]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-intel.c
Assemble(): don't dup_super() before we need it.
[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 /* supports RAID0 */
45 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
46 /* supports RAID1 */
47 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
48 /* supports RAID10 */
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 /* supports RAID1E */
51 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
52 /* supports RAID5 */
53 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
54 /* supports RAID CNG */
55 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
56 /* supports expanded stripe sizes of 256K, 512K and 1MB */
57 #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040)
58
59 /* The OROM Support RST Caching of Volumes */
60 #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000)
61 /* The OROM supports creating disks greater than 2TB */
62 #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000)
63 /* The OROM supports Bad Block Management */
64 #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000)
65
66 /* THe OROM Supports NVM Caching of Volumes */
67 #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000)
68 /* The OROM supports creating volumes greater than 2TB */
69 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
70 /* originally for PMP, now it's wasted b/c. Never use this bit! */
71 #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000)
72 /* Verify MPB contents against checksum after reading MPB */
73 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
74
75 /* Define all supported attributes that have to be accepted by mdadm
76 */
77 #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \
78 MPB_ATTRIB_2TB | \
79 MPB_ATTRIB_2TB_DISK | \
80 MPB_ATTRIB_RAID0 | \
81 MPB_ATTRIB_RAID1 | \
82 MPB_ATTRIB_RAID10 | \
83 MPB_ATTRIB_RAID5 | \
84 MPB_ATTRIB_EXP_STRIPE_SIZE)
85
86 /* Define attributes that are unused but not harmful */
87 #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE)
88
89 #define MPB_SECTOR_CNT 2210
90 #define IMSM_RESERVED_SECTORS 4096
91 #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056
92 #define SECT_PER_MB_SHIFT 11
93
94 /* Disk configuration info. */
95 #define IMSM_MAX_DEVICES 255
96 struct imsm_disk {
97 __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
98 __u32 total_blocks; /* 0xE8 - 0xEB total blocks */
99 __u32 scsi_id; /* 0xEC - 0xEF scsi ID */
100 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
101 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
102 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
103 __u32 status; /* 0xF0 - 0xF3 */
104 __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
105 #define IMSM_DISK_FILLERS 4
106 __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
107 };
108
109 /* RAID map configuration infos. */
110 struct imsm_map {
111 __u32 pba_of_lba0; /* start address of partition */
112 __u32 blocks_per_member;/* blocks per member */
113 __u32 num_data_stripes; /* number of data stripes */
114 __u16 blocks_per_strip;
115 __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */
116 #define IMSM_T_STATE_NORMAL 0
117 #define IMSM_T_STATE_UNINITIALIZED 1
118 #define IMSM_T_STATE_DEGRADED 2
119 #define IMSM_T_STATE_FAILED 3
120 __u8 raid_level;
121 #define IMSM_T_RAID0 0
122 #define IMSM_T_RAID1 1
123 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
124 __u8 num_members; /* number of member disks */
125 __u8 num_domains; /* number of parity domains */
126 __u8 failed_disk_num; /* valid only when state is degraded */
127 __u8 ddf;
128 __u32 filler[7]; /* expansion area */
129 #define IMSM_ORD_REBUILD (1 << 24)
130 __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
131 * top byte contains some flags
132 */
133 } __attribute__ ((packed));
134
135 struct imsm_vol {
136 __u32 curr_migr_unit;
137 __u32 checkpoint_id; /* id to access curr_migr_unit */
138 __u8 migr_state; /* Normal or Migrating */
139 #define MIGR_INIT 0
140 #define MIGR_REBUILD 1
141 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
142 #define MIGR_GEN_MIGR 3
143 #define MIGR_STATE_CHANGE 4
144 #define MIGR_REPAIR 5
145 __u8 migr_type; /* Initializing, Rebuilding, ... */
146 __u8 dirty;
147 __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */
148 __u16 verify_errors; /* number of mismatches */
149 __u16 bad_blocks; /* number of bad blocks during verify */
150 __u32 filler[4];
151 struct imsm_map map[1];
152 /* here comes another one if migr_state */
153 } __attribute__ ((packed));
154
155 struct imsm_dev {
156 __u8 volume[MAX_RAID_SERIAL_LEN];
157 __u32 size_low;
158 __u32 size_high;
159 #define DEV_BOOTABLE __cpu_to_le32(0x01)
160 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
161 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
162 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
163 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
164 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
165 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
166 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
167 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
168 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
169 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
170 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
171 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
172 __u32 status; /* Persistent RaidDev status */
173 __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
174 __u8 migr_priority;
175 __u8 num_sub_vols;
176 __u8 tid;
177 __u8 cng_master_disk;
178 __u16 cache_policy;
179 __u8 cng_state;
180 __u8 cng_sub_state;
181 #define IMSM_DEV_FILLERS 10
182 __u32 filler[IMSM_DEV_FILLERS];
183 struct imsm_vol vol;
184 } __attribute__ ((packed));
185
186 struct imsm_super {
187 __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
188 __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */
189 __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */
190 __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */
191 __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
192 __u32 error_log_size; /* 0x30 - 0x33 in bytes */
193 __u32 attributes; /* 0x34 - 0x37 */
194 __u8 num_disks; /* 0x38 Number of configured disks */
195 __u8 num_raid_devs; /* 0x39 Number of configured volumes */
196 __u8 error_log_pos; /* 0x3A */
197 __u8 fill[1]; /* 0x3B */
198 __u32 cache_size; /* 0x3c - 0x40 in mb */
199 __u32 orig_family_num; /* 0x40 - 0x43 original family num */
200 __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */
201 __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
202 #define IMSM_FILLERS 35
203 __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
204 struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */
205 /* here comes imsm_dev[num_raid_devs] */
206 /* here comes BBM logs */
207 } __attribute__ ((packed));
208
209 #define BBM_LOG_MAX_ENTRIES 254
210
211 struct bbm_log_entry {
212 __u64 defective_block_start;
213 #define UNREADABLE 0xFFFFFFFF
214 __u32 spare_block_offset;
215 __u16 remapped_marked_count;
216 __u16 disk_ordinal;
217 } __attribute__ ((__packed__));
218
219 struct bbm_log {
220 __u32 signature; /* 0xABADB10C */
221 __u32 entry_count;
222 __u32 reserved_spare_block_count; /* 0 */
223 __u32 reserved; /* 0xFFFF */
224 __u64 first_spare_lba;
225 struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
226 } __attribute__ ((__packed__));
227
228
229 #ifndef MDASSEMBLE
230 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
231 #endif
232
233 #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209
234
235 #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */
236
237 #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must
238 * be recovered using srcMap */
239 #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has
240 * already been migrated and must
241 * be recovered from checkpoint area */
242 struct migr_record {
243 __u32 rec_status; /* Status used to determine how to restart
244 * migration in case it aborts
245 * in some fashion */
246 __u32 curr_migr_unit; /* 0..numMigrUnits-1 */
247 __u32 family_num; /* Family number of MPB
248 * containing the RaidDev
249 * that is migrating */
250 __u32 ascending_migr; /* True if migrating in increasing
251 * order of lbas */
252 __u32 blocks_per_unit; /* Num disk blocks per unit of operation */
253 __u32 dest_depth_per_unit; /* Num member blocks each destMap
254 * member disk
255 * advances per unit-of-operation */
256 __u32 ckpt_area_pba; /* Pba of first block of ckpt copy area */
257 __u32 dest_1st_member_lba; /* First member lba on first
258 * stripe of destination */
259 __u32 num_migr_units; /* Total num migration units-of-op */
260 __u32 post_migr_vol_cap; /* Size of volume after
261 * migration completes */
262 __u32 post_migr_vol_cap_hi; /* Expansion space for LBA64 */
263 __u32 ckpt_read_disk_num; /* Which member disk in destSubMap[0] the
264 * migration ckpt record was read from
265 * (for recovered migrations) */
266 } __attribute__ ((__packed__));
267
268 static __u8 migr_type(struct imsm_dev *dev)
269 {
270 if (dev->vol.migr_type == MIGR_VERIFY &&
271 dev->status & DEV_VERIFY_AND_FIX)
272 return MIGR_REPAIR;
273 else
274 return dev->vol.migr_type;
275 }
276
277 static void set_migr_type(struct imsm_dev *dev, __u8 migr_type)
278 {
279 /* for compatibility with older oroms convert MIGR_REPAIR, into
280 * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status
281 */
282 if (migr_type == MIGR_REPAIR) {
283 dev->vol.migr_type = MIGR_VERIFY;
284 dev->status |= DEV_VERIFY_AND_FIX;
285 } else {
286 dev->vol.migr_type = migr_type;
287 dev->status &= ~DEV_VERIFY_AND_FIX;
288 }
289 }
290
291 static unsigned int sector_count(__u32 bytes)
292 {
293 return ((bytes + (512-1)) & (~(512-1))) / 512;
294 }
295
296 static unsigned int mpb_sectors(struct imsm_super *mpb)
297 {
298 return sector_count(__le32_to_cpu(mpb->mpb_size));
299 }
300
301 struct intel_dev {
302 struct imsm_dev *dev;
303 struct intel_dev *next;
304 unsigned index;
305 };
306
307 struct intel_hba {
308 enum sys_dev_type type;
309 char *path;
310 char *pci_id;
311 struct intel_hba *next;
312 };
313
314 enum action {
315 DISK_REMOVE = 1,
316 DISK_ADD
317 };
318 /* internal representation of IMSM metadata */
319 struct intel_super {
320 union {
321 void *buf; /* O_DIRECT buffer for reading/writing metadata */
322 struct imsm_super *anchor; /* immovable parameters */
323 };
324 union {
325 void *migr_rec_buf; /* buffer for I/O operations */
326 struct migr_record *migr_rec; /* migration record */
327 };
328 size_t len; /* size of the 'buf' allocation */
329 void *next_buf; /* for realloc'ing buf from the manager */
330 size_t next_len;
331 int updates_pending; /* count of pending updates for mdmon */
332 int current_vol; /* index of raid device undergoing creation */
333 __u32 create_offset; /* common start for 'current_vol' */
334 __u32 random; /* random data for seeding new family numbers */
335 struct intel_dev *devlist;
336 struct dl {
337 struct dl *next;
338 int index;
339 __u8 serial[MAX_RAID_SERIAL_LEN];
340 int major, minor;
341 char *devname;
342 struct imsm_disk disk;
343 int fd;
344 int extent_cnt;
345 struct extent *e; /* for determining freespace @ create */
346 int raiddisk; /* slot to fill in autolayout */
347 enum action action;
348 } *disks, *current_disk;
349 struct dl *disk_mgmt_list; /* list of disks to add/remove while mdmon
350 active */
351 struct dl *missing; /* disks removed while we weren't looking */
352 struct bbm_log *bbm_log;
353 struct intel_hba *hba; /* device path of the raid controller for this metadata */
354 const struct imsm_orom *orom; /* platform firmware support */
355 struct intel_super *next; /* (temp) list for disambiguating family_num */
356 };
357
358 struct intel_disk {
359 struct imsm_disk disk;
360 #define IMSM_UNKNOWN_OWNER (-1)
361 int owner;
362 struct intel_disk *next;
363 };
364
365 struct extent {
366 unsigned long long start, size;
367 };
368
369 /* definitions of reshape process types */
370 enum imsm_reshape_type {
371 CH_TAKEOVER,
372 CH_MIGRATION,
373 };
374
375 /* definition of messages passed to imsm_process_update */
376 enum imsm_update_type {
377 update_activate_spare,
378 update_create_array,
379 update_kill_array,
380 update_rename_array,
381 update_add_remove_disk,
382 update_reshape_container_disks,
383 update_reshape_migration,
384 update_takeover,
385 update_general_migration_checkpoint,
386 };
387
388 struct imsm_update_activate_spare {
389 enum imsm_update_type type;
390 struct dl *dl;
391 int slot;
392 int array;
393 struct imsm_update_activate_spare *next;
394 };
395
396 struct geo_params {
397 int dev_id;
398 char *dev_name;
399 long long size;
400 int level;
401 int layout;
402 int chunksize;
403 int raid_disks;
404 };
405
406 enum takeover_direction {
407 R10_TO_R0,
408 R0_TO_R10
409 };
410 struct imsm_update_takeover {
411 enum imsm_update_type type;
412 int subarray;
413 enum takeover_direction direction;
414 };
415
416 struct imsm_update_reshape {
417 enum imsm_update_type type;
418 int old_raid_disks;
419 int new_raid_disks;
420
421 int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */
422 };
423
424 struct imsm_update_reshape_migration {
425 enum imsm_update_type type;
426 int old_raid_disks;
427 int new_raid_disks;
428 /* fields for array migration changes
429 */
430 int subdev;
431 int new_level;
432 int new_layout;
433 int new_chunksize;
434
435 int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */
436 };
437
438 struct imsm_update_general_migration_checkpoint {
439 enum imsm_update_type type;
440 __u32 curr_migr_unit;
441 };
442
443 struct disk_info {
444 __u8 serial[MAX_RAID_SERIAL_LEN];
445 };
446
447 struct imsm_update_create_array {
448 enum imsm_update_type type;
449 int dev_idx;
450 struct imsm_dev dev;
451 };
452
453 struct imsm_update_kill_array {
454 enum imsm_update_type type;
455 int dev_idx;
456 };
457
458 struct imsm_update_rename_array {
459 enum imsm_update_type type;
460 __u8 name[MAX_RAID_SERIAL_LEN];
461 int dev_idx;
462 };
463
464 struct imsm_update_add_remove_disk {
465 enum imsm_update_type type;
466 };
467
468
469 static const char *_sys_dev_type[] = {
470 [SYS_DEV_UNKNOWN] = "Unknown",
471 [SYS_DEV_SAS] = "SAS",
472 [SYS_DEV_SATA] = "SATA"
473 };
474
475 const char *get_sys_dev_type(enum sys_dev_type type)
476 {
477 if (type >= SYS_DEV_MAX)
478 type = SYS_DEV_UNKNOWN;
479
480 return _sys_dev_type[type];
481 }
482
483 static struct intel_hba * alloc_intel_hba(struct sys_dev *device)
484 {
485 struct intel_hba *result = malloc(sizeof(*result));
486 if (result) {
487 result->type = device->type;
488 result->path = strdup(device->path);
489 result->next = NULL;
490 if (result->path && (result->pci_id = strrchr(result->path, '/')) != NULL)
491 result->pci_id++;
492 }
493 return result;
494 }
495
496 static struct intel_hba * find_intel_hba(struct intel_hba *hba, struct sys_dev *device)
497 {
498 struct intel_hba *result=NULL;
499 for (result = hba; result; result = result->next) {
500 if (result->type == device->type && strcmp(result->path, device->path) == 0)
501 break;
502 }
503 return result;
504 }
505
506 static int attach_hba_to_super(struct intel_super *super, struct sys_dev *device)
507 {
508 struct intel_hba *hba;
509
510 /* check if disk attached to Intel HBA */
511 hba = find_intel_hba(super->hba, device);
512 if (hba != NULL)
513 return 1;
514 /* Check if HBA is already attached to super */
515 if (super->hba == NULL) {
516 super->hba = alloc_intel_hba(device);
517 return 1;
518 }
519
520 hba = super->hba;
521 /* Intel metadata allows for all disks attached to the same type HBA.
522 * Do not sypport odf HBA types mixing
523 */
524 if (device->type != hba->type)
525 return 2;
526
527 while (hba->next)
528 hba = hba->next;
529
530 hba->next = alloc_intel_hba(device);
531 return 1;
532 }
533
534 static struct sys_dev* find_disk_attached_hba(int fd, const char *devname)
535 {
536 struct sys_dev *list, *elem, *prev;
537 char *disk_path;
538
539 if ((list = find_intel_devices()) == NULL)
540 return 0;
541
542 if (fd < 0)
543 disk_path = (char *) devname;
544 else
545 disk_path = diskfd_to_devpath(fd);
546
547 if (!disk_path) {
548 free_sys_dev(&list);
549 return 0;
550 }
551
552 for (prev = NULL, elem = list; elem; prev = elem, elem = elem->next) {
553 if (path_attached_to_hba(disk_path, elem->path)) {
554 if (prev == NULL)
555 list = list->next;
556 else
557 prev->next = elem->next;
558 elem->next = NULL;
559 if (disk_path != devname)
560 free(disk_path);
561 free_sys_dev(&list);
562 return elem;
563 }
564 }
565 if (disk_path != devname)
566 free(disk_path);
567 free_sys_dev(&list);
568
569 return NULL;
570 }
571
572
573 static int find_intel_hba_capability(int fd, struct intel_super *super,
574 char *devname);
575
576 static struct supertype *match_metadata_desc_imsm(char *arg)
577 {
578 struct supertype *st;
579
580 if (strcmp(arg, "imsm") != 0 &&
581 strcmp(arg, "default") != 0
582 )
583 return NULL;
584
585 st = malloc(sizeof(*st));
586 if (!st)
587 return NULL;
588 memset(st, 0, sizeof(*st));
589 st->container_dev = NoMdDev;
590 st->ss = &super_imsm;
591 st->max_devs = IMSM_MAX_DEVICES;
592 st->minor_version = 0;
593 st->sb = NULL;
594 return st;
595 }
596
597 #ifndef MDASSEMBLE
598 static __u8 *get_imsm_version(struct imsm_super *mpb)
599 {
600 return &mpb->sig[MPB_SIG_LEN];
601 }
602 #endif
603
604 /* retrieve a disk directly from the anchor when the anchor is known to be
605 * up-to-date, currently only at load time
606 */
607 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
608 {
609 if (index >= mpb->num_disks)
610 return NULL;
611 return &mpb->disk[index];
612 }
613
614 /* retrieve the disk description based on a index of the disk
615 * in the sub-array
616 */
617 static struct dl *get_imsm_dl_disk(struct intel_super *super, __u8 index)
618 {
619 struct dl *d;
620
621 for (d = super->disks; d; d = d->next)
622 if (d->index == index)
623 return d;
624
625 return NULL;
626 }
627 /* retrieve a disk from the parsed metadata */
628 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
629 {
630 struct dl *dl;
631
632 dl = get_imsm_dl_disk(super, index);
633 if (dl)
634 return &dl->disk;
635
636 return NULL;
637 }
638
639 /* generate a checksum directly from the anchor when the anchor is known to be
640 * up-to-date, currently only at load or write_super after coalescing
641 */
642 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
643 {
644 __u32 end = mpb->mpb_size / sizeof(end);
645 __u32 *p = (__u32 *) mpb;
646 __u32 sum = 0;
647
648 while (end--) {
649 sum += __le32_to_cpu(*p);
650 p++;
651 }
652
653 return sum - __le32_to_cpu(mpb->check_sum);
654 }
655
656 static size_t sizeof_imsm_map(struct imsm_map *map)
657 {
658 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
659 }
660
661 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
662 {
663 /* A device can have 2 maps if it is in the middle of a migration.
664 * If second_map is:
665 * 0 - we return the first map
666 * 1 - we return the second map if it exists, else NULL
667 * -1 - we return the second map if it exists, else the first
668 */
669 struct imsm_map *map = &dev->vol.map[0];
670
671 if (second_map == 1 && !dev->vol.migr_state)
672 return NULL;
673 else if (second_map == 1 ||
674 (second_map < 0 && dev->vol.migr_state)) {
675 void *ptr = map;
676
677 return ptr + sizeof_imsm_map(map);
678 } else
679 return map;
680
681 }
682
683 /* return the size of the device.
684 * migr_state increases the returned size if map[0] were to be duplicated
685 */
686 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
687 {
688 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
689 sizeof_imsm_map(get_imsm_map(dev, 0));
690
691 /* migrating means an additional map */
692 if (dev->vol.migr_state)
693 size += sizeof_imsm_map(get_imsm_map(dev, 1));
694 else if (migr_state)
695 size += sizeof_imsm_map(get_imsm_map(dev, 0));
696
697 return size;
698 }
699
700 #ifndef MDASSEMBLE
701 /* retrieve disk serial number list from a metadata update */
702 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
703 {
704 void *u = update;
705 struct disk_info *inf;
706
707 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
708 sizeof_imsm_dev(&update->dev, 0);
709
710 return inf;
711 }
712 #endif
713
714 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
715 {
716 int offset;
717 int i;
718 void *_mpb = mpb;
719
720 if (index >= mpb->num_raid_devs)
721 return NULL;
722
723 /* devices start after all disks */
724 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
725
726 for (i = 0; i <= index; i++)
727 if (i == index)
728 return _mpb + offset;
729 else
730 offset += sizeof_imsm_dev(_mpb + offset, 0);
731
732 return NULL;
733 }
734
735 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
736 {
737 struct intel_dev *dv;
738
739 if (index >= super->anchor->num_raid_devs)
740 return NULL;
741 for (dv = super->devlist; dv; dv = dv->next)
742 if (dv->index == index)
743 return dv->dev;
744 return NULL;
745 }
746
747 /*
748 * for second_map:
749 * == 0 get first map
750 * == 1 get second map
751 * == -1 than get map according to the current migr_state
752 */
753 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev,
754 int slot,
755 int second_map)
756 {
757 struct imsm_map *map;
758
759 map = get_imsm_map(dev, second_map);
760
761 /* top byte identifies disk under rebuild */
762 return __le32_to_cpu(map->disk_ord_tbl[slot]);
763 }
764
765 #define ord_to_idx(ord) (((ord) << 8) >> 8)
766 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot, int second_map)
767 {
768 __u32 ord = get_imsm_ord_tbl_ent(dev, slot, second_map);
769
770 return ord_to_idx(ord);
771 }
772
773 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
774 {
775 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
776 }
777
778 static int get_imsm_disk_slot(struct imsm_map *map, unsigned idx)
779 {
780 int slot;
781 __u32 ord;
782
783 for (slot = 0; slot < map->num_members; slot++) {
784 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
785 if (ord_to_idx(ord) == idx)
786 return slot;
787 }
788
789 return -1;
790 }
791
792 static int get_imsm_raid_level(struct imsm_map *map)
793 {
794 if (map->raid_level == 1) {
795 if (map->num_members == 2)
796 return 1;
797 else
798 return 10;
799 }
800
801 return map->raid_level;
802 }
803
804 static int cmp_extent(const void *av, const void *bv)
805 {
806 const struct extent *a = av;
807 const struct extent *b = bv;
808 if (a->start < b->start)
809 return -1;
810 if (a->start > b->start)
811 return 1;
812 return 0;
813 }
814
815 static int count_memberships(struct dl *dl, struct intel_super *super)
816 {
817 int memberships = 0;
818 int i;
819
820 for (i = 0; i < super->anchor->num_raid_devs; i++) {
821 struct imsm_dev *dev = get_imsm_dev(super, i);
822 struct imsm_map *map = get_imsm_map(dev, 0);
823
824 if (get_imsm_disk_slot(map, dl->index) >= 0)
825 memberships++;
826 }
827
828 return memberships;
829 }
830
831 static __u32 imsm_min_reserved_sectors(struct intel_super *super);
832
833 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
834 {
835 /* find a list of used extents on the given physical device */
836 struct extent *rv, *e;
837 int i;
838 int memberships = count_memberships(dl, super);
839 __u32 reservation;
840
841 /* trim the reserved area for spares, so they can join any array
842 * regardless of whether the OROM has assigned sectors from the
843 * IMSM_RESERVED_SECTORS region
844 */
845 if (dl->index == -1)
846 reservation = imsm_min_reserved_sectors(super);
847 else
848 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
849
850 rv = malloc(sizeof(struct extent) * (memberships + 1));
851 if (!rv)
852 return NULL;
853 e = rv;
854
855 for (i = 0; i < super->anchor->num_raid_devs; i++) {
856 struct imsm_dev *dev = get_imsm_dev(super, i);
857 struct imsm_map *map = get_imsm_map(dev, 0);
858
859 if (get_imsm_disk_slot(map, dl->index) >= 0) {
860 e->start = __le32_to_cpu(map->pba_of_lba0);
861 e->size = __le32_to_cpu(map->blocks_per_member);
862 e++;
863 }
864 }
865 qsort(rv, memberships, sizeof(*rv), cmp_extent);
866
867 /* determine the start of the metadata
868 * when no raid devices are defined use the default
869 * ...otherwise allow the metadata to truncate the value
870 * as is the case with older versions of imsm
871 */
872 if (memberships) {
873 struct extent *last = &rv[memberships - 1];
874 __u32 remainder;
875
876 remainder = __le32_to_cpu(dl->disk.total_blocks) -
877 (last->start + last->size);
878 /* round down to 1k block to satisfy precision of the kernel
879 * 'size' interface
880 */
881 remainder &= ~1UL;
882 /* make sure remainder is still sane */
883 if (remainder < (unsigned)ROUND_UP(super->len, 512) >> 9)
884 remainder = ROUND_UP(super->len, 512) >> 9;
885 if (reservation > remainder)
886 reservation = remainder;
887 }
888 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
889 e->size = 0;
890 return rv;
891 }
892
893 /* try to determine how much space is reserved for metadata from
894 * the last get_extents() entry, otherwise fallback to the
895 * default
896 */
897 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
898 {
899 struct extent *e;
900 int i;
901 __u32 rv;
902
903 /* for spares just return a minimal reservation which will grow
904 * once the spare is picked up by an array
905 */
906 if (dl->index == -1)
907 return MPB_SECTOR_CNT;
908
909 e = get_extents(super, dl);
910 if (!e)
911 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
912
913 /* scroll to last entry */
914 for (i = 0; e[i].size; i++)
915 continue;
916
917 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
918
919 free(e);
920
921 return rv;
922 }
923
924 static int is_spare(struct imsm_disk *disk)
925 {
926 return (disk->status & SPARE_DISK) == SPARE_DISK;
927 }
928
929 static int is_configured(struct imsm_disk *disk)
930 {
931 return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK;
932 }
933
934 static int is_failed(struct imsm_disk *disk)
935 {
936 return (disk->status & FAILED_DISK) == FAILED_DISK;
937 }
938
939 /* try to determine how much space is reserved for metadata from
940 * the last get_extents() entry on the smallest active disk,
941 * otherwise fallback to the default
942 */
943 static __u32 imsm_min_reserved_sectors(struct intel_super *super)
944 {
945 struct extent *e;
946 int i;
947 __u32 min_active, remainder;
948 __u32 rv = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
949 struct dl *dl, *dl_min = NULL;
950
951 if (!super)
952 return rv;
953
954 min_active = 0;
955 for (dl = super->disks; dl; dl = dl->next) {
956 if (dl->index < 0)
957 continue;
958 if (dl->disk.total_blocks < min_active || min_active == 0) {
959 dl_min = dl;
960 min_active = dl->disk.total_blocks;
961 }
962 }
963 if (!dl_min)
964 return rv;
965
966 /* find last lba used by subarrays on the smallest active disk */
967 e = get_extents(super, dl_min);
968 if (!e)
969 return rv;
970 for (i = 0; e[i].size; i++)
971 continue;
972
973 remainder = min_active - e[i].start;
974 free(e);
975
976 /* to give priority to recovery we should not require full
977 IMSM_RESERVED_SECTORS from the spare */
978 rv = MPB_SECTOR_CNT + NUM_BLOCKS_DIRTY_STRIPE_REGION;
979
980 /* if real reservation is smaller use that value */
981 return (remainder < rv) ? remainder : rv;
982 }
983
984 /* Return minimum size of a spare that can be used in this array*/
985 static unsigned long long min_acceptable_spare_size_imsm(struct supertype *st)
986 {
987 struct intel_super *super = st->sb;
988 struct dl *dl;
989 struct extent *e;
990 int i;
991 unsigned long long rv = 0;
992
993 if (!super)
994 return rv;
995 /* find first active disk in array */
996 dl = super->disks;
997 while (dl && (is_failed(&dl->disk) || dl->index == -1))
998 dl = dl->next;
999 if (!dl)
1000 return rv;
1001 /* find last lba used by subarrays */
1002 e = get_extents(super, dl);
1003 if (!e)
1004 return rv;
1005 for (i = 0; e[i].size; i++)
1006 continue;
1007 if (i > 0)
1008 rv = e[i-1].start + e[i-1].size;
1009 free(e);
1010
1011 /* add the amount of space needed for metadata */
1012 rv = rv + imsm_min_reserved_sectors(super);
1013
1014 return rv * 512;
1015 }
1016
1017 #ifndef MDASSEMBLE
1018 static __u64 blocks_per_migr_unit(struct intel_super *super,
1019 struct imsm_dev *dev);
1020
1021 static void print_imsm_dev(struct intel_super *super,
1022 struct imsm_dev *dev,
1023 char *uuid,
1024 int disk_idx)
1025 {
1026 __u64 sz;
1027 int slot, i;
1028 struct imsm_map *map = get_imsm_map(dev, 0);
1029 struct imsm_map *map2 = get_imsm_map(dev, 1);
1030 __u32 ord;
1031
1032 printf("\n");
1033 printf("[%.16s]:\n", dev->volume);
1034 printf(" UUID : %s\n", uuid);
1035 printf(" RAID Level : %d", get_imsm_raid_level(map));
1036 if (map2)
1037 printf(" <-- %d", get_imsm_raid_level(map2));
1038 printf("\n");
1039 printf(" Members : %d", map->num_members);
1040 if (map2)
1041 printf(" <-- %d", map2->num_members);
1042 printf("\n");
1043 printf(" Slots : [");
1044 for (i = 0; i < map->num_members; i++) {
1045 ord = get_imsm_ord_tbl_ent(dev, i, 0);
1046 printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
1047 }
1048 printf("]");
1049 if (map2) {
1050 printf(" <-- [");
1051 for (i = 0; i < map2->num_members; i++) {
1052 ord = get_imsm_ord_tbl_ent(dev, i, 1);
1053 printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U");
1054 }
1055 printf("]");
1056 }
1057 printf("\n");
1058 printf(" Failed disk : ");
1059 if (map->failed_disk_num == 0xff)
1060 printf("none");
1061 else
1062 printf("%i", map->failed_disk_num);
1063 printf("\n");
1064 slot = get_imsm_disk_slot(map, disk_idx);
1065 if (slot >= 0) {
1066 ord = get_imsm_ord_tbl_ent(dev, slot, -1);
1067 printf(" This Slot : %d%s\n", slot,
1068 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
1069 } else
1070 printf(" This Slot : ?\n");
1071 sz = __le32_to_cpu(dev->size_high);
1072 sz <<= 32;
1073 sz += __le32_to_cpu(dev->size_low);
1074 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
1075 human_size(sz * 512));
1076 sz = __le32_to_cpu(map->blocks_per_member);
1077 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
1078 human_size(sz * 512));
1079 printf(" Sector Offset : %u\n",
1080 __le32_to_cpu(map->pba_of_lba0));
1081 printf(" Num Stripes : %u\n",
1082 __le32_to_cpu(map->num_data_stripes));
1083 printf(" Chunk Size : %u KiB",
1084 __le16_to_cpu(map->blocks_per_strip) / 2);
1085 if (map2)
1086 printf(" <-- %u KiB",
1087 __le16_to_cpu(map2->blocks_per_strip) / 2);
1088 printf("\n");
1089 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
1090 printf(" Migrate State : ");
1091 if (dev->vol.migr_state) {
1092 if (migr_type(dev) == MIGR_INIT)
1093 printf("initialize\n");
1094 else if (migr_type(dev) == MIGR_REBUILD)
1095 printf("rebuild\n");
1096 else if (migr_type(dev) == MIGR_VERIFY)
1097 printf("check\n");
1098 else if (migr_type(dev) == MIGR_GEN_MIGR)
1099 printf("general migration\n");
1100 else if (migr_type(dev) == MIGR_STATE_CHANGE)
1101 printf("state change\n");
1102 else if (migr_type(dev) == MIGR_REPAIR)
1103 printf("repair\n");
1104 else
1105 printf("<unknown:%d>\n", migr_type(dev));
1106 } else
1107 printf("idle\n");
1108 printf(" Map State : %s", map_state_str[map->map_state]);
1109 if (dev->vol.migr_state) {
1110 struct imsm_map *map = get_imsm_map(dev, 1);
1111
1112 printf(" <-- %s", map_state_str[map->map_state]);
1113 printf("\n Checkpoint : %u (%llu)",
1114 __le32_to_cpu(dev->vol.curr_migr_unit),
1115 (unsigned long long)blocks_per_migr_unit(super, dev));
1116 }
1117 printf("\n");
1118 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
1119 }
1120
1121 static void print_imsm_disk(struct imsm_disk *disk, int index, __u32 reserved)
1122 {
1123 char str[MAX_RAID_SERIAL_LEN + 1];
1124 __u64 sz;
1125
1126 if (index < -1 || !disk)
1127 return;
1128
1129 printf("\n");
1130 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
1131 if (index >= 0)
1132 printf(" Disk%02d Serial : %s\n", index, str);
1133 else
1134 printf(" Disk Serial : %s\n", str);
1135 printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "",
1136 is_configured(disk) ? " active" : "",
1137 is_failed(disk) ? " failed" : "");
1138 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
1139 sz = __le32_to_cpu(disk->total_blocks) - reserved;
1140 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
1141 human_size(sz * 512));
1142 }
1143
1144 static int is_gen_migration(struct imsm_dev *dev);
1145
1146 void examine_migr_rec_imsm(struct intel_super *super)
1147 {
1148 struct migr_record *migr_rec = super->migr_rec;
1149 struct imsm_super *mpb = super->anchor;
1150 int i;
1151
1152 for (i = 0; i < mpb->num_raid_devs; i++) {
1153 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
1154 if (is_gen_migration(dev) == 0)
1155 continue;
1156
1157 printf("\nMigration Record Information:");
1158 if (super->disks->index > 1) {
1159 printf(" Empty\n ");
1160 printf("Examine one of first two disks in array\n");
1161 break;
1162 }
1163 printf("\n Status : ");
1164 if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL)
1165 printf("Normal\n");
1166 else
1167 printf("Contains Data\n");
1168 printf(" Current Unit : %u\n",
1169 __le32_to_cpu(migr_rec->curr_migr_unit));
1170 printf(" Family : %u\n",
1171 __le32_to_cpu(migr_rec->family_num));
1172 printf(" Ascending : %u\n",
1173 __le32_to_cpu(migr_rec->ascending_migr));
1174 printf(" Blocks Per Unit : %u\n",
1175 __le32_to_cpu(migr_rec->blocks_per_unit));
1176 printf(" Dest. Depth Per Unit : %u\n",
1177 __le32_to_cpu(migr_rec->dest_depth_per_unit));
1178 printf(" Checkpoint Area pba : %u\n",
1179 __le32_to_cpu(migr_rec->ckpt_area_pba));
1180 printf(" First member lba : %u\n",
1181 __le32_to_cpu(migr_rec->dest_1st_member_lba));
1182 printf(" Total Number of Units : %u\n",
1183 __le32_to_cpu(migr_rec->num_migr_units));
1184 printf(" Size of volume : %u\n",
1185 __le32_to_cpu(migr_rec->post_migr_vol_cap));
1186 printf(" Expansion space for LBA64 : %u\n",
1187 __le32_to_cpu(migr_rec->post_migr_vol_cap_hi));
1188 printf(" Record was read from : %u\n",
1189 __le32_to_cpu(migr_rec->ckpt_read_disk_num));
1190
1191 break;
1192 }
1193 }
1194 #endif /* MDASSEMBLE */
1195 /*******************************************************************************
1196 * function: imsm_check_attributes
1197 * Description: Function checks if features represented by attributes flags
1198 * are supported by mdadm.
1199 * Parameters:
1200 * attributes - Attributes read from metadata
1201 * Returns:
1202 * 0 - passed attributes contains unsupported features flags
1203 * 1 - all features are supported
1204 ******************************************************************************/
1205 static int imsm_check_attributes(__u32 attributes)
1206 {
1207 int ret_val = 1;
1208 __u32 not_supported = MPB_ATTRIB_SUPPORTED^0xffffffff;
1209
1210 not_supported &= ~MPB_ATTRIB_IGNORED;
1211
1212 not_supported &= attributes;
1213 if (not_supported) {
1214 fprintf(stderr, Name "(IMSM): Unsupported attributes : %x\n",
1215 (unsigned)__le32_to_cpu(not_supported));
1216 if (not_supported & MPB_ATTRIB_CHECKSUM_VERIFY) {
1217 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n");
1218 not_supported ^= MPB_ATTRIB_CHECKSUM_VERIFY;
1219 }
1220 if (not_supported & MPB_ATTRIB_2TB) {
1221 dprintf("\t\tMPB_ATTRIB_2TB\n");
1222 not_supported ^= MPB_ATTRIB_2TB;
1223 }
1224 if (not_supported & MPB_ATTRIB_RAID0) {
1225 dprintf("\t\tMPB_ATTRIB_RAID0\n");
1226 not_supported ^= MPB_ATTRIB_RAID0;
1227 }
1228 if (not_supported & MPB_ATTRIB_RAID1) {
1229 dprintf("\t\tMPB_ATTRIB_RAID1\n");
1230 not_supported ^= MPB_ATTRIB_RAID1;
1231 }
1232 if (not_supported & MPB_ATTRIB_RAID10) {
1233 dprintf("\t\tMPB_ATTRIB_RAID10\n");
1234 not_supported ^= MPB_ATTRIB_RAID10;
1235 }
1236 if (not_supported & MPB_ATTRIB_RAID1E) {
1237 dprintf("\t\tMPB_ATTRIB_RAID1E\n");
1238 not_supported ^= MPB_ATTRIB_RAID1E;
1239 }
1240 if (not_supported & MPB_ATTRIB_RAID5) {
1241 dprintf("\t\tMPB_ATTRIB_RAID5\n");
1242 not_supported ^= MPB_ATTRIB_RAID5;
1243 }
1244 if (not_supported & MPB_ATTRIB_RAIDCNG) {
1245 dprintf("\t\tMPB_ATTRIB_RAIDCNG\n");
1246 not_supported ^= MPB_ATTRIB_RAIDCNG;
1247 }
1248 if (not_supported & MPB_ATTRIB_BBM) {
1249 dprintf("\t\tMPB_ATTRIB_BBM\n");
1250 not_supported ^= MPB_ATTRIB_BBM;
1251 }
1252 if (not_supported & MPB_ATTRIB_CHECKSUM_VERIFY) {
1253 dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n");
1254 not_supported ^= MPB_ATTRIB_CHECKSUM_VERIFY;
1255 }
1256 if (not_supported & MPB_ATTRIB_EXP_STRIPE_SIZE) {
1257 dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n");
1258 not_supported ^= MPB_ATTRIB_EXP_STRIPE_SIZE;
1259 }
1260 if (not_supported & MPB_ATTRIB_2TB_DISK) {
1261 dprintf("\t\tMPB_ATTRIB_2TB_DISK\n");
1262 not_supported ^= MPB_ATTRIB_2TB_DISK;
1263 }
1264 if (not_supported & MPB_ATTRIB_NEVER_USE2) {
1265 dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n");
1266 not_supported ^= MPB_ATTRIB_NEVER_USE2;
1267 }
1268 if (not_supported & MPB_ATTRIB_NEVER_USE) {
1269 dprintf("\t\tMPB_ATTRIB_NEVER_USE\n");
1270 not_supported ^= MPB_ATTRIB_NEVER_USE;
1271 }
1272
1273 if (not_supported)
1274 dprintf(Name "(IMSM): Unknown attributes : %x\n", not_supported);
1275
1276 ret_val = 0;
1277 }
1278
1279 return ret_val;
1280 }
1281
1282 #ifndef MDASSEMBLE
1283 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map);
1284
1285 static void examine_super_imsm(struct supertype *st, char *homehost)
1286 {
1287 struct intel_super *super = st->sb;
1288 struct imsm_super *mpb = super->anchor;
1289 char str[MAX_SIGNATURE_LENGTH];
1290 int i;
1291 struct mdinfo info;
1292 char nbuf[64];
1293 __u32 sum;
1294 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1295 struct dl *dl;
1296
1297 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
1298 printf(" Magic : %s\n", str);
1299 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
1300 printf(" Version : %s\n", get_imsm_version(mpb));
1301 printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num));
1302 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
1303 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
1304 printf(" Attributes : ");
1305 if (imsm_check_attributes(mpb->attributes))
1306 printf("All supported\n");
1307 else
1308 printf("not supported\n");
1309 getinfo_super_imsm(st, &info, NULL);
1310 fname_from_uuid(st, &info, nbuf, ':');
1311 printf(" UUID : %s\n", nbuf + 5);
1312 sum = __le32_to_cpu(mpb->check_sum);
1313 printf(" Checksum : %08x %s\n", sum,
1314 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
1315 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
1316 printf(" Disks : %d\n", mpb->num_disks);
1317 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
1318 print_imsm_disk(__get_imsm_disk(mpb, super->disks->index), super->disks->index, reserved);
1319 if (super->bbm_log) {
1320 struct bbm_log *log = super->bbm_log;
1321
1322 printf("\n");
1323 printf("Bad Block Management Log:\n");
1324 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
1325 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
1326 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
1327 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
1328 printf(" First Spare : %llx\n",
1329 (unsigned long long) __le64_to_cpu(log->first_spare_lba));
1330 }
1331 for (i = 0; i < mpb->num_raid_devs; i++) {
1332 struct mdinfo info;
1333 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
1334
1335 super->current_vol = i;
1336 getinfo_super_imsm(st, &info, NULL);
1337 fname_from_uuid(st, &info, nbuf, ':');
1338 print_imsm_dev(super, dev, nbuf + 5, super->disks->index);
1339 }
1340 for (i = 0; i < mpb->num_disks; i++) {
1341 if (i == super->disks->index)
1342 continue;
1343 print_imsm_disk(__get_imsm_disk(mpb, i), i, reserved);
1344 }
1345
1346 for (dl = super->disks; dl; dl = dl->next)
1347 if (dl->index == -1)
1348 print_imsm_disk(&dl->disk, -1, reserved);
1349
1350 examine_migr_rec_imsm(super);
1351 }
1352
1353 static void brief_examine_super_imsm(struct supertype *st, int verbose)
1354 {
1355 /* We just write a generic IMSM ARRAY entry */
1356 struct mdinfo info;
1357 char nbuf[64];
1358 struct intel_super *super = st->sb;
1359
1360 if (!super->anchor->num_raid_devs) {
1361 printf("ARRAY metadata=imsm\n");
1362 return;
1363 }
1364
1365 getinfo_super_imsm(st, &info, NULL);
1366 fname_from_uuid(st, &info, nbuf, ':');
1367 printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5);
1368 }
1369
1370 static void brief_examine_subarrays_imsm(struct supertype *st, int verbose)
1371 {
1372 /* We just write a generic IMSM ARRAY entry */
1373 struct mdinfo info;
1374 char nbuf[64];
1375 char nbuf1[64];
1376 struct intel_super *super = st->sb;
1377 int i;
1378
1379 if (!super->anchor->num_raid_devs)
1380 return;
1381
1382 getinfo_super_imsm(st, &info, NULL);
1383 fname_from_uuid(st, &info, nbuf, ':');
1384 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1385 struct imsm_dev *dev = get_imsm_dev(super, i);
1386
1387 super->current_vol = i;
1388 getinfo_super_imsm(st, &info, NULL);
1389 fname_from_uuid(st, &info, nbuf1, ':');
1390 printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n",
1391 dev->volume, nbuf + 5, i, nbuf1 + 5);
1392 }
1393 }
1394
1395 static void export_examine_super_imsm(struct supertype *st)
1396 {
1397 struct intel_super *super = st->sb;
1398 struct imsm_super *mpb = super->anchor;
1399 struct mdinfo info;
1400 char nbuf[64];
1401
1402 getinfo_super_imsm(st, &info, NULL);
1403 fname_from_uuid(st, &info, nbuf, ':');
1404 printf("MD_METADATA=imsm\n");
1405 printf("MD_LEVEL=container\n");
1406 printf("MD_UUID=%s\n", nbuf+5);
1407 printf("MD_DEVICES=%u\n", mpb->num_disks);
1408 }
1409
1410 static void detail_super_imsm(struct supertype *st, char *homehost)
1411 {
1412 struct mdinfo info;
1413 char nbuf[64];
1414
1415 getinfo_super_imsm(st, &info, NULL);
1416 fname_from_uuid(st, &info, nbuf, ':');
1417 printf("\n UUID : %s\n", nbuf + 5);
1418 }
1419
1420 static void brief_detail_super_imsm(struct supertype *st)
1421 {
1422 struct mdinfo info;
1423 char nbuf[64];
1424 getinfo_super_imsm(st, &info, NULL);
1425 fname_from_uuid(st, &info, nbuf, ':');
1426 printf(" UUID=%s", nbuf + 5);
1427 }
1428
1429 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
1430 static void fd2devname(int fd, char *name);
1431
1432 static int ahci_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
1433 {
1434 /* dump an unsorted list of devices attached to AHCI Intel storage
1435 * controller, as well as non-connected ports
1436 */
1437 int hba_len = strlen(hba_path) + 1;
1438 struct dirent *ent;
1439 DIR *dir;
1440 char *path = NULL;
1441 int err = 0;
1442 unsigned long port_mask = (1 << port_count) - 1;
1443
1444 if (port_count > (int)sizeof(port_mask) * 8) {
1445 if (verbose)
1446 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
1447 return 2;
1448 }
1449
1450 /* scroll through /sys/dev/block looking for devices attached to
1451 * this hba
1452 */
1453 dir = opendir("/sys/dev/block");
1454 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1455 int fd;
1456 char model[64];
1457 char vendor[64];
1458 char buf[1024];
1459 int major, minor;
1460 char *device;
1461 char *c;
1462 int port;
1463 int type;
1464
1465 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
1466 continue;
1467 path = devt_to_devpath(makedev(major, minor));
1468 if (!path)
1469 continue;
1470 if (!path_attached_to_hba(path, hba_path)) {
1471 free(path);
1472 path = NULL;
1473 continue;
1474 }
1475
1476 /* retrieve the scsi device type */
1477 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
1478 if (verbose)
1479 fprintf(stderr, Name ": failed to allocate 'device'\n");
1480 err = 2;
1481 break;
1482 }
1483 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
1484 if (load_sys(device, buf) != 0) {
1485 if (verbose)
1486 fprintf(stderr, Name ": failed to read device type for %s\n",
1487 path);
1488 err = 2;
1489 free(device);
1490 break;
1491 }
1492 type = strtoul(buf, NULL, 10);
1493
1494 /* if it's not a disk print the vendor and model */
1495 if (!(type == 0 || type == 7 || type == 14)) {
1496 vendor[0] = '\0';
1497 model[0] = '\0';
1498 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
1499 if (load_sys(device, buf) == 0) {
1500 strncpy(vendor, buf, sizeof(vendor));
1501 vendor[sizeof(vendor) - 1] = '\0';
1502 c = (char *) &vendor[sizeof(vendor) - 1];
1503 while (isspace(*c) || *c == '\0')
1504 *c-- = '\0';
1505
1506 }
1507 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
1508 if (load_sys(device, buf) == 0) {
1509 strncpy(model, buf, sizeof(model));
1510 model[sizeof(model) - 1] = '\0';
1511 c = (char *) &model[sizeof(model) - 1];
1512 while (isspace(*c) || *c == '\0')
1513 *c-- = '\0';
1514 }
1515
1516 if (vendor[0] && model[0])
1517 sprintf(buf, "%.64s %.64s", vendor, model);
1518 else
1519 switch (type) { /* numbers from hald/linux/device.c */
1520 case 1: sprintf(buf, "tape"); break;
1521 case 2: sprintf(buf, "printer"); break;
1522 case 3: sprintf(buf, "processor"); break;
1523 case 4:
1524 case 5: sprintf(buf, "cdrom"); break;
1525 case 6: sprintf(buf, "scanner"); break;
1526 case 8: sprintf(buf, "media_changer"); break;
1527 case 9: sprintf(buf, "comm"); break;
1528 case 12: sprintf(buf, "raid"); break;
1529 default: sprintf(buf, "unknown");
1530 }
1531 } else
1532 buf[0] = '\0';
1533 free(device);
1534
1535 /* chop device path to 'host%d' and calculate the port number */
1536 c = strchr(&path[hba_len], '/');
1537 if (!c) {
1538 if (verbose)
1539 fprintf(stderr, Name ": %s - invalid path name\n", path + hba_len);
1540 err = 2;
1541 break;
1542 }
1543 *c = '\0';
1544 if (sscanf(&path[hba_len], "host%d", &port) == 1)
1545 port -= host_base;
1546 else {
1547 if (verbose) {
1548 *c = '/'; /* repair the full string */
1549 fprintf(stderr, Name ": failed to determine port number for %s\n",
1550 path);
1551 }
1552 err = 2;
1553 break;
1554 }
1555
1556 /* mark this port as used */
1557 port_mask &= ~(1 << port);
1558
1559 /* print out the device information */
1560 if (buf[0]) {
1561 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
1562 continue;
1563 }
1564
1565 fd = dev_open(ent->d_name, O_RDONLY);
1566 if (fd < 0)
1567 printf(" Port%d : - disk info unavailable -\n", port);
1568 else {
1569 fd2devname(fd, buf);
1570 printf(" Port%d : %s", port, buf);
1571 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
1572 printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN, buf);
1573 else
1574 printf(" ()\n");
1575 }
1576 close(fd);
1577 free(path);
1578 path = NULL;
1579 }
1580 if (path)
1581 free(path);
1582 if (dir)
1583 closedir(dir);
1584 if (err == 0) {
1585 int i;
1586
1587 for (i = 0; i < port_count; i++)
1588 if (port_mask & (1 << i))
1589 printf(" Port%d : - no device attached -\n", i);
1590 }
1591
1592 return err;
1593 }
1594
1595 static void print_found_intel_controllers(struct sys_dev *elem)
1596 {
1597 for (; elem; elem = elem->next) {
1598 fprintf(stderr, Name ": found Intel(R) ");
1599 if (elem->type == SYS_DEV_SATA)
1600 fprintf(stderr, "SATA ");
1601 else if (elem->type == SYS_DEV_SAS)
1602 fprintf(stderr, "SAS ");
1603 fprintf(stderr, "RAID controller");
1604 if (elem->pci_id)
1605 fprintf(stderr, " at %s", elem->pci_id);
1606 fprintf(stderr, ".\n");
1607 }
1608 fflush(stderr);
1609 }
1610
1611 static int ahci_get_port_count(const char *hba_path, int *port_count)
1612 {
1613 struct dirent *ent;
1614 DIR *dir;
1615 int host_base = -1;
1616
1617 *port_count = 0;
1618 if ((dir = opendir(hba_path)) == NULL)
1619 return -1;
1620
1621 for (ent = readdir(dir); ent; ent = readdir(dir)) {
1622 int host;
1623
1624 if (sscanf(ent->d_name, "host%d", &host) != 1)
1625 continue;
1626 if (*port_count == 0)
1627 host_base = host;
1628 else if (host < host_base)
1629 host_base = host;
1630
1631 if (host + 1 > *port_count + host_base)
1632 *port_count = host + 1 - host_base;
1633 }
1634 closedir(dir);
1635 return host_base;
1636 }
1637
1638 static void print_imsm_capability(const struct imsm_orom *orom)
1639 {
1640 printf(" Platform : Intel(R) Matrix Storage Manager\n");
1641 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
1642 orom->hotfix_ver, orom->build);
1643 printf(" RAID Levels :%s%s%s%s%s\n",
1644 imsm_orom_has_raid0(orom) ? " raid0" : "",
1645 imsm_orom_has_raid1(orom) ? " raid1" : "",
1646 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
1647 imsm_orom_has_raid10(orom) ? " raid10" : "",
1648 imsm_orom_has_raid5(orom) ? " raid5" : "");
1649 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
1650 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
1651 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
1652 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
1653 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
1654 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
1655 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
1656 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
1657 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
1658 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
1659 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
1660 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
1661 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
1662 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
1663 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
1664 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
1665 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1666 printf(" Max Disks : %d\n", orom->tds);
1667 printf(" Max Volumes : %d\n", orom->vpa);
1668 return;
1669 }
1670
1671 static int detail_platform_imsm(int verbose, int enumerate_only)
1672 {
1673 /* There are two components to imsm platform support, the ahci SATA
1674 * controller and the option-rom. To find the SATA controller we
1675 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
1676 * controller with the Intel vendor id is present. This approach
1677 * allows mdadm to leverage the kernel's ahci detection logic, with the
1678 * caveat that if ahci.ko is not loaded mdadm will not be able to
1679 * detect platform raid capabilities. The option-rom resides in a
1680 * platform "Adapter ROM". We scan for its signature to retrieve the
1681 * platform capabilities. If raid support is disabled in the BIOS the
1682 * option-rom capability structure will not be available.
1683 */
1684 const struct imsm_orom *orom;
1685 struct sys_dev *list, *hba;
1686 int host_base = 0;
1687 int port_count = 0;
1688 int result=0;
1689
1690 if (enumerate_only) {
1691 if (check_env("IMSM_NO_PLATFORM"))
1692 return 0;
1693 list = find_intel_devices();
1694 if (!list)
1695 return 2;
1696 for (hba = list; hba; hba = hba->next) {
1697 orom = find_imsm_capability(hba->type);
1698 if (!orom) {
1699 result = 2;
1700 break;
1701 }
1702 }
1703 free_sys_dev(&list);
1704 return result;
1705 }
1706
1707 list = find_intel_devices();
1708 if (!list) {
1709 if (verbose)
1710 fprintf(stderr, Name ": no active Intel(R) RAID "
1711 "controller found.\n");
1712 free_sys_dev(&list);
1713 return 2;
1714 } else if (verbose)
1715 print_found_intel_controllers(list);
1716
1717 for (hba = list; hba; hba = hba->next) {
1718 orom = find_imsm_capability(hba->type);
1719 if (!orom)
1720 fprintf(stderr, Name ": imsm capabilities not found for controller: %s (type %s)\n",
1721 hba->path, get_sys_dev_type(hba->type));
1722 else
1723 print_imsm_capability(orom);
1724 }
1725
1726 for (hba = list; hba; hba = hba->next) {
1727 printf(" I/O Controller : %s (%s)\n",
1728 hba->path, get_sys_dev_type(hba->type));
1729
1730 if (hba->type == SYS_DEV_SATA) {
1731 host_base = ahci_get_port_count(hba->path, &port_count);
1732 if (ahci_enumerate_ports(hba->path, port_count, host_base, verbose)) {
1733 if (verbose)
1734 fprintf(stderr, Name ": failed to enumerate "
1735 "ports on SATA controller at %s.", hba->pci_id);
1736 result |= 2;
1737 }
1738 }
1739 }
1740
1741 free_sys_dev(&list);
1742 return result;
1743 }
1744 #endif
1745
1746 static int match_home_imsm(struct supertype *st, char *homehost)
1747 {
1748 /* the imsm metadata format does not specify any host
1749 * identification information. We return -1 since we can never
1750 * confirm nor deny whether a given array is "meant" for this
1751 * host. We rely on compare_super and the 'family_num' fields to
1752 * exclude member disks that do not belong, and we rely on
1753 * mdadm.conf to specify the arrays that should be assembled.
1754 * Auto-assembly may still pick up "foreign" arrays.
1755 */
1756
1757 return -1;
1758 }
1759
1760 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1761 {
1762 /* The uuid returned here is used for:
1763 * uuid to put into bitmap file (Create, Grow)
1764 * uuid for backup header when saving critical section (Grow)
1765 * comparing uuids when re-adding a device into an array
1766 * In these cases the uuid required is that of the data-array,
1767 * not the device-set.
1768 * uuid to recognise same set when adding a missing device back
1769 * to an array. This is a uuid for the device-set.
1770 *
1771 * For each of these we can make do with a truncated
1772 * or hashed uuid rather than the original, as long as
1773 * everyone agrees.
1774 * In each case the uuid required is that of the data-array,
1775 * not the device-set.
1776 */
1777 /* imsm does not track uuid's so we synthesis one using sha1 on
1778 * - The signature (Which is constant for all imsm array, but no matter)
1779 * - the orig_family_num of the container
1780 * - the index number of the volume
1781 * - the 'serial' number of the volume.
1782 * Hopefully these are all constant.
1783 */
1784 struct intel_super *super = st->sb;
1785
1786 char buf[20];
1787 struct sha1_ctx ctx;
1788 struct imsm_dev *dev = NULL;
1789 __u32 family_num;
1790
1791 /* some mdadm versions failed to set ->orig_family_num, in which
1792 * case fall back to ->family_num. orig_family_num will be
1793 * fixed up with the first metadata update.
1794 */
1795 family_num = super->anchor->orig_family_num;
1796 if (family_num == 0)
1797 family_num = super->anchor->family_num;
1798 sha1_init_ctx(&ctx);
1799 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1800 sha1_process_bytes(&family_num, sizeof(__u32), &ctx);
1801 if (super->current_vol >= 0)
1802 dev = get_imsm_dev(super, super->current_vol);
1803 if (dev) {
1804 __u32 vol = super->current_vol;
1805 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1806 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1807 }
1808 sha1_finish_ctx(&ctx, buf);
1809 memcpy(uuid, buf, 4*4);
1810 }
1811
1812 #if 0
1813 static void
1814 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1815 {
1816 __u8 *v = get_imsm_version(mpb);
1817 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1818 char major[] = { 0, 0, 0 };
1819 char minor[] = { 0 ,0, 0 };
1820 char patch[] = { 0, 0, 0 };
1821 char *ver_parse[] = { major, minor, patch };
1822 int i, j;
1823
1824 i = j = 0;
1825 while (*v != '\0' && v < end) {
1826 if (*v != '.' && j < 2)
1827 ver_parse[i][j++] = *v;
1828 else {
1829 i++;
1830 j = 0;
1831 }
1832 v++;
1833 }
1834
1835 *m = strtol(minor, NULL, 0);
1836 *p = strtol(patch, NULL, 0);
1837 }
1838 #endif
1839
1840 static __u32 migr_strip_blocks_resync(struct imsm_dev *dev)
1841 {
1842 /* migr_strip_size when repairing or initializing parity */
1843 struct imsm_map *map = get_imsm_map(dev, 0);
1844 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1845
1846 switch (get_imsm_raid_level(map)) {
1847 case 5:
1848 case 10:
1849 return chunk;
1850 default:
1851 return 128*1024 >> 9;
1852 }
1853 }
1854
1855 static __u32 migr_strip_blocks_rebuild(struct imsm_dev *dev)
1856 {
1857 /* migr_strip_size when rebuilding a degraded disk, no idea why
1858 * this is different than migr_strip_size_resync(), but it's good
1859 * to be compatible
1860 */
1861 struct imsm_map *map = get_imsm_map(dev, 1);
1862 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1863
1864 switch (get_imsm_raid_level(map)) {
1865 case 1:
1866 case 10:
1867 if (map->num_members % map->num_domains == 0)
1868 return 128*1024 >> 9;
1869 else
1870 return chunk;
1871 case 5:
1872 return max((__u32) 64*1024 >> 9, chunk);
1873 default:
1874 return 128*1024 >> 9;
1875 }
1876 }
1877
1878 static __u32 num_stripes_per_unit_resync(struct imsm_dev *dev)
1879 {
1880 struct imsm_map *lo = get_imsm_map(dev, 0);
1881 struct imsm_map *hi = get_imsm_map(dev, 1);
1882 __u32 lo_chunk = __le32_to_cpu(lo->blocks_per_strip);
1883 __u32 hi_chunk = __le32_to_cpu(hi->blocks_per_strip);
1884
1885 return max((__u32) 1, hi_chunk / lo_chunk);
1886 }
1887
1888 static __u32 num_stripes_per_unit_rebuild(struct imsm_dev *dev)
1889 {
1890 struct imsm_map *lo = get_imsm_map(dev, 0);
1891 int level = get_imsm_raid_level(lo);
1892
1893 if (level == 1 || level == 10) {
1894 struct imsm_map *hi = get_imsm_map(dev, 1);
1895
1896 return hi->num_domains;
1897 } else
1898 return num_stripes_per_unit_resync(dev);
1899 }
1900
1901 static __u8 imsm_num_data_members(struct imsm_dev *dev, int second_map)
1902 {
1903 /* named 'imsm_' because raid0, raid1 and raid10
1904 * counter-intuitively have the same number of data disks
1905 */
1906 struct imsm_map *map = get_imsm_map(dev, second_map);
1907
1908 switch (get_imsm_raid_level(map)) {
1909 case 0:
1910 case 1:
1911 case 10:
1912 return map->num_members;
1913 case 5:
1914 return map->num_members - 1;
1915 default:
1916 dprintf("%s: unsupported raid level\n", __func__);
1917 return 0;
1918 }
1919 }
1920
1921 static __u32 parity_segment_depth(struct imsm_dev *dev)
1922 {
1923 struct imsm_map *map = get_imsm_map(dev, 0);
1924 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1925
1926 switch(get_imsm_raid_level(map)) {
1927 case 1:
1928 case 10:
1929 return chunk * map->num_domains;
1930 case 5:
1931 return chunk * map->num_members;
1932 default:
1933 return chunk;
1934 }
1935 }
1936
1937 static __u32 map_migr_block(struct imsm_dev *dev, __u32 block)
1938 {
1939 struct imsm_map *map = get_imsm_map(dev, 1);
1940 __u32 chunk = __le32_to_cpu(map->blocks_per_strip);
1941 __u32 strip = block / chunk;
1942
1943 switch (get_imsm_raid_level(map)) {
1944 case 1:
1945 case 10: {
1946 __u32 vol_strip = (strip * map->num_domains) + 1;
1947 __u32 vol_stripe = vol_strip / map->num_members;
1948
1949 return vol_stripe * chunk + block % chunk;
1950 } case 5: {
1951 __u32 stripe = strip / (map->num_members - 1);
1952
1953 return stripe * chunk + block % chunk;
1954 }
1955 default:
1956 return 0;
1957 }
1958 }
1959
1960 static __u64 blocks_per_migr_unit(struct intel_super *super,
1961 struct imsm_dev *dev)
1962 {
1963 /* calculate the conversion factor between per member 'blocks'
1964 * (md/{resync,rebuild}_start) and imsm migration units, return
1965 * 0 for the 'not migrating' and 'unsupported migration' cases
1966 */
1967 if (!dev->vol.migr_state)
1968 return 0;
1969
1970 switch (migr_type(dev)) {
1971 case MIGR_GEN_MIGR: {
1972 struct migr_record *migr_rec = super->migr_rec;
1973 return __le32_to_cpu(migr_rec->blocks_per_unit);
1974 }
1975 case MIGR_VERIFY:
1976 case MIGR_REPAIR:
1977 case MIGR_INIT: {
1978 struct imsm_map *map = get_imsm_map(dev, 0);
1979 __u32 stripes_per_unit;
1980 __u32 blocks_per_unit;
1981 __u32 parity_depth;
1982 __u32 migr_chunk;
1983 __u32 block_map;
1984 __u32 block_rel;
1985 __u32 segment;
1986 __u32 stripe;
1987 __u8 disks;
1988
1989 /* yes, this is really the translation of migr_units to
1990 * per-member blocks in the 'resync' case
1991 */
1992 stripes_per_unit = num_stripes_per_unit_resync(dev);
1993 migr_chunk = migr_strip_blocks_resync(dev);
1994 disks = imsm_num_data_members(dev, 0);
1995 blocks_per_unit = stripes_per_unit * migr_chunk * disks;
1996 stripe = __le16_to_cpu(map->blocks_per_strip) * disks;
1997 segment = blocks_per_unit / stripe;
1998 block_rel = blocks_per_unit - segment * stripe;
1999 parity_depth = parity_segment_depth(dev);
2000 block_map = map_migr_block(dev, block_rel);
2001 return block_map + parity_depth * segment;
2002 }
2003 case MIGR_REBUILD: {
2004 __u32 stripes_per_unit;
2005 __u32 migr_chunk;
2006
2007 stripes_per_unit = num_stripes_per_unit_rebuild(dev);
2008 migr_chunk = migr_strip_blocks_rebuild(dev);
2009 return migr_chunk * stripes_per_unit;
2010 }
2011 case MIGR_STATE_CHANGE:
2012 default:
2013 return 0;
2014 }
2015 }
2016
2017 static int imsm_level_to_layout(int level)
2018 {
2019 switch (level) {
2020 case 0:
2021 case 1:
2022 return 0;
2023 case 5:
2024 case 6:
2025 return ALGORITHM_LEFT_ASYMMETRIC;
2026 case 10:
2027 return 0x102;
2028 }
2029 return UnSet;
2030 }
2031
2032 /*******************************************************************************
2033 * Function: read_imsm_migr_rec
2034 * Description: Function reads imsm migration record from last sector of disk
2035 * Parameters:
2036 * fd : disk descriptor
2037 * super : metadata info
2038 * Returns:
2039 * 0 : success,
2040 * -1 : fail
2041 ******************************************************************************/
2042 static int read_imsm_migr_rec(int fd, struct intel_super *super)
2043 {
2044 int ret_val = -1;
2045 unsigned long long dsize;
2046
2047 get_dev_size(fd, NULL, &dsize);
2048 if (lseek64(fd, dsize - 512, SEEK_SET) < 0) {
2049 fprintf(stderr,
2050 Name ": Cannot seek to anchor block: %s\n",
2051 strerror(errno));
2052 goto out;
2053 }
2054 if (read(fd, super->migr_rec_buf, 512) != 512) {
2055 fprintf(stderr,
2056 Name ": Cannot read migr record block: %s\n",
2057 strerror(errno));
2058 goto out;
2059 }
2060 ret_val = 0;
2061
2062 out:
2063 return ret_val;
2064 }
2065
2066 /*******************************************************************************
2067 * Function: load_imsm_migr_rec
2068 * Description: Function reads imsm migration record (it is stored at the last
2069 * sector of disk)
2070 * Parameters:
2071 * super : imsm internal array info
2072 * info : general array info
2073 * Returns:
2074 * 0 : success
2075 * -1 : fail
2076 ******************************************************************************/
2077 static int load_imsm_migr_rec(struct intel_super *super, struct mdinfo *info)
2078 {
2079 struct mdinfo *sd;
2080 struct dl *dl = NULL;
2081 char nm[30];
2082 int retval = -1;
2083 int fd = -1;
2084
2085 if (info) {
2086 for (sd = info->devs ; sd ; sd = sd->next) {
2087 /* read only from one of the first two slots */
2088 if ((sd->disk.raid_disk > 1) ||
2089 (sd->disk.raid_disk < 0))
2090 continue;
2091 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2092 fd = dev_open(nm, O_RDONLY);
2093 if (fd >= 0)
2094 break;
2095 }
2096 }
2097 if (fd < 0) {
2098 for (dl = super->disks; dl; dl = dl->next) {
2099 /* read only from one of the first two slots */
2100 if (dl->index > 1)
2101 continue;
2102 sprintf(nm, "%d:%d", dl->major, dl->minor);
2103 fd = dev_open(nm, O_RDONLY);
2104 if (fd >= 0)
2105 break;
2106 }
2107 }
2108 if (fd < 0)
2109 goto out;
2110 retval = read_imsm_migr_rec(fd, super);
2111
2112 out:
2113 if (fd >= 0)
2114 close(fd);
2115 return retval;
2116 }
2117
2118 #ifndef MDASSEMBLE
2119 /*******************************************************************************
2120 * function: imsm_create_metadata_checkpoint_update
2121 * Description: It creates update for checkpoint change.
2122 * Parameters:
2123 * super : imsm internal array info
2124 * u : pointer to prepared update
2125 * Returns:
2126 * Uptate length.
2127 * If length is equal to 0, input pointer u contains no update
2128 ******************************************************************************/
2129 static int imsm_create_metadata_checkpoint_update(
2130 struct intel_super *super,
2131 struct imsm_update_general_migration_checkpoint **u)
2132 {
2133
2134 int update_memory_size = 0;
2135
2136 dprintf("imsm_create_metadata_checkpoint_update(enter)\n");
2137
2138 if (u == NULL)
2139 return 0;
2140 *u = NULL;
2141
2142 /* size of all update data without anchor */
2143 update_memory_size =
2144 sizeof(struct imsm_update_general_migration_checkpoint);
2145
2146 *u = calloc(1, update_memory_size);
2147 if (*u == NULL) {
2148 dprintf("error: cannot get memory for "
2149 "imsm_create_metadata_checkpoint_update update\n");
2150 return 0;
2151 }
2152 (*u)->type = update_general_migration_checkpoint;
2153 (*u)->curr_migr_unit = __le32_to_cpu(super->migr_rec->curr_migr_unit);
2154 dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n",
2155 (*u)->curr_migr_unit);
2156
2157 return update_memory_size;
2158 }
2159
2160
2161 static void imsm_update_metadata_locally(struct supertype *st,
2162 void *buf, int len);
2163
2164 /*******************************************************************************
2165 * Function: write_imsm_migr_rec
2166 * Description: Function writes imsm migration record
2167 * (at the last sector of disk)
2168 * Parameters:
2169 * super : imsm internal array info
2170 * Returns:
2171 * 0 : success
2172 * -1 : if fail
2173 ******************************************************************************/
2174 static int write_imsm_migr_rec(struct supertype *st)
2175 {
2176 struct intel_super *super = st->sb;
2177 unsigned long long dsize;
2178 char nm[30];
2179 int fd = -1;
2180 int retval = -1;
2181 struct dl *sd;
2182 int len;
2183 struct imsm_update_general_migration_checkpoint *u;
2184
2185 for (sd = super->disks ; sd ; sd = sd->next) {
2186 /* write to 2 first slots only */
2187 if ((sd->index < 0) || (sd->index > 1))
2188 continue;
2189 sprintf(nm, "%d:%d", sd->major, sd->minor);
2190 fd = dev_open(nm, O_RDWR);
2191 if (fd < 0)
2192 continue;
2193 get_dev_size(fd, NULL, &dsize);
2194 if (lseek64(fd, dsize - 512, SEEK_SET) < 0) {
2195 fprintf(stderr,
2196 Name ": Cannot seek to anchor block: %s\n",
2197 strerror(errno));
2198 goto out;
2199 }
2200 if (write(fd, super->migr_rec_buf, 512) != 512) {
2201 fprintf(stderr,
2202 Name ": Cannot write migr record block: %s\n",
2203 strerror(errno));
2204 goto out;
2205 }
2206 close(fd);
2207 fd = -1;
2208 }
2209 /* update checkpoint information in metadata */
2210 len = imsm_create_metadata_checkpoint_update(super, &u);
2211
2212 if (len <= 0) {
2213 dprintf("imsm: Cannot prepare update\n");
2214 goto out;
2215 }
2216 /* update metadata locally */
2217 imsm_update_metadata_locally(st, u, len);
2218 /* and possibly remotely */
2219 if (st->update_tail) {
2220 append_metadata_update(st, u, len);
2221 /* during reshape we do all work inside metadata handler
2222 * manage_reshape(), so metadata update has to be triggered
2223 * insida it
2224 */
2225 flush_metadata_updates(st);
2226 st->update_tail = &st->updates;
2227 } else
2228 free(u);
2229
2230 retval = 0;
2231 out:
2232 if (fd >= 0)
2233 close(fd);
2234 return retval;
2235 }
2236 #endif /* MDASSEMBLE */
2237
2238 /* spare/missing disks activations are not allowe when
2239 * array/container performs reshape operation, because
2240 * all arrays in container works on the same disks set
2241 */
2242 int imsm_reshape_blocks_arrays_changes(struct intel_super *super)
2243 {
2244 int rv = 0;
2245 struct intel_dev *i_dev;
2246 struct imsm_dev *dev;
2247
2248 /* check whole container
2249 */
2250 for (i_dev = super->devlist; i_dev; i_dev = i_dev->next) {
2251 dev = i_dev->dev;
2252 if (is_gen_migration(dev)) {
2253 /* No repair during any migration in container
2254 */
2255 rv = 1;
2256 break;
2257 }
2258 }
2259 return rv;
2260 }
2261
2262 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info, char *dmap)
2263 {
2264 struct intel_super *super = st->sb;
2265 struct migr_record *migr_rec = super->migr_rec;
2266 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
2267 struct imsm_map *map = get_imsm_map(dev, 0);
2268 struct imsm_map *prev_map = get_imsm_map(dev, 1);
2269 struct imsm_map *map_to_analyse = map;
2270 struct dl *dl;
2271 char *devname;
2272 unsigned int component_size_alligment;
2273 int map_disks = info->array.raid_disks;
2274
2275 memset(info, 0, sizeof(*info));
2276 if (prev_map)
2277 map_to_analyse = prev_map;
2278
2279 dl = super->current_disk;
2280
2281 info->container_member = super->current_vol;
2282 info->array.raid_disks = map->num_members;
2283 info->array.level = get_imsm_raid_level(map_to_analyse);
2284 info->array.layout = imsm_level_to_layout(info->array.level);
2285 info->array.md_minor = -1;
2286 info->array.ctime = 0;
2287 info->array.utime = 0;
2288 info->array.chunk_size =
2289 __le16_to_cpu(map_to_analyse->blocks_per_strip) << 9;
2290 info->array.state = !dev->vol.dirty;
2291 info->custom_array_size = __le32_to_cpu(dev->size_high);
2292 info->custom_array_size <<= 32;
2293 info->custom_array_size |= __le32_to_cpu(dev->size_low);
2294 info->recovery_blocked = imsm_reshape_blocks_arrays_changes(st->sb);
2295
2296 if (prev_map && map->map_state == prev_map->map_state &&
2297 (migr_type(dev) == MIGR_GEN_MIGR)) {
2298 info->reshape_active = 1;
2299 info->new_level = get_imsm_raid_level(map);
2300 info->new_layout = imsm_level_to_layout(info->new_level);
2301 info->new_chunk = __le16_to_cpu(map->blocks_per_strip) << 9;
2302 info->delta_disks = map->num_members - prev_map->num_members;
2303 if (info->delta_disks) {
2304 /* this needs to be applied to every array
2305 * in the container.
2306 */
2307 info->reshape_active = CONTAINER_RESHAPE;
2308 }
2309 /* We shape information that we give to md might have to be
2310 * modify to cope with md's requirement for reshaping arrays.
2311 * For example, when reshaping a RAID0, md requires it to be
2312 * presented as a degraded RAID4.
2313 * Also if a RAID0 is migrating to a RAID5 we need to specify
2314 * the array as already being RAID5, but the 'before' layout
2315 * is a RAID4-like layout.
2316 */
2317 switch (info->array.level) {
2318 case 0:
2319 switch(info->new_level) {
2320 case 0:
2321 /* conversion is happening as RAID4 */
2322 info->array.level = 4;
2323 info->array.raid_disks += 1;
2324 break;
2325 case 5:
2326 /* conversion is happening as RAID5 */
2327 info->array.level = 5;
2328 info->array.layout = ALGORITHM_PARITY_N;
2329 info->delta_disks -= 1;
2330 break;
2331 default:
2332 /* FIXME error message */
2333 info->array.level = UnSet;
2334 break;
2335 }
2336 break;
2337 }
2338 } else {
2339 info->new_level = UnSet;
2340 info->new_layout = UnSet;
2341 info->new_chunk = info->array.chunk_size;
2342 info->delta_disks = 0;
2343 }
2344
2345 if (dl) {
2346 info->disk.major = dl->major;
2347 info->disk.minor = dl->minor;
2348 info->disk.number = dl->index;
2349 info->disk.raid_disk = get_imsm_disk_slot(map_to_analyse,
2350 dl->index);
2351 }
2352
2353 info->data_offset = __le32_to_cpu(map_to_analyse->pba_of_lba0);
2354 info->component_size =
2355 __le32_to_cpu(map_to_analyse->blocks_per_member);
2356
2357 /* check component size aligment
2358 */
2359 component_size_alligment =
2360 info->component_size % (info->array.chunk_size/512);
2361
2362 if (component_size_alligment &&
2363 (info->array.level != 1) && (info->array.level != UnSet)) {
2364 dprintf("imsm: reported component size alligned from %llu ",
2365 info->component_size);
2366 info->component_size -= component_size_alligment;
2367 dprintf("to %llu (%i).\n",
2368 info->component_size, component_size_alligment);
2369 }
2370
2371 memset(info->uuid, 0, sizeof(info->uuid));
2372 info->recovery_start = MaxSector;
2373
2374 info->reshape_progress = 0;
2375 info->resync_start = MaxSector;
2376 if ((map_to_analyse->map_state == IMSM_T_STATE_UNINITIALIZED ||
2377 dev->vol.dirty) &&
2378 imsm_reshape_blocks_arrays_changes(super) == 0) {
2379 info->resync_start = 0;
2380 }
2381 if (dev->vol.migr_state) {
2382 switch (migr_type(dev)) {
2383 case MIGR_REPAIR:
2384 case MIGR_INIT: {
2385 __u64 blocks_per_unit = blocks_per_migr_unit(super,
2386 dev);
2387 __u64 units = __le32_to_cpu(dev->vol.curr_migr_unit);
2388
2389 info->resync_start = blocks_per_unit * units;
2390 break;
2391 }
2392 case MIGR_GEN_MIGR: {
2393 __u64 blocks_per_unit = blocks_per_migr_unit(super,
2394 dev);
2395 __u64 units = __le32_to_cpu(migr_rec->curr_migr_unit);
2396 unsigned long long array_blocks;
2397 int used_disks;
2398
2399 if (__le32_to_cpu(migr_rec->ascending_migr) &&
2400 (units <
2401 (__le32_to_cpu(migr_rec->num_migr_units)-1)) &&
2402 (super->migr_rec->rec_status ==
2403 __cpu_to_le32(UNIT_SRC_IN_CP_AREA)))
2404 units++;
2405
2406 info->reshape_progress = blocks_per_unit * units;
2407
2408 dprintf("IMSM: General Migration checkpoint : %llu "
2409 "(%llu) -> read reshape progress : %llu\n",
2410 (unsigned long long)units,
2411 (unsigned long long)blocks_per_unit,
2412 info->reshape_progress);
2413
2414 used_disks = imsm_num_data_members(dev, 1);
2415 if (used_disks > 0) {
2416 array_blocks = map->blocks_per_member *
2417 used_disks;
2418 /* round array size down to closest MB
2419 */
2420 info->custom_array_size = (array_blocks
2421 >> SECT_PER_MB_SHIFT)
2422 << SECT_PER_MB_SHIFT;
2423 }
2424 }
2425 case MIGR_VERIFY:
2426 /* we could emulate the checkpointing of
2427 * 'sync_action=check' migrations, but for now
2428 * we just immediately complete them
2429 */
2430 case MIGR_REBUILD:
2431 /* this is handled by container_content_imsm() */
2432 case MIGR_STATE_CHANGE:
2433 /* FIXME handle other migrations */
2434 default:
2435 /* we are not dirty, so... */
2436 info->resync_start = MaxSector;
2437 }
2438 }
2439
2440 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
2441 info->name[MAX_RAID_SERIAL_LEN] = 0;
2442
2443 info->array.major_version = -1;
2444 info->array.minor_version = -2;
2445 devname = devnum2devname(st->container_dev);
2446 *info->text_version = '\0';
2447 if (devname)
2448 sprintf(info->text_version, "/%s/%d", devname, info->container_member);
2449 free(devname);
2450 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
2451 uuid_from_super_imsm(st, info->uuid);
2452
2453 if (dmap) {
2454 int i, j;
2455 for (i=0; i<map_disks; i++) {
2456 dmap[i] = 0;
2457 if (i < info->array.raid_disks) {
2458 struct imsm_disk *dsk;
2459 j = get_imsm_disk_idx(dev, i, -1);
2460 dsk = get_imsm_disk(super, j);
2461 if (dsk && (dsk->status & CONFIGURED_DISK))
2462 dmap[i] = 1;
2463 }
2464 }
2465 }
2466 }
2467
2468 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed);
2469 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev);
2470
2471 static struct imsm_disk *get_imsm_missing(struct intel_super *super, __u8 index)
2472 {
2473 struct dl *d;
2474
2475 for (d = super->missing; d; d = d->next)
2476 if (d->index == index)
2477 return &d->disk;
2478 return NULL;
2479 }
2480
2481 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map)
2482 {
2483 struct intel_super *super = st->sb;
2484 struct imsm_disk *disk;
2485 int map_disks = info->array.raid_disks;
2486 int max_enough = -1;
2487 int i;
2488 struct imsm_super *mpb;
2489
2490 if (super->current_vol >= 0) {
2491 getinfo_super_imsm_volume(st, info, map);
2492 return;
2493 }
2494 memset(info, 0, sizeof(*info));
2495
2496 /* Set raid_disks to zero so that Assemble will always pull in valid
2497 * spares
2498 */
2499 info->array.raid_disks = 0;
2500 info->array.level = LEVEL_CONTAINER;
2501 info->array.layout = 0;
2502 info->array.md_minor = -1;
2503 info->array.ctime = 0; /* N/A for imsm */
2504 info->array.utime = 0;
2505 info->array.chunk_size = 0;
2506
2507 info->disk.major = 0;
2508 info->disk.minor = 0;
2509 info->disk.raid_disk = -1;
2510 info->reshape_active = 0;
2511 info->array.major_version = -1;
2512 info->array.minor_version = -2;
2513 strcpy(info->text_version, "imsm");
2514 info->safe_mode_delay = 0;
2515 info->disk.number = -1;
2516 info->disk.state = 0;
2517 info->name[0] = 0;
2518 info->recovery_start = MaxSector;
2519 info->recovery_blocked = imsm_reshape_blocks_arrays_changes(st->sb);
2520
2521 /* do we have the all the insync disks that we expect? */
2522 mpb = super->anchor;
2523
2524 for (i = 0; i < mpb->num_raid_devs; i++) {
2525 struct imsm_dev *dev = get_imsm_dev(super, i);
2526 int failed, enough, j, missing = 0;
2527 struct imsm_map *map;
2528 __u8 state;
2529
2530 failed = imsm_count_failed(super, dev);
2531 state = imsm_check_degraded(super, dev, failed);
2532 map = get_imsm_map(dev, dev->vol.migr_state);
2533
2534 /* any newly missing disks?
2535 * (catches single-degraded vs double-degraded)
2536 */
2537 for (j = 0; j < map->num_members; j++) {
2538 __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1);
2539 __u32 idx = ord_to_idx(ord);
2540
2541 if (!(ord & IMSM_ORD_REBUILD) &&
2542 get_imsm_missing(super, idx)) {
2543 missing = 1;
2544 break;
2545 }
2546 }
2547
2548 if (state == IMSM_T_STATE_FAILED)
2549 enough = -1;
2550 else if (state == IMSM_T_STATE_DEGRADED &&
2551 (state != map->map_state || missing))
2552 enough = 0;
2553 else /* we're normal, or already degraded */
2554 enough = 1;
2555
2556 /* in the missing/failed disk case check to see
2557 * if at least one array is runnable
2558 */
2559 max_enough = max(max_enough, enough);
2560 }
2561 dprintf("%s: enough: %d\n", __func__, max_enough);
2562 info->container_enough = max_enough;
2563
2564 if (super->disks) {
2565 __u32 reserved = imsm_reserved_sectors(super, super->disks);
2566
2567 disk = &super->disks->disk;
2568 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
2569 info->component_size = reserved;
2570 info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0;
2571 /* we don't change info->disk.raid_disk here because
2572 * this state will be finalized in mdmon after we have
2573 * found the 'most fresh' version of the metadata
2574 */
2575 info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
2576 info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
2577 }
2578
2579 /* only call uuid_from_super_imsm when this disk is part of a populated container,
2580 * ->compare_super may have updated the 'num_raid_devs' field for spares
2581 */
2582 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
2583 uuid_from_super_imsm(st, info->uuid);
2584 else
2585 memcpy(info->uuid, uuid_zero, sizeof(uuid_zero));
2586
2587 /* I don't know how to compute 'map' on imsm, so use safe default */
2588 if (map) {
2589 int i;
2590 for (i = 0; i < map_disks; i++)
2591 map[i] = 1;
2592 }
2593
2594 }
2595
2596 /* allocates memory and fills disk in mdinfo structure
2597 * for each disk in array */
2598 struct mdinfo *getinfo_super_disks_imsm(struct supertype *st)
2599 {
2600 struct mdinfo *mddev = NULL;
2601 struct intel_super *super = st->sb;
2602 struct imsm_disk *disk;
2603 int count = 0;
2604 struct dl *dl;
2605 if (!super || !super->disks)
2606 return NULL;
2607 dl = super->disks;
2608 mddev = malloc(sizeof(*mddev));
2609 if (!mddev) {
2610 fprintf(stderr, Name ": Failed to allocate memory.\n");
2611 return NULL;
2612 }
2613 memset(mddev, 0, sizeof(*mddev));
2614 while (dl) {
2615 struct mdinfo *tmp;
2616 disk = &dl->disk;
2617 tmp = malloc(sizeof(*tmp));
2618 if (!tmp) {
2619 fprintf(stderr, Name ": Failed to allocate memory.\n");
2620 if (mddev)
2621 sysfs_free(mddev);
2622 return NULL;
2623 }
2624 memset(tmp, 0, sizeof(*tmp));
2625 if (mddev->devs)
2626 tmp->next = mddev->devs;
2627 mddev->devs = tmp;
2628 tmp->disk.number = count++;
2629 tmp->disk.major = dl->major;
2630 tmp->disk.minor = dl->minor;
2631 tmp->disk.state = is_configured(disk) ?
2632 (1 << MD_DISK_ACTIVE) : 0;
2633 tmp->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0;
2634 tmp->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC);
2635 tmp->disk.raid_disk = -1;
2636 dl = dl->next;
2637 }
2638 return mddev;
2639 }
2640
2641 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
2642 char *update, char *devname, int verbose,
2643 int uuid_set, char *homehost)
2644 {
2645 /* For 'assemble' and 'force' we need to return non-zero if any
2646 * change was made. For others, the return value is ignored.
2647 * Update options are:
2648 * force-one : This device looks a bit old but needs to be included,
2649 * update age info appropriately.
2650 * assemble: clear any 'faulty' flag to allow this device to
2651 * be assembled.
2652 * force-array: Array is degraded but being forced, mark it clean
2653 * if that will be needed to assemble it.
2654 *
2655 * newdev: not used ????
2656 * grow: Array has gained a new device - this is currently for
2657 * linear only
2658 * resync: mark as dirty so a resync will happen.
2659 * name: update the name - preserving the homehost
2660 * uuid: Change the uuid of the array to match watch is given
2661 *
2662 * Following are not relevant for this imsm:
2663 * sparc2.2 : update from old dodgey metadata
2664 * super-minor: change the preferred_minor number
2665 * summaries: update redundant counters.
2666 * homehost: update the recorded homehost
2667 * _reshape_progress: record new reshape_progress position.
2668 */
2669 int rv = 1;
2670 struct intel_super *super = st->sb;
2671 struct imsm_super *mpb;
2672
2673 /* we can only update container info */
2674 if (!super || super->current_vol >= 0 || !super->anchor)
2675 return 1;
2676
2677 mpb = super->anchor;
2678
2679 if (strcmp(update, "uuid") == 0 && uuid_set && !info->update_private)
2680 rv = -1;
2681 else if (strcmp(update, "uuid") == 0 && uuid_set && info->update_private) {
2682 mpb->orig_family_num = *((__u32 *) info->update_private);
2683 rv = 0;
2684 } else if (strcmp(update, "uuid") == 0) {
2685 __u32 *new_family = malloc(sizeof(*new_family));
2686
2687 /* update orig_family_number with the incoming random
2688 * data, report the new effective uuid, and store the
2689 * new orig_family_num for future updates.
2690 */
2691 if (new_family) {
2692 memcpy(&mpb->orig_family_num, info->uuid, sizeof(__u32));
2693 uuid_from_super_imsm(st, info->uuid);
2694 *new_family = mpb->orig_family_num;
2695 info->update_private = new_family;
2696 rv = 0;
2697 }
2698 } else if (strcmp(update, "assemble") == 0)
2699 rv = 0;
2700 else
2701 rv = -1;
2702
2703 /* successful update? recompute checksum */
2704 if (rv == 0)
2705 mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb));
2706
2707 return rv;
2708 }
2709
2710 static size_t disks_to_mpb_size(int disks)
2711 {
2712 size_t size;
2713
2714 size = sizeof(struct imsm_super);
2715 size += (disks - 1) * sizeof(struct imsm_disk);
2716 size += 2 * sizeof(struct imsm_dev);
2717 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
2718 size += (4 - 2) * sizeof(struct imsm_map);
2719 /* 4 possible disk_ord_tbl's */
2720 size += 4 * (disks - 1) * sizeof(__u32);
2721
2722 return size;
2723 }
2724
2725 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
2726 {
2727 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
2728 return 0;
2729
2730 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
2731 }
2732
2733 static void free_devlist(struct intel_super *super)
2734 {
2735 struct intel_dev *dv;
2736
2737 while (super->devlist) {
2738 dv = super->devlist->next;
2739 free(super->devlist->dev);
2740 free(super->devlist);
2741 super->devlist = dv;
2742 }
2743 }
2744
2745 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
2746 {
2747 memcpy(dest, src, sizeof_imsm_dev(src, 0));
2748 }
2749
2750 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
2751 {
2752 /*
2753 * return:
2754 * 0 same, or first was empty, and second was copied
2755 * 1 second had wrong number
2756 * 2 wrong uuid
2757 * 3 wrong other info
2758 */
2759 struct intel_super *first = st->sb;
2760 struct intel_super *sec = tst->sb;
2761
2762 if (!first) {
2763 st->sb = tst->sb;
2764 tst->sb = NULL;
2765 return 0;
2766 }
2767 /* in platform dependent environment test if the disks
2768 * use the same Intel hba
2769 */
2770 if (!check_env("IMSM_NO_PLATFORM")) {
2771 if (!first->hba || !sec->hba ||
2772 (first->hba->type != sec->hba->type)) {
2773 fprintf(stderr,
2774 "HBAs of devices does not match %s != %s\n",
2775 first->hba ? get_sys_dev_type(first->hba->type) : NULL,
2776 sec->hba ? get_sys_dev_type(sec->hba->type) : NULL);
2777 return 3;
2778 }
2779 }
2780
2781 /* if an anchor does not have num_raid_devs set then it is a free
2782 * floating spare
2783 */
2784 if (first->anchor->num_raid_devs > 0 &&
2785 sec->anchor->num_raid_devs > 0) {
2786 /* Determine if these disks might ever have been
2787 * related. Further disambiguation can only take place
2788 * in load_super_imsm_all
2789 */
2790 __u32 first_family = first->anchor->orig_family_num;
2791 __u32 sec_family = sec->anchor->orig_family_num;
2792
2793 if (memcmp(first->anchor->sig, sec->anchor->sig,
2794 MAX_SIGNATURE_LENGTH) != 0)
2795 return 3;
2796
2797 if (first_family == 0)
2798 first_family = first->anchor->family_num;
2799 if (sec_family == 0)
2800 sec_family = sec->anchor->family_num;
2801
2802 if (first_family != sec_family)
2803 return 3;
2804
2805 }
2806
2807
2808 /* if 'first' is a spare promote it to a populated mpb with sec's
2809 * family number
2810 */
2811 if (first->anchor->num_raid_devs == 0 &&
2812 sec->anchor->num_raid_devs > 0) {
2813 int i;
2814 struct intel_dev *dv;
2815 struct imsm_dev *dev;
2816
2817 /* we need to copy raid device info from sec if an allocation
2818 * fails here we don't associate the spare
2819 */
2820 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
2821 dv = malloc(sizeof(*dv));
2822 if (!dv)
2823 break;
2824 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
2825 if (!dev) {
2826 free(dv);
2827 break;
2828 }
2829 dv->dev = dev;
2830 dv->index = i;
2831 dv->next = first->devlist;
2832 first->devlist = dv;
2833 }
2834 if (i < sec->anchor->num_raid_devs) {
2835 /* allocation failure */
2836 free_devlist(first);
2837 fprintf(stderr, "imsm: failed to associate spare\n");
2838 return 3;
2839 }
2840 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
2841 first->anchor->orig_family_num = sec->anchor->orig_family_num;
2842 first->anchor->family_num = sec->anchor->family_num;
2843 memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH);
2844 for (i = 0; i < sec->anchor->num_raid_devs; i++)
2845 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
2846 }
2847
2848 return 0;
2849 }
2850
2851 static void fd2devname(int fd, char *name)
2852 {
2853 struct stat st;
2854 char path[256];
2855 char dname[PATH_MAX];
2856 char *nm;
2857 int rv;
2858
2859 name[0] = '\0';
2860 if (fstat(fd, &st) != 0)
2861 return;
2862 sprintf(path, "/sys/dev/block/%d:%d",
2863 major(st.st_rdev), minor(st.st_rdev));
2864
2865 rv = readlink(path, dname, sizeof(dname)-1);
2866 if (rv <= 0)
2867 return;
2868
2869 dname[rv] = '\0';
2870 nm = strrchr(dname, '/');
2871 nm++;
2872 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
2873 }
2874
2875 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
2876
2877 static int imsm_read_serial(int fd, char *devname,
2878 __u8 serial[MAX_RAID_SERIAL_LEN])
2879 {
2880 unsigned char scsi_serial[255];
2881 int rv;
2882 int rsp_len;
2883 int len;
2884 char *dest;
2885 char *src;
2886 char *rsp_buf;
2887 int i;
2888
2889 memset(scsi_serial, 0, sizeof(scsi_serial));
2890
2891 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
2892
2893 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
2894 memset(serial, 0, MAX_RAID_SERIAL_LEN);
2895 fd2devname(fd, (char *) serial);
2896 return 0;
2897 }
2898
2899 if (rv != 0) {
2900 if (devname)
2901 fprintf(stderr,
2902 Name ": Failed to retrieve serial for %s\n",
2903 devname);
2904 return rv;
2905 }
2906
2907 rsp_len = scsi_serial[3];
2908 if (!rsp_len) {
2909 if (devname)
2910 fprintf(stderr,
2911 Name ": Failed to retrieve serial for %s\n",
2912 devname);
2913 return 2;
2914 }
2915 rsp_buf = (char *) &scsi_serial[4];
2916
2917 /* trim all whitespace and non-printable characters and convert
2918 * ':' to ';'
2919 */
2920 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
2921 src = &rsp_buf[i];
2922 if (*src > 0x20) {
2923 /* ':' is reserved for use in placeholder serial
2924 * numbers for missing disks
2925 */
2926 if (*src == ':')
2927 *dest++ = ';';
2928 else
2929 *dest++ = *src;
2930 }
2931 }
2932 len = dest - rsp_buf;
2933 dest = rsp_buf;
2934
2935 /* truncate leading characters */
2936 if (len > MAX_RAID_SERIAL_LEN) {
2937 dest += len - MAX_RAID_SERIAL_LEN;
2938 len = MAX_RAID_SERIAL_LEN;
2939 }
2940
2941 memset(serial, 0, MAX_RAID_SERIAL_LEN);
2942 memcpy(serial, dest, len);
2943
2944 return 0;
2945 }
2946
2947 static int serialcmp(__u8 *s1, __u8 *s2)
2948 {
2949 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
2950 }
2951
2952 static void serialcpy(__u8 *dest, __u8 *src)
2953 {
2954 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
2955 }
2956
2957 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
2958 {
2959 struct dl *dl;
2960
2961 for (dl = super->disks; dl; dl = dl->next)
2962 if (serialcmp(dl->serial, serial) == 0)
2963 break;
2964
2965 return dl;
2966 }
2967
2968 static struct imsm_disk *
2969 __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx)
2970 {
2971 int i;
2972
2973 for (i = 0; i < mpb->num_disks; i++) {
2974 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
2975
2976 if (serialcmp(disk->serial, serial) == 0) {
2977 if (idx)
2978 *idx = i;
2979 return disk;
2980 }
2981 }
2982
2983 return NULL;
2984 }
2985
2986 static int
2987 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
2988 {
2989 struct imsm_disk *disk;
2990 struct dl *dl;
2991 struct stat stb;
2992 int rv;
2993 char name[40];
2994 __u8 serial[MAX_RAID_SERIAL_LEN];
2995
2996 rv = imsm_read_serial(fd, devname, serial);
2997
2998 if (rv != 0)
2999 return 2;
3000
3001 dl = calloc(1, sizeof(*dl));
3002 if (!dl) {
3003 if (devname)
3004 fprintf(stderr,
3005 Name ": failed to allocate disk buffer for %s\n",
3006 devname);
3007 return 2;
3008 }
3009
3010 fstat(fd, &stb);
3011 dl->major = major(stb.st_rdev);
3012 dl->minor = minor(stb.st_rdev);
3013 dl->next = super->disks;
3014 dl->fd = keep_fd ? fd : -1;
3015 assert(super->disks == NULL);
3016 super->disks = dl;
3017 serialcpy(dl->serial, serial);
3018 dl->index = -2;
3019 dl->e = NULL;
3020 fd2devname(fd, name);
3021 if (devname)
3022 dl->devname = strdup(devname);
3023 else
3024 dl->devname = strdup(name);
3025
3026 /* look up this disk's index in the current anchor */
3027 disk = __serial_to_disk(dl->serial, super->anchor, &dl->index);
3028 if (disk) {
3029 dl->disk = *disk;
3030 /* only set index on disks that are a member of a
3031 * populated contianer, i.e. one with raid_devs
3032 */
3033 if (is_failed(&dl->disk))
3034 dl->index = -2;
3035 else if (is_spare(&dl->disk))
3036 dl->index = -1;
3037 }
3038
3039 return 0;
3040 }
3041
3042 #ifndef MDASSEMBLE
3043 /* When migrating map0 contains the 'destination' state while map1
3044 * contains the current state. When not migrating map0 contains the
3045 * current state. This routine assumes that map[0].map_state is set to
3046 * the current array state before being called.
3047 *
3048 * Migration is indicated by one of the following states
3049 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
3050 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
3051 * map1state=unitialized)
3052 * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal
3053 * map1state=normal)
3054 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
3055 * map1state=degraded)
3056 * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal
3057 * map1state=normal)
3058 */
3059 static void migrate(struct imsm_dev *dev, struct intel_super *super,
3060 __u8 to_state, int migr_type)
3061 {
3062 struct imsm_map *dest;
3063 struct imsm_map *src = get_imsm_map(dev, 0);
3064
3065 dev->vol.migr_state = 1;
3066 set_migr_type(dev, migr_type);
3067 dev->vol.curr_migr_unit = 0;
3068 dest = get_imsm_map(dev, 1);
3069
3070 /* duplicate and then set the target end state in map[0] */
3071 memcpy(dest, src, sizeof_imsm_map(src));
3072 if ((migr_type == MIGR_REBUILD) ||
3073 (migr_type == MIGR_GEN_MIGR)) {
3074 __u32 ord;
3075 int i;
3076
3077 for (i = 0; i < src->num_members; i++) {
3078 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
3079 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
3080 }
3081 }
3082
3083 if (migr_type == MIGR_GEN_MIGR)
3084 /* Clear migration record */
3085 memset(super->migr_rec, 0, sizeof(struct migr_record));
3086
3087 src->map_state = to_state;
3088 }
3089
3090 static void end_migration(struct imsm_dev *dev, __u8 map_state)
3091 {
3092 struct imsm_map *map = get_imsm_map(dev, 0);
3093 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3094 int i, j;
3095
3096 /* merge any IMSM_ORD_REBUILD bits that were not successfully
3097 * completed in the last migration.
3098 *
3099 * FIXME add support for raid-level-migration
3100 */
3101 for (i = 0; i < prev->num_members; i++)
3102 for (j = 0; j < map->num_members; j++)
3103 /* during online capacity expansion
3104 * disks position can be changed if takeover is used
3105 */
3106 if (ord_to_idx(map->disk_ord_tbl[j]) ==
3107 ord_to_idx(prev->disk_ord_tbl[i])) {
3108 map->disk_ord_tbl[j] |= prev->disk_ord_tbl[i];
3109 break;
3110 }
3111
3112 dev->vol.migr_state = 0;
3113 set_migr_type(dev, 0);
3114 dev->vol.curr_migr_unit = 0;
3115 map->map_state = map_state;
3116 }
3117 #endif
3118
3119 static int parse_raid_devices(struct intel_super *super)
3120 {
3121 int i;
3122 struct imsm_dev *dev_new;
3123 size_t len, len_migr;
3124 size_t max_len = 0;
3125 size_t space_needed = 0;
3126 struct imsm_super *mpb = super->anchor;
3127
3128 for (i = 0; i < super->anchor->num_raid_devs; i++) {
3129 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
3130 struct intel_dev *dv;
3131
3132 len = sizeof_imsm_dev(dev_iter, 0);
3133 len_migr = sizeof_imsm_dev(dev_iter, 1);
3134 if (len_migr > len)
3135 space_needed += len_migr - len;
3136
3137 dv = malloc(sizeof(*dv));
3138 if (!dv)
3139 return 1;
3140 if (max_len < len_migr)
3141 max_len = len_migr;
3142 if (max_len > len_migr)
3143 space_needed += max_len - len_migr;
3144 dev_new = malloc(max_len);
3145 if (!dev_new) {
3146 free(dv);
3147 return 1;
3148 }
3149 imsm_copy_dev(dev_new, dev_iter);
3150 dv->dev = dev_new;
3151 dv->index = i;
3152 dv->next = super->devlist;
3153 super->devlist = dv;
3154 }
3155
3156 /* ensure that super->buf is large enough when all raid devices
3157 * are migrating
3158 */
3159 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
3160 void *buf;
3161
3162 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
3163 if (posix_memalign(&buf, 512, len) != 0)
3164 return 1;
3165
3166 memcpy(buf, super->buf, super->len);
3167 memset(buf + super->len, 0, len - super->len);
3168 free(super->buf);
3169 super->buf = buf;
3170 super->len = len;
3171 }
3172
3173 return 0;
3174 }
3175
3176 /* retrieve a pointer to the bbm log which starts after all raid devices */
3177 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
3178 {
3179 void *ptr = NULL;
3180
3181 if (__le32_to_cpu(mpb->bbm_log_size)) {
3182 ptr = mpb;
3183 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
3184 }
3185
3186 return ptr;
3187 }
3188
3189 /*******************************************************************************
3190 * Function: check_mpb_migr_compatibility
3191 * Description: Function checks for unsupported migration features:
3192 * - migration optimization area (pba_of_lba0)
3193 * - descending reshape (ascending_migr)
3194 * Parameters:
3195 * super : imsm metadata information
3196 * Returns:
3197 * 0 : migration is compatible
3198 * -1 : migration is not compatible
3199 ******************************************************************************/
3200 int check_mpb_migr_compatibility(struct intel_super *super)
3201 {
3202 struct imsm_map *map0, *map1;
3203 struct migr_record *migr_rec = super->migr_rec;
3204 int i;
3205
3206 for (i = 0; i < super->anchor->num_raid_devs; i++) {
3207 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
3208
3209 if (dev_iter &&
3210 dev_iter->vol.migr_state == 1 &&
3211 dev_iter->vol.migr_type == MIGR_GEN_MIGR) {
3212 /* This device is migrating */
3213 map0 = get_imsm_map(dev_iter, 0);
3214 map1 = get_imsm_map(dev_iter, 1);
3215 if (map0->pba_of_lba0 != map1->pba_of_lba0)
3216 /* migration optimization area was used */
3217 return -1;
3218 if (migr_rec->ascending_migr == 0
3219 && migr_rec->dest_depth_per_unit > 0)
3220 /* descending reshape not supported yet */
3221 return -1;
3222 }
3223 }
3224 return 0;
3225 }
3226
3227 static void __free_imsm(struct intel_super *super, int free_disks);
3228
3229 /* load_imsm_mpb - read matrix metadata
3230 * allocates super->mpb to be freed by free_imsm
3231 */
3232 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
3233 {
3234 unsigned long long dsize;
3235 unsigned long long sectors;
3236 struct stat;
3237 struct imsm_super *anchor;
3238 __u32 check_sum;
3239
3240 get_dev_size(fd, NULL, &dsize);
3241 if (dsize < 1024) {
3242 if (devname)
3243 fprintf(stderr,
3244 Name ": %s: device to small for imsm\n",
3245 devname);
3246 return 1;
3247 }
3248
3249 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
3250 if (devname)
3251 fprintf(stderr, Name
3252 ": Cannot seek to anchor block on %s: %s\n",
3253 devname, strerror(errno));
3254 return 1;
3255 }
3256
3257 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
3258 if (devname)
3259 fprintf(stderr,
3260 Name ": Failed to allocate imsm anchor buffer"
3261 " on %s\n", devname);
3262 return 1;
3263 }
3264 if (read(fd, anchor, 512) != 512) {
3265 if (devname)
3266 fprintf(stderr,
3267 Name ": Cannot read anchor block on %s: %s\n",
3268 devname, strerror(errno));
3269 free(anchor);
3270 return 1;
3271 }
3272
3273 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
3274 if (devname)
3275 fprintf(stderr,
3276 Name ": no IMSM anchor on %s\n", devname);
3277 free(anchor);
3278 return 2;
3279 }
3280
3281 __free_imsm(super, 0);
3282 /* reload capability and hba */
3283
3284 /* capability and hba must be updated with new super allocation */
3285 find_intel_hba_capability(fd, super, devname);
3286 super->len = ROUND_UP(anchor->mpb_size, 512);
3287 if (posix_memalign(&super->buf, 512, super->len) != 0) {
3288 if (devname)
3289 fprintf(stderr,
3290 Name ": unable to allocate %zu byte mpb buffer\n",
3291 super->len);
3292 free(anchor);
3293 return 2;
3294 }
3295 memcpy(super->buf, anchor, 512);
3296
3297 sectors = mpb_sectors(anchor) - 1;
3298 free(anchor);
3299
3300 if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) {
3301 fprintf(stderr, Name
3302 ": %s could not allocate migr_rec buffer\n", __func__);
3303 free(super->buf);
3304 return 2;
3305 }
3306
3307 if (!sectors) {
3308 check_sum = __gen_imsm_checksum(super->anchor);
3309 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
3310 if (devname)
3311 fprintf(stderr,
3312 Name ": IMSM checksum %x != %x on %s\n",
3313 check_sum,
3314 __le32_to_cpu(super->anchor->check_sum),
3315 devname);
3316 return 2;
3317 }
3318
3319 return 0;
3320 }
3321
3322 /* read the extended mpb */
3323 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
3324 if (devname)
3325 fprintf(stderr,
3326 Name ": Cannot seek to extended mpb on %s: %s\n",
3327 devname, strerror(errno));
3328 return 1;
3329 }
3330
3331 if ((unsigned)read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
3332 if (devname)
3333 fprintf(stderr,
3334 Name ": Cannot read extended mpb on %s: %s\n",
3335 devname, strerror(errno));
3336 return 2;
3337 }
3338
3339 check_sum = __gen_imsm_checksum(super->anchor);
3340 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
3341 if (devname)
3342 fprintf(stderr,
3343 Name ": IMSM checksum %x != %x on %s\n",
3344 check_sum, __le32_to_cpu(super->anchor->check_sum),
3345 devname);
3346 return 3;
3347 }
3348
3349 /* FIXME the BBM log is disk specific so we cannot use this global
3350 * buffer for all disks. Ok for now since we only look at the global
3351 * bbm_log_size parameter to gate assembly
3352 */
3353 super->bbm_log = __get_imsm_bbm_log(super->anchor);
3354
3355 return 0;
3356 }
3357
3358 static int read_imsm_migr_rec(int fd, struct intel_super *super);
3359
3360 static int
3361 load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd)
3362 {
3363 int err;
3364
3365 err = load_imsm_mpb(fd, super, devname);
3366 if (err)
3367 return err;
3368 err = load_imsm_disk(fd, super, devname, keep_fd);
3369 if (err)
3370 return err;
3371 err = parse_raid_devices(super);
3372
3373 return err;
3374 }
3375
3376 static void __free_imsm_disk(struct dl *d)
3377 {
3378 if (d->fd >= 0)
3379 close(d->fd);
3380 if (d->devname)
3381 free(d->devname);
3382 if (d->e)
3383 free(d->e);
3384 free(d);
3385
3386 }
3387
3388 static void free_imsm_disks(struct intel_super *super)
3389 {
3390 struct dl *d;
3391
3392 while (super->disks) {
3393 d = super->disks;
3394 super->disks = d->next;
3395 __free_imsm_disk(d);
3396 }
3397 while (super->disk_mgmt_list) {
3398 d = super->disk_mgmt_list;
3399 super->disk_mgmt_list = d->next;
3400 __free_imsm_disk(d);
3401 }
3402 while (super->missing) {
3403 d = super->missing;
3404 super->missing = d->next;
3405 __free_imsm_disk(d);
3406 }
3407
3408 }
3409
3410 /* free all the pieces hanging off of a super pointer */
3411 static void __free_imsm(struct intel_super *super, int free_disks)
3412 {
3413 struct intel_hba *elem, *next;
3414
3415 if (super->buf) {
3416 free(super->buf);
3417 super->buf = NULL;
3418 }
3419 /* unlink capability description */
3420 super->orom = NULL;
3421 if (super->migr_rec_buf) {
3422 free(super->migr_rec_buf);
3423 super->migr_rec_buf = NULL;
3424 }
3425 if (free_disks)
3426 free_imsm_disks(super);
3427 free_devlist(super);
3428 elem = super->hba;
3429 while (elem) {
3430 if (elem->path)
3431 free((void *)elem->path);
3432 next = elem->next;
3433 free(elem);
3434 elem = next;
3435 }
3436 super->hba = NULL;
3437 }
3438
3439 static void free_imsm(struct intel_super *super)
3440 {
3441 __free_imsm(super, 1);
3442 free(super);
3443 }
3444
3445 static void free_super_imsm(struct supertype *st)
3446 {
3447 struct intel_super *super = st->sb;
3448
3449 if (!super)
3450 return;
3451
3452 free_imsm(super);
3453 st->sb = NULL;
3454 }
3455
3456 static struct intel_super *alloc_super(void)
3457 {
3458 struct intel_super *super = malloc(sizeof(*super));
3459
3460 if (super) {
3461 memset(super, 0, sizeof(*super));
3462 super->current_vol = -1;
3463 super->create_offset = ~((__u32 ) 0);
3464 }
3465 return super;
3466 }
3467
3468 /*
3469 * find and allocate hba and OROM/EFI based on valid fd of RAID component device
3470 */
3471 static int find_intel_hba_capability(int fd, struct intel_super *super, char *devname)
3472 {
3473 struct sys_dev *hba_name;
3474 int rv = 0;
3475
3476 if ((fd < 0) || check_env("IMSM_NO_PLATFORM")) {
3477 super->orom = NULL;
3478 super->hba = NULL;
3479 return 0;
3480 }
3481 hba_name = find_disk_attached_hba(fd, NULL);
3482 if (!hba_name) {
3483 if (devname)
3484 fprintf(stderr,
3485 Name ": %s is not attached to Intel(R) RAID controller.\n",
3486 devname);
3487 return 1;
3488 }
3489 rv = attach_hba_to_super(super, hba_name);
3490 if (rv == 2) {
3491 if (devname) {
3492 struct intel_hba *hba = super->hba;
3493
3494 fprintf(stderr, Name ": %s is attached to Intel(R) %s RAID "
3495 "controller (%s),\n"
3496 " but the container is assigned to Intel(R) "
3497 "%s RAID controller (",
3498 devname,
3499 hba_name->path,
3500 hba_name->pci_id ? : "Err!",
3501 get_sys_dev_type(hba_name->type));
3502
3503 while (hba) {
3504 fprintf(stderr, "%s", hba->pci_id ? : "Err!");
3505 if (hba->next)
3506 fprintf(stderr, ", ");
3507 hba = hba->next;
3508 }
3509
3510 fprintf(stderr, ").\n"
3511 " Mixing devices attached to different controllers "
3512 "is not allowed.\n");
3513 }
3514 free_sys_dev(&hba_name);
3515 return 2;
3516 }
3517 super->orom = find_imsm_capability(hba_name->type);
3518 free_sys_dev(&hba_name);
3519 if (!super->orom)
3520 return 3;
3521 return 0;
3522 }
3523
3524 /* find_missing - helper routine for load_super_imsm_all that identifies
3525 * disks that have disappeared from the system. This routine relies on
3526 * the mpb being uptodate, which it is at load time.
3527 */
3528 static int find_missing(struct intel_super *super)
3529 {
3530 int i;
3531 struct imsm_super *mpb = super->anchor;
3532 struct dl *dl;
3533 struct imsm_disk *disk;
3534
3535 for (i = 0; i < mpb->num_disks; i++) {
3536 disk = __get_imsm_disk(mpb, i);
3537 dl = serial_to_dl(disk->serial, super);
3538 if (dl)
3539 continue;
3540
3541 dl = malloc(sizeof(*dl));
3542 if (!dl)
3543 return 1;
3544 dl->major = 0;
3545 dl->minor = 0;
3546 dl->fd = -1;
3547 dl->devname = strdup("missing");
3548 dl->index = i;
3549 serialcpy(dl->serial, disk->serial);
3550 dl->disk = *disk;
3551 dl->e = NULL;
3552 dl->next = super->missing;
3553 super->missing = dl;
3554 }
3555
3556 return 0;
3557 }
3558
3559 #ifndef MDASSEMBLE
3560 static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list)
3561 {
3562 struct intel_disk *idisk = disk_list;
3563
3564 while (idisk) {
3565 if (serialcmp(idisk->disk.serial, serial) == 0)
3566 break;
3567 idisk = idisk->next;
3568 }
3569
3570 return idisk;
3571 }
3572
3573 static int __prep_thunderdome(struct intel_super **table, int tbl_size,
3574 struct intel_super *super,
3575 struct intel_disk **disk_list)
3576 {
3577 struct imsm_disk *d = &super->disks->disk;
3578 struct imsm_super *mpb = super->anchor;
3579 int i, j;
3580
3581 for (i = 0; i < tbl_size; i++) {
3582 struct imsm_super *tbl_mpb = table[i]->anchor;
3583 struct imsm_disk *tbl_d = &table[i]->disks->disk;
3584
3585 if (tbl_mpb->family_num == mpb->family_num) {
3586 if (tbl_mpb->check_sum == mpb->check_sum) {
3587 dprintf("%s: mpb from %d:%d matches %d:%d\n",
3588 __func__, super->disks->major,
3589 super->disks->minor,
3590 table[i]->disks->major,
3591 table[i]->disks->minor);
3592 break;
3593 }
3594
3595 if (((is_configured(d) && !is_configured(tbl_d)) ||
3596 is_configured(d) == is_configured(tbl_d)) &&
3597 tbl_mpb->generation_num < mpb->generation_num) {
3598 /* current version of the mpb is a
3599 * better candidate than the one in
3600 * super_table, but copy over "cross
3601 * generational" status
3602 */
3603 struct intel_disk *idisk;
3604
3605 dprintf("%s: mpb from %d:%d replaces %d:%d\n",
3606 __func__, super->disks->major,
3607 super->disks->minor,
3608 table[i]->disks->major,
3609 table[i]->disks->minor);
3610
3611 idisk = disk_list_get(tbl_d->serial, *disk_list);
3612 if (idisk && is_failed(&idisk->disk))
3613 tbl_d->status |= FAILED_DISK;
3614 break;
3615 } else {
3616 struct intel_disk *idisk;
3617 struct imsm_disk *disk;
3618
3619 /* tbl_mpb is more up to date, but copy
3620 * over cross generational status before
3621 * returning
3622 */
3623 disk = __serial_to_disk(d->serial, mpb, NULL);
3624 if (disk && is_failed(disk))
3625 d->status |= FAILED_DISK;
3626
3627 idisk = disk_list_get(d->serial, *disk_list);
3628 if (idisk) {
3629 idisk->owner = i;
3630 if (disk && is_configured(disk))
3631 idisk->disk.status |= CONFIGURED_DISK;
3632 }
3633
3634 dprintf("%s: mpb from %d:%d prefer %d:%d\n",
3635 __func__, super->disks->major,
3636 super->disks->minor,
3637 table[i]->disks->major,
3638 table[i]->disks->minor);
3639
3640 return tbl_size;
3641 }
3642 }
3643 }
3644
3645 if (i >= tbl_size)
3646 table[tbl_size++] = super;
3647 else
3648 table[i] = super;
3649
3650 /* update/extend the merged list of imsm_disk records */
3651 for (j = 0; j < mpb->num_disks; j++) {
3652 struct imsm_disk *disk = __get_imsm_disk(mpb, j);
3653 struct intel_disk *idisk;
3654
3655 idisk = disk_list_get(disk->serial, *disk_list);
3656 if (idisk) {
3657 idisk->disk.status |= disk->status;
3658 if (is_configured(&idisk->disk) ||
3659 is_failed(&idisk->disk))
3660 idisk->disk.status &= ~(SPARE_DISK);
3661 } else {
3662 idisk = calloc(1, sizeof(*idisk));
3663 if (!idisk)
3664 return -1;
3665 idisk->owner = IMSM_UNKNOWN_OWNER;
3666 idisk->disk = *disk;
3667 idisk->next = *disk_list;
3668 *disk_list = idisk;
3669 }
3670
3671 if (serialcmp(idisk->disk.serial, d->serial) == 0)
3672 idisk->owner = i;
3673 }
3674
3675 return tbl_size;
3676 }
3677
3678 static struct intel_super *
3679 validate_members(struct intel_super *super, struct intel_disk *disk_list,
3680 const int owner)
3681 {
3682 struct imsm_super *mpb = super->anchor;
3683 int ok_count = 0;
3684 int i;
3685
3686 for (i = 0; i < mpb->num_disks; i++) {
3687 struct imsm_disk *disk = __get_imsm_disk(mpb, i);
3688 struct intel_disk *idisk;
3689
3690 idisk = disk_list_get(disk->serial, disk_list);
3691 if (idisk) {
3692 if (idisk->owner == owner ||
3693 idisk->owner == IMSM_UNKNOWN_OWNER)
3694 ok_count++;
3695 else
3696 dprintf("%s: '%.16s' owner %d != %d\n",
3697 __func__, disk->serial, idisk->owner,
3698 owner);
3699 } else {
3700 dprintf("%s: unknown disk %x [%d]: %.16s\n",
3701 __func__, __le32_to_cpu(mpb->family_num), i,
3702 disk->serial);
3703 break;
3704 }
3705 }
3706
3707 if (ok_count == mpb->num_disks)
3708 return super;
3709 return NULL;
3710 }
3711
3712 static void show_conflicts(__u32 family_num, struct intel_super *super_list)
3713 {
3714 struct intel_super *s;
3715
3716 for (s = super_list; s; s = s->next) {
3717 if (family_num != s->anchor->family_num)
3718 continue;
3719 fprintf(stderr, "Conflict, offlining family %#x on '%s'\n",
3720 __le32_to_cpu(family_num), s->disks->devname);
3721 }
3722 }
3723
3724 static struct intel_super *
3725 imsm_thunderdome(struct intel_super **super_list, int len)
3726 {
3727 struct intel_super *super_table[len];
3728 struct intel_disk *disk_list = NULL;
3729 struct intel_super *champion, *spare;
3730 struct intel_super *s, **del;
3731 int tbl_size = 0;
3732 int conflict;
3733 int i;
3734
3735 memset(super_table, 0, sizeof(super_table));
3736 for (s = *super_list; s; s = s->next)
3737 tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list);
3738
3739 for (i = 0; i < tbl_size; i++) {
3740 struct imsm_disk *d;
3741 struct intel_disk *idisk;
3742 struct imsm_super *mpb = super_table[i]->anchor;
3743
3744 s = super_table[i];
3745 d = &s->disks->disk;
3746
3747 /* 'd' must appear in merged disk list for its
3748 * configuration to be valid
3749 */
3750 idisk = disk_list_get(d->serial, disk_list);
3751 if (idisk && idisk->owner == i)
3752 s = validate_members(s, disk_list, i);
3753 else
3754 s = NULL;
3755
3756 if (!s)
3757 dprintf("%s: marking family: %#x from %d:%d offline\n",
3758 __func__, mpb->family_num,
3759 super_table[i]->disks->major,
3760 super_table[i]->disks->minor);
3761 super_table[i] = s;
3762 }
3763
3764 /* This is where the mdadm implementation differs from the Windows
3765 * driver which has no strict concept of a container. We can only
3766 * assemble one family from a container, so when returning a prodigal
3767 * array member to this system the code will not be able to disambiguate
3768 * the container contents that should be assembled ("foreign" versus
3769 * "local"). It requires user intervention to set the orig_family_num
3770 * to a new value to establish a new container. The Windows driver in
3771 * this situation fixes up the volume name in place and manages the
3772 * foreign array as an independent entity.
3773 */
3774 s = NULL;
3775 spare = NULL;
3776 conflict = 0;
3777 for (i = 0; i < tbl_size; i++) {
3778 struct intel_super *tbl_ent = super_table[i];
3779 int is_spare = 0;
3780
3781 if (!tbl_ent)
3782 continue;
3783
3784 if (tbl_ent->anchor->num_raid_devs == 0) {
3785 spare = tbl_ent;
3786 is_spare = 1;
3787 }
3788
3789 if (s && !is_spare) {
3790 show_conflicts(tbl_ent->anchor->family_num, *super_list);
3791 conflict++;
3792 } else if (!s && !is_spare)
3793 s = tbl_ent;
3794 }
3795
3796 if (!s)
3797 s = spare;
3798 if (!s) {
3799 champion = NULL;
3800 goto out;
3801 }
3802 champion = s;
3803
3804 if (conflict)
3805 fprintf(stderr, "Chose family %#x on '%s', "
3806 "assemble conflicts to new container with '--update=uuid'\n",
3807 __le32_to_cpu(s->anchor->family_num), s->disks->devname);
3808
3809 /* collect all dl's onto 'champion', and update them to
3810 * champion's version of the status
3811 */
3812 for (s = *super_list; s; s = s->next) {
3813 struct imsm_super *mpb = champion->anchor;
3814 struct dl *dl = s->disks;
3815
3816 if (s == champion)
3817 continue;
3818
3819 for (i = 0; i < mpb->num_disks; i++) {
3820 struct imsm_disk *disk;
3821
3822 disk = __serial_to_disk(dl->serial, mpb, &dl->index);
3823 if (disk) {
3824 dl->disk = *disk;
3825 /* only set index on disks that are a member of
3826 * a populated contianer, i.e. one with
3827 * raid_devs
3828 */
3829 if (is_failed(&dl->disk))
3830 dl->index = -2;
3831 else if (is_spare(&dl->disk))
3832 dl->index = -1;
3833 break;
3834 }
3835 }
3836
3837 if (i >= mpb->num_disks) {
3838 struct intel_disk *idisk;
3839
3840 idisk = disk_list_get(dl->serial, disk_list);
3841 if (idisk && is_spare(&idisk->disk) &&
3842 !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
3843 dl->index = -1;
3844 else {
3845 dl->index = -2;
3846 continue;
3847 }
3848 }
3849
3850 dl->next = champion->disks;
3851 champion->disks = dl;
3852 s->disks = NULL;
3853 }
3854
3855 /* delete 'champion' from super_list */
3856 for (del = super_list; *del; ) {
3857 if (*del == champion) {
3858 *del = (*del)->next;
3859 break;
3860 } else
3861 del = &(*del)->next;
3862 }
3863 champion->next = NULL;
3864
3865 out:
3866 while (disk_list) {
3867 struct intel_disk *idisk = disk_list;
3868
3869 disk_list = disk_list->next;
3870 free(idisk);
3871 }
3872
3873 return champion;
3874 }
3875
3876 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
3877 char *devname)
3878 {
3879 struct mdinfo *sra;
3880 struct intel_super *super_list = NULL;
3881 struct intel_super *super = NULL;
3882 int devnum = fd2devnum(fd);
3883 struct mdinfo *sd;
3884 int retry;
3885 int err = 0;
3886 int i;
3887
3888 /* check if 'fd' an opened container */
3889 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
3890 if (!sra)
3891 return 1;
3892
3893 if (sra->array.major_version != -1 ||
3894 sra->array.minor_version != -2 ||
3895 strcmp(sra->text_version, "imsm") != 0) {
3896 err = 1;
3897 goto error;
3898 }
3899 /* load all mpbs */
3900 for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) {
3901 struct intel_super *s = alloc_super();
3902 char nm[32];
3903 int dfd;
3904 int rv;
3905
3906 err = 1;
3907 if (!s)
3908 goto error;
3909 s->next = super_list;
3910 super_list = s;
3911
3912 err = 2;
3913 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3914 dfd = dev_open(nm, O_RDWR);
3915 if (dfd < 0)
3916 goto error;
3917
3918 rv = find_intel_hba_capability(dfd, s, devname);
3919 /* no orom/efi or non-intel hba of the disk */
3920 if (rv != 0)
3921 goto error;
3922
3923 err = load_and_parse_mpb(dfd, s, NULL, 1);
3924
3925 /* retry the load if we might have raced against mdmon */
3926 if (err == 3 && mdmon_running(devnum))
3927 for (retry = 0; retry < 3; retry++) {
3928 usleep(3000);
3929 err = load_and_parse_mpb(dfd, s, NULL, 1);
3930 if (err != 3)
3931 break;
3932 }
3933 if (err)
3934 goto error;
3935 }
3936
3937 /* all mpbs enter, maybe one leaves */
3938 super = imsm_thunderdome(&super_list, i);
3939 if (!super) {
3940 err = 1;
3941 goto error;
3942 }
3943
3944 if (find_missing(super) != 0) {
3945 free_imsm(super);
3946 err = 2;
3947 goto error;
3948 }
3949
3950 /* load migration record */
3951 err = load_imsm_migr_rec(super, NULL);
3952 if (err) {
3953 err = 4;
3954 goto error;
3955 }
3956
3957 /* Check migration compatibility */
3958 if (check_mpb_migr_compatibility(super) != 0) {
3959 fprintf(stderr, Name ": Unsupported migration detected");
3960 if (devname)
3961 fprintf(stderr, " on %s\n", devname);
3962 else
3963 fprintf(stderr, " (IMSM).\n");
3964
3965 err = 5;
3966 goto error;
3967 }
3968
3969 err = 0;
3970
3971 error:
3972 while (super_list) {
3973 struct intel_super *s = super_list;
3974
3975 super_list = super_list->next;
3976 free_imsm(s);
3977 }
3978 sysfs_free(sra);
3979
3980 if (err)
3981 return err;
3982
3983 *sbp = super;
3984 st->container_dev = devnum;
3985 if (err == 0 && st->ss == NULL) {
3986 st->ss = &super_imsm;
3987 st->minor_version = 0;
3988 st->max_devs = IMSM_MAX_DEVICES;
3989 }
3990 return 0;
3991 }
3992
3993 static int load_container_imsm(struct supertype *st, int fd, char *devname)
3994 {
3995 return load_super_imsm_all(st, fd, &st->sb, devname);
3996 }
3997 #endif
3998
3999 static int load_super_imsm(struct supertype *st, int fd, char *devname)
4000 {
4001 struct intel_super *super;
4002 int rv;
4003
4004 if (test_partition(fd))
4005 /* IMSM not allowed on partitions */
4006 return 1;
4007
4008 free_super_imsm(st);
4009
4010 super = alloc_super();
4011 if (!super) {
4012 fprintf(stderr,
4013 Name ": malloc of %zu failed.\n",
4014 sizeof(*super));
4015 return 1;
4016 }
4017 /* Load hba and capabilities if they exist.
4018 * But do not preclude loading metadata in case capabilities or hba are
4019 * non-compliant and ignore_hw_compat is set.
4020 */
4021 rv = find_intel_hba_capability(fd, super, devname);
4022 /* no orom/efi or non-intel hba of the disk */
4023 if ((rv != 0) && (st->ignore_hw_compat == 0)) {
4024 if (devname)
4025 fprintf(stderr,
4026 Name ": No OROM/EFI properties for %s\n", devname);
4027 free_imsm(super);
4028 return 2;
4029 }
4030 rv = load_and_parse_mpb(fd, super, devname, 0);
4031
4032 if (rv) {
4033 if (devname)
4034 fprintf(stderr,
4035 Name ": Failed to load all information "
4036 "sections on %s\n", devname);
4037 free_imsm(super);
4038 return rv;
4039 }
4040
4041 st->sb = super;
4042 if (st->ss == NULL) {
4043 st->ss = &super_imsm;
4044 st->minor_version = 0;
4045 st->max_devs = IMSM_MAX_DEVICES;
4046 }
4047
4048 /* load migration record */
4049 if (load_imsm_migr_rec(super, NULL) == 0) {
4050 /* Check for unsupported migration features */
4051 if (check_mpb_migr_compatibility(super) != 0) {
4052 fprintf(stderr,
4053 Name ": Unsupported migration detected");
4054 if (devname)
4055 fprintf(stderr, " on %s\n", devname);
4056 else
4057 fprintf(stderr, " (IMSM).\n");
4058 return 3;
4059 }
4060 }
4061
4062 return 0;
4063 }
4064
4065 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
4066 {
4067 if (info->level == 1)
4068 return 128;
4069 return info->chunk_size >> 9;
4070 }
4071
4072 static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains)
4073 {
4074 __u32 num_stripes;
4075
4076 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
4077 num_stripes /= num_domains;
4078
4079 return num_stripes;
4080 }
4081
4082 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
4083 {
4084 if (info->level == 1)
4085 return info->size * 2;
4086 else
4087 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
4088 }
4089
4090 static void imsm_update_version_info(struct intel_super *super)
4091 {
4092 /* update the version and attributes */
4093 struct imsm_super *mpb = super->anchor;
4094 char *version;
4095 struct imsm_dev *dev;
4096 struct imsm_map *map;
4097 int i;
4098
4099 for (i = 0; i < mpb->num_raid_devs; i++) {
4100 dev = get_imsm_dev(super, i);
4101 map = get_imsm_map(dev, 0);
4102 if (__le32_to_cpu(dev->size_high) > 0)
4103 mpb->attributes |= MPB_ATTRIB_2TB;
4104
4105 /* FIXME detect when an array spans a port multiplier */
4106 #if 0
4107 mpb->attributes |= MPB_ATTRIB_PM;
4108 #endif
4109
4110 if (mpb->num_raid_devs > 1 ||
4111 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
4112 version = MPB_VERSION_ATTRIBS;
4113 switch (get_imsm_raid_level(map)) {
4114 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
4115 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
4116 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
4117 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
4118 }
4119 } else {
4120 if (map->num_members >= 5)
4121 version = MPB_VERSION_5OR6_DISK_ARRAY;
4122 else if (dev->status == DEV_CLONE_N_GO)
4123 version = MPB_VERSION_CNG;
4124 else if (get_imsm_raid_level(map) == 5)
4125 version = MPB_VERSION_RAID5;
4126 else if (map->num_members >= 3)
4127 version = MPB_VERSION_3OR4_DISK_ARRAY;
4128 else if (get_imsm_raid_level(map) == 1)
4129 version = MPB_VERSION_RAID1;
4130 else
4131 version = MPB_VERSION_RAID0;
4132 }
4133 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
4134 }
4135 }
4136
4137 static int check_name(struct intel_super *super, char *name, int quiet)
4138 {
4139 struct imsm_super *mpb = super->anchor;
4140 char *reason = NULL;
4141 int i;
4142
4143 if (strlen(name) > MAX_RAID_SERIAL_LEN)
4144 reason = "must be 16 characters or less";
4145
4146 for (i = 0; i < mpb->num_raid_devs; i++) {
4147 struct imsm_dev *dev = get_imsm_dev(super, i);
4148
4149 if (strncmp((char *) dev->volume, name, MAX_RAID_SERIAL_LEN) == 0) {
4150 reason = "already exists";
4151 break;
4152 }
4153 }
4154
4155 if (reason && !quiet)
4156 fprintf(stderr, Name ": imsm volume name %s\n", reason);
4157
4158 return !reason;
4159 }
4160
4161 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
4162 unsigned long long size, char *name,
4163 char *homehost, int *uuid)
4164 {
4165 /* We are creating a volume inside a pre-existing container.
4166 * so st->sb is already set.
4167 */
4168 struct intel_super *super = st->sb;
4169 struct imsm_super *mpb = super->anchor;
4170 struct intel_dev *dv;
4171 struct imsm_dev *dev;
4172 struct imsm_vol *vol;
4173 struct imsm_map *map;
4174 int idx = mpb->num_raid_devs;
4175 int i;
4176 unsigned long long array_blocks;
4177 size_t size_old, size_new;
4178 __u32 num_data_stripes;
4179
4180 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
4181 fprintf(stderr, Name": This imsm-container already has the "
4182 "maximum of %d volumes\n", super->orom->vpa);
4183 return 0;
4184 }
4185
4186 /* ensure the mpb is large enough for the new data */
4187 size_old = __le32_to_cpu(mpb->mpb_size);
4188 size_new = disks_to_mpb_size(info->nr_disks);
4189 if (size_new > size_old) {
4190 void *mpb_new;
4191 size_t size_round = ROUND_UP(size_new, 512);
4192
4193 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
4194 fprintf(stderr, Name": could not allocate new mpb\n");
4195 return 0;
4196 }
4197 if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) {
4198 fprintf(stderr, Name
4199 ": %s could not allocate migr_rec buffer\n",
4200 __func__);
4201 free(super->buf);
4202 free(super);
4203 free(mpb_new);
4204 return 0;
4205 }
4206 memcpy(mpb_new, mpb, size_old);
4207 free(mpb);
4208 mpb = mpb_new;
4209 super->anchor = mpb_new;
4210 mpb->mpb_size = __cpu_to_le32(size_new);
4211 memset(mpb_new + size_old, 0, size_round - size_old);
4212 }
4213 super->current_vol = idx;
4214
4215 /* handle 'failed_disks' by either:
4216 * a) create dummy disk entries in the table if this the first
4217 * volume in the array. We add them here as this is the only
4218 * opportunity to add them. add_to_super_imsm_volume()
4219 * handles the non-failed disks and continues incrementing
4220 * mpb->num_disks.
4221 * b) validate that 'failed_disks' matches the current number
4222 * of missing disks if the container is populated
4223 */
4224 if (super->current_vol == 0) {
4225 mpb->num_disks = 0;
4226 for (i = 0; i < info->failed_disks; i++) {
4227 struct imsm_disk *disk;
4228
4229 mpb->num_disks++;
4230 disk = __get_imsm_disk(mpb, i);
4231 disk->status = CONFIGURED_DISK | FAILED_DISK;
4232 disk->scsi_id = __cpu_to_le32(~(__u32)0);
4233 snprintf((char *) disk->serial, MAX_RAID_SERIAL_LEN,
4234 "missing:%d", i);
4235 }
4236 find_missing(super);
4237 } else {
4238 int missing = 0;
4239 struct dl *d;
4240
4241 for (d = super->missing; d; d = d->next)
4242 missing++;
4243 if (info->failed_disks > missing) {
4244 fprintf(stderr, Name": unable to add 'missing' disk to container\n");
4245 return 0;
4246 }
4247 }
4248
4249 if (!check_name(super, name, 0))
4250 return 0;
4251 dv = malloc(sizeof(*dv));
4252 if (!dv) {
4253 fprintf(stderr, Name ": failed to allocate device list entry\n");
4254 return 0;
4255 }
4256 dev = calloc(1, sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
4257 if (!dev) {
4258 free(dv);
4259 fprintf(stderr, Name": could not allocate raid device\n");
4260 return 0;
4261 }
4262
4263 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
4264 if (info->level == 1)
4265 array_blocks = info_to_blocks_per_member(info);
4266 else
4267 array_blocks = calc_array_size(info->level, info->raid_disks,
4268 info->layout, info->chunk_size,
4269 info->size*2);
4270 /* round array size down to closest MB */
4271 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
4272
4273 dev->size_low = __cpu_to_le32((__u32) array_blocks);
4274 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
4275 dev->status = (DEV_READ_COALESCING | DEV_WRITE_COALESCING);
4276 vol = &dev->vol;
4277 vol->migr_state = 0;
4278 set_migr_type(dev, MIGR_INIT);
4279 vol->dirty = !info->state;
4280 vol->curr_migr_unit = 0;
4281 map = get_imsm_map(dev, 0);
4282 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
4283 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
4284 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
4285 map->failed_disk_num = ~0;
4286 map->map_state = info->failed_disks ? IMSM_T_STATE_DEGRADED : IMSM_T_STATE_NORMAL;
4287 map->ddf = 1;
4288
4289 if (info->level == 1 && info->raid_disks > 2) {
4290 free(dev);
4291 free(dv);
4292 fprintf(stderr, Name": imsm does not support more than 2 disks"
4293 "in a raid1 volume\n");
4294 return 0;
4295 }
4296
4297 map->raid_level = info->level;
4298 if (info->level == 10) {
4299 map->raid_level = 1;
4300 map->num_domains = info->raid_disks / 2;
4301 } else if (info->level == 1)
4302 map->num_domains = info->raid_disks;
4303 else
4304 map->num_domains = 1;
4305
4306 num_data_stripes = info_to_num_data_stripes(info, map->num_domains);
4307 map->num_data_stripes = __cpu_to_le32(num_data_stripes);
4308
4309 map->num_members = info->raid_disks;
4310 for (i = 0; i < map->num_members; i++) {
4311 /* initialized in add_to_super */
4312 set_imsm_ord_tbl_ent(map, i, IMSM_ORD_REBUILD);
4313 }
4314 mpb->num_raid_devs++;
4315
4316 dv->dev = dev;
4317 dv->index = super->current_vol;
4318 dv->next = super->devlist;
4319 super->devlist = dv;
4320
4321 imsm_update_version_info(super);
4322
4323 return 1;
4324 }
4325
4326 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
4327 unsigned long long size, char *name,
4328 char *homehost, int *uuid)
4329 {
4330 /* This is primarily called by Create when creating a new array.
4331 * We will then get add_to_super called for each component, and then
4332 * write_init_super called to write it out to each device.
4333 * For IMSM, Create can create on fresh devices or on a pre-existing
4334 * array.
4335 * To create on a pre-existing array a different method will be called.
4336 * This one is just for fresh drives.
4337 */
4338 struct intel_super *super;
4339 struct imsm_super *mpb;
4340 size_t mpb_size;
4341 char *version;
4342
4343 if (st->sb)
4344 return init_super_imsm_volume(st, info, size, name, homehost, uuid);
4345
4346 if (info)
4347 mpb_size = disks_to_mpb_size(info->nr_disks);
4348 else
4349 mpb_size = 512;
4350
4351 super = alloc_super();
4352 if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) {
4353 free(super);
4354 super = NULL;
4355 }
4356 if (!super) {
4357 fprintf(stderr, Name
4358 ": %s could not allocate superblock\n", __func__);
4359 return 0;
4360 }
4361 if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) {
4362 fprintf(stderr, Name
4363 ": %s could not allocate migr_rec buffer\n", __func__);
4364 free(super->buf);
4365 free(super);
4366 return 0;
4367 }
4368 memset(super->buf, 0, mpb_size);
4369 mpb = super->buf;
4370 mpb->mpb_size = __cpu_to_le32(mpb_size);
4371 st->sb = super;
4372
4373 if (info == NULL) {
4374 /* zeroing superblock */
4375 return 0;
4376 }
4377
4378 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
4379
4380 version = (char *) mpb->sig;
4381 strcpy(version, MPB_SIGNATURE);
4382 version += strlen(MPB_SIGNATURE);
4383 strcpy(version, MPB_VERSION_RAID0);
4384
4385 return 1;
4386 }
4387
4388 #ifndef MDASSEMBLE
4389 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
4390 int fd, char *devname)
4391 {
4392 struct intel_super *super = st->sb;
4393 struct imsm_super *mpb = super->anchor;
4394 struct imsm_disk *_disk;
4395 struct imsm_dev *dev;
4396 struct imsm_map *map;
4397 struct dl *dl, *df;
4398 int slot;
4399
4400 dev = get_imsm_dev(super, super->current_vol);
4401 map = get_imsm_map(dev, 0);
4402
4403 if (! (dk->state & (1<<MD_DISK_SYNC))) {
4404 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
4405 devname);
4406 return 1;
4407 }
4408
4409 if (fd == -1) {
4410 /* we're doing autolayout so grab the pre-marked (in
4411 * validate_geometry) raid_disk
4412 */
4413 for (dl = super->disks; dl; dl = dl->next)
4414 if (dl->raiddisk == dk->raid_disk)
4415 break;
4416 } else {
4417 for (dl = super->disks; dl ; dl = dl->next)
4418 if (dl->major == dk->major &&
4419 dl->minor == dk->minor)
4420 break;
4421 }
4422
4423 if (!dl) {
4424 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
4425 return 1;
4426 }
4427
4428 /* add a pristine spare to the metadata */
4429 if (dl->index < 0) {
4430 dl->index = super->anchor->num_disks;
4431 super->anchor->num_disks++;
4432 }
4433 /* Check the device has not already been added */
4434 slot = get_imsm_disk_slot(map, dl->index);
4435 if (slot >= 0 &&
4436 (get_imsm_ord_tbl_ent(dev, slot, -1) & IMSM_ORD_REBUILD) == 0) {
4437 fprintf(stderr, Name ": %s has been included in this array twice\n",
4438 devname);
4439 return 1;
4440 }
4441 set_imsm_ord_tbl_ent(map, dk->raid_disk, dl->index);
4442 dl->disk.status = CONFIGURED_DISK;
4443
4444 /* update size of 'missing' disks to be at least as large as the
4445 * largest acitve member (we only have dummy missing disks when
4446 * creating the first volume)
4447 */
4448 if (super->current_vol == 0) {
4449 for (df = super->missing; df; df = df->next) {
4450 if (dl->disk.total_blocks > df->disk.total_blocks)
4451 df->disk.total_blocks = dl->disk.total_blocks;
4452 _disk = __get_imsm_disk(mpb, df->index);
4453 *_disk = df->disk;
4454 }
4455 }
4456
4457 /* refresh unset/failed slots to point to valid 'missing' entries */
4458 for (df = super->missing; df; df = df->next)
4459 for (slot = 0; slot < mpb->num_disks; slot++) {
4460 __u32 ord = get_imsm_ord_tbl_ent(dev, slot, -1);
4461
4462 if ((ord & IMSM_ORD_REBUILD) == 0)
4463 continue;
4464 set_imsm_ord_tbl_ent(map, slot, df->index | IMSM_ORD_REBUILD);
4465 dprintf("set slot:%d to missing disk:%d\n", slot, df->index);
4466 break;
4467 }
4468
4469 /* if we are creating the first raid device update the family number */
4470 if (super->current_vol == 0) {
4471 __u32 sum;
4472 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
4473
4474 _disk = __get_imsm_disk(mpb, dl->index);
4475 if (!_dev || !_disk) {
4476 fprintf(stderr, Name ": BUG mpb setup error\n");
4477 return 1;
4478 }
4479 *_dev = *dev;
4480 *_disk = dl->disk;
4481 sum = random32();
4482 sum += __gen_imsm_checksum(mpb);
4483 mpb->family_num = __cpu_to_le32(sum);
4484 mpb->orig_family_num = mpb->family_num;
4485 }
4486 super->current_disk = dl;
4487 return 0;
4488 }
4489
4490 /* mark_spare()
4491 * Function marks disk as spare and restores disk serial
4492 * in case it was previously marked as failed by takeover operation
4493 * reruns:
4494 * -1 : critical error
4495 * 0 : disk is marked as spare but serial is not set
4496 * 1 : success
4497 */
4498 int mark_spare(struct dl *disk)
4499 {
4500 __u8 serial[MAX_RAID_SERIAL_LEN];
4501 int ret_val = -1;
4502
4503 if (!disk)
4504 return ret_val;
4505
4506 ret_val = 0;
4507 if (!imsm_read_serial(disk->fd, NULL, serial)) {
4508 /* Restore disk serial number, because takeover marks disk
4509 * as failed and adds to serial ':0' before it becomes
4510 * a spare disk.
4511 */
4512 serialcpy(disk->serial, serial);
4513 serialcpy(disk->disk.serial, serial);
4514 ret_val = 1;
4515 }
4516 disk->disk.status = SPARE_DISK;
4517 disk->index = -1;
4518
4519 return ret_val;
4520 }
4521
4522 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
4523 int fd, char *devname)
4524 {
4525 struct intel_super *super = st->sb;
4526 struct dl *dd;
4527 unsigned long long size;
4528 __u32 id;
4529 int rv;
4530 struct stat stb;
4531
4532 /* If we are on an RAID enabled platform check that the disk is
4533 * attached to the raid controller.
4534 * We do not need to test disks attachment for container based additions,
4535 * they shall be already tested when container was created/assembled.
4536 */
4537 rv = find_intel_hba_capability(fd, super, devname);
4538 /* no orom/efi or non-intel hba of the disk */
4539 if (rv != 0) {
4540 dprintf("capability: %p fd: %d ret: %d\n",
4541 super->orom, fd, rv);
4542 return 1;
4543 }
4544
4545 if (super->current_vol >= 0)
4546 return add_to_super_imsm_volume(st, dk, fd, devname);
4547
4548 fstat(fd, &stb);
4549 dd = malloc(sizeof(*dd));
4550 if (!dd) {
4551 fprintf(stderr,
4552 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
4553 return 1;
4554 }
4555 memset(dd, 0, sizeof(*dd));
4556 dd->major = major(stb.st_rdev);
4557 dd->minor = minor(stb.st_rdev);
4558 dd->devname = devname ? strdup(devname) : NULL;
4559 dd->fd = fd;
4560 dd->e = NULL;
4561 dd->action = DISK_ADD;
4562 rv = imsm_read_serial(fd, devname, dd->serial);
4563 if (rv) {
4564 fprintf(stderr,
4565 Name ": failed to retrieve scsi serial, aborting\n");
4566 free(dd);
4567 abort();
4568 }
4569
4570 get_dev_size(fd, NULL, &size);
4571 size /= 512;
4572 serialcpy(dd->disk.serial, dd->serial);
4573 dd->disk.total_blocks = __cpu_to_le32(size);
4574 mark_spare(dd);
4575 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
4576 dd->disk.scsi_id = __cpu_to_le32(id);
4577 else
4578 dd->disk.scsi_id = __cpu_to_le32(0);
4579
4580 if (st->update_tail) {
4581 dd->next = super->disk_mgmt_list;
4582 super->disk_mgmt_list = dd;
4583 } else {
4584 dd->next = super->disks;
4585 super->disks = dd;
4586 super->updates_pending++;
4587 }
4588
4589 return 0;
4590 }
4591
4592
4593 static int remove_from_super_imsm(struct supertype *st, mdu_disk_info_t *dk)
4594 {
4595 struct intel_super *super = st->sb;
4596 struct dl *dd;
4597
4598 /* remove from super works only in mdmon - for communication
4599 * manager - monitor. Check if communication memory buffer
4600 * is prepared.
4601 */
4602 if (!st->update_tail) {
4603 fprintf(stderr,
4604 Name ": %s shall be used in mdmon context only"
4605 "(line %d).\n", __func__, __LINE__);
4606 return 1;
4607 }
4608 dd = malloc(sizeof(*dd));
4609 if (!dd) {
4610 fprintf(stderr,
4611 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
4612 return 1;
4613 }
4614 memset(dd, 0, sizeof(*dd));
4615 dd->major = dk->major;
4616 dd->minor = dk->minor;
4617 dd->fd = -1;
4618 mark_spare(dd);
4619 dd->action = DISK_REMOVE;
4620
4621 dd->next = super->disk_mgmt_list;
4622 super->disk_mgmt_list = dd;
4623
4624
4625 return 0;
4626 }
4627
4628 static int store_imsm_mpb(int fd, struct imsm_super *mpb);
4629
4630 static union {
4631 char buf[512];
4632 struct imsm_super anchor;
4633 } spare_record __attribute__ ((aligned(512)));
4634
4635 /* spare records have their own family number and do not have any defined raid
4636 * devices
4637 */
4638 static int write_super_imsm_spares(struct intel_super *super, int doclose)
4639 {
4640 struct imsm_super *mpb = super->anchor;
4641 struct imsm_super *spare = &spare_record.anchor;
4642 __u32 sum;
4643 struct dl *d;
4644
4645 spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)),
4646 spare->generation_num = __cpu_to_le32(1UL),
4647 spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
4648 spare->num_disks = 1,
4649 spare->num_raid_devs = 0,
4650 spare->cache_size = mpb->cache_size,
4651 spare->pwr_cycle_count = __cpu_to_le32(1),
4652
4653 snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH,
4654 MPB_SIGNATURE MPB_VERSION_RAID0);
4655
4656 for (d = super->disks; d; d = d->next) {
4657 if (d->index != -1)
4658 continue;
4659
4660 spare->disk[0] = d->disk;
4661 sum = __gen_imsm_checksum(spare);
4662 spare->family_num = __cpu_to_le32(sum);
4663 spare->orig_family_num = 0;
4664 sum = __gen_imsm_checksum(spare);
4665 spare->check_sum = __cpu_to_le32(sum);
4666
4667 if (store_imsm_mpb(d->fd, spare)) {
4668 fprintf(stderr, "%s: failed for device %d:%d %s\n",
4669 __func__, d->major, d->minor, strerror(errno));
4670 return 1;
4671 }
4672 if (doclose) {
4673 close(d->fd);
4674 d->fd = -1;
4675 }
4676 }
4677
4678 return 0;
4679 }
4680
4681 static int write_super_imsm(struct supertype *st, int doclose)
4682 {
4683 struct intel_super *super = st->sb;
4684 struct imsm_super *mpb = super->anchor;
4685 struct dl *d;
4686 __u32 generation;
4687 __u32 sum;
4688 int spares = 0;
4689 int i;
4690 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
4691 int num_disks = 0;
4692 int clear_migration_record = 1;
4693
4694 /* 'generation' is incremented everytime the metadata is written */
4695 generation = __le32_to_cpu(mpb->generation_num);
4696 generation++;
4697 mpb->generation_num = __cpu_to_le32(generation);
4698
4699 /* fix up cases where previous mdadm releases failed to set
4700 * orig_family_num
4701 */
4702 if (mpb->orig_family_num == 0)
4703 mpb->orig_family_num = mpb->family_num;
4704
4705 for (d = super->disks; d; d = d->next) {
4706 if (d->index == -1)
4707 spares++;
4708 else {
4709 mpb->disk[d->index] = d->disk;
4710 num_disks++;
4711 }
4712 }
4713 for (d = super->missing; d; d = d->next) {
4714 mpb->disk[d->index] = d->disk;
4715 num_disks++;
4716 }
4717 mpb->num_disks = num_disks;
4718 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
4719
4720 for (i = 0; i < mpb->num_raid_devs; i++) {
4721 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
4722 struct imsm_dev *dev2 = get_imsm_dev(super, i);
4723 if (dev && dev2) {
4724 imsm_copy_dev(dev, dev2);
4725 mpb_size += sizeof_imsm_dev(dev, 0);
4726 }
4727 if (is_gen_migration(dev2))
4728 clear_migration_record = 0;
4729 }
4730 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
4731 mpb->mpb_size = __cpu_to_le32(mpb_size);
4732
4733 /* recalculate checksum */
4734 sum = __gen_imsm_checksum(mpb);
4735 mpb->check_sum = __cpu_to_le32(sum);
4736
4737 if (clear_migration_record)
4738 memset(super->migr_rec_buf, 0, 512);
4739
4740 /* write the mpb for disks that compose raid devices */
4741 for (d = super->disks; d ; d = d->next) {
4742 if (d->index < 0 || is_failed(&d->disk))
4743 continue;
4744 if (store_imsm_mpb(d->fd, mpb))
4745 fprintf(stderr, "%s: failed for device %d:%d %s\n",
4746 __func__, d->major, d->minor, strerror(errno));
4747 if (clear_migration_record) {
4748 unsigned long long dsize;
4749
4750 get_dev_size(d->fd, NULL, &dsize);
4751 if (lseek64(d->fd, dsize - 512, SEEK_SET) >= 0) {
4752 if (write(d->fd, super->migr_rec_buf, 512) != 512)
4753 perror("Write migr_rec failed");
4754 }
4755 }
4756 if (doclose) {
4757 close(d->fd);
4758 d->fd = -1;
4759 }
4760 }
4761
4762 if (spares)
4763 return write_super_imsm_spares(super, doclose);
4764
4765 return 0;
4766 }
4767
4768
4769 static int create_array(struct supertype *st, int dev_idx)
4770 {
4771 size_t len;
4772 struct imsm_update_create_array *u;
4773 struct intel_super *super = st->sb;
4774 struct imsm_dev *dev = get_imsm_dev(super, dev_idx);
4775 struct imsm_map *map = get_imsm_map(dev, 0);
4776 struct disk_info *inf;
4777 struct imsm_disk *disk;
4778 int i;
4779
4780 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
4781 sizeof(*inf) * map->num_members;
4782 u = malloc(len);
4783 if (!u) {
4784 fprintf(stderr, "%s: failed to allocate update buffer\n",
4785 __func__);
4786 return 1;
4787 }
4788
4789 u->type = update_create_array;
4790 u->dev_idx = dev_idx;
4791 imsm_copy_dev(&u->dev, dev);
4792 inf = get_disk_info(u);
4793 for (i = 0; i < map->num_members; i++) {
4794 int idx = get_imsm_disk_idx(dev, i, -1);
4795
4796 disk = get_imsm_disk(super, idx);
4797 serialcpy(inf[i].serial, disk->serial);
4798 }
4799 append_metadata_update(st, u, len);
4800
4801 return 0;
4802 }
4803
4804 static int mgmt_disk(struct supertype *st)
4805 {
4806 struct intel_super *super = st->sb;
4807 size_t len;
4808 struct imsm_update_add_remove_disk *u;
4809
4810 if (!super->disk_mgmt_list)
4811 return 0;
4812
4813 len = sizeof(*u);
4814 u = malloc(len);
4815 if (!u) {
4816 fprintf(stderr, "%s: failed to allocate update buffer\n",
4817 __func__);
4818 return 1;
4819 }
4820
4821 u->type = update_add_remove_disk;
4822 append_metadata_update(st, u, len);
4823
4824 return 0;
4825 }
4826
4827 static int write_init_super_imsm(struct supertype *st)
4828 {
4829 struct intel_super *super = st->sb;
4830 int current_vol = super->current_vol;
4831
4832 /* we are done with current_vol reset it to point st at the container */
4833 super->current_vol = -1;
4834
4835 if (st->update_tail) {
4836 /* queue the recently created array / added disk
4837 * as a metadata update */
4838 int rv;
4839
4840 /* determine if we are creating a volume or adding a disk */
4841 if (current_vol < 0) {
4842 /* in the mgmt (add/remove) disk case we are running
4843 * in mdmon context, so don't close fd's
4844 */
4845 return mgmt_disk(st);
4846 } else
4847 rv = create_array(st, current_vol);
4848
4849 return rv;
4850 } else {
4851 struct dl *d;
4852 for (d = super->disks; d; d = d->next)
4853 Kill(d->devname, NULL, 0, 1, 1);
4854 return write_super_imsm(st, 1);
4855 }
4856 }
4857 #endif
4858
4859 static int store_super_imsm(struct supertype *st, int fd)
4860 {
4861 struct intel_super *super = st->sb;
4862 struct imsm_super *mpb = super ? super->anchor : NULL;
4863
4864 if (!mpb)
4865 return 1;
4866
4867 #ifndef MDASSEMBLE
4868 return store_imsm_mpb(fd, mpb);
4869 #else
4870 return 1;
4871 #endif
4872 }
4873
4874 static int imsm_bbm_log_size(struct imsm_super *mpb)
4875 {
4876 return __le32_to_cpu(mpb->bbm_log_size);
4877 }
4878
4879 #ifndef MDASSEMBLE
4880 static int validate_geometry_imsm_container(struct supertype *st, int level,
4881 int layout, int raiddisks, int chunk,
4882 unsigned long long size, char *dev,
4883 unsigned long long *freesize,
4884 int verbose)
4885 {
4886 int fd;
4887 unsigned long long ldsize;
4888 struct intel_super *super=NULL;
4889 int rv = 0;
4890
4891 if (level != LEVEL_CONTAINER)
4892 return 0;
4893 if (!dev)
4894 return 1;
4895
4896 fd = open(dev, O_RDONLY|O_EXCL, 0);
4897 if (fd < 0) {
4898 if (verbose)
4899 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
4900 dev, strerror(errno));
4901 return 0;
4902 }
4903 if (!get_dev_size(fd, dev, &ldsize)) {
4904 close(fd);
4905 return 0;
4906 }
4907
4908 /* capabilities retrieve could be possible
4909 * note that there is no fd for the disks in array.
4910 */
4911 super = alloc_super();
4912 if (!super) {
4913 fprintf(stderr,
4914 Name ": malloc of %zu failed.\n",
4915 sizeof(*super));
4916 close(fd);
4917 return 0;
4918 }
4919
4920 rv = find_intel_hba_capability(fd, super, verbose ? dev : NULL);
4921 if (rv != 0) {
4922 #if DEBUG
4923 char str[256];
4924 fd2devname(fd, str);
4925 dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n",
4926 fd, str, super->orom, rv, raiddisks);
4927 #endif
4928 /* no orom/efi or non-intel hba of the disk */
4929 close(fd);
4930 free_imsm(super);
4931 return 0;
4932 }
4933 close(fd);
4934 if (super->orom && raiddisks > super->orom->tds) {
4935 if (verbose)
4936 fprintf(stderr, Name ": %d exceeds maximum number of"
4937 " platform supported disks: %d\n",
4938 raiddisks, super->orom->tds);
4939
4940 free_imsm(super);
4941 return 0;
4942 }
4943
4944 *freesize = avail_size_imsm(st, ldsize >> 9);
4945 free_imsm(super);
4946
4947 return 1;
4948 }
4949
4950 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
4951 {
4952 const unsigned long long base_start = e[*idx].start;
4953 unsigned long long end = base_start + e[*idx].size;
4954 int i;
4955
4956 if (base_start == end)
4957 return 0;
4958
4959 *idx = *idx + 1;
4960 for (i = *idx; i < num_extents; i++) {
4961 /* extend overlapping extents */
4962 if (e[i].start >= base_start &&
4963 e[i].start <= end) {
4964 if (e[i].size == 0)
4965 return 0;
4966 if (e[i].start + e[i].size > end)
4967 end = e[i].start + e[i].size;
4968 } else if (e[i].start > end) {
4969 *idx = i;
4970 break;
4971 }
4972 }
4973
4974 return end - base_start;
4975 }
4976
4977 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
4978 {
4979 /* build a composite disk with all known extents and generate a new
4980 * 'maxsize' given the "all disks in an array must share a common start
4981 * offset" constraint
4982 */
4983 struct extent *e = calloc(sum_extents, sizeof(*e));
4984 struct dl *dl;
4985 int i, j;
4986 int start_extent;
4987 unsigned long long pos;
4988 unsigned long long start = 0;
4989 unsigned long long maxsize;
4990 unsigned long reserve;
4991
4992 if (!e)
4993 return 0;
4994
4995 /* coalesce and sort all extents. also, check to see if we need to
4996 * reserve space between member arrays
4997 */
4998 j = 0;
4999 for (dl = super->disks; dl; dl = dl->next) {
5000 if (!dl->e)
5001 continue;
5002 for (i = 0; i < dl->extent_cnt; i++)
5003 e[j++] = dl->e[i];
5004 }
5005 qsort(e, sum_extents, sizeof(*e), cmp_extent);
5006
5007 /* merge extents */
5008 i = 0;
5009 j = 0;
5010 while (i < sum_extents) {
5011 e[j].start = e[i].start;
5012 e[j].size = find_size(e, &i, sum_extents);
5013 j++;
5014 if (e[j-1].size == 0)
5015 break;
5016 }
5017
5018 pos = 0;
5019 maxsize = 0;
5020 start_extent = 0;
5021 i = 0;
5022 do {
5023 unsigned long long esize;
5024
5025 esize = e[i].start - pos;
5026 if (esize >= maxsize) {
5027 maxsize = esize;
5028 start = pos;
5029 start_extent = i;
5030 }
5031 pos = e[i].start + e[i].size;
5032 i++;
5033 } while (e[i-1].size);
5034 free(e);
5035
5036 if (maxsize == 0)
5037 return 0;
5038
5039 /* FIXME assumes volume at offset 0 is the first volume in a
5040 * container
5041 */
5042 if (start_extent > 0)
5043 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
5044 else
5045 reserve = 0;
5046
5047 if (maxsize < reserve)
5048 return 0;
5049
5050 super->create_offset = ~((__u32) 0);
5051 if (start + reserve > super->create_offset)
5052 return 0; /* start overflows create_offset */
5053 super->create_offset = start + reserve;
5054
5055 return maxsize - reserve;
5056 }
5057
5058 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
5059 {
5060 if (level < 0 || level == 6 || level == 4)
5061 return 0;
5062
5063 /* if we have an orom prevent invalid raid levels */
5064 if (orom)
5065 switch (level) {
5066 case 0: return imsm_orom_has_raid0(orom);
5067 case 1:
5068 if (raiddisks > 2)
5069 return imsm_orom_has_raid1e(orom);
5070 return imsm_orom_has_raid1(orom) && raiddisks == 2;
5071 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
5072 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
5073 }
5074 else
5075 return 1; /* not on an Intel RAID platform so anything goes */
5076
5077 return 0;
5078 }
5079
5080 static int imsm_default_chunk(const struct imsm_orom *orom)
5081 {
5082 /* up to 512 if the plaform supports it, otherwise the platform max.
5083 * 128 if no platform detected
5084 */
5085 int fs = max(7, orom ? fls(orom->sss) : 0);
5086
5087 return min(512, (1 << fs));
5088 }
5089
5090 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
5091 static int
5092 validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
5093 int raiddisks, int *chunk, int verbose)
5094 {
5095 /* check/set platform and metadata limits/defaults */
5096 if (super->orom && raiddisks > super->orom->dpa) {
5097 pr_vrb(": platform supports a maximum of %d disks per array\n",
5098 super->orom->dpa);
5099 return 0;
5100 }
5101
5102 /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */
5103 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
5104 pr_vrb(": platform does not support raid%d with %d disk%s\n",
5105 level, raiddisks, raiddisks > 1 ? "s" : "");
5106 return 0;
5107 }
5108
5109 if (chunk && (*chunk == 0 || *chunk == UnSet))
5110 *chunk = imsm_default_chunk(super->orom);
5111
5112 if (super->orom && chunk && !imsm_orom_has_chunk(super->orom, *chunk)) {
5113 pr_vrb(": platform does not support a chunk size of: "
5114 "%d\n", *chunk);
5115 return 0;
5116 }
5117
5118 if (layout != imsm_level_to_layout(level)) {
5119 if (level == 5)
5120 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
5121 else if (level == 10)
5122 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
5123 else
5124 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
5125 layout, level);
5126 return 0;
5127 }
5128 return 1;
5129 }
5130
5131 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
5132 * FIX ME add ahci details
5133 */
5134 static int validate_geometry_imsm_volume(struct supertype *st, int level,
5135 int layout, int raiddisks, int *chunk,
5136 unsigned long long size, char *dev,
5137 unsigned long long *freesize,
5138 int verbose)
5139 {
5140 struct stat stb;
5141 struct intel_super *super = st->sb;
5142 struct imsm_super *mpb = super->anchor;
5143 struct dl *dl;
5144 unsigned long long pos = 0;
5145 unsigned long long maxsize;
5146 struct extent *e;
5147 int i;
5148
5149 /* We must have the container info already read in. */
5150 if (!super)
5151 return 0;
5152
5153 if (mpb->num_raid_devs > 0 && mpb->num_disks != raiddisks) {
5154 fprintf(stderr, Name ": the option-rom requires all "
5155 "member disks to be a member of all volumes.\n");
5156 return 0;
5157 }
5158
5159 if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose)) {
5160 fprintf(stderr, Name ": RAID gemetry validation failed. "
5161 "Cannot proceed with the action(s).\n");
5162 return 0;
5163 }
5164 if (!dev) {
5165 /* General test: make sure there is space for
5166 * 'raiddisks' device extents of size 'size' at a given
5167 * offset
5168 */
5169 unsigned long long minsize = size;
5170 unsigned long long start_offset = MaxSector;
5171 int dcnt = 0;
5172 if (minsize == 0)
5173 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
5174 for (dl = super->disks; dl ; dl = dl->next) {
5175 int found = 0;
5176
5177 pos = 0;
5178 i = 0;
5179 e = get_extents(super, dl);
5180 if (!e) continue;
5181 do {
5182 unsigned long long esize;
5183 esize = e[i].start - pos;
5184 if (esize >= minsize)
5185 found = 1;
5186 if (found && start_offset == MaxSector) {
5187 start_offset = pos;
5188 break;
5189 } else if (found && pos != start_offset) {
5190 found = 0;
5191 break;
5192 }
5193 pos = e[i].start + e[i].size;
5194 i++;
5195 } while (e[i-1].size);
5196 if (found)
5197 dcnt++;
5198 free(e);
5199 }
5200 if (dcnt < raiddisks) {
5201 if (verbose)
5202 fprintf(stderr, Name ": imsm: Not enough "
5203 "devices with space for this array "
5204 "(%d < %d)\n",
5205 dcnt, raiddisks);
5206 return 0;
5207 }
5208 return 1;
5209 }
5210
5211 /* This device must be a member of the set */
5212 if (stat(dev, &stb) < 0)
5213 return 0;
5214 if ((S_IFMT & stb.st_mode) != S_IFBLK)
5215 return 0;
5216 for (dl = super->disks ; dl ; dl = dl->next) {
5217 if (dl->major == (int)major(stb.st_rdev) &&
5218 dl->minor == (int)minor(stb.st_rdev))
5219 break;
5220 }
5221 if (!dl) {
5222 if (verbose)
5223 fprintf(stderr, Name ": %s is not in the "
5224 "same imsm set\n", dev);
5225 return 0;
5226 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
5227 /* If a volume is present then the current creation attempt
5228 * cannot incorporate new spares because the orom may not
5229 * understand this configuration (all member disks must be
5230 * members of each array in the container).
5231 */
5232 fprintf(stderr, Name ": %s is a spare and a volume"
5233 " is already defined for this container\n", dev);
5234 fprintf(stderr, Name ": The option-rom requires all member"
5235 " disks to be a member of all volumes\n");
5236 return 0;
5237 }
5238
5239 /* retrieve the largest free space block */
5240 e = get_extents(super, dl);
5241 maxsize = 0;
5242 i = 0;
5243 if (e) {
5244 do {
5245 unsigned long long esize;
5246
5247 esize = e[i].start - pos;
5248 if (esize >= maxsize)
5249 maxsize = esize;
5250 pos = e[i].start + e[i].size;
5251 i++;
5252 } while (e[i-1].size);
5253 dl->e = e;
5254 dl->extent_cnt = i;
5255 } else {
5256 if (verbose)
5257 fprintf(stderr, Name ": unable to determine free space for: %s\n",
5258 dev);
5259 return 0;
5260 }
5261 if (maxsize < size) {
5262 if (verbose)
5263 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
5264 dev, maxsize, size);
5265 return 0;
5266 }
5267
5268 /* count total number of extents for merge */
5269 i = 0;
5270 for (dl = super->disks; dl; dl = dl->next)
5271 if (dl->e)
5272 i += dl->extent_cnt;
5273
5274 maxsize = merge_extents(super, i);
5275 if (maxsize < size || maxsize == 0) {
5276 if (verbose)
5277 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
5278 maxsize, size);
5279 return 0;
5280 }
5281
5282 *freesize = maxsize;
5283
5284 return 1;
5285 }
5286
5287 static int reserve_space(struct supertype *st, int raiddisks,
5288 unsigned long long size, int chunk,
5289 unsigned long long *freesize)
5290 {
5291 struct intel_super *super = st->sb;
5292 struct imsm_super *mpb = super->anchor;
5293 struct dl *dl;
5294 int i;
5295 int extent_cnt;
5296 struct extent *e;
5297 unsigned long long maxsize;
5298 unsigned long long minsize;
5299 int cnt;
5300 int used;
5301
5302 /* find the largest common start free region of the possible disks */
5303 used = 0;
5304 extent_cnt = 0;
5305 cnt = 0;
5306 for (dl = super->disks; dl; dl = dl->next) {
5307 dl->raiddisk = -1;
5308
5309 if (dl->index >= 0)
5310 used++;
5311
5312 /* don't activate new spares if we are orom constrained
5313 * and there is already a volume active in the container
5314 */
5315 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
5316 continue;
5317
5318 e = get_extents(super, dl);
5319 if (!e)
5320 continue;
5321 for (i = 1; e[i-1].size; i++)
5322 ;
5323 dl->e = e;
5324 dl->extent_cnt = i;
5325 extent_cnt += i;
5326 cnt++;
5327 }
5328
5329 maxsize = merge_extents(super, extent_cnt);
5330 minsize = size;
5331 if (size == 0)
5332 /* chunk is in K */
5333 minsize = chunk * 2;
5334
5335 if (cnt < raiddisks ||
5336 (super->orom && used && used != raiddisks) ||
5337 maxsize < minsize ||
5338 maxsize == 0) {
5339 fprintf(stderr, Name ": not enough devices with space to create array.\n");
5340 return 0; /* No enough free spaces large enough */
5341 }
5342
5343 if (size == 0) {
5344 size = maxsize;
5345 if (chunk) {
5346 size /= 2 * chunk;
5347 size *= 2 * chunk;
5348 }
5349 }
5350
5351 cnt = 0;
5352 for (dl = super->disks; dl; dl = dl->next)
5353 if (dl->e)
5354 dl->raiddisk = cnt++;
5355
5356 *freesize = size;
5357
5358 return 1;
5359 }
5360
5361 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
5362 int raiddisks, int *chunk, unsigned long long size,
5363 char *dev, unsigned long long *freesize,
5364 int verbose)
5365 {
5366 int fd, cfd;
5367 struct mdinfo *sra;
5368 int is_member = 0;
5369
5370 /* load capability
5371 * if given unused devices create a container
5372 * if given given devices in a container create a member volume
5373 */
5374 if (level == LEVEL_CONTAINER) {
5375 /* Must be a fresh device to add to a container */
5376 return validate_geometry_imsm_container(st, level, layout,
5377 raiddisks,
5378 chunk?*chunk:0, size,
5379 dev, freesize,
5380 verbose);
5381 }
5382
5383 if (!dev) {
5384 if (st->sb && freesize) {
5385 /* we are being asked to automatically layout a
5386 * new volume based on the current contents of
5387 * the container. If the the parameters can be
5388 * satisfied reserve_space will record the disks,
5389 * start offset, and size of the volume to be
5390 * created. add_to_super and getinfo_super
5391 * detect when autolayout is in progress.
5392 */
5393 if (!validate_geometry_imsm_orom(st->sb, level, layout,
5394 raiddisks, chunk,
5395 verbose))
5396 return 0;
5397 return reserve_space(st, raiddisks, size,
5398 chunk?*chunk:0, freesize);
5399 }
5400 return 1;
5401 }
5402 if (st->sb) {
5403 /* creating in a given container */
5404 return validate_geometry_imsm_volume(st, level, layout,
5405 raiddisks, chunk, size,
5406 dev, freesize, verbose);
5407 }
5408
5409 /* This device needs to be a device in an 'imsm' container */
5410 fd = open(dev, O_RDONLY|O_EXCL, 0);
5411 if (fd >= 0) {
5412 if (verbose)
5413 fprintf(stderr,
5414 Name ": Cannot create this array on device %s\n",
5415 dev);
5416 close(fd);
5417 return 0;
5418 }
5419 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
5420 if (verbose)
5421 fprintf(stderr, Name ": Cannot open %s: %s\n",
5422 dev, strerror(errno));
5423 return 0;
5424 }
5425 /* Well, it is in use by someone, maybe an 'imsm' container. */
5426 cfd = open_container(fd);
5427 close(fd);
5428 if (cfd < 0) {
5429 if (verbose)
5430 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
5431 dev);
5432 return 0;
5433 }
5434 sra = sysfs_read(cfd, 0, GET_VERSION);
5435 if (sra && sra->array.major_version == -1 &&
5436 strcmp(sra->text_version, "imsm") == 0)
5437 is_member = 1;
5438 sysfs_free(sra);
5439 if (is_member) {
5440 /* This is a member of a imsm container. Load the container
5441 * and try to create a volume
5442 */
5443 struct intel_super *super;
5444
5445 if (load_super_imsm_all(st, cfd, (void **) &super, NULL) == 0) {
5446 st->sb = super;
5447 st->container_dev = fd2devnum(cfd);
5448 close(cfd);
5449 return validate_geometry_imsm_volume(st, level, layout,
5450 raiddisks, chunk,
5451 size, dev,
5452 freesize, 1)
5453 ? 1 : -1;
5454 }
5455 }
5456
5457 if (verbose)
5458 fprintf(stderr, Name ": failed container membership check\n");
5459
5460 close(cfd);
5461 return 0;
5462 }
5463
5464 static void default_geometry_imsm(struct supertype *st, int *level, int *layout, int *chunk)
5465 {
5466 struct intel_super *super = st->sb;
5467
5468 if (level && *level == UnSet)
5469 *level = LEVEL_CONTAINER;
5470
5471 if (level && layout && *layout == UnSet)
5472 *layout = imsm_level_to_layout(*level);
5473
5474 if (chunk && (*chunk == UnSet || *chunk == 0))
5475 *chunk = imsm_default_chunk(super->orom);
5476 }
5477
5478 static void handle_missing(struct intel_super *super, struct imsm_dev *dev);
5479
5480 static int kill_subarray_imsm(struct supertype *st)
5481 {
5482 /* remove the subarray currently referenced by ->current_vol */
5483 __u8 i;
5484 struct intel_dev **dp;
5485 struct intel_super *super = st->sb;
5486 __u8 current_vol = super->current_vol;
5487 struct imsm_super *mpb = super->anchor;
5488
5489 if (super->current_vol < 0)
5490 return 2;
5491 super->current_vol = -1; /* invalidate subarray cursor */
5492
5493 /* block deletions that would change the uuid of active subarrays
5494 *
5495 * FIXME when immutable ids are available, but note that we'll
5496 * also need to fixup the invalidated/active subarray indexes in
5497 * mdstat
5498 */
5499 for (i = 0; i < mpb->num_raid_devs; i++) {
5500 char subarray[4];
5501
5502 if (i < current_vol)
5503 continue;
5504 sprintf(subarray, "%u", i);
5505 if (is_subarray_active(subarray, st->devname)) {
5506 fprintf(stderr,
5507 Name ": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n",
5508 current_vol, i);
5509
5510 return 2;
5511 }
5512 }
5513
5514 if (st->update_tail) {
5515 struct imsm_update_kill_array *u = malloc(sizeof(*u));
5516
5517 if (!u)
5518 return 2;
5519 u->type = update_kill_array;
5520 u->dev_idx = current_vol;
5521 append_metadata_update(st, u, sizeof(*u));
5522
5523 return 0;
5524 }
5525
5526 for (dp = &super->devlist; *dp;)
5527 if ((*dp)->index == current_vol) {
5528 *dp = (*dp)->next;
5529 } else {
5530 handle_missing(super, (*dp)->dev);
5531 if ((*dp)->index > current_vol)
5532 (*dp)->index--;
5533 dp = &(*dp)->next;
5534 }
5535
5536 /* no more raid devices, all active components are now spares,
5537 * but of course failed are still failed
5538 */
5539 if (--mpb->num_raid_devs == 0) {
5540 struct dl *d;
5541
5542 for (d = super->disks; d; d = d->next)
5543 if (d->index > -2)
5544 mark_spare(d);
5545 }
5546
5547 super->updates_pending++;
5548
5549 return 0;
5550 }
5551
5552 static int update_subarray_imsm(struct supertype *st, char *subarray,
5553 char *update, struct mddev_ident *ident)
5554 {
5555 /* update the subarray currently referenced by ->current_vol */
5556 struct intel_super *super = st->sb;
5557 struct imsm_super *mpb = super->anchor;
5558
5559 if (strcmp(update, "name") == 0) {
5560 char *name = ident->name;
5561 char *ep;
5562 int vol;
5563
5564 if (is_subarray_active(subarray, st->devname)) {
5565 fprintf(stderr,
5566 Name ": Unable to update name of active subarray\n");
5567 return 2;
5568 }
5569
5570 if (!check_name(super, name, 0))
5571 return 2;
5572
5573 vol = strtoul(subarray, &ep, 10);
5574 if (*ep != '\0' || vol >= super->anchor->num_raid_devs)
5575 return 2;
5576
5577 if (st->update_tail) {
5578 struct imsm_update_rename_array *u = malloc(sizeof(*u));
5579
5580 if (!u)
5581 return 2;
5582 u->type = update_rename_array;
5583 u->dev_idx = vol;
5584 snprintf((char *) u->name, MAX_RAID_SERIAL_LEN, "%s", name);
5585 append_metadata_update(st, u, sizeof(*u));
5586 } else {
5587 struct imsm_dev *dev;
5588 int i;
5589
5590 dev = get_imsm_dev(super, vol);
5591 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
5592 for (i = 0; i < mpb->num_raid_devs; i++) {
5593 dev = get_imsm_dev(super, i);
5594 handle_missing(super, dev);
5595 }
5596 super->updates_pending++;
5597 }
5598 } else
5599 return 2;
5600
5601 return 0;
5602 }
5603
5604 static int is_gen_migration(struct imsm_dev *dev)
5605 {
5606 if (dev == NULL)
5607 return 0;
5608
5609 if (!dev->vol.migr_state)
5610 return 0;
5611
5612 if (migr_type(dev) == MIGR_GEN_MIGR)
5613 return 1;
5614
5615 return 0;
5616 }
5617 #endif /* MDASSEMBLE */
5618
5619 static int is_rebuilding(struct imsm_dev *dev)
5620 {
5621 struct imsm_map *migr_map;
5622
5623 if (!dev->vol.migr_state)
5624 return 0;
5625
5626 if (migr_type(dev) != MIGR_REBUILD)
5627 return 0;
5628
5629 migr_map = get_imsm_map(dev, 1);
5630
5631 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
5632 return 1;
5633 else
5634 return 0;
5635 }
5636
5637 static void update_recovery_start(struct intel_super *super,
5638 struct imsm_dev *dev,
5639 struct mdinfo *array)
5640 {
5641 struct mdinfo *rebuild = NULL;
5642 struct mdinfo *d;
5643 __u32 units;
5644
5645 if (!is_rebuilding(dev))
5646 return;
5647
5648 /* Find the rebuild target, but punt on the dual rebuild case */
5649 for (d = array->devs; d; d = d->next)
5650 if (d->recovery_start == 0) {
5651 if (rebuild)
5652 return;
5653 rebuild = d;
5654 }
5655
5656 if (!rebuild) {
5657 /* (?) none of the disks are marked with
5658 * IMSM_ORD_REBUILD, so assume they are missing and the
5659 * disk_ord_tbl was not correctly updated
5660 */
5661 dprintf("%s: failed to locate out-of-sync disk\n", __func__);
5662 return;
5663 }
5664
5665 units = __le32_to_cpu(dev->vol.curr_migr_unit);
5666 rebuild->recovery_start = units * blocks_per_migr_unit(super, dev);
5667 }
5668
5669 #ifndef MDASSEMBLE
5670 static int recover_backup_imsm(struct supertype *st, struct mdinfo *info);
5671 #endif
5672
5673 static struct mdinfo *container_content_imsm(struct supertype *st, char *subarray)
5674 {
5675 /* Given a container loaded by load_super_imsm_all,
5676 * extract information about all the arrays into
5677 * an mdinfo tree.
5678 * If 'subarray' is given, just extract info about that array.
5679 *
5680 * For each imsm_dev create an mdinfo, fill it in,
5681 * then look for matching devices in super->disks
5682 * and create appropriate device mdinfo.
5683 */
5684 struct intel_super *super = st->sb;
5685 struct imsm_super *mpb = super->anchor;
5686 struct mdinfo *rest = NULL;
5687 unsigned int i;
5688 int sb_errors = 0;
5689 struct dl *d;
5690 int spare_disks = 0;
5691
5692 /* do not assemble arrays when not all attributes are supported */
5693 if (imsm_check_attributes(mpb->attributes) == 0) {
5694 sb_errors = 1;
5695 fprintf(stderr, Name ": Unsupported attributes in IMSM metadata."
5696 "Arrays activation is blocked.\n");
5697 }
5698
5699 /* check for bad blocks */
5700 if (imsm_bbm_log_size(super->anchor)) {
5701 fprintf(stderr, Name ": BBM log found in IMSM metadata."
5702 "Arrays activation is blocked.\n");
5703 sb_errors = 1;
5704 }
5705
5706
5707 /* count spare devices, not used in maps
5708 */
5709 for (d = super->disks; d; d = d->next)
5710 if (d->index == -1)
5711 spare_disks++;
5712
5713 for (i = 0; i < mpb->num_raid_devs; i++) {
5714 struct imsm_dev *dev;
5715 struct imsm_map *map;
5716 struct imsm_map *map2;
5717 struct mdinfo *this;
5718 int slot, chunk;
5719 char *ep;
5720
5721 if (subarray &&
5722 (i != strtoul(subarray, &ep, 10) || *ep != '\0'))
5723 continue;
5724
5725 dev = get_imsm_dev(super, i);
5726 map = get_imsm_map(dev, 0);
5727 map2 = get_imsm_map(dev, 1);
5728
5729 /* do not publish arrays that are in the middle of an
5730 * unsupported migration
5731 */
5732 if (dev->vol.migr_state &&
5733 (migr_type(dev) == MIGR_STATE_CHANGE)) {
5734 fprintf(stderr, Name ": cannot assemble volume '%.16s':"
5735 " unsupported migration in progress\n",
5736 dev->volume);
5737 continue;
5738 }
5739 /* do not publish arrays that are not support by controller's
5740 * OROM/EFI
5741 */
5742
5743 chunk = __le16_to_cpu(map->blocks_per_strip) >> 1;
5744 this = malloc(sizeof(*this));
5745 if (!this) {
5746 fprintf(stderr, Name ": failed to allocate %zu bytes\n",
5747 sizeof(*this));
5748 break;
5749 }
5750
5751 super->current_vol = i;
5752 getinfo_super_imsm_volume(st, this, NULL);
5753 this->next = rest;
5754 #ifndef MDASSEMBLE
5755 /* mdadm does not support all metadata features- set the bit in all arrays state */
5756 if (!validate_geometry_imsm_orom(super,
5757 get_imsm_raid_level(map), /* RAID level */
5758 imsm_level_to_layout(get_imsm_raid_level(map)),
5759 map->num_members, /* raid disks */
5760 &chunk,
5761 1 /* verbose */)) {
5762 fprintf(stderr, Name ": IMSM RAID geometry validation"
5763 " failed. Array %s activation is blocked.\n",
5764 dev->volume);
5765 this->array.state |=
5766 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE) |
5767 (1<<MD_SB_BLOCK_VOLUME);
5768 }
5769 #endif
5770
5771 /* if array has bad blocks, set suitable bit in all arrays state */
5772 if (sb_errors)
5773 this->array.state |=
5774 (1<<MD_SB_BLOCK_CONTAINER_RESHAPE) |
5775 (1<<MD_SB_BLOCK_VOLUME);
5776
5777 for (slot = 0 ; slot < map->num_members; slot++) {
5778 unsigned long long recovery_start;
5779 struct mdinfo *info_d;
5780 struct dl *d;
5781 int idx;
5782 int skip;
5783 __u32 ord;
5784
5785 skip = 0;
5786 idx = get_imsm_disk_idx(dev, slot, 0);
5787 ord = get_imsm_ord_tbl_ent(dev, slot, -1);
5788 for (d = super->disks; d ; d = d->next)
5789 if (d->index == idx)
5790 break;
5791
5792 recovery_start = MaxSector;
5793 if (d == NULL)
5794 skip = 1;
5795 if (d && is_failed(&d->disk))
5796 skip = 1;
5797 if (ord & IMSM_ORD_REBUILD)
5798 recovery_start = 0;
5799
5800 /*
5801 * if we skip some disks the array will be assmebled degraded;
5802 * reset resync start to avoid a dirty-degraded
5803 * situation when performing the intial sync
5804 *
5805 * FIXME handle dirty degraded
5806 */
5807 if ((skip || recovery_start == 0) && !dev->vol.dirty)
5808 this->resync_start = MaxSector;
5809 if (skip)
5810 continue;
5811
5812 info_d = calloc(1, sizeof(*info_d));
5813 if (!info_d) {
5814 fprintf(stderr, Name ": failed to allocate disk"
5815 " for volume %.16s\n", dev->volume);
5816 info_d = this->devs;
5817 while (info_d) {
5818 struct mdinfo *d = info_d->next;
5819
5820 free(info_d);
5821 info_d = d;
5822 }
5823 free(this);
5824 this = rest;
5825 break;
5826 }
5827 info_d->next = this->devs;
5828 this->devs = info_d;
5829
5830 info_d->disk.number = d->index;
5831 info_d->disk.major = d->major;
5832 info_d->disk.minor = d->minor;
5833 info_d->disk.raid_disk = slot;
5834 info_d->recovery_start = recovery_start;
5835 if (map2) {
5836 if (slot < map2->num_members)
5837 info_d->disk.state = (1 << MD_DISK_ACTIVE);
5838 else
5839 this->array.spare_disks++;
5840 } else {
5841 if (slot < map->num_members)
5842 info_d->disk.state = (1 << MD_DISK_ACTIVE);
5843 else
5844 this->array.spare_disks++;
5845 }
5846 if (info_d->recovery_start == MaxSector)
5847 this->array.working_disks++;
5848
5849 info_d->events = __le32_to_cpu(mpb->generation_num);
5850 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
5851 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
5852 }
5853 /* now that the disk list is up-to-date fixup recovery_start */
5854 update_recovery_start(super, dev, this);
5855 this->array.spare_disks += spare_disks;
5856
5857 #ifndef MDASSEMBLE
5858 /* check for reshape */
5859 if (this->reshape_active == 1)
5860 recover_backup_imsm(st, this);
5861 #endif
5862 rest = this;
5863 }
5864
5865 return rest;
5866 }
5867
5868
5869 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
5870 {
5871 struct imsm_map *map = get_imsm_map(dev, 0);
5872
5873 if (!failed)
5874 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
5875 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
5876
5877 switch (get_imsm_raid_level(map)) {
5878 case 0:
5879 return IMSM_T_STATE_FAILED;
5880 break;
5881 case 1:
5882 if (failed < map->num_members)
5883 return IMSM_T_STATE_DEGRADED;
5884 else
5885 return IMSM_T_STATE_FAILED;
5886 break;
5887 case 10:
5888 {
5889 /**
5890 * check to see if any mirrors have failed, otherwise we
5891 * are degraded. Even numbered slots are mirrored on
5892 * slot+1
5893 */
5894 int i;
5895 /* gcc -Os complains that this is unused */
5896 int insync = insync;
5897
5898 for (i = 0; i < map->num_members; i++) {
5899 __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1);
5900 int idx = ord_to_idx(ord);
5901 struct imsm_disk *disk;
5902
5903 /* reset the potential in-sync count on even-numbered
5904 * slots. num_copies is always 2 for imsm raid10
5905 */
5906 if ((i & 1) == 0)
5907 insync = 2;
5908
5909 disk = get_imsm_disk(super, idx);
5910 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
5911 insync--;
5912
5913 /* no in-sync disks left in this mirror the
5914 * array has failed
5915 */
5916 if (insync == 0)
5917 return IMSM_T_STATE_FAILED;
5918 }
5919
5920 return IMSM_T_STATE_DEGRADED;
5921 }
5922 case 5:
5923 if (failed < 2)
5924 return IMSM_T_STATE_DEGRADED;
5925 else
5926 return IMSM_T_STATE_FAILED;
5927 break;
5928 default:
5929 break;
5930 }
5931
5932 return map->map_state;
5933 }
5934
5935 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
5936 {
5937 int i;
5938 int failed = 0;
5939 struct imsm_disk *disk;
5940 struct imsm_map *map = get_imsm_map(dev, 0);
5941 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
5942 __u32 ord;
5943 int idx;
5944
5945 /* at the beginning of migration we set IMSM_ORD_REBUILD on
5946 * disks that are being rebuilt. New failures are recorded to
5947 * map[0]. So we look through all the disks we started with and
5948 * see if any failures are still present, or if any new ones
5949 * have arrived
5950 *
5951 * FIXME add support for online capacity expansion and
5952 * raid-level-migration
5953 */
5954 for (i = 0; i < prev->num_members; i++) {
5955 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
5956 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
5957 idx = ord_to_idx(ord);
5958
5959 disk = get_imsm_disk(super, idx);
5960 if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD)
5961 failed++;
5962 }
5963
5964 return failed;
5965 }
5966
5967 #ifndef MDASSEMBLE
5968 static int imsm_open_new(struct supertype *c, struct active_array *a,
5969 char *inst)
5970 {
5971 struct intel_super *super = c->sb;
5972 struct imsm_super *mpb = super->anchor;
5973
5974 if (atoi(inst) >= mpb->num_raid_devs) {
5975 fprintf(stderr, "%s: subarry index %d, out of range\n",
5976 __func__, atoi(inst));
5977 return -ENODEV;
5978 }
5979
5980 dprintf("imsm: open_new %s\n", inst);
5981 a->info.container_member = atoi(inst);
5982 return 0;
5983 }
5984
5985 static int is_resyncing(struct imsm_dev *dev)
5986 {
5987 struct imsm_map *migr_map;
5988
5989 if (!dev->vol.migr_state)
5990 return 0;
5991
5992 if (migr_type(dev) == MIGR_INIT ||
5993 migr_type(dev) == MIGR_REPAIR)
5994 return 1;
5995
5996 if (migr_type(dev) == MIGR_GEN_MIGR)
5997 return 0;
5998
5999 migr_map = get_imsm_map(dev, 1);
6000
6001 if ((migr_map->map_state == IMSM_T_STATE_NORMAL) &&
6002 (dev->vol.migr_type != MIGR_GEN_MIGR))
6003 return 1;
6004 else
6005 return 0;
6006 }
6007
6008 /* return true if we recorded new information */
6009 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
6010 {
6011 __u32 ord;
6012 int slot;
6013 struct imsm_map *map;
6014 char buf[MAX_RAID_SERIAL_LEN+3];
6015 unsigned int len, shift = 0;
6016
6017 /* new failures are always set in map[0] */
6018 map = get_imsm_map(dev, 0);
6019
6020 slot = get_imsm_disk_slot(map, idx);
6021 if (slot < 0)
6022 return 0;
6023
6024 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
6025 if (is_failed(disk) && (ord & IMSM_ORD_REBUILD))
6026 return 0;
6027
6028 memcpy(buf, disk->serial, MAX_RAID_SERIAL_LEN);
6029 buf[MAX_RAID_SERIAL_LEN] = '\000';
6030 strcat(buf, ":0");
6031 if ((len = strlen(buf)) >= MAX_RAID_SERIAL_LEN)
6032 shift = len - MAX_RAID_SERIAL_LEN + 1;
6033 strncpy((char *)disk->serial, &buf[shift], MAX_RAID_SERIAL_LEN);
6034
6035 disk->status |= FAILED_DISK;
6036 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
6037 if (map->failed_disk_num == 0xff)
6038 map->failed_disk_num = slot;
6039 return 1;
6040 }
6041
6042 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
6043 {
6044 mark_failure(dev, disk, idx);
6045
6046 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
6047 return;
6048
6049 disk->scsi_id = __cpu_to_le32(~(__u32)0);
6050 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
6051 }
6052
6053 static void handle_missing(struct intel_super *super, struct imsm_dev *dev)
6054 {
6055 __u8 map_state;
6056 struct dl *dl;
6057 int failed;
6058
6059 if (!super->missing)
6060 return;
6061 failed = imsm_count_failed(super, dev);
6062 map_state = imsm_check_degraded(super, dev, failed);
6063
6064 dprintf("imsm: mark missing\n");
6065 end_migration(dev, map_state);
6066 for (dl = super->missing; dl; dl = dl->next)
6067 mark_missing(dev, &dl->disk, dl->index);
6068 super->updates_pending++;
6069 }
6070
6071 static unsigned long long imsm_set_array_size(struct imsm_dev *dev)
6072 {
6073 int used_disks = imsm_num_data_members(dev, 0);
6074 unsigned long long array_blocks;
6075 struct imsm_map *map;
6076
6077 if (used_disks == 0) {
6078 /* when problems occures
6079 * return current array_blocks value
6080 */
6081 array_blocks = __le32_to_cpu(dev->size_high);
6082 array_blocks = array_blocks << 32;
6083 array_blocks += __le32_to_cpu(dev->size_low);
6084
6085 return array_blocks;
6086 }
6087
6088 /* set array size in metadata
6089 */
6090 map = get_imsm_map(dev, 0);
6091 array_blocks = map->blocks_per_member * used_disks;
6092
6093 /* round array size down to closest MB
6094 */
6095 array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT;
6096 dev->size_low = __cpu_to_le32((__u32)array_blocks);
6097 dev->size_high = __cpu_to_le32((__u32)(array_blocks >> 32));
6098
6099 return array_blocks;
6100 }
6101
6102 static void imsm_set_disk(struct active_array *a, int n, int state);
6103
6104 static void imsm_progress_container_reshape(struct intel_super *super)
6105 {
6106 /* if no device has a migr_state, but some device has a
6107 * different number of members than the previous device, start
6108 * changing the number of devices in this device to match
6109 * previous.
6110 */
6111 struct imsm_super *mpb = super->anchor;
6112 int prev_disks = -1;
6113 int i;
6114 int copy_map_size;
6115
6116 for (i = 0; i < mpb->num_raid_devs; i++) {
6117 struct imsm_dev *dev = get_imsm_dev(super, i);
6118 struct imsm_map *map = get_imsm_map(dev, 0);
6119 struct imsm_map *map2;
6120 int prev_num_members;
6121
6122 if (dev->vol.migr_state)
6123 return;
6124
6125 if (prev_disks == -1)
6126 prev_disks = map->num_members;
6127 if (prev_disks == map->num_members)
6128 continue;
6129
6130 /* OK, this array needs to enter reshape mode.
6131 * i.e it needs a migr_state
6132 */
6133
6134 copy_map_size = sizeof_imsm_map(map);
6135 prev_num_members = map->num_members;
6136 map->num_members = prev_disks;
6137 dev->vol.migr_state = 1;
6138 dev->vol.curr_migr_unit = 0;
6139 set_migr_type(dev, MIGR_GEN_MIGR);
6140 for (i = prev_num_members;
6141 i < map->num_members; i++)
6142 set_imsm_ord_tbl_ent(map, i, i);
6143 map2 = get_imsm_map(dev, 1);
6144 /* Copy the current map */
6145 memcpy(map2, map, copy_map_size);
6146 map2->num_members = prev_num_members;
6147
6148 imsm_set_array_size(dev);
6149 super->updates_pending++;
6150 }
6151 }
6152
6153 /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild
6154 * states are handled in imsm_set_disk() with one exception, when a
6155 * resync is stopped due to a new failure this routine will set the
6156 * 'degraded' state for the array.
6157 */
6158 static int imsm_set_array_state(struct active_array *a, int consistent)
6159 {
6160 int inst = a->info.container_member;
6161 struct intel_super *super = a->container->sb;
6162 struct imsm_dev *dev = get_imsm_dev(super, inst);
6163 struct imsm_map *map = get_imsm_map(dev, 0);
6164 int failed = imsm_count_failed(super, dev);
6165 __u8 map_state = imsm_check_degraded(super, dev, failed);
6166 __u32 blocks_per_unit;
6167
6168 if (dev->vol.migr_state &&
6169 dev->vol.migr_type == MIGR_GEN_MIGR) {
6170 /* array state change is blocked due to reshape action
6171 * We might need to
6172 * - abort the reshape (if last_checkpoint is 0 and action!= reshape)
6173 * - finish the reshape (if last_checkpoint is big and action != reshape)
6174 * - update curr_migr_unit
6175 */
6176 if (a->curr_action == reshape) {
6177 /* still reshaping, maybe update curr_migr_unit */
6178 goto mark_checkpoint;
6179 } else {
6180 if (a->last_checkpoint == 0 && a->prev_action == reshape) {
6181 /* for some reason we aborted the reshape.
6182 *
6183 * disable automatic metadata rollback
6184 * user action is required to recover process
6185 */
6186 if (0) {
6187 struct imsm_map *map2 = get_imsm_map(dev, 1);
6188 dev->vol.migr_state = 0;
6189 set_migr_type(dev, 0);
6190 dev->vol.curr_migr_unit = 0;
6191 memcpy(map, map2, sizeof_imsm_map(map2));
6192 super->updates_pending++;
6193 }
6194 }
6195 if (a->last_checkpoint >= a->info.component_size) {
6196 unsigned long long array_blocks;
6197 int used_disks;
6198 struct mdinfo *mdi;
6199
6200 used_disks = imsm_num_data_members(dev, 0);
6201 if (used_disks > 0) {
6202 array_blocks =
6203 map->blocks_per_member *
6204 used_disks;
6205 /* round array size down to closest MB
6206 */
6207 array_blocks = (array_blocks
6208 >> SECT_PER_MB_SHIFT)
6209 << SECT_PER_MB_SHIFT;
6210 a->info.custom_array_size = array_blocks;
6211 /* encourage manager to update array
6212 * size
6213 */
6214
6215 a->check_reshape = 1;
6216 }
6217 /* finalize online capacity expansion/reshape */
6218 for (mdi = a->info.devs; mdi; mdi = mdi->next)
6219 imsm_set_disk(a,
6220 mdi->disk.raid_disk,
6221 mdi->curr_state);
6222
6223 imsm_progress_container_reshape(super);
6224 }
6225 }
6226 }
6227
6228 /* before we activate this array handle any missing disks */
6229 if (consistent == 2)
6230 handle_missing(super, dev);
6231
6232 if (consistent == 2 &&
6233 (!is_resync_complete(&a->info) ||
6234 map_state != IMSM_T_STATE_NORMAL ||
6235 dev->vol.migr_state))
6236 consistent = 0;
6237
6238 if (is_resync_complete(&a->info)) {
6239 /* complete intialization / resync,
6240 * recovery and interrupted recovery is completed in
6241 * ->set_disk
6242 */
6243 if (is_resyncing(dev)) {
6244 dprintf("imsm: mark resync done\n");
6245 end_migration(dev, map_state);
6246 super->updates_pending++;
6247 a->last_checkpoint = 0;
6248 }
6249 } else if ((!is_resyncing(dev) && !failed) &&
6250 (imsm_reshape_blocks_arrays_changes(super) == 0)) {
6251 /* mark the start of the init process if nothing is failed */
6252 dprintf("imsm: mark resync start\n");
6253 if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
6254 migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_INIT);
6255 else
6256 migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_REPAIR);
6257 super->updates_pending++;
6258 }
6259
6260 mark_checkpoint:
6261 /* skip checkpointing for general migration,
6262 * it is controlled in mdadm
6263 */
6264 if (is_gen_migration(dev))
6265 goto skip_mark_checkpoint;
6266
6267 /* check if we can update curr_migr_unit from resync_start, recovery_start */
6268 blocks_per_unit = blocks_per_migr_unit(super, dev);
6269 if (blocks_per_unit) {
6270 __u32 units32;
6271 __u64 units;
6272
6273 units = a->last_checkpoint / blocks_per_unit;
6274 units32 = units;
6275
6276 /* check that we did not overflow 32-bits, and that
6277 * curr_migr_unit needs updating
6278 */
6279 if (units32 == units &&
6280 units32 != 0 &&
6281 __le32_to_cpu(dev->vol.curr_migr_unit) != units32) {
6282 dprintf("imsm: mark checkpoint (%u)\n", units32);
6283 dev->vol.curr_migr_unit = __cpu_to_le32(units32);
6284 super->updates_pending++;
6285 }
6286 }
6287
6288 skip_mark_checkpoint:
6289 /* mark dirty / clean */
6290 if (dev->vol.dirty != !consistent) {
6291 dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty");
6292 if (consistent)
6293 dev->vol.dirty = 0;
6294 else
6295 dev->vol.dirty = 1;
6296 super->updates_pending++;
6297 }
6298
6299 return consistent;
6300 }
6301
6302 static void imsm_set_disk(struct active_array *a, int n, int state)
6303 {
6304 int inst = a->info.container_member;
6305 struct intel_super *super = a->container->sb;
6306 struct imsm_dev *dev = get_imsm_dev(super, inst);
6307 struct imsm_map *map = get_imsm_map(dev, 0);
6308 struct imsm_disk *disk;
6309 int failed;
6310 __u32 ord;
6311 __u8 map_state;
6312
6313 if (n > map->num_members)
6314 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
6315 n, map->num_members - 1);
6316
6317 if (n < 0)
6318 return;
6319
6320 dprintf("imsm: set_disk %d:%x\n", n, state);
6321
6322 ord = get_imsm_ord_tbl_ent(dev, n, -1);
6323 disk = get_imsm_disk(super, ord_to_idx(ord));
6324
6325 /* check for new failures */
6326 if (state & DS_FAULTY) {
6327 if (mark_failure(dev, disk, ord_to_idx(ord)))
6328 super->updates_pending++;
6329 }
6330
6331 /* check if in_sync */
6332 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
6333 struct imsm_map *migr_map = get_imsm_map(dev, 1);
6334
6335 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
6336 super->updates_pending++;
6337 }
6338
6339 failed = imsm_count_failed(super, dev);
6340 map_state = imsm_check_degraded(super, dev, failed);
6341
6342 /* check if recovery complete, newly degraded, or failed */
6343 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
6344 end_migration(dev, map_state);
6345 map = get_imsm_map(dev, 0);
6346 map->failed_disk_num = ~0;
6347 super->updates_pending++;
6348 a->last_checkpoint = 0;
6349 } else if (map_state == IMSM_T_STATE_DEGRADED &&
6350 map->map_state != map_state &&
6351 !dev->vol.migr_state) {
6352 dprintf("imsm: mark degraded\n");
6353 map->map_state = map_state;
6354 super->updates_pending++;
6355 a->last_checkpoint = 0;
6356 } else if (map_state == IMSM_T_STATE_FAILED &&
6357 map->map_state != map_state) {
6358 dprintf("imsm: mark failed\n");
6359 end_migration(dev, map_state);
6360 super->updates_pending++;
6361 a->last_checkpoint = 0;
6362 } else if (is_gen_migration(dev)) {
6363 dprintf("imsm: Detected General Migration in state: ");
6364 if (map_state == IMSM_T_STATE_NORMAL) {
6365 end_migration(dev, map_state);
6366 map = get_imsm_map(dev, 0);
6367 map->failed_disk_num = ~0;
6368 dprintf("normal\n");
6369 } else {
6370 if (map_state == IMSM_T_STATE_DEGRADED) {
6371 printf("degraded\n");
6372 end_migration(dev, map_state);
6373 } else {
6374 dprintf("failed\n");
6375 }
6376 map->map_state = map_state;
6377 }
6378 super->updates_pending++;
6379 }
6380 }
6381
6382 static int store_imsm_mpb(int fd, struct imsm_super *mpb)
6383 {
6384 void *buf = mpb;
6385 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
6386 unsigned long long dsize;
6387 unsigned long long sectors;
6388
6389 get_dev_size(fd, NULL, &dsize);
6390
6391 if (mpb_size > 512) {
6392 /* -1 to account for anchor */
6393 sectors = mpb_sectors(mpb) - 1;
6394
6395 /* write the extended mpb to the sectors preceeding the anchor */
6396 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
6397 return 1;
6398
6399 if ((unsigned long long)write(fd, buf + 512, 512 * sectors)
6400 != 512 * sectors)
6401 return 1;
6402 }
6403
6404 /* first block is stored on second to last sector of the disk */
6405 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
6406 return 1;
6407
6408 if (write(fd, buf, 512) != 512)
6409 return 1;
6410
6411 return 0;
6412 }
6413
6414 static void imsm_sync_metadata(struct supertype *container)
6415 {
6416 struct intel_super *super = container->sb;
6417
6418 dprintf("sync metadata: %d\n", super->updates_pending);
6419 if (!super->updates_pending)
6420 return;
6421
6422 write_super_imsm(container, 0);
6423
6424 super->updates_pending = 0;
6425 }
6426
6427 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
6428 {
6429 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
6430 int i = get_imsm_disk_idx(dev, idx, -1);
6431 struct dl *dl;
6432
6433 for (dl = super->disks; dl; dl = dl->next)
6434 if (dl->index == i)
6435 break;
6436
6437 if (dl && is_failed(&dl->disk))
6438 dl = NULL;
6439
6440 if (dl)
6441 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
6442
6443 return dl;
6444 }
6445
6446 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
6447 struct active_array *a, int activate_new,
6448 struct mdinfo *additional_test_list)
6449 {
6450 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
6451 int idx = get_imsm_disk_idx(dev, slot, -1);
6452 struct imsm_super *mpb = super->anchor;
6453 struct imsm_map *map;
6454 unsigned long long pos;
6455 struct mdinfo *d;
6456 struct extent *ex;
6457 int i, j;
6458 int found;
6459 __u32 array_start = 0;
6460 __u32 array_end = 0;
6461 struct dl *dl;
6462 struct mdinfo *test_list;
6463
6464 for (dl = super->disks; dl; dl = dl->next) {
6465 /* If in this array, skip */
6466 for (d = a->info.devs ; d ; d = d->next)
6467 if (d->state_fd >= 0 &&
6468 d->disk.major == dl->major &&
6469 d->disk.minor == dl->minor) {
6470 dprintf("%x:%x already in array\n",
6471 dl->major, dl->minor);
6472 break;
6473 }
6474 if (d)
6475 continue;
6476 test_list = additional_test_list;
6477 while (test_list) {
6478 if (test_list->disk.major == dl->major &&
6479 test_list->disk.minor == dl->minor) {
6480 dprintf("%x:%x already in additional test list\n",
6481 dl->major, dl->minor);
6482 break;
6483 }
6484 test_list = test_list->next;
6485 }
6486 if (test_list)
6487 continue;
6488
6489 /* skip in use or failed drives */
6490 if (is_failed(&dl->disk) || idx == dl->index ||
6491 dl->index == -2) {
6492 dprintf("%x:%x status (failed: %d index: %d)\n",
6493 dl->major, dl->minor, is_failed(&dl->disk), idx);
6494 continue;
6495 }
6496
6497 /* skip pure spares when we are looking for partially
6498 * assimilated drives
6499 */
6500 if (dl->index == -1 && !activate_new)
6501 continue;
6502
6503 /* Does this unused device have the requisite free space?
6504 * It needs to be able to cover all member volumes
6505 */
6506 ex = get_extents(super, dl);
6507 if (!ex) {
6508 dprintf("cannot get extents\n");
6509 continue;
6510 }
6511 for (i = 0; i < mpb->num_raid_devs; i++) {
6512 dev = get_imsm_dev(super, i);
6513 map = get_imsm_map(dev, 0);
6514
6515 /* check if this disk is already a member of
6516 * this array
6517 */
6518 if (get_imsm_disk_slot(map, dl->index) >= 0)
6519 continue;
6520
6521 found = 0;
6522 j = 0;
6523 pos = 0;
6524 array_start = __le32_to_cpu(map->pba_of_lba0);
6525 array_end = array_start +
6526 __le32_to_cpu(map->blocks_per_member) - 1;
6527
6528 do {
6529 /* check that we can start at pba_of_lba0 with
6530 * blocks_per_member of space
6531 */
6532 if (array_start >= pos && array_end < ex[j].start) {
6533 found = 1;
6534 break;
6535 }
6536 pos = ex[j].start + ex[j].size;
6537 j++;
6538 } while (ex[j-1].size);
6539
6540 if (!found)
6541 break;
6542 }
6543
6544 free(ex);
6545 if (i < mpb->num_raid_devs) {
6546 dprintf("%x:%x does not have %u to %u available\n",
6547 dl->major, dl->minor, array_start, array_end);
6548 /* No room */
6549 continue;
6550 }
6551 return dl;
6552 }
6553
6554 return dl;
6555 }
6556
6557
6558 static int imsm_rebuild_allowed(struct supertype *cont, int dev_idx, int failed)
6559 {
6560 struct imsm_dev *dev2;
6561 struct imsm_map *map;
6562 struct dl *idisk;
6563 int slot;
6564 int idx;
6565 __u8 state;
6566
6567 dev2 = get_imsm_dev(cont->sb, dev_idx);
6568 if (dev2) {
6569 state = imsm_check_degraded(cont->sb, dev2, failed);
6570 if (state == IMSM_T_STATE_FAILED) {
6571 map = get_imsm_map(dev2, 0);
6572 if (!map)
6573 return 1;
6574 for (slot = 0; slot < map->num_members; slot++) {
6575 /*
6576 * Check if failed disks are deleted from intel
6577 * disk list or are marked to be deleted
6578 */
6579 idx = get_imsm_disk_idx(dev2, slot, -1);
6580 idisk = get_imsm_dl_disk(cont->sb, idx);
6581 /*
6582 * Do not rebuild the array if failed disks
6583 * from failed sub-array are not removed from
6584 * container.
6585 */
6586 if (idisk &&
6587 is_failed(&idisk->disk) &&
6588 (idisk->action != DISK_REMOVE))
6589 return 0;
6590 }
6591 }
6592 }
6593 return 1;
6594 }
6595
6596 static struct mdinfo *imsm_activate_spare(struct active_array *a,
6597 struct metadata_update **updates)
6598 {
6599 /**
6600 * Find a device with unused free space and use it to replace a
6601 * failed/vacant region in an array. We replace failed regions one a
6602 * array at a time. The result is that a new spare disk will be added
6603 * to the first failed array and after the monitor has finished
6604 * propagating failures the remainder will be consumed.
6605 *
6606 * FIXME add a capability for mdmon to request spares from another
6607 * container.
6608 */
6609
6610 struct intel_super *super = a->container->sb;
6611 int inst = a->info.container_member;
6612 struct imsm_dev *dev = get_imsm_dev(super, inst);
6613 struct imsm_map *map = get_imsm_map(dev, 0);
6614 int failed = a->info.array.raid_disks;
6615 struct mdinfo *rv = NULL;
6616 struct mdinfo *d;
6617 struct mdinfo *di;
6618 struct metadata_update *mu;
6619 struct dl *dl;
6620 struct imsm_update_activate_spare *u;
6621 int num_spares = 0;
6622 int i;
6623 int allowed;
6624
6625 for (d = a->info.devs ; d ; d = d->next) {
6626 if ((d->curr_state & DS_FAULTY) &&
6627 d->state_fd >= 0)
6628 /* wait for Removal to happen */
6629 return NULL;
6630 if (d->state_fd >= 0)
6631 failed--;
6632 }
6633
6634 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
6635 inst, failed, a->info.array.raid_disks, a->info.array.level);
6636
6637 if (imsm_reshape_blocks_arrays_changes(super))
6638 return NULL;
6639
6640 if (a->info.array.level == 4)
6641 /* No repair for takeovered array
6642 * imsm doesn't support raid4
6643 */
6644 return NULL;
6645
6646 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
6647 return NULL;
6648
6649 /*
6650 * If there are any failed disks check state of the other volume.
6651 * Block rebuild if the another one is failed until failed disks
6652 * are removed from container.
6653 */
6654 if (failed) {
6655 dprintf("found failed disks in %.*s, check if there another"
6656 "failed sub-array.\n",
6657 MAX_RAID_SERIAL_LEN, dev->volume);
6658 /* check if states of the other volumes allow for rebuild */
6659 for (i = 0; i < super->anchor->num_raid_devs; i++) {
6660 if (i != inst) {
6661 allowed = imsm_rebuild_allowed(a->container,
6662 i, failed);
6663 if (!allowed)
6664 return NULL;
6665 }
6666 }
6667 }
6668
6669 /* For each slot, if it is not working, find a spare */
6670 for (i = 0; i < a->info.array.raid_disks; i++) {
6671 for (d = a->info.devs ; d ; d = d->next)
6672 if (d->disk.raid_disk == i)
6673 break;
6674 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
6675 if (d && (d->state_fd >= 0))
6676 continue;
6677
6678 /*
6679 * OK, this device needs recovery. Try to re-add the
6680 * previous occupant of this slot, if this fails see if
6681 * we can continue the assimilation of a spare that was
6682 * partially assimilated, finally try to activate a new
6683 * spare.
6684 */
6685 dl = imsm_readd(super, i, a);
6686 if (!dl)
6687 dl = imsm_add_spare(super, i, a, 0, rv);
6688 if (!dl)
6689 dl = imsm_add_spare(super, i, a, 1, rv);
6690 if (!dl)
6691 continue;
6692
6693 /* found a usable disk with enough space */
6694 di = malloc(sizeof(*di));
6695 if (!di)
6696 continue;
6697 memset(di, 0, sizeof(*di));
6698
6699 /* dl->index will be -1 in the case we are activating a
6700 * pristine spare. imsm_process_update() will create a
6701 * new index in this case. Once a disk is found to be
6702 * failed in all member arrays it is kicked from the
6703 * metadata
6704 */
6705 di->disk.number = dl->index;
6706
6707 /* (ab)use di->devs to store a pointer to the device
6708 * we chose
6709 */
6710 di->devs = (struct mdinfo *) dl;
6711
6712 di->disk.raid_disk = i;
6713 di->disk.major = dl->major;
6714 di->disk.minor = dl->minor;
6715 di->disk.state = 0;
6716 di->recovery_start = 0;
6717 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
6718 di->component_size = a->info.component_size;
6719 di->container_member = inst;
6720 super->random = random32();
6721 di->next = rv;
6722 rv = di;
6723 num_spares++;
6724 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
6725 i, di->data_offset);
6726 }
6727
6728 if (!rv)
6729 /* No spares found */
6730 return rv;
6731 /* Now 'rv' has a list of devices to return.
6732 * Create a metadata_update record to update the
6733 * disk_ord_tbl for the array
6734 */
6735 mu = malloc(sizeof(*mu));
6736 if (mu) {
6737 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
6738 if (mu->buf == NULL) {
6739 free(mu);
6740 mu = NULL;
6741 }
6742 }
6743 if (!mu) {
6744 while (rv) {
6745 struct mdinfo *n = rv->next;
6746
6747 free(rv);
6748 rv = n;
6749 }
6750 return NULL;
6751 }
6752
6753 mu->space = NULL;
6754 mu->space_list = NULL;
6755 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
6756 mu->next = *updates;
6757 u = (struct imsm_update_activate_spare *) mu->buf;
6758
6759 for (di = rv ; di ; di = di->next) {
6760 u->type = update_activate_spare;
6761 u->dl = (struct dl *) di->devs;
6762 di->devs = NULL;
6763 u->slot = di->disk.raid_disk;
6764 u->array = inst;
6765 u->next = u + 1;
6766 u++;
6767 }
6768 (u-1)->next = NULL;
6769 *updates = mu;
6770
6771 return rv;
6772 }
6773
6774 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
6775 {
6776 struct imsm_dev *dev = get_imsm_dev(super, idx);
6777 struct imsm_map *map = get_imsm_map(dev, 0);
6778 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
6779 struct disk_info *inf = get_disk_info(u);
6780 struct imsm_disk *disk;
6781 int i;
6782 int j;
6783
6784 for (i = 0; i < map->num_members; i++) {
6785 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i, -1));
6786 for (j = 0; j < new_map->num_members; j++)
6787 if (serialcmp(disk->serial, inf[j].serial) == 0)
6788 return 1;
6789 }
6790
6791 return 0;
6792 }
6793
6794
6795 static struct dl *get_disk_super(struct intel_super *super, int major, int minor)
6796 {
6797 struct dl *dl = NULL;
6798 for (dl = super->disks; dl; dl = dl->next)
6799 if ((dl->major == major) && (dl->minor == minor))
6800 return dl;
6801 return NULL;
6802 }
6803
6804 static int remove_disk_super(struct intel_super *super, int major, int minor)
6805 {
6806 struct dl *prev = NULL;
6807 struct dl *dl;
6808
6809 prev = NULL;
6810 for (dl = super->disks; dl; dl = dl->next) {
6811 if ((dl->major == major) && (dl->minor == minor)) {
6812 /* remove */
6813 if (prev)
6814 prev->next = dl->next;
6815 else
6816 super->disks = dl->next;
6817 dl->next = NULL;
6818 __free_imsm_disk(dl);
6819 dprintf("%s: removed %x:%x\n",
6820 __func__, major, minor);
6821 break;
6822 }
6823 prev = dl;
6824 }
6825 return 0;
6826 }
6827
6828 static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index);
6829
6830 static int add_remove_disk_update(struct intel_super *super)
6831 {
6832 int check_degraded = 0;
6833 struct dl *disk = NULL;
6834 /* add/remove some spares to/from the metadata/contrainer */
6835 while (super->disk_mgmt_list) {
6836 struct dl *disk_cfg;
6837
6838 disk_cfg = super->disk_mgmt_list;
6839 super->disk_mgmt_list = disk_cfg->next;
6840 disk_cfg->next = NULL;
6841
6842 if (disk_cfg->action == DISK_ADD) {
6843 disk_cfg->next = super->disks;
6844 super->disks = disk_cfg;
6845 check_degraded = 1;
6846 dprintf("%s: added %x:%x\n",
6847 __func__, disk_cfg->major,
6848 disk_cfg->minor);
6849 } else if (disk_cfg->action == DISK_REMOVE) {
6850 dprintf("Disk remove action processed: %x.%x\n",
6851 disk_cfg->major, disk_cfg->minor);
6852 disk = get_disk_super(super,
6853 disk_cfg->major,
6854 disk_cfg->minor);
6855 if (disk) {
6856 /* store action status */
6857 disk->action = DISK_REMOVE;
6858 /* remove spare disks only */
6859 if (disk->index == -1) {
6860 remove_disk_super(super,
6861 disk_cfg->major,
6862 disk_cfg->minor);
6863 }
6864 }
6865 /* release allocate disk structure */
6866 __free_imsm_disk(disk_cfg);
6867 }
6868 }
6869 return check_degraded;
6870 }
6871
6872
6873 static int apply_reshape_migration_update(struct imsm_update_reshape_migration *u,
6874 struct intel_super *super,
6875 void ***space_list)
6876 {
6877 struct intel_dev *id;
6878 void **tofree = NULL;
6879 int ret_val = 0;
6880
6881 dprintf("apply_reshape_migration_update()\n");
6882 if ((u->subdev < 0) ||
6883 (u->subdev > 1)) {
6884 dprintf("imsm: Error: Wrong subdev: %i\n", u->subdev);
6885 return ret_val;
6886 }
6887 if ((space_list == NULL) || (*space_list == NULL)) {
6888 dprintf("imsm: Error: Memory is not allocated\n");
6889 return ret_val;
6890 }
6891
6892 for (id = super->devlist ; id; id = id->next) {
6893 if (id->index == (unsigned)u->subdev) {
6894 struct imsm_dev *dev = get_imsm_dev(super, u->subdev);
6895 struct imsm_map *map;
6896 struct imsm_dev *new_dev =
6897 (struct imsm_dev *)*space_list;
6898 struct imsm_map *migr_map = get_imsm_map(dev, 1);
6899 int to_state;
6900 struct dl *new_disk;
6901
6902 if (new_dev == NULL)
6903 return ret_val;
6904 *space_list = **space_list;
6905 memcpy(new_dev, dev, sizeof_imsm_dev(dev, 0));
6906 map = get_imsm_map(new_dev, 0);
6907 if (migr_map) {
6908 dprintf("imsm: Error: migration in progress");
6909 return ret_val;
6910 }
6911
6912 to_state = map->map_state;
6913 if ((u->new_level == 5) && (map->raid_level == 0)) {
6914 map->num_members++;
6915 /* this should not happen */
6916 if (u->new_disks[0] < 0) {
6917 map->failed_disk_num =
6918 map->num_members - 1;
6919 to_state = IMSM_T_STATE_DEGRADED;
6920 } else
6921 to_state = IMSM_T_STATE_NORMAL;
6922 }
6923 migrate(new_dev, super, to_state, MIGR_GEN_MIGR);
6924 if (u->new_level > -1)
6925 map->raid_level = u->new_level;
6926 migr_map = get_imsm_map(new_dev, 1);
6927 if ((u->new_level == 5) &&
6928 (migr_map->raid_level == 0)) {
6929 int ord = map->num_members - 1;
6930 migr_map->num_members--;
6931 if (u->new_disks[0] < 0)
6932 ord |= IMSM_ORD_REBUILD;
6933 set_imsm_ord_tbl_ent(map,
6934 map->num_members - 1,
6935 ord);
6936 }
6937 id->dev = new_dev;
6938 tofree = (void **)dev;
6939
6940 /* update chunk size
6941 */
6942 if (u->new_chunksize > 0)
6943 map->blocks_per_strip =
6944 __cpu_to_le16(u->new_chunksize * 2);
6945
6946 /* add disk
6947 */
6948 if ((u->new_level != 5) ||
6949 (migr_map->raid_level != 0) ||
6950 (migr_map->raid_level == map->raid_level))
6951 goto skip_disk_add;
6952
6953 if (u->new_disks[0] >= 0) {
6954 /* use passes spare
6955 */
6956 new_disk = get_disk_super(super,
6957 major(u->new_disks[0]),
6958 minor(u->new_disks[0]));
6959 dprintf("imsm: new disk for reshape is: %i:%i "
6960 "(%p, index = %i)\n",
6961 major(u->new_disks[0]),
6962 minor(u->new_disks[0]),
6963 new_disk, new_disk->index);
6964 if (new_disk == NULL)
6965 goto error_disk_add;
6966
6967 new_disk->index = map->num_members - 1;
6968 /* slot to fill in autolayout
6969 */
6970 new_disk->raiddisk = new_disk->index;
6971 new_disk->disk.status |= CONFIGURED_DISK;
6972 new_disk->disk.status &= ~SPARE_DISK;
6973 } else
6974 goto error_disk_add;
6975
6976 skip_disk_add:
6977 *tofree = *space_list;
6978 /* calculate new size
6979 */
6980 imsm_set_array_size(new_dev);
6981
6982 ret_val = 1;
6983 }
6984 }
6985
6986 if (tofree)
6987 *space_list = tofree;
6988 return ret_val;
6989
6990 error_disk_add:
6991 dprintf("Error: imsm: Cannot find disk.\n");
6992 return ret_val;
6993 }
6994
6995 static int apply_update_activate_spare(struct imsm_update_activate_spare *u,
6996 struct intel_super *super,
6997 struct active_array *active_array)
6998 {
6999 struct imsm_super *mpb = super->anchor;
7000 struct imsm_dev *dev = get_imsm_dev(super, u->array);
7001 struct imsm_map *map = get_imsm_map(dev, 0);
7002 struct imsm_map *migr_map;
7003 struct active_array *a;
7004 struct imsm_disk *disk;
7005 __u8 to_state;
7006 struct dl *dl;
7007 unsigned int found;
7008 int failed;
7009 int victim;
7010 int i;
7011 int second_map_created = 0;
7012
7013 for (; u; u = u->next) {
7014 victim = get_imsm_disk_idx(dev, u->slot, -1);
7015
7016 if (victim < 0)
7017 return 0;
7018
7019 for (dl = super->disks; dl; dl = dl->next)
7020 if (dl == u->dl)
7021 break;
7022
7023 if (!dl) {
7024 fprintf(stderr, "error: imsm_activate_spare passed "
7025 "an unknown disk (index: %d)\n",
7026 u->dl->index);
7027 return 0;
7028 }
7029
7030 /* count failures (excluding rebuilds and the victim)
7031 * to determine map[0] state
7032 */
7033 failed = 0;
7034 for (i = 0; i < map->num_members; i++) {
7035 if (i == u->slot)
7036 continue;
7037 disk = get_imsm_disk(super,
7038 get_imsm_disk_idx(dev, i, -1));
7039 if (!disk || is_failed(disk))
7040 failed++;
7041 }
7042
7043 /* adding a pristine spare, assign a new index */
7044 if (dl->index < 0) {
7045 dl->index = super->anchor->num_disks;
7046 super->anchor->num_disks++;
7047 }
7048 disk = &dl->disk;
7049 disk->status |= CONFIGURED_DISK;
7050 disk->status &= ~SPARE_DISK;
7051
7052 /* mark rebuild */
7053 to_state = imsm_check_degraded(super, dev, failed);
7054 if (!second_map_created) {
7055 second_map_created = 1;
7056 map->map_state = IMSM_T_STATE_DEGRADED;
7057 migrate(dev, super, to_state, MIGR_REBUILD);
7058 } else
7059 map->map_state = to_state;
7060 migr_map = get_imsm_map(dev, 1);
7061 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
7062 set_imsm_ord_tbl_ent(migr_map, u->slot,
7063 dl->index | IMSM_ORD_REBUILD);
7064
7065 /* update the family_num to mark a new container
7066 * generation, being careful to record the existing
7067 * family_num in orig_family_num to clean up after
7068 * earlier mdadm versions that neglected to set it.
7069 */
7070 if (mpb->orig_family_num == 0)
7071 mpb->orig_family_num = mpb->family_num;
7072 mpb->family_num += super->random;
7073
7074 /* count arrays using the victim in the metadata */
7075 found = 0;
7076 for (a = active_array; a ; a = a->next) {
7077 dev = get_imsm_dev(super, a->info.container_member);
7078 map = get_imsm_map(dev, 0);
7079
7080 if (get_imsm_disk_slot(map, victim) >= 0)
7081 found++;
7082 }
7083
7084 /* delete the victim if it is no longer being
7085 * utilized anywhere
7086 */
7087 if (!found) {
7088 struct dl **dlp;
7089
7090 /* We know that 'manager' isn't touching anything,
7091 * so it is safe to delete
7092 */
7093 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
7094 if ((*dlp)->index == victim)
7095 break;
7096
7097 /* victim may be on the missing list */
7098 if (!*dlp)
7099 for (dlp = &super->missing; *dlp;
7100 dlp = &(*dlp)->next)
7101 if ((*dlp)->index == victim)
7102 break;
7103 imsm_delete(super, dlp, victim);
7104 }
7105 }
7106
7107 return 1;
7108 }
7109
7110 static int apply_reshape_container_disks_update(struct imsm_update_reshape *u,
7111 struct intel_super *super,
7112 void ***space_list)
7113 {
7114 struct dl *new_disk;
7115 struct intel_dev *id;
7116 int i;
7117 int delta_disks = u->new_raid_disks - u->old_raid_disks;
7118 int disk_count = u->old_raid_disks;
7119 void **tofree = NULL;
7120 int devices_to_reshape = 1;
7121 struct imsm_super *mpb = super->anchor;
7122 int ret_val = 0;
7123 unsigned int dev_id;
7124
7125 dprintf("imsm: apply_reshape_container_disks_update()\n");
7126
7127 /* enable spares to use in array */
7128 for (i = 0; i < delta_disks; i++) {
7129 new_disk = get_disk_super(super,
7130 major(u->new_disks[i]),
7131 minor(u->new_disks[i]));
7132 dprintf("imsm: new disk for reshape is: %i:%i "
7133 "(%p, index = %i)\n",
7134 major(u->new_disks[i]), minor(u->new_disks[i]),
7135 new_disk, new_disk->index);
7136 if ((new_disk == NULL) ||
7137 ((new_disk->index >= 0) &&
7138 (new_disk->index < u->old_raid_disks)))
7139 goto update_reshape_exit;
7140 new_disk->index = disk_count++;
7141 /* slot to fill in autolayout
7142 */
7143 new_disk->raiddisk = new_disk->index;
7144 new_disk->disk.status |=
7145 CONFIGURED_DISK;
7146 new_disk->disk.status &= ~SPARE_DISK;
7147 }
7148
7149 dprintf("imsm: volume set mpb->num_raid_devs = %i\n",
7150 mpb->num_raid_devs);
7151 /* manage changes in volume
7152 */
7153 for (dev_id = 0; dev_id < mpb->num_raid_devs; dev_id++) {
7154 void **sp = *space_list;
7155 struct imsm_dev *newdev;
7156 struct imsm_map *newmap, *oldmap;
7157
7158 for (id = super->devlist ; id; id = id->next) {
7159 if (id->index == dev_id)
7160 break;
7161 }
7162 if (id == NULL)
7163 break;
7164 if (!sp)
7165 continue;
7166 *space_list = *sp;
7167 newdev = (void*)sp;
7168 /* Copy the dev, but not (all of) the map */
7169 memcpy(newdev, id->dev, sizeof(*newdev));
7170 oldmap = get_imsm_map(id->dev, 0);
7171 newmap = get_imsm_map(newdev, 0);
7172 /* Copy the current map */
7173 memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
7174 /* update one device only
7175 */
7176 if (devices_to_reshape) {
7177 dprintf("imsm: modifying subdev: %i\n",
7178 id->index);
7179 devices_to_reshape--;
7180 newdev->vol.migr_state = 1;
7181 newdev->vol.curr_migr_unit = 0;
7182 set_migr_type(newdev, MIGR_GEN_MIGR);
7183 newmap->num_members = u->new_raid_disks;
7184 for (i = 0; i < delta_disks; i++) {
7185 set_imsm_ord_tbl_ent(newmap,
7186 u->old_raid_disks + i,
7187 u->old_raid_disks + i);
7188 }
7189 /* New map is correct, now need to save old map
7190 */
7191 newmap = get_imsm_map(newdev, 1);
7192 memcpy(newmap, oldmap, sizeof_imsm_map(oldmap));
7193
7194 imsm_set_array_size(newdev);
7195 }
7196
7197 sp = (void **)id->dev;
7198 id->dev = newdev;
7199 *sp = tofree;
7200 tofree = sp;
7201
7202 /* Clear migration record */
7203 memset(super->migr_rec, 0, sizeof(struct migr_record));
7204 }
7205 if (tofree)
7206 *space_list = tofree;
7207 ret_val = 1;
7208
7209 update_reshape_exit:
7210
7211 return ret_val;
7212 }
7213
7214 static int apply_takeover_update(struct imsm_update_takeover *u,
7215 struct intel_super *super,
7216 void ***space_list)
7217 {
7218 struct imsm_dev *dev = NULL;
7219 struct intel_dev *dv;
7220 struct imsm_dev *dev_new;
7221 struct imsm_map *map;
7222 struct dl *dm, *du;
7223 int i;
7224
7225 for (dv = super->devlist; dv; dv = dv->next)
7226 if (dv->index == (unsigned int)u->subarray) {
7227 dev = dv->dev;
7228 break;
7229 }
7230
7231 if (dev == NULL)
7232 return 0;
7233
7234 map = get_imsm_map(dev, 0);
7235
7236 if (u->direction == R10_TO_R0) {
7237 /* Number of failed disks must be half of initial disk number */
7238 if (imsm_count_failed(super, dev) != (map->num_members / 2))
7239 return 0;
7240
7241 /* iterate through devices to mark removed disks as spare */
7242 for (dm = super->disks; dm; dm = dm->next) {
7243 if (dm->disk.status & FAILED_DISK) {
7244 int idx = dm->index;
7245 /* update indexes on the disk list */
7246 /* FIXME this loop-with-the-loop looks wrong, I'm not convinced
7247 the index values will end up being correct.... NB */
7248 for (du = super->disks; du; du = du->next)
7249 if (du->index > idx)
7250 du->index--;
7251 /* mark as spare disk */
7252 mark_spare(dm);
7253 }
7254 }
7255 /* update map */
7256 map->num_members = map->num_members / 2;
7257 map->map_state = IMSM_T_STATE_NORMAL;
7258 map->num_domains = 1;
7259 map->raid_level = 0;
7260 map->failed_disk_num = -1;
7261 }
7262
7263 if (u->direction == R0_TO_R10) {
7264 void **space;
7265 /* update slots in current disk list */
7266 for (dm = super->disks; dm; dm = dm->next) {
7267 if (dm->index >= 0)
7268 dm->index *= 2;
7269 }
7270 /* create new *missing* disks */
7271 for (i = 0; i < map->num_members; i++) {
7272 space = *space_list;
7273 if (!space)
7274 continue;
7275 *space_list = *space;
7276 du = (void *)space;
7277 memcpy(du, super->disks, sizeof(*du));
7278 du->fd = -1;
7279 du->minor = 0;
7280 du->major = 0;
7281 du->index = (i * 2) + 1;
7282 sprintf((char *)du->disk.serial,
7283 " MISSING_%d", du->index);
7284 sprintf((char *)du->serial,
7285 "MISSING_%d", du->index);
7286 du->next = super->missing;
7287 super->missing = du;
7288 }
7289 /* create new dev and map */
7290 space = *space_list;
7291 if (!space)
7292 return 0;
7293 *space_list = *space;
7294 dev_new = (void *)space;
7295 memcpy(dev_new, dev, sizeof(*dev));
7296 /* update new map */
7297 map = get_imsm_map(dev_new, 0);
7298 map->num_members = map->num_members * 2;
7299 map->map_state = IMSM_T_STATE_DEGRADED;
7300 map->num_domains = 2;
7301 map->raid_level = 1;
7302 /* replace dev<->dev_new */
7303 dv->dev = dev_new;
7304 }
7305 /* update disk order table */
7306 for (du = super->disks; du; du = du->next)
7307 if (du->index >= 0)
7308 set_imsm_ord_tbl_ent(map, du->index, du->index);
7309 for (du = super->missing; du; du = du->next)
7310 if (du->index >= 0) {
7311 set_imsm_ord_tbl_ent(map, du->index, du->index);
7312 mark_missing(dv->dev, &du->disk, du->index);
7313 }
7314
7315 return 1;
7316 }
7317
7318 static void imsm_process_update(struct supertype *st,
7319 struct metadata_update *update)
7320 {
7321 /**
7322 * crack open the metadata_update envelope to find the update record
7323 * update can be one of:
7324 * update_reshape_container_disks - all the arrays in the container
7325 * are being reshaped to have more devices. We need to mark
7326 * the arrays for general migration and convert selected spares
7327 * into active devices.
7328 * update_activate_spare - a spare device has replaced a failed
7329 * device in an array, update the disk_ord_tbl. If this disk is
7330 * present in all member arrays then also clear the SPARE_DISK
7331 * flag
7332 * update_create_array
7333 * update_kill_array
7334 * update_rename_array
7335 * update_add_remove_disk
7336 */
7337 struct intel_super *super = st->sb;
7338 struct imsm_super *mpb;
7339 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
7340
7341 /* update requires a larger buf but the allocation failed */
7342 if (super->next_len && !super->next_buf) {
7343 super->next_len = 0;
7344 return;
7345 }
7346
7347 if (super->next_buf) {
7348 memcpy(super->next_buf, super->buf, super->len);
7349 free(super->buf);
7350 super->len = super->next_len;
7351 super->buf = super->next_buf;
7352
7353 super->next_len = 0;
7354 super->next_buf = NULL;
7355 }
7356
7357 mpb = super->anchor;
7358
7359 switch (type) {
7360 case update_general_migration_checkpoint: {
7361 struct intel_dev *id;
7362 struct imsm_update_general_migration_checkpoint *u =
7363 (void *)update->buf;
7364
7365 dprintf("imsm: process_update() "
7366 "for update_general_migration_checkpoint called\n");
7367
7368 /* find device under general migration */
7369 for (id = super->devlist ; id; id = id->next) {
7370 if (is_gen_migration(id->dev)) {
7371 id->dev->vol.curr_migr_unit =
7372 __cpu_to_le32(u->curr_migr_unit);
7373 super->updates_pending++;
7374 }
7375 }
7376 break;
7377 }
7378 case update_takeover: {
7379 struct imsm_update_takeover *u = (void *)update->buf;
7380 if (apply_takeover_update(u, super, &update->space_list)) {
7381 imsm_update_version_info(super);
7382 super->updates_pending++;
7383 }
7384 break;
7385 }
7386
7387 case update_reshape_container_disks: {
7388 struct imsm_update_reshape *u = (void *)update->buf;
7389 if (apply_reshape_container_disks_update(
7390 u, super, &update->space_list))
7391 super->updates_pending++;
7392 break;
7393 }
7394 case update_reshape_migration: {
7395 struct imsm_update_reshape_migration *u = (void *)update->buf;
7396 if (apply_reshape_migration_update(
7397 u, super, &update->space_list))
7398 super->updates_pending++;
7399 break;
7400 }
7401 case update_activate_spare: {
7402 struct imsm_update_activate_spare *u = (void *) update->buf;
7403 if (apply_update_activate_spare(u, super, st->arrays))
7404 super->updates_pending++;
7405 break;
7406 }
7407 case update_create_array: {
7408 /* someone wants to create a new array, we need to be aware of
7409 * a few races/collisions:
7410 * 1/ 'Create' called by two separate instances of mdadm
7411 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
7412 * devices that have since been assimilated via
7413 * activate_spare.
7414 * In the event this update can not be carried out mdadm will
7415 * (FIX ME) notice that its update did not take hold.
7416 */
7417 struct imsm_update_create_array *u = (void *) update->buf;
7418 struct intel_dev *dv;
7419 struct imsm_dev *dev;
7420 struct imsm_map *map, *new_map;
7421 unsigned long long start, end;
7422 unsigned long long new_start, new_end;
7423 int i;
7424 struct disk_info *inf;
7425 struct dl *dl;
7426
7427 /* handle racing creates: first come first serve */
7428 if (u->dev_idx < mpb->num_raid_devs) {
7429 dprintf("%s: subarray %d already defined\n",
7430 __func__, u->dev_idx);
7431 goto create_error;
7432 }
7433
7434 /* check update is next in sequence */
7435 if (u->dev_idx != mpb->num_raid_devs) {
7436 dprintf("%s: can not create array %d expected index %d\n",
7437 __func__, u->dev_idx, mpb->num_raid_devs);
7438 goto create_error;
7439 }
7440
7441 new_map = get_imsm_map(&u->dev, 0);
7442 new_start = __le32_to_cpu(new_map->pba_of_lba0);
7443 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
7444 inf = get_disk_info(u);
7445
7446 /* handle activate_spare versus create race:
7447 * check to make sure that overlapping arrays do not include
7448 * overalpping disks
7449 */
7450 for (i = 0; i < mpb->num_raid_devs; i++) {
7451 dev = get_imsm_dev(super, i);
7452 map = get_imsm_map(dev, 0);
7453 start = __le32_to_cpu(map->pba_of_lba0);
7454 end = start + __le32_to_cpu(map->blocks_per_member);
7455 if ((new_start >= start && new_start <= end) ||
7456 (start >= new_start && start <= new_end))
7457 /* overlap */;
7458 else
7459 continue;
7460
7461 if (disks_overlap(super, i, u)) {
7462 dprintf("%s: arrays overlap\n", __func__);
7463 goto create_error;
7464 }
7465 }
7466
7467 /* check that prepare update was successful */
7468 if (!update->space) {
7469 dprintf("%s: prepare update failed\n", __func__);
7470 goto create_error;
7471 }
7472
7473 /* check that all disks are still active before committing
7474 * changes. FIXME: could we instead handle this by creating a
7475 * degraded array? That's probably not what the user expects,
7476 * so better to drop this update on the floor.
7477 */
7478 for (i = 0; i < new_map->num_members; i++) {
7479 dl = serial_to_dl(inf[i].serial, super);
7480 if (!dl) {
7481 dprintf("%s: disk disappeared\n", __func__);
7482 goto create_error;
7483 }
7484 }
7485
7486 super->updates_pending++;
7487
7488 /* convert spares to members and fixup ord_tbl */
7489 for (i = 0; i < new_map->num_members; i++) {
7490 dl = serial_to_dl(inf[i].serial, super);
7491 if (dl->index == -1) {
7492 dl->index = mpb->num_disks;
7493 mpb->num_disks++;
7494 dl->disk.status |= CONFIGURED_DISK;
7495 dl->disk.status &= ~SPARE_DISK;
7496 }
7497 set_imsm_ord_tbl_ent(new_map, i, dl->index);
7498 }
7499
7500 dv = update->space;
7501 dev = dv->dev;
7502 update->space = NULL;
7503 imsm_copy_dev(dev, &u->dev);
7504 dv->index = u->dev_idx;
7505 dv->next = super->devlist;
7506 super->devlist = dv;
7507 mpb->num_raid_devs++;
7508
7509 imsm_update_version_info(super);
7510 break;
7511 create_error:
7512 /* mdmon knows how to release update->space, but not
7513 * ((struct intel_dev *) update->space)->dev
7514 */
7515 if (update->space) {
7516 dv = update->space;
7517 free(dv->dev);
7518 }
7519 break;
7520 }
7521 case update_kill_array: {
7522 struct imsm_update_kill_array *u = (void *) update->buf;
7523 int victim = u->dev_idx;
7524 struct active_array *a;
7525 struct intel_dev **dp;
7526 struct imsm_dev *dev;
7527
7528 /* sanity check that we are not affecting the uuid of
7529 * active arrays, or deleting an active array
7530 *
7531 * FIXME when immutable ids are available, but note that
7532 * we'll also need to fixup the invalidated/active
7533 * subarray indexes in mdstat
7534 */
7535 for (a = st->arrays; a; a = a->next)
7536 if (a->info.container_member >= victim)
7537 break;
7538 /* by definition if mdmon is running at least one array
7539 * is active in the container, so checking
7540 * mpb->num_raid_devs is just extra paranoia
7541 */
7542 dev = get_imsm_dev(super, victim);
7543 if (a || !dev || mpb->num_raid_devs == 1) {
7544 dprintf("failed to delete subarray-%d\n", victim);
7545 break;
7546 }
7547
7548 for (dp = &super->devlist; *dp;)
7549 if ((*dp)->index == (unsigned)super->current_vol) {
7550 *dp = (*dp)->next;
7551 } else {
7552 if ((*dp)->index > (unsigned)victim)
7553 (*dp)->index--;
7554 dp = &(*dp)->next;
7555 }
7556 mpb->num_raid_devs--;
7557 super->updates_pending++;
7558 break;
7559 }
7560 case update_rename_array: {
7561 struct imsm_update_rename_array *u = (void *) update->buf;
7562 char name[MAX_RAID_SERIAL_LEN+1];
7563 int target = u->dev_idx;
7564 struct active_array *a;
7565 struct imsm_dev *dev;
7566
7567 /* sanity check that we are not affecting the uuid of
7568 * an active array
7569 */
7570 snprintf(name, MAX_RAID_SERIAL_LEN, "%s", (char *) u->name);
7571 name[MAX_RAID_SERIAL_LEN] = '\0';
7572 for (a = st->arrays; a; a = a->next)
7573 if (a->info.container_member == target)
7574 break;
7575 dev = get_imsm_dev(super, u->dev_idx);
7576 if (a || !dev || !check_name(super, name, 1)) {
7577 dprintf("failed to rename subarray-%d\n", target);
7578 break;
7579 }
7580
7581 snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name);
7582 super->updates_pending++;
7583 break;
7584 }
7585 case update_add_remove_disk: {
7586 /* we may be able to repair some arrays if disks are
7587 * being added, check teh status of add_remove_disk
7588 * if discs has been added.
7589 */
7590 if (add_remove_disk_update(super)) {
7591 struct active_array *a;
7592
7593 super->updates_pending++;
7594 for (a = st->arrays; a; a = a->next)
7595 a->check_degraded = 1;
7596 }
7597 break;
7598 }
7599 default:
7600 fprintf(stderr, "error: unsuported process update type:"
7601 "(type: %d)\n", type);
7602 }
7603 }
7604
7605 static struct mdinfo *get_spares_for_grow(struct supertype *st);
7606
7607 static void imsm_prepare_update(struct supertype *st,
7608 struct metadata_update *update)
7609 {
7610 /**
7611 * Allocate space to hold new disk entries, raid-device entries or a new
7612 * mpb if necessary. The manager synchronously waits for updates to
7613 * complete in the monitor, so new mpb buffers allocated here can be
7614 * integrated by the monitor thread without worrying about live pointers
7615 * in the manager thread.
7616 */
7617 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
7618 struct intel_super *super = st->sb;
7619 struct imsm_super *mpb = super->anchor;
7620 size_t buf_len;
7621 size_t len = 0;
7622
7623 switch (type) {
7624 case update_general_migration_checkpoint:
7625 dprintf("imsm: prepare_update() "
7626 "for update_general_migration_checkpoint called\n");
7627 break;
7628 case update_takeover: {
7629 struct imsm_update_takeover *u = (void *)update->buf;
7630 if (u->direction == R0_TO_R10) {
7631 void **tail = (void **)&update->space_list;
7632 struct imsm_dev *dev = get_imsm_dev(super, u->subarray);
7633 struct imsm_map *map = get_imsm_map(dev, 0);
7634 int num_members = map->num_members;
7635 void *space;
7636 int size, i;
7637 int err = 0;
7638 /* allocate memory for added disks */
7639 for (i = 0; i < num_members; i++) {
7640 size = sizeof(struct dl);
7641 space = malloc(size);
7642 if (!space) {
7643 err++;
7644 break;
7645 }
7646 *tail = space;
7647 tail = space;
7648 *tail = NULL;
7649 }
7650 /* allocate memory for new device */
7651 size = sizeof_imsm_dev(super->devlist->dev, 0) +
7652 (num_members * sizeof(__u32));
7653 space = malloc(size);
7654 if (!space)
7655 err++;
7656 else {
7657 *tail = space;
7658 tail = space;
7659 *tail = NULL;
7660 }
7661 if (!err) {
7662 len = disks_to_mpb_size(num_members * 2);
7663 } else {
7664 /* if allocation didn't success, free buffer */
7665 while (update->space_list) {
7666 void **sp = update->space_list;
7667 update->space_list = *sp;
7668 free(sp);
7669 }
7670 }
7671 }
7672
7673 break;
7674 }
7675 case update_reshape_container_disks: {
7676 /* Every raid device in the container is about to
7677 * gain some more devices, and we will enter a
7678 * reconfiguration.
7679 * So each 'imsm_map' will be bigger, and the imsm_vol
7680 * will now hold 2 of them.
7681 * Thus we need new 'struct imsm_dev' allocations sized
7682 * as sizeof_imsm_dev but with more devices in both maps.
7683 */
7684 struct imsm_update_reshape *u = (void *)update->buf;
7685 struct intel_dev *dl;
7686 void **space_tail = (void**)&update->space_list;
7687
7688 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7689
7690 for (dl = super->devlist; dl; dl = dl->next) {
7691 int size = sizeof_imsm_dev(dl->dev, 1);
7692 void *s;
7693 if (u->new_raid_disks > u->old_raid_disks)
7694 size += sizeof(__u32)*2*
7695 (u->new_raid_disks - u->old_raid_disks);
7696 s = malloc(size);
7697 if (!s)
7698 break;
7699 *space_tail = s;
7700 space_tail = s;
7701 *space_tail = NULL;
7702 }
7703
7704 len = disks_to_mpb_size(u->new_raid_disks);
7705 dprintf("New anchor length is %llu\n", (unsigned long long)len);
7706 break;
7707 }
7708 case update_reshape_migration: {
7709 /* for migration level 0->5 we need to add disks
7710 * so the same as for container operation we will copy
7711 * device to the bigger location.
7712 * in memory prepared device and new disk area are prepared
7713 * for usage in process update
7714 */
7715 struct imsm_update_reshape_migration *u = (void *)update->buf;
7716 struct intel_dev *id;
7717 void **space_tail = (void **)&update->space_list;
7718 int size;
7719 void *s;
7720 int current_level = -1;
7721
7722 dprintf("imsm: imsm_prepare_update() for update_reshape\n");
7723
7724 /* add space for bigger array in update
7725 */
7726 for (id = super->devlist; id; id = id->next) {
7727 if (id->index == (unsigned)u->subdev) {
7728 size = sizeof_imsm_dev(id->dev, 1);
7729 if (u->new_raid_disks > u->old_raid_disks)
7730 size += sizeof(__u32)*2*
7731 (u->new_raid_disks - u->old_raid_disks);
7732 s = malloc(size);
7733 if (!s)
7734 break;
7735 *space_tail = s;
7736 space_tail = s;
7737 *space_tail = NULL;
7738 break;
7739 }
7740 }
7741 if (update->space_list == NULL)
7742 break;
7743
7744 /* add space for disk in update
7745 */
7746 size = sizeof(struct dl);
7747 s = malloc(size);
7748 if (!s) {
7749 free(update->space_list);
7750 update->space_list = NULL;
7751 break;
7752 }
7753 *space_tail = s;
7754 space_tail = s;
7755 *space_tail = NULL;
7756
7757 /* add spare device to update
7758 */
7759 for (id = super->devlist ; id; id = id->next)
7760 if (id->index == (unsigned)u->subdev) {
7761 struct imsm_dev *dev;
7762 struct imsm_map *map;
7763
7764 dev = get_imsm_dev(super, u->subdev);
7765 map = get_imsm_map(dev, 0);
7766 current_level = map->raid_level;
7767 break;
7768 }
7769 if ((u->new_level == 5) && (u->new_level != current_level)) {
7770 struct mdinfo *spares;
7771
7772 spares = get_spares_for_grow(st);
7773 if (spares) {
7774 struct dl *dl;
7775 struct mdinfo *dev;
7776
7777 dev = spares->devs;
7778 if (dev) {
7779 u->new_disks[0] =
7780 makedev(dev->disk.major,
7781 dev->disk.minor);
7782 dl = get_disk_super(super,
7783 dev->disk.major,
7784 dev->disk.minor);
7785 dl->index = u->old_raid_disks;
7786 dev = dev->next;
7787 }
7788 sysfs_free(spares);
7789 }
7790 }
7791 len = disks_to_mpb_size(u->new_raid_disks);
7792 dprintf("New anchor length is %llu\n", (unsigned long long)len);
7793 break;
7794 }
7795 case update_create_array: {
7796 struct imsm_update_create_array *u = (void *) update->buf;
7797 struct intel_dev *dv;
7798 struct imsm_dev *dev = &u->dev;
7799 struct imsm_map *map = get_imsm_map(dev, 0);
7800 struct dl *dl;
7801 struct disk_info *inf;
7802 int i;
7803 int activate = 0;
7804
7805 inf = get_disk_info(u);
7806 len = sizeof_imsm_dev(dev, 1);
7807 /* allocate a new super->devlist entry */
7808 dv = malloc(sizeof(*dv));
7809 if (dv) {
7810 dv->dev = malloc(len);
7811 if (dv->dev)
7812 update->space = dv;
7813 else {
7814 free(dv);
7815 update->space = NULL;
7816 }
7817 }
7818
7819 /* count how many spares will be converted to members */
7820 for (i = 0; i < map->num_members; i++) {
7821 dl = serial_to_dl(inf[i].serial, super);
7822 if (!dl) {
7823 /* hmm maybe it failed?, nothing we can do about
7824 * it here
7825 */
7826 continue;
7827 }
7828 if (count_memberships(dl, super) == 0)
7829 activate++;
7830 }
7831 len += activate * sizeof(struct imsm_disk);
7832 break;
7833 default:
7834 break;
7835 }
7836 }
7837
7838 /* check if we need a larger metadata buffer */
7839 if (super->next_buf)
7840 buf_len = super->next_len;
7841 else
7842 buf_len = super->len;
7843
7844 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
7845 /* ok we need a larger buf than what is currently allocated
7846 * if this allocation fails process_update will notice that
7847 * ->next_len is set and ->next_buf is NULL
7848 */
7849 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
7850 if (super->next_buf)
7851 free(super->next_buf);
7852
7853 super->next_len = buf_len;
7854 if (posix_memalign(&super->next_buf, 512, buf_len) == 0)
7855 memset(super->next_buf, 0, buf_len);
7856 else
7857 super->next_buf = NULL;
7858 }
7859 }
7860
7861 /* must be called while manager is quiesced */
7862 static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index)
7863 {
7864 struct imsm_super *mpb = super->anchor;
7865 struct dl *iter;
7866 struct imsm_dev *dev;
7867 struct imsm_map *map;
7868 int i, j, num_members;
7869 __u32 ord;
7870
7871 dprintf("%s: deleting device[%d] from imsm_super\n",
7872 __func__, index);
7873
7874 /* shift all indexes down one */
7875 for (iter = super->disks; iter; iter = iter->next)
7876 if (iter->index > (int)index)
7877 iter->index--;
7878 for (iter = super->missing; iter; iter = iter->next)
7879 if (iter->index > (int)index)
7880 iter->index--;
7881
7882 for (i = 0; i < mpb->num_raid_devs; i++) {
7883 dev = get_imsm_dev(super, i);
7884 map = get_imsm_map(dev, 0);
7885 num_members = map->num_members;
7886 for (j = 0; j < num_members; j++) {
7887 /* update ord entries being careful not to propagate
7888 * ord-flags to the first map
7889 */
7890 ord = get_imsm_ord_tbl_ent(dev, j, -1);
7891
7892 if (ord_to_idx(ord) <= index)
7893 continue;
7894
7895 map = get_imsm_map(dev, 0);
7896 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
7897 map = get_imsm_map(dev, 1);
7898 if (map)
7899 set_imsm_ord_tbl_ent(map, j, ord - 1);
7900 }
7901 }
7902
7903 mpb->num_disks--;
7904 super->updates_pending++;
7905 if (*dlp) {
7906 struct dl *dl = *dlp;
7907
7908 *dlp = (*dlp)->next;
7909 __free_imsm_disk(dl);
7910 }
7911 }
7912 #endif /* MDASSEMBLE */
7913 /*******************************************************************************
7914 * Function: open_backup_targets
7915 * Description: Function opens file descriptors for all devices given in
7916 * info->devs
7917 * Parameters:
7918 * info : general array info
7919 * raid_disks : number of disks
7920 * raid_fds : table of device's file descriptors
7921 * Returns:
7922 * 0 : success
7923 * -1 : fail
7924 ******************************************************************************/
7925 int open_backup_targets(struct mdinfo *info, int raid_disks, int *raid_fds)
7926 {
7927 struct mdinfo *sd;
7928
7929 for (sd = info->devs ; sd ; sd = sd->next) {
7930 char *dn;
7931
7932 if (sd->disk.state & (1<<MD_DISK_FAULTY)) {
7933 dprintf("disk is faulty!!\n");
7934 continue;
7935 }
7936
7937 if ((sd->disk.raid_disk >= raid_disks) ||
7938 (sd->disk.raid_disk < 0))
7939 continue;
7940
7941 dn = map_dev(sd->disk.major,
7942 sd->disk.minor, 1);
7943 raid_fds[sd->disk.raid_disk] = dev_open(dn, O_RDWR);
7944 if (raid_fds[sd->disk.raid_disk] < 0) {
7945 fprintf(stderr, "cannot open component\n");
7946 return -1;
7947 }
7948 }
7949 return 0;
7950 }
7951
7952 #ifndef MDASSEMBLE
7953 /*******************************************************************************
7954 * Function: init_migr_record_imsm
7955 * Description: Function inits imsm migration record
7956 * Parameters:
7957 * super : imsm internal array info
7958 * dev : device under migration
7959 * info : general array info to find the smallest device
7960 * Returns:
7961 * none
7962 ******************************************************************************/
7963 void init_migr_record_imsm(struct supertype *st, struct imsm_dev *dev,
7964 struct mdinfo *info)
7965 {
7966 struct intel_super *super = st->sb;
7967 struct migr_record *migr_rec = super->migr_rec;
7968 int new_data_disks;
7969 unsigned long long dsize, dev_sectors;
7970 long long unsigned min_dev_sectors = -1LLU;
7971 struct mdinfo *sd;
7972 char nm[30];
7973 int fd;
7974 struct imsm_map *map_dest = get_imsm_map(dev, 0);
7975 struct imsm_map *map_src = get_imsm_map(dev, 1);
7976 unsigned long long num_migr_units;
7977 unsigned long long array_blocks;
7978
7979 memset(migr_rec, 0, sizeof(struct migr_record));
7980 migr_rec->family_num = __cpu_to_le32(super->anchor->family_num);
7981
7982 /* only ascending reshape supported now */
7983 migr_rec->ascending_migr = __cpu_to_le32(1);
7984
7985 migr_rec->dest_depth_per_unit = GEN_MIGR_AREA_SIZE /
7986 max(map_dest->blocks_per_strip, map_src->blocks_per_strip);
7987 migr_rec->dest_depth_per_unit *= map_dest->blocks_per_strip;
7988 new_data_disks = imsm_num_data_members(dev, 0);
7989 migr_rec->blocks_per_unit =
7990 __cpu_to_le32(migr_rec->dest_depth_per_unit * new_data_disks);
7991 migr_rec->dest_depth_per_unit =
7992 __cpu_to_le32(migr_rec->dest_depth_per_unit);
7993 array_blocks = info->component_size * new_data_disks;
7994 num_migr_units =
7995 array_blocks / __le32_to_cpu(migr_rec->blocks_per_unit);
7996
7997 if (array_blocks % __le32_to_cpu(migr_rec->blocks_per_unit))
7998 num_migr_units++;
7999 migr_rec->num_migr_units = __cpu_to_le32(num_migr_units);
8000
8001 migr_rec->post_migr_vol_cap = dev->size_low;
8002 migr_rec->post_migr_vol_cap_hi = dev->size_high;
8003
8004
8005 /* Find the smallest dev */
8006 for (sd = info->devs ; sd ; sd = sd->next) {
8007 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
8008 fd = dev_open(nm, O_RDONLY);
8009 if (fd < 0)
8010 continue;
8011 get_dev_size(fd, NULL, &dsize);
8012 dev_sectors = dsize / 512;
8013 if (dev_sectors < min_dev_sectors)
8014 min_dev_sectors = dev_sectors;
8015 close(fd);
8016 }
8017 migr_rec->ckpt_area_pba = __cpu_to_le32(min_dev_sectors -
8018 RAID_DISK_RESERVED_BLOCKS_IMSM_HI);
8019
8020 write_imsm_migr_rec(st);
8021
8022 return;
8023 }
8024
8025 /*******************************************************************************
8026 * Function: save_backup_imsm
8027 * Description: Function saves critical data stripes to Migration Copy Area
8028 * and updates the current migration unit status.
8029 * Use restore_stripes() to form a destination stripe,
8030 * and to write it to the Copy Area.
8031 * Parameters:
8032 * st : supertype information
8033 * dev : imsm device that backup is saved for
8034 * info : general array info
8035 * buf : input buffer
8036 * length : length of data to backup (blocks_per_unit)
8037 * Returns:
8038 * 0 : success
8039 *, -1 : fail
8040 ******************************************************************************/
8041 int save_backup_imsm(struct supertype *st,
8042 struct imsm_dev *dev,
8043 struct mdinfo *info,
8044 void *buf,
8045 int length)
8046 {
8047 int rv = -1;
8048 struct intel_super *super = st->sb;
8049 unsigned long long *target_offsets = NULL;
8050 int *targets = NULL;
8051 int i;
8052 struct imsm_map *map_dest = get_imsm_map(dev, 0);
8053 int new_disks = map_dest->num_members;
8054 int dest_layout = 0;
8055 int dest_chunk;
8056 unsigned long long start;
8057 int data_disks = imsm_num_data_members(dev, 0);
8058
8059 targets = malloc(new_disks * sizeof(int));
8060 if (!targets)
8061 goto abort;
8062
8063 for (i = 0; i < new_disks; i++)
8064 targets[i] = -1;
8065
8066 target_offsets = malloc(new_disks * sizeof(unsigned long long));
8067 if (!target_offsets)
8068 goto abort;
8069
8070 start = info->reshape_progress * 512;
8071 for (i = 0; i < new_disks; i++) {
8072 target_offsets[i] = (unsigned long long)
8073 __le32_to_cpu(super->migr_rec->ckpt_area_pba) * 512;
8074 /* move back copy area adderss, it will be moved forward
8075 * in restore_stripes() using start input variable
8076 */
8077 target_offsets[i] -= start/data_disks;
8078 }
8079
8080 if (open_backup_targets(info, new_disks, targets))
8081 goto abort;
8082
8083 dest_layout = imsm_level_to_layout(map_dest->raid_level);
8084 dest_chunk = __le16_to_cpu(map_dest->blocks_per_strip) * 512;
8085
8086 if (restore_stripes(targets, /* list of dest devices */
8087 target_offsets, /* migration record offsets */
8088 new_disks,
8089 dest_chunk,
8090 map_dest->raid_level,
8091 dest_layout,
8092 -1, /* source backup file descriptor */
8093 0, /* input buf offset
8094 * always 0 buf is already offseted */
8095 start,
8096 length,
8097 buf) != 0) {
8098 fprintf(stderr, Name ": Error restoring stripes\n");
8099 goto abort;
8100 }
8101
8102 rv = 0;
8103
8104 abort:
8105 if (targets) {
8106 for (i = 0; i < new_disks; i++)
8107 if (targets[i] >= 0)
8108 close(targets[i]);
8109 free(targets);
8110 }
8111 free(target_offsets);
8112
8113 return rv;
8114 }
8115
8116 /*******************************************************************************
8117 * Function: save_checkpoint_imsm
8118 * Description: Function called for current unit status update
8119 * in the migration record. It writes it to disk.
8120 * Parameters:
8121 * super : imsm internal array info
8122 * info : general array info
8123 * Returns:
8124 * 0: success
8125 * 1: failure
8126 * 2: failure, means no valid migration record
8127 * / no general migration in progress /
8128 ******************************************************************************/
8129 int save_checkpoint_imsm(struct supertype *st, struct mdinfo *info, int state)
8130 {
8131 struct intel_super *super = st->sb;
8132 unsigned long long blocks_per_unit;
8133 unsigned long long curr_migr_unit;
8134
8135 if (load_imsm_migr_rec(super, info) != 0) {
8136 dprintf("imsm: ERROR: Cannot read migration record "
8137 "for checkpoint save.\n");
8138 return 1;
8139 }
8140
8141 blocks_per_unit = __le32_to_cpu(super->migr_rec->blocks_per_unit);
8142 if (blocks_per_unit == 0) {
8143 dprintf("imsm: no migration in progress.\n");
8144 return 2;
8145 }
8146 curr_migr_unit = info->reshape_progress / blocks_per_unit;
8147 /* check if array is alligned to copy area
8148 * if it is not alligned, add one to current migration unit value
8149 * this can happend on array reshape finish only
8150 */
8151 if (info->reshape_progress % blocks_per_unit)
8152 curr_migr_unit++;
8153
8154 super->migr_rec->curr_migr_unit =
8155 __cpu_to_le32(curr_migr_unit);
8156 super->migr_rec->rec_status = __cpu_to_le32(state);
8157 super->migr_rec->dest_1st_member_lba =
8158 __cpu_to_le32(curr_migr_unit *
8159 __le32_to_cpu(super->migr_rec->dest_depth_per_unit));
8160 if (write_imsm_migr_rec(st) < 0) {
8161 dprintf("imsm: Cannot write migration record "
8162 "outside backup area\n");
8163 return 1;
8164 }
8165
8166 return 0;
8167 }
8168
8169 /*******************************************************************************
8170 * Function: recover_backup_imsm
8171 * Description: Function recovers critical data from the Migration Copy Area
8172 * while assembling an array.
8173 * Parameters:
8174 * super : imsm internal array info
8175 * info : general array info
8176 * Returns:
8177 * 0 : success (or there is no data to recover)
8178 * 1 : fail
8179 ******************************************************************************/
8180 int recover_backup_imsm(struct supertype *st, struct mdinfo *info)
8181 {
8182 struct intel_super *super = st->sb;
8183 struct migr_record *migr_rec = super->migr_rec;
8184 struct imsm_map *map_dest = NULL;
8185 struct intel_dev *id = NULL;
8186 unsigned long long read_offset;
8187 unsigned long long write_offset;
8188 unsigned unit_len;
8189 int *targets = NULL;
8190 int new_disks, i, err;
8191 char *buf = NULL;
8192 int retval = 1;
8193 unsigned long curr_migr_unit = __le32_to_cpu(migr_rec->curr_migr_unit);
8194 unsigned long num_migr_units = __le32_to_cpu(migr_rec->num_migr_units);
8195 char buffer[20];
8196 int skipped_disks = 0;
8197 int max_degradation;
8198
8199 err = sysfs_get_str(info, NULL, "array_state", (char *)buffer, 20);
8200 if (err < 1)
8201 return 1;
8202
8203 /* recover data only during assemblation */
8204 if (strncmp(buffer, "inactive", 8) != 0)
8205 return 0;
8206 /* no data to recover */
8207 if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL)
8208 return 0;
8209 if (curr_migr_unit >= num_migr_units)
8210 return 1;
8211
8212 /* find device during reshape */
8213 for (id = super->devlist; id; id = id->next)
8214 if (is_gen_migration(id->dev))
8215 break;
8216 if (id == NULL)
8217 return 1;
8218
8219 map_dest = get_imsm_map(id->dev, 0);
8220 new_disks = map_dest->num_members;
8221 max_degradation = new_disks - imsm_num_data_members(id->dev, 0);
8222
8223 read_offset = (unsigned long long)
8224 __le32_to_cpu(migr_rec->ckpt_area_pba) * 512;
8225
8226 write_offset = ((unsigned long long)
8227 __le32_to_cpu(migr_rec->dest_1st_member_lba) +
8228 __le32_to_cpu(map_dest->pba_of_lba0)) * 512;
8229
8230 unit_len = __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512;
8231 if (posix_memalign((void **)&buf, 512, unit_len) != 0)
8232 goto abort;
8233 targets = malloc(new_disks * sizeof(int));
8234 if (!targets)
8235 goto abort;
8236
8237 open_backup_targets(info, new_disks, targets);
8238
8239 for (i = 0; i < new_disks; i++) {
8240 if (targets[i] < 0) {
8241 skipped_disks++;
8242 continue;
8243 }
8244 if (lseek64(targets[i], read_offset, SEEK_SET) < 0) {
8245 fprintf(stderr,
8246 Name ": Cannot seek to block: %s\n",
8247 strerror(errno));
8248 goto abort;
8249 }
8250 if ((unsigned)read(targets[i], buf, unit_len) != unit_len) {
8251 fprintf(stderr,
8252 Name ": Cannot read copy area block: %s\n",
8253 strerror(errno));
8254 goto abort;
8255 }
8256 if (lseek64(targets[i], write_offset, SEEK_SET) < 0) {
8257 fprintf(stderr,
8258 Name ": Cannot seek to block: %s\n",
8259 strerror(errno));
8260 goto abort;
8261 }
8262 if ((unsigned)write(targets[i], buf, unit_len) != unit_len) {
8263 fprintf(stderr,
8264 Name ": Cannot restore block: %s\n",
8265 strerror(errno));
8266 goto abort;
8267 }
8268 }
8269
8270 if (skipped_disks > max_degradation) {
8271 fprintf(stderr,
8272 Name ": Cannot restore data from backup."
8273 " Too many failed disks\n");
8274 goto abort;
8275 }
8276
8277 if (save_checkpoint_imsm(st, info, UNIT_SRC_NORMAL)) {
8278 /* ignore error == 2, this can mean end of reshape here
8279 */
8280 dprintf("imsm: Cannot write checkpoint to "
8281 "migration record (UNIT_SRC_NORMAL) during restart\n");
8282 } else
8283 retval = 0;
8284
8285 abort:
8286 if (targets) {
8287 for (i = 0; i < new_disks; i++)
8288 if (targets[i])
8289 close(targets[i]);
8290 free(targets);
8291 }
8292 free(buf);
8293 return retval;
8294 }
8295
8296 static char disk_by_path[] = "/dev/disk/by-path/";
8297
8298 static const char *imsm_get_disk_controller_domain(const char *path)
8299 {
8300 char disk_path[PATH_MAX];
8301 char *drv=NULL;
8302 struct stat st;
8303
8304 strncpy(disk_path, disk_by_path, PATH_MAX - 1);
8305 strncat(disk_path, path, PATH_MAX - strlen(disk_path) - 1);
8306 if (stat(disk_path, &st) == 0) {
8307 struct sys_dev* hba;
8308 char *path=NULL;
8309
8310 path = devt_to_devpath(st.st_rdev);
8311 if (path == NULL)
8312 return "unknown";
8313 hba = find_disk_attached_hba(-1, path);
8314 if (hba && hba->type == SYS_DEV_SAS)
8315 drv = "isci";
8316 else if (hba && hba->type == SYS_DEV_SATA)
8317 drv = "ahci";
8318 else
8319 drv = "unknown";
8320 dprintf("path: %s hba: %s attached: %s\n",
8321 path, (hba) ? hba->path : "NULL", drv);
8322 free(path);
8323 if (hba)
8324 free_sys_dev(&hba);
8325 }
8326 return drv;
8327 }
8328
8329 static int imsm_find_array_minor_by_subdev(int subdev, int container, int *minor)
8330 {
8331 char subdev_name[20];
8332 struct mdstat_ent *mdstat;
8333
8334 sprintf(subdev_name, "%d", subdev);
8335 mdstat = mdstat_by_subdev(subdev_name, container);
8336 if (!mdstat)
8337 return -1;
8338
8339 *minor = mdstat->devnum;
8340 free_mdstat(mdstat);
8341 return 0;
8342 }
8343
8344 static int imsm_reshape_is_allowed_on_container(struct supertype *st,
8345 struct geo_params *geo,
8346 int *old_raid_disks)
8347 {
8348 /* currently we only support increasing the number of devices
8349 * for a container. This increases the number of device for each
8350 * member array. They must all be RAID0 or RAID5.
8351 */
8352 int ret_val = 0;
8353 struct mdinfo *info, *member;
8354 int devices_that_can_grow = 0;
8355
8356 dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): "
8357 "st->devnum = (%i)\n",
8358 st->devnum);
8359
8360 if (geo->size != -1 ||
8361 geo->level != UnSet ||
8362 geo->layout != UnSet ||
8363 geo->chunksize != 0 ||
8364 geo->raid_disks == UnSet) {
8365 dprintf("imsm: Container operation is allowed for "
8366 "raid disks number change only.\n");
8367 return ret_val;
8368 }
8369
8370 info = container_content_imsm(st, NULL);
8371 for (member = info; member; member = member->next) {
8372 int result;
8373 int minor;
8374
8375 dprintf("imsm: checking device_num: %i\n",
8376 member->container_member);
8377
8378 if (geo->raid_disks <= member->array.raid_disks) {
8379 /* we work on container for Online Capacity Expansion
8380 * only so raid_disks has to grow
8381 */
8382 dprintf("imsm: for container operation raid disks "
8383 "increase is required\n");
8384 break;
8385 }
8386
8387 if ((info->array.level != 0) &&
8388 (info->array.level != 5)) {
8389 /* we cannot use this container with other raid level
8390 */
8391 dprintf("imsm: for container operation wrong"
8392 " raid level (%i) detected\n",
8393 info->array.level);
8394 break;
8395 } else {
8396 /* check for platform support
8397 * for this raid level configuration
8398 */
8399 struct intel_super *super = st->sb;
8400 if (!is_raid_level_supported(super->orom,
8401 member->array.level,
8402 geo->raid_disks)) {
8403 dprintf("platform does not support raid%d with"
8404 " %d disk%s\n",
8405 info->array.level,
8406 geo->raid_disks,
8407 geo->raid_disks > 1 ? "s" : "");
8408 break;
8409 }
8410 /* check if component size is aligned to chunk size
8411 */
8412 if (info->component_size %
8413 (info->array.chunk_size/512)) {
8414 dprintf("Component size is not aligned to "
8415 "chunk size\n");
8416 break;
8417 }
8418 }
8419
8420 if (*old_raid_disks &&
8421 info->array.raid_disks != *old_raid_disks)
8422 break;
8423 *old_raid_disks = info->array.raid_disks;
8424
8425 /* All raid5 and raid0 volumes in container
8426 * have to be ready for Online Capacity Expansion
8427 * so they need to be assembled. We have already
8428 * checked that no recovery etc is happening.
8429 */
8430 result = imsm_find_array_minor_by_subdev(member->container_member,
8431 st->container_dev,
8432 &minor);
8433 if (result < 0) {
8434 dprintf("imsm: cannot find array\n");
8435 break;
8436 }
8437 devices_that_can_grow++;
8438 }
8439 sysfs_free(info);
8440 if (!member && devices_that_can_grow)
8441 ret_val = 1;
8442
8443 if (ret_val)
8444 dprintf("\tContainer operation allowed\n");
8445 else
8446 dprintf("\tError: %i\n", ret_val);
8447
8448 return ret_val;
8449 }
8450
8451 /* Function: get_spares_for_grow
8452 * Description: Allocates memory and creates list of spare devices
8453 * avaliable in container. Checks if spare drive size is acceptable.
8454 * Parameters: Pointer to the supertype structure
8455 * Returns: Pointer to the list of spare devices (mdinfo structure) on success,
8456 * NULL if fail
8457 */
8458 static struct mdinfo *get_spares_for_grow(struct supertype *st)
8459 {
8460 unsigned long long min_size = min_acceptable_spare_size_imsm(st);
8461 return container_choose_spares(st, min_size, NULL, NULL, NULL, 0);
8462 }
8463
8464 /******************************************************************************
8465 * function: imsm_create_metadata_update_for_reshape
8466 * Function creates update for whole IMSM container.
8467 *
8468 ******************************************************************************/
8469 static int imsm_create_metadata_update_for_reshape(
8470 struct supertype *st,
8471 struct geo_params *geo,
8472 int old_raid_disks,
8473 struct imsm_update_reshape **updatep)
8474 {
8475 struct intel_super *super = st->sb;
8476 struct imsm_super *mpb = super->anchor;
8477 int update_memory_size = 0;
8478 struct imsm_update_reshape *u = NULL;
8479 struct mdinfo *spares = NULL;
8480 int i;
8481 int delta_disks = 0;
8482 struct mdinfo *dev;
8483
8484 dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n",
8485 geo->raid_disks);
8486
8487 delta_disks = geo->raid_disks - old_raid_disks;
8488
8489 /* size of all update data without anchor */
8490 update_memory_size = sizeof(struct imsm_update_reshape);
8491
8492 /* now add space for spare disks that we need to add. */
8493 update_memory_size += sizeof(u->new_disks[0]) * (delta_disks - 1);
8494
8495 u = calloc(1, update_memory_size);
8496 if (u == NULL) {
8497 dprintf("error: "
8498 "cannot get memory for imsm_update_reshape update\n");
8499 return 0;
8500 }
8501 u->type = update_reshape_container_disks;
8502 u->old_raid_disks = old_raid_disks;
8503 u->new_raid_disks = geo->raid_disks;
8504
8505 /* now get spare disks list
8506 */
8507 spares = get_spares_for_grow(st);
8508
8509 if (spares == NULL
8510 || delta_disks > spares->array.spare_disks) {
8511 fprintf(stderr, Name ": imsm: ERROR: Cannot get spare devices "
8512 "for %s.\n", geo->dev_name);
8513 i = -1;
8514 goto abort;
8515 }
8516
8517 /* we have got spares
8518 * update disk list in imsm_disk list table in anchor
8519 */
8520 dprintf("imsm: %i spares are available.\n\n",
8521 spares->array.spare_disks);
8522
8523 dev = spares->devs;
8524 for (i = 0; i < delta_disks; i++) {
8525 struct dl *dl;
8526
8527 if (dev == NULL)
8528 break;
8529 u->new_disks[i] = makedev(dev->disk.major,
8530 dev->disk.minor);
8531 dl = get_disk_super(super, dev->disk.major, dev->disk.minor);
8532 dl->index = mpb->num_disks;
8533 mpb->num_disks++;
8534 dev = dev->next;
8535 }
8536
8537 abort:
8538 /* free spares
8539 */
8540 sysfs_free(spares);
8541
8542 dprintf("imsm: reshape update preparation :");
8543 if (i == delta_disks) {
8544 dprintf(" OK\n");
8545 *updatep = u;
8546 return update_memory_size;
8547 }
8548 free(u);
8549 dprintf(" Error\n");
8550
8551 return 0;
8552 }
8553
8554 /******************************************************************************
8555 * function: imsm_create_metadata_update_for_migration()
8556 * Creates update for IMSM array.
8557 *
8558 ******************************************************************************/
8559 static int imsm_create_metadata_update_for_migration(
8560 struct supertype *st,
8561 struct geo_params *geo,
8562 struct imsm_update_reshape_migration **updatep)
8563 {
8564 struct intel_super *super = st->sb;
8565 int update_memory_size = 0;
8566 struct imsm_update_reshape_migration *u = NULL;
8567 struct imsm_dev *dev;
8568 int previous_level = -1;
8569
8570 dprintf("imsm_create_metadata_update_for_migration(enter)"
8571 " New Level = %i\n", geo->level);
8572
8573 /* size of all update data without anchor */
8574 update_memory_size = sizeof(struct imsm_update_reshape_migration);
8575
8576 u = calloc(1, update_memory_size);
8577 if (u == NULL) {
8578 dprintf("error: cannot get memory for "
8579 "imsm_create_metadata_update_for_migration\n");
8580 return 0;
8581 }
8582 u->type = update_reshape_migration;
8583 u->subdev = super->current_vol;
8584 u->new_level = geo->level;
8585 u->new_layout = geo->layout;
8586 u->new_raid_disks = u->old_raid_disks = geo->raid_disks;
8587 u->new_disks[0] = -1;
8588 u->new_chunksize = -1;
8589
8590 dev = get_imsm_dev(super, u->subdev);
8591 if (dev) {
8592 struct imsm_map *map;
8593
8594 map = get_imsm_map(dev, 0);
8595 if (map) {
8596 int current_chunk_size =
8597 __le16_to_cpu(map->blocks_per_strip) / 2;
8598
8599 if (geo->chunksize != current_chunk_size) {
8600 u->new_chunksize = geo->chunksize / 1024;
8601 dprintf("imsm: "
8602 "chunk size change from %i to %i\n",
8603 current_chunk_size, u->new_chunksize);
8604 }
8605 previous_level = map->raid_level;
8606 }
8607 }
8608 if ((geo->level == 5) && (previous_level == 0)) {
8609 struct mdinfo *spares = NULL;
8610
8611 u->new_raid_disks++;
8612 spares = get_spares_for_grow(st);
8613 if ((spares == NULL) || (spares->array.spare_disks < 1)) {
8614 free(u);
8615 sysfs_free(spares);
8616 update_memory_size = 0;
8617 dprintf("error: cannot get spare device "
8618 "for requested migration");
8619 return 0;
8620 }
8621 sysfs_free(spares);
8622 }
8623 dprintf("imsm: reshape update preparation : OK\n");
8624 *updatep = u;
8625
8626 return update_memory_size;
8627 }
8628
8629 static void imsm_update_metadata_locally(struct supertype *st,
8630 void *buf, int len)
8631 {
8632 struct metadata_update mu;
8633
8634 mu.buf = buf;
8635 mu.len = len;
8636 mu.space = NULL;
8637 mu.space_list = NULL;
8638 mu.next = NULL;
8639 imsm_prepare_update(st, &mu);
8640 imsm_process_update(st, &mu);
8641
8642 while (mu.space_list) {
8643 void **space = mu.space_list;
8644 mu.space_list = *space;
8645 free(space);
8646 }
8647 }
8648
8649 /***************************************************************************
8650 * Function: imsm_analyze_change
8651 * Description: Function analyze change for single volume
8652 * and validate if transition is supported
8653 * Parameters: Geometry parameters, supertype structure
8654 * Returns: Operation type code on success, -1 if fail
8655 ****************************************************************************/
8656 enum imsm_reshape_type imsm_analyze_change(struct supertype *st,
8657 struct geo_params *geo)
8658 {
8659 struct mdinfo info;
8660 int change = -1;
8661 int check_devs = 0;
8662 int chunk;
8663
8664 getinfo_super_imsm_volume(st, &info, NULL);
8665 if ((geo->level != info.array.level) &&
8666 (geo->level >= 0) &&
8667 (geo->level != UnSet)) {
8668 switch (info.array.level) {
8669 case 0:
8670 if (geo->level == 5) {
8671 change = CH_MIGRATION;
8672 if (geo->layout != ALGORITHM_LEFT_ASYMMETRIC) {
8673 fprintf(stderr,
8674 Name " Error. Requested Layout "
8675 "not supported (left-asymmetric layout "
8676 "is supported only)!\n");
8677 change = -1;
8678 goto analyse_change_exit;
8679 }
8680 check_devs = 1;
8681 }
8682 if (geo->level == 10) {
8683 change = CH_TAKEOVER;
8684 check_devs = 1;
8685 }
8686 break;
8687 case 1:
8688 if (geo->level == 0) {
8689 change = CH_TAKEOVER;
8690 check_devs = 1;
8691 }
8692 break;
8693 case 10:
8694 if (geo->level == 0) {
8695 change = CH_TAKEOVER;
8696 check_devs = 1;
8697 }
8698 break;
8699 }
8700 if (change == -1) {
8701 fprintf(stderr,
8702 Name " Error. Level Migration from %d to %d "
8703 "not supported!\n",
8704 info.array.level, geo->level);
8705 goto analyse_change_exit;
8706 }
8707 } else
8708 geo->level = info.array.level;
8709
8710 if ((geo->layout != info.array.layout)
8711 && ((geo->layout != UnSet) && (geo->layout != -1))) {
8712 change = CH_MIGRATION;
8713 if ((info.array.layout == 0)
8714 && (info.array.level == 5)
8715 && (geo->layout == 5)) {
8716 /* reshape 5 -> 4 */
8717 } else if ((info.array.layout == 5)
8718 && (info.array.level == 5)
8719 && (geo->layout == 0)) {
8720 /* reshape 4 -> 5 */
8721 geo->layout = 0;
8722 geo->level = 5;
8723 } else {
8724 fprintf(stderr,
8725 Name " Error. Layout Migration from %d to %d "
8726 "not supported!\n",
8727 info.array.layout, geo->layout);
8728 change = -1;
8729 goto analyse_change_exit;
8730 }
8731 } else
8732 geo->layout = info.array.layout;
8733
8734 if ((geo->chunksize > 0) && (geo->chunksize != UnSet)
8735 && (geo->chunksize != info.array.chunk_size))
8736 change = CH_MIGRATION;
8737 else
8738 geo->chunksize = info.array.chunk_size;
8739
8740 chunk = geo->chunksize / 1024;
8741 if (!validate_geometry_imsm(st,
8742 geo->level,
8743 geo->layout,
8744 geo->raid_disks,
8745 &chunk,
8746 geo->size,
8747 0, 0, 1))
8748 change = -1;
8749
8750 if (check_devs) {
8751 struct intel_super *super = st->sb;
8752 struct imsm_super *mpb = super->anchor;
8753
8754 if (mpb->num_raid_devs > 1) {
8755 fprintf(stderr,
8756 Name " Error. Cannot perform operation on %s"
8757 "- for this operation it MUST be single "
8758 "array in container\n",
8759 geo->dev_name);
8760 change = -1;
8761 }
8762 }
8763
8764 analyse_change_exit:
8765
8766 return change;
8767 }
8768
8769 int imsm_takeover(struct supertype *st, struct geo_params *geo)
8770 {
8771 struct intel_super *super = st->sb;
8772 struct imsm_update_takeover *u;
8773
8774 u = malloc(sizeof(struct imsm_update_takeover));
8775 if (u == NULL)
8776 return 1;
8777
8778 u->type = update_takeover;
8779 u->subarray = super->current_vol;
8780
8781 /* 10->0 transition */
8782 if (geo->level == 0)
8783 u->direction = R10_TO_R0;
8784
8785 /* 0->10 transition */
8786 if (geo->level == 10)
8787 u->direction = R0_TO_R10;
8788
8789 /* update metadata locally */
8790 imsm_update_metadata_locally(st, u,
8791 sizeof(struct imsm_update_takeover));
8792 /* and possibly remotely */
8793 if (st->update_tail)
8794 append_metadata_update(st, u,
8795 sizeof(struct imsm_update_takeover));
8796 else
8797 free(u);
8798
8799 return 0;
8800 }
8801
8802 static int imsm_reshape_super(struct supertype *st, long long size, int level,
8803 int layout, int chunksize, int raid_disks,
8804 int delta_disks, char *backup, char *dev,
8805 int verbose)
8806 {
8807 int ret_val = 1;
8808 struct geo_params geo;
8809
8810 dprintf("imsm: reshape_super called.\n");
8811
8812 memset(&geo, 0, sizeof(struct geo_params));
8813
8814 geo.dev_name = dev;
8815 geo.dev_id = st->devnum;
8816 geo.size = size;
8817 geo.level = level;
8818 geo.layout = layout;
8819 geo.chunksize = chunksize;
8820 geo.raid_disks = raid_disks;
8821 if (delta_disks != UnSet)
8822 geo.raid_disks += delta_disks;
8823
8824 dprintf("\tfor level : %i\n", geo.level);
8825 dprintf("\tfor raid_disks : %i\n", geo.raid_disks);
8826
8827 if (experimental() == 0)
8828 return ret_val;
8829
8830 if (st->container_dev == st->devnum) {
8831 /* On container level we can only increase number of devices. */
8832 dprintf("imsm: info: Container operation\n");
8833 int old_raid_disks = 0;
8834
8835 if (imsm_reshape_is_allowed_on_container(
8836 st, &geo, &old_raid_disks)) {
8837 struct imsm_update_reshape *u = NULL;
8838 int len;
8839
8840 len = imsm_create_metadata_update_for_reshape(
8841 st, &geo, old_raid_disks, &u);
8842
8843 if (len <= 0) {
8844 dprintf("imsm: Cannot prepare update\n");
8845 goto exit_imsm_reshape_super;
8846 }
8847
8848 ret_val = 0;
8849 /* update metadata locally */
8850 imsm_update_metadata_locally(st, u, len);
8851 /* and possibly remotely */
8852 if (st->update_tail)
8853 append_metadata_update(st, u, len);
8854 else
8855 free(u);
8856
8857 } else {
8858 fprintf(stderr, Name ": (imsm) Operation "
8859 "is not allowed on this container\n");
8860 }
8861 } else {
8862 /* On volume level we support following operations
8863 * - takeover: raid10 -> raid0; raid0 -> raid10
8864 * - chunk size migration
8865 * - migration: raid5 -> raid0; raid0 -> raid5
8866 */
8867 struct intel_super *super = st->sb;
8868 struct intel_dev *dev = super->devlist;
8869 int change, devnum;
8870 dprintf("imsm: info: Volume operation\n");
8871 /* find requested device */
8872 while (dev) {
8873 if (imsm_find_array_minor_by_subdev(
8874 dev->index, st->container_dev, &devnum) == 0
8875 && devnum == geo.dev_id)
8876 break;
8877 dev = dev->next;
8878 }
8879 if (dev == NULL) {
8880 fprintf(stderr, Name " Cannot find %s (%i) subarray\n",
8881 geo.dev_name, geo.dev_id);
8882 goto exit_imsm_reshape_super;
8883 }
8884 super->current_vol = dev->index;
8885 change = imsm_analyze_change(st, &geo);
8886 switch (change) {
8887 case CH_TAKEOVER:
8888 ret_val = imsm_takeover(st, &geo);
8889 break;
8890 case CH_MIGRATION: {
8891 struct imsm_update_reshape_migration *u = NULL;
8892 int len =
8893 imsm_create_metadata_update_for_migration(
8894 st, &geo, &u);
8895 if (len < 1) {
8896 dprintf("imsm: "
8897 "Cannot prepare update\n");
8898 break;
8899 }
8900 ret_val = 0;
8901 /* update metadata locally */
8902 imsm_update_metadata_locally(st, u, len);
8903 /* and possibly remotely */
8904 if (st->update_tail)
8905 append_metadata_update(st, u, len);
8906 else
8907 free(u);
8908 }
8909 break;
8910 default:
8911 ret_val = 1;
8912 }
8913 }
8914
8915 exit_imsm_reshape_super:
8916 dprintf("imsm: reshape_super Exit code = %i\n", ret_val);
8917 return ret_val;
8918 }
8919
8920 /*******************************************************************************
8921 * Function: wait_for_reshape_imsm
8922 * Description: Function writes new sync_max value and waits until
8923 * reshape process reach new position
8924 * Parameters:
8925 * sra : general array info
8926 * ndata : number of disks in new array's layout
8927 * Returns:
8928 * 0 : success,
8929 * 1 : there is no reshape in progress,
8930 * -1 : fail
8931 ******************************************************************************/
8932 int wait_for_reshape_imsm(struct mdinfo *sra, int ndata)
8933 {
8934 int fd = sysfs_get_fd(sra, NULL, "reshape_position");
8935 unsigned long long completed;
8936 /* to_complete : new sync_max position */
8937 unsigned long long to_complete = sra->reshape_progress;
8938 unsigned long long position_to_set = to_complete / ndata;
8939
8940 if (fd < 0) {
8941 dprintf("imsm: wait_for_reshape_imsm() "
8942 "cannot open reshape_position\n");
8943 return 1;
8944 }
8945
8946 if (sysfs_fd_get_ll(fd, &completed) < 0) {
8947 dprintf("imsm: wait_for_reshape_imsm() "
8948 "cannot read reshape_position (no reshape in progres)\n");
8949 close(fd);
8950 return 0;
8951 }
8952
8953 if (completed > to_complete) {
8954 dprintf("imsm: wait_for_reshape_imsm() "
8955 "wrong next position to set %llu (%llu)\n",
8956 to_complete, completed);
8957 close(fd);
8958 return -1;
8959 }
8960 dprintf("Position set: %llu\n", position_to_set);
8961 if (sysfs_set_num(sra, NULL, "sync_max",
8962 position_to_set) != 0) {
8963 dprintf("imsm: wait_for_reshape_imsm() "
8964 "cannot set reshape position to %llu\n",
8965 position_to_set);
8966 close(fd);
8967 return -1;
8968 }
8969
8970 do {
8971 char action[20];
8972 fd_set rfds;
8973 FD_ZERO(&rfds);
8974 FD_SET(fd, &rfds);
8975 select(fd+1, &rfds, NULL, NULL, NULL);
8976 if (sysfs_get_str(sra, NULL, "sync_action",
8977 action, 20) > 0 &&
8978 strncmp(action, "reshape", 7) != 0)
8979 break;
8980 if (sysfs_fd_get_ll(fd, &completed) < 0) {
8981 dprintf("imsm: wait_for_reshape_imsm() "
8982 "cannot read reshape_position (in loop)\n");
8983 close(fd);
8984 return 1;
8985 }
8986 } while (completed < to_complete);
8987 close(fd);
8988 return 0;
8989
8990 }
8991
8992 /*******************************************************************************
8993 * Function: check_degradation_change
8994 * Description: Check that array hasn't become failed.
8995 * Parameters:
8996 * info : for sysfs access
8997 * sources : source disks descriptors
8998 * degraded: previous degradation level
8999 * Returns:
9000 * degradation level
9001 ******************************************************************************/
9002 int check_degradation_change(struct mdinfo *info,
9003 int *sources,
9004 int degraded)
9005 {
9006 unsigned long long new_degraded;
9007 sysfs_get_ll(info, NULL, "degraded", &new_degraded);
9008 if (new_degraded != (unsigned long long)degraded) {
9009 /* check each device to ensure it is still working */
9010 struct mdinfo *sd;
9011 new_degraded = 0;
9012 for (sd = info->devs ; sd ; sd = sd->next) {
9013 if (sd->disk.state & (1<<MD_DISK_FAULTY))
9014 continue;
9015 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
9016 char sbuf[20];
9017 if (sysfs_get_str(info,
9018 sd, "state", sbuf, 20) < 0 ||
9019 strstr(sbuf, "faulty") ||
9020 strstr(sbuf, "in_sync") == NULL) {
9021 /* this device is dead */
9022 sd->disk.state = (1<<MD_DISK_FAULTY);
9023 if (sd->disk.raid_disk >= 0 &&
9024 sources[sd->disk.raid_disk] >= 0) {
9025 close(sources[
9026 sd->disk.raid_disk]);
9027 sources[sd->disk.raid_disk] =
9028 -1;
9029 }
9030 new_degraded++;
9031 }
9032 }
9033 }
9034 }
9035
9036 return new_degraded;
9037 }
9038
9039 /*******************************************************************************
9040 * Function: imsm_manage_reshape
9041 * Description: Function finds array under reshape and it manages reshape
9042 * process. It creates stripes backups (if required) and sets
9043 * checheckpoits.
9044 * Parameters:
9045 * afd : Backup handle (nattive) - not used
9046 * sra : general array info
9047 * reshape : reshape parameters - not used
9048 * st : supertype structure
9049 * blocks : size of critical section [blocks]
9050 * fds : table of source device descriptor
9051 * offsets : start of array (offest per devices)
9052 * dests : not used
9053 * destfd : table of destination device descriptor
9054 * destoffsets : table of destination offsets (per device)
9055 * Returns:
9056 * 1 : success, reshape is done
9057 * 0 : fail
9058 ******************************************************************************/
9059 static int imsm_manage_reshape(
9060 int afd, struct mdinfo *sra, struct reshape *reshape,
9061 struct supertype *st, unsigned long backup_blocks,
9062 int *fds, unsigned long long *offsets,
9063 int dests, int *destfd, unsigned long long *destoffsets)
9064 {
9065 int ret_val = 0;
9066 struct intel_super *super = st->sb;
9067 struct intel_dev *dv = NULL;
9068 struct imsm_dev *dev = NULL;
9069 struct imsm_map *map_src;
9070 int migr_vol_qan = 0;
9071 int ndata, odata; /* [bytes] */
9072 int chunk; /* [bytes] */
9073 struct migr_record *migr_rec;
9074 char *buf = NULL;
9075 unsigned int buf_size; /* [bytes] */
9076 unsigned long long max_position; /* array size [bytes] */
9077 unsigned long long next_step; /* [blocks]/[bytes] */
9078 unsigned long long old_data_stripe_length;
9079 unsigned long long start_src; /* [bytes] */
9080 unsigned long long start; /* [bytes] */
9081 unsigned long long start_buf_shift; /* [bytes] */
9082 int degraded = 0;
9083 int source_layout = 0;
9084
9085 if (!fds || !offsets || !sra)
9086 goto abort;
9087
9088 /* Find volume during the reshape */
9089 for (dv = super->devlist; dv; dv = dv->next) {
9090 if (dv->dev->vol.migr_type == MIGR_GEN_MIGR
9091 && dv->dev->vol.migr_state == 1) {
9092 dev = dv->dev;
9093 migr_vol_qan++;
9094 }
9095 }
9096 /* Only one volume can migrate at the same time */
9097 if (migr_vol_qan != 1) {
9098 fprintf(stderr, Name " : %s", migr_vol_qan ?
9099 "Number of migrating volumes greater than 1\n" :
9100 "There is no volume during migrationg\n");
9101 goto abort;
9102 }
9103
9104 map_src = get_imsm_map(dev, 1);
9105 if (map_src == NULL)
9106 goto abort;
9107
9108 ndata = imsm_num_data_members(dev, 0);
9109 odata = imsm_num_data_members(dev, 1);
9110
9111 chunk = __le16_to_cpu(map_src->blocks_per_strip) * 512;
9112 old_data_stripe_length = odata * chunk;
9113
9114 migr_rec = super->migr_rec;
9115
9116 /* initialize migration record for start condition */
9117 if (sra->reshape_progress == 0)
9118 init_migr_record_imsm(st, dev, sra);
9119 else {
9120 if (__le32_to_cpu(migr_rec->rec_status) != UNIT_SRC_NORMAL) {
9121 dprintf("imsm: cannot restart migration when data "
9122 "are present in copy area.\n");
9123 goto abort;
9124 }
9125 }
9126
9127 /* size for data */
9128 buf_size = __le32_to_cpu(migr_rec->blocks_per_unit) * 512;
9129 /* extend buffer size for parity disk */
9130 buf_size += __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512;
9131 /* add space for stripe aligment */
9132 buf_size += old_data_stripe_length;
9133 if (posix_memalign((void **)&buf, 4096, buf_size)) {
9134 dprintf("imsm: Cannot allocate checpoint buffer\n");
9135 goto abort;
9136 }
9137
9138 max_position = sra->component_size * ndata;
9139 source_layout = imsm_level_to_layout(map_src->raid_level);
9140
9141 while (__le32_to_cpu(migr_rec->curr_migr_unit) <
9142 __le32_to_cpu(migr_rec->num_migr_units)) {
9143 /* current reshape position [blocks] */
9144 unsigned long long current_position =
9145 __le32_to_cpu(migr_rec->blocks_per_unit)
9146 * __le32_to_cpu(migr_rec->curr_migr_unit);
9147 unsigned long long border;
9148
9149 /* Check that array hasn't become failed.
9150 */
9151 degraded = check_degradation_change(sra, fds, degraded);
9152 if (degraded > 1) {
9153 dprintf("imsm: Abort reshape due to degradation"
9154 " level (%i)\n", degraded);
9155 goto abort;
9156 }
9157
9158 next_step = __le32_to_cpu(migr_rec->blocks_per_unit);
9159
9160 if ((current_position + next_step) > max_position)
9161 next_step = max_position - current_position;
9162
9163 start = current_position * 512;
9164
9165 /* allign reading start to old geometry */
9166 start_buf_shift = start % old_data_stripe_length;
9167 start_src = start - start_buf_shift;
9168
9169 border = (start_src / odata) - (start / ndata);
9170 border /= 512;
9171 if (border <= __le32_to_cpu(migr_rec->dest_depth_per_unit)) {
9172 /* save critical stripes to buf
9173 * start - start address of current unit
9174 * to backup [bytes]
9175 * start_src - start address of current unit
9176 * to backup alligned to source array
9177 * [bytes]
9178 */
9179 unsigned long long next_step_filler = 0;
9180 unsigned long long copy_length = next_step * 512;
9181
9182 /* allign copy area length to stripe in old geometry */
9183 next_step_filler = ((copy_length + start_buf_shift)
9184 % old_data_stripe_length);
9185 if (next_step_filler)
9186 next_step_filler = (old_data_stripe_length
9187 - next_step_filler);
9188 dprintf("save_stripes() parameters: start = %llu,"
9189 "\tstart_src = %llu,\tnext_step*512 = %llu,"
9190 "\tstart_in_buf_shift = %llu,"
9191 "\tnext_step_filler = %llu\n",
9192 start, start_src, copy_length,
9193 start_buf_shift, next_step_filler);
9194
9195 if (save_stripes(fds, offsets, map_src->num_members,
9196 chunk, map_src->raid_level,
9197 source_layout, 0, NULL, start_src,
9198 copy_length +
9199 next_step_filler + start_buf_shift,
9200 buf)) {
9201 dprintf("imsm: Cannot save stripes"
9202 " to buffer\n");
9203 goto abort;
9204 }
9205 /* Convert data to destination format and store it
9206 * in backup general migration area
9207 */
9208 if (save_backup_imsm(st, dev, sra,
9209 buf + start_buf_shift, copy_length)) {
9210 dprintf("imsm: Cannot save stripes to "
9211 "target devices\n");
9212 goto abort;
9213 }
9214 if (save_checkpoint_imsm(st, sra,
9215 UNIT_SRC_IN_CP_AREA)) {
9216 dprintf("imsm: Cannot write checkpoint to "
9217 "migration record (UNIT_SRC_IN_CP_AREA)\n");
9218 goto abort;
9219 }
9220 } else {
9221 /* set next step to use whole border area */
9222 border /= next_step;
9223 if (border > 1)
9224 next_step *= border;
9225 }
9226 /* When data backed up, checkpoint stored,
9227 * kick the kernel to reshape unit of data
9228 */
9229 next_step = next_step + sra->reshape_progress;
9230 /* limit next step to array max position */
9231 if (next_step > max_position)
9232 next_step = max_position;
9233 sysfs_set_num(sra, NULL, "suspend_lo", sra->reshape_progress);
9234 sysfs_set_num(sra, NULL, "suspend_hi", next_step);
9235 sra->reshape_progress = next_step;
9236
9237 /* wait until reshape finish */
9238 if (wait_for_reshape_imsm(sra, ndata) < 0) {
9239 dprintf("wait_for_reshape_imsm returned error!\n");
9240 goto abort;
9241 }
9242
9243 if (save_checkpoint_imsm(st, sra, UNIT_SRC_NORMAL) == 1) {
9244 /* ignore error == 2, this can mean end of reshape here
9245 */
9246 dprintf("imsm: Cannot write checkpoint to "
9247 "migration record (UNIT_SRC_NORMAL)\n");
9248 goto abort;
9249 }
9250
9251 }
9252
9253 /* return '1' if done */
9254 ret_val = 1;
9255 abort:
9256 free(buf);
9257 abort_reshape(sra);
9258
9259 return ret_val;
9260 }
9261 #endif /* MDASSEMBLE */
9262
9263 struct superswitch super_imsm = {
9264 #ifndef MDASSEMBLE
9265 .examine_super = examine_super_imsm,
9266 .brief_examine_super = brief_examine_super_imsm,
9267 .brief_examine_subarrays = brief_examine_subarrays_imsm,
9268 .export_examine_super = export_examine_super_imsm,
9269 .detail_super = detail_super_imsm,
9270 .brief_detail_super = brief_detail_super_imsm,
9271 .write_init_super = write_init_super_imsm,
9272 .validate_geometry = validate_geometry_imsm,
9273 .add_to_super = add_to_super_imsm,
9274 .remove_from_super = remove_from_super_imsm,
9275 .detail_platform = detail_platform_imsm,
9276 .kill_subarray = kill_subarray_imsm,
9277 .update_subarray = update_subarray_imsm,
9278 .load_container = load_container_imsm,
9279 .default_geometry = default_geometry_imsm,
9280 .get_disk_controller_domain = imsm_get_disk_controller_domain,
9281 .reshape_super = imsm_reshape_super,
9282 .manage_reshape = imsm_manage_reshape,
9283 .recover_backup = recover_backup_imsm,
9284 #endif
9285 .match_home = match_home_imsm,
9286 .uuid_from_super= uuid_from_super_imsm,
9287 .getinfo_super = getinfo_super_imsm,
9288 .getinfo_super_disks = getinfo_super_disks_imsm,
9289 .update_super = update_super_imsm,
9290
9291 .avail_size = avail_size_imsm,
9292 .min_acceptable_spare_size = min_acceptable_spare_size_imsm,
9293
9294 .compare_super = compare_super_imsm,
9295
9296 .load_super = load_super_imsm,
9297 .init_super = init_super_imsm,
9298 .store_super = store_super_imsm,
9299 .free_super = free_super_imsm,
9300 .match_metadata_desc = match_metadata_desc_imsm,
9301 .container_content = container_content_imsm,
9302
9303
9304 .external = 1,
9305 .name = "imsm",
9306
9307 #ifndef MDASSEMBLE
9308 /* for mdmon */
9309 .open_new = imsm_open_new,
9310 .set_array_state= imsm_set_array_state,
9311 .set_disk = imsm_set_disk,
9312 .sync_metadata = imsm_sync_metadata,
9313 .activate_spare = imsm_activate_spare,
9314 .process_update = imsm_process_update,
9315 .prepare_update = imsm_prepare_update,
9316 #endif /* MDASSEMBLE */
9317 };