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