2545
2546 dev = get_imsm_dev(super, super->current_vol);
2547 map = get_imsm_map(dev, 0);
2548
2549 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2550 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2551 devname);
2552 return 1;
2553 }
2554
2555 if (fd == -1) {
2556 /* we're doing autolayout so grab the pre-marked (in
2557 * validate_geometry) raid_disk
2558 */
2559 for (dl = super->disks; dl; dl = dl->next)
2560 if (dl->raiddisk == dk->raid_disk)
2561 break;
2562 } else {
2563 for (dl = super->disks; dl ; dl = dl->next)
2564 if (dl->major == dk->major &&
2565 dl->minor == dk->minor)
2566 break;
2567 }
2568
2569 if (!dl) {
2570 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2571 return 1;
2572 }
2573
2574 /* add a pristine spare to the metadata */
2575 if (dl->index < 0) {
2576 dl->index = super->anchor->num_disks;
2577 super->anchor->num_disks++;
2578 }
2579 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2580 dl->disk.status = CONFIGURED_DISK;
2581
2582 /* if we are creating the first raid device update the family number */
2583 if (super->current_vol == 0) {
2584 __u32 sum;
2585 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2586 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2587
2588 *_dev = *dev;
2589 *_disk = dl->disk;
2590 sum = random32();
2591 sum += __gen_imsm_checksum(mpb);
2592 mpb->family_num = __cpu_to_le32(sum);
2593 mpb->orig_family_num = mpb->family_num;
2594 }
2595
2596 return 0;
2597 }
2598
2599 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2600 int fd, char *devname)
2601 {
2602 struct intel_super *super = st->sb;
2603 struct dl *dd;
2604 unsigned long long size;
2605 __u32 id;
2606 int rv;
2607 struct stat stb;
2608
2609 /* if we are on an RAID enabled platform check that the disk is
2610 * attached to the raid controller
2611 */
2612 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2613 fprintf(stderr,
2614 Name ": %s is not attached to the raid controller: %s\n",
2615 devname ? : "disk", super->hba);
2616 return 1;
2617 }
2618
2619 if (super->current_vol >= 0)
2620 return add_to_super_imsm_volume(st, dk, fd, devname);
2621
2622 fstat(fd, &stb);
2623 dd = malloc(sizeof(*dd));
2624 if (!dd) {
2625 fprintf(stderr,
2626 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2627 return 1;
2628 }
2629 memset(dd, 0, sizeof(*dd));
2630 dd->major = major(stb.st_rdev);
2631 dd->minor = minor(stb.st_rdev);
2632 dd->index = -1;
2633 dd->devname = devname ? strdup(devname) : NULL;
2634 dd->fd = fd;
2635 dd->e = NULL;
2636 rv = imsm_read_serial(fd, devname, dd->serial);
2637 if (rv) {
2638 fprintf(stderr,
2639 Name ": failed to retrieve scsi serial, aborting\n");
2640 free(dd);
2641 abort();
2642 }
2643
2644 get_dev_size(fd, NULL, &size);
2645 size /= 512;
2646 serialcpy(dd->disk.serial, dd->serial);
2647 dd->disk.total_blocks = __cpu_to_le32(size);
2648 dd->disk.status = SPARE_DISK;
2649 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
2650 dd->disk.scsi_id = __cpu_to_le32(id);
2651 else
2652 dd->disk.scsi_id = __cpu_to_le32(0);
2653
2654 if (st->update_tail) {
2655 dd->next = super->add;
2656 super->add = dd;
2657 } else {
2658 dd->next = super->disks;
2659 super->disks = dd;
2660 }
2661
2662 return 0;
2663 }
2664
2665 static int store_imsm_mpb(int fd, struct intel_super *super);
2666
2667 /* spare records have their own family number and do not have any defined raid
2668 * devices
2669 */
2670 static int write_super_imsm_spares(struct intel_super *super, int doclose)
2671 {
2672 struct imsm_super mpb_save;
2673 struct imsm_super *mpb = super->anchor;
2674 __u32 sum;
2675 struct dl *d;
2676
2677 mpb_save = *mpb;
2678 mpb->num_raid_devs = 0;
2679 mpb->num_disks = 1;
2680 mpb->mpb_size = sizeof(struct imsm_super);
2681 mpb->generation_num = __cpu_to_le32(1UL);
2682
2683 for (d = super->disks; d; d = d->next) {
2684 if (d->index != -1)
2685 continue;
2686
2687 mpb->disk[0] = d->disk;
2688 sum = __gen_imsm_checksum(mpb);
2689 mpb->family_num = __cpu_to_le32(sum);
2690 mpb->orig_family_num = 0;
2691 sum = __gen_imsm_checksum(mpb);
2692 mpb->check_sum = __cpu_to_le32(sum);
2693
2694 if (store_imsm_mpb(d->fd, super)) {
2695 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2696 __func__, d->major, d->minor, strerror(errno));
2697 *mpb = mpb_save;
2698 return 1;
2699 }
2700 if (doclose) {
2701 close(d->fd);
2702 d->fd = -1;
2703 }
2704 }
2705
2706 *mpb = mpb_save;
2707 return 0;
2708 }
2709
2710 static int write_super_imsm(struct intel_super *super, int doclose)
2711 {
2712 struct imsm_super *mpb = super->anchor;
2713 struct dl *d;
2714 __u32 generation;
2715 __u32 sum;
2716 int spares = 0;
2717 int i;
2718 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
2719
2720 /* 'generation' is incremented everytime the metadata is written */
2721 generation = __le32_to_cpu(mpb->generation_num);
2722 generation++;
2723 mpb->generation_num = __cpu_to_le32(generation);
2724
2725 /* fix up cases where previous mdadm releases failed to set
2726 * orig_family_num
2727 */
2728 if (mpb->orig_family_num == 0)
2729 mpb->orig_family_num = mpb->family_num;
2730
2731 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
2732 for (d = super->disks; d; d = d->next) {
2733 if (d->index == -1)
2734 spares++;
2735 else
2736 mpb->disk[d->index] = d->disk;
2737 }
2738 for (d = super->missing; d; d = d->next)
2739 mpb->disk[d->index] = d->disk;
2740
2741 for (i = 0; i < mpb->num_raid_devs; i++) {
2742 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
2743
2744 imsm_copy_dev(dev, get_imsm_dev(super, i));
2745 mpb_size += sizeof_imsm_dev(dev, 0);
2746 }
2747 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
2748 mpb->mpb_size = __cpu_to_le32(mpb_size);
2749
2750 /* recalculate checksum */
2751 sum = __gen_imsm_checksum(mpb);
2752 mpb->check_sum = __cpu_to_le32(sum);
2753
2754 /* write the mpb for disks that compose raid devices */
2755 for (d = super->disks; d ; d = d->next) {
2756 if (d->index < 0)
2757 continue;
2758 if (store_imsm_mpb(d->fd, super))
2759 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2760 __func__, d->major, d->minor, strerror(errno));
2761 if (doclose) {
2762 close(d->fd);
2763 d->fd = -1;
2764 }
2765 }
2766
2767 if (spares)
2768 return write_super_imsm_spares(super, doclose);
2769
2770 return 0;
2771 }
2772
2773
2774 static int create_array(struct supertype *st, int dev_idx)
2775 {
2776 size_t len;
2777 struct imsm_update_create_array *u;
2778 struct intel_super *super = st->sb;
2779 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
2780 struct imsm_map *map = get_imsm_map(dev, 0);
2781 struct disk_info *inf;
2782 struct imsm_disk *disk;
2783 int i;
2784
2785 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
2786 sizeof(*inf) * map->num_members;
2787 u = malloc(len);
2788 if (!u) {
2789 fprintf(stderr, "%s: failed to allocate update buffer\n",
2790 __func__);
2791 return 1;
2792 }
2793
2794 u->type = update_create_array;
2795 u->dev_idx = dev_idx;
2796 imsm_copy_dev(&u->dev, dev);
2797 inf = get_disk_info(u);
2798 for (i = 0; i < map->num_members; i++) {
2799 int idx = get_imsm_disk_idx(dev, i);
2800
2801 disk = get_imsm_disk(super, idx);
2802 serialcpy(inf[i].serial, disk->serial);
2803 }
2804 append_metadata_update(st, u, len);
2805
2806 return 0;
2807 }
2808
2809 static int _add_disk(struct supertype *st)
2810 {
2811 struct intel_super *super = st->sb;
2812 size_t len;
2813 struct imsm_update_add_disk *u;
2814
2815 if (!super->add)
2816 return 0;
2817
2818 len = sizeof(*u);
2819 u = malloc(len);
2820 if (!u) {
2821 fprintf(stderr, "%s: failed to allocate update buffer\n",
2822 __func__);
2823 return 1;
2824 }
2825
2826 u->type = update_add_disk;
2827 append_metadata_update(st, u, len);
2828
2829 return 0;
2830 }
2831
2832 static int write_init_super_imsm(struct supertype *st)
2833 {
2834 struct intel_super *super = st->sb;
2835 int current_vol = super->current_vol;
2836
2837 /* we are done with current_vol reset it to point st at the container */
2838 super->current_vol = -1;
2839
2840 if (st->update_tail) {
2841 /* queue the recently created array / added disk
2842 * as a metadata update */
2843 struct dl *d;
2844 int rv;
2845
2846 /* determine if we are creating a volume or adding a disk */
2847 if (current_vol < 0) {
2848 /* in the add disk case we are running in mdmon
2849 * context, so don't close fd's
2850 */
2851 return _add_disk(st);
2852 } else
2853 rv = create_array(st, current_vol);
2854
2855 for (d = super->disks; d ; d = d->next) {
2856 close(d->fd);
2857 d->fd = -1;
2858 }
2859
2860 return rv;
2861 } else
2862 return write_super_imsm(st->sb, 1);
2863 }
2864 #endif
2865
2866 static int store_zero_imsm(struct supertype *st, int fd)
2867 {
2868 unsigned long long dsize;
2869 void *buf;
2870
2871 get_dev_size(fd, NULL, &dsize);
2872
2873 /* first block is stored on second to last sector of the disk */
2874 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
2875 return 1;
2876
2877 if (posix_memalign(&buf, 512, 512) != 0)
2878 return 1;
2879
2880 memset(buf, 0, 512);
2881 if (write(fd, buf, 512) != 512)
2882 return 1;
2883 return 0;
2884 }
2885
2886 static int imsm_bbm_log_size(struct imsm_super *mpb)
2887 {
2888 return __le32_to_cpu(mpb->bbm_log_size);
2889 }
2890
2891 #ifndef MDASSEMBLE
2892 static int validate_geometry_imsm_container(struct supertype *st, int level,
2893 int layout, int raiddisks, int chunk,
2894 unsigned long long size, char *dev,
2895 unsigned long long *freesize,
2896 int verbose)
2897 {
2898 int fd;
2899 unsigned long long ldsize;
2900 const struct imsm_orom *orom;
2901
2902 if (level != LEVEL_CONTAINER)
2903 return 0;
2904 if (!dev)
2905 return 1;
2906
2907 if (check_env("IMSM_NO_PLATFORM"))
2908 orom = NULL;
2909 else
2910 orom = find_imsm_orom();
2911 if (orom && raiddisks > orom->tds) {
2912 if (verbose)
2913 fprintf(stderr, Name ": %d exceeds maximum number of"
2914 " platform supported disks: %d\n",
2915 raiddisks, orom->tds);
2916 return 0;
2917 }
2918
2919 fd = open(dev, O_RDONLY|O_EXCL, 0);
2920 if (fd < 0) {
2921 if (verbose)
2922 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
2923 dev, strerror(errno));
2924 return 0;
2925 }
2926 if (!get_dev_size(fd, dev, &ldsize)) {
2927 close(fd);
2928 return 0;
2929 }
2930 close(fd);
2931
2932 *freesize = avail_size_imsm(st, ldsize >> 9);
2933
2934 return 1;
2935 }
2936
2937 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
2938 {
2939 const unsigned long long base_start = e[*idx].start;
2940 unsigned long long end = base_start + e[*idx].size;
2941 int i;
2942
2943 if (base_start == end)
2944 return 0;
2945
2946 *idx = *idx + 1;
2947 for (i = *idx; i < num_extents; i++) {
2948 /* extend overlapping extents */
2949 if (e[i].start >= base_start &&
2950 e[i].start <= end) {
2951 if (e[i].size == 0)
2952 return 0;
2953 if (e[i].start + e[i].size > end)
2954 end = e[i].start + e[i].size;
2955 } else if (e[i].start > end) {
2956 *idx = i;
2957 break;
2958 }
2959 }
2960
2961 return end - base_start;
2962 }
2963
2964 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
2965 {
2966 /* build a composite disk with all known extents and generate a new
2967 * 'maxsize' given the "all disks in an array must share a common start
2968 * offset" constraint
2969 */
2970 struct extent *e = calloc(sum_extents, sizeof(*e));
2971 struct dl *dl;
2972 int i, j;
2973 int start_extent;
2974 unsigned long long pos;
2975 unsigned long long start = 0;
2976 unsigned long long maxsize;
2977 unsigned long reserve;
2978
2979 if (!e)
2980 return ~0ULL; /* error */
2981
2982 /* coalesce and sort all extents. also, check to see if we need to
2983 * reserve space between member arrays
2984 */
2985 j = 0;
2986 for (dl = super->disks; dl; dl = dl->next) {
2987 if (!dl->e)
2988 continue;
2989 for (i = 0; i < dl->extent_cnt; i++)
2990 e[j++] = dl->e[i];
2991 }
2992 qsort(e, sum_extents, sizeof(*e), cmp_extent);
2993
2994 /* merge extents */
2995 i = 0;
2996 j = 0;
2997 while (i < sum_extents) {
2998 e[j].start = e[i].start;
2999 e[j].size = find_size(e, &i, sum_extents);
3000 j++;
3001 if (e[j-1].size == 0)
3002 break;
3003 }
3004
3005 pos = 0;
3006 maxsize = 0;
3007 start_extent = 0;
3008 i = 0;
3009 do {
3010 unsigned long long esize;
3011
3012 esize = e[i].start - pos;
3013 if (esize >= maxsize) {
3014 maxsize = esize;
3015 start = pos;
3016 start_extent = i;
3017 }
3018 pos = e[i].start + e[i].size;
3019 i++;
3020 } while (e[i-1].size);
3021 free(e);
3022
3023 if (start_extent > 0)
3024 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
3025 else
3026 reserve = 0;
3027
3028 if (maxsize < reserve)
3029 return ~0ULL;
3030
3031 super->create_offset = ~((__u32) 0);
3032 if (start + reserve > super->create_offset)
3033 return ~0ULL; /* start overflows create_offset */
3034 super->create_offset = start + reserve;
3035
3036 return maxsize - reserve;
3037 }
3038
3039 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
3040 {
3041 if (level < 0 || level == 6 || level == 4)
3042 return 0;
3043
3044 /* if we have an orom prevent invalid raid levels */
3045 if (orom)
3046 switch (level) {
3047 case 0: return imsm_orom_has_raid0(orom);
3048 case 1:
3049 if (raiddisks > 2)
3050 return imsm_orom_has_raid1e(orom);
3051 return imsm_orom_has_raid1(orom) && raiddisks == 2;
3052 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
3053 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
3054 }
3055 else
3056 return 1; /* not on an Intel RAID platform so anything goes */
3057
3058 return 0;
3059 }
3060
3061 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3062 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3063 * FIX ME add ahci details
3064 */
3065 static int validate_geometry_imsm_volume(struct supertype *st, int level,
3066 int layout, int raiddisks, int chunk,
3067 unsigned long long size, char *dev,
3068 unsigned long long *freesize,
3069 int verbose)
3070 {
3071 struct stat stb;
3072 struct intel_super *super = st->sb;
3073 struct imsm_super *mpb = super->anchor;
3074 struct dl *dl;
3075 unsigned long long pos = 0;
3076 unsigned long long maxsize;
3077 struct extent *e;
3078 int i;
3079
3080 /* We must have the container info already read in. */
3081 if (!super)
3082 return 0;
3083
3084 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
3085 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3086 level, raiddisks, raiddisks > 1 ? "s" : "");
3087 return 0;
3088 }
3089 if (super->orom && level != 1 &&
3090 !imsm_orom_has_chunk(super->orom, chunk)) {
3091 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
3092 return 0;
3093 }
3094 if (layout != imsm_level_to_layout(level)) {
3095 if (level == 5)
3096 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3097 else if (level == 10)
3098 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3099 else
3100 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3101 layout, level);
3102 return 0;
3103 }
3104
3105 if (!dev) {
3106 /* General test: make sure there is space for
3107 * 'raiddisks' device extents of size 'size' at a given
3108 * offset
3109 */
3110 unsigned long long minsize = size;
3111 unsigned long long start_offset = ~0ULL;
3112 int dcnt = 0;
3113 if (minsize == 0)
3114 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
3115 for (dl = super->disks; dl ; dl = dl->next) {
3116 int found = 0;
3117
3118 pos = 0;
3119 i = 0;
3120 e = get_extents(super, dl);
3121 if (!e) continue;
3122 do {
3123 unsigned long long esize;
3124 esize = e[i].start - pos;
3125 if (esize >= minsize)
3126 found = 1;
3127 if (found && start_offset == ~0ULL) {
3128 start_offset = pos;
3129 break;
3130 } else if (found && pos != start_offset) {
3131 found = 0;
3132 break;
3133 }
3134 pos = e[i].start + e[i].size;
3135 i++;
3136 } while (e[i-1].size);
3137 if (found)
3138 dcnt++;
3139 free(e);
3140 }
3141 if (dcnt < raiddisks) {
3142 if (verbose)
3143 fprintf(stderr, Name ": imsm: Not enough "
3144 "devices with space for this array "
3145 "(%d < %d)\n",
3146 dcnt, raiddisks);
3147 return 0;
3148 }
3149 return 1;
3150 }
3151
3152 /* This device must be a member of the set */
3153 if (stat(dev, &stb) < 0)
3154 return 0;
3155 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3156 return 0;
3157 for (dl = super->disks ; dl ; dl = dl->next) {
3158 if (dl->major == major(stb.st_rdev) &&
3159 dl->minor == minor(stb.st_rdev))
3160 break;
3161 }
3162 if (!dl) {
3163 if (verbose)
3164 fprintf(stderr, Name ": %s is not in the "
3165 "same imsm set\n", dev);
3166 return 0;
3167 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
3168 /* If a volume is present then the current creation attempt
3169 * cannot incorporate new spares because the orom may not
3170 * understand this configuration (all member disks must be
3171 * members of each array in the container).
3172 */
3173 fprintf(stderr, Name ": %s is a spare and a volume"
3174 " is already defined for this container\n", dev);
3175 fprintf(stderr, Name ": The option-rom requires all member"
3176 " disks to be a member of all volumes\n");
3177 return 0;
3178 }
3179
3180 /* retrieve the largest free space block */
3181 e = get_extents(super, dl);
3182 maxsize = 0;
3183 i = 0;
3184 if (e) {
3185 do {
3186 unsigned long long esize;
3187
3188 esize = e[i].start - pos;
3189 if (esize >= maxsize)
3190 maxsize = esize;
3191 pos = e[i].start + e[i].size;
3192 i++;
3193 } while (e[i-1].size);
3194 dl->e = e;
3195 dl->extent_cnt = i;
3196 } else {
3197 if (verbose)
3198 fprintf(stderr, Name ": unable to determine free space for: %s\n",
3199 dev);
3200 return 0;
3201 }
3202 if (maxsize < size) {
3203 if (verbose)
3204 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
3205 dev, maxsize, size);
3206 return 0;
3207 }
3208
3209 /* count total number of extents for merge */
3210 i = 0;
3211 for (dl = super->disks; dl; dl = dl->next)
3212 if (dl->e)
3213 i += dl->extent_cnt;
3214
3215 maxsize = merge_extents(super, i);
3216 if (maxsize < size) {
3217 if (verbose)
3218 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
3219 maxsize, size);
3220 return 0;
3221 } else if (maxsize == ~0ULL) {
3222 if (verbose)
3223 fprintf(stderr, Name ": failed to merge %d extents\n", i);
3224 return 0;
3225 }
3226
3227 *freesize = maxsize;
3228
3229 return 1;
3230 }
3231
3232 static int reserve_space(struct supertype *st, int raiddisks,
3233 unsigned long long size, int chunk,
3234 unsigned long long *freesize)
3235 {
3236 struct intel_super *super = st->sb;
3237 struct imsm_super *mpb = super->anchor;
3238 struct dl *dl;
3239 int i;
3240 int extent_cnt;
3241 struct extent *e;
3242 unsigned long long maxsize;
3243 unsigned long long minsize;
3244 int cnt;
3245 int used;
3246
3247 /* find the largest common start free region of the possible disks */
3248 used = 0;
3249 extent_cnt = 0;
3250 cnt = 0;
3251 for (dl = super->disks; dl; dl = dl->next) {
3252 dl->raiddisk = -1;
3253
3254 if (dl->index >= 0)
3255 used++;
3256
3257 /* don't activate new spares if we are orom constrained
3258 * and there is already a volume active in the container
3259 */
3260 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
3261 continue;
3262
3263 e = get_extents(super, dl);
3264 if (!e)
3265 continue;
3266 for (i = 1; e[i-1].size; i++)
3267 ;
3268 dl->e = e;
3269 dl->extent_cnt = i;
3270 extent_cnt += i;
3271 cnt++;
3272 }
3273
3274 maxsize = merge_extents(super, extent_cnt);
3275 minsize = size;
3276 if (size == 0)
3277 minsize = chunk;
3278
3279 if (cnt < raiddisks ||
3280 (super->orom && used && used != raiddisks) ||
3281 maxsize < minsize) {
3282 fprintf(stderr, Name ": not enough devices with space to create array.\n");
3283 return 0; /* No enough free spaces large enough */
3284 }
3285
3286 if (size == 0) {
3287 size = maxsize;
3288 if (chunk) {
3289 size /= chunk;
3290 size *= chunk;
3291 }
3292 }
3293
3294 cnt = 0;
3295 for (dl = super->disks; dl; dl = dl->next)
3296 if (dl->e)
3297 dl->raiddisk = cnt++;
3298
3299 *freesize = size;
3300
3301 return 1;
3302 }
3303
3304 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
3305 int raiddisks, int chunk, unsigned long long size,
3306 char *dev, unsigned long long *freesize,
3307 int verbose)
3308 {
3309 int fd, cfd;
3310 struct mdinfo *sra;
3311
3312 /* if given unused devices create a container
3313 * if given given devices in a container create a member volume
3314 */
3315 if (level == LEVEL_CONTAINER) {
3316 /* Must be a fresh device to add to a container */
3317 return validate_geometry_imsm_container(st, level, layout,
3318 raiddisks, chunk, size,
3319 dev, freesize,
3320 verbose);
3321 }
3322
3323 if (!dev) {
3324 if (st->sb && freesize) {
3325 /* we are being asked to automatically layout a
3326 * new volume based on the current contents of
3327 * the container. If the the parameters can be
3328 * satisfied reserve_space will record the disks,
3329 * start offset, and size of the volume to be
3330 * created. add_to_super and getinfo_super
3331 * detect when autolayout is in progress.
3332 */
3333 return reserve_space(st, raiddisks, size, chunk, freesize);
3334 }
3335 return 1;
3336 }
3337 if (st->sb) {
3338 /* creating in a given container */
3339 return validate_geometry_imsm_volume(st, level, layout,
3340 raiddisks, chunk, size,
3341 dev, freesize, verbose);
3342 }
3343
3344 /* limit creation to the following levels */
3345 if (!dev)
3346 switch (level) {
3347 case 0:
3348 case 1:
3349 case 10:
3350 case 5:
3351 break;
3352 default:
3353 return 1;
3354 }
3355
3356 /* This device needs to be a device in an 'imsm' container */
3357 fd = open(dev, O_RDONLY|O_EXCL, 0);
3358 if (fd >= 0) {
3359 if (verbose)
3360 fprintf(stderr,
3361 Name ": Cannot create this array on device %s\n",
3362 dev);
3363 close(fd);
3364 return 0;
3365 }
3366 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3367 if (verbose)
3368 fprintf(stderr, Name ": Cannot open %s: %s\n",
3369 dev, strerror(errno));
3370 return 0;
3371 }
3372 /* Well, it is in use by someone, maybe an 'imsm' container. */
3373 cfd = open_container(fd);
3374 if (cfd < 0) {
3375 close(fd);
3376 if (verbose)
3377 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3378 dev);
3379 return 0;
3380 }
3381 sra = sysfs_read(cfd, 0, GET_VERSION);
3382 close(fd);
3383 if (sra && sra->array.major_version == -1 &&
3384 strcmp(sra->text_version, "imsm") == 0) {
3385 /* This is a member of a imsm container. Load the container
3386 * and try to create a volume
3387 */
3388 struct intel_super *super;
3389
3390 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3391 st->sb = super;
3392 st->container_dev = fd2devnum(cfd);
3393 close(cfd);
3394 return validate_geometry_imsm_volume(st, level, layout,
3395 raiddisks, chunk,
3396 size, dev,
3397 freesize, verbose);
3398 }
3399 close(cfd);
3400 } else /* may belong to another container */
3401 return 0;
3402
3403 return 1;
3404 }
3405 #endif /* MDASSEMBLE */
3406
3407 static struct mdinfo *container_content_imsm(struct supertype *st)
3408 {
3409 /* Given a container loaded by load_super_imsm_all,
3410 * extract information about all the arrays into
3411 * an mdinfo tree.
3412 *
3413 * For each imsm_dev create an mdinfo, fill it in,
3414 * then look for matching devices in super->disks
3415 * and create appropriate device mdinfo.
3416 */
3417 struct intel_super *super = st->sb;
3418 struct imsm_super *mpb = super->anchor;
3419 struct mdinfo *rest = NULL;
3420 int i;
3421
3422 /* do not assemble arrays that might have bad blocks */
3423 if (imsm_bbm_log_size(super->anchor)) {
3424 fprintf(stderr, Name ": BBM log found in metadata. "
3425 "Cannot activate array(s).\n");
3426 return NULL;
3427 }
3428
3429 for (i = 0; i < mpb->num_raid_devs; i++) {
3430 struct imsm_dev *dev = get_imsm_dev(super, i);
3431 struct imsm_map *map = get_imsm_map(dev, 0);
3432 struct mdinfo *this;
3433 int slot;
3434
3435 /* do not publish arrays that are in the middle of an
3436 * unsupported migration
3437 */
3438 if (dev->vol.migr_state &&
3439 (migr_type(dev) == MIGR_GEN_MIGR ||
3440 migr_type(dev) == MIGR_STATE_CHANGE)) {
3441 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
3442 " unsupported migration in progress\n",
3443 dev->volume);
3444 continue;
3445 }
3446
3447 this = malloc(sizeof(*this));
3448 memset(this, 0, sizeof(*this));
3449 this->next = rest;
3450
3451 super->current_vol = i;
3452 getinfo_super_imsm_volume(st, this);
3453 for (slot = 0 ; slot < map->num_members; slot++) {
3454 struct mdinfo *info_d;
3455 struct dl *d;
3456 int idx;
3457 int skip;
3458 __u32 ord;
3459
3460 skip = 0;
3461 idx = get_imsm_disk_idx(dev, slot);
3462 ord = get_imsm_ord_tbl_ent(dev, slot);
3463 for (d = super->disks; d ; d = d->next)
3464 if (d->index == idx)
3465 break;
3466
3467 if (d == NULL)
3468 skip = 1;
3469 if (d && is_failed(&d->disk))
3470 skip = 1;
3471 if (ord & IMSM_ORD_REBUILD)
3472 skip = 1;
3473
3474 /*
3475 * if we skip some disks the array will be assmebled degraded;
3476 * reset resync start to avoid a dirty-degraded situation
3477 *
3478 * FIXME handle dirty degraded
3479 */
3480 if (skip && !dev->vol.dirty)
3481 this->resync_start = ~0ULL;
3482 if (skip)
3483 continue;
3484
3485 info_d = malloc(sizeof(*info_d));
3486 if (!info_d) {
3487 fprintf(stderr, Name ": failed to allocate disk"
3488 " for volume %.16s\n", dev->volume);
3489 free(this);
3490 this = rest;
3491 break;
3492 }
3493 memset(info_d, 0, sizeof(*info_d));
3494 info_d->next = this->devs;
3495 this->devs = info_d;
3496
3497 info_d->disk.number = d->index;
3498 info_d->disk.major = d->major;
3499 info_d->disk.minor = d->minor;
3500 info_d->disk.raid_disk = slot;
3501
3502 this->array.working_disks++;
3503
3504 info_d->events = __le32_to_cpu(mpb->generation_num);
3505 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3506 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3507 if (d->devname)
3508 strcpy(info_d->name, d->devname);
3509 }
3510 rest = this;
3511 }
3512
3513 return rest;
3514 }
3515
3516
3517 #ifndef MDASSEMBLE
3518 static int imsm_open_new(struct supertype *c, struct active_array *a,
3519 char *inst)
3520 {
3521 struct intel_super *super = c->sb;
3522 struct imsm_super *mpb = super->anchor;
3523
3524 if (atoi(inst) >= mpb->num_raid_devs) {
3525 fprintf(stderr, "%s: subarry index %d, out of range\n",
3526 __func__, atoi(inst));
3527 return -ENODEV;
3528 }
3529
3530 dprintf("imsm: open_new %s\n", inst);
3531 a->info.container_member = atoi(inst);
3532 return 0;
3533 }
3534
3535 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3536 {
3537 struct imsm_map *map = get_imsm_map(dev, 0);
3538
3539 if (!failed)
3540 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3541 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3542
3543 switch (get_imsm_raid_level(map)) {
3544 case 0:
3545 return IMSM_T_STATE_FAILED;
3546 break;
3547 case 1:
3548 if (failed < map->num_members)
3549 return IMSM_T_STATE_DEGRADED;
3550 else
3551 return IMSM_T_STATE_FAILED;
3552 break;
3553 case 10:
3554 {
3555 /**
3556 * check to see if any mirrors have failed, otherwise we
3557 * are degraded. Even numbered slots are mirrored on
3558 * slot+1
3559 */
3560 int i;
3561 /* gcc -Os complains that this is unused */
3562 int insync = insync;
3563
3564 for (i = 0; i < map->num_members; i++) {
3565 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3566 int idx = ord_to_idx(ord);
3567 struct imsm_disk *disk;
3568
3569 /* reset the potential in-sync count on even-numbered
3570 * slots. num_copies is always 2 for imsm raid10
3571 */
3572 if ((i & 1) == 0)
3573 insync = 2;
3574
3575 disk = get_imsm_disk(super, idx);
3576 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
3577 insync--;
3578
3579 /* no in-sync disks left in this mirror the
3580 * array has failed
3581 */
3582 if (insync == 0)
3583 return IMSM_T_STATE_FAILED;
3584 }
3585
3586 return IMSM_T_STATE_DEGRADED;
3587 }
3588 case 5:
3589 if (failed < 2)
3590 return IMSM_T_STATE_DEGRADED;
3591 else
3592 return IMSM_T_STATE_FAILED;
3593 break;
3594 default:
3595 break;
3596 }
3597
3598 return map->map_state;
3599 }
3600
3601 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3602 {
3603 int i;
3604 int failed = 0;
3605 struct imsm_disk *disk;
3606 struct imsm_map *map = get_imsm_map(dev, 0);
3607 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3608 __u32 ord;
3609 int idx;
3610
3611 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3612 * disks that are being rebuilt. New failures are recorded to
3613 * map[0]. So we look through all the disks we started with and
3614 * see if any failures are still present, or if any new ones
3615 * have arrived
3616 *
3617 * FIXME add support for online capacity expansion and
3618 * raid-level-migration
3619 */
3620 for (i = 0; i < prev->num_members; i++) {
3621 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
3622 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
3623 idx = ord_to_idx(ord);
3624
3625 disk = get_imsm_disk(super, idx);
3626 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
3627 failed++;
3628 }
3629
3630 return failed;
3631 }
3632
3633 static int is_resyncing(struct imsm_dev *dev)
3634 {
3635 struct imsm_map *migr_map;
3636
3637 if (!dev->vol.migr_state)
3638 return 0;
3639
3640 if (migr_type(dev) == MIGR_INIT ||
3641 migr_type(dev) == MIGR_REPAIR)
3642 return 1;
3643
3644 migr_map = get_imsm_map(dev, 1);
3645
3646 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
3647 return 1;
3648 else
3649 return 0;
3650 }
3651
3652 static int is_rebuilding(struct imsm_dev *dev)
3653 {
3654 struct imsm_map *migr_map;
3655
3656 if (!dev->vol.migr_state)
3657 return 0;
3658
3659 if (migr_type(dev) != MIGR_REBUILD)
3660 return 0;
3661
3662 migr_map = get_imsm_map(dev, 1);
3663
3664 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
3665 return 1;
3666 else
3667 return 0;
3668 }
3669
3670 /* return true if we recorded new information */
3671 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
3672 {
3673 __u32 ord;
3674 int slot;
3675 struct imsm_map *map;
3676
3677 /* new failures are always set in map[0] */
3678 map = get_imsm_map(dev, 0);
3679
3680 slot = get_imsm_disk_slot(map, idx);
3681 if (slot < 0)
3682 return 0;
3683
3684 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
3685 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
3686 return 0;
3687
3688 disk->status |= FAILED_DISK;
3689 disk->status &= ~CONFIGURED_DISK;
3690 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
3691 if (~map->failed_disk_num == 0)
3692 map->failed_disk_num = slot;
3693 return 1;
3694 }
3695
3696 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
3697 {
3698 mark_failure(dev, disk, idx);
3699
3700 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
3701 return;
3702
3703 disk->scsi_id = __cpu_to_le32(~(__u32)0);
3704 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
3705 }
3706
3707 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3708 * states are handled in imsm_set_disk() with one exception, when a
3709 * resync is stopped due to a new failure this routine will set the
3710 * 'degraded' state for the array.
3711 */
3712 static int imsm_set_array_state(struct active_array *a, int consistent)
3713 {
3714 int inst = a->info.container_member;
3715 struct intel_super *super = a->container->sb;
3716 struct imsm_dev *dev = get_imsm_dev(super, inst);
3717 struct imsm_map *map = get_imsm_map(dev, 0);
3718 int failed = imsm_count_failed(super, dev);
3719 __u8 map_state = imsm_check_degraded(super, dev, failed);
3720
3721 /* before we activate this array handle any missing disks */
3722 if (consistent == 2 && super->missing) {
3723 struct dl *dl;
3724
3725 dprintf("imsm: mark missing\n");
3726 end_migration(dev, map_state);
3727 for (dl = super->missing; dl; dl = dl->next)
3728 mark_missing(dev, &dl->disk, dl->index);
3729 super->updates_pending++;
3730 }
3731
3732 if (consistent == 2 &&
3733 (!is_resync_complete(a) ||
3734 map_state != IMSM_T_STATE_NORMAL ||
3735 dev->vol.migr_state))
3736 consistent = 0;
3737
3738 if (is_resync_complete(a)) {
3739 /* complete intialization / resync,
3740 * recovery and interrupted recovery is completed in
3741 * ->set_disk
3742 */
3743 if (is_resyncing(dev)) {
3744 dprintf("imsm: mark resync done\n");
3745 end_migration(dev, map_state);
3746 super->updates_pending++;
3747 }
3748 } else if (!is_resyncing(dev) && !failed) {
3749 /* mark the start of the init process if nothing is failed */
3750 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
3751 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
3752 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
3753 else
3754 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
3755 super->updates_pending++;
3756 }
3757
3758 /* FIXME check if we can update curr_migr_unit from resync_start */
3759
3760 /* mark dirty / clean */
3761 if (dev->vol.dirty != !consistent) {
3762 dprintf("imsm: mark '%s' (%llu)\n",
3763 consistent ? "clean" : "dirty", a->resync_start);
3764 if (consistent)
3765 dev->vol.dirty = 0;
3766 else
3767 dev->vol.dirty = 1;
3768 super->updates_pending++;
3769 }
3770 return consistent;
3771 }
3772
3773 static void imsm_set_disk(struct active_array *a, int n, int state)
3774 {
3775 int inst = a->info.container_member;
3776 struct intel_super *super = a->container->sb;
3777 struct imsm_dev *dev = get_imsm_dev(super, inst);
3778 struct imsm_map *map = get_imsm_map(dev, 0);
3779 struct imsm_disk *disk;
3780 int failed;
3781 __u32 ord;
3782 __u8 map_state;
3783
3784 if (n > map->num_members)
3785 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
3786 n, map->num_members - 1);
3787
3788 if (n < 0)
3789 return;
3790
3791 dprintf("imsm: set_disk %d:%x\n", n, state);
3792
3793 ord = get_imsm_ord_tbl_ent(dev, n);
3794 disk = get_imsm_disk(super, ord_to_idx(ord));
3795
3796 /* check for new failures */
3797 if (state & DS_FAULTY) {
3798 if (mark_failure(dev, disk, ord_to_idx(ord)))
3799 super->updates_pending++;
3800 }
3801
3802 /* check if in_sync */
3803 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
3804 struct imsm_map *migr_map = get_imsm_map(dev, 1);
3805
3806 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
3807 super->updates_pending++;
3808 }
3809
3810 failed = imsm_count_failed(super, dev);
3811 map_state = imsm_check_degraded(super, dev, failed);
3812
3813 /* check if recovery complete, newly degraded, or failed */
3814 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
3815 end_migration(dev, map_state);
3816 map = get_imsm_map(dev, 0);
3817 map->failed_disk_num = ~0;
3818 super->updates_pending++;
3819 } else if (map_state == IMSM_T_STATE_DEGRADED &&
3820 map->map_state != map_state &&
3821 !dev->vol.migr_state) {
3822 dprintf("imsm: mark degraded\n");
3823 map->map_state = map_state;
3824 super->updates_pending++;
3825 } else if (map_state == IMSM_T_STATE_FAILED &&
3826 map->map_state != map_state) {
3827 dprintf("imsm: mark failed\n");
3828 end_migration(dev, map_state);
3829 super->updates_pending++;
3830 }
3831 }
3832
3833 static int store_imsm_mpb(int fd, struct intel_super *super)
3834 {
3835 struct imsm_super *mpb = super->anchor;
3836 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
3837 unsigned long long dsize;
3838 unsigned long long sectors;
3839
3840 get_dev_size(fd, NULL, &dsize);
3841
3842 if (mpb_size > 512) {
3843 /* -1 to account for anchor */
3844 sectors = mpb_sectors(mpb) - 1;
3845
3846 /* write the extended mpb to the sectors preceeding the anchor */
3847 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
3848 return 1;
3849
3850 if (write(fd, super->buf + 512, 512 * sectors) != 512 * sectors)
3851 return 1;
3852 }
3853
3854 /* first block is stored on second to last sector of the disk */
3855 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
3856 return 1;
3857
3858 if (write(fd, super->buf, 512) != 512)
3859 return 1;
3860
3861 return 0;
3862 }
3863
3864 static void imsm_sync_metadata(struct supertype *container)
3865 {
3866 struct intel_super *super = container->sb;
3867
3868 if (!super->updates_pending)
3869 return;
3870
3871 write_super_imsm(super, 0);
3872
3873 super->updates_pending = 0;
3874 }
3875
3876 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
3877 {
3878 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3879 int i = get_imsm_disk_idx(dev, idx);
3880 struct dl *dl;
3881
3882 for (dl = super->disks; dl; dl = dl->next)
3883 if (dl->index == i)
3884 break;
3885
3886 if (dl && is_failed(&dl->disk))
3887 dl = NULL;
3888
3889 if (dl)
3890 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
3891
3892 return dl;
3893 }
3894
3895 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
3896 struct active_array *a, int activate_new)
3897 {
3898 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3899 int idx = get_imsm_disk_idx(dev, slot);
3900 struct imsm_super *mpb = super->anchor;
3901 struct imsm_map *map;
3902 unsigned long long pos;
3903 struct mdinfo *d;
3904 struct extent *ex;
3905 int i, j;
3906 int found;
3907 __u32 array_start;
3908 __u32 array_end;
3909 struct dl *dl;
3910
3911 for (dl = super->disks; dl; dl = dl->next) {
3912 /* If in this array, skip */
3913 for (d = a->info.devs ; d ; d = d->next)
3914 if (d->state_fd >= 0 &&
3915 d->disk.major == dl->major &&
3916 d->disk.minor == dl->minor) {
3917 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3918 break;
3919 }
3920 if (d)
3921 continue;
3922
3923 /* skip in use or failed drives */
3924 if (is_failed(&dl->disk) || idx == dl->index ||
3925 dl->index == -2) {
3926 dprintf("%x:%x status (failed: %d index: %d)\n",
3927 dl->major, dl->minor, is_failed(&dl->disk), idx);
3928 continue;
3929 }
3930
3931 /* skip pure spares when we are looking for partially
3932 * assimilated drives
3933 */
3934 if (dl->index == -1 && !activate_new)
3935 continue;
3936
3937 /* Does this unused device have the requisite free space?
3938 * It needs to be able to cover all member volumes
3939 */
3940 ex = get_extents(super, dl);
3941 if (!ex) {
3942 dprintf("cannot get extents\n");
3943 continue;
3944 }
3945 for (i = 0; i < mpb->num_raid_devs; i++) {
3946 dev = get_imsm_dev(super, i);
3947 map = get_imsm_map(dev, 0);
3948
3949 /* check if this disk is already a member of
3950 * this array
3951 */
3952 if (get_imsm_disk_slot(map, dl->index) >= 0)
3953 continue;
3954
3955 found = 0;
3956 j = 0;
3957 pos = 0;
3958 array_start = __le32_to_cpu(map->pba_of_lba0);
3959 array_end = array_start +
3960 __le32_to_cpu(map->blocks_per_member) - 1;
3961
3962 do {
3963 /* check that we can start at pba_of_lba0 with
3964 * blocks_per_member of space
3965 */
3966 if (array_start >= pos && array_end < ex[j].start) {
3967 found = 1;
3968 break;
3969 }
3970 pos = ex[j].start + ex[j].size;
3971 j++;
3972 } while (ex[j-1].size);
3973
3974 if (!found)
3975 break;
3976 }
3977
3978 free(ex);
3979 if (i < mpb->num_raid_devs) {
3980 dprintf("%x:%x does not have %u to %u available\n",
3981 dl->major, dl->minor, array_start, array_end);
3982 /* No room */
3983 continue;
3984 }
3985 return dl;
3986 }
3987
3988 return dl;
3989 }
3990
3991 static struct mdinfo *imsm_activate_spare(struct active_array *a,
3992 struct metadata_update **updates)
3993 {
3994 /**
3995 * Find a device with unused free space and use it to replace a
3996 * failed/vacant region in an array. We replace failed regions one a
3997 * array at a time. The result is that a new spare disk will be added
3998 * to the first failed array and after the monitor has finished
3999 * propagating failures the remainder will be consumed.
4000 *
4001 * FIXME add a capability for mdmon to request spares from another
4002 * container.
4003 */
4004
4005 struct intel_super *super = a->container->sb;
4006 int inst = a->info.container_member;
4007 struct imsm_dev *dev = get_imsm_dev(super, inst);
4008 struct imsm_map *map = get_imsm_map(dev, 0);
4009 int failed = a->info.array.raid_disks;
4010 struct mdinfo *rv = NULL;
4011 struct mdinfo *d;
4012 struct mdinfo *di;
4013 struct metadata_update *mu;
4014 struct dl *dl;
4015 struct imsm_update_activate_spare *u;
4016 int num_spares = 0;
4017 int i;
4018
4019 for (d = a->info.devs ; d ; d = d->next) {
4020 if ((d->curr_state & DS_FAULTY) &&
4021 d->state_fd >= 0)
4022 /* wait for Removal to happen */
4023 return NULL;
4024 if (d->state_fd >= 0)
4025 failed--;
4026 }
4027
4028 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
4029 inst, failed, a->info.array.raid_disks, a->info.array.level);
4030 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
4031 return NULL;
4032
4033 /* For each slot, if it is not working, find a spare */
4034 for (i = 0; i < a->info.array.raid_disks; i++) {
4035 for (d = a->info.devs ; d ; d = d->next)
4036 if (d->disk.raid_disk == i)
4037 break;
4038 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
4039 if (d && (d->state_fd >= 0))
4040 continue;
4041
4042 /*
4043 * OK, this device needs recovery. Try to re-add the
4044 * previous occupant of this slot, if this fails see if
4045 * we can continue the assimilation of a spare that was
4046 * partially assimilated, finally try to activate a new
4047 * spare.
4048 */
4049 dl = imsm_readd(super, i, a);
4050 if (!dl)
4051 dl = imsm_add_spare(super, i, a, 0);
4052 if (!dl)
4053 dl = imsm_add_spare(super, i, a, 1);
4054 if (!dl)
4055 continue;
4056
4057 /* found a usable disk with enough space */
4058 di = malloc(sizeof(*di));
4059 if (!di)
4060 continue;
4061 memset(di, 0, sizeof(*di));
4062
4063 /* dl->index will be -1 in the case we are activating a
4064 * pristine spare. imsm_process_update() will create a
4065 * new index in this case. Once a disk is found to be
4066 * failed in all member arrays it is kicked from the
4067 * metadata
4068 */
4069 di->disk.number = dl->index;
4070
4071 /* (ab)use di->devs to store a pointer to the device
4072 * we chose
4073 */
4074 di->devs = (struct mdinfo *) dl;
4075
4076 di->disk.raid_disk = i;
4077 di->disk.major = dl->major;
4078 di->disk.minor = dl->minor;
4079 di->disk.state = 0;
4080 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
4081 di->component_size = a->info.component_size;
4082 di->container_member = inst;
4083 super->random = random32();
4084 di->next = rv;
4085 rv = di;
4086 num_spares++;
4087 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
4088 i, di->data_offset);
4089
4090 break;
4091 }
4092
4093 if (!rv)
4094 /* No spares found */
4095 return rv;
4096 /* Now 'rv' has a list of devices to return.
4097 * Create a metadata_update record to update the
4098 * disk_ord_tbl for the array
4099 */
4100 mu = malloc(sizeof(*mu));
4101 if (mu) {
4102 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
4103 if (mu->buf == NULL) {
4104 free(mu);
4105 mu = NULL;
4106 }
4107 }
4108 if (!mu) {
4109 while (rv) {
4110 struct mdinfo *n = rv->next;
4111
4112 free(rv);
4113 rv = n;
4114 }
4115 return NULL;
4116 }
4117
4118 mu->space = NULL;
4119 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
4120 mu->next = *updates;
4121 u = (struct imsm_update_activate_spare *) mu->buf;
4122
4123 for (di = rv ; di ; di = di->next) {
4124 u->type = update_activate_spare;
4125 u->dl = (struct dl *) di->devs;
4126 di->devs = NULL;
4127 u->slot = di->disk.raid_disk;
4128 u->array = inst;
4129 u->next = u + 1;
4130 u++;
4131 }
4132 (u-1)->next = NULL;
4133 *updates = mu;
4134
4135 return rv;
4136 }
4137
4138 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
4139 {
4140 struct imsm_dev *dev = get_imsm_dev(super, idx);
4141 struct imsm_map *map = get_imsm_map(dev, 0);
4142 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
4143 struct disk_info *inf = get_disk_info(u);
4144 struct imsm_disk *disk;
4145 int i;
4146 int j;
4147
4148 for (i = 0; i < map->num_members; i++) {
4149 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4150 for (j = 0; j < new_map->num_members; j++)
4151 if (serialcmp(disk->serial, inf[j].serial) == 0)
4152 return 1;
4153 }
4154
4155 return 0;
4156 }
4157
4158 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
4159
4160 static void imsm_process_update(struct supertype *st,
4161 struct metadata_update *update)
4162 {
4163 /**
4164 * crack open the metadata_update envelope to find the update record
4165 * update can be one of:
4166 * update_activate_spare - a spare device has replaced a failed
4167 * device in an array, update the disk_ord_tbl. If this disk is
4168 * present in all member arrays then also clear the SPARE_DISK
4169 * flag
4170 */
4171 struct intel_super *super = st->sb;
4172 struct imsm_super *mpb;
4173 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4174
4175 /* update requires a larger buf but the allocation failed */
4176 if (super->next_len && !super->next_buf) {
4177 super->next_len = 0;
4178 return;
4179 }
4180
4181 if (super->next_buf) {
4182 memcpy(super->next_buf, super->buf, super->len);
4183 free(super->buf);
4184 super->len = super->next_len;
4185 super->buf = super->next_buf;
4186
4187 super->next_len = 0;
4188 super->next_buf = NULL;
4189 }
4190
4191 mpb = super->anchor;
4192
4193 switch (type) {
4194 case update_activate_spare: {
4195 struct imsm_update_activate_spare *u = (void *) update->buf;
4196 struct imsm_dev *dev = get_imsm_dev(super, u->array);
4197 struct imsm_map *map = get_imsm_map(dev, 0);
4198 struct imsm_map *migr_map;
4199 struct active_array *a;
4200 struct imsm_disk *disk;
4201 __u8 to_state;
4202 struct dl *dl;
4203 unsigned int found;
4204 int failed;
4205 int victim = get_imsm_disk_idx(dev, u->slot);
4206 int i;
4207
4208 for (dl = super->disks; dl; dl = dl->next)
4209 if (dl == u->dl)
4210 break;
4211
4212 if (!dl) {
4213 fprintf(stderr, "error: imsm_activate_spare passed "
4214 "an unknown disk (index: %d)\n",
4215 u->dl->index);
4216 return;
4217 }
4218
4219 super->updates_pending++;
4220
4221 /* count failures (excluding rebuilds and the victim)
4222 * to determine map[0] state
4223 */
4224 failed = 0;
4225 for (i = 0; i < map->num_members; i++) {
4226 if (i == u->slot)
4227 continue;
4228 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4229 if (!disk || is_failed(disk))
4230 failed++;
4231 }
4232
4233 /* adding a pristine spare, assign a new index */
4234 if (dl->index < 0) {
4235 dl->index = super->anchor->num_disks;
4236 super->anchor->num_disks++;
4237 }
4238 disk = &dl->disk;
4239 disk->status |= CONFIGURED_DISK;
4240 disk->status &= ~SPARE_DISK;
4241
4242 /* mark rebuild */
4243 to_state = imsm_check_degraded(super, dev, failed);
4244 map->map_state = IMSM_T_STATE_DEGRADED;
4245 migrate(dev, to_state, MIGR_REBUILD);
4246 migr_map = get_imsm_map(dev, 1);
4247 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
4248 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
4249
4250 /* update the family_num to mark a new container
4251 * generation, being careful to record the existing
4252 * family_num in orig_family_num to clean up after
4253 * earlier mdadm versions that neglected to set it.
4254 */
4255 if (mpb->orig_family_num == 0)
4256 mpb->orig_family_num = mpb->family_num;
4257 mpb->family_num += super->random;
4258
4259 /* count arrays using the victim in the metadata */
4260 found = 0;
4261 for (a = st->arrays; a ; a = a->next) {
4262 dev = get_imsm_dev(super, a->info.container_member);
4263 map = get_imsm_map(dev, 0);
4264
4265 if (get_imsm_disk_slot(map, victim) >= 0)
4266 found++;
4267 }
4268
4269 /* delete the victim if it is no longer being
4270 * utilized anywhere
4271 */
4272 if (!found) {
4273 struct dl **dlp;
4274
4275 /* We know that 'manager' isn't touching anything,
4276 * so it is safe to delete
4277 */
4278 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
4279 if ((*dlp)->index == victim)
4280 break;
4281
4282 /* victim may be on the missing list */
4283 if (!*dlp)
4284 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
4285 if ((*dlp)->index == victim)
4286 break;
4287 imsm_delete(super, dlp, victim);
4288 }
4289 break;
4290 }
4291 case update_create_array: {
4292 /* someone wants to create a new array, we need to be aware of
4293 * a few races/collisions:
4294 * 1/ 'Create' called by two separate instances of mdadm
4295 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4296 * devices that have since been assimilated via
4297 * activate_spare.
4298 * In the event this update can not be carried out mdadm will
4299 * (FIX ME) notice that its update did not take hold.
4300 */
4301 struct imsm_update_create_array *u = (void *) update->buf;
4302 struct intel_dev *dv;
4303 struct imsm_dev *dev;
4304 struct imsm_map *map, *new_map;
4305 unsigned long long start, end;
4306 unsigned long long new_start, new_end;
4307 int i;
4308 struct disk_info *inf;
4309 struct dl *dl;
4310
4311 /* handle racing creates: first come first serve */
4312 if (u->dev_idx < mpb->num_raid_devs) {
4313 dprintf("%s: subarray %d already defined\n",
4314 __func__, u->dev_idx);
4315 goto create_error;
4316 }
4317
4318 /* check update is next in sequence */
4319 if (u->dev_idx != mpb->num_raid_devs) {
4320 dprintf("%s: can not create array %d expected index %d\n",
4321 __func__, u->dev_idx, mpb->num_raid_devs);
4322 goto create_error;
4323 }
4324
4325 new_map = get_imsm_map(&u->dev, 0);
4326 new_start = __le32_to_cpu(new_map->pba_of_lba0);
4327 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
4328 inf = get_disk_info(u);
4329
4330 /* handle activate_spare versus create race:
4331 * check to make sure that overlapping arrays do not include
4332 * overalpping disks
4333 */
4334 for (i = 0; i < mpb->num_raid_devs; i++) {
4335 dev = get_imsm_dev(super, i);
4336 map = get_imsm_map(dev, 0);
4337 start = __le32_to_cpu(map->pba_of_lba0);
4338 end = start + __le32_to_cpu(map->blocks_per_member);
4339 if ((new_start >= start && new_start <= end) ||
4340 (start >= new_start && start <= new_end))
4341 /* overlap */;
4342 else
4343 continue;
4344
4345 if (disks_overlap(super, i, u)) {
4346 dprintf("%s: arrays overlap\n", __func__);
4347 goto create_error;
4348 }
4349 }
4350
4351 /* check that prepare update was successful */
4352 if (!update->space) {
4353 dprintf("%s: prepare update failed\n", __func__);
4354 goto create_error;
4355 }
4356
4357 /* check that all disks are still active before committing
4358 * changes. FIXME: could we instead handle this by creating a
4359 * degraded array? That's probably not what the user expects,
4360 * so better to drop this update on the floor.
4361 */
4362 for (i = 0; i < new_map->num_members; i++) {
4363 dl = serial_to_dl(inf[i].serial, super);
4364 if (!dl) {
4365 dprintf("%s: disk disappeared\n", __func__);
4366 goto create_error;
4367 }
4368 }
4369
4370 super->updates_pending++;
4371
4372 /* convert spares to members and fixup ord_tbl */
4373 for (i = 0; i < new_map->num_members; i++) {
4374 dl = serial_to_dl(inf[i].serial, super);
4375 if (dl->index == -1) {
4376 dl->index = mpb->num_disks;
4377 mpb->num_disks++;
4378 dl->disk.status |= CONFIGURED_DISK;
4379 dl->disk.status &= ~SPARE_DISK;
4380 }
4381 set_imsm_ord_tbl_ent(new_map, i, dl->index);
4382 }
4383
4384 dv = update->space;
4385 dev = dv->dev;
4386 update->space = NULL;
4387 imsm_copy_dev(dev, &u->dev);
4388 dv->index = u->dev_idx;
4389 dv->next = super->devlist;
4390 super->devlist = dv;
4391 mpb->num_raid_devs++;
4392
4393 imsm_update_version_info(super);
4394 break;
4395 create_error:
4396 /* mdmon knows how to release update->space, but not
4397 * ((struct intel_dev *) update->space)->dev
4398 */
4399 if (update->space) {
4400 dv = update->space;
4401 free(dv->dev);
4402 }
4403 break;
4404 }
4405 case update_add_disk:
4406
4407 /* we may be able to repair some arrays if disks are
4408 * being added */
4409 if (super->add) {
4410 struct active_array *a;
4411
4412 super->updates_pending++;
4413 for (a = st->arrays; a; a = a->next)
4414 a->check_degraded = 1;
4415 }
4416 /* add some spares to the metadata */
4417 while (super->add) {
4418 struct dl *al;
4419
4420 al = super->add;
4421 super->add = al->next;
4422 al->next = super->disks;
4423 super->disks = al;
4424 dprintf("%s: added %x:%x\n",
4425 __func__, al->major, al->minor);
4426 }
4427
4428 break;
4429 }
4430 }
4431
4432 static void imsm_prepare_update(struct supertype *st,
4433 struct metadata_update *update)
4434 {
4435 /**
4436 * Allocate space to hold new disk entries, raid-device entries or a new
4437 * mpb if necessary. The manager synchronously waits for updates to
4438 * complete in the monitor, so new mpb buffers allocated here can be
4439 * integrated by the monitor thread without worrying about live pointers
4440 * in the manager thread.
4441 */
4442 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4443 struct intel_super *super = st->sb;
4444 struct imsm_super *mpb = super->anchor;
4445 size_t buf_len;
4446 size_t len = 0;
4447
4448 switch (type) {
4449 case update_create_array: {
4450 struct imsm_update_create_array *u = (void *) update->buf;
4451 struct intel_dev *dv;
4452 struct imsm_dev *dev = &u->dev;
4453 struct imsm_map *map = get_imsm_map(dev, 0);
4454 struct dl *dl;
4455 struct disk_info *inf;
4456 int i;
4457 int activate = 0;
4458
4459 inf = get_disk_info(u);
4460 len = sizeof_imsm_dev(dev, 1);
4461 /* allocate a new super->devlist entry */
4462 dv = malloc(sizeof(*dv));
4463 if (dv) {
4464 dv->dev = malloc(len);
4465 if (dv->dev)
4466 update->space = dv;
4467 else {
4468 free(dv);
4469 update->space = NULL;
4470 }
4471 }
4472
4473 /* count how many spares will be converted to members */
4474 for (i = 0; i < map->num_members; i++) {
4475 dl = serial_to_dl(inf[i].serial, super);
4476 if (!dl) {
4477 /* hmm maybe it failed?, nothing we can do about
4478 * it here
4479 */
4480 continue;
4481 }
4482 if (count_memberships(dl, super) == 0)
4483 activate++;
4484 }
4485 len += activate * sizeof(struct imsm_disk);
4486 break;
4487 default:
4488 break;
4489 }
4490 }
4491
4492 /* check if we need a larger metadata buffer */
4493 if (super->next_buf)
4494 buf_len = super->next_len;
4495 else
4496 buf_len = super->len;
4497
4498 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4499 /* ok we need a larger buf than what is currently allocated
4500 * if this allocation fails process_update will notice that
4501 * ->next_len is set and ->next_buf is NULL
4502 */
4503 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4504 if (super->next_buf)
4505 free(super->next_buf);
4506
4507 super->next_len = buf_len;
4508 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
4509 memset(super->next_buf, 0, buf_len);
4510 else
4511 super->next_buf = NULL;
4512 }
4513 }
4514
4515 /* must be called while manager is quiesced */
4516 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4517 {
4518 struct imsm_super *mpb = super->anchor;
4519 struct dl *iter;
4520 struct imsm_dev *dev;
4521 struct imsm_map *map;
4522 int i, j, num_members;
4523 __u32 ord;
4524
4525 dprintf("%s: deleting device[%d] from imsm_super\n",
4526 __func__, index);
4527
4528 /* shift all indexes down one */
4529 for (iter = super->disks; iter; iter = iter->next)
4530 if (iter->index > index)
4531 iter->index--;
4532 for (iter = super->missing; iter; iter = iter->next)
4533 if (iter->index > index)
4534 iter->index--;
4535
4536 for (i = 0; i < mpb->num_raid_devs; i++) {
4537 dev = get_imsm_dev(super, i);
4538 map = get_imsm_map(dev, 0);
4539 num_members = map->num_members;
4540 for (j = 0; j < num_members; j++) {
4541 /* update ord entries being careful not to propagate
4542 * ord-flags to the first map
4543 */
4544 ord = get_imsm_ord_tbl_ent(dev, j);
4545
4546 if (ord_to_idx(ord) <= index)
4547 continue;
4548
4549 map = get_imsm_map(dev, 0);
4550 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4551 map = get_imsm_map(dev, 1);
4552 if (map)
4553 set_imsm_ord_tbl_ent(map, j, ord - 1);
4554 }
4555 }
4556
4557 mpb->num_disks--;
4558 super->updates_pending++;
4559 if (*dlp) {
4560 struct dl *dl = *dlp;
4561
4562 *dlp = (*dlp)->next;
4563 __free_imsm_disk(dl);
4564 }
4565 }
4566 #endif /* MDASSEMBLE */
4567
4568 struct superswitch super_imsm = {
4569 #ifndef MDASSEMBLE
4570 .examine_super = examine_super_imsm,
4571 .brief_examine_super = brief_examine_super_imsm,
4572 .brief_examine_subarrays = brief_examine_subarrays_imsm,
4573 .export_examine_super = export_examine_super_imsm,
4574 .detail_super = detail_super_imsm,
4575 .brief_detail_super = brief_detail_super_imsm,
4576 .write_init_super = write_init_super_imsm,
4577 .validate_geometry = validate_geometry_imsm,
4578 .add_to_super = add_to_super_imsm,
4579 .detail_platform = detail_platform_imsm,
4580 #endif
4581 .match_home = match_home_imsm,
4582 .uuid_from_super= uuid_from_super_imsm,
4583 .getinfo_super = getinfo_super_imsm,
4584 .update_super = update_super_imsm,
4585
4586 .avail_size = avail_size_imsm,
4587
4588 .compare_super = compare_super_imsm,
4589
4590 .load_super = load_super_imsm,
4591 .init_super = init_super_imsm,
4592 .store_super = store_zero_imsm,
4593 .free_super = free_super_imsm,
4594 .match_metadata_desc = match_metadata_desc_imsm,
4595 .container_content = container_content_imsm,
4596 .default_layout = imsm_level_to_layout,
4597
4598 .external = 1,
4599 .name = "imsm",
4600
4601 #ifndef MDASSEMBLE
4602 /* for mdmon */
4603 .open_new = imsm_open_new,
4604 .load_super = load_super_imsm,
4605 .set_array_state= imsm_set_array_state,
4606 .set_disk = imsm_set_disk,
4607 .sync_metadata = imsm_sync_metadata,
4608 .activate_spare = imsm_activate_spare,
4609 .process_update = imsm_process_update,
4610 .prepare_update = imsm_prepare_update,
4611 #endif /* MDASSEMBLE */
4612 };
Unnamed repository; edit this file 'description' to name the repository.