]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-intel.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
imsm: FIX: mdadm should process local data
[thirdparty/mdadm.git] / super-intel.c
1 /*
2 * mdadm - Intel(R) Matrix Storage Manager Support
3 *
4 * Copyright (C) 2002-2008 Intel Corporation
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define HAVE_STDINT_H 1
21 #include "mdadm.h"
22 #include "mdmon.h"
23 #include "sha1.h"
24 #include "platform-intel.h"
25 #include <values.h>
26 #include <scsi/sg.h>
27 #include <ctype.h>
28 #include <dirent.h>
29
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
43
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
53
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56 #define SECT_PER_MB_SHIFT 11
57
58 /* Disk configuration info. */
59 #define IMSM_MAX_DEVICES 255
60 struct imsm_disk {
61 __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
62 __u32 total_blocks; /* 0xE8 - 0xEB total blocks */
63 __u32 scsi_id; /* 0xEC - 0xEF scsi ID */
64 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
65 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
66 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
67 __u32 status; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
71 };
72
73 /* RAID map configuration infos. */
74 struct imsm_map {
75 __u32 pba_of_lba0; /* start address of partition */
76 __u32 blocks_per_member;/* blocks per member */
77 __u32 num_data_stripes; /* number of data stripes */
78 __u16 blocks_per_strip;
79 __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
84 __u8 raid_level;
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members; /* number of member disks */
89 __u8 num_domains; /* number of parity domains */
90 __u8 failed_disk_num; /* valid only when state is degraded */
91 __u8 ddf;
92 __u32 filler[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
96 */
97 } __attribute__ ((packed));
98
99 struct imsm_vol {
100 __u32 curr_migr_unit;
101 __u32 checkpoint_id; /* id to access curr_migr_unit */
102 __u8 migr_state; /* Normal or Migrating */
103 #define MIGR_INIT 0
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 #define MIGR_REPAIR 5
109 __u8 migr_type; /* Initializing, Rebuilding, ... */
110 __u8 dirty;
111 __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */
112 __u16 verify_errors; /* number of mismatches */
113 __u16 bad_blocks; /* number of bad blocks during verify */
114 __u32 filler[4];
115 struct imsm_map map[1];
116 /* here comes another one if migr_state */
117 } __attribute__ ((packed));
118
119 struct imsm_dev {
120 __u8 volume[MAX_RAID_SERIAL_LEN];
121 __u32 size_low;
122 __u32 size_high;
123 #define DEV_BOOTABLE __cpu_to_le32(0x01)
124 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
125 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
126 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
127 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
128 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
129 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
130 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
131 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
132 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
133 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
134 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
135 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
136 __u32 status; /* Persistent RaidDev status */
137 __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
138 __u8 migr_priority;
139 __u8 num_sub_vols;
140 __u8 tid;
141 __u8 cng_master_disk;
142 __u16 cache_policy;
143 __u8 cng_state;
144 __u8 cng_sub_state;
145 #define IMSM_DEV_FILLERS 10
146 __u32 filler[IMSM_DEV_FILLERS];
147 struct imsm_vol vol;
148 } __attribute__ ((packed));
149
150 struct imsm_super {
151 __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
152 __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */
153 __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */
154 __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */
155 __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
156 __u32 error_log_size; /* 0x30 - 0x33 in bytes */
157 __u32 attributes; /* 0x34 - 0x37 */
158 __u8 num_disks; /* 0x38 Number of configured disks */
159 __u8 num_raid_devs; /* 0x39 Number of configured volumes */
160 __u8 error_log_pos; /* 0x3A */
161 __u8 fill[1]; /* 0x3B */
162 __u32 cache_size; /* 0x3c - 0x40 in mb */
163 __u32 orig_family_num; /* 0x40 - 0x43 original family num */
164 __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */
165 __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
166 #define IMSM_FILLERS 35
167 __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
168 struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */
169 /* here comes imsm_dev[num_raid_devs] */
170 /* here comes BBM logs */
171 } __attribute__ ((packed));
172
173 #define BBM_LOG_MAX_ENTRIES 254
174
175 struct bbm_log_entry {
176 __u64 defective_block_start;
177 #define UNREADABLE 0xFFFFFFFF
178 __u32 spare_block_offset;
179 __u16 remapped_marked_count;
180 __u16 disk_ordinal;
181 } __attribute__ ((__packed__));
182
183 struct bbm_log {
184 __u32 signature; /* 0xABADB10C */
185 __u32 entry_count;
186 __u32 reserved_spare_block_count; /* 0 */
187 __u32 reserved; /* 0xFFFF */
188 __u64 first_spare_lba;
189 struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
190 } __attribute__ ((__packed__));
191
192
193 #ifndef MDASSEMBLE
194 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
195 #endif
196
197 static __u8 migr_type(struct imsm_dev *dev)
198 {
199 if (dev->vol.migr_type == MIGR_VERIFY &&
200 dev->status & DEV_VERIFY_AND_FIX)
201 return MIGR_REPAIR;
202 else
203 return dev->vol.migr_type;
204 }
205
206 static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
207 {
208 /* for compatibility with older oroms convert MIGR_REPAIR, into
209 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
210 */
211 if (migr_type == MIGR_REPAIR) {
212 dev->vol.migr_type = MIGR_VERIFY;
213 dev->status |= DEV_VERIFY_AND_FIX;
214 } else {
215 dev->vol.migr_type = migr_type;
216 dev->status &= ~DEV_VERIFY_AND_FIX;
217 }
218 }
219
220 static unsigned int sector_count(__u32 bytes)
221 {
222 return ((bytes + (512-1)) & (~(512-1))) / 512;
223 }
224
225 static unsigned int mpb_sectors(struct imsm_super *mpb)
226 {
227 return sector_count(__le32_to_cpu(mpb->mpb_size));
228 }
229
230 struct intel_dev {
231 struct imsm_dev *dev;
232 struct intel_dev *next;
233 unsigned index;
234 };
235
236 enum action {
237 DISK_REMOVE = 1,
238 DISK_ADD
239 };
240 /* internal representation of IMSM metadata */
241 struct intel_super {
242 union {
243 void *buf; /* O_DIRECT buffer for reading/writing metadata */
244 struct imsm_super *anchor; /* immovable parameters */
245 };
246 size_t len; /* size of the 'buf' allocation */
247 void *next_buf; /* for realloc'ing buf from the manager */
248 size_t next_len;
249 int updates_pending; /* count of pending updates for mdmon */
250 int current_vol; /* index of raid device undergoing creation */
251 __u32 create_offset; /* common start for 'current_vol' */
252 __u32 random; /* random data for seeding new family numbers */
253 struct intel_dev *devlist;
254 struct dl {
255 struct dl *next;
256 int index;
257 __u8 serial[MAX_RAID_SERIAL_LEN];
258 int major, minor;
259 char *devname;
260 struct imsm_disk disk;
261 int fd;
262 int extent_cnt;
263 struct extent *e; /* for determining freespace @ create */
264 int raiddisk; /* slot to fill in autolayout */
265 enum action action;
266 } *disks;
267 struct dl *disk_mgmt_list; /* list of disks to add/remove while mdmon
268 active */
269 struct dl *missing; /* disks removed while we weren't looking */
270 struct bbm_log *bbm_log;
271 const char *hba; /* device path of the raid controller for this metadata */
272 const struct imsm_orom *orom; /* platform firmware support */
273 struct intel_super *next; /* (temp) list for disambiguating family_num */
274 };
275
276 struct intel_disk {
277 struct imsm_disk disk;
278 #define IMSM_UNKNOWN_OWNER (-1)
279 int owner;
280 struct intel_disk *next;
281 };
282
283 struct extent {
284 unsigned long long start, size;
285 };
286
287 /* definition of messages passed to imsm_process_update */
288 enum imsm_update_type {
289 update_activate_spare,
290 update_create_array,
291 update_kill_array,
292 update_rename_array,
293 update_add_remove_disk,
294 update_reshape_container_disks,
295 };
296
297 struct imsm_update_activate_spare {
298 enum imsm_update_type type;
299 struct dl *dl;
300 int slot;
301 int array;
302 struct imsm_update_activate_spare *next;
303 };
304
305 struct geo_params {
306 int dev_id;
307 char *dev_name;
308 long long size;
309 int level;
310 int layout;
311 int chunksize;
312 int raid_disks;
313 };
314
315
316 struct imsm_update_reshape {
317 enum imsm_update_type type;
318 int old_raid_disks;
319 int new_raid_disks;
320 int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */
321 };
322
323 struct disk_info {
324 __u8 serial[MAX_RAID_SERIAL_LEN];
325 };
326
327 struct imsm_update_create_array {
328 enum imsm_update_type type;
329 int dev_idx;
330 struct imsm_dev dev;
331 };
332
333 struct imsm_update_kill_array {
334 enum imsm_update_type type;
335 int dev_idx;
336 };
337
338 struct imsm_update_rename_array {
339 enum imsm_update_type type;
340 __u8 name[MAX_RAID_SERIAL_LEN];
341 int dev_idx;
342 };
343
344 struct imsm_update_add_remove_disk {
345 enum imsm_update_type type;
346 };
347
348 static struct supertype *match_metadata_desc_imsm(char *arg)
349 {
350 struct supertype *st;
351
352 if (strcmp(arg, "imsm") != 0 &&
353 strcmp(arg, "default") != 0
354 )
355 return NULL;
356
357 st = malloc(sizeof(*st));
358 if (!st)
359 return NULL;
360 memset(st, 0, sizeof(*st));
361 st->container_dev = NoMdDev;
362 st->ss = &super_imsm;
363 st->max_devs = IMSM_MAX_DEVICES;
364 st->minor_version = 0;
365 st->sb = NULL;
366 return st;
367 }
368
369 #ifndef MDASSEMBLE
370 static __u8 *get_imsm_version(struct imsm_super *mpb)
371 {
372 return &mpb->sig[MPB_SIG_LEN];
373 }
374 #endif
375
376 /* retrieve a disk directly from the anchor when the anchor is known to be
377 * up-to-date, currently only at load time
378 */
379 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
380 {
381 if (index >= mpb->num_disks)
382 return NULL;
383 return &mpb->disk[index];
384 }
385
386 /* retrieve the disk description based on a index of the disk
387 * in the sub-array
388 */
389 static struct dl *get_imsm_dl_disk(struct intel_super *super, __u8 index)
390 {
391 struct dl *d;
392
393 for (d = super->disks; d; d = d->next)
394 if (d->index == index)
395 return d;
396
397 return NULL;
398 }
399 /* retrieve a disk from the parsed metadata */
400 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
401 {
402 struct dl *dl;
403
404 dl = get_imsm_dl_disk(super, index);
405 if (dl)
406 return &dl->disk;
407
408 return NULL;
409 }
410
411 /* generate a checksum directly from the anchor when the anchor is known to be
412 * up-to-date, currently only at load or write_super after coalescing
413 */
414 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
415 {
416 __u32 end = mpb->mpb_size / sizeof(end);
417 __u32 *p = (__u32 *) mpb;
418 __u32 sum = 0;
419
420 while (end--) {
421 sum += __le32_to_cpu(*p);
422 p++;
423 }
424
425 return sum - __le32_to_cpu(mpb->check_sum);
426 }
427
428 static size_t sizeof_imsm_map(struct imsm_map *map)
429 {
430 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
431 }
432
433 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
434 {
435 struct imsm_map *map = &dev->vol.map[0];
436
437 if (second_map && !dev->vol.migr_state)
438 return NULL;
439 else if (second_map) {
440 void *ptr = map;
441
442 return ptr + sizeof_imsm_map(map);
443 } else
444 return map;
445
446 }
447
448 /* return the size of the device.
449 * migr_state increases the returned size if map[0] were to be duplicated
450 */
451 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
452 {
453 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
454 sizeof_imsm_map(get_imsm_map(dev, 0));
455
456 /* migrating means an additional map */
457 if (dev->vol.migr_state)
458 size += sizeof_imsm_map(get_imsm_map(dev, 1));
459 else if (migr_state)
460 size += sizeof_imsm_map(get_imsm_map(dev, 0));
461
462 return size;
463 }
464
465 #ifndef MDASSEMBLE
466 /* retrieve disk serial number list from a metadata update */
467 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
468 {
469 void *u = update;
470 struct disk_info *inf;
471
472 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
473 sizeof_imsm_dev(&update->dev, 0);
474
475 return inf;
476 }
477 #endif
478
479 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
480 {
481 int offset;
482 int i;
483 void *_mpb = mpb;
484
485 if (index >= mpb->num_raid_devs)
486 return NULL;
487
488 /* devices start after all disks */
489 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
490
491 for (i = 0; i <= index; i++)
492 if (i == index)
493 return _mpb + offset;
494 else
495 offset += sizeof_imsm_dev(_mpb + offset, 0);
496
497 return NULL;
498 }
499
500 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
501 {
502 struct intel_dev *dv;
503
504 if (index >= super->anchor->num_raid_devs)
505 return NULL;
506 for (dv = super->devlist; dv; dv = dv->next)
507 if (dv->index == index)
508 return dv->dev;
509 return NULL;
510 }
511
512 /*
513 * for second_map:
514 * == 0 get first map
515 * == 1 get second map
516 * == -1 than get map according to the current migr_state
517 */
518 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev,
519 int slot,
520 int second_map)
521 {
522 struct imsm_map *map;
523
524 if (second_map == -1) {
525 if (dev->vol.migr_state)
526 map = get_imsm_map(dev, 1);
527 else
528 map = get_imsm_map(dev, 0);
529 } else {
530 map = get_imsm_map(dev, second_map);
531 }
532
533 /* top byte identifies disk under rebuild */
534 return __le32_to_cpu(map->disk_ord_tbl[slot]);
535 }
536
537 #define ord_to_idx(ord) (((ord) << 8) >> 8)
538 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot, int second_map)
539 {
540 __u32 ord = get_imsm_ord_tbl_ent(dev, slot, second_map);
541
542 return ord_to_idx(ord);
543 }
544
545 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
546 {
547 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
548 }
549
550 static int get_imsm_disk_slot(struct imsm_map *map, unsigned idx)
551 {
552 int slot;
553 __u32 ord;
554
555 for (slot = 0; slot < map->num_members; slot++) {
556 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
557 if (ord_to_idx(ord) == idx)
558 return slot;
559 }
560
561 return -1;
562 }
563
564 static int get_imsm_raid_level(struct imsm_map *map)
565 {
566 if (map->raid_level == 1) {
567 if (map->num_members == 2)
568 return 1;
569 else
570 return 10;
571 }
572
573 return map->raid_level;
574 }
575
576 static int cmp_extent(const void *av, const void *bv)
577 {
578 const struct extent *a = av;
579 const struct extent *b = bv;
580 if (a->start < b->start)
581 return -1;
582 if (a->start > b->start)
583 return 1;
584 return 0;
585 }
586
587 static int count_memberships(struct dl *dl, struct intel_super *super)
588 {
589 int memberships = 0;
590 int i;
591
592 for (i = 0; i < super->anchor->num_raid_devs; i++) {
593 struct imsm_dev *dev = get_imsm_dev(super, i);
594 struct imsm_map *map = get_imsm_map(dev, 0);
595
596 if (get_imsm_disk_slot(map, dl->index) >= 0)
597 memberships++;
598 }
599
600 return memberships;
601 }
602
603 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
604 {
605 /* find a list of used extents on the given physical device */
606 struct extent *rv, *e;
607 int i;
608 int memberships = count_memberships(dl, super);
609 __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
610
611 rv = malloc(sizeof(struct extent) * (memberships + 1));
612 if (!rv)
613 return NULL;
614 e = rv;
615
616 for (i = 0; i < super->anchor->num_raid_devs; i++) {
617 struct imsm_dev *dev = get_imsm_dev(super, i);
618 struct imsm_map *map = get_imsm_map(dev, 0);
619
620 if (get_imsm_disk_slot(map, dl->index) >= 0) {
621 e->start = __le32_to_cpu(map->pba_of_lba0);
622 e->size = __le32_to_cpu(map->blocks_per_member);
623 e++;
624 }
625 }
626 qsort(rv, memberships, sizeof(*rv), cmp_extent);
627
628 /* determine the start of the metadata
629 * when no raid devices are defined use the default
630 * ...otherwise allow the metadata to truncate the value
631 * as is the case with older versions of imsm
632 */
633 if (memberships) {
634 struct extent *last = &rv[memberships - 1];
635 __u32 remainder;
636
637 remainder = __le32_to_cpu(dl->disk.total_blocks) -
638 (last->start + last->size);
639 /* round down to 1k block to satisfy precision of the kernel
640 * 'size' interface
641 */
642 remainder &= ~1UL;
643 /* make sure remainder is still sane */
644 if (remainder < (unsigned)ROUND_UP(super->len, 512) >> 9)
645 remainder = ROUND_UP(super->len, 512) >> 9;
646 if (reservation > remainder)
647 reservation = remainder;
648 }
649 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
650 e->size = 0;
651 return rv;
652 }
653
654 /* try to determine how much space is reserved for metadata from
655 * the last get_extents() entry, otherwise fallback to the
656 * default
657 */
658 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
659 {
660 struct extent *e;
661 int i;
662 __u32 rv;
663
664 /* for spares just return a minimal reservation which will grow
665 * once the spare is picked up by an array
666 */
667 if (dl->index == -1)
668 return MPB_SECTOR_CNT;
669
670 e = get_extents(super, dl);
671 if (!e)
672 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
673
674 /* scroll to last entry */
675 for (i = 0; e[i].size; i++)
676 continue;
677
678 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
679
680 free(e);
681
682 return rv;
683 }
684
685 static int is_spare(struct imsm_disk *disk)
686 {
687 return (disk->status & SPARE_DISK) == SPARE_DISK;
688 }
689
690 static int is_configured(struct imsm_disk *disk)
691 {
692 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
693 }
694
695 static int is_failed(struct imsm_disk *disk)
696 {
697 return (disk->status & FAILED_DISK) == FAILED_DISK;
698 }
699
700 /* Return minimum size of a spare that can be used in this array*/
701 static unsigned long long min_acceptable_spare_size_imsm(struct supertype *st)
702 {
703 struct intel_super *super = st->sb;
704 struct dl *dl;
705 struct extent *e;
706 int i;
707 unsigned long long rv = 0;
708
709 if (!super)
710 return rv;
711 /* find first active disk in array */
712 dl = super->disks;
713 while (dl && (is_failed(&dl->disk) || dl->index == -1))
714 dl = dl->next;
715 if (!dl)
716 return rv;
717 /* find last lba used by subarrays */
718 e = get_extents(super, dl);
719 if (!e)
720 return rv;
721 for (i = 0; e[i].size; i++)
722 continue;
723 if (i > 0)
724 rv = e[i-1].start + e[i-1].size;
725 free(e);
726 /* add the amount of space needed for metadata */
727 rv = rv + MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
728 return rv * 512;
729 }
730
731 #ifndef MDASSEMBLE
732 static __u64 blocks_per_migr_unit(struct imsm_dev *dev);
733
734 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
735 {
736 __u64 sz;
737 int slot, i;
738 struct imsm_map *map = get_imsm_map(dev, 0);
739 struct imsm_map *map2 = get_imsm_map(dev, 1);
740 __u32 ord;
741
742 printf("\n");
743 printf("[%.16s]:\n", dev->volume);
744 printf(" UUID : %s\n", uuid);
745 printf(" RAID Level : %d", get_imsm_raid_level(map));
746 if (map2)
747 printf(" <-- %d", get_imsm_raid_level(map2));
748 printf("\n");
749 printf(" Members : %d", map->num_members);
750 if (map2)
751 printf(" <-- %d", map2->num_members);
752 printf("\n");
753 printf(" Slots : [");
754 for (i = 0; i < map->num_members; i++) {
755 ord = get_imsm_ord_tbl_ent(dev, i, 0);
756 printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
757 }
758 printf("]");
759 if (map2) {
760 printf(" <-- [");
761 for (i = 0; i < map2->num_members; i++) {
762 ord = get_imsm_ord_tbl_ent(dev, i, 1);
763 printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
764 }
765 printf("]");
766 }
767 printf("\n");
768 slot = get_imsm_disk_slot(map, disk_idx);
769 if (slot >= 0) {
770 ord = get_imsm_ord_tbl_ent(dev, slot, -1);
771 printf(" This Slot : %d%s\n", slot,
772 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
773 } else
774 printf(" This Slot : ?\n");
775 sz = __le32_to_cpu(dev->size_high);
776 sz <<= 32;
777 sz += __le32_to_cpu(dev->size_low);
778 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
779 human_size(sz * 512));
780 sz = __le32_to_cpu(map->blocks_per_member);
781 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
782 human_size(sz * 512));
783 printf(" Sector Offset : %u\n",
784 __le32_to_cpu(map->pba_of_lba0));
785 printf(" Num Stripes : %u\n",
786 __le32_to_cpu(map->num_data_stripes));
787 printf(" Chunk Size : %u KiB",
788 __le16_to_cpu(map->blocks_per_strip) / 2);
789 if (map2)
790 printf(" <-- %u KiB",
791 __le16_to_cpu(map2->blocks_per_strip) / 2);
792 printf("\n");
793 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
794 printf(" Migrate State : ");
795 if (dev->vol.migr_state) {
796 if (migr_type(dev) == MIGR_INIT)
797 printf("initialize\n");
798 else if (migr_type(dev) == MIGR_REBUILD)
799 printf("rebuild\n");
800 else if (migr_type(dev) == MIGR_VERIFY)
801 printf("check\n");
802 else if (migr_type(dev) == MIGR_GEN_MIGR)
803 printf("general migration\n");
804 else if (migr_type(dev) == MIGR_STATE_CHANGE)
805 printf("state change\n");
806 else if (migr_type(dev) == MIGR_REPAIR)
807 printf("repair\n");
808 else
809 printf("<unknown:%d>\n", migr_type(dev));
810 } else
811 printf("idle\n");
812 printf(" Map State : %s", map_state_str[map->map_state]);
813 if (dev->vol.migr_state) {
814 struct imsm_map *map = get_imsm_map(dev, 1);
815
816 printf(" <-- %s", map_state_str[map->map_state]);
817 printf("\n Checkpoint : %u (%llu)",
818 __le32_to_cpu(dev->vol.curr_migr_unit),
819 (unsigned long long)blocks_per_migr_unit(dev));
820 }
821 printf("\n");
822 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
823 }
824
825 static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved)
826 {
827 struct imsm_disk *disk = __get_imsm_disk(mpb, index);
828 char str[MAX_RAID_SERIAL_LEN + 1];
829 __u64 sz;
830
831 if (index < 0 || !disk)
832 return;
833
834 printf("\n");
835 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
836 printf(" Disk%02d Serial : %s\n", index, str);
837 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
838 is_configured(disk) ? " active" : "",
839 is_failed(disk) ? " failed" : "");
840 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
841 sz = __le32_to_cpu(disk->total_blocks) - reserved;
842 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
843 human_size(sz * 512));
844 }
845
846 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map);
847
848 static void examine_super_imsm(struct supertype *st, char *homehost)
849 {
850 struct intel_super *super = st->sb;
851 struct imsm_super *mpb = super->anchor;
852 char str[MAX_SIGNATURE_LENGTH];
853 int i;
854 struct mdinfo info;
855 char nbuf[64];
856 __u32 sum;
857 __u32 reserved = imsm_reserved_sectors(super, super->disks);
858 struct dl *dl;
859
860 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
861 printf(" Magic : %s\n", str);
862 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
863 printf(" Version : %s\n", get_imsm_version(mpb));
864 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
865 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
866 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
867 getinfo_super_imsm(st, &info, NULL);
868 fname_from_uuid(st, &info, nbuf, ':');
869 printf(" UUID : %s\n", nbuf + 5);
870 sum = __le32_to_cpu(mpb->check_sum);
871 printf(" Checksum : %08x %s\n", sum,
872 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
873 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
874 printf(" Disks : %d\n", mpb->num_disks);
875 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
876 print_imsm_disk(mpb, super->disks->index, reserved);
877 if (super->bbm_log) {
878 struct bbm_log *log = super->bbm_log;
879
880 printf("\n");
881 printf("Bad Block Management Log:\n");
882 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
883 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
884 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
885 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
886 printf(" First Spare : %llx\n",
887 (unsigned long long) __le64_to_cpu(log->first_spare_lba));
888 }
889 for (i = 0; i < mpb->num_raid_devs; i++) {
890 struct mdinfo info;
891 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
892
893 super->current_vol = i;
894 getinfo_super_imsm(st, &info, NULL);
895 fname_from_uuid(st, &info, nbuf, ':');
896 print_imsm_dev(dev, nbuf + 5, super->disks->index);
897 }
898 for (i = 0; i < mpb->num_disks; i++) {
899 if (i == super->disks->index)
900 continue;
901 print_imsm_disk(mpb, i, reserved);
902 }
903 for (dl = super->disks ; dl; dl = dl->next) {
904 struct imsm_disk *disk;
905 char str[MAX_RAID_SERIAL_LEN + 1];
906 __u64 sz;
907
908 if (dl->index >= 0)
909 continue;
910
911 disk = &dl->disk;
912 printf("\n");
913 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
914 printf(" Disk Serial : %s\n", str);
915 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
916 is_configured(disk) ? " active" : "",
917 is_failed(disk) ? " failed" : "");
918 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
919 sz = __le32_to_cpu(disk->total_blocks) - reserved;
920 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
921 human_size(sz * 512));
922 }
923 }
924
925 static void brief_examine_super_imsm(struct supertype *st, int verbose)
926 {
927 /* We just write a generic IMSM ARRAY entry */
928 struct mdinfo info;
929 char nbuf[64];
930 struct intel_super *super = st->sb;
931
932 if (!super->anchor->num_raid_devs) {
933 printf("ARRAY metadata=imsm\n");
934 return;
935 }
936
937 getinfo_super_imsm(st, &info, NULL);
938 fname_from_uuid(st, &info, nbuf, ':');
939 printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
940 }
941
942 static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
943 {
944 /* We just write a generic IMSM ARRAY entry */
945 struct mdinfo info;
946 char nbuf[64];
947 char nbuf1[64];
948 struct intel_super *super = st->sb;
949 int i;
950
951 if (!super->anchor->num_raid_devs)
952 return;
953
954 getinfo_super_imsm(st, &info, NULL);
955 fname_from_uuid(st, &info, nbuf, ':');
956 for (i = 0; i < super->anchor->num_raid_devs; i++) {
957 struct imsm_dev *dev = get_imsm_dev(super, i);
958
959 super->current_vol = i;
960 getinfo_super_imsm(st, &info, NULL);
961 fname_from_uuid(st, &info, nbuf1, ':');
962 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
963 dev->volume, nbuf + 5, i, nbuf1 + 5);
964 }
965 }
966
967 static void export_examine_super_imsm(struct supertype *st)
968 {
969 struct intel_super *super = st->sb;
970 struct imsm_super *mpb = super->anchor;
971 struct mdinfo info;
972 char nbuf[64];
973
974 getinfo_super_imsm(st, &info, NULL);
975 fname_from_uuid(st, &info, nbuf, ':');
976 printf("MD_METADATA=imsm\n");
977 printf("MD_LEVEL=container\n");
978 printf("MD_UUID=%s\n", nbuf+5);
979 printf("MD_DEVICES=%u\n", mpb->num_disks);
980 }
981
982 static void detail_super_imsm(struct supertype *st, char *homehost)
983 {
984 struct mdinfo info;
985 char nbuf[64];
986
987 getinfo_super_imsm(st, &info, NULL);
988 fname_from_uuid(st, &info, nbuf, ':');
989 printf("\n UUID : %s\n", nbuf + 5);
990 }
991
992 static void brief_detail_super_imsm(struct supertype *st)
993 {
994 struct mdinfo info;
995 char nbuf[64];
996 getinfo_super_imsm(st, &info, NULL);
997 fname_from_uuid(st, &info, nbuf, ':');
998 printf(" UUID=%s", nbuf + 5);
999 }
1000
1001 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
1002 static void fd2devname(int fd, char *name);
1003
1004 static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
1005 {
1006 /* dump an unsorted list of devices attached to ahci, as well as
1007 * non-connected ports
1008 */
1009 int hba_len = strlen(hba_path) + 1;
1010 struct dirent *ent;
1011 DIR *dir;
1012 char *path = NULL;
1013 int err = 0;
1014 unsigned long port_mask = (1 << port_count) - 1;
1015
1016 if (port_count > (int)sizeof(port_mask) * 8) {
1017 if (verbose)
1018 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
1019 return 2;
1020 }
1021
1022 /* scroll through /sys/dev/block looking for devices attached to
1023 * this hba
1024 */
1025 dir = opendir("/sys/dev/block");
1026 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1027 int fd;
1028 char model[64];
1029 char vendor[64];
1030 char buf[1024];
1031 int major, minor;
1032 char *device;
1033 char *c;
1034 int port;
1035 int type;
1036
1037 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
1038 continue;
1039 path = devt_to_devpath(makedev(major, minor));
1040 if (!path)
1041 continue;
1042 if (!path_attached_to_hba(path, hba_path)) {
1043 free(path);
1044 path = NULL;
1045 continue;
1046 }
1047
1048 /* retrieve the scsi device type */
1049 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
1050 if (verbose)
1051 fprintf(stderr, Name ": failed to allocate 'device'\n");
1052 err = 2;
1053 break;
1054 }
1055 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
1056 if (load_sys(device, buf) != 0) {
1057 if (verbose)
1058 fprintf(stderr, Name ": failed to read device type for %s\n",
1059 path);
1060 err = 2;
1061 free(device);
1062 break;
1063 }
1064 type = strtoul(buf, NULL, 10);
1065
1066 /* if it's not a disk print the vendor and model */
1067 if (!(type == 0 || type == 7 || type == 14)) {
1068 vendor[0] = '\0';
1069 model[0] = '\0';
1070 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
1071 if (load_sys(device, buf) == 0) {
1072 strncpy(vendor, buf, sizeof(vendor));
1073 vendor[sizeof(vendor) - 1] = '\0';
1074 c = (char *) &vendor[sizeof(vendor) - 1];
1075 while (isspace(*c) || *c == '\0')
1076 *c-- = '\0';
1077
1078 }
1079 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
1080 if (load_sys(device, buf) == 0) {
1081 strncpy(model, buf, sizeof(model));
1082 model[sizeof(model) - 1] = '\0';
1083 c = (char *) &model[sizeof(model) - 1];
1084 while (isspace(*c) || *c == '\0')
1085 *c-- = '\0';
1086 }
1087
1088 if (vendor[0] && model[0])
1089 sprintf(buf, "%.64s %.64s", vendor, model);
1090 else
1091 switch (type) { /* numbers from hald/linux/device.c */
1092 case 1: sprintf(buf, "tape"); break;
1093 case 2: sprintf(buf, "printer"); break;
1094 case 3: sprintf(buf, "processor"); break;
1095 case 4:
1096 case 5: sprintf(buf, "cdrom"); break;
1097 case 6: sprintf(buf, "scanner"); break;
1098 case 8: sprintf(buf, "media_changer"); break;
1099 case 9: sprintf(buf, "comm"); break;
1100 case 12: sprintf(buf, "raid"); break;
1101 default: sprintf(buf, "unknown");
1102 }
1103 } else
1104 buf[0] = '\0';
1105 free(device);
1106
1107 /* chop device path to 'host%d' and calculate the port number */
1108 c = strchr(&path[hba_len], '/');
1109 if (!c) {
1110 if (verbose)
1111 fprintf(stderr, Name ": %s - invalid path name\n", path + hba_len);
1112 err = 2;
1113 break;
1114 }
1115 *c = '\0';
1116 if (sscanf(&path[hba_len], "host%d", &port) == 1)
1117 port -= host_base;
1118 else {
1119 if (verbose) {
1120 *c = '/'; /* repair the full string */
1121 fprintf(stderr, Name ": failed to determine port number for %s\n",
1122 path);
1123 }
1124 err = 2;
1125 break;
1126 }
1127
1128 /* mark this port as used */
1129 port_mask &= ~(1 << port);
1130
1131 /* print out the device information */
1132 if (buf[0]) {
1133 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
1134 continue;
1135 }
1136
1137 fd = dev_open(ent->d_name, O_RDONLY);
1138 if (fd < 0)
1139 printf(" Port%d : - disk info unavailable -\n", port);
1140 else {
1141 fd2devname(fd, buf);
1142 printf(" Port%d : %s", port, buf);
1143 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
1144 printf(" (%s)\n", buf);
1145 else
1146 printf("()\n");
1147 }
1148 close(fd);
1149 free(path);
1150 path = NULL;
1151 }
1152 if (path)
1153 free(path);
1154 if (dir)
1155 closedir(dir);
1156 if (err == 0) {
1157 int i;
1158
1159 for (i = 0; i < port_count; i++)
1160 if (port_mask & (1 << i))
1161 printf(" Port%d : - no device attached -\n", i);
1162 }
1163
1164 return err;
1165 }
1166
1167 static int detail_platform_imsm(int verbose, int enumerate_only)
1168 {
1169 /* There are two components to imsm platform support, the ahci SATA
1170 * controller and the option-rom. To find the SATA controller we
1171 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1172 * controller with the Intel vendor id is present. This approach
1173 * allows mdadm to leverage the kernel's ahci detection logic, with the
1174 * caveat that if ahci.ko is not loaded mdadm will not be able to
1175 * detect platform raid capabilities. The option-rom resides in a
1176 * platform "Adapter ROM". We scan for its signature to retrieve the
1177 * platform capabilities. If raid support is disabled in the BIOS the
1178 * option-rom capability structure will not be available.
1179 */
1180 const struct imsm_orom *orom;
1181 struct sys_dev *list, *hba;
1182 DIR *dir;
1183 struct dirent *ent;
1184 const char *hba_path;
1185 int host_base = 0;
1186 int port_count = 0;
1187
1188 if (enumerate_only) {
1189 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
1190 return 0;
1191 return 2;
1192 }
1193
1194 list = find_driver_devices("pci", "ahci");
1195 for (hba = list; hba; hba = hba->next)
1196 if (devpath_to_vendor(hba->path) == 0x8086)
1197 break;
1198
1199 if (!hba) {
1200 if (verbose)
1201 fprintf(stderr, Name ": unable to find active ahci controller\n");
1202 free_sys_dev(&list);
1203 return 2;
1204 } else if (verbose)
1205 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
1206 hba_path = hba->path;
1207 hba->path = NULL;
1208 free_sys_dev(&list);
1209
1210 orom = find_imsm_orom();
1211 if (!orom) {
1212 if (verbose)
1213 fprintf(stderr, Name ": imsm option-rom not found\n");
1214 return 2;
1215 }
1216
1217 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1218 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
1219 orom->hotfix_ver, orom->build);
1220 printf(" RAID Levels :%s%s%s%s%s\n",
1221 imsm_orom_has_raid0(orom) ? " raid0" : "",
1222 imsm_orom_has_raid1(orom) ? " raid1" : "",
1223 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
1224 imsm_orom_has_raid10(orom) ? " raid10" : "",
1225 imsm_orom_has_raid5(orom) ? " raid5" : "");
1226 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1227 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
1228 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
1229 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
1230 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
1231 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
1232 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
1233 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
1234 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
1235 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
1236 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
1237 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
1238 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
1239 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
1240 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
1241 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
1242 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1243 printf(" Max Disks : %d\n", orom->tds);
1244 printf(" Max Volumes : %d\n", orom->vpa);
1245 printf(" I/O Controller : %s\n", hba_path);
1246
1247 /* find the smallest scsi host number to determine a port number base */
1248 dir = opendir(hba_path);
1249 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1250 int host;
1251
1252 if (sscanf(ent->d_name, "host%d", &host) != 1)
1253 continue;
1254 if (port_count == 0)
1255 host_base = host;
1256 else if (host < host_base)
1257 host_base = host;
1258
1259 if (host + 1 > port_count + host_base)
1260 port_count = host + 1 - host_base;
1261
1262 }
1263 if (dir)
1264 closedir(dir);
1265
1266 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1267 host_base, verbose) != 0) {
1268 if (verbose)
1269 fprintf(stderr, Name ": failed to enumerate ports\n");
1270 return 2;
1271 }
1272
1273 return 0;
1274 }
1275 #endif
1276
1277 static int match_home_imsm(struct supertype *st, char *homehost)
1278 {
1279 /* the imsm metadata format does not specify any host
1280 * identification information. We return -1 since we can never
1281 * confirm nor deny whether a given array is "meant" for this
1282 * host. We rely on compare_super and the 'family_num' fields to
1283 * exclude member disks that do not belong, and we rely on
1284 * mdadm.conf to specify the arrays that should be assembled.
1285 * Auto-assembly may still pick up "foreign" arrays.
1286 */
1287
1288 return -1;
1289 }
1290
1291 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1292 {
1293 /* The uuid returned here is used for:
1294 * uuid to put into bitmap file (Create, Grow)
1295 * uuid for backup header when saving critical section (Grow)
1296 * comparing uuids when re-adding a device into an array
1297 * In these cases the uuid required is that of the data-array,
1298 * not the device-set.
1299 * uuid to recognise same set when adding a missing device back
1300 * to an array. This is a uuid for the device-set.
1301 *
1302 * For each of these we can make do with a truncated
1303 * or hashed uuid rather than the original, as long as
1304 * everyone agrees.
1305 * In each case the uuid required is that of the data-array,
1306 * not the device-set.
1307 */
1308 /* imsm does not track uuid's so we synthesis one using sha1 on
1309 * - The signature (Which is constant for all imsm array, but no matter)
1310 * - the orig_family_num of the container
1311 * - the index number of the volume
1312 * - the 'serial' number of the volume.
1313 * Hopefully these are all constant.
1314 */
1315 struct intel_super *super = st->sb;
1316
1317 char buf[20];
1318 struct sha1_ctx ctx;
1319 struct imsm_dev *dev = NULL;
1320 __u32 family_num;
1321
1322 /* some mdadm versions failed to set ->orig_family_num, in which
1323 * case fall back to ->family_num. orig_family_num will be
1324 * fixed up with the first metadata update.
1325 */
1326 family_num = super->anchor->orig_family_num;
1327 if (family_num == 0)
1328 family_num = super->anchor->family_num;
1329 sha1_init_ctx(&ctx);
1330 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1331 sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
1332 if (super->current_vol >= 0)
1333 dev = get_imsm_dev(super, super->current_vol);
1334 if (dev) {
1335 __u32 vol = super->current_vol;
1336 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1337 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1338 }
1339 sha1_finish_ctx(&ctx, buf);
1340 memcpy(uuid, buf, 4*4);
1341 }
1342
1343 #if 0
1344 static void
1345 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1346 {
1347 __u8 *v = get_imsm_version(mpb);
1348 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1349 char major[] = { 0, 0, 0 };
1350 char minor[] = { 0 ,0, 0 };
1351 char patch[] = { 0, 0, 0 };
1352 char *ver_parse[] = { major, minor, patch };
1353 int i, j;
1354
1355 i = j = 0;
1356 while (*v != '\0' && v < end) {
1357 if (*v != '.' && j < 2)
1358 ver_parse[i][j++] = *v;
1359 else {
1360 i++;
1361 j = 0;
1362 }
1363 v++;
1364 }
1365
1366 *m = strtol(minor, NULL, 0);
1367 *p = strtol(patch, NULL, 0);
1368 }
1369 #endif
1370
1371 static __u32 migr_strip_blocks_resync(struct imsm_dev *dev)
1372 {
1373 /* migr_strip_size when repairing or initializing parity */
1374 struct imsm_map *map = get_imsm_map(dev, 0);
1375 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1376
1377 switch (get_imsm_raid_level(map)) {
1378 case 5:
1379 case 10:
1380 return chunk;
1381 default:
1382 return 128*1024 >> 9;
1383 }
1384 }
1385
1386 static __u32 migr_strip_blocks_rebuild(struct imsm_dev *dev)
1387 {
1388 /* migr_strip_size when rebuilding a degraded disk, no idea why
1389 * this is different than migr_strip_size_resync(), but it's good
1390 * to be compatible
1391 */
1392 struct imsm_map *map = get_imsm_map(dev, 1);
1393 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1394
1395 switch (get_imsm_raid_level(map)) {
1396 case 1:
1397 case 10:
1398 if (map->num_members % map->num_domains == 0)
1399 return 128*1024 >> 9;
1400 else
1401 return chunk;
1402 case 5:
1403 return max((__u32) 64*1024 >> 9, chunk);
1404 default:
1405 return 128*1024 >> 9;
1406 }
1407 }
1408
1409 static __u32 num_stripes_per_unit_resync(struct imsm_dev *dev)
1410 {
1411 struct imsm_map *lo = get_imsm_map(dev, 0);
1412 struct imsm_map *hi = get_imsm_map(dev, 1);
1413 __u32 lo_chunk = __le32_to_cpu(lo->blocks_per_strip);
1414 __u32 hi_chunk = __le32_to_cpu(hi->blocks_per_strip);
1415
1416 return max((__u32) 1, hi_chunk / lo_chunk);
1417 }
1418
1419 static __u32 num_stripes_per_unit_rebuild(struct imsm_dev *dev)
1420 {
1421 struct imsm_map *lo = get_imsm_map(dev, 0);
1422 int level = get_imsm_raid_level(lo);
1423
1424 if (level == 1 || level == 10) {
1425 struct imsm_map *hi = get_imsm_map(dev, 1);
1426
1427 return hi->num_domains;
1428 } else
1429 return num_stripes_per_unit_resync(dev);
1430 }
1431
1432 static __u8 imsm_num_data_members(struct imsm_dev *dev, int second_map)
1433 {
1434 /* named 'imsm_' because raid0, raid1 and raid10
1435 * counter-intuitively have the same number of data disks
1436 */
1437 struct imsm_map *map = get_imsm_map(dev, second_map);
1438
1439 switch (get_imsm_raid_level(map)) {
1440 case 0:
1441 case 1:
1442 case 10:
1443 return map->num_members;
1444 case 5:
1445 return map->num_members - 1;
1446 default:
1447 dprintf("%s: unsupported raid level\n", __func__);
1448 return 0;
1449 }
1450 }
1451
1452 static __u32 parity_segment_depth(struct imsm_dev *dev)
1453 {
1454 struct imsm_map *map = get_imsm_map(dev, 0);
1455 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1456
1457 switch(get_imsm_raid_level(map)) {
1458 case 1:
1459 case 10:
1460 return chunk * map->num_domains;
1461 case 5:
1462 return chunk * map->num_members;
1463 default:
1464 return chunk;
1465 }
1466 }
1467
1468 static __u32 map_migr_block(struct imsm_dev *dev, __u32 block)
1469 {
1470 struct imsm_map *map = get_imsm_map(dev, 1);
1471 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1472 __u32 strip = block / chunk;
1473
1474 switch (get_imsm_raid_level(map)) {
1475 case 1:
1476 case 10: {
1477 __u32 vol_strip = (strip * map->num_domains) + 1;
1478 __u32 vol_stripe = vol_strip / map->num_members;
1479
1480 return vol_stripe * chunk + block % chunk;
1481 } case 5: {
1482 __u32 stripe = strip / (map->num_members - 1);
1483
1484 return stripe * chunk + block % chunk;
1485 }
1486 default:
1487 return 0;
1488 }
1489 }
1490
1491 static __u64 blocks_per_migr_unit(struct imsm_dev *dev)
1492 {
1493 /* calculate the conversion factor between per member 'blocks'
1494 * (md/{resync,rebuild}_start) and imsm migration units, return
1495 * 0 for the 'not migrating' and 'unsupported migration' cases
1496 */
1497 if (!dev->vol.migr_state)
1498 return 0;
1499
1500 switch (migr_type(dev)) {
1501 case MIGR_GEN_MIGR:
1502 case MIGR_VERIFY:
1503 case MIGR_REPAIR:
1504 case MIGR_INIT: {
1505 struct imsm_map *map = get_imsm_map(dev, 0);
1506 __u32 stripes_per_unit;
1507 __u32 blocks_per_unit;
1508 __u32 parity_depth;
1509 __u32 migr_chunk;
1510 __u32 block_map;
1511 __u32 block_rel;
1512 __u32 segment;
1513 __u32 stripe;
1514 __u8 disks;
1515
1516 /* yes, this is really the translation of migr_units to
1517 * per-member blocks in the 'resync' case
1518 */
1519 stripes_per_unit = num_stripes_per_unit_resync(dev);
1520 migr_chunk = migr_strip_blocks_resync(dev);
1521 disks = imsm_num_data_members(dev, 0);
1522 blocks_per_unit = stripes_per_unit * migr_chunk * disks;
1523 stripe = __le32_to_cpu(map->blocks_per_strip) * disks;
1524 segment = blocks_per_unit / stripe;
1525 block_rel = blocks_per_unit - segment * stripe;
1526 parity_depth = parity_segment_depth(dev);
1527 block_map = map_migr_block(dev, block_rel);
1528 return block_map + parity_depth * segment;
1529 }
1530 case MIGR_REBUILD: {
1531 __u32 stripes_per_unit;
1532 __u32 migr_chunk;
1533
1534 stripes_per_unit = num_stripes_per_unit_rebuild(dev);
1535 migr_chunk = migr_strip_blocks_rebuild(dev);
1536 return migr_chunk * stripes_per_unit;
1537 }
1538 case MIGR_STATE_CHANGE:
1539 default:
1540 return 0;
1541 }
1542 }
1543
1544 static int imsm_level_to_layout(int level)
1545 {
1546 switch (level) {
1547 case 0:
1548 case 1:
1549 return 0;
1550 case 5:
1551 case 6:
1552 return ALGORITHM_LEFT_ASYMMETRIC;
1553 case 10:
1554 return 0x102;
1555 }
1556 return UnSet;
1557 }
1558
1559 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info, char *dmap)
1560 {
1561 struct intel_super *super = st->sb;
1562 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1563 struct imsm_map *map = get_imsm_map(dev, 0);
1564 struct imsm_map *prev_map = get_imsm_map(dev, 1);
1565 struct imsm_map *map_to_analyse = map;
1566 struct dl *dl;
1567 char *devname;
1568 int map_disks = info->array.raid_disks;
1569
1570 if (prev_map)
1571 map_to_analyse = prev_map;
1572
1573 for (dl = super->disks; dl; dl = dl->next)
1574 if (dl->raiddisk == info->disk.raid_disk)
1575 break;
1576 info->container_member = super->current_vol;
1577 info->array.raid_disks = map_to_analyse->num_members;
1578 info->array.level = get_imsm_raid_level(map_to_analyse);
1579 info->array.layout = imsm_level_to_layout(info->array.level);
1580 info->array.md_minor = -1;
1581 info->array.ctime = 0;
1582 info->array.utime = 0;
1583 info->array.chunk_size =
1584 __le16_to_cpu(map_to_analyse->blocks_per_strip) << 9;
1585 info->array.state = !dev->vol.dirty;
1586 info->custom_array_size = __le32_to_cpu(dev->size_high);
1587 info->custom_array_size <<= 32;
1588 info->custom_array_size |= __le32_to_cpu(dev->size_low);
1589 if (prev_map) {
1590 info->new_level = get_imsm_raid_level(map);
1591 info->new_layout = imsm_level_to_layout(info->new_level);
1592 info->new_chunk = __le16_to_cpu(map->blocks_per_strip) << 9;
1593 } else {
1594 info->new_level = UnSet;
1595 info->new_layout = UnSet;
1596 info->new_chunk = info->array.chunk_size;
1597 }
1598 info->disk.major = 0;
1599 info->disk.minor = 0;
1600 if (dl) {
1601 info->disk.major = dl->major;
1602 info->disk.minor = dl->minor;
1603 }
1604
1605 info->data_offset = __le32_to_cpu(map_to_analyse->pba_of_lba0);
1606 info->component_size =
1607 __le32_to_cpu(map_to_analyse->blocks_per_member);
1608 memset(info->uuid, 0, sizeof(info->uuid));
1609 info->recovery_start = MaxSector;
1610 info->reshape_active = (prev_map != NULL);
1611 if (info->reshape_active)
1612 info->delta_disks = map->num_members - prev_map->num_members;
1613 else
1614 info->delta_disks = 0;
1615
1616 if (map_to_analyse->map_state == IMSM_T_STATE_UNINITIALIZED ||
1617 dev->vol.dirty) {
1618 info->resync_start = 0;
1619 } else if (dev->vol.migr_state) {
1620 switch (migr_type(dev)) {
1621 case MIGR_REPAIR:
1622 case MIGR_INIT: {
1623 __u64 blocks_per_unit = blocks_per_migr_unit(dev);
1624 __u64 units = __le32_to_cpu(dev->vol.curr_migr_unit);
1625
1626 info->resync_start = blocks_per_unit * units;
1627 break;
1628 }
1629 case MIGR_VERIFY:
1630 /* we could emulate the checkpointing of
1631 * 'sync_action=check' migrations, but for now
1632 * we just immediately complete them
1633 */
1634 case MIGR_REBUILD:
1635 /* this is handled by container_content_imsm() */
1636 case MIGR_GEN_MIGR:
1637 case MIGR_STATE_CHANGE:
1638 /* FIXME handle other migrations */
1639 default:
1640 /* we are not dirty, so... */
1641 info->resync_start = MaxSector;
1642 }
1643 } else
1644 info->resync_start = MaxSector;
1645
1646 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1647 info->name[MAX_RAID_SERIAL_LEN] = 0;
1648
1649 info->array.major_version = -1;
1650 info->array.minor_version = -2;
1651 devname = devnum2devname(st->container_dev);
1652 *info->text_version = '\0';
1653 if (devname)
1654 sprintf(info->text_version, "/%s/%d", devname, info->container_member);
1655 free(devname);
1656 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1657 uuid_from_super_imsm(st, info->uuid);
1658
1659 if (dmap) {
1660 int i, j;
1661 for (i=0; i<map_disks; i++) {
1662 dmap[i] = 0;
1663 if (i < info->array.raid_disks) {
1664 struct imsm_disk *dsk;
1665 j = get_imsm_disk_idx(dev, i, -1);
1666 dsk = get_imsm_disk(super, j);
1667 if (dsk && (dsk->status & CONFIGURED_DISK))
1668 dmap[i] = 1;
1669 }
1670 }
1671 }
1672 }
1673
1674 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed);
1675 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev);
1676
1677 static struct imsm_disk *get_imsm_missing(struct intel_super *super, __u8 index)
1678 {
1679 struct dl *d;
1680
1681 for (d = super->missing; d; d = d->next)
1682 if (d->index == index)
1683 return &d->disk;
1684 return NULL;
1685 }
1686
1687 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map)
1688 {
1689 struct intel_super *super = st->sb;
1690 struct imsm_disk *disk;
1691 int map_disks = info->array.raid_disks;
1692 int max_enough = -1;
1693 int i;
1694 struct imsm_super *mpb;
1695
1696 if (super->current_vol >= 0) {
1697 getinfo_super_imsm_volume(st, info, map);
1698 return;
1699 }
1700
1701 /* Set raid_disks to zero so that Assemble will always pull in valid
1702 * spares
1703 */
1704 info->array.raid_disks = 0;
1705 info->array.level = LEVEL_CONTAINER;
1706 info->array.layout = 0;
1707 info->array.md_minor = -1;
1708 info->array.ctime = 0; /* N/A for imsm */
1709 info->array.utime = 0;
1710 info->array.chunk_size = 0;
1711
1712 info->disk.major = 0;
1713 info->disk.minor = 0;
1714 info->disk.raid_disk = -1;
1715 info->reshape_active = 0;
1716 info->array.major_version = -1;
1717 info->array.minor_version = -2;
1718 strcpy(info->text_version, "imsm");
1719 info->safe_mode_delay = 0;
1720 info->disk.number = -1;
1721 info->disk.state = 0;
1722 info->name[0] = 0;
1723 info->recovery_start = MaxSector;
1724
1725 /* do we have the all the insync disks that we expect? */
1726 mpb = super->anchor;
1727
1728 for (i = 0; i < mpb->num_raid_devs; i++) {
1729 struct imsm_dev *dev = get_imsm_dev(super, i);
1730 int failed, enough, j, missing = 0;
1731 struct imsm_map *map;
1732 __u8 state;
1733
1734 failed = imsm_count_failed(super, dev);
1735 state = imsm_check_degraded(super, dev, failed);
1736 map = get_imsm_map(dev, dev->vol.migr_state);
1737
1738 /* any newly missing disks?
1739 * (catches single-degraded vs double-degraded)
1740 */
1741 for (j = 0; j < map->num_members; j++) {
1742 __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1);
1743 __u32 idx = ord_to_idx(ord);
1744
1745 if (!(ord & IMSM_ORD_REBUILD) &&
1746 get_imsm_missing(super, idx)) {
1747 missing = 1;
1748 break;
1749 }
1750 }
1751
1752 if (state == IMSM_T_STATE_FAILED)
1753 enough = -1;
1754 else if (state == IMSM_T_STATE_DEGRADED &&
1755 (state != map->map_state || missing))
1756 enough = 0;
1757 else /* we're normal, or already degraded */
1758 enough = 1;
1759
1760 /* in the missing/failed disk case check to see
1761 * if at least one array is runnable
1762 */
1763 max_enough = max(max_enough, enough);
1764 }
1765 dprintf("%s: enough: %d\n", __func__, max_enough);
1766 info->container_enough = max_enough;
1767
1768 if (super->disks) {
1769 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1770
1771 disk = &super->disks->disk;
1772 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1773 info->component_size = reserved;
1774 info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
1775 /* we don't change info->disk.raid_disk here because
1776 * this state will be finalized in mdmon after we have
1777 * found the 'most fresh' version of the metadata
1778 */
1779 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
1780 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
1781 }
1782
1783 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1784 * ->compare_super may have updated the 'num_raid_devs' field for spares
1785 */
1786 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1787 uuid_from_super_imsm(st, info->uuid);
1788 else
1789 memcpy(info->uuid, uuid_zero, sizeof(uuid_zero));
1790
1791 /* I don't know how to compute 'map' on imsm, so use safe default */
1792 if (map) {
1793 int i;
1794 for (i = 0; i < map_disks; i++)
1795 map[i] = 1;
1796 }
1797
1798 }
1799
1800 /* allocates memory and fills disk in mdinfo structure
1801 * for each disk in array */
1802 struct mdinfo *getinfo_super_disks_imsm(struct supertype *st)
1803 {
1804 struct mdinfo *mddev = NULL;
1805 struct intel_super *super = st->sb;
1806 struct imsm_disk *disk;
1807 int count = 0;
1808 struct dl *dl;
1809 if (!super || !super->disks)
1810 return NULL;
1811 dl = super->disks;
1812 mddev = malloc(sizeof(*mddev));
1813 if (!mddev) {
1814 fprintf(stderr, Name ": Failed to allocate memory.\n");
1815 return NULL;
1816 }
1817 memset(mddev, 0, sizeof(*mddev));
1818 while (dl) {
1819 struct mdinfo *tmp;
1820 disk = &dl->disk;
1821 tmp = malloc(sizeof(*tmp));
1822 if (!tmp) {
1823 fprintf(stderr, Name ": Failed to allocate memory.\n");
1824 if (mddev)
1825 sysfs_free(mddev);
1826 return NULL;
1827 }
1828 memset(tmp, 0, sizeof(*tmp));
1829 if (mddev->devs)
1830 tmp->next = mddev->devs;
1831 mddev->devs = tmp;
1832 tmp->disk.number = count++;
1833 tmp->disk.major = dl->major;
1834 tmp->disk.minor = dl->minor;
1835 tmp->disk.state = is_configured(disk) ?
1836 (1 << MD_DISK_ACTIVE) : 0;
1837 tmp->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
1838 tmp->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
1839 tmp->disk.raid_disk = -1;
1840 dl = dl->next;
1841 }
1842 return mddev;
1843 }
1844
1845 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1846 char *update, char *devname, int verbose,
1847 int uuid_set, char *homehost)
1848 {
1849 /* For 'assemble' and 'force' we need to return non-zero if any
1850 * change was made. For others, the return value is ignored.
1851 * Update options are:
1852 * force-one : This device looks a bit old but needs to be included,
1853 * update age info appropriately.
1854 * assemble: clear any 'faulty' flag to allow this device to
1855 * be assembled.
1856 * force-array: Array is degraded but being forced, mark it clean
1857 * if that will be needed to assemble it.
1858 *
1859 * newdev: not used ????
1860 * grow: Array has gained a new device - this is currently for
1861 * linear only
1862 * resync: mark as dirty so a resync will happen.
1863 * name: update the name - preserving the homehost
1864 * uuid: Change the uuid of the array to match watch is given
1865 *
1866 * Following are not relevant for this imsm:
1867 * sparc2.2 : update from old dodgey metadata
1868 * super-minor: change the preferred_minor number
1869 * summaries: update redundant counters.
1870 * homehost: update the recorded homehost
1871 * _reshape_progress: record new reshape_progress position.
1872 */
1873 int rv = 1;
1874 struct intel_super *super = st->sb;
1875 struct imsm_super *mpb;
1876
1877 /* we can only update container info */
1878 if (!super || super->current_vol >= 0 || !super->anchor)
1879 return 1;
1880
1881 mpb = super->anchor;
1882
1883 if (strcmp(update, "uuid") == 0 && uuid_set && !info->update_private)
1884 rv = -1;
1885 else if (strcmp(update, "uuid") == 0 && uuid_set && info->update_private) {
1886 mpb->orig_family_num = *((__u32 *) info->update_private);
1887 rv = 0;
1888 } else if (strcmp(update, "uuid") == 0) {
1889 __u32 *new_family = malloc(sizeof(*new_family));
1890
1891 /* update orig_family_number with the incoming random
1892 * data, report the new effective uuid, and store the
1893 * new orig_family_num for future updates.
1894 */
1895 if (new_family) {
1896 memcpy(&mpb->orig_family_num, info->uuid, sizeof(__u32));
1897 uuid_from_super_imsm(st, info->uuid);
1898 *new_family = mpb->orig_family_num;
1899 info->update_private = new_family;
1900 rv = 0;
1901 }
1902 } else if (strcmp(update, "assemble") == 0)
1903 rv = 0;
1904 else
1905 rv = -1;
1906
1907 /* successful update? recompute checksum */
1908 if (rv == 0)
1909 mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb));
1910
1911 return rv;
1912 }
1913
1914 static size_t disks_to_mpb_size(int disks)
1915 {
1916 size_t size;
1917
1918 size = sizeof(struct imsm_super);
1919 size += (disks - 1) * sizeof(struct imsm_disk);
1920 size += 2 * sizeof(struct imsm_dev);
1921 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1922 size += (4 - 2) * sizeof(struct imsm_map);
1923 /* 4 possible disk_ord_tbl's */
1924 size += 4 * (disks - 1) * sizeof(__u32);
1925
1926 return size;
1927 }
1928
1929 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1930 {
1931 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1932 return 0;
1933
1934 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1935 }
1936
1937 static void free_devlist(struct intel_super *super)
1938 {
1939 struct intel_dev *dv;
1940
1941 while (super->devlist) {
1942 dv = super->devlist->next;
1943 free(super->devlist->dev);
1944 free(super->devlist);
1945 super->devlist = dv;
1946 }
1947 }
1948
1949 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1950 {
1951 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1952 }
1953
1954 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1955 {
1956 /*
1957 * return:
1958 * 0 same, or first was empty, and second was copied
1959 * 1 second had wrong number
1960 * 2 wrong uuid
1961 * 3 wrong other info
1962 */
1963 struct intel_super *first = st->sb;
1964 struct intel_super *sec = tst->sb;
1965
1966 if (!first) {
1967 st->sb = tst->sb;
1968 tst->sb = NULL;
1969 return 0;
1970 }
1971
1972 /* if an anchor does not have num_raid_devs set then it is a free
1973 * floating spare
1974 */
1975 if (first->anchor->num_raid_devs > 0 &&
1976 sec->anchor->num_raid_devs > 0) {
1977 /* Determine if these disks might ever have been
1978 * related. Further disambiguation can only take place
1979 * in load_super_imsm_all
1980 */
1981 __u32 first_family = first->anchor->orig_family_num;
1982 __u32 sec_family = sec->anchor->orig_family_num;
1983
1984 if (memcmp(first->anchor->sig, sec->anchor->sig,
1985 MAX_SIGNATURE_LENGTH) != 0)
1986 return 3;
1987
1988 if (first_family == 0)
1989 first_family = first->anchor->family_num;
1990 if (sec_family == 0)
1991 sec_family = sec->anchor->family_num;
1992
1993 if (first_family != sec_family)
1994 return 3;
1995
1996 }
1997
1998
1999 /* if 'first' is a spare promote it to a populated mpb with sec's
2000 * family number
2001 */
2002 if (first->anchor->num_raid_devs == 0 &&
2003 sec->anchor->num_raid_devs > 0) {
2004 int i;
2005 struct intel_dev *dv;
2006 struct imsm_dev *dev;
2007
2008 /* we need to copy raid device info from sec if an allocation
2009 * fails here we don't associate the spare
2010 */
2011 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
2012 dv = malloc(sizeof(*dv));
2013 if (!dv)
2014 break;
2015 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
2016 if (!dev) {
2017 free(dv);
2018 break;
2019 }
2020 dv->dev = dev;
2021 dv->index = i;
2022 dv->next = first->devlist;
2023 first->devlist = dv;
2024 }
2025 if (i < sec->anchor->num_raid_devs) {
2026 /* allocation failure */
2027 free_devlist(first);
2028 fprintf(stderr, "imsm: failed to associate spare\n");
2029 return 3;
2030 }
2031 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
2032 first->anchor->orig_family_num = sec->anchor->orig_family_num;
2033 first->anchor->family_num = sec->anchor->family_num;
2034 memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH);
2035 for (i = 0; i < sec->anchor->num_raid_devs; i++)
2036 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
2037 }
2038
2039 return 0;
2040 }
2041
2042 static void fd2devname(int fd, char *name)
2043 {
2044 struct stat st;
2045 char path[256];
2046 char dname[PATH_MAX];
2047 char *nm;
2048 int rv;
2049
2050 name[0] = '\0';
2051 if (fstat(fd, &st) != 0)
2052 return;
2053 sprintf(path, "/sys/dev/block/%d:%d",
2054 major(st.st_rdev), minor(st.st_rdev));
2055
2056 rv = readlink(path, dname, sizeof(dname));
2057 if (rv <= 0)
2058 return;
2059
2060 dname[rv] = '\0';
2061 nm = strrchr(dname, '/');
2062 nm++;
2063 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
2064 }
2065
2066 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
2067
2068 static int imsm_read_serial(int fd, char *devname,
2069 __u8 serial[MAX_RAID_SERIAL_LEN])
2070 {
2071 unsigned char scsi_serial[255];
2072 int rv;
2073 int rsp_len;
2074 int len;
2075 char *dest;
2076 char *src;
2077 char *rsp_buf;
2078 int i;
2079
2080 memset(scsi_serial, 0, sizeof(scsi_serial));
2081
2082 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
2083
2084 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
2085 memset(serial, 0, MAX_RAID_SERIAL_LEN);
2086 fd2devname(fd, (char *) serial);
2087 return 0;
2088 }
2089
2090 if (rv != 0) {
2091 if (devname)
2092 fprintf(stderr,
2093 Name ": Failed to retrieve serial for %s\n",
2094 devname);
2095 return rv;
2096 }
2097
2098 rsp_len = scsi_serial[3];
2099 if (!rsp_len) {
2100 if (devname)
2101 fprintf(stderr,
2102 Name ": Failed to retrieve serial for %s\n",
2103 devname);
2104 return 2;
2105 }
2106 rsp_buf = (char *) &scsi_serial[4];
2107
2108 /* trim all whitespace and non-printable characters and convert
2109 * ':' to ';'
2110 */
2111 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
2112 src = &rsp_buf[i];
2113 if (*src > 0x20) {
2114 /* ':' is reserved for use in placeholder serial
2115 * numbers for missing disks
2116 */
2117 if (*src == ':')
2118 *dest++ = ';';
2119 else
2120 *dest++ = *src;
2121 }
2122 }
2123 len = dest - rsp_buf;
2124 dest = rsp_buf;
2125
2126 /* truncate leading characters */
2127 if (len > MAX_RAID_SERIAL_LEN) {
2128 dest += len - MAX_RAID_SERIAL_LEN;
2129 len = MAX_RAID_SERIAL_LEN;
2130 }
2131
2132 memset(serial, 0, MAX_RAID_SERIAL_LEN);
2133 memcpy(serial, dest, len);
2134
2135 return 0;
2136 }
2137
2138 static int serialcmp(__u8 *s1, __u8 *s2)
2139 {
2140 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
2141 }
2142
2143 static void serialcpy(__u8 *dest, __u8 *src)
2144 {
2145 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
2146 }
2147
2148 #ifndef MDASSEMBLE
2149 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
2150 {
2151 struct dl *dl;
2152
2153 for (dl = super->disks; dl; dl = dl->next)
2154 if (serialcmp(dl->serial, serial) == 0)
2155 break;
2156
2157 return dl;
2158 }
2159 #endif
2160
2161 static struct imsm_disk *
2162 __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
2163 {
2164 int i;
2165
2166 for (i = 0; i < mpb->num_disks; i++) {
2167 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2168
2169 if (serialcmp(disk->serial, serial) == 0) {
2170 if (idx)
2171 *idx = i;
2172 return disk;
2173 }
2174 }
2175
2176 return NULL;
2177 }
2178
2179 static int
2180 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
2181 {
2182 struct imsm_disk *disk;
2183 struct dl *dl;
2184 struct stat stb;
2185 int rv;
2186 char name[40];
2187 __u8 serial[MAX_RAID_SERIAL_LEN];
2188
2189 rv = imsm_read_serial(fd, devname, serial);
2190
2191 if (rv != 0)
2192 return 2;
2193
2194 dl = calloc(1, sizeof(*dl));
2195 if (!dl) {
2196 if (devname)
2197 fprintf(stderr,
2198 Name ": failed to allocate disk buffer for %s\n",
2199 devname);
2200 return 2;
2201 }
2202
2203 fstat(fd, &stb);
2204 dl->major = major(stb.st_rdev);
2205 dl->minor = minor(stb.st_rdev);
2206 dl->next = super->disks;
2207 dl->fd = keep_fd ? fd : -1;
2208 assert(super->disks == NULL);
2209 super->disks = dl;
2210 serialcpy(dl->serial, serial);
2211 dl->index = -2;
2212 dl->e = NULL;
2213 fd2devname(fd, name);
2214 if (devname)
2215 dl->devname = strdup(devname);
2216 else
2217 dl->devname = strdup(name);
2218
2219 /* look up this disk's index in the current anchor */
2220 disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
2221 if (disk) {
2222 dl->disk = *disk;
2223 /* only set index on disks that are a member of a
2224 * populated contianer, i.e. one with raid_devs
2225 */
2226 if (is_failed(&dl->disk))
2227 dl->index = -2;
2228 else if (is_spare(&dl->disk))
2229 dl->index = -1;
2230 }
2231
2232 return 0;
2233 }
2234
2235 #ifndef MDASSEMBLE
2236 /* When migrating map0 contains the 'destination' state while map1
2237 * contains the current state. When not migrating map0 contains the
2238 * current state. This routine assumes that map[0].map_state is set to
2239 * the current array state before being called.
2240 *
2241 * Migration is indicated by one of the following states
2242 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
2243 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
2244 * map1state=unitialized)
2245 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
2246 * map1state=normal)
2247 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
2248 * map1state=degraded)
2249 */
2250 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
2251 {
2252 struct imsm_map *dest;
2253 struct imsm_map *src = get_imsm_map(dev, 0);
2254
2255 dev->vol.migr_state = 1;
2256 set_migr_type(dev, migr_type);
2257 dev->vol.curr_migr_unit = 0;
2258 dest = get_imsm_map(dev, 1);
2259
2260 /* duplicate and then set the target end state in map[0] */
2261 memcpy(dest, src, sizeof_imsm_map(src));
2262 if ((migr_type == MIGR_REBUILD) ||
2263 (migr_type == MIGR_GEN_MIGR)) {
2264 __u32 ord;
2265 int i;
2266
2267 for (i = 0; i < src->num_members; i++) {
2268 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
2269 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
2270 }
2271 }
2272
2273 src->map_state = to_state;
2274 }
2275
2276 static void end_migration(struct imsm_dev *dev, __u8 map_state)
2277 {
2278 struct imsm_map *map = get_imsm_map(dev, 0);
2279 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
2280 int i, j;
2281
2282 /* merge any IMSM_ORD_REBUILD bits that were not successfully
2283 * completed in the last migration.
2284 *
2285 * FIXME add support for raid-level-migration
2286 */
2287 for (i = 0; i < prev->num_members; i++)
2288 for (j = 0; j < map->num_members; j++)
2289 /* during online capacity expansion
2290 * disks position can be changed if takeover is used
2291 */
2292 if (ord_to_idx(map->disk_ord_tbl[j]) ==
2293 ord_to_idx(prev->disk_ord_tbl[i])) {
2294 map->disk_ord_tbl[j] |= prev->disk_ord_tbl[i];
2295 break;
2296 }
2297
2298 dev->vol.migr_state = 0;
2299 dev->vol.migr_type = 0;
2300 dev->vol.curr_migr_unit = 0;
2301 map->map_state = map_state;
2302 }
2303 #endif
2304
2305 static int parse_raid_devices(struct intel_super *super)
2306 {
2307 int i;
2308 struct imsm_dev *dev_new;
2309 size_t len, len_migr;
2310 size_t space_needed = 0;
2311 struct imsm_super *mpb = super->anchor;
2312
2313 for (i = 0; i < super->anchor->num_raid_devs; i++) {
2314 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
2315 struct intel_dev *dv;
2316
2317 len = sizeof_imsm_dev(dev_iter, 0);
2318 len_migr = sizeof_imsm_dev(dev_iter, 1);
2319 if (len_migr > len)
2320 space_needed += len_migr - len;
2321
2322 dv = malloc(sizeof(*dv));
2323 if (!dv)
2324 return 1;
2325 dev_new = malloc(len_migr);
2326 if (!dev_new) {
2327 free(dv);
2328 return 1;
2329 }
2330 imsm_copy_dev(dev_new, dev_iter);
2331 dv->dev = dev_new;
2332 dv->index = i;
2333 dv->next = super->devlist;
2334 super->devlist = dv;
2335 }
2336
2337 /* ensure that super->buf is large enough when all raid devices
2338 * are migrating
2339 */
2340 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
2341 void *buf;
2342
2343 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
2344 if (posix_memalign(&buf, 512, len) != 0)
2345 return 1;
2346
2347 memcpy(buf, super->buf, super->len);
2348 memset(buf + super->len, 0, len - super->len);
2349 free(super->buf);
2350 super->buf = buf;
2351 super->len = len;
2352 }
2353
2354 return 0;
2355 }
2356
2357 /* retrieve a pointer to the bbm log which starts after all raid devices */
2358 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
2359 {
2360 void *ptr = NULL;
2361
2362 if (__le32_to_cpu(mpb->bbm_log_size)) {
2363 ptr = mpb;
2364 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
2365 }
2366
2367 return ptr;
2368 }
2369
2370 static void __free_imsm(struct intel_super *super, int free_disks);
2371
2372 /* load_imsm_mpb - read matrix metadata
2373 * allocates super->mpb to be freed by free_super
2374 */
2375 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
2376 {
2377 unsigned long long dsize;
2378 unsigned long long sectors;
2379 struct stat;
2380 struct imsm_super *anchor;
2381 __u32 check_sum;
2382
2383 get_dev_size(fd, NULL, &dsize);
2384 if (dsize < 1024) {
2385 if (devname)
2386 fprintf(stderr,
2387 Name ": %s: device to small for imsm\n",
2388 devname);
2389 return 1;
2390 }
2391
2392 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
2393 if (devname)
2394 fprintf(stderr,
2395 Name ": Cannot seek to anchor block on %s: %s\n",
2396 devname, strerror(errno));
2397 return 1;
2398 }
2399
2400 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
2401 if (devname)
2402 fprintf(stderr,
2403 Name ": Failed to allocate imsm anchor buffer"
2404 " on %s\n", devname);
2405 return 1;
2406 }
2407 if (read(fd, anchor, 512) != 512) {
2408 if (devname)
2409 fprintf(stderr,
2410 Name ": Cannot read anchor block on %s: %s\n",
2411 devname, strerror(errno));
2412 free(anchor);
2413 return 1;
2414 }
2415
2416 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
2417 if (devname)
2418 fprintf(stderr,
2419 Name ": no IMSM anchor on %s\n", devname);
2420 free(anchor);
2421 return 2;
2422 }
2423
2424 __free_imsm(super, 0);
2425 super->len = ROUND_UP(anchor->mpb_size, 512);
2426 if (posix_memalign(&super->buf, 512, super->len) != 0) {
2427 if (devname)
2428 fprintf(stderr,
2429 Name ": unable to allocate %zu byte mpb buffer\n",
2430 super->len);
2431 free(anchor);
2432 return 2;
2433 }
2434 memcpy(super->buf, anchor, 512);
2435
2436 sectors = mpb_sectors(anchor) - 1;
2437 free(anchor);
2438 if (!sectors) {
2439 check_sum = __gen_imsm_checksum(super->anchor);
2440 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
2441 if (devname)
2442 fprintf(stderr,
2443 Name ": IMSM checksum %x != %x on %s\n",
2444 check_sum,
2445 __le32_to_cpu(super->anchor->check_sum),
2446 devname);
2447 return 2;
2448 }
2449
2450 return 0;
2451 }
2452
2453 /* read the extended mpb */
2454 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
2455 if (devname)
2456 fprintf(stderr,
2457 Name ": Cannot seek to extended mpb on %s: %s\n",
2458 devname, strerror(errno));
2459 return 1;
2460 }
2461
2462 if ((unsigned)read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
2463 if (devname)
2464 fprintf(stderr,
2465 Name ": Cannot read extended mpb on %s: %s\n",
2466 devname, strerror(errno));
2467 return 2;
2468 }
2469
2470 check_sum = __gen_imsm_checksum(super->anchor);
2471 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
2472 if (devname)
2473 fprintf(stderr,
2474 Name ": IMSM checksum %x != %x on %s\n",
2475 check_sum, __le32_to_cpu(super->anchor->check_sum),
2476 devname);
2477 return 3;
2478 }
2479
2480 /* FIXME the BBM log is disk specific so we cannot use this global
2481 * buffer for all disks. Ok for now since we only look at the global
2482 * bbm_log_size parameter to gate assembly
2483 */
2484 super->bbm_log = __get_imsm_bbm_log(super->anchor);
2485
2486 return 0;
2487 }
2488
2489 static int
2490 load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
2491 {
2492 int err;
2493
2494 err = load_imsm_mpb(fd, super, devname);
2495 if (err)
2496 return err;
2497 err = load_imsm_disk(fd, super, devname, keep_fd);
2498 if (err)
2499 return err;
2500 err = parse_raid_devices(super);
2501
2502 return err;
2503 }
2504
2505 static void __free_imsm_disk(struct dl *d)
2506 {
2507 if (d->fd >= 0)
2508 close(d->fd);
2509 if (d->devname)
2510 free(d->devname);
2511 if (d->e)
2512 free(d->e);
2513 free(d);
2514
2515 }
2516
2517 static void free_imsm_disks(struct intel_super *super)
2518 {
2519 struct dl *d;
2520
2521 while (super->disks) {
2522 d = super->disks;
2523 super->disks = d->next;
2524 __free_imsm_disk(d);
2525 }
2526 while (super->missing) {
2527 d = super->missing;
2528 super->missing = d->next;
2529 __free_imsm_disk(d);
2530 }
2531
2532 }
2533
2534 /* free all the pieces hanging off of a super pointer */
2535 static void __free_imsm(struct intel_super *super, int free_disks)
2536 {
2537 if (super->buf) {
2538 free(super->buf);
2539 super->buf = NULL;
2540 }
2541 if (free_disks)
2542 free_imsm_disks(super);
2543 free_devlist(super);
2544 if (super->hba) {
2545 free((void *) super->hba);
2546 super->hba = NULL;
2547 }
2548 }
2549
2550 static void free_imsm(struct intel_super *super)
2551 {
2552 __free_imsm(super, 1);
2553 free(super);
2554 }
2555
2556 static void free_super_imsm(struct supertype *st)
2557 {
2558 struct intel_super *super = st->sb;
2559
2560 if (!super)
2561 return;
2562
2563 free_imsm(super);
2564 st->sb = NULL;
2565 }
2566
2567 static struct intel_super *alloc_super(void)
2568 {
2569 struct intel_super *super = malloc(sizeof(*super));
2570
2571 if (super) {
2572 memset(super, 0, sizeof(*super));
2573 super->current_vol = -1;
2574 super->create_offset = ~((__u32 ) 0);
2575 if (!check_env("IMSM_NO_PLATFORM"))
2576 super->orom = find_imsm_orom();
2577 if (super->orom && !check_env("IMSM_TEST_OROM")) {
2578 struct sys_dev *list, *ent;
2579
2580 /* find the first intel ahci controller */
2581 list = find_driver_devices("pci", "ahci");
2582 for (ent = list; ent; ent = ent->next)
2583 if (devpath_to_vendor(ent->path) == 0x8086)
2584 break;
2585 if (ent) {
2586 super->hba = ent->path;
2587 ent->path = NULL;
2588 }
2589 free_sys_dev(&list);
2590 }
2591 }
2592
2593 return super;
2594 }
2595
2596 #ifndef MDASSEMBLE
2597 /* find_missing - helper routine for load_super_imsm_all that identifies
2598 * disks that have disappeared from the system. This routine relies on
2599 * the mpb being uptodate, which it is at load time.
2600 */
2601 static int find_missing(struct intel_super *super)
2602 {
2603 int i;
2604 struct imsm_super *mpb = super->anchor;
2605 struct dl *dl;
2606 struct imsm_disk *disk;
2607
2608 for (i = 0; i < mpb->num_disks; i++) {
2609 disk = __get_imsm_disk(mpb, i);
2610 dl = serial_to_dl(disk->serial, super);
2611 if (dl)
2612 continue;
2613
2614 dl = malloc(sizeof(*dl));
2615 if (!dl)
2616 return 1;
2617 dl->major = 0;
2618 dl->minor = 0;
2619 dl->fd = -1;
2620 dl->devname = strdup("missing");
2621 dl->index = i;
2622 serialcpy(dl->serial, disk->serial);
2623 dl->disk = *disk;
2624 dl->e = NULL;
2625 dl->next = super->missing;
2626 super->missing = dl;
2627 }
2628
2629 return 0;
2630 }
2631
2632 static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
2633 {
2634 struct intel_disk *idisk = disk_list;
2635
2636 while (idisk) {
2637 if (serialcmp(idisk->disk.serial, serial) == 0)
2638 break;
2639 idisk = idisk->next;
2640 }
2641
2642 return idisk;
2643 }
2644
2645 static int __prep_thunderdome(struct intel_super **table, int tbl_size,
2646 struct intel_super *super,
2647 struct intel_disk **disk_list)
2648 {
2649 struct imsm_disk *d = &super->disks->disk;
2650 struct imsm_super *mpb = super->anchor;
2651 int i, j;
2652
2653 for (i = 0; i < tbl_size; i++) {
2654 struct imsm_super *tbl_mpb = table[i]->anchor;
2655 struct imsm_disk *tbl_d = &table[i]->disks->disk;
2656
2657 if (tbl_mpb->family_num == mpb->family_num) {
2658 if (tbl_mpb->check_sum == mpb->check_sum) {
2659 dprintf("%s: mpb from %d:%d matches %d:%d\n",
2660 __func__, super->disks->major,
2661 super->disks->minor,
2662 table[i]->disks->major,
2663 table[i]->disks->minor);
2664 break;
2665 }
2666
2667 if (((is_configured(d) && !is_configured(tbl_d)) ||
2668 is_configured(d) == is_configured(tbl_d)) &&
2669 tbl_mpb->generation_num < mpb->generation_num) {
2670 /* current version of the mpb is a
2671 * better candidate than the one in
2672 * super_table, but copy over "cross
2673 * generational" status
2674 */
2675 struct intel_disk *idisk;
2676
2677 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
2678 __func__, super->disks->major,
2679 super->disks->minor,
2680 table[i]->disks->major,
2681 table[i]->disks->minor);
2682
2683 idisk = disk_list_get(tbl_d->serial, *disk_list);
2684 if (idisk && is_failed(&idisk->disk))
2685 tbl_d->status |= FAILED_DISK;
2686 break;
2687 } else {
2688 struct intel_disk *idisk;
2689 struct imsm_disk *disk;
2690
2691 /* tbl_mpb is more up to date, but copy
2692 * over cross generational status before
2693 * returning
2694 */
2695 disk = __serial_to_disk(d->serial, mpb, NULL);
2696 if (disk && is_failed(disk))
2697 d->status |= FAILED_DISK;
2698
2699 idisk = disk_list_get(d->serial, *disk_list);
2700 if (idisk) {
2701 idisk->owner = i;
2702 if (disk && is_configured(disk))
2703 idisk->disk.status |= CONFIGURED_DISK;
2704 }
2705
2706 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
2707 __func__, super->disks->major,
2708 super->disks->minor,
2709 table[i]->disks->major,
2710 table[i]->disks->minor);
2711
2712 return tbl_size;
2713 }
2714 }
2715 }
2716
2717 if (i >= tbl_size)
2718 table[tbl_size++] = super;
2719 else
2720 table[i] = super;
2721
2722 /* update/extend the merged list of imsm_disk records */
2723 for (j = 0; j < mpb->num_disks; j++) {
2724 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
2725 struct intel_disk *idisk;
2726
2727 idisk = disk_list_get(disk->serial, *disk_list);
2728 if (idisk) {
2729 idisk->disk.status |= disk->status;
2730 if (is_configured(&idisk->disk) ||
2731 is_failed(&idisk->disk))
2732 idisk->disk.status &= ~(SPARE_DISK);
2733 } else {
2734 idisk = calloc(1, sizeof(*idisk));
2735 if (!idisk)
2736 return -1;
2737 idisk->owner = IMSM_UNKNOWN_OWNER;
2738 idisk->disk = *disk;
2739 idisk->next = *disk_list;
2740 *disk_list = idisk;
2741 }
2742
2743 if (serialcmp(idisk->disk.serial, d->serial) == 0)
2744 idisk->owner = i;
2745 }
2746
2747 return tbl_size;
2748 }
2749
2750 static struct intel_super *
2751 validate_members(struct intel_super *super, struct intel_disk *disk_list,
2752 const int owner)
2753 {
2754 struct imsm_super *mpb = super->anchor;
2755 int ok_count = 0;
2756 int i;
2757
2758 for (i = 0; i < mpb->num_disks; i++) {
2759 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2760 struct intel_disk *idisk;
2761
2762 idisk = disk_list_get(disk->serial, disk_list);
2763 if (idisk) {
2764 if (idisk->owner == owner ||
2765 idisk->owner == IMSM_UNKNOWN_OWNER)
2766 ok_count++;
2767 else
2768 dprintf("%s: '%.16s' owner %d != %d\n",
2769 __func__, disk->serial, idisk->owner,
2770 owner);
2771 } else {
2772 dprintf("%s: unknown disk %x [%d]: %.16s\n",
2773 __func__, __le32_to_cpu(mpb->family_num), i,
2774 disk->serial);
2775 break;
2776 }
2777 }
2778
2779 if (ok_count == mpb->num_disks)
2780 return super;
2781 return NULL;
2782 }
2783
2784 static void show_conflicts(__u32 family_num, struct intel_super *super_list)
2785 {
2786 struct intel_super *s;
2787
2788 for (s = super_list; s; s = s->next) {
2789 if (family_num != s->anchor->family_num)
2790 continue;
2791 fprintf(stderr, "Conflict, offlining family %#x on '%s'\n",
2792 __le32_to_cpu(family_num), s->disks->devname);
2793 }
2794 }
2795
2796 static struct intel_super *
2797 imsm_thunderdome(struct intel_super **super_list, int len)
2798 {
2799 struct intel_super *super_table[len];
2800 struct intel_disk *disk_list = NULL;
2801 struct intel_super *champion, *spare;
2802 struct intel_super *s, **del;
2803 int tbl_size = 0;
2804 int conflict;
2805 int i;
2806
2807 memset(super_table, 0, sizeof(super_table));
2808 for (s = *super_list; s; s = s->next)
2809 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
2810
2811 for (i = 0; i < tbl_size; i++) {
2812 struct imsm_disk *d;
2813 struct intel_disk *idisk;
2814 struct imsm_super *mpb = super_table[i]->anchor;
2815
2816 s = super_table[i];
2817 d = &s->disks->disk;
2818
2819 /* 'd' must appear in merged disk list for its
2820 * configuration to be valid
2821 */
2822 idisk = disk_list_get(d->serial, disk_list);
2823 if (idisk && idisk->owner == i)
2824 s = validate_members(s, disk_list, i);
2825 else
2826 s = NULL;
2827
2828 if (!s)
2829 dprintf("%s: marking family: %#x from %d:%d offline\n",
2830 __func__, mpb->family_num,
2831 super_table[i]->disks->major,
2832 super_table[i]->disks->minor);
2833 super_table[i] = s;
2834 }
2835
2836 /* This is where the mdadm implementation differs from the Windows
2837 * driver which has no strict concept of a container. We can only
2838 * assemble one family from a container, so when returning a prodigal
2839 * array member to this system the code will not be able to disambiguate
2840 * the container contents that should be assembled ("foreign" versus
2841 * "local"). It requires user intervention to set the orig_family_num
2842 * to a new value to establish a new container. The Windows driver in
2843 * this situation fixes up the volume name in place and manages the
2844 * foreign array as an independent entity.
2845 */
2846 s = NULL;
2847 spare = NULL;
2848 conflict = 0;
2849 for (i = 0; i < tbl_size; i++) {
2850 struct intel_super *tbl_ent = super_table[i];
2851 int is_spare = 0;
2852
2853 if (!tbl_ent)
2854 continue;
2855
2856 if (tbl_ent->anchor->num_raid_devs == 0) {
2857 spare = tbl_ent;
2858 is_spare = 1;
2859 }
2860
2861 if (s && !is_spare) {
2862 show_conflicts(tbl_ent->anchor->family_num, *super_list);
2863 conflict++;
2864 } else if (!s && !is_spare)
2865 s = tbl_ent;
2866 }
2867
2868 if (!s)
2869 s = spare;
2870 if (!s) {
2871 champion = NULL;
2872 goto out;
2873 }
2874 champion = s;
2875
2876 if (conflict)
2877 fprintf(stderr, "Chose family %#x on '%s', "
2878 "assemble conflicts to new container with '--update=uuid'\n",
2879 __le32_to_cpu(s->anchor->family_num), s->disks->devname);
2880
2881 /* collect all dl's onto 'champion', and update them to
2882 * champion's version of the status
2883 */
2884 for (s = *super_list; s; s = s->next) {
2885 struct imsm_super *mpb = champion->anchor;
2886 struct dl *dl = s->disks;
2887
2888 if (s == champion)
2889 continue;
2890
2891 for (i = 0; i < mpb->num_disks; i++) {
2892 struct imsm_disk *disk;
2893
2894 disk = __serial_to_disk(dl->serial, mpb, &dl->index);
2895 if (disk) {
2896 dl->disk = *disk;
2897 /* only set index on disks that are a member of
2898 * a populated contianer, i.e. one with
2899 * raid_devs
2900 */
2901 if (is_failed(&dl->disk))
2902 dl->index = -2;
2903 else if (is_spare(&dl->disk))
2904 dl->index = -1;
2905 break;
2906 }
2907 }
2908
2909 if (i >= mpb->num_disks) {
2910 struct intel_disk *idisk;
2911
2912 idisk = disk_list_get(dl->serial, disk_list);
2913 if (idisk && is_spare(&idisk->disk) &&
2914 !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
2915 dl->index = -1;
2916 else {
2917 dl->index = -2;
2918 continue;
2919 }
2920 }
2921
2922 dl->next = champion->disks;
2923 champion->disks = dl;
2924 s->disks = NULL;
2925 }
2926
2927 /* delete 'champion' from super_list */
2928 for (del = super_list; *del; ) {
2929 if (*del == champion) {
2930 *del = (*del)->next;
2931 break;
2932 } else
2933 del = &(*del)->next;
2934 }
2935 champion->next = NULL;
2936
2937 out:
2938 while (disk_list) {
2939 struct intel_disk *idisk = disk_list;
2940
2941 disk_list = disk_list->next;
2942 free(idisk);
2943 }
2944
2945 return champion;
2946 }
2947
2948 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
2949 char *devname)
2950 {
2951 struct mdinfo *sra;
2952 struct intel_super *super_list = NULL;
2953 struct intel_super *super = NULL;
2954 int devnum = fd2devnum(fd);
2955 struct mdinfo *sd;
2956 int retry;
2957 int err = 0;
2958 int i;
2959
2960 /* check if 'fd' an opened container */
2961 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2962 if (!sra)
2963 return 1;
2964
2965 if (sra->array.major_version != -1 ||
2966 sra->array.minor_version != -2 ||
2967 strcmp(sra->text_version, "imsm") != 0) {
2968 err = 1;
2969 goto error;
2970 }
2971 /* load all mpbs */
2972 for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
2973 struct intel_super *s = alloc_super();
2974 char nm[32];
2975 int dfd;
2976
2977 err = 1;
2978 if (!s)
2979 goto error;
2980 s->next = super_list;
2981 super_list = s;
2982
2983 err = 2;
2984 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2985 dfd = dev_open(nm, O_RDWR);
2986 if (dfd < 0)
2987 goto error;
2988
2989 err = load_and_parse_mpb(dfd, s, NULL, 1);
2990
2991 /* retry the load if we might have raced against mdmon */
2992 if (err == 3 && mdmon_running(devnum))
2993 for (retry = 0; retry < 3; retry++) {
2994 usleep(3000);
2995 err = load_and_parse_mpb(dfd, s, NULL, 1);
2996 if (err != 3)
2997 break;
2998 }
2999 if (err)
3000 goto error;
3001 }
3002
3003 /* all mpbs enter, maybe one leaves */
3004 super = imsm_thunderdome(&super_list, i);
3005 if (!super) {
3006 err = 1;
3007 goto error;
3008 }
3009
3010 if (find_missing(super) != 0) {
3011 free_imsm(super);
3012 err = 2;
3013 goto error;
3014 }
3015 err = 0;
3016
3017 error:
3018 while (super_list) {
3019 struct intel_super *s = super_list;
3020
3021 super_list = super_list->next;
3022 free_imsm(s);
3023 }
3024 sysfs_free(sra);
3025
3026 if (err)
3027 return err;
3028
3029 *sbp = super;
3030 st->container_dev = devnum;
3031 if (err == 0 && st->ss == NULL) {
3032 st->ss = &super_imsm;
3033 st->minor_version = 0;
3034 st->max_devs = IMSM_MAX_DEVICES;
3035 }
3036 return 0;
3037 }
3038
3039 static int load_container_imsm(struct supertype *st, int fd, char *devname)
3040 {
3041 return load_super_imsm_all(st, fd, &st->sb, devname);
3042 }
3043 #endif
3044
3045 static int load_super_imsm(struct supertype *st, int fd, char *devname)
3046 {
3047 struct intel_super *super;
3048 int rv;
3049
3050 if (test_partition(fd))
3051 /* IMSM not allowed on partitions */
3052 return 1;
3053
3054 free_super_imsm(st);
3055
3056 super = alloc_super();
3057 if (!super) {
3058 fprintf(stderr,
3059 Name ": malloc of %zu failed.\n",
3060 sizeof(*super));
3061 return 1;
3062 }
3063
3064 rv = load_and_parse_mpb(fd, super, devname, 0);
3065
3066 if (rv) {
3067 if (devname)
3068 fprintf(stderr,
3069 Name ": Failed to load all information "
3070 "sections on %s\n", devname);
3071 free_imsm(super);
3072 return rv;
3073 }
3074
3075 st->sb = super;
3076 if (st->ss == NULL) {
3077 st->ss = &super_imsm;
3078 st->minor_version = 0;
3079 st->max_devs = IMSM_MAX_DEVICES;
3080 }
3081 return 0;
3082 }
3083
3084 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
3085 {
3086 if (info->level == 1)
3087 return 128;
3088 return info->chunk_size >> 9;
3089 }
3090
3091 static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
3092 {
3093 __u32 num_stripes;
3094
3095 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
3096 num_stripes /= num_domains;
3097
3098 return num_stripes;
3099 }
3100
3101 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
3102 {
3103 if (info->level == 1)
3104 return info->size * 2;
3105 else
3106 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
3107 }
3108
3109 static void imsm_update_version_info(struct intel_super *super)
3110 {
3111 /* update the version and attributes */
3112 struct imsm_super *mpb = super->anchor;
3113 char *version;
3114 struct imsm_dev *dev;
3115 struct imsm_map *map;
3116 int i;
3117
3118 for (i = 0; i < mpb->num_raid_devs; i++) {
3119 dev = get_imsm_dev(super, i);
3120 map = get_imsm_map(dev, 0);
3121 if (__le32_to_cpu(dev->size_high) > 0)
3122 mpb->attributes |= MPB_ATTRIB_2TB;
3123
3124 /* FIXME detect when an array spans a port multiplier */
3125 #if 0
3126 mpb->attributes |= MPB_ATTRIB_PM;
3127 #endif
3128
3129 if (mpb->num_raid_devs > 1 ||
3130 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
3131 version = MPB_VERSION_ATTRIBS;
3132 switch (get_imsm_raid_level(map)) {
3133 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
3134 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
3135 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
3136 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
3137 }
3138 } else {
3139 if (map->num_members >= 5)
3140 version = MPB_VERSION_5OR6_DISK_ARRAY;
3141 else if (dev->status == DEV_CLONE_N_GO)
3142 version = MPB_VERSION_CNG;
3143 else if (get_imsm_raid_level(map) == 5)
3144 version = MPB_VERSION_RAID5;
3145 else if (map->num_members >= 3)
3146 version = MPB_VERSION_3OR4_DISK_ARRAY;
3147 else if (get_imsm_raid_level(map) == 1)
3148 version = MPB_VERSION_RAID1;
3149 else
3150 version = MPB_VERSION_RAID0;
3151 }
3152 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
3153 }
3154 }
3155
3156 static int check_name(struct intel_super *super, char *name, int quiet)
3157 {
3158 struct imsm_super *mpb = super->anchor;
3159 char *reason = NULL;
3160 int i;
3161
3162 if (strlen(name) > MAX_RAID_SERIAL_LEN)
3163 reason = "must be 16 characters or less";
3164
3165 for (i = 0; i < mpb->num_raid_devs; i++) {
3166 struct imsm_dev *dev = get_imsm_dev(super, i);
3167
3168 if (strncmp((char *) dev->volume, name, MAX_RAID_SERIAL_LEN) == 0) {
3169 reason = "already exists";
3170 break;
3171 }
3172 }
3173
3174 if (reason && !quiet)
3175 fprintf(stderr, Name ": imsm volume name %s\n", reason);
3176
3177 return !reason;
3178 }
3179
3180 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
3181 unsigned long long size, char *name,
3182 char *homehost, int *uuid)
3183 {
3184 /* We are creating a volume inside a pre-existing container.
3185 * so st->sb is already set.
3186 */
3187 struct intel_super *super = st->sb;
3188 struct imsm_super *mpb = super->anchor;
3189 struct intel_dev *dv;
3190 struct imsm_dev *dev;
3191 struct imsm_vol *vol;
3192 struct imsm_map *map;
3193 int idx = mpb->num_raid_devs;
3194 int i;
3195 unsigned long long array_blocks;
3196 size_t size_old, size_new;
3197 __u32 num_data_stripes;
3198
3199 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
3200 fprintf(stderr, Name": This imsm-container already has the "
3201 "maximum of %d volumes\n", super->orom->vpa);
3202 return 0;
3203 }
3204
3205 /* ensure the mpb is large enough for the new data */
3206 size_old = __le32_to_cpu(mpb->mpb_size);
3207 size_new = disks_to_mpb_size(info->nr_disks);
3208 if (size_new > size_old) {
3209 void *mpb_new;
3210 size_t size_round = ROUND_UP(size_new, 512);
3211
3212 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
3213 fprintf(stderr, Name": could not allocate new mpb\n");
3214 return 0;
3215 }
3216 memcpy(mpb_new, mpb, size_old);
3217 free(mpb);
3218 mpb = mpb_new;
3219 super->anchor = mpb_new;
3220 mpb->mpb_size = __cpu_to_le32(size_new);
3221 memset(mpb_new + size_old, 0, size_round - size_old);
3222 }
3223 super->current_vol = idx;
3224 /* when creating the first raid device in this container set num_disks
3225 * to zero, i.e. delete this spare and add raid member devices in
3226 * add_to_super_imsm_volume()
3227 */
3228 if (super->current_vol == 0)
3229 mpb->num_disks = 0;
3230
3231 if (!check_name(super, name, 0))
3232 return 0;
3233 dv = malloc(sizeof(*dv));
3234 if (!dv) {
3235 fprintf(stderr, Name ": failed to allocate device list entry\n");
3236 return 0;
3237 }
3238 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
3239 if (!dev) {
3240 free(dv);
3241 fprintf(stderr, Name": could not allocate raid device\n");
3242 return 0;
3243 }
3244 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
3245 if (info->level == 1)
3246 array_blocks = info_to_blocks_per_member(info);
3247 else
3248 array_blocks = calc_array_size(info->level, info->raid_disks,
3249 info->layout, info->chunk_size,
3250 info->size*2);
3251 /* round array size down to closest MB */
3252 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
3253
3254 dev->size_low = __cpu_to_le32((__u32) array_blocks);
3255 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
3256 dev->status = __cpu_to_le32(0);
3257 dev->reserved_blocks = __cpu_to_le32(0);
3258 vol = &dev->vol;
3259 vol->migr_state = 0;
3260 set_migr_type(dev, MIGR_INIT);
3261 vol->dirty = 0;
3262 vol->curr_migr_unit = 0;
3263 map = get_imsm_map(dev, 0);
3264 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
3265 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
3266 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
3267 map->failed_disk_num = ~0;
3268 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
3269 IMSM_T_STATE_NORMAL;
3270 map->ddf = 1;
3271
3272 if (info->level == 1 && info->raid_disks > 2) {
3273 free(dev);
3274 free(dv);
3275 fprintf(stderr, Name": imsm does not support more than 2 disks"
3276 "in a raid1 volume\n");
3277 return 0;
3278 }
3279
3280 map->raid_level = info->level;
3281 if (info->level == 10) {
3282 map->raid_level = 1;
3283 map->num_domains = info->raid_disks / 2;
3284 } else if (info->level == 1)
3285 map->num_domains = info->raid_disks;
3286 else
3287 map->num_domains = 1;
3288
3289 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
3290 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
3291
3292 map->num_members = info->raid_disks;
3293 for (i = 0; i < map->num_members; i++) {
3294 /* initialized in add_to_super */
3295 set_imsm_ord_tbl_ent(map, i, IMSM_ORD_REBUILD);
3296 }
3297 mpb->num_raid_devs++;
3298
3299 dv->dev = dev;
3300 dv->index = super->current_vol;
3301 dv->next = super->devlist;
3302 super->devlist = dv;
3303
3304 imsm_update_version_info(super);
3305
3306 return 1;
3307 }
3308
3309 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
3310 unsigned long long size, char *name,
3311 char *homehost, int *uuid)
3312 {
3313 /* This is primarily called by Create when creating a new array.
3314 * We will then get add_to_super called for each component, and then
3315 * write_init_super called to write it out to each device.
3316 * For IMSM, Create can create on fresh devices or on a pre-existing
3317 * array.
3318 * To create on a pre-existing array a different method will be called.
3319 * This one is just for fresh drives.
3320 */
3321 struct intel_super *super;
3322 struct imsm_super *mpb;
3323 size_t mpb_size;
3324 char *version;
3325
3326 if (st->sb)
3327 return init_super_imsm_volume(st, info, size, name, homehost, uuid);
3328
3329 if (info)
3330 mpb_size = disks_to_mpb_size(info->nr_disks);
3331 else
3332 mpb_size = 512;
3333
3334 super = alloc_super();
3335 if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) {
3336 free(super);
3337 super = NULL;
3338 }
3339 if (!super) {
3340 fprintf(stderr, Name
3341 ": %s could not allocate superblock\n", __func__);
3342 return 0;
3343 }
3344 memset(super->buf, 0, mpb_size);
3345 mpb = super->buf;
3346 mpb->mpb_size = __cpu_to_le32(mpb_size);
3347 st->sb = super;
3348
3349 if (info == NULL) {
3350 /* zeroing superblock */
3351 return 0;
3352 }
3353
3354 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
3355
3356 version = (char *) mpb->sig;
3357 strcpy(version, MPB_SIGNATURE);
3358 version += strlen(MPB_SIGNATURE);
3359 strcpy(version, MPB_VERSION_RAID0);
3360
3361 return 1;
3362 }
3363
3364 #ifndef MDASSEMBLE
3365 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
3366 int fd, char *devname)
3367 {
3368 struct intel_super *super = st->sb;
3369 struct imsm_super *mpb = super->anchor;
3370 struct dl *dl;
3371 struct imsm_dev *dev;
3372 struct imsm_map *map;
3373 int slot;
3374
3375 dev = get_imsm_dev(super, super->current_vol);
3376 map = get_imsm_map(dev, 0);
3377
3378 if (! (dk->state & (1<<MD_DISK_SYNC))) {
3379 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
3380 devname);
3381 return 1;
3382 }
3383
3384 if (fd == -1) {
3385 /* we're doing autolayout so grab the pre-marked (in
3386 * validate_geometry) raid_disk
3387 */
3388 for (dl = super->disks; dl; dl = dl->next)
3389 if (dl->raiddisk == dk->raid_disk)
3390 break;
3391 } else {
3392 for (dl = super->disks; dl ; dl = dl->next)
3393 if (dl->major == dk->major &&
3394 dl->minor == dk->minor)
3395 break;
3396 }
3397
3398 if (!dl) {
3399 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
3400 return 1;
3401 }
3402
3403 /* add a pristine spare to the metadata */
3404 if (dl->index < 0) {
3405 dl->index = super->anchor->num_disks;
3406 super->anchor->num_disks++;
3407 }
3408 /* Check the device has not already been added */
3409 slot = get_imsm_disk_slot(map, dl->index);
3410 if (slot >= 0 &&
3411 (get_imsm_ord_tbl_ent(dev, slot, -1) & IMSM_ORD_REBUILD) == 0) {
3412 fprintf(stderr, Name ": %s has been included in this array twice\n",
3413 devname);
3414 return 1;
3415 }
3416 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
3417 dl->disk.status = CONFIGURED_DISK;
3418
3419 /* if we are creating the first raid device update the family number */
3420 if (super->current_vol == 0) {
3421 __u32 sum;
3422 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
3423 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
3424
3425 if (!_dev || !_disk) {
3426 fprintf(stderr, Name ": BUG mpb setup error\n");
3427 return 1;
3428 }
3429 *_dev = *dev;
3430 *_disk = dl->disk;
3431 sum = random32();
3432 sum += __gen_imsm_checksum(mpb);
3433 mpb->family_num = __cpu_to_le32(sum);
3434 mpb->orig_family_num = mpb->family_num;
3435 }
3436
3437 return 0;
3438 }
3439
3440 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
3441 int fd, char *devname)
3442 {
3443 struct intel_super *super = st->sb;
3444 struct dl *dd;
3445 unsigned long long size;
3446 __u32 id;
3447 int rv;
3448 struct stat stb;
3449
3450 /* if we are on an RAID enabled platform check that the disk is
3451 * attached to the raid controller
3452 */
3453 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
3454 fprintf(stderr,
3455 Name ": %s is not attached to the raid controller: %s\n",
3456 devname ? : "disk", super->hba);
3457 return 1;
3458 }
3459
3460 if (super->current_vol >= 0)
3461 return add_to_super_imsm_volume(st, dk, fd, devname);
3462
3463 fstat(fd, &stb);
3464 dd = malloc(sizeof(*dd));
3465 if (!dd) {
3466 fprintf(stderr,
3467 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
3468 return 1;
3469 }
3470 memset(dd, 0, sizeof(*dd));
3471 dd->major = major(stb.st_rdev);
3472 dd->minor = minor(stb.st_rdev);
3473 dd->index = -1;
3474 dd->devname = devname ? strdup(devname) : NULL;
3475 dd->fd = fd;
3476 dd->e = NULL;
3477 dd->action = DISK_ADD;
3478 rv = imsm_read_serial(fd, devname, dd->serial);
3479 if (rv) {
3480 fprintf(stderr,
3481 Name ": failed to retrieve scsi serial, aborting\n");
3482 free(dd);
3483 abort();
3484 }
3485
3486 get_dev_size(fd, NULL, &size);
3487 size /= 512;
3488 serialcpy(dd->disk.serial, dd->serial);
3489 dd->disk.total_blocks = __cpu_to_le32(size);
3490 dd->disk.status = SPARE_DISK;
3491 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
3492 dd->disk.scsi_id = __cpu_to_le32(id);
3493 else
3494 dd->disk.scsi_id = __cpu_to_le32(0);
3495
3496 if (st->update_tail) {
3497 dd->next = super->disk_mgmt_list;
3498 super->disk_mgmt_list = dd;
3499 } else {
3500 dd->next = super->disks;
3501 super->disks = dd;
3502 }
3503
3504 return 0;
3505 }
3506
3507
3508 static int remove_from_super_imsm(struct supertype *st, mdu_disk_info_t *dk)
3509 {
3510 struct intel_super *super = st->sb;
3511 struct dl *dd;
3512
3513 /* remove from super works only in mdmon - for communication
3514 * manager - monitor. Check if communication memory buffer
3515 * is prepared.
3516 */
3517 if (!st->update_tail) {
3518 fprintf(stderr,
3519 Name ": %s shall be used in mdmon context only"
3520 "(line %d).\n", __func__, __LINE__);
3521 return 1;
3522 }
3523 dd = malloc(sizeof(*dd));
3524 if (!dd) {
3525 fprintf(stderr,
3526 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
3527 return 1;
3528 }
3529 memset(dd, 0, sizeof(*dd));
3530 dd->major = dk->major;
3531 dd->minor = dk->minor;
3532 dd->index = -1;
3533 dd->fd = -1;
3534 dd->disk.status = SPARE_DISK;
3535 dd->action = DISK_REMOVE;
3536
3537 dd->next = super->disk_mgmt_list;
3538 super->disk_mgmt_list = dd;
3539
3540
3541 return 0;
3542 }
3543
3544 static int store_imsm_mpb(int fd, struct imsm_super *mpb);
3545
3546 static union {
3547 char buf[512];
3548 struct imsm_super anchor;
3549 } spare_record __attribute__ ((aligned(512)));
3550
3551 /* spare records have their own family number and do not have any defined raid
3552 * devices
3553 */
3554 static int write_super_imsm_spares(struct intel_super *super, int doclose)
3555 {
3556 struct imsm_super *mpb = super->anchor;
3557 struct imsm_super *spare = &spare_record.anchor;
3558 __u32 sum;
3559 struct dl *d;
3560
3561 spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)),
3562 spare->generation_num = __cpu_to_le32(1UL),
3563 spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
3564 spare->num_disks = 1,
3565 spare->num_raid_devs = 0,
3566 spare->cache_size = mpb->cache_size,
3567 spare->pwr_cycle_count = __cpu_to_le32(1),
3568
3569 snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
3570 MPB_SIGNATURE MPB_VERSION_RAID0);
3571
3572 for (d = super->disks; d; d = d->next) {
3573 if (d->index != -1)
3574 continue;
3575
3576 spare->disk[0] = d->disk;
3577 sum = __gen_imsm_checksum(spare);
3578 spare->family_num = __cpu_to_le32(sum);
3579 spare->orig_family_num = 0;
3580 sum = __gen_imsm_checksum(spare);
3581 spare->check_sum = __cpu_to_le32(sum);
3582
3583 if (store_imsm_mpb(d->fd, spare)) {
3584 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3585 __func__, d->major, d->minor, strerror(errno));
3586 return 1;
3587 }
3588 if (doclose) {
3589 close(d->fd);
3590 d->fd = -1;
3591 }
3592 }
3593
3594 return 0;
3595 }
3596
3597 static int write_super_imsm(struct supertype *st, int doclose)
3598 {
3599 struct intel_super *super = st->sb;
3600 struct imsm_super *mpb = super->anchor;
3601 struct dl *d;
3602 __u32 generation;
3603 __u32 sum;
3604 int spares = 0;
3605 int i;
3606 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
3607 int num_disks = 0;
3608
3609 /* 'generation' is incremented everytime the metadata is written */
3610 generation = __le32_to_cpu(mpb->generation_num);
3611 generation++;
3612 mpb->generation_num = __cpu_to_le32(generation);
3613
3614 /* fix up cases where previous mdadm releases failed to set
3615 * orig_family_num
3616 */
3617 if (mpb->orig_family_num == 0)
3618 mpb->orig_family_num = mpb->family_num;
3619
3620 for (d = super->disks; d; d = d->next) {
3621 if (d->index == -1)
3622 spares++;
3623 else {
3624 mpb->disk[d->index] = d->disk;
3625 num_disks++;
3626 }
3627 }
3628 for (d = super->missing; d; d = d->next) {
3629 mpb->disk[d->index] = d->disk;
3630 num_disks++;
3631 }
3632 mpb->num_disks = num_disks;
3633 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
3634
3635 for (i = 0; i < mpb->num_raid_devs; i++) {
3636 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
3637 struct imsm_dev *dev2 = get_imsm_dev(super, i);
3638 if (dev && dev2) {
3639 imsm_copy_dev(dev, dev2);
3640 mpb_size += sizeof_imsm_dev(dev, 0);
3641 }
3642 }
3643 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
3644 mpb->mpb_size = __cpu_to_le32(mpb_size);
3645
3646 /* recalculate checksum */
3647 sum = __gen_imsm_checksum(mpb);
3648 mpb->check_sum = __cpu_to_le32(sum);
3649
3650 /* write the mpb for disks that compose raid devices */
3651 for (d = super->disks; d ; d = d->next) {
3652 if (d->index < 0)
3653 continue;
3654 if (store_imsm_mpb(d->fd, mpb))
3655 fprintf(stderr, "%s: failed for device %d:%d %s\n",
3656 __func__, d->major, d->minor, strerror(errno));
3657 if (doclose) {
3658 close(d->fd);
3659 d->fd = -1;
3660 }
3661 }
3662
3663 if (spares)
3664 return write_super_imsm_spares(super, doclose);
3665
3666 return 0;
3667 }
3668
3669
3670 static int create_array(struct supertype *st, int dev_idx)
3671 {
3672 size_t len;
3673 struct imsm_update_create_array *u;
3674 struct intel_super *super = st->sb;
3675 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
3676 struct imsm_map *map = get_imsm_map(dev, 0);
3677 struct disk_info *inf;
3678 struct imsm_disk *disk;
3679 int i;
3680
3681 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
3682 sizeof(*inf) * map->num_members;
3683 u = malloc(len);
3684 if (!u) {
3685 fprintf(stderr, "%s: failed to allocate update buffer\n",
3686 __func__);
3687 return 1;
3688 }
3689
3690 u->type = update_create_array;
3691 u->dev_idx = dev_idx;
3692 imsm_copy_dev(&u->dev, dev);
3693 inf = get_disk_info(u);
3694 for (i = 0; i < map->num_members; i++) {
3695 int idx = get_imsm_disk_idx(dev, i, -1);
3696
3697 disk = get_imsm_disk(super, idx);
3698 serialcpy(inf[i].serial, disk->serial);
3699 }
3700 append_metadata_update(st, u, len);
3701
3702 return 0;
3703 }
3704
3705 static int mgmt_disk(struct supertype *st)
3706 {
3707 struct intel_super *super = st->sb;
3708 size_t len;
3709 struct imsm_update_add_remove_disk *u;
3710
3711 if (!super->disk_mgmt_list)
3712 return 0;
3713
3714 len = sizeof(*u);
3715 u = malloc(len);
3716 if (!u) {
3717 fprintf(stderr, "%s: failed to allocate update buffer\n",
3718 __func__);
3719 return 1;
3720 }
3721
3722 u->type = update_add_remove_disk;
3723 append_metadata_update(st, u, len);
3724
3725 return 0;
3726 }
3727
3728 static int write_init_super_imsm(struct supertype *st)
3729 {
3730 struct intel_super *super = st->sb;
3731 int current_vol = super->current_vol;
3732
3733 /* we are done with current_vol reset it to point st at the container */
3734 super->current_vol = -1;
3735
3736 if (st->update_tail) {
3737 /* queue the recently created array / added disk
3738 * as a metadata update */
3739 struct dl *d;
3740 int rv;
3741
3742 /* determine if we are creating a volume or adding a disk */
3743 if (current_vol < 0) {
3744 /* in the mgmt (add/remove) disk case we are running
3745 * in mdmon context, so don't close fd's
3746 */
3747 return mgmt_disk(st);
3748 } else
3749 rv = create_array(st, current_vol);
3750
3751 for (d = super->disks; d ; d = d->next) {
3752 close(d->fd);
3753 d->fd = -1;
3754 }
3755
3756 return rv;
3757 } else {
3758 struct dl *d;
3759 for (d = super->disks; d; d = d->next)
3760 Kill(d->devname, NULL, 0, 1, 1);
3761 return write_super_imsm(st, 1);
3762 }
3763 }
3764 #endif
3765
3766 static int store_super_imsm(struct supertype *st, int fd)
3767 {
3768 struct intel_super *super = st->sb;
3769 struct imsm_super *mpb = super ? super->anchor : NULL;
3770
3771 if (!mpb)
3772 return 1;
3773
3774 #ifndef MDASSEMBLE
3775 return store_imsm_mpb(fd, mpb);
3776 #else
3777 return 1;
3778 #endif
3779 }
3780
3781 static int imsm_bbm_log_size(struct imsm_super *mpb)
3782 {
3783 return __le32_to_cpu(mpb->bbm_log_size);
3784 }
3785
3786 #ifndef MDASSEMBLE
3787 static int validate_geometry_imsm_container(struct supertype *st, int level,
3788 int layout, int raiddisks, int chunk,
3789 unsigned long long size, char *dev,
3790 unsigned long long *freesize,
3791 int verbose)
3792 {
3793 int fd;
3794 unsigned long long ldsize;
3795 const struct imsm_orom *orom;
3796
3797 if (level != LEVEL_CONTAINER)
3798 return 0;
3799 if (!dev)
3800 return 1;
3801
3802 if (check_env("IMSM_NO_PLATFORM"))
3803 orom = NULL;
3804 else
3805 orom = find_imsm_orom();
3806 if (orom && raiddisks > orom->tds) {
3807 if (verbose)
3808 fprintf(stderr, Name ": %d exceeds maximum number of"
3809 " platform supported disks: %d\n",
3810 raiddisks, orom->tds);
3811 return 0;
3812 }
3813
3814 fd = open(dev, O_RDONLY|O_EXCL, 0);
3815 if (fd < 0) {
3816 if (verbose)
3817 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
3818 dev, strerror(errno));
3819 return 0;
3820 }
3821 if (!get_dev_size(fd, dev, &ldsize)) {
3822 close(fd);
3823 return 0;
3824 }
3825 close(fd);
3826
3827 *freesize = avail_size_imsm(st, ldsize >> 9);
3828
3829 return 1;
3830 }
3831
3832 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
3833 {
3834 const unsigned long long base_start = e[*idx].start;
3835 unsigned long long end = base_start + e[*idx].size;
3836 int i;
3837
3838 if (base_start == end)
3839 return 0;
3840
3841 *idx = *idx + 1;
3842 for (i = *idx; i < num_extents; i++) {
3843 /* extend overlapping extents */
3844 if (e[i].start >= base_start &&
3845 e[i].start <= end) {
3846 if (e[i].size == 0)
3847 return 0;
3848 if (e[i].start + e[i].size > end)
3849 end = e[i].start + e[i].size;
3850 } else if (e[i].start > end) {
3851 *idx = i;
3852 break;
3853 }
3854 }
3855
3856 return end - base_start;
3857 }
3858
3859 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
3860 {
3861 /* build a composite disk with all known extents and generate a new
3862 * 'maxsize' given the "all disks in an array must share a common start
3863 * offset" constraint
3864 */
3865 struct extent *e = calloc(sum_extents, sizeof(*e));
3866 struct dl *dl;
3867 int i, j;
3868 int start_extent;
3869 unsigned long long pos;
3870 unsigned long long start = 0;
3871 unsigned long long maxsize;
3872 unsigned long reserve;
3873
3874 if (!e)
3875 return 0;
3876
3877 /* coalesce and sort all extents. also, check to see if we need to
3878 * reserve space between member arrays
3879 */
3880 j = 0;
3881 for (dl = super->disks; dl; dl = dl->next) {
3882 if (!dl->e)
3883 continue;
3884 for (i = 0; i < dl->extent_cnt; i++)
3885 e[j++] = dl->e[i];
3886 }
3887 qsort(e, sum_extents, sizeof(*e), cmp_extent);
3888
3889 /* merge extents */
3890 i = 0;
3891 j = 0;
3892 while (i < sum_extents) {
3893 e[j].start = e[i].start;
3894 e[j].size = find_size(e, &i, sum_extents);
3895 j++;
3896 if (e[j-1].size == 0)
3897 break;
3898 }
3899
3900 pos = 0;
3901 maxsize = 0;
3902 start_extent = 0;
3903 i = 0;
3904 do {
3905 unsigned long long esize;
3906
3907 esize = e[i].start - pos;
3908 if (esize >= maxsize) {
3909 maxsize = esize;
3910 start = pos;
3911 start_extent = i;
3912 }
3913 pos = e[i].start + e[i].size;
3914 i++;
3915 } while (e[i-1].size);
3916 free(e);
3917
3918 if (maxsize == 0)
3919 return 0;
3920
3921 /* FIXME assumes volume at offset 0 is the first volume in a
3922 * container
3923 */
3924 if (start_extent > 0)
3925 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
3926 else
3927 reserve = 0;
3928
3929 if (maxsize < reserve)
3930 return 0;
3931
3932 super->create_offset = ~((__u32) 0);
3933 if (start + reserve > super->create_offset)
3934 return 0; /* start overflows create_offset */
3935 super->create_offset = start + reserve;
3936
3937 return maxsize - reserve;
3938 }
3939
3940 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
3941 {
3942 if (level < 0 || level == 6 || level == 4)
3943 return 0;
3944
3945 /* if we have an orom prevent invalid raid levels */
3946 if (orom)
3947 switch (level) {
3948 case 0: return imsm_orom_has_raid0(orom);
3949 case 1:
3950 if (raiddisks > 2)
3951 return imsm_orom_has_raid1e(orom);
3952 return imsm_orom_has_raid1(orom) && raiddisks == 2;
3953 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
3954 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
3955 }
3956 else
3957 return 1; /* not on an Intel RAID platform so anything goes */
3958
3959 return 0;
3960 }
3961
3962 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
3963 static int
3964 validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
3965 int raiddisks, int chunk, int verbose)
3966 {
3967 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
3968 pr_vrb(": platform does not support raid%d with %d disk%s\n",
3969 level, raiddisks, raiddisks > 1 ? "s" : "");
3970 return 0;
3971 }
3972 if (super->orom && level != 1 &&
3973 !imsm_orom_has_chunk(super->orom, chunk)) {
3974 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
3975 return 0;
3976 }
3977 if (layout != imsm_level_to_layout(level)) {
3978 if (level == 5)
3979 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
3980 else if (level == 10)
3981 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
3982 else
3983 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
3984 layout, level);
3985 return 0;
3986 }
3987
3988 return 1;
3989 }
3990
3991 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
3992 * FIX ME add ahci details
3993 */
3994 static int validate_geometry_imsm_volume(struct supertype *st, int level,
3995 int layout, int raiddisks, int chunk,
3996 unsigned long long size, char *dev,
3997 unsigned long long *freesize,
3998 int verbose)
3999 {
4000 struct stat stb;
4001 struct intel_super *super = st->sb;
4002 struct imsm_super *mpb = super->anchor;
4003 struct dl *dl;
4004 unsigned long long pos = 0;
4005 unsigned long long maxsize;
4006 struct extent *e;
4007 int i;
4008
4009 /* We must have the container info already read in. */
4010 if (!super)
4011 return 0;
4012
4013 if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose))
4014 return 0;
4015
4016 if (!dev) {
4017 /* General test: make sure there is space for
4018 * 'raiddisks' device extents of size 'size' at a given
4019 * offset
4020 */
4021 unsigned long long minsize = size;
4022 unsigned long long start_offset = MaxSector;
4023 int dcnt = 0;
4024 if (minsize == 0)
4025 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
4026 for (dl = super->disks; dl ; dl = dl->next) {
4027 int found = 0;
4028
4029 pos = 0;
4030 i = 0;
4031 e = get_extents(super, dl);
4032 if (!e) continue;
4033 do {
4034 unsigned long long esize;
4035 esize = e[i].start - pos;
4036 if (esize >= minsize)
4037 found = 1;
4038 if (found && start_offset == MaxSector) {
4039 start_offset = pos;
4040 break;
4041 } else if (found && pos != start_offset) {
4042 found = 0;
4043 break;
4044 }
4045 pos = e[i].start + e[i].size;
4046 i++;
4047 } while (e[i-1].size);
4048 if (found)
4049 dcnt++;
4050 free(e);
4051 }
4052 if (dcnt < raiddisks) {
4053 if (verbose)
4054 fprintf(stderr, Name ": imsm: Not enough "
4055 "devices with space for this array "
4056 "(%d < %d)\n",
4057 dcnt, raiddisks);
4058 return 0;
4059 }
4060 return 1;
4061 }
4062
4063 /* This device must be a member of the set */
4064 if (stat(dev, &stb) < 0)
4065 return 0;
4066 if ((S_IFMT & stb.st_mode) != S_IFBLK)
4067 return 0;
4068 for (dl = super->disks ; dl ; dl = dl->next) {
4069 if (dl->major == (int)major(stb.st_rdev) &&
4070 dl->minor == (int)minor(stb.st_rdev))
4071 break;
4072 }
4073 if (!dl) {
4074 if (verbose)
4075 fprintf(stderr, Name ": %s is not in the "
4076 "same imsm set\n", dev);
4077 return 0;
4078 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
4079 /* If a volume is present then the current creation attempt
4080 * cannot incorporate new spares because the orom may not
4081 * understand this configuration (all member disks must be
4082 * members of each array in the container).
4083 */
4084 fprintf(stderr, Name ": %s is a spare and a volume"
4085 " is already defined for this container\n", dev);
4086 fprintf(stderr, Name ": The option-rom requires all member"
4087 " disks to be a member of all volumes\n");
4088 return 0;
4089 }
4090
4091 /* retrieve the largest free space block */
4092 e = get_extents(super, dl);
4093 maxsize = 0;
4094 i = 0;
4095 if (e) {
4096 do {
4097 unsigned long long esize;
4098
4099 esize = e[i].start - pos;
4100 if (esize >= maxsize)
4101 maxsize = esize;
4102 pos = e[i].start + e[i].size;
4103 i++;
4104 } while (e[i-1].size);
4105 dl->e = e;
4106 dl->extent_cnt = i;
4107 } else {
4108 if (verbose)
4109 fprintf(stderr, Name ": unable to determine free space for: %s\n",
4110 dev);
4111 return 0;
4112 }
4113 if (maxsize < size) {
4114 if (verbose)
4115 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
4116 dev, maxsize, size);
4117 return 0;
4118 }
4119
4120 /* count total number of extents for merge */
4121 i = 0;
4122 for (dl = super->disks; dl; dl = dl->next)
4123 if (dl->e)
4124 i += dl->extent_cnt;
4125
4126 maxsize = merge_extents(super, i);
4127 if (maxsize < size || maxsize == 0) {
4128 if (verbose)
4129 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
4130 maxsize, size);
4131 return 0;
4132 }
4133
4134 *freesize = maxsize;
4135
4136 return 1;
4137 }
4138
4139 static int reserve_space(struct supertype *st, int raiddisks,
4140 unsigned long long size, int chunk,
4141 unsigned long long *freesize)
4142 {
4143 struct intel_super *super = st->sb;
4144 struct imsm_super *mpb = super->anchor;
4145 struct dl *dl;
4146 int i;
4147 int extent_cnt;
4148 struct extent *e;
4149 unsigned long long maxsize;
4150 unsigned long long minsize;
4151 int cnt;
4152 int used;
4153
4154 /* find the largest common start free region of the possible disks */
4155 used = 0;
4156 extent_cnt = 0;
4157 cnt = 0;
4158 for (dl = super->disks; dl; dl = dl->next) {
4159 dl->raiddisk = -1;
4160
4161 if (dl->index >= 0)
4162 used++;
4163
4164 /* don't activate new spares if we are orom constrained
4165 * and there is already a volume active in the container
4166 */
4167 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
4168 continue;
4169
4170 e = get_extents(super, dl);
4171 if (!e)
4172 continue;
4173 for (i = 1; e[i-1].size; i++)
4174 ;
4175 dl->e = e;
4176 dl->extent_cnt = i;
4177 extent_cnt += i;
4178 cnt++;
4179 }
4180
4181 maxsize = merge_extents(super, extent_cnt);
4182 minsize = size;
4183 if (size == 0)
4184 minsize = chunk;
4185
4186 if (cnt < raiddisks ||
4187 (super->orom && used && used != raiddisks) ||
4188 maxsize < minsize ||
4189 maxsize == 0) {
4190 fprintf(stderr, Name ": not enough devices with space to create array.\n");
4191 return 0; /* No enough free spaces large enough */
4192 }
4193
4194 if (size == 0) {
4195 size = maxsize;
4196 if (chunk) {
4197 size /= chunk;
4198 size *= chunk;
4199 }
4200 }
4201
4202 cnt = 0;
4203 for (dl = super->disks; dl; dl = dl->next)
4204 if (dl->e)
4205 dl->raiddisk = cnt++;
4206
4207 *freesize = size;
4208
4209 return 1;
4210 }
4211
4212 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
4213 int raiddisks, int chunk, unsigned long long size,
4214 char *dev, unsigned long long *freesize,
4215 int verbose)
4216 {
4217 int fd, cfd;
4218 struct mdinfo *sra;
4219 int is_member = 0;
4220
4221 /* if given unused devices create a container
4222 * if given given devices in a container create a member volume
4223 */
4224 if (level == LEVEL_CONTAINER) {
4225 /* Must be a fresh device to add to a container */
4226 return validate_geometry_imsm_container(st, level, layout,
4227 raiddisks, chunk, size,
4228 dev, freesize,
4229 verbose);
4230 }
4231
4232 if (!dev) {
4233 if (st->sb && freesize) {
4234 /* we are being asked to automatically layout a
4235 * new volume based on the current contents of
4236 * the container. If the the parameters can be
4237 * satisfied reserve_space will record the disks,
4238 * start offset, and size of the volume to be
4239 * created. add_to_super and getinfo_super
4240 * detect when autolayout is in progress.
4241 */
4242 if (!validate_geometry_imsm_orom(st->sb, level, layout,
4243 raiddisks, chunk,
4244 verbose))
4245 return 0;
4246 return reserve_space(st, raiddisks, size, chunk, freesize);
4247 }
4248 return 1;
4249 }
4250 if (st->sb) {
4251 /* creating in a given container */
4252 return validate_geometry_imsm_volume(st, level, layout,
4253 raiddisks, chunk, size,
4254 dev, freesize, verbose);
4255 }
4256
4257 /* This device needs to be a device in an 'imsm' container */
4258 fd = open(dev, O_RDONLY|O_EXCL, 0);
4259 if (fd >= 0) {
4260 if (verbose)
4261 fprintf(stderr,
4262 Name ": Cannot create this array on device %s\n",
4263 dev);
4264 close(fd);
4265 return 0;
4266 }
4267 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
4268 if (verbose)
4269 fprintf(stderr, Name ": Cannot open %s: %s\n",
4270 dev, strerror(errno));
4271 return 0;
4272 }
4273 /* Well, it is in use by someone, maybe an 'imsm' container. */
4274 cfd = open_container(fd);
4275 close(fd);
4276 if (cfd < 0) {
4277 if (verbose)
4278 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
4279 dev);
4280 return 0;
4281 }
4282 sra = sysfs_read(cfd, 0, GET_VERSION);
4283 if (sra && sra->array.major_version == -1 &&
4284 strcmp(sra->text_version, "imsm") == 0)
4285 is_member = 1;
4286 sysfs_free(sra);
4287 if (is_member) {
4288 /* This is a member of a imsm container. Load the container
4289 * and try to create a volume
4290 */
4291 struct intel_super *super;
4292
4293 if (load_super_imsm_all(st, cfd, (void **) &super, NULL) == 0) {
4294 st->sb = super;
4295 st->container_dev = fd2devnum(cfd);
4296 close(cfd);
4297 return validate_geometry_imsm_volume(st, level, layout,
4298 raiddisks, chunk,
4299 size, dev,
4300 freesize, verbose);
4301 }
4302 }
4303
4304 if (verbose)
4305 fprintf(stderr, Name ": failed container membership check\n");
4306
4307 close(cfd);
4308 return 0;
4309 }
4310
4311 static void default_geometry_imsm(struct supertype *st, int *level, int *layout, int *chunk)
4312 {
4313 struct intel_super *super = st->sb;
4314
4315 if (level && *level == UnSet)
4316 *level = LEVEL_CONTAINER;
4317
4318 if (level && layout && *layout == UnSet)
4319 *layout = imsm_level_to_layout(*level);
4320
4321 if (chunk && (*chunk == UnSet || *chunk == 0) &&
4322 super && super->orom)
4323 *chunk = imsm_orom_default_chunk(super->orom);
4324 }
4325
4326 static void handle_missing(struct intel_super *super, struct imsm_dev *dev);
4327
4328 static int kill_subarray_imsm(struct supertype *st)
4329 {
4330 /* remove the subarray currently referenced by ->current_vol */
4331 __u8 i;
4332 struct intel_dev **dp;
4333 struct intel_super *super = st->sb;
4334 __u8 current_vol = super->current_vol;
4335 struct imsm_super *mpb = super->anchor;
4336
4337 if (super->current_vol < 0)
4338 return 2;
4339 super->current_vol = -1; /* invalidate subarray cursor */
4340
4341 /* block deletions that would change the uuid of active subarrays
4342 *
4343 * FIXME when immutable ids are available, but note that we'll
4344 * also need to fixup the invalidated/active subarray indexes in
4345 * mdstat
4346 */
4347 for (i = 0; i < mpb->num_raid_devs; i++) {
4348 char subarray[4];
4349
4350 if (i < current_vol)
4351 continue;
4352 sprintf(subarray, "%u", i);
4353 if (is_subarray_active(subarray, st->devname)) {
4354 fprintf(stderr,
4355 Name ": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
4356 current_vol, i);
4357
4358 return 2;
4359 }
4360 }
4361
4362 if (st->update_tail) {
4363 struct imsm_update_kill_array *u = malloc(sizeof(*u));
4364
4365 if (!u)
4366 return 2;
4367 u->type = update_kill_array;
4368 u->dev_idx = current_vol;
4369 append_metadata_update(st, u, sizeof(*u));
4370
4371 return 0;
4372 }
4373
4374 for (dp = &super->devlist; *dp;)
4375 if ((*dp)->index == current_vol) {
4376 *dp = (*dp)->next;
4377 } else {
4378 handle_missing(super, (*dp)->dev);
4379 if ((*dp)->index > current_vol)
4380 (*dp)->index--;
4381 dp = &(*dp)->next;
4382 }
4383
4384 /* no more raid devices, all active components are now spares,
4385 * but of course failed are still failed
4386 */
4387 if (--mpb->num_raid_devs == 0) {
4388 struct dl *d;
4389
4390 for (d = super->disks; d; d = d->next)
4391 if (d->index > -2) {
4392 d->index = -1;
4393 d->disk.status = SPARE_DISK;
4394 }
4395 }
4396
4397 super->updates_pending++;
4398
4399 return 0;
4400 }
4401
4402 static int update_subarray_imsm(struct supertype *st, char *subarray,
4403 char *update, struct mddev_ident *ident)
4404 {
4405 /* update the subarray currently referenced by ->current_vol */
4406 struct intel_super *super = st->sb;
4407 struct imsm_super *mpb = super->anchor;
4408
4409 if (strcmp(update, "name") == 0) {
4410 char *name = ident->name;
4411 char *ep;
4412 int vol;
4413
4414 if (is_subarray_active(subarray, st->devname)) {
4415 fprintf(stderr,
4416 Name ": Unable to update name of active subarray\n");
4417 return 2;
4418 }
4419
4420 if (!check_name(super, name, 0))
4421 return 2;
4422
4423 vol = strtoul(subarray, &ep, 10);
4424 if (*ep != '\0' || vol >= super->anchor->num_raid_devs)
4425 return 2;
4426
4427 if (st->update_tail) {
4428 struct imsm_update_rename_array *u = malloc(sizeof(*u));
4429
4430 if (!u)
4431 return 2;
4432 u->type = update_rename_array;
4433 u->dev_idx = vol;
4434 snprintf((char *) u->name, MAX_RAID_SERIAL_LEN, "%s", name);
4435 append_metadata_update(st, u, sizeof(*u));
4436 } else {
4437 struct imsm_dev *dev;
4438 int i;
4439
4440 dev = get_imsm_dev(super, vol);
4441 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
4442 for (i = 0; i < mpb->num_raid_devs; i++) {
4443 dev = get_imsm_dev(super, i);
4444 handle_missing(super, dev);
4445 }
4446 super->updates_pending++;
4447 }
4448 } else
4449 return 2;
4450
4451 return 0;
4452 }
4453 #endif /* MDASSEMBLE */
4454
4455 static int is_gen_migration(struct imsm_dev *dev)
4456 {
4457 if (!dev->vol.migr_state)
4458 return 0;
4459
4460 if (migr_type(dev) == MIGR_GEN_MIGR)
4461 return 1;
4462
4463 return 0;
4464 }
4465
4466 static int is_rebuilding(struct imsm_dev *dev)
4467 {
4468 struct imsm_map *migr_map;
4469
4470 if (!dev->vol.migr_state)
4471 return 0;
4472
4473 if (migr_type(dev) != MIGR_REBUILD)
4474 return 0;
4475
4476 migr_map = get_imsm_map(dev, 1);
4477
4478 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
4479 return 1;
4480 else
4481 return 0;
4482 }
4483
4484 static void update_recovery_start(struct imsm_dev *dev, struct mdinfo *array)
4485 {
4486 struct mdinfo *rebuild = NULL;
4487 struct mdinfo *d;
4488 __u32 units;
4489
4490 if (!is_rebuilding(dev))
4491 return;
4492
4493 /* Find the rebuild target, but punt on the dual rebuild case */
4494 for (d = array->devs; d; d = d->next)
4495 if (d->recovery_start == 0) {
4496 if (rebuild)
4497 return;
4498 rebuild = d;
4499 }
4500
4501 if (!rebuild) {
4502 /* (?) none of the disks are marked with
4503 * IMSM_ORD_REBUILD, so assume they are missing and the
4504 * disk_ord_tbl was not correctly updated
4505 */
4506 dprintf("%s: failed to locate out-of-sync disk\n", __func__);
4507 return;
4508 }
4509
4510 units = __le32_to_cpu(dev->vol.curr_migr_unit);
4511 rebuild->recovery_start = units * blocks_per_migr_unit(dev);
4512 }
4513
4514
4515 static struct mdinfo *container_content_imsm(struct supertype *st, char *subarray)
4516 {
4517 /* Given a container loaded by load_super_imsm_all,
4518 * extract information about all the arrays into
4519 * an mdinfo tree.
4520 * If 'subarray' is given, just extract info about that array.
4521 *
4522 * For each imsm_dev create an mdinfo, fill it in,
4523 * then look for matching devices in super->disks
4524 * and create appropriate device mdinfo.
4525 */
4526 struct intel_super *super = st->sb;
4527 struct imsm_super *mpb = super->anchor;
4528 struct mdinfo *rest = NULL;
4529 unsigned int i;
4530 int bbm_errors = 0;
4531
4532 /* check for bad blocks */
4533 if (imsm_bbm_log_size(super->anchor))
4534 bbm_errors = 1;
4535
4536 for (i = 0; i < mpb->num_raid_devs; i++) {
4537 struct imsm_dev *dev;
4538 struct imsm_map *map;
4539 struct imsm_map *map2;
4540 struct mdinfo *this;
4541 int slot;
4542 char *ep;
4543
4544 if (subarray &&
4545 (i != strtoul(subarray, &ep, 10) || *ep != '\0'))
4546 continue;
4547
4548 dev = get_imsm_dev(super, i);
4549 map = get_imsm_map(dev, 0);
4550 map2 = get_imsm_map(dev, 1);
4551
4552 /* do not publish arrays that are in the middle of an
4553 * unsupported migration
4554 */
4555 if (dev->vol.migr_state &&
4556 (migr_type(dev) == MIGR_STATE_CHANGE)) {
4557 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
4558 " unsupported migration in progress\n",
4559 dev->volume);
4560 continue;
4561 }
4562
4563 this = malloc(sizeof(*this));
4564 if (!this) {
4565 fprintf(stderr, Name ": failed to allocate %zu bytes\n",
4566 sizeof(*this));
4567 break;
4568 }
4569 memset(this, 0, sizeof(*this));
4570 this->next = rest;
4571
4572 super->current_vol = i;
4573 getinfo_super_imsm_volume(st, this, NULL);
4574 for (slot = 0 ; slot < map->num_members; slot++) {
4575 unsigned long long recovery_start;
4576 struct mdinfo *info_d;
4577 struct dl *d;
4578 int idx;
4579 int skip;
4580 __u32 ord;
4581
4582 skip = 0;
4583 idx = get_imsm_disk_idx(dev, slot, 0);
4584 ord = get_imsm_ord_tbl_ent(dev, slot, 0);
4585 for (d = super->disks; d ; d = d->next)
4586 if (d->index == idx)
4587 break;
4588
4589 recovery_start = MaxSector;
4590 if (d == NULL)
4591 skip = 1;
4592 if (d && is_failed(&d->disk))
4593 skip = 1;
4594 if (ord & IMSM_ORD_REBUILD)
4595 recovery_start = 0;
4596
4597 /*
4598 * if we skip some disks the array will be assmebled degraded;
4599 * reset resync start to avoid a dirty-degraded
4600 * situation when performing the intial sync
4601 *
4602 * FIXME handle dirty degraded
4603 */
4604 if ((skip || recovery_start == 0) && !dev->vol.dirty)
4605 this->resync_start = MaxSector;
4606 if (skip)
4607 continue;
4608
4609 info_d = calloc(1, sizeof(*info_d));
4610 if (!info_d) {
4611 fprintf(stderr, Name ": failed to allocate disk"
4612 " for volume %.16s\n", dev->volume);
4613 info_d = this->devs;
4614 while (info_d) {
4615 struct mdinfo *d = info_d->next;
4616
4617 free(info_d);
4618 info_d = d;
4619 }
4620 free(this);
4621 this = rest;
4622 break;
4623 }
4624 info_d->next = this->devs;
4625 this->devs = info_d;
4626
4627 info_d->disk.number = d->index;
4628 info_d->disk.major = d->major;
4629 info_d->disk.minor = d->minor;
4630 info_d->disk.raid_disk = slot;
4631 info_d->recovery_start = recovery_start;
4632 if (map2) {
4633 if (slot < map2->num_members)
4634 info_d->disk.state = (1 << MD_DISK_ACTIVE);
4635 } else {
4636 if (slot < map->num_members)
4637 info_d->disk.state = (1 << MD_DISK_ACTIVE);
4638 }
4639 if (info_d->recovery_start == MaxSector)
4640 this->array.working_disks++;
4641
4642 info_d->events = __le32_to_cpu(mpb->generation_num);
4643 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
4644 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
4645 }
4646 /* now that the disk list is up-to-date fixup recovery_start */
4647 update_recovery_start(dev, this);
4648 rest = this;
4649 }
4650
4651 /* if array has bad blocks, set suitable bit in array status */
4652 if (bbm_errors)
4653 rest->array.state |= (1<<MD_SB_BBM_ERRORS);
4654
4655 return rest;
4656 }
4657
4658
4659 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
4660 {
4661 struct imsm_map *map = get_imsm_map(dev, 0);
4662
4663 if (!failed)
4664 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
4665 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
4666
4667 switch (get_imsm_raid_level(map)) {
4668 case 0:
4669 return IMSM_T_STATE_FAILED;
4670 break;
4671 case 1:
4672 if (failed < map->num_members)
4673 return IMSM_T_STATE_DEGRADED;
4674 else
4675 return IMSM_T_STATE_FAILED;
4676 break;
4677 case 10:
4678 {
4679 /**
4680 * check to see if any mirrors have failed, otherwise we
4681 * are degraded. Even numbered slots are mirrored on
4682 * slot+1
4683 */
4684 int i;
4685 /* gcc -Os complains that this is unused */
4686 int insync = insync;
4687
4688 for (i = 0; i < map->num_members; i++) {
4689 __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1);
4690 int idx = ord_to_idx(ord);
4691 struct imsm_disk *disk;
4692
4693 /* reset the potential in-sync count on even-numbered
4694 * slots. num_copies is always 2 for imsm raid10
4695 */
4696 if ((i & 1) == 0)
4697 insync = 2;
4698
4699 disk = get_imsm_disk(super, idx);
4700 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
4701 insync--;
4702
4703 /* no in-sync disks left in this mirror the
4704 * array has failed
4705 */
4706 if (insync == 0)
4707 return IMSM_T_STATE_FAILED;
4708 }
4709
4710 return IMSM_T_STATE_DEGRADED;
4711 }
4712 case 5:
4713 if (failed < 2)
4714 return IMSM_T_STATE_DEGRADED;
4715 else
4716 return IMSM_T_STATE_FAILED;
4717 break;
4718 default:
4719 break;
4720 }
4721
4722 return map->map_state;
4723 }
4724
4725 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
4726 {
4727 int i;
4728 int failed = 0;
4729 struct imsm_disk *disk;
4730 struct imsm_map *map = get_imsm_map(dev, 0);
4731 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
4732 __u32 ord;
4733 int idx;
4734
4735 /* at the beginning of migration we set IMSM_ORD_REBUILD on
4736 * disks that are being rebuilt. New failures are recorded to
4737 * map[0]. So we look through all the disks we started with and
4738 * see if any failures are still present, or if any new ones
4739 * have arrived
4740 *
4741 * FIXME add support for online capacity expansion and
4742 * raid-level-migration
4743 */
4744 for (i = 0; i < prev->num_members; i++) {
4745 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
4746 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
4747 idx = ord_to_idx(ord);
4748
4749 disk = get_imsm_disk(super, idx);
4750 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
4751 failed++;
4752 }
4753
4754 return failed;
4755 }
4756
4757 #ifndef MDASSEMBLE
4758 static int imsm_open_new(struct supertype *c, struct active_array *a,
4759 char *inst)
4760 {
4761 struct intel_super *super = c->sb;
4762 struct imsm_super *mpb = super->anchor;
4763
4764 if (atoi(inst) >= mpb->num_raid_devs) {
4765 fprintf(stderr, "%s: subarry index %d, out of range\n",
4766 __func__, atoi(inst));
4767 return -ENODEV;
4768 }
4769
4770 dprintf("imsm: open_new %s\n", inst);
4771 a->info.container_member = atoi(inst);
4772 return 0;
4773 }
4774
4775 static int is_resyncing(struct imsm_dev *dev)
4776 {
4777 struct imsm_map *migr_map;
4778
4779 if (!dev->vol.migr_state)
4780 return 0;
4781
4782 if (migr_type(dev) == MIGR_INIT ||
4783 migr_type(dev) == MIGR_REPAIR)
4784 return 1;
4785
4786 if (migr_type(dev) == MIGR_GEN_MIGR)
4787 return 0;
4788
4789 migr_map = get_imsm_map(dev, 1);
4790
4791 if ((migr_map->map_state == IMSM_T_STATE_NORMAL) &&
4792 (dev->vol.migr_type != MIGR_GEN_MIGR))
4793 return 1;
4794 else
4795 return 0;
4796 }
4797
4798 /* return true if we recorded new information */
4799 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4800 {
4801 __u32 ord;
4802 int slot;
4803 struct imsm_map *map;
4804
4805 /* new failures are always set in map[0] */
4806 map = get_imsm_map(dev, 0);
4807
4808 slot = get_imsm_disk_slot(map, idx);
4809 if (slot < 0)
4810 return 0;
4811
4812 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
4813 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
4814 return 0;
4815
4816 disk->status |= FAILED_DISK;
4817 disk->status &= ~CONFIGURED_DISK;
4818 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
4819 if (map->failed_disk_num == 0xff)
4820 map->failed_disk_num = slot;
4821 return 1;
4822 }
4823
4824 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
4825 {
4826 mark_failure(dev, disk, idx);
4827
4828 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
4829 return;
4830
4831 disk->scsi_id = __cpu_to_le32(~(__u32)0);
4832 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
4833 }
4834
4835 static void handle_missing(struct intel_super *super, struct imsm_dev *dev)
4836 {
4837 __u8 map_state;
4838 struct dl *dl;
4839 int failed;
4840
4841 if (!super->missing)
4842 return;
4843 failed = imsm_count_failed(super, dev);
4844 map_state = imsm_check_degraded(super, dev, failed);
4845
4846 dprintf("imsm: mark missing\n");
4847 end_migration(dev, map_state);
4848 for (dl = super->missing; dl; dl = dl->next)
4849 mark_missing(dev, &dl->disk, dl->index);
4850 super->updates_pending++;
4851 }
4852
4853 static void imsm_set_disk(struct active_array *a, int n, int state);
4854
4855 static void imsm_progress_container_reshape(struct intel_super *super)
4856 {
4857 /* if no device has a migr_state, but some device has a
4858 * different number of members than the previous device, start
4859 * changing the number of devices in this device to match
4860 * previous.
4861 */
4862 struct imsm_super *mpb = super->anchor;
4863 int prev_disks = -1;
4864 int i;
4865
4866 for (i = 0; i < mpb->num_raid_devs; i++) {
4867 struct imsm_dev *dev = get_imsm_dev(super, i);
4868 struct imsm_map *map = get_imsm_map(dev, 0);
4869 struct imsm_map *map2;
4870 int prev_num_members;
4871 int used_disks;
4872
4873 if (dev->vol.migr_state)
4874 return;
4875
4876 if (prev_disks == -1)
4877 prev_disks = map->num_members;
4878 if (prev_disks == map->num_members)
4879 continue;
4880
4881 /* OK, this array needs to enter reshape mode.
4882 * i.e it needs a migr_state
4883 */
4884
4885 prev_num_members = map->num_members;
4886 map->num_members = prev_disks;
4887 dev->vol.migr_state = 1;
4888 dev->vol.curr_migr_unit = 0;
4889 dev->vol.migr_type = MIGR_GEN_MIGR;
4890 for (i = prev_num_members;
4891 i < map->num_members; i++)
4892 set_imsm_ord_tbl_ent(map, i, i);
4893 map2 = get_imsm_map(dev, 1);
4894 /* Copy the current map */
4895 memcpy(map2, map, sizeof_imsm_map(map));
4896 map2->num_members = prev_num_members;
4897
4898 /* calculate new size
4899 */
4900 used_disks = imsm_num_data_members(dev, 0);
4901 if (used_disks) {
4902 unsigned long long array_blocks;
4903
4904 array_blocks =
4905 map->blocks_per_member
4906 * used_disks;
4907 /* round array size down to closest MB
4908 */
4909 array_blocks = (array_blocks
4910 >> SECT_PER_MB_SHIFT)
4911 << SECT_PER_MB_SHIFT;
4912 dev->size_low =
4913 __cpu_to_le32((__u32)array_blocks);
4914 dev->size_high =
4915 __cpu_to_le32(
4916 (__u32)(array_blocks >> 32));
4917 }
4918 super->updates_pending++;
4919 }
4920 }
4921
4922 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
4923 * states are handled in imsm_set_disk() with one exception, when a
4924 * resync is stopped due to a new failure this routine will set the
4925 * 'degraded' state for the array.
4926 */
4927 static int imsm_set_array_state(struct active_array *a, int consistent)
4928 {
4929 int inst = a->info.container_member;
4930 struct intel_super *super = a->container->sb;
4931 struct imsm_dev *dev = get_imsm_dev(super, inst);
4932 struct imsm_map *map = get_imsm_map(dev, 0);
4933 int failed = imsm_count_failed(super, dev);
4934 __u8 map_state = imsm_check_degraded(super, dev, failed);
4935 __u32 blocks_per_unit;
4936
4937 if (dev->vol.migr_state &&
4938 dev->vol.migr_type == MIGR_GEN_MIGR) {
4939 /* array state change is blocked due to reshape action
4940 * We might need to
4941 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
4942 * - finish the reshape (if last_checkpoint is big and action != reshape)
4943 * - update curr_migr_unit
4944 */
4945 if (a->curr_action == reshape) {
4946 /* still reshaping, maybe update curr_migr_unit */
4947 long long blocks_per_unit = blocks_per_migr_unit(dev);
4948 long long unit = a->last_checkpoint;
4949 if (blocks_per_unit) {
4950 unit /= blocks_per_unit;
4951 if (unit >
4952 __le32_to_cpu(dev->vol.curr_migr_unit)) {
4953 dev->vol.curr_migr_unit =
4954 __cpu_to_le32(unit);
4955 super->updates_pending++;
4956 }
4957 }
4958 return 0;
4959 } else {
4960 if (a->last_checkpoint == 0 && a->prev_action == reshape) {
4961 /* for some reason we aborted the reshape.
4962 * Better clean up
4963 */
4964 struct imsm_map *map2 = get_imsm_map(dev, 1);
4965 dev->vol.migr_state = 0;
4966 dev->vol.migr_type = 0;
4967 dev->vol.curr_migr_unit = 0;
4968 memcpy(map, map2, sizeof_imsm_map(map2));
4969 super->updates_pending++;
4970 }
4971 if (a->last_checkpoint >= a->info.component_size) {
4972 unsigned long long array_blocks;
4973 int used_disks;
4974 /* it seems the reshape is all done */
4975 dev->vol.migr_state = 0;
4976 dev->vol.migr_type = 0;
4977 dev->vol.curr_migr_unit = 0;
4978
4979 used_disks = imsm_num_data_members(dev, -1);
4980 array_blocks = map->blocks_per_member * used_disks;
4981 /* round array size down to closest MB */
4982 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT)
4983 << SECT_PER_MB_SHIFT;
4984 dev->size_low = __cpu_to_le32((__u32) array_blocks);
4985 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
4986 a->info.custom_array_size = array_blocks;
4987 a->check_reshape = 1; /* encourage manager to update
4988 * array size
4989 */
4990 super->updates_pending++;
4991 imsm_progress_container_reshape(super);
4992 }
4993 }
4994 }
4995
4996 /* before we activate this array handle any missing disks */
4997 if (consistent == 2)
4998 handle_missing(super, dev);
4999
5000 if (consistent == 2 &&
5001 (!is_resync_complete(&a->info) ||
5002 map_state != IMSM_T_STATE_NORMAL ||
5003 dev->vol.migr_state))
5004 consistent = 0;
5005
5006 if (is_resync_complete(&a->info)) {
5007 /* complete intialization / resync,
5008 * recovery and interrupted recovery is completed in
5009 * ->set_disk
5010 */
5011 if (is_resyncing(dev)) {
5012 dprintf("imsm: mark resync done\n");
5013 end_migration(dev, map_state);
5014 super->updates_pending++;
5015 a->last_checkpoint = 0;
5016 }
5017 } else if (!is_resyncing(dev) && !failed) {
5018 /* mark the start of the init process if nothing is failed */
5019 dprintf("imsm: mark resync start\n");
5020 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
5021 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
5022 else
5023 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
5024 super->updates_pending++;
5025 }
5026
5027 /* check if we can update curr_migr_unit from resync_start, recovery_start */
5028 blocks_per_unit = blocks_per_migr_unit(dev);
5029 if (blocks_per_unit) {
5030 __u32 units32;
5031 __u64 units;
5032
5033 units = a->last_checkpoint / blocks_per_unit;
5034 units32 = units;
5035
5036 /* check that we did not overflow 32-bits, and that
5037 * curr_migr_unit needs updating
5038 */
5039 if (units32 == units &&
5040 __le32_to_cpu(dev->vol.curr_migr_unit) != units32) {
5041 dprintf("imsm: mark checkpoint (%u)\n", units32);
5042 dev->vol.curr_migr_unit = __cpu_to_le32(units32);
5043 super->updates_pending++;
5044 }
5045 }
5046
5047 /* mark dirty / clean */
5048 if (dev->vol.dirty != !consistent) {
5049 dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty");
5050 if (consistent)
5051 dev->vol.dirty = 0;
5052 else
5053 dev->vol.dirty = 1;
5054 super->updates_pending++;
5055 }
5056
5057 /* finalize online capacity expansion/reshape */
5058 if ((a->curr_action != reshape) &&
5059 (a->prev_action == reshape)) {
5060 struct mdinfo *mdi;
5061
5062 for (mdi = a->info.devs; mdi; mdi = mdi->next)
5063 imsm_set_disk(a, mdi->disk.raid_disk, mdi->curr_state);
5064 }
5065
5066 return consistent;
5067 }
5068
5069 static void imsm_set_disk(struct active_array *a, int n, int state)
5070 {
5071 int inst = a->info.container_member;
5072 struct intel_super *super = a->container->sb;
5073 struct imsm_dev *dev = get_imsm_dev(super, inst);
5074 struct imsm_map *map = get_imsm_map(dev, 0);
5075 struct imsm_disk *disk;
5076 int failed;
5077 __u32 ord;
5078 __u8 map_state;
5079
5080 if (n > map->num_members)
5081 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
5082 n, map->num_members - 1);
5083
5084 if (n < 0)
5085 return;
5086
5087 dprintf("imsm: set_disk %d:%x\n", n, state);
5088
5089 ord = get_imsm_ord_tbl_ent(dev, n, -1);
5090 disk = get_imsm_disk(super, ord_to_idx(ord));
5091
5092 /* check for new failures */
5093 if (state & DS_FAULTY) {
5094 if (mark_failure(dev, disk, ord_to_idx(ord)))
5095 super->updates_pending++;
5096 }
5097
5098 /* check if in_sync */
5099 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
5100 struct imsm_map *migr_map = get_imsm_map(dev, 1);
5101
5102 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
5103 super->updates_pending++;
5104 }
5105
5106 failed = imsm_count_failed(super, dev);
5107 map_state = imsm_check_degraded(super, dev, failed);
5108
5109 /* check if recovery complete, newly degraded, or failed */
5110 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
5111 end_migration(dev, map_state);
5112 map = get_imsm_map(dev, 0);
5113 map->failed_disk_num = ~0;
5114 super->updates_pending++;
5115 a->last_checkpoint = 0;
5116 } else if (map_state == IMSM_T_STATE_DEGRADED &&
5117 map->map_state != map_state &&
5118 !dev->vol.migr_state) {
5119 dprintf("imsm: mark degraded\n");
5120 map->map_state = map_state;
5121 super->updates_pending++;
5122 a->last_checkpoint = 0;
5123 } else if (map_state == IMSM_T_STATE_FAILED &&
5124 map->map_state != map_state) {
5125 dprintf("imsm: mark failed\n");
5126 end_migration(dev, map_state);
5127 super->updates_pending++;
5128 a->last_checkpoint = 0;
5129 } else if (is_gen_migration(dev)) {
5130 dprintf("imsm: Detected General Migration in state: ");
5131 if (map_state == IMSM_T_STATE_NORMAL) {
5132 end_migration(dev, map_state);
5133 map = get_imsm_map(dev, 0);
5134 map->failed_disk_num = ~0;
5135 dprintf("normal\n");
5136 } else {
5137 if (map_state == IMSM_T_STATE_DEGRADED) {
5138 printf("degraded\n");
5139 end_migration(dev, map_state);
5140 } else {
5141 dprintf("failed\n");
5142 }
5143 map->map_state = map_state;
5144 }
5145 super->updates_pending++;
5146 }
5147 }
5148
5149 static int store_imsm_mpb(int fd, struct imsm_super *mpb)
5150 {
5151 void *buf = mpb;
5152 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
5153 unsigned long long dsize;
5154 unsigned long long sectors;
5155
5156 get_dev_size(fd, NULL, &dsize);
5157
5158 if (mpb_size > 512) {
5159 /* -1 to account for anchor */
5160 sectors = mpb_sectors(mpb) - 1;
5161
5162 /* write the extended mpb to the sectors preceeding the anchor */
5163 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
5164 return 1;
5165
5166 if ((unsigned long long)write(fd, buf + 512, 512 * sectors)
5167 != 512 * sectors)
5168 return 1;
5169 }
5170
5171 /* first block is stored on second to last sector of the disk */
5172 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
5173 return 1;
5174
5175 if (write(fd, buf, 512) != 512)
5176 return 1;
5177
5178 return 0;
5179 }
5180
5181 static void imsm_sync_metadata(struct supertype *container)
5182 {
5183 struct intel_super *super = container->sb;
5184
5185 dprintf("sync metadata: %d\n", super->updates_pending);
5186 if (!super->updates_pending)
5187 return;
5188
5189 write_super_imsm(container, 0);
5190
5191 super->updates_pending = 0;
5192 }
5193
5194 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
5195 {
5196 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
5197 int i = get_imsm_disk_idx(dev, idx, -1);
5198 struct dl *dl;
5199
5200 for (dl = super->disks; dl; dl = dl->next)
5201 if (dl->index == i)
5202 break;
5203
5204 if (dl && is_failed(&dl->disk))
5205 dl = NULL;
5206
5207 if (dl)
5208 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
5209
5210 return dl;
5211 }
5212
5213 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
5214 struct active_array *a, int activate_new,
5215 struct mdinfo *additional_test_list)
5216 {
5217 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
5218 int idx = get_imsm_disk_idx(dev, slot, -1);
5219 struct imsm_super *mpb = super->anchor;
5220 struct imsm_map *map;
5221 unsigned long long pos;
5222 struct mdinfo *d;
5223 struct extent *ex;
5224 int i, j;
5225 int found;
5226 __u32 array_start = 0;
5227 __u32 array_end = 0;
5228 struct dl *dl;
5229 struct mdinfo *test_list;
5230
5231 for (dl = super->disks; dl; dl = dl->next) {
5232 /* If in this array, skip */
5233 for (d = a->info.devs ; d ; d = d->next)
5234 if (d->state_fd >= 0 &&
5235 d->disk.major == dl->major &&
5236 d->disk.minor == dl->minor) {
5237 dprintf("%x:%x already in array\n",
5238 dl->major, dl->minor);
5239 break;
5240 }
5241 if (d)
5242 continue;
5243 test_list = additional_test_list;
5244 while (test_list) {
5245 if (test_list->disk.major == dl->major &&
5246 test_list->disk.minor == dl->minor) {
5247 dprintf("%x:%x already in additional test list\n",
5248 dl->major, dl->minor);
5249 break;
5250 }
5251 test_list = test_list->next;
5252 }
5253 if (test_list)
5254 continue;
5255
5256 /* skip in use or failed drives */
5257 if (is_failed(&dl->disk) || idx == dl->index ||
5258 dl->index == -2) {
5259 dprintf("%x:%x status (failed: %d index: %d)\n",
5260 dl->major, dl->minor, is_failed(&dl->disk), idx);
5261 continue;
5262 }
5263
5264 /* skip pure spares when we are looking for partially
5265 * assimilated drives
5266 */
5267 if (dl->index == -1 && !activate_new)
5268 continue;
5269
5270 /* Does this unused device have the requisite free space?
5271 * It needs to be able to cover all member volumes
5272 */
5273 ex = get_extents(super, dl);
5274 if (!ex) {
5275 dprintf("cannot get extents\n");
5276 continue;
5277 }
5278 for (i = 0; i < mpb->num_raid_devs; i++) {
5279 dev = get_imsm_dev(super, i);
5280 map = get_imsm_map(dev, 0);
5281
5282 /* check if this disk is already a member of
5283 * this array
5284 */
5285 if (get_imsm_disk_slot(map, dl->index) >= 0)
5286 continue;
5287
5288 found = 0;
5289 j = 0;
5290 pos = 0;
5291 array_start = __le32_to_cpu(map->pba_of_lba0);
5292 array_end = array_start +
5293 __le32_to_cpu(map->blocks_per_member) - 1;
5294
5295 do {
5296 /* check that we can start at pba_of_lba0 with
5297 * blocks_per_member of space
5298 */
5299 if (array_start >= pos && array_end < ex[j].start) {
5300 found = 1;
5301 break;
5302 }
5303 pos = ex[j].start + ex[j].size;
5304 j++;
5305 } while (ex[j-1].size);
5306
5307 if (!found)
5308 break;
5309 }
5310
5311 free(ex);
5312 if (i < mpb->num_raid_devs) {
5313 dprintf("%x:%x does not have %u to %u available\n",
5314 dl->major, dl->minor, array_start, array_end);
5315 /* No room */
5316 continue;
5317 }
5318 return dl;
5319 }
5320
5321 return dl;
5322 }
5323
5324
5325 static int imsm_rebuild_allowed(struct supertype *cont, int dev_idx, int failed)
5326 {
5327 struct imsm_dev *dev2;
5328 struct imsm_map *map;
5329 struct dl *idisk;
5330 int slot;
5331 int idx;
5332 __u8 state;
5333
5334 dev2 = get_imsm_dev(cont->sb, dev_idx);
5335 if (dev2) {
5336 state = imsm_check_degraded(cont->sb, dev2, failed);
5337 if (state == IMSM_T_STATE_FAILED) {
5338 map = get_imsm_map(dev2, 0);
5339 if (!map)
5340 return 1;
5341 for (slot = 0; slot < map->num_members; slot++) {
5342 /*
5343 * Check if failed disks are deleted from intel
5344 * disk list or are marked to be deleted
5345 */
5346 idx = get_imsm_disk_idx(dev2, slot, -1);
5347 idisk = get_imsm_dl_disk(cont->sb, idx);
5348 /*
5349 * Do not rebuild the array if failed disks
5350 * from failed sub-array are not removed from
5351 * container.
5352 */
5353 if (idisk &&
5354 is_failed(&idisk->disk) &&
5355 (idisk->action != DISK_REMOVE))
5356 return 0;
5357 }
5358 }
5359 }
5360 return 1;
5361 }
5362
5363 static struct mdinfo *imsm_activate_spare(struct active_array *a,
5364 struct metadata_update **updates)
5365 {
5366 /**
5367 * Find a device with unused free space and use it to replace a
5368 * failed/vacant region in an array. We replace failed regions one a
5369 * array at a time. The result is that a new spare disk will be added
5370 * to the first failed array and after the monitor has finished
5371 * propagating failures the remainder will be consumed.
5372 *
5373 * FIXME add a capability for mdmon to request spares from another
5374 * container.
5375 */
5376
5377 struct intel_super *super = a->container->sb;
5378 int inst = a->info.container_member;
5379 struct imsm_dev *dev = get_imsm_dev(super, inst);
5380 struct imsm_map *map = get_imsm_map(dev, 0);
5381 int failed = a->info.array.raid_disks;
5382 struct mdinfo *rv = NULL;
5383 struct mdinfo *d;
5384 struct mdinfo *di;
5385 struct metadata_update *mu;
5386 struct dl *dl;
5387 struct imsm_update_activate_spare *u;
5388 int num_spares = 0;
5389 int i;
5390 int allowed;
5391
5392 for (d = a->info.devs ; d ; d = d->next) {
5393 if ((d->curr_state & DS_FAULTY) &&
5394 d->state_fd >= 0)
5395 /* wait for Removal to happen */
5396 return NULL;
5397 if (d->state_fd >= 0)
5398 failed--;
5399 }
5400
5401 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
5402 inst, failed, a->info.array.raid_disks, a->info.array.level);
5403
5404 if (dev->vol.migr_state &&
5405 dev->vol.migr_type == MIGR_GEN_MIGR)
5406 /* No repair during migration */
5407 return NULL;
5408
5409 if (a->info.array.level == 4)
5410 /* No repair for takeovered array
5411 * imsm doesn't support raid4
5412 */
5413 return NULL;
5414
5415 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
5416 return NULL;
5417
5418 /*
5419 * If there are any failed disks check state of the other volume.
5420 * Block rebuild if the another one is failed until failed disks
5421 * are removed from container.
5422 */
5423 if (failed) {
5424 dprintf("found failed disks in %s, check if there another"
5425 "failed sub-array.\n",
5426 dev->volume);
5427 /* check if states of the other volumes allow for rebuild */
5428 for (i = 0; i < super->anchor->num_raid_devs; i++) {
5429 if (i != inst) {
5430 allowed = imsm_rebuild_allowed(a->container,
5431 i, failed);
5432 if (!allowed)
5433 return NULL;
5434 }
5435 }
5436 }
5437
5438 /* For each slot, if it is not working, find a spare */
5439 for (i = 0; i < a->info.array.raid_disks; i++) {
5440 for (d = a->info.devs ; d ; d = d->next)
5441 if (d->disk.raid_disk == i)
5442 break;
5443 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
5444 if (d && (d->state_fd >= 0))
5445 continue;
5446
5447 /*
5448 * OK, this device needs recovery. Try to re-add the
5449 * previous occupant of this slot, if this fails see if
5450 * we can continue the assimilation of a spare that was
5451 * partially assimilated, finally try to activate a new
5452 * spare.
5453 */
5454 dl = imsm_readd(super, i, a);
5455 if (!dl)
5456 dl = imsm_add_spare(super, i, a, 0, NULL);
5457 if (!dl)
5458 dl = imsm_add_spare(super, i, a, 1, NULL);
5459 if (!dl)
5460 continue;
5461
5462 /* found a usable disk with enough space */
5463 di = malloc(sizeof(*di));
5464 if (!di)
5465 continue;
5466 memset(di, 0, sizeof(*di));
5467
5468 /* dl->index will be -1 in the case we are activating a
5469 * pristine spare. imsm_process_update() will create a
5470 * new index in this case. Once a disk is found to be
5471 * failed in all member arrays it is kicked from the
5472 * metadata
5473 */
5474 di->disk.number = dl->index;
5475
5476 /* (ab)use di->devs to store a pointer to the device
5477 * we chose
5478 */
5479 di->devs = (struct mdinfo *) dl;
5480
5481 di->disk.raid_disk = i;
5482 di->disk.major = dl->major;
5483 di->disk.minor = dl->minor;
5484 di->disk.state = 0;
5485 di->recovery_start = 0;
5486 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
5487 di->component_size = a->info.component_size;
5488 di->container_member = inst;
5489 super->random = random32();
5490 di->next = rv;
5491 rv = di;
5492 num_spares++;
5493 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
5494 i, di->data_offset);
5495
5496 break;
5497 }
5498
5499 if (!rv)
5500 /* No spares found */
5501 return rv;
5502 /* Now 'rv' has a list of devices to return.
5503 * Create a metadata_update record to update the
5504 * disk_ord_tbl for the array
5505 */
5506 mu = malloc(sizeof(*mu));
5507 if (mu) {
5508 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
5509 if (mu->buf == NULL) {
5510 free(mu);
5511 mu = NULL;
5512 }
5513 }
5514 if (!mu) {
5515 while (rv) {
5516 struct mdinfo *n = rv->next;
5517
5518 free(rv);
5519 rv = n;
5520 }
5521 return NULL;
5522 }
5523
5524 mu->space = NULL;
5525 mu->space_list = NULL;
5526 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
5527 mu->next = *updates;
5528 u = (struct imsm_update_activate_spare *) mu->buf;
5529
5530 for (di = rv ; di ; di = di->next) {
5531 u->type = update_activate_spare;
5532 u->dl = (struct dl *) di->devs;
5533 di->devs = NULL;
5534 u->slot = di->disk.raid_disk;
5535 u->array = inst;
5536 u->next = u + 1;
5537 u++;
5538 }
5539 (u-1)->next = NULL;
5540 *updates = mu;
5541
5542 return rv;
5543 }
5544
5545 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
5546 {
5547 struct imsm_dev *dev = get_imsm_dev(super, idx);
5548 struct imsm_map *map = get_imsm_map(dev, 0);
5549 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
5550 struct disk_info *inf = get_disk_info(u);
5551 struct imsm_disk *disk;
5552 int i;
5553 int j;
5554
5555 for (i = 0; i < map->num_members; i++) {
5556 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i, -1));
5557 for (j = 0; j < new_map->num_members; j++)
5558 if (serialcmp(disk->serial, inf[j].serial) == 0)
5559 return 1;
5560 }
5561
5562 return 0;
5563 }
5564
5565
5566 static struct dl *get_disk_super(struct intel_super *super, int major, int minor)
5567 {
5568 struct dl *dl = NULL;
5569 for (dl = super->disks; dl; dl = dl->next)
5570 if ((dl->major == major) && (dl->minor == minor))
5571 return dl;
5572 return NULL;
5573 }
5574
5575 static int remove_disk_super(struct intel_super *super, int major, int minor)
5576 {
5577 struct dl *prev = NULL;
5578 struct dl *dl;
5579
5580 prev = NULL;
5581 for (dl = super->disks; dl; dl = dl->next) {
5582 if ((dl->major == major) && (dl->minor == minor)) {
5583 /* remove */
5584 if (prev)
5585 prev->next = dl->next;
5586 else
5587 super->disks = dl->next;
5588 dl->next = NULL;
5589 __free_imsm_disk(dl);
5590 dprintf("%s: removed %x:%x\n",
5591 __func__, major, minor);
5592 break;
5593 }
5594 prev = dl;
5595 }
5596 return 0;
5597 }
5598
5599 static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index);
5600
5601 static int add_remove_disk_update(struct intel_super *super)
5602 {
5603 int check_degraded = 0;
5604 struct dl *disk = NULL;
5605 /* add/remove some spares to/from the metadata/contrainer */
5606 while (super->disk_mgmt_list) {
5607 struct dl *disk_cfg;
5608
5609 disk_cfg = super->disk_mgmt_list;
5610 super->disk_mgmt_list = disk_cfg->next;
5611 disk_cfg->next = NULL;
5612
5613 if (disk_cfg->action == DISK_ADD) {
5614 disk_cfg->next = super->disks;
5615 super->disks = disk_cfg;
5616 check_degraded = 1;
5617 dprintf("%s: added %x:%x\n",
5618 __func__, disk_cfg->major,
5619 disk_cfg->minor);
5620 } else if (disk_cfg->action == DISK_REMOVE) {
5621 dprintf("Disk remove action processed: %x.%x\n",
5622 disk_cfg->major, disk_cfg->minor);
5623 disk = get_disk_super(super,
5624 disk_cfg->major,
5625 disk_cfg->minor);
5626 if (disk) {
5627 /* store action status */
5628 disk->action = DISK_REMOVE;
5629 /* remove spare disks only */
5630 if (disk->index == -1) {
5631 remove_disk_super(super,
5632 disk_cfg->major,
5633 disk_cfg->minor);
5634 }
5635 }
5636 /* release allocate disk structure */
5637 __free_imsm_disk(disk_cfg);
5638 }
5639 }
5640 return check_degraded;
5641 }
5642
5643 static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
5644 struct intel_super *super,
5645 void ***space_list)
5646 {
5647 struct dl *new_disk;
5648 struct intel_dev *id;
5649 int i;
5650 int delta_disks = u->new_raid_disks - u->old_raid_disks;
5651 int disk_count = u->old_raid_disks;
5652 void **tofree = NULL;
5653 int devices_to_reshape = 1;
5654 struct imsm_super *mpb = super->anchor;
5655 int ret_val = 0;
5656
5657 dprintf("imsm: imsm_process_update() for update_reshape\n");
5658
5659 /* enable spares to use in array */
5660 for (i = 0; i < delta_disks; i++) {
5661 new_disk = get_disk_super(super,
5662 major(u->new_disks[i]),
5663 minor(u->new_disks[i]));
5664 dprintf("imsm: imsm_process_update(): new disk "
5665 "for reshape is: %i:%i (%p, index = %i)\n",
5666 major(u->new_disks[i]), minor(u->new_disks[i]),
5667 new_disk, new_disk->index);
5668 if ((new_disk == NULL) ||
5669 ((new_disk->index >= 0) &&
5670 (new_disk->index < u->old_raid_disks)))
5671 goto update_reshape_exit;
5672 new_disk->index = disk_count++;
5673 /* slot to fill in autolayout
5674 */
5675 new_disk->raiddisk = new_disk->index;
5676 new_disk->disk.status |=
5677 CONFIGURED_DISK;
5678 new_disk->disk.status &= ~SPARE_DISK;
5679 }
5680
5681 dprintf("imsm: process_update(): update_reshape: volume set"
5682 " mpb->num_raid_devs = %i\n", mpb->num_raid_devs);
5683 /* manage changes in volume
5684 */
5685 for (id = super->devlist ; id; id = id->next) {
5686 void **sp = *space_list;
5687 struct imsm_dev *newdev;
5688 struct imsm_map *newmap, *oldmap;
5689
5690 if (!sp)
5691 continue;
5692 *space_list = *sp;
5693 newdev = (void*)sp;
5694 /* Copy the dev, but not (all of) the map */
5695 memcpy(newdev, id->dev, sizeof(*newdev));
5696 oldmap = get_imsm_map(id->dev, 0);
5697 newmap = get_imsm_map(newdev, 0);
5698 /* Copy the current map */
5699 memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
5700 /* update one device only
5701 */
5702 if (devices_to_reshape) {
5703 int used_disks;
5704
5705 dprintf("process_update(): modifying "
5706 "subdev: %i\n", id->index);
5707 devices_to_reshape--;
5708 newdev->vol.migr_state = 1;
5709 newdev->vol.curr_migr_unit = 0;
5710 newdev->vol.migr_type = MIGR_GEN_MIGR;
5711 newmap->num_members = u->new_raid_disks;
5712 for (i = 0; i < delta_disks; i++) {
5713 set_imsm_ord_tbl_ent(newmap,
5714 u->old_raid_disks + i,
5715 u->old_raid_disks + i);
5716 }
5717 /* New map is correct, now need to save old map
5718 */
5719 newmap = get_imsm_map(newdev, 1);
5720 memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
5721
5722 /* calculate new size
5723 */
5724 used_disks = imsm_num_data_members(newdev, 0);
5725 if (used_disks) {
5726 unsigned long long array_blocks;
5727
5728 array_blocks =
5729 newmap->blocks_per_member * used_disks;
5730 /* round array size down to closest MB
5731 */
5732 array_blocks = (array_blocks
5733 >> SECT_PER_MB_SHIFT)
5734 << SECT_PER_MB_SHIFT;
5735 newdev->size_low =
5736 __cpu_to_le32((__u32)array_blocks);
5737 newdev->size_high =
5738 __cpu_to_le32((__u32)(array_blocks >> 32));
5739 }
5740 }
5741
5742 sp = (void **)id->dev;
5743 id->dev = newdev;
5744 *sp = tofree;
5745 tofree = sp;
5746 }
5747 ret_val = 1;
5748
5749 update_reshape_exit:
5750
5751 return ret_val;
5752 }
5753
5754 static void imsm_process_update(struct supertype *st,
5755 struct metadata_update *update)
5756 {
5757 /**
5758 * crack open the metadata_update envelope to find the update record
5759 * update can be one of:
5760 * update_reshape_container_disks - all the arrays in the container
5761 * are being reshaped to have more devices. We need to mark
5762 * the arrays for general migration and convert selected spares
5763 * into active devices.
5764 * update_activate_spare - a spare device has replaced a failed
5765 * device in an array, update the disk_ord_tbl. If this disk is
5766 * present in all member arrays then also clear the SPARE_DISK
5767 * flag
5768 * update_create_array
5769 * update_kill_array
5770 * update_rename_array
5771 * update_add_remove_disk
5772 */
5773 struct intel_super *super = st->sb;
5774 struct imsm_super *mpb;
5775 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
5776
5777 /* update requires a larger buf but the allocation failed */
5778 if (super->next_len && !super->next_buf) {
5779 super->next_len = 0;
5780 return;
5781 }
5782
5783 if (super->next_buf) {
5784 memcpy(super->next_buf, super->buf, super->len);
5785 free(super->buf);
5786 super->len = super->next_len;
5787 super->buf = super->next_buf;
5788
5789 super->next_len = 0;
5790 super->next_buf = NULL;
5791 }
5792
5793 mpb = super->anchor;
5794
5795 switch (type) {
5796 case update_reshape_container_disks: {
5797 struct imsm_update_reshape *u = (void *)update->buf;
5798 if (apply_reshape_container_disks_update(
5799 u, super, &update->space_list))
5800 super->updates_pending++;
5801 break;
5802 }
5803 case update_activate_spare: {
5804 struct imsm_update_activate_spare *u = (void *) update->buf;
5805 struct imsm_dev *dev = get_imsm_dev(super, u->array);
5806 struct imsm_map *map = get_imsm_map(dev, 0);
5807 struct imsm_map *migr_map;
5808 struct active_array *a;
5809 struct imsm_disk *disk;
5810 __u8 to_state;
5811 struct dl *dl;
5812 unsigned int found;
5813 int failed;
5814 int victim = get_imsm_disk_idx(dev, u->slot, -1);
5815 int i;
5816
5817 for (dl = super->disks; dl; dl = dl->next)
5818 if (dl == u->dl)
5819 break;
5820
5821 if (!dl) {
5822 fprintf(stderr, "error: imsm_activate_spare passed "
5823 "an unknown disk (index: %d)\n",
5824 u->dl->index);
5825 return;
5826 }
5827
5828 super->updates_pending++;
5829
5830 /* count failures (excluding rebuilds and the victim)
5831 * to determine map[0] state
5832 */
5833 failed = 0;
5834 for (i = 0; i < map->num_members; i++) {
5835 if (i == u->slot)
5836 continue;
5837 disk = get_imsm_disk(super,
5838 get_imsm_disk_idx(dev, i, -1));
5839 if (!disk || is_failed(disk))
5840 failed++;
5841 }
5842
5843 /* adding a pristine spare, assign a new index */
5844 if (dl->index < 0) {
5845 dl->index = super->anchor->num_disks;
5846 super->anchor->num_disks++;
5847 }
5848 disk = &dl->disk;
5849 disk->status |= CONFIGURED_DISK;
5850 disk->status &= ~SPARE_DISK;
5851
5852 /* mark rebuild */
5853 to_state = imsm_check_degraded(super, dev, failed);
5854 map->map_state = IMSM_T_STATE_DEGRADED;
5855 migrate(dev, to_state, MIGR_REBUILD);
5856 migr_map = get_imsm_map(dev, 1);
5857 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
5858 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
5859
5860 /* update the family_num to mark a new container
5861 * generation, being careful to record the existing
5862 * family_num in orig_family_num to clean up after
5863 * earlier mdadm versions that neglected to set it.
5864 */
5865 if (mpb->orig_family_num == 0)
5866 mpb->orig_family_num = mpb->family_num;
5867 mpb->family_num += super->random;
5868
5869 /* count arrays using the victim in the metadata */
5870 found = 0;
5871 for (a = st->arrays; a ; a = a->next) {
5872 dev = get_imsm_dev(super, a->info.container_member);
5873 map = get_imsm_map(dev, 0);
5874
5875 if (get_imsm_disk_slot(map, victim) >= 0)
5876 found++;
5877 }
5878
5879 /* delete the victim if it is no longer being
5880 * utilized anywhere
5881 */
5882 if (!found) {
5883 struct dl **dlp;
5884
5885 /* We know that 'manager' isn't touching anything,
5886 * so it is safe to delete
5887 */
5888 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
5889 if ((*dlp)->index == victim)
5890 break;
5891
5892 /* victim may be on the missing list */
5893 if (!*dlp)
5894 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
5895 if ((*dlp)->index == victim)
5896 break;
5897 imsm_delete(super, dlp, victim);
5898 }
5899 break;
5900 }
5901 case update_create_array: {
5902 /* someone wants to create a new array, we need to be aware of
5903 * a few races/collisions:
5904 * 1/ 'Create' called by two separate instances of mdadm
5905 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
5906 * devices that have since been assimilated via
5907 * activate_spare.
5908 * In the event this update can not be carried out mdadm will
5909 * (FIX ME) notice that its update did not take hold.
5910 */
5911 struct imsm_update_create_array *u = (void *) update->buf;
5912 struct intel_dev *dv;
5913 struct imsm_dev *dev;
5914 struct imsm_map *map, *new_map;
5915 unsigned long long start, end;
5916 unsigned long long new_start, new_end;
5917 int i;
5918 struct disk_info *inf;
5919 struct dl *dl;
5920
5921 /* handle racing creates: first come first serve */
5922 if (u->dev_idx < mpb->num_raid_devs) {
5923 dprintf("%s: subarray %d already defined\n",
5924 __func__, u->dev_idx);
5925 goto create_error;
5926 }
5927
5928 /* check update is next in sequence */
5929 if (u->dev_idx != mpb->num_raid_devs) {
5930 dprintf("%s: can not create array %d expected index %d\n",
5931 __func__, u->dev_idx, mpb->num_raid_devs);
5932 goto create_error;
5933 }
5934
5935 new_map = get_imsm_map(&u->dev, 0);
5936 new_start = __le32_to_cpu(new_map->pba_of_lba0);
5937 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
5938 inf = get_disk_info(u);
5939
5940 /* handle activate_spare versus create race:
5941 * check to make sure that overlapping arrays do not include
5942 * overalpping disks
5943 */
5944 for (i = 0; i < mpb->num_raid_devs; i++) {
5945 dev = get_imsm_dev(super, i);
5946 map = get_imsm_map(dev, 0);
5947 start = __le32_to_cpu(map->pba_of_lba0);
5948 end = start + __le32_to_cpu(map->blocks_per_member);
5949 if ((new_start >= start && new_start <= end) ||
5950 (start >= new_start && start <= new_end))
5951 /* overlap */;
5952 else
5953 continue;
5954
5955 if (disks_overlap(super, i, u)) {
5956 dprintf("%s: arrays overlap\n", __func__);
5957 goto create_error;
5958 }
5959 }
5960
5961 /* check that prepare update was successful */
5962 if (!update->space) {
5963 dprintf("%s: prepare update failed\n", __func__);
5964 goto create_error;
5965 }
5966
5967 /* check that all disks are still active before committing
5968 * changes. FIXME: could we instead handle this by creating a
5969 * degraded array? That's probably not what the user expects,
5970 * so better to drop this update on the floor.
5971 */
5972 for (i = 0; i < new_map->num_members; i++) {
5973 dl = serial_to_dl(inf[i].serial, super);
5974 if (!dl) {
5975 dprintf("%s: disk disappeared\n", __func__);
5976 goto create_error;
5977 }
5978 }
5979
5980 super->updates_pending++;
5981
5982 /* convert spares to members and fixup ord_tbl */
5983 for (i = 0; i < new_map->num_members; i++) {
5984 dl = serial_to_dl(inf[i].serial, super);
5985 if (dl->index == -1) {
5986 dl->index = mpb->num_disks;
5987 mpb->num_disks++;
5988 dl->disk.status |= CONFIGURED_DISK;
5989 dl->disk.status &= ~SPARE_DISK;
5990 }
5991 set_imsm_ord_tbl_ent(new_map, i, dl->index);
5992 }
5993
5994 dv = update->space;
5995 dev = dv->dev;
5996 update->space = NULL;
5997 imsm_copy_dev(dev, &u->dev);
5998 dv->index = u->dev_idx;
5999 dv->next = super->devlist;
6000 super->devlist = dv;
6001 mpb->num_raid_devs++;
6002
6003 imsm_update_version_info(super);
6004 break;
6005 create_error:
6006 /* mdmon knows how to release update->space, but not
6007 * ((struct intel_dev *) update->space)->dev
6008 */
6009 if (update->space) {
6010 dv = update->space;
6011 free(dv->dev);
6012 }
6013 break;
6014 }
6015 case update_kill_array: {
6016 struct imsm_update_kill_array *u = (void *) update->buf;
6017 int victim = u->dev_idx;
6018 struct active_array *a;
6019 struct intel_dev **dp;
6020 struct imsm_dev *dev;
6021
6022 /* sanity check that we are not affecting the uuid of
6023 * active arrays, or deleting an active array
6024 *
6025 * FIXME when immutable ids are available, but note that
6026 * we'll also need to fixup the invalidated/active
6027 * subarray indexes in mdstat
6028 */
6029 for (a = st->arrays; a; a = a->next)
6030 if (a->info.container_member >= victim)
6031 break;
6032 /* by definition if mdmon is running at least one array
6033 * is active in the container, so checking
6034 * mpb->num_raid_devs is just extra paranoia
6035 */
6036 dev = get_imsm_dev(super, victim);
6037 if (a || !dev || mpb->num_raid_devs == 1) {
6038 dprintf("failed to delete subarray-%d\n", victim);
6039 break;
6040 }
6041
6042 for (dp = &super->devlist; *dp;)
6043 if ((*dp)->index == (unsigned)super->current_vol) {
6044 *dp = (*dp)->next;
6045 } else {
6046 if ((*dp)->index > (unsigned)victim)
6047 (*dp)->index--;
6048 dp = &(*dp)->next;
6049 }
6050 mpb->num_raid_devs--;
6051 super->updates_pending++;
6052 break;
6053 }
6054 case update_rename_array: {
6055 struct imsm_update_rename_array *u = (void *) update->buf;
6056 char name[MAX_RAID_SERIAL_LEN+1];
6057 int target = u->dev_idx;
6058 struct active_array *a;
6059 struct imsm_dev *dev;
6060
6061 /* sanity check that we are not affecting the uuid of
6062 * an active array
6063 */
6064 snprintf(name, MAX_RAID_SERIAL_LEN, "%s", (char *) u->name);
6065 name[MAX_RAID_SERIAL_LEN] = '\0';
6066 for (a = st->arrays; a; a = a->next)
6067 if (a->info.container_member == target)
6068 break;
6069 dev = get_imsm_dev(super, u->dev_idx);
6070 if (a || !dev || !check_name(super, name, 1)) {
6071 dprintf("failed to rename subarray-%d\n", target);
6072 break;
6073 }
6074
6075 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
6076 super->updates_pending++;
6077 break;
6078 }
6079 case update_add_remove_disk: {
6080 /* we may be able to repair some arrays if disks are
6081 * being added, check teh status of add_remove_disk
6082 * if discs has been added.
6083 */
6084 if (add_remove_disk_update(super)) {
6085 struct active_array *a;
6086
6087 super->updates_pending++;
6088 for (a = st->arrays; a; a = a->next)
6089 a->check_degraded = 1;
6090 }
6091 break;
6092 }
6093 default:
6094 fprintf(stderr, "error: unsuported process update type:"
6095 "(type: %d)\n", type);
6096 }
6097 }
6098
6099 static void imsm_prepare_update(struct supertype *st,
6100 struct metadata_update *update)
6101 {
6102 /**
6103 * Allocate space to hold new disk entries, raid-device entries or a new
6104 * mpb if necessary. The manager synchronously waits for updates to
6105 * complete in the monitor, so new mpb buffers allocated here can be
6106 * integrated by the monitor thread without worrying about live pointers
6107 * in the manager thread.
6108 */
6109 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
6110 struct intel_super *super = st->sb;
6111 struct imsm_super *mpb = super->anchor;
6112 size_t buf_len;
6113 size_t len = 0;
6114
6115 switch (type) {
6116 case update_reshape_container_disks: {
6117 /* Every raid device in the container is about to
6118 * gain some more devices, and we will enter a
6119 * reconfiguration.
6120 * So each 'imsm_map' will be bigger, and the imsm_vol
6121 * will now hold 2 of them.
6122 * Thus we need new 'struct imsm_dev' allocations sized
6123 * as sizeof_imsm_dev but with more devices in both maps.
6124 */
6125 struct imsm_update_reshape *u = (void *)update->buf;
6126 struct intel_dev *dl;
6127 void **space_tail = (void**)&update->space_list;
6128
6129 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
6130
6131 for (dl = super->devlist; dl; dl = dl->next) {
6132 int size = sizeof_imsm_dev(dl->dev, 1);
6133 void *s;
6134 if (u->new_raid_disks > u->old_raid_disks)
6135 size += sizeof(__u32)*2*
6136 (u->new_raid_disks - u->old_raid_disks);
6137 s = malloc(size);
6138 if (!s)
6139 break;
6140 *space_tail = s;
6141 space_tail = s;
6142 *space_tail = NULL;
6143 }
6144
6145 len = disks_to_mpb_size(u->new_raid_disks);
6146 dprintf("New anchor length is %llu\n", (unsigned long long)len);
6147 break;
6148 }
6149 case update_create_array: {
6150 struct imsm_update_create_array *u = (void *) update->buf;
6151 struct intel_dev *dv;
6152 struct imsm_dev *dev = &u->dev;
6153 struct imsm_map *map = get_imsm_map(dev, 0);
6154 struct dl *dl;
6155 struct disk_info *inf;
6156 int i;
6157 int activate = 0;
6158
6159 inf = get_disk_info(u);
6160 len = sizeof_imsm_dev(dev, 1);
6161 /* allocate a new super->devlist entry */
6162 dv = malloc(sizeof(*dv));
6163 if (dv) {
6164 dv->dev = malloc(len);
6165 if (dv->dev)
6166 update->space = dv;
6167 else {
6168 free(dv);
6169 update->space = NULL;
6170 }
6171 }
6172
6173 /* count how many spares will be converted to members */
6174 for (i = 0; i < map->num_members; i++) {
6175 dl = serial_to_dl(inf[i].serial, super);
6176 if (!dl) {
6177 /* hmm maybe it failed?, nothing we can do about
6178 * it here
6179 */
6180 continue;
6181 }
6182 if (count_memberships(dl, super) == 0)
6183 activate++;
6184 }
6185 len += activate * sizeof(struct imsm_disk);
6186 break;
6187 default:
6188 break;
6189 }
6190 }
6191
6192 /* check if we need a larger metadata buffer */
6193 if (super->next_buf)
6194 buf_len = super->next_len;
6195 else
6196 buf_len = super->len;
6197
6198 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
6199 /* ok we need a larger buf than what is currently allocated
6200 * if this allocation fails process_update will notice that
6201 * ->next_len is set and ->next_buf is NULL
6202 */
6203 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
6204 if (super->next_buf)
6205 free(super->next_buf);
6206
6207 super->next_len = buf_len;
6208 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
6209 memset(super->next_buf, 0, buf_len);
6210 else
6211 super->next_buf = NULL;
6212 }
6213 }
6214
6215 /* must be called while manager is quiesced */
6216 static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index)
6217 {
6218 struct imsm_super *mpb = super->anchor;
6219 struct dl *iter;
6220 struct imsm_dev *dev;
6221 struct imsm_map *map;
6222 int i, j, num_members;
6223 __u32 ord;
6224
6225 dprintf("%s: deleting device[%d] from imsm_super\n",
6226 __func__, index);
6227
6228 /* shift all indexes down one */
6229 for (iter = super->disks; iter; iter = iter->next)
6230 if (iter->index > (int)index)
6231 iter->index--;
6232 for (iter = super->missing; iter; iter = iter->next)
6233 if (iter->index > (int)index)
6234 iter->index--;
6235
6236 for (i = 0; i < mpb->num_raid_devs; i++) {
6237 dev = get_imsm_dev(super, i);
6238 map = get_imsm_map(dev, 0);
6239 num_members = map->num_members;
6240 for (j = 0; j < num_members; j++) {
6241 /* update ord entries being careful not to propagate
6242 * ord-flags to the first map
6243 */
6244 ord = get_imsm_ord_tbl_ent(dev, j, -1);
6245
6246 if (ord_to_idx(ord) <= index)
6247 continue;
6248
6249 map = get_imsm_map(dev, 0);
6250 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
6251 map = get_imsm_map(dev, 1);
6252 if (map)
6253 set_imsm_ord_tbl_ent(map, j, ord - 1);
6254 }
6255 }
6256
6257 mpb->num_disks--;
6258 super->updates_pending++;
6259 if (*dlp) {
6260 struct dl *dl = *dlp;
6261
6262 *dlp = (*dlp)->next;
6263 __free_imsm_disk(dl);
6264 }
6265 }
6266 #endif /* MDASSEMBLE */
6267
6268 static char disk_by_path[] = "/dev/disk/by-path/";
6269
6270 static const char *imsm_get_disk_controller_domain(const char *path)
6271 {
6272 struct sys_dev *list, *hba = NULL;
6273 char disk_path[PATH_MAX];
6274 int ahci = 0;
6275 char *dpath = NULL;
6276
6277 list = find_driver_devices("pci", "ahci");
6278 for (hba = list; hba; hba = hba->next)
6279 if (devpath_to_vendor(hba->path) == 0x8086)
6280 break;
6281
6282 if (hba) {
6283 struct stat st;
6284
6285 strncpy(disk_path, disk_by_path, PATH_MAX - 1);
6286 strncat(disk_path, path, PATH_MAX - strlen(disk_path) - 1);
6287 if (stat(disk_path, &st) == 0) {
6288 dpath = devt_to_devpath(st.st_rdev);
6289 if (dpath)
6290 ahci = path_attached_to_hba(dpath, hba->path);
6291 }
6292 }
6293 dprintf("path: %s(%s) hba: %s attached: %d\n",
6294 path, dpath, (hba) ? hba->path : "NULL", ahci);
6295 free_sys_dev(&list);
6296 if (ahci)
6297 return "ahci";
6298 else
6299 return NULL;
6300 }
6301
6302 static int imsm_find_array_minor_by_subdev(int subdev, int container, int *minor)
6303 {
6304 char subdev_name[20];
6305 struct mdstat_ent *mdstat;
6306
6307 sprintf(subdev_name, "%d", subdev);
6308 mdstat = mdstat_by_subdev(subdev_name, container);
6309 if (!mdstat)
6310 return -1;
6311
6312 *minor = mdstat->devnum;
6313 free_mdstat(mdstat);
6314 return 0;
6315 }
6316
6317 static int imsm_reshape_is_allowed_on_container(struct supertype *st,
6318 struct geo_params *geo,
6319 int *old_raid_disks)
6320 {
6321 int ret_val = 0;
6322 struct mdinfo *info, *member;
6323 int devices_that_can_grow = 0;
6324
6325 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
6326 "st->devnum = (%i)\n",
6327 st->devnum);
6328
6329 if (geo->size != -1 ||
6330 geo->level != UnSet ||
6331 geo->layout != UnSet ||
6332 geo->chunksize != 0 ||
6333 geo->raid_disks == UnSet) {
6334 dprintf("imsm: Container operation is allowed for "
6335 "raid disks number change only.\n");
6336 return ret_val;
6337 }
6338
6339 info = container_content_imsm(st, NULL);
6340 for (member = info; member; member = member->next) {
6341 int result;
6342 int minor;
6343
6344 dprintf("imsm: checking device_num: %i\n",
6345 member->container_member);
6346
6347 if (geo->raid_disks < member->array.raid_disks) {
6348 /* we work on container for Online Capacity Expansion
6349 * only so raid_disks has to grow
6350 */
6351 dprintf("imsm: for container operation raid disks "
6352 "increase is required\n");
6353 break;
6354 }
6355
6356 if ((info->array.level != 0) &&
6357 (info->array.level != 5)) {
6358 /* we cannot use this container with other raid level
6359 */
6360 dprintf("imsm: for container operation wrong"
6361 " raid level (%i) detected\n",
6362 info->array.level);
6363 break;
6364 } else {
6365 /* check for platform support
6366 * for this raid level configuration
6367 */
6368 struct intel_super *super = st->sb;
6369 if (!is_raid_level_supported(super->orom,
6370 member->array.level,
6371 geo->raid_disks)) {
6372 dprintf("platform does not support raid%d with"
6373 " %d disk%s\n",
6374 info->array.level,
6375 geo->raid_disks,
6376 geo->raid_disks > 1 ? "s" : "");
6377 break;
6378 }
6379 }
6380
6381 if (*old_raid_disks &&
6382 info->array.raid_disks != *old_raid_disks)
6383 break;
6384 *old_raid_disks = info->array.raid_disks;
6385
6386 /* All raid5 and raid0 volumes in container
6387 * have to be ready for Online Capacity Expansion
6388 * so they need to be assembled. We have already
6389 * checked that no recovery etc is happening.
6390 */
6391 result = imsm_find_array_minor_by_subdev(member->container_member,
6392 st->container_dev,
6393 &minor);
6394 if (result < 0) {
6395 dprintf("imsm: cannot find array\n");
6396 break;
6397 }
6398 devices_that_can_grow++;
6399 }
6400 sysfs_free(info);
6401 if (!member && devices_that_can_grow)
6402 ret_val = 1;
6403
6404 if (ret_val)
6405 dprintf("\tContainer operation allowed\n");
6406 else
6407 dprintf("\tError: %i\n", ret_val);
6408
6409 return ret_val;
6410 }
6411
6412 /* Function: get_spares_for_grow
6413 * Description: Allocates memory and creates list of spare devices
6414 * avaliable in container. Checks if spare drive size is acceptable.
6415 * Parameters: Pointer to the supertype structure
6416 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
6417 * NULL if fail
6418 */
6419 static struct mdinfo *get_spares_for_grow(struct supertype *st)
6420 {
6421 unsigned long long min_size = min_acceptable_spare_size_imsm(st);
6422 return container_choose_spares(st, min_size, NULL, NULL, NULL, 0);
6423 }
6424
6425 /******************************************************************************
6426 * function: imsm_create_metadata_update_for_reshape
6427 * Function creates update for whole IMSM container.
6428 *
6429 ******************************************************************************/
6430 static int imsm_create_metadata_update_for_reshape(
6431 struct supertype *st,
6432 struct geo_params *geo,
6433 int old_raid_disks,
6434 struct imsm_update_reshape **updatep)
6435 {
6436 struct intel_super *super = st->sb;
6437 struct imsm_super *mpb = super->anchor;
6438 int update_memory_size = 0;
6439 struct imsm_update_reshape *u = NULL;
6440 struct mdinfo *spares = NULL;
6441 int i;
6442 int delta_disks = 0;
6443 struct mdinfo *dev;
6444
6445 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
6446 geo->raid_disks);
6447
6448 delta_disks = geo->raid_disks - old_raid_disks;
6449
6450 /* size of all update data without anchor */
6451 update_memory_size = sizeof(struct imsm_update_reshape);
6452
6453 /* now add space for spare disks that we need to add. */
6454 update_memory_size += sizeof(u->new_disks[0]) * (delta_disks - 1);
6455
6456 u = calloc(1, update_memory_size);
6457 if (u == NULL) {
6458 dprintf("error: "
6459 "cannot get memory for imsm_update_reshape update\n");
6460 return 0;
6461 }
6462 u->type = update_reshape_container_disks;
6463 u->old_raid_disks = old_raid_disks;
6464 u->new_raid_disks = geo->raid_disks;
6465
6466 /* now get spare disks list
6467 */
6468 spares = get_spares_for_grow(st);
6469
6470 if (spares == NULL
6471 || delta_disks > spares->array.spare_disks) {
6472 dprintf("imsm: ERROR: Cannot get spare devices.\n");
6473 goto abort;
6474 }
6475
6476 /* we have got spares
6477 * update disk list in imsm_disk list table in anchor
6478 */
6479 dprintf("imsm: %i spares are available.\n\n",
6480 spares->array.spare_disks);
6481
6482 dev = spares->devs;
6483 for (i = 0; i < delta_disks; i++) {
6484 struct dl *dl;
6485
6486 if (dev == NULL)
6487 break;
6488 u->new_disks[i] = makedev(dev->disk.major,
6489 dev->disk.minor);
6490 dl = get_disk_super(super, dev->disk.major, dev->disk.minor);
6491 dl->index = mpb->num_disks;
6492 mpb->num_disks++;
6493 dev = dev->next;
6494 }
6495 /* Now update the metadata so that container_content will find
6496 * the new devices
6497 */
6498 for (i = 0; i < mpb->num_raid_devs; i++) {
6499 int d;
6500 struct imsm_dev *dev = get_imsm_dev(super, i);
6501 struct imsm_map *map = get_imsm_map(dev, 0);
6502 map->num_members = geo->raid_disks;
6503 for (d = 0; d < delta_disks; d++) {
6504 set_imsm_ord_tbl_ent(map, old_raid_disks + d,
6505 mpb->num_disks - delta_disks + d);
6506 }
6507 }
6508
6509 abort:
6510 /* free spares
6511 */
6512 sysfs_free(spares);
6513
6514 dprintf("imsm: reshape update preparation :");
6515 if (i == delta_disks) {
6516 dprintf(" OK\n");
6517 *updatep = u;
6518 return update_memory_size;
6519 }
6520 free(u);
6521 dprintf(" Error\n");
6522
6523 return 0;
6524 }
6525
6526 static void imsm_update_metadata_locally(struct supertype *st,
6527 void *buf, int len)
6528 {
6529 struct metadata_update mu;
6530
6531 mu.buf = buf;
6532 mu.len = len;
6533 mu.space = NULL;
6534 mu.space_list = NULL;
6535 mu.next = NULL;
6536 imsm_prepare_update(st, &mu);
6537 imsm_process_update(st, &mu);
6538
6539 while (mu.space_list) {
6540 void **space = mu.space_list;
6541 mu.space_list = *space;
6542 free(space);
6543 }
6544 }
6545
6546 static int imsm_reshape_super(struct supertype *st, long long size, int level,
6547 int layout, int chunksize, int raid_disks,
6548 char *backup, char *dev, int verbose)
6549 {
6550 /* currently we only support increasing the number of devices
6551 * for a container. This increases the number of device for each
6552 * member array. They must all be RAID0 or RAID5.
6553 */
6554
6555 int ret_val = 1;
6556 struct geo_params geo;
6557
6558 dprintf("imsm: reshape_super called.\n");
6559
6560 memset(&geo, sizeof(struct geo_params), 0);
6561
6562 geo.dev_name = dev;
6563 geo.size = size;
6564 geo.level = level;
6565 geo.layout = layout;
6566 geo.chunksize = chunksize;
6567 geo.raid_disks = raid_disks;
6568
6569 dprintf("\tfor level : %i\n", geo.level);
6570 dprintf("\tfor raid_disks : %i\n", geo.raid_disks);
6571
6572 if (experimental() == 0)
6573 return ret_val;
6574
6575 /* verify reshape conditions
6576 * on container level we can only increase number of devices.
6577 */
6578 if (st->container_dev == st->devnum) {
6579 /* check for delta_disks > 0
6580 * and supported raid levels 0 and 5 only in container
6581 */
6582 int old_raid_disks = 0;
6583 if (imsm_reshape_is_allowed_on_container(
6584 st, &geo, &old_raid_disks)) {
6585 struct imsm_update_reshape *u = NULL;
6586 int len;
6587
6588 len = imsm_create_metadata_update_for_reshape(
6589 st, &geo, old_raid_disks, &u);
6590
6591 if (len <= 0) {
6592 dprintf("imsm: Cannot prepare update\n");
6593 goto exit_imsm_reshape_super;
6594 }
6595
6596 ret_val = 0;
6597 /* update metadata locally */
6598 imsm_update_metadata_locally(st, u, len);
6599 /* and possibly remotely */
6600 if (st->update_tail)
6601 append_metadata_update(st, u, len);
6602 else
6603 free(u);
6604
6605 } else
6606 fprintf(stderr, Name "imsm: Operation is not allowed "
6607 "on this container\n");
6608 } else
6609 fprintf(stderr, Name "imsm: not a container operation\n");
6610
6611 exit_imsm_reshape_super:
6612 dprintf("imsm: reshape_super Exit code = %i\n", ret_val);
6613 return ret_val;
6614 }
6615
6616 static int imsm_manage_reshape(
6617 int afd, struct mdinfo *sra, struct reshape *reshape,
6618 struct supertype *st, unsigned long stripes,
6619 int *fds, unsigned long long *offsets,
6620 int dests, int *destfd, unsigned long long *destoffsets)
6621 {
6622 /* Just use child_monitor for now */
6623 return child_monitor(
6624 afd, sra, reshape, st, stripes,
6625 fds, offsets, dests, destfd, destoffsets);
6626 }
6627
6628 struct superswitch super_imsm = {
6629 #ifndef MDASSEMBLE
6630 .examine_super = examine_super_imsm,
6631 .brief_examine_super = brief_examine_super_imsm,
6632 .brief_examine_subarrays = brief_examine_subarrays_imsm,
6633 .export_examine_super = export_examine_super_imsm,
6634 .detail_super = detail_super_imsm,
6635 .brief_detail_super = brief_detail_super_imsm,
6636 .write_init_super = write_init_super_imsm,
6637 .validate_geometry = validate_geometry_imsm,
6638 .add_to_super = add_to_super_imsm,
6639 .remove_from_super = remove_from_super_imsm,
6640 .detail_platform = detail_platform_imsm,
6641 .kill_subarray = kill_subarray_imsm,
6642 .update_subarray = update_subarray_imsm,
6643 .load_container = load_container_imsm,
6644 #endif
6645 .match_home = match_home_imsm,
6646 .uuid_from_super= uuid_from_super_imsm,
6647 .getinfo_super = getinfo_super_imsm,
6648 .getinfo_super_disks = getinfo_super_disks_imsm,
6649 .update_super = update_super_imsm,
6650
6651 .avail_size = avail_size_imsm,
6652 .min_acceptable_spare_size = min_acceptable_spare_size_imsm,
6653
6654 .compare_super = compare_super_imsm,
6655
6656 .load_super = load_super_imsm,
6657 .init_super = init_super_imsm,
6658 .store_super = store_super_imsm,
6659 .free_super = free_super_imsm,
6660 .match_metadata_desc = match_metadata_desc_imsm,
6661 .container_content = container_content_imsm,
6662 .default_geometry = default_geometry_imsm,
6663 .get_disk_controller_domain = imsm_get_disk_controller_domain,
6664 .reshape_super = imsm_reshape_super,
6665 .manage_reshape = imsm_manage_reshape,
6666
6667 .external = 1,
6668 .name = "imsm",
6669
6670 #ifndef MDASSEMBLE
6671 /* for mdmon */
6672 .open_new = imsm_open_new,
6673 .set_array_state= imsm_set_array_state,
6674 .set_disk = imsm_set_disk,
6675 .sync_metadata = imsm_sync_metadata,
6676 .activate_spare = imsm_activate_spare,
6677 .process_update = imsm_process_update,
6678 .prepare_update = imsm_prepare_update,
6679 #endif /* MDASSEMBLE */
6680 };
Unnamed repository; edit this file 'description' to name the repository.