]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - Intel(R) Matrix Storage Manager Support | |
3 | * | |
4 | * Copyright (C) 2002-2008 Intel Corporation | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms and conditions of the GNU General Public License, | |
8 | * version 2, as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
18 | */ | |
19 | ||
20 | #define HAVE_STDINT_H 1 | |
21 | #include "mdadm.h" | |
22 | #include "mdmon.h" | |
23 | #include "sha1.h" | |
24 | #include "platform-intel.h" | |
25 | #include <values.h> | |
26 | #include <scsi/sg.h> | |
27 | #include <ctype.h> | |
28 | #include <dirent.h> | |
29 | ||
30 | /* MPB == Metadata Parameter Block */ | |
31 | #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. " | |
32 | #define MPB_SIG_LEN (strlen(MPB_SIGNATURE)) | |
33 | #define MPB_VERSION_RAID0 "1.0.00" | |
34 | #define MPB_VERSION_RAID1 "1.1.00" | |
35 | #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00" | |
36 | #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01" | |
37 | #define MPB_VERSION_RAID5 "1.2.02" | |
38 | #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04" | |
39 | #define MPB_VERSION_CNG "1.2.06" | |
40 | #define MPB_VERSION_ATTRIBS "1.3.00" | |
41 | #define MAX_SIGNATURE_LENGTH 32 | |
42 | #define MAX_RAID_SERIAL_LEN 16 | |
43 | ||
44 | #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000) | |
45 | #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000) | |
46 | #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000) | |
47 | #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001) | |
48 | #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002) | |
49 | #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004) | |
50 | #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008) | |
51 | #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010) | |
52 | #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020) | |
53 | ||
54 | #define MPB_SECTOR_CNT 418 | |
55 | #define IMSM_RESERVED_SECTORS 4096 | |
56 | #define SECT_PER_MB_SHIFT 11 | |
57 | ||
58 | /* Disk configuration info. */ | |
59 | #define IMSM_MAX_DEVICES 255 | |
60 | struct imsm_disk { | |
61 | __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */ | |
62 | __u32 total_blocks; /* 0xE8 - 0xEB total blocks */ | |
63 | __u32 scsi_id; /* 0xEC - 0xEF scsi ID */ | |
64 | #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */ | |
65 | #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */ | |
66 | #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */ | |
67 | __u32 status; /* 0xF0 - 0xF3 */ | |
68 | __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */ | |
69 | #define IMSM_DISK_FILLERS 4 | |
70 | __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */ | |
71 | }; | |
72 | ||
73 | /* RAID map configuration infos. */ | |
74 | struct imsm_map { | |
75 | __u32 pba_of_lba0; /* start address of partition */ | |
76 | __u32 blocks_per_member;/* blocks per member */ | |
77 | __u32 num_data_stripes; /* number of data stripes */ | |
78 | __u16 blocks_per_strip; | |
79 | __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */ | |
80 | #define IMSM_T_STATE_NORMAL 0 | |
81 | #define IMSM_T_STATE_UNINITIALIZED 1 | |
82 | #define IMSM_T_STATE_DEGRADED 2 | |
83 | #define IMSM_T_STATE_FAILED 3 | |
84 | __u8 raid_level; | |
85 | #define IMSM_T_RAID0 0 | |
86 | #define IMSM_T_RAID1 1 | |
87 | #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */ | |
88 | __u8 num_members; /* number of member disks */ | |
89 | __u8 num_domains; /* number of parity domains */ | |
90 | __u8 failed_disk_num; /* valid only when state is degraded */ | |
91 | __u8 ddf; | |
92 | __u32 filler[7]; /* expansion area */ | |
93 | #define IMSM_ORD_REBUILD (1 << 24) | |
94 | __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members], | |
95 | * top byte contains some flags | |
96 | */ | |
97 | } __attribute__ ((packed)); | |
98 | ||
99 | struct imsm_vol { | |
100 | __u32 curr_migr_unit; | |
101 | __u32 checkpoint_id; /* id to access curr_migr_unit */ | |
102 | __u8 migr_state; /* Normal or Migrating */ | |
103 | #define MIGR_INIT 0 | |
104 | #define MIGR_REBUILD 1 | |
105 | #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */ | |
106 | #define MIGR_GEN_MIGR 3 | |
107 | #define MIGR_STATE_CHANGE 4 | |
108 | #define MIGR_REPAIR 5 | |
109 | __u8 migr_type; /* Initializing, Rebuilding, ... */ | |
110 | __u8 dirty; | |
111 | __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */ | |
112 | __u16 verify_errors; /* number of mismatches */ | |
113 | __u16 bad_blocks; /* number of bad blocks during verify */ | |
114 | __u32 filler[4]; | |
115 | struct imsm_map map[1]; | |
116 | /* here comes another one if migr_state */ | |
117 | } __attribute__ ((packed)); | |
118 | ||
119 | struct imsm_dev { | |
120 | __u8 volume[MAX_RAID_SERIAL_LEN]; | |
121 | __u32 size_low; | |
122 | __u32 size_high; | |
123 | #define DEV_BOOTABLE __cpu_to_le32(0x01) | |
124 | #define DEV_BOOT_DEVICE __cpu_to_le32(0x02) | |
125 | #define DEV_READ_COALESCING __cpu_to_le32(0x04) | |
126 | #define DEV_WRITE_COALESCING __cpu_to_le32(0x08) | |
127 | #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10) | |
128 | #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20) | |
129 | #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40) | |
130 | #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80) | |
131 | #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100) | |
132 | #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200) | |
133 | #define DEV_CLONE_N_GO __cpu_to_le32(0x400) | |
134 | #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800) | |
135 | #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000) | |
136 | __u32 status; /* Persistent RaidDev status */ | |
137 | __u32 reserved_blocks; /* Reserved blocks at beginning of volume */ | |
138 | __u8 migr_priority; | |
139 | __u8 num_sub_vols; | |
140 | __u8 tid; | |
141 | __u8 cng_master_disk; | |
142 | __u16 cache_policy; | |
143 | __u8 cng_state; | |
144 | __u8 cng_sub_state; | |
145 | #define IMSM_DEV_FILLERS 10 | |
146 | __u32 filler[IMSM_DEV_FILLERS]; | |
147 | struct imsm_vol vol; | |
148 | } __attribute__ ((packed)); | |
149 | ||
150 | struct imsm_super { | |
151 | __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */ | |
152 | __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */ | |
153 | __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */ | |
154 | __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */ | |
155 | __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */ | |
156 | __u32 error_log_size; /* 0x30 - 0x33 in bytes */ | |
157 | __u32 attributes; /* 0x34 - 0x37 */ | |
158 | __u8 num_disks; /* 0x38 Number of configured disks */ | |
159 | __u8 num_raid_devs; /* 0x39 Number of configured volumes */ | |
160 | __u8 error_log_pos; /* 0x3A */ | |
161 | __u8 fill[1]; /* 0x3B */ | |
162 | __u32 cache_size; /* 0x3c - 0x40 in mb */ | |
163 | __u32 orig_family_num; /* 0x40 - 0x43 original family num */ | |
164 | __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */ | |
165 | __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */ | |
166 | #define IMSM_FILLERS 35 | |
167 | __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */ | |
168 | struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */ | |
169 | /* here comes imsm_dev[num_raid_devs] */ | |
170 | /* here comes BBM logs */ | |
171 | } __attribute__ ((packed)); | |
172 | ||
173 | #define BBM_LOG_MAX_ENTRIES 254 | |
174 | ||
175 | struct bbm_log_entry { | |
176 | __u64 defective_block_start; | |
177 | #define UNREADABLE 0xFFFFFFFF | |
178 | __u32 spare_block_offset; | |
179 | __u16 remapped_marked_count; | |
180 | __u16 disk_ordinal; | |
181 | } __attribute__ ((__packed__)); | |
182 | ||
183 | struct bbm_log { | |
184 | __u32 signature; /* 0xABADB10C */ | |
185 | __u32 entry_count; | |
186 | __u32 reserved_spare_block_count; /* 0 */ | |
187 | __u32 reserved; /* 0xFFFF */ | |
188 | __u64 first_spare_lba; | |
189 | struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES]; | |
190 | } __attribute__ ((__packed__)); | |
191 | ||
192 | ||
193 | #ifndef MDASSEMBLE | |
194 | static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" }; | |
195 | #endif | |
196 | ||
197 | static __u8 migr_type(struct imsm_dev *dev) | |
198 | { | |
199 | if (dev->vol.migr_type == MIGR_VERIFY && | |
200 | dev->status & DEV_VERIFY_AND_FIX) | |
201 | return MIGR_REPAIR; | |
202 | else | |
203 | return dev->vol.migr_type; | |
204 | } | |
205 | ||
206 | static void set_migr_type(struct imsm_dev *dev, __u8 migr_type) | |
207 | { | |
208 | /* for compatibility with older oroms convert MIGR_REPAIR, into | |
209 | * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status | |
210 | */ | |
211 | if (migr_type == MIGR_REPAIR) { | |
212 | dev->vol.migr_type = MIGR_VERIFY; | |
213 | dev->status |= DEV_VERIFY_AND_FIX; | |
214 | } else { | |
215 | dev->vol.migr_type = migr_type; | |
216 | dev->status &= ~DEV_VERIFY_AND_FIX; | |
217 | } | |
218 | } | |
219 | ||
220 | static unsigned int sector_count(__u32 bytes) | |
221 | { | |
222 | return ((bytes + (512-1)) & (~(512-1))) / 512; | |
223 | } | |
224 | ||
225 | static unsigned int mpb_sectors(struct imsm_super *mpb) | |
226 | { | |
227 | return sector_count(__le32_to_cpu(mpb->mpb_size)); | |
228 | } | |
229 | ||
230 | struct intel_dev { | |
231 | struct imsm_dev *dev; | |
232 | struct intel_dev *next; | |
233 | int index; | |
234 | }; | |
235 | ||
236 | /* internal representation of IMSM metadata */ | |
237 | struct intel_super { | |
238 | union { | |
239 | void *buf; /* O_DIRECT buffer for reading/writing metadata */ | |
240 | struct imsm_super *anchor; /* immovable parameters */ | |
241 | }; | |
242 | size_t len; /* size of the 'buf' allocation */ | |
243 | void *next_buf; /* for realloc'ing buf from the manager */ | |
244 | size_t next_len; | |
245 | int updates_pending; /* count of pending updates for mdmon */ | |
246 | int creating_imsm; /* flag to indicate container creation */ | |
247 | int current_vol; /* index of raid device undergoing creation */ | |
248 | __u32 create_offset; /* common start for 'current_vol' */ | |
249 | __u32 random; /* random data for seeding new family numbers */ | |
250 | struct intel_dev *devlist; | |
251 | struct dl { | |
252 | struct dl *next; | |
253 | int index; | |
254 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
255 | int major, minor; | |
256 | char *devname; | |
257 | struct imsm_disk disk; | |
258 | int fd; | |
259 | int extent_cnt; | |
260 | struct extent *e; /* for determining freespace @ create */ | |
261 | int raiddisk; /* slot to fill in autolayout */ | |
262 | } *disks; | |
263 | struct dl *add; /* list of disks to add while mdmon active */ | |
264 | struct dl *missing; /* disks removed while we weren't looking */ | |
265 | struct bbm_log *bbm_log; | |
266 | const char *hba; /* device path of the raid controller for this metadata */ | |
267 | const struct imsm_orom *orom; /* platform firmware support */ | |
268 | struct intel_super *next; /* (temp) list for disambiguating family_num */ | |
269 | }; | |
270 | ||
271 | struct intel_disk { | |
272 | struct imsm_disk disk; | |
273 | #define IMSM_UNKNOWN_OWNER (-1) | |
274 | int owner; | |
275 | struct intel_disk *next; | |
276 | }; | |
277 | ||
278 | struct extent { | |
279 | unsigned long long start, size; | |
280 | }; | |
281 | ||
282 | /* definition of messages passed to imsm_process_update */ | |
283 | enum imsm_update_type { | |
284 | update_activate_spare, | |
285 | update_create_array, | |
286 | update_add_disk, | |
287 | }; | |
288 | ||
289 | struct imsm_update_activate_spare { | |
290 | enum imsm_update_type type; | |
291 | struct dl *dl; | |
292 | int slot; | |
293 | int array; | |
294 | struct imsm_update_activate_spare *next; | |
295 | }; | |
296 | ||
297 | struct disk_info { | |
298 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
299 | }; | |
300 | ||
301 | struct imsm_update_create_array { | |
302 | enum imsm_update_type type; | |
303 | int dev_idx; | |
304 | struct imsm_dev dev; | |
305 | }; | |
306 | ||
307 | struct imsm_update_add_disk { | |
308 | enum imsm_update_type type; | |
309 | }; | |
310 | ||
311 | static struct supertype *match_metadata_desc_imsm(char *arg) | |
312 | { | |
313 | struct supertype *st; | |
314 | ||
315 | if (strcmp(arg, "imsm") != 0 && | |
316 | strcmp(arg, "default") != 0 | |
317 | ) | |
318 | return NULL; | |
319 | ||
320 | st = malloc(sizeof(*st)); | |
321 | memset(st, 0, sizeof(*st)); | |
322 | st->ss = &super_imsm; | |
323 | st->max_devs = IMSM_MAX_DEVICES; | |
324 | st->minor_version = 0; | |
325 | st->sb = NULL; | |
326 | return st; | |
327 | } | |
328 | ||
329 | #ifndef MDASSEMBLE | |
330 | static __u8 *get_imsm_version(struct imsm_super *mpb) | |
331 | { | |
332 | return &mpb->sig[MPB_SIG_LEN]; | |
333 | } | |
334 | #endif | |
335 | ||
336 | /* retrieve a disk directly from the anchor when the anchor is known to be | |
337 | * up-to-date, currently only at load time | |
338 | */ | |
339 | static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index) | |
340 | { | |
341 | if (index >= mpb->num_disks) | |
342 | return NULL; | |
343 | return &mpb->disk[index]; | |
344 | } | |
345 | ||
346 | #ifndef MDASSEMBLE | |
347 | /* retrieve a disk from the parsed metadata */ | |
348 | static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index) | |
349 | { | |
350 | struct dl *d; | |
351 | ||
352 | for (d = super->disks; d; d = d->next) | |
353 | if (d->index == index) | |
354 | return &d->disk; | |
355 | ||
356 | return NULL; | |
357 | } | |
358 | #endif | |
359 | ||
360 | /* generate a checksum directly from the anchor when the anchor is known to be | |
361 | * up-to-date, currently only at load or write_super after coalescing | |
362 | */ | |
363 | static __u32 __gen_imsm_checksum(struct imsm_super *mpb) | |
364 | { | |
365 | __u32 end = mpb->mpb_size / sizeof(end); | |
366 | __u32 *p = (__u32 *) mpb; | |
367 | __u32 sum = 0; | |
368 | ||
369 | while (end--) { | |
370 | sum += __le32_to_cpu(*p); | |
371 | p++; | |
372 | } | |
373 | ||
374 | return sum - __le32_to_cpu(mpb->check_sum); | |
375 | } | |
376 | ||
377 | static size_t sizeof_imsm_map(struct imsm_map *map) | |
378 | { | |
379 | return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1); | |
380 | } | |
381 | ||
382 | struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map) | |
383 | { | |
384 | struct imsm_map *map = &dev->vol.map[0]; | |
385 | ||
386 | if (second_map && !dev->vol.migr_state) | |
387 | return NULL; | |
388 | else if (second_map) { | |
389 | void *ptr = map; | |
390 | ||
391 | return ptr + sizeof_imsm_map(map); | |
392 | } else | |
393 | return map; | |
394 | ||
395 | } | |
396 | ||
397 | /* return the size of the device. | |
398 | * migr_state increases the returned size if map[0] were to be duplicated | |
399 | */ | |
400 | static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state) | |
401 | { | |
402 | size_t size = sizeof(*dev) - sizeof(struct imsm_map) + | |
403 | sizeof_imsm_map(get_imsm_map(dev, 0)); | |
404 | ||
405 | /* migrating means an additional map */ | |
406 | if (dev->vol.migr_state) | |
407 | size += sizeof_imsm_map(get_imsm_map(dev, 1)); | |
408 | else if (migr_state) | |
409 | size += sizeof_imsm_map(get_imsm_map(dev, 0)); | |
410 | ||
411 | return size; | |
412 | } | |
413 | ||
414 | #ifndef MDASSEMBLE | |
415 | /* retrieve disk serial number list from a metadata update */ | |
416 | static struct disk_info *get_disk_info(struct imsm_update_create_array *update) | |
417 | { | |
418 | void *u = update; | |
419 | struct disk_info *inf; | |
420 | ||
421 | inf = u + sizeof(*update) - sizeof(struct imsm_dev) + | |
422 | sizeof_imsm_dev(&update->dev, 0); | |
423 | ||
424 | return inf; | |
425 | } | |
426 | #endif | |
427 | ||
428 | static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index) | |
429 | { | |
430 | int offset; | |
431 | int i; | |
432 | void *_mpb = mpb; | |
433 | ||
434 | if (index >= mpb->num_raid_devs) | |
435 | return NULL; | |
436 | ||
437 | /* devices start after all disks */ | |
438 | offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb; | |
439 | ||
440 | for (i = 0; i <= index; i++) | |
441 | if (i == index) | |
442 | return _mpb + offset; | |
443 | else | |
444 | offset += sizeof_imsm_dev(_mpb + offset, 0); | |
445 | ||
446 | return NULL; | |
447 | } | |
448 | ||
449 | static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index) | |
450 | { | |
451 | struct intel_dev *dv; | |
452 | ||
453 | if (index >= super->anchor->num_raid_devs) | |
454 | return NULL; | |
455 | for (dv = super->devlist; dv; dv = dv->next) | |
456 | if (dv->index == index) | |
457 | return dv->dev; | |
458 | return NULL; | |
459 | } | |
460 | ||
461 | static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, int slot) | |
462 | { | |
463 | struct imsm_map *map; | |
464 | ||
465 | if (dev->vol.migr_state) | |
466 | map = get_imsm_map(dev, 1); | |
467 | else | |
468 | map = get_imsm_map(dev, 0); | |
469 | ||
470 | /* top byte identifies disk under rebuild */ | |
471 | return __le32_to_cpu(map->disk_ord_tbl[slot]); | |
472 | } | |
473 | ||
474 | #define ord_to_idx(ord) (((ord) << 8) >> 8) | |
475 | static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot) | |
476 | { | |
477 | __u32 ord = get_imsm_ord_tbl_ent(dev, slot); | |
478 | ||
479 | return ord_to_idx(ord); | |
480 | } | |
481 | ||
482 | static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord) | |
483 | { | |
484 | map->disk_ord_tbl[slot] = __cpu_to_le32(ord); | |
485 | } | |
486 | ||
487 | static int get_imsm_disk_slot(struct imsm_map *map, int idx) | |
488 | { | |
489 | int slot; | |
490 | __u32 ord; | |
491 | ||
492 | for (slot = 0; slot < map->num_members; slot++) { | |
493 | ord = __le32_to_cpu(map->disk_ord_tbl[slot]); | |
494 | if (ord_to_idx(ord) == idx) | |
495 | return slot; | |
496 | } | |
497 | ||
498 | return -1; | |
499 | } | |
500 | ||
501 | static int get_imsm_raid_level(struct imsm_map *map) | |
502 | { | |
503 | if (map->raid_level == 1) { | |
504 | if (map->num_members == 2) | |
505 | return 1; | |
506 | else | |
507 | return 10; | |
508 | } | |
509 | ||
510 | return map->raid_level; | |
511 | } | |
512 | ||
513 | static int cmp_extent(const void *av, const void *bv) | |
514 | { | |
515 | const struct extent *a = av; | |
516 | const struct extent *b = bv; | |
517 | if (a->start < b->start) | |
518 | return -1; | |
519 | if (a->start > b->start) | |
520 | return 1; | |
521 | return 0; | |
522 | } | |
523 | ||
524 | static int count_memberships(struct dl *dl, struct intel_super *super) | |
525 | { | |
526 | int memberships = 0; | |
527 | int i; | |
528 | ||
529 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
530 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
531 | struct imsm_map *map = get_imsm_map(dev, 0); | |
532 | ||
533 | if (get_imsm_disk_slot(map, dl->index) >= 0) | |
534 | memberships++; | |
535 | } | |
536 | ||
537 | return memberships; | |
538 | } | |
539 | ||
540 | static struct extent *get_extents(struct intel_super *super, struct dl *dl) | |
541 | { | |
542 | /* find a list of used extents on the given physical device */ | |
543 | struct extent *rv, *e; | |
544 | int i; | |
545 | int memberships = count_memberships(dl, super); | |
546 | __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
547 | ||
548 | rv = malloc(sizeof(struct extent) * (memberships + 1)); | |
549 | if (!rv) | |
550 | return NULL; | |
551 | e = rv; | |
552 | ||
553 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
554 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
555 | struct imsm_map *map = get_imsm_map(dev, 0); | |
556 | ||
557 | if (get_imsm_disk_slot(map, dl->index) >= 0) { | |
558 | e->start = __le32_to_cpu(map->pba_of_lba0); | |
559 | e->size = __le32_to_cpu(map->blocks_per_member); | |
560 | e++; | |
561 | } | |
562 | } | |
563 | qsort(rv, memberships, sizeof(*rv), cmp_extent); | |
564 | ||
565 | /* determine the start of the metadata | |
566 | * when no raid devices are defined use the default | |
567 | * ...otherwise allow the metadata to truncate the value | |
568 | * as is the case with older versions of imsm | |
569 | */ | |
570 | if (memberships) { | |
571 | struct extent *last = &rv[memberships - 1]; | |
572 | __u32 remainder; | |
573 | ||
574 | remainder = __le32_to_cpu(dl->disk.total_blocks) - | |
575 | (last->start + last->size); | |
576 | /* round down to 1k block to satisfy precision of the kernel | |
577 | * 'size' interface | |
578 | */ | |
579 | remainder &= ~1UL; | |
580 | /* make sure remainder is still sane */ | |
581 | if (remainder < ROUND_UP(super->len, 512) >> 9) | |
582 | remainder = ROUND_UP(super->len, 512) >> 9; | |
583 | if (reservation > remainder) | |
584 | reservation = remainder; | |
585 | } | |
586 | e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation; | |
587 | e->size = 0; | |
588 | return rv; | |
589 | } | |
590 | ||
591 | /* try to determine how much space is reserved for metadata from | |
592 | * the last get_extents() entry, otherwise fallback to the | |
593 | * default | |
594 | */ | |
595 | static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl) | |
596 | { | |
597 | struct extent *e; | |
598 | int i; | |
599 | __u32 rv; | |
600 | ||
601 | /* for spares just return a minimal reservation which will grow | |
602 | * once the spare is picked up by an array | |
603 | */ | |
604 | if (dl->index == -1) | |
605 | return MPB_SECTOR_CNT; | |
606 | ||
607 | e = get_extents(super, dl); | |
608 | if (!e) | |
609 | return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
610 | ||
611 | /* scroll to last entry */ | |
612 | for (i = 0; e[i].size; i++) | |
613 | continue; | |
614 | ||
615 | rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start; | |
616 | ||
617 | free(e); | |
618 | ||
619 | return rv; | |
620 | } | |
621 | ||
622 | static int is_spare(struct imsm_disk *disk) | |
623 | { | |
624 | return (disk->status & SPARE_DISK) == SPARE_DISK; | |
625 | } | |
626 | ||
627 | static int is_configured(struct imsm_disk *disk) | |
628 | { | |
629 | return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK; | |
630 | } | |
631 | ||
632 | static int is_failed(struct imsm_disk *disk) | |
633 | { | |
634 | return (disk->status & FAILED_DISK) == FAILED_DISK; | |
635 | } | |
636 | ||
637 | #ifndef MDASSEMBLE | |
638 | static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx) | |
639 | { | |
640 | __u64 sz; | |
641 | int slot; | |
642 | struct imsm_map *map = get_imsm_map(dev, 0); | |
643 | __u32 ord; | |
644 | ||
645 | printf("\n"); | |
646 | printf("[%.16s]:\n", dev->volume); | |
647 | printf(" UUID : %s\n", uuid); | |
648 | printf(" RAID Level : %d\n", get_imsm_raid_level(map)); | |
649 | printf(" Members : %d\n", map->num_members); | |
650 | slot = get_imsm_disk_slot(map, disk_idx); | |
651 | if (slot >= 0) { | |
652 | ord = get_imsm_ord_tbl_ent(dev, slot); | |
653 | printf(" This Slot : %d%s\n", slot, | |
654 | ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : ""); | |
655 | } else | |
656 | printf(" This Slot : ?\n"); | |
657 | sz = __le32_to_cpu(dev->size_high); | |
658 | sz <<= 32; | |
659 | sz += __le32_to_cpu(dev->size_low); | |
660 | printf(" Array Size : %llu%s\n", (unsigned long long)sz, | |
661 | human_size(sz * 512)); | |
662 | sz = __le32_to_cpu(map->blocks_per_member); | |
663 | printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz, | |
664 | human_size(sz * 512)); | |
665 | printf(" Sector Offset : %u\n", | |
666 | __le32_to_cpu(map->pba_of_lba0)); | |
667 | printf(" Num Stripes : %u\n", | |
668 | __le32_to_cpu(map->num_data_stripes)); | |
669 | printf(" Chunk Size : %u KiB\n", | |
670 | __le16_to_cpu(map->blocks_per_strip) / 2); | |
671 | printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks)); | |
672 | printf(" Migrate State : "); | |
673 | if (dev->vol.migr_state) { | |
674 | if (migr_type(dev) == MIGR_INIT) | |
675 | printf("initialize\n"); | |
676 | else if (migr_type(dev) == MIGR_REBUILD) | |
677 | printf("rebuild\n"); | |
678 | else if (migr_type(dev) == MIGR_VERIFY) | |
679 | printf("check\n"); | |
680 | else if (migr_type(dev) == MIGR_GEN_MIGR) | |
681 | printf("general migration\n"); | |
682 | else if (migr_type(dev) == MIGR_STATE_CHANGE) | |
683 | printf("state change\n"); | |
684 | else if (migr_type(dev) == MIGR_REPAIR) | |
685 | printf("repair\n"); | |
686 | else | |
687 | printf("<unknown:%d>\n", migr_type(dev)); | |
688 | } else | |
689 | printf("idle\n"); | |
690 | printf(" Map State : %s", map_state_str[map->map_state]); | |
691 | if (dev->vol.migr_state) { | |
692 | struct imsm_map *map = get_imsm_map(dev, 1); | |
693 | printf(" <-- %s", map_state_str[map->map_state]); | |
694 | } | |
695 | printf("\n"); | |
696 | printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean"); | |
697 | } | |
698 | ||
699 | static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved) | |
700 | { | |
701 | struct imsm_disk *disk = __get_imsm_disk(mpb, index); | |
702 | char str[MAX_RAID_SERIAL_LEN + 1]; | |
703 | __u64 sz; | |
704 | ||
705 | if (index < 0) | |
706 | return; | |
707 | ||
708 | printf("\n"); | |
709 | snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial); | |
710 | printf(" Disk%02d Serial : %s\n", index, str); | |
711 | printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "", | |
712 | is_configured(disk) ? " active" : "", | |
713 | is_failed(disk) ? " failed" : ""); | |
714 | printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id)); | |
715 | sz = __le32_to_cpu(disk->total_blocks) - reserved; | |
716 | printf(" Usable Size : %llu%s\n", (unsigned long long)sz, | |
717 | human_size(sz * 512)); | |
718 | } | |
719 | ||
720 | static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info); | |
721 | ||
722 | static void examine_super_imsm(struct supertype *st, char *homehost) | |
723 | { | |
724 | struct intel_super *super = st->sb; | |
725 | struct imsm_super *mpb = super->anchor; | |
726 | char str[MAX_SIGNATURE_LENGTH]; | |
727 | int i; | |
728 | struct mdinfo info; | |
729 | char nbuf[64]; | |
730 | __u32 sum; | |
731 | __u32 reserved = imsm_reserved_sectors(super, super->disks); | |
732 | ||
733 | ||
734 | snprintf(str, MPB_SIG_LEN, "%s", mpb->sig); | |
735 | printf(" Magic : %s\n", str); | |
736 | snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb)); | |
737 | printf(" Version : %s\n", get_imsm_version(mpb)); | |
738 | printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num)); | |
739 | printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num)); | |
740 | printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num)); | |
741 | getinfo_super_imsm(st, &info); | |
742 | fname_from_uuid(st, &info, nbuf, ':'); | |
743 | printf(" UUID : %s\n", nbuf + 5); | |
744 | sum = __le32_to_cpu(mpb->check_sum); | |
745 | printf(" Checksum : %08x %s\n", sum, | |
746 | __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect"); | |
747 | printf(" MPB Sectors : %d\n", mpb_sectors(mpb)); | |
748 | printf(" Disks : %d\n", mpb->num_disks); | |
749 | printf(" RAID Devices : %d\n", mpb->num_raid_devs); | |
750 | print_imsm_disk(mpb, super->disks->index, reserved); | |
751 | if (super->bbm_log) { | |
752 | struct bbm_log *log = super->bbm_log; | |
753 | ||
754 | printf("\n"); | |
755 | printf("Bad Block Management Log:\n"); | |
756 | printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size)); | |
757 | printf(" Signature : %x\n", __le32_to_cpu(log->signature)); | |
758 | printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count)); | |
759 | printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count)); | |
760 | printf(" First Spare : %llx\n", | |
761 | (unsigned long long) __le64_to_cpu(log->first_spare_lba)); | |
762 | } | |
763 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
764 | struct mdinfo info; | |
765 | struct imsm_dev *dev = __get_imsm_dev(mpb, i); | |
766 | ||
767 | super->current_vol = i; | |
768 | getinfo_super_imsm(st, &info); | |
769 | fname_from_uuid(st, &info, nbuf, ':'); | |
770 | print_imsm_dev(dev, nbuf + 5, super->disks->index); | |
771 | } | |
772 | for (i = 0; i < mpb->num_disks; i++) { | |
773 | if (i == super->disks->index) | |
774 | continue; | |
775 | print_imsm_disk(mpb, i, reserved); | |
776 | } | |
777 | } | |
778 | ||
779 | static void brief_examine_super_imsm(struct supertype *st, int verbose) | |
780 | { | |
781 | /* We just write a generic IMSM ARRAY entry */ | |
782 | struct mdinfo info; | |
783 | char nbuf[64]; | |
784 | struct intel_super *super = st->sb; | |
785 | ||
786 | if (!super->anchor->num_raid_devs) { | |
787 | printf("ARRAY metadata=imsm\n"); | |
788 | return; | |
789 | } | |
790 | ||
791 | getinfo_super_imsm(st, &info); | |
792 | fname_from_uuid(st, &info, nbuf, ':'); | |
793 | printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5); | |
794 | } | |
795 | ||
796 | static void brief_examine_subarrays_imsm(struct supertype *st, int verbose) | |
797 | { | |
798 | /* We just write a generic IMSM ARRAY entry */ | |
799 | struct mdinfo info; | |
800 | char nbuf[64]; | |
801 | char nbuf1[64]; | |
802 | struct intel_super *super = st->sb; | |
803 | int i; | |
804 | ||
805 | if (!super->anchor->num_raid_devs) | |
806 | return; | |
807 | ||
808 | getinfo_super_imsm(st, &info); | |
809 | fname_from_uuid(st, &info, nbuf, ':'); | |
810 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
811 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
812 | ||
813 | super->current_vol = i; | |
814 | getinfo_super_imsm(st, &info); | |
815 | fname_from_uuid(st, &info, nbuf1, ':'); | |
816 | printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n", | |
817 | dev->volume, nbuf + 5, i, nbuf1 + 5); | |
818 | } | |
819 | } | |
820 | ||
821 | static void export_examine_super_imsm(struct supertype *st) | |
822 | { | |
823 | struct intel_super *super = st->sb; | |
824 | struct imsm_super *mpb = super->anchor; | |
825 | struct mdinfo info; | |
826 | char nbuf[64]; | |
827 | ||
828 | getinfo_super_imsm(st, &info); | |
829 | fname_from_uuid(st, &info, nbuf, ':'); | |
830 | printf("MD_METADATA=imsm\n"); | |
831 | printf("MD_LEVEL=container\n"); | |
832 | printf("MD_UUID=%s\n", nbuf+5); | |
833 | printf("MD_DEVICES=%u\n", mpb->num_disks); | |
834 | } | |
835 | ||
836 | static void detail_super_imsm(struct supertype *st, char *homehost) | |
837 | { | |
838 | struct mdinfo info; | |
839 | char nbuf[64]; | |
840 | ||
841 | getinfo_super_imsm(st, &info); | |
842 | fname_from_uuid(st, &info, nbuf, ':'); | |
843 | printf("\n UUID : %s\n", nbuf + 5); | |
844 | } | |
845 | ||
846 | static void brief_detail_super_imsm(struct supertype *st) | |
847 | { | |
848 | struct mdinfo info; | |
849 | char nbuf[64]; | |
850 | getinfo_super_imsm(st, &info); | |
851 | fname_from_uuid(st, &info, nbuf, ':'); | |
852 | printf(" UUID=%s", nbuf + 5); | |
853 | } | |
854 | ||
855 | static int imsm_read_serial(int fd, char *devname, __u8 *serial); | |
856 | static void fd2devname(int fd, char *name); | |
857 | ||
858 | static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose) | |
859 | { | |
860 | /* dump an unsorted list of devices attached to ahci, as well as | |
861 | * non-connected ports | |
862 | */ | |
863 | int hba_len = strlen(hba_path) + 1; | |
864 | struct dirent *ent; | |
865 | DIR *dir; | |
866 | char *path = NULL; | |
867 | int err = 0; | |
868 | unsigned long port_mask = (1 << port_count) - 1; | |
869 | ||
870 | if (port_count > sizeof(port_mask) * 8) { | |
871 | if (verbose) | |
872 | fprintf(stderr, Name ": port_count %d out of range\n", port_count); | |
873 | return 2; | |
874 | } | |
875 | ||
876 | /* scroll through /sys/dev/block looking for devices attached to | |
877 | * this hba | |
878 | */ | |
879 | dir = opendir("/sys/dev/block"); | |
880 | for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) { | |
881 | int fd; | |
882 | char model[64]; | |
883 | char vendor[64]; | |
884 | char buf[1024]; | |
885 | int major, minor; | |
886 | char *device; | |
887 | char *c; | |
888 | int port; | |
889 | int type; | |
890 | ||
891 | if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2) | |
892 | continue; | |
893 | path = devt_to_devpath(makedev(major, minor)); | |
894 | if (!path) | |
895 | continue; | |
896 | if (!path_attached_to_hba(path, hba_path)) { | |
897 | free(path); | |
898 | path = NULL; | |
899 | continue; | |
900 | } | |
901 | ||
902 | /* retrieve the scsi device type */ | |
903 | if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) { | |
904 | if (verbose) | |
905 | fprintf(stderr, Name ": failed to allocate 'device'\n"); | |
906 | err = 2; | |
907 | break; | |
908 | } | |
909 | sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor); | |
910 | if (load_sys(device, buf) != 0) { | |
911 | if (verbose) | |
912 | fprintf(stderr, Name ": failed to read device type for %s\n", | |
913 | path); | |
914 | err = 2; | |
915 | free(device); | |
916 | break; | |
917 | } | |
918 | type = strtoul(buf, NULL, 10); | |
919 | ||
920 | /* if it's not a disk print the vendor and model */ | |
921 | if (!(type == 0 || type == 7 || type == 14)) { | |
922 | vendor[0] = '\0'; | |
923 | model[0] = '\0'; | |
924 | sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor); | |
925 | if (load_sys(device, buf) == 0) { | |
926 | strncpy(vendor, buf, sizeof(vendor)); | |
927 | vendor[sizeof(vendor) - 1] = '\0'; | |
928 | c = (char *) &vendor[sizeof(vendor) - 1]; | |
929 | while (isspace(*c) || *c == '\0') | |
930 | *c-- = '\0'; | |
931 | ||
932 | } | |
933 | sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor); | |
934 | if (load_sys(device, buf) == 0) { | |
935 | strncpy(model, buf, sizeof(model)); | |
936 | model[sizeof(model) - 1] = '\0'; | |
937 | c = (char *) &model[sizeof(model) - 1]; | |
938 | while (isspace(*c) || *c == '\0') | |
939 | *c-- = '\0'; | |
940 | } | |
941 | ||
942 | if (vendor[0] && model[0]) | |
943 | sprintf(buf, "%.64s %.64s", vendor, model); | |
944 | else | |
945 | switch (type) { /* numbers from hald/linux/device.c */ | |
946 | case 1: sprintf(buf, "tape"); break; | |
947 | case 2: sprintf(buf, "printer"); break; | |
948 | case 3: sprintf(buf, "processor"); break; | |
949 | case 4: | |
950 | case 5: sprintf(buf, "cdrom"); break; | |
951 | case 6: sprintf(buf, "scanner"); break; | |
952 | case 8: sprintf(buf, "media_changer"); break; | |
953 | case 9: sprintf(buf, "comm"); break; | |
954 | case 12: sprintf(buf, "raid"); break; | |
955 | default: sprintf(buf, "unknown"); | |
956 | } | |
957 | } else | |
958 | buf[0] = '\0'; | |
959 | free(device); | |
960 | ||
961 | /* chop device path to 'host%d' and calculate the port number */ | |
962 | c = strchr(&path[hba_len], '/'); | |
963 | *c = '\0'; | |
964 | if (sscanf(&path[hba_len], "host%d", &port) == 1) | |
965 | port -= host_base; | |
966 | else { | |
967 | if (verbose) { | |
968 | *c = '/'; /* repair the full string */ | |
969 | fprintf(stderr, Name ": failed to determine port number for %s\n", | |
970 | path); | |
971 | } | |
972 | err = 2; | |
973 | break; | |
974 | } | |
975 | ||
976 | /* mark this port as used */ | |
977 | port_mask &= ~(1 << port); | |
978 | ||
979 | /* print out the device information */ | |
980 | if (buf[0]) { | |
981 | printf(" Port%d : - non-disk device (%s) -\n", port, buf); | |
982 | continue; | |
983 | } | |
984 | ||
985 | fd = dev_open(ent->d_name, O_RDONLY); | |
986 | if (fd < 0) | |
987 | printf(" Port%d : - disk info unavailable -\n", port); | |
988 | else { | |
989 | fd2devname(fd, buf); | |
990 | printf(" Port%d : %s", port, buf); | |
991 | if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0) | |
992 | printf(" (%s)\n", buf); | |
993 | else | |
994 | printf("()\n"); | |
995 | } | |
996 | close(fd); | |
997 | free(path); | |
998 | path = NULL; | |
999 | } | |
1000 | if (path) | |
1001 | free(path); | |
1002 | if (dir) | |
1003 | closedir(dir); | |
1004 | if (err == 0) { | |
1005 | int i; | |
1006 | ||
1007 | for (i = 0; i < port_count; i++) | |
1008 | if (port_mask & (1 << i)) | |
1009 | printf(" Port%d : - no device attached -\n", i); | |
1010 | } | |
1011 | ||
1012 | return err; | |
1013 | } | |
1014 | ||
1015 | static int detail_platform_imsm(int verbose, int enumerate_only) | |
1016 | { | |
1017 | /* There are two components to imsm platform support, the ahci SATA | |
1018 | * controller and the option-rom. To find the SATA controller we | |
1019 | * simply look in /sys/bus/pci/drivers/ahci to see if an ahci | |
1020 | * controller with the Intel vendor id is present. This approach | |
1021 | * allows mdadm to leverage the kernel's ahci detection logic, with the | |
1022 | * caveat that if ahci.ko is not loaded mdadm will not be able to | |
1023 | * detect platform raid capabilities. The option-rom resides in a | |
1024 | * platform "Adapter ROM". We scan for its signature to retrieve the | |
1025 | * platform capabilities. If raid support is disabled in the BIOS the | |
1026 | * option-rom capability structure will not be available. | |
1027 | */ | |
1028 | const struct imsm_orom *orom; | |
1029 | struct sys_dev *list, *hba; | |
1030 | DIR *dir; | |
1031 | struct dirent *ent; | |
1032 | const char *hba_path; | |
1033 | int host_base = 0; | |
1034 | int port_count = 0; | |
1035 | ||
1036 | if (enumerate_only) { | |
1037 | if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom()) | |
1038 | return 0; | |
1039 | return 2; | |
1040 | } | |
1041 | ||
1042 | list = find_driver_devices("pci", "ahci"); | |
1043 | for (hba = list; hba; hba = hba->next) | |
1044 | if (devpath_to_vendor(hba->path) == 0x8086) | |
1045 | break; | |
1046 | ||
1047 | if (!hba) { | |
1048 | if (verbose) | |
1049 | fprintf(stderr, Name ": unable to find active ahci controller\n"); | |
1050 | free_sys_dev(&list); | |
1051 | return 2; | |
1052 | } else if (verbose) | |
1053 | fprintf(stderr, Name ": found Intel SATA AHCI Controller\n"); | |
1054 | hba_path = hba->path; | |
1055 | hba->path = NULL; | |
1056 | free_sys_dev(&list); | |
1057 | ||
1058 | orom = find_imsm_orom(); | |
1059 | if (!orom) { | |
1060 | if (verbose) | |
1061 | fprintf(stderr, Name ": imsm option-rom not found\n"); | |
1062 | return 2; | |
1063 | } | |
1064 | ||
1065 | printf(" Platform : Intel(R) Matrix Storage Manager\n"); | |
1066 | printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver, | |
1067 | orom->hotfix_ver, orom->build); | |
1068 | printf(" RAID Levels :%s%s%s%s%s\n", | |
1069 | imsm_orom_has_raid0(orom) ? " raid0" : "", | |
1070 | imsm_orom_has_raid1(orom) ? " raid1" : "", | |
1071 | imsm_orom_has_raid1e(orom) ? " raid1e" : "", | |
1072 | imsm_orom_has_raid10(orom) ? " raid10" : "", | |
1073 | imsm_orom_has_raid5(orom) ? " raid5" : ""); | |
1074 | printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n", | |
1075 | imsm_orom_has_chunk(orom, 2) ? " 2k" : "", | |
1076 | imsm_orom_has_chunk(orom, 4) ? " 4k" : "", | |
1077 | imsm_orom_has_chunk(orom, 8) ? " 8k" : "", | |
1078 | imsm_orom_has_chunk(orom, 16) ? " 16k" : "", | |
1079 | imsm_orom_has_chunk(orom, 32) ? " 32k" : "", | |
1080 | imsm_orom_has_chunk(orom, 64) ? " 64k" : "", | |
1081 | imsm_orom_has_chunk(orom, 128) ? " 128k" : "", | |
1082 | imsm_orom_has_chunk(orom, 256) ? " 256k" : "", | |
1083 | imsm_orom_has_chunk(orom, 512) ? " 512k" : "", | |
1084 | imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "", | |
1085 | imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "", | |
1086 | imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "", | |
1087 | imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "", | |
1088 | imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "", | |
1089 | imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "", | |
1090 | imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : ""); | |
1091 | printf(" Max Disks : %d\n", orom->tds); | |
1092 | printf(" Max Volumes : %d\n", orom->vpa); | |
1093 | printf(" I/O Controller : %s\n", hba_path); | |
1094 | ||
1095 | /* find the smallest scsi host number to determine a port number base */ | |
1096 | dir = opendir(hba_path); | |
1097 | for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) { | |
1098 | int host; | |
1099 | ||
1100 | if (sscanf(ent->d_name, "host%d", &host) != 1) | |
1101 | continue; | |
1102 | if (port_count == 0) | |
1103 | host_base = host; | |
1104 | else if (host < host_base) | |
1105 | host_base = host; | |
1106 | ||
1107 | if (host + 1 > port_count + host_base) | |
1108 | port_count = host + 1 - host_base; | |
1109 | ||
1110 | } | |
1111 | if (dir) | |
1112 | closedir(dir); | |
1113 | ||
1114 | if (!port_count || imsm_enumerate_ports(hba_path, port_count, | |
1115 | host_base, verbose) != 0) { | |
1116 | if (verbose) | |
1117 | fprintf(stderr, Name ": failed to enumerate ports\n"); | |
1118 | return 2; | |
1119 | } | |
1120 | ||
1121 | return 0; | |
1122 | } | |
1123 | #endif | |
1124 | ||
1125 | static int match_home_imsm(struct supertype *st, char *homehost) | |
1126 | { | |
1127 | /* the imsm metadata format does not specify any host | |
1128 | * identification information. We return -1 since we can never | |
1129 | * confirm nor deny whether a given array is "meant" for this | |
1130 | * host. We rely on compare_super and the 'family_num' fields to | |
1131 | * exclude member disks that do not belong, and we rely on | |
1132 | * mdadm.conf to specify the arrays that should be assembled. | |
1133 | * Auto-assembly may still pick up "foreign" arrays. | |
1134 | */ | |
1135 | ||
1136 | return -1; | |
1137 | } | |
1138 | ||
1139 | static void uuid_from_super_imsm(struct supertype *st, int uuid[4]) | |
1140 | { | |
1141 | /* The uuid returned here is used for: | |
1142 | * uuid to put into bitmap file (Create, Grow) | |
1143 | * uuid for backup header when saving critical section (Grow) | |
1144 | * comparing uuids when re-adding a device into an array | |
1145 | * In these cases the uuid required is that of the data-array, | |
1146 | * not the device-set. | |
1147 | * uuid to recognise same set when adding a missing device back | |
1148 | * to an array. This is a uuid for the device-set. | |
1149 | * | |
1150 | * For each of these we can make do with a truncated | |
1151 | * or hashed uuid rather than the original, as long as | |
1152 | * everyone agrees. | |
1153 | * In each case the uuid required is that of the data-array, | |
1154 | * not the device-set. | |
1155 | */ | |
1156 | /* imsm does not track uuid's so we synthesis one using sha1 on | |
1157 | * - The signature (Which is constant for all imsm array, but no matter) | |
1158 | * - the orig_family_num of the container | |
1159 | * - the index number of the volume | |
1160 | * - the 'serial' number of the volume. | |
1161 | * Hopefully these are all constant. | |
1162 | */ | |
1163 | struct intel_super *super = st->sb; | |
1164 | ||
1165 | char buf[20]; | |
1166 | struct sha1_ctx ctx; | |
1167 | struct imsm_dev *dev = NULL; | |
1168 | __u32 family_num; | |
1169 | ||
1170 | /* some mdadm versions failed to set ->orig_family_num, in which | |
1171 | * case fall back to ->family_num. orig_family_num will be | |
1172 | * fixed up with the first metadata update. | |
1173 | */ | |
1174 | family_num = super->anchor->orig_family_num; | |
1175 | if (family_num == 0) | |
1176 | family_num = super->anchor->family_num; | |
1177 | sha1_init_ctx(&ctx); | |
1178 | sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx); | |
1179 | sha1_process_bytes(&family_num, sizeof(__u32), &ctx); | |
1180 | if (super->current_vol >= 0) | |
1181 | dev = get_imsm_dev(super, super->current_vol); | |
1182 | if (dev) { | |
1183 | __u32 vol = super->current_vol; | |
1184 | sha1_process_bytes(&vol, sizeof(vol), &ctx); | |
1185 | sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx); | |
1186 | } | |
1187 | sha1_finish_ctx(&ctx, buf); | |
1188 | memcpy(uuid, buf, 4*4); | |
1189 | } | |
1190 | ||
1191 | #if 0 | |
1192 | static void | |
1193 | get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p) | |
1194 | { | |
1195 | __u8 *v = get_imsm_version(mpb); | |
1196 | __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH; | |
1197 | char major[] = { 0, 0, 0 }; | |
1198 | char minor[] = { 0 ,0, 0 }; | |
1199 | char patch[] = { 0, 0, 0 }; | |
1200 | char *ver_parse[] = { major, minor, patch }; | |
1201 | int i, j; | |
1202 | ||
1203 | i = j = 0; | |
1204 | while (*v != '\0' && v < end) { | |
1205 | if (*v != '.' && j < 2) | |
1206 | ver_parse[i][j++] = *v; | |
1207 | else { | |
1208 | i++; | |
1209 | j = 0; | |
1210 | } | |
1211 | v++; | |
1212 | } | |
1213 | ||
1214 | *m = strtol(minor, NULL, 0); | |
1215 | *p = strtol(patch, NULL, 0); | |
1216 | } | |
1217 | #endif | |
1218 | ||
1219 | static int imsm_level_to_layout(int level) | |
1220 | { | |
1221 | switch (level) { | |
1222 | case 0: | |
1223 | case 1: | |
1224 | return 0; | |
1225 | case 5: | |
1226 | case 6: | |
1227 | return ALGORITHM_LEFT_ASYMMETRIC; | |
1228 | case 10: | |
1229 | return 0x102; | |
1230 | } | |
1231 | return UnSet; | |
1232 | } | |
1233 | ||
1234 | static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info) | |
1235 | { | |
1236 | struct intel_super *super = st->sb; | |
1237 | struct imsm_dev *dev = get_imsm_dev(super, super->current_vol); | |
1238 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1239 | struct dl *dl; | |
1240 | ||
1241 | for (dl = super->disks; dl; dl = dl->next) | |
1242 | if (dl->raiddisk == info->disk.raid_disk) | |
1243 | break; | |
1244 | info->container_member = super->current_vol; | |
1245 | info->array.raid_disks = map->num_members; | |
1246 | info->array.level = get_imsm_raid_level(map); | |
1247 | info->array.layout = imsm_level_to_layout(info->array.level); | |
1248 | info->array.md_minor = -1; | |
1249 | info->array.ctime = 0; | |
1250 | info->array.utime = 0; | |
1251 | info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9; | |
1252 | info->array.state = !dev->vol.dirty; | |
1253 | info->custom_array_size = __le32_to_cpu(dev->size_high); | |
1254 | info->custom_array_size <<= 32; | |
1255 | info->custom_array_size |= __le32_to_cpu(dev->size_low); | |
1256 | ||
1257 | info->disk.major = 0; | |
1258 | info->disk.minor = 0; | |
1259 | if (dl) { | |
1260 | info->disk.major = dl->major; | |
1261 | info->disk.minor = dl->minor; | |
1262 | } | |
1263 | ||
1264 | info->data_offset = __le32_to_cpu(map->pba_of_lba0); | |
1265 | info->component_size = __le32_to_cpu(map->blocks_per_member); | |
1266 | memset(info->uuid, 0, sizeof(info->uuid)); | |
1267 | ||
1268 | if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty) | |
1269 | info->resync_start = 0; | |
1270 | else if (dev->vol.migr_state) | |
1271 | /* FIXME add curr_migr_unit to resync_start conversion */ | |
1272 | info->resync_start = 0; | |
1273 | else | |
1274 | info->resync_start = MaxSector; | |
1275 | ||
1276 | strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN); | |
1277 | info->name[MAX_RAID_SERIAL_LEN] = 0; | |
1278 | ||
1279 | info->array.major_version = -1; | |
1280 | info->array.minor_version = -2; | |
1281 | sprintf(info->text_version, "/%s/%d", | |
1282 | devnum2devname(st->container_dev), | |
1283 | info->container_member); | |
1284 | info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */ | |
1285 | uuid_from_super_imsm(st, info->uuid); | |
1286 | } | |
1287 | ||
1288 | /* check the config file to see if we can return a real uuid for this spare */ | |
1289 | static void fixup_container_spare_uuid(struct mdinfo *inf) | |
1290 | { | |
1291 | struct mddev_ident_s *array_list; | |
1292 | ||
1293 | if (inf->array.level != LEVEL_CONTAINER || | |
1294 | memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0) | |
1295 | return; | |
1296 | ||
1297 | array_list = conf_get_ident(NULL); | |
1298 | ||
1299 | for (; array_list; array_list = array_list->next) { | |
1300 | if (array_list->uuid_set) { | |
1301 | struct supertype *_sst; /* spare supertype */ | |
1302 | struct supertype *_cst; /* container supertype */ | |
1303 | ||
1304 | _cst = array_list->st; | |
1305 | if (_cst) | |
1306 | _sst = _cst->ss->match_metadata_desc(inf->text_version); | |
1307 | else | |
1308 | _sst = NULL; | |
1309 | ||
1310 | if (_sst) { | |
1311 | memcpy(inf->uuid, array_list->uuid, sizeof(int[4])); | |
1312 | free(_sst); | |
1313 | break; | |
1314 | } | |
1315 | } | |
1316 | } | |
1317 | } | |
1318 | ||
1319 | static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info) | |
1320 | { | |
1321 | struct intel_super *super = st->sb; | |
1322 | struct imsm_disk *disk; | |
1323 | ||
1324 | if (super->current_vol >= 0) { | |
1325 | getinfo_super_imsm_volume(st, info); | |
1326 | return; | |
1327 | } | |
1328 | ||
1329 | /* Set raid_disks to zero so that Assemble will always pull in valid | |
1330 | * spares | |
1331 | */ | |
1332 | info->array.raid_disks = 0; | |
1333 | info->array.level = LEVEL_CONTAINER; | |
1334 | info->array.layout = 0; | |
1335 | info->array.md_minor = -1; | |
1336 | info->array.ctime = 0; /* N/A for imsm */ | |
1337 | info->array.utime = 0; | |
1338 | info->array.chunk_size = 0; | |
1339 | ||
1340 | info->disk.major = 0; | |
1341 | info->disk.minor = 0; | |
1342 | info->disk.raid_disk = -1; | |
1343 | info->reshape_active = 0; | |
1344 | info->array.major_version = -1; | |
1345 | info->array.minor_version = -2; | |
1346 | strcpy(info->text_version, "imsm"); | |
1347 | info->safe_mode_delay = 0; | |
1348 | info->disk.number = -1; | |
1349 | info->disk.state = 0; | |
1350 | info->name[0] = 0; | |
1351 | ||
1352 | if (super->disks) { | |
1353 | __u32 reserved = imsm_reserved_sectors(super, super->disks); | |
1354 | ||
1355 | disk = &super->disks->disk; | |
1356 | info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved; | |
1357 | info->component_size = reserved; | |
1358 | info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0; | |
1359 | /* we don't change info->disk.raid_disk here because | |
1360 | * this state will be finalized in mdmon after we have | |
1361 | * found the 'most fresh' version of the metadata | |
1362 | */ | |
1363 | info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0; | |
1364 | info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC); | |
1365 | } | |
1366 | ||
1367 | /* only call uuid_from_super_imsm when this disk is part of a populated container, | |
1368 | * ->compare_super may have updated the 'num_raid_devs' field for spares | |
1369 | */ | |
1370 | if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs) | |
1371 | uuid_from_super_imsm(st, info->uuid); | |
1372 | else { | |
1373 | memcpy(info->uuid, uuid_match_any, sizeof(int[4])); | |
1374 | fixup_container_spare_uuid(info); | |
1375 | } | |
1376 | } | |
1377 | ||
1378 | static int update_super_imsm(struct supertype *st, struct mdinfo *info, | |
1379 | char *update, char *devname, int verbose, | |
1380 | int uuid_set, char *homehost) | |
1381 | { | |
1382 | /* For 'assemble' and 'force' we need to return non-zero if any | |
1383 | * change was made. For others, the return value is ignored. | |
1384 | * Update options are: | |
1385 | * force-one : This device looks a bit old but needs to be included, | |
1386 | * update age info appropriately. | |
1387 | * assemble: clear any 'faulty' flag to allow this device to | |
1388 | * be assembled. | |
1389 | * force-array: Array is degraded but being forced, mark it clean | |
1390 | * if that will be needed to assemble it. | |
1391 | * | |
1392 | * newdev: not used ???? | |
1393 | * grow: Array has gained a new device - this is currently for | |
1394 | * linear only | |
1395 | * resync: mark as dirty so a resync will happen. | |
1396 | * name: update the name - preserving the homehost | |
1397 | * uuid: Change the uuid of the array to match watch is given | |
1398 | * | |
1399 | * Following are not relevant for this imsm: | |
1400 | * sparc2.2 : update from old dodgey metadata | |
1401 | * super-minor: change the preferred_minor number | |
1402 | * summaries: update redundant counters. | |
1403 | * homehost: update the recorded homehost | |
1404 | * _reshape_progress: record new reshape_progress position. | |
1405 | */ | |
1406 | int rv = 1; | |
1407 | struct intel_super *super = st->sb; | |
1408 | struct imsm_super *mpb; | |
1409 | ||
1410 | /* we can only update container info */ | |
1411 | if (!super || super->current_vol >= 0 || !super->anchor) | |
1412 | return 1; | |
1413 | ||
1414 | mpb = super->anchor; | |
1415 | ||
1416 | if (strcmp(update, "uuid") == 0 && uuid_set && !info->update_private) | |
1417 | fprintf(stderr, | |
1418 | Name ": '--uuid' not supported for imsm metadata\n"); | |
1419 | else if (strcmp(update, "uuid") == 0 && uuid_set && info->update_private) { | |
1420 | mpb->orig_family_num = *((__u32 *) info->update_private); | |
1421 | rv = 0; | |
1422 | } else if (strcmp(update, "uuid") == 0) { | |
1423 | __u32 *new_family = malloc(sizeof(*new_family)); | |
1424 | ||
1425 | /* update orig_family_number with the incoming random | |
1426 | * data, report the new effective uuid, and store the | |
1427 | * new orig_family_num for future updates. | |
1428 | */ | |
1429 | if (new_family) { | |
1430 | memcpy(&mpb->orig_family_num, info->uuid, sizeof(__u32)); | |
1431 | uuid_from_super_imsm(st, info->uuid); | |
1432 | *new_family = mpb->orig_family_num; | |
1433 | info->update_private = new_family; | |
1434 | rv = 0; | |
1435 | } | |
1436 | } else if (strcmp(update, "assemble") == 0) | |
1437 | rv = 0; | |
1438 | else | |
1439 | fprintf(stderr, | |
1440 | Name ": '--update=%s' not supported for imsm metadata\n", | |
1441 | update); | |
1442 | ||
1443 | /* successful update? recompute checksum */ | |
1444 | if (rv == 0) | |
1445 | mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb)); | |
1446 | ||
1447 | return rv; | |
1448 | } | |
1449 | ||
1450 | static size_t disks_to_mpb_size(int disks) | |
1451 | { | |
1452 | size_t size; | |
1453 | ||
1454 | size = sizeof(struct imsm_super); | |
1455 | size += (disks - 1) * sizeof(struct imsm_disk); | |
1456 | size += 2 * sizeof(struct imsm_dev); | |
1457 | /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */ | |
1458 | size += (4 - 2) * sizeof(struct imsm_map); | |
1459 | /* 4 possible disk_ord_tbl's */ | |
1460 | size += 4 * (disks - 1) * sizeof(__u32); | |
1461 | ||
1462 | return size; | |
1463 | } | |
1464 | ||
1465 | static __u64 avail_size_imsm(struct supertype *st, __u64 devsize) | |
1466 | { | |
1467 | if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS)) | |
1468 | return 0; | |
1469 | ||
1470 | return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS); | |
1471 | } | |
1472 | ||
1473 | static void free_devlist(struct intel_super *super) | |
1474 | { | |
1475 | struct intel_dev *dv; | |
1476 | ||
1477 | while (super->devlist) { | |
1478 | dv = super->devlist->next; | |
1479 | free(super->devlist->dev); | |
1480 | free(super->devlist); | |
1481 | super->devlist = dv; | |
1482 | } | |
1483 | } | |
1484 | ||
1485 | static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src) | |
1486 | { | |
1487 | memcpy(dest, src, sizeof_imsm_dev(src, 0)); | |
1488 | } | |
1489 | ||
1490 | static int compare_super_imsm(struct supertype *st, struct supertype *tst) | |
1491 | { | |
1492 | /* | |
1493 | * return: | |
1494 | * 0 same, or first was empty, and second was copied | |
1495 | * 1 second had wrong number | |
1496 | * 2 wrong uuid | |
1497 | * 3 wrong other info | |
1498 | */ | |
1499 | struct intel_super *first = st->sb; | |
1500 | struct intel_super *sec = tst->sb; | |
1501 | ||
1502 | if (!first) { | |
1503 | st->sb = tst->sb; | |
1504 | tst->sb = NULL; | |
1505 | return 0; | |
1506 | } | |
1507 | ||
1508 | /* if an anchor does not have num_raid_devs set then it is a free | |
1509 | * floating spare | |
1510 | */ | |
1511 | if (first->anchor->num_raid_devs > 0 && | |
1512 | sec->anchor->num_raid_devs > 0) { | |
1513 | /* Determine if these disks might ever have been | |
1514 | * related. Further disambiguation can only take place | |
1515 | * in load_super_imsm_all | |
1516 | */ | |
1517 | __u32 first_family = first->anchor->orig_family_num; | |
1518 | __u32 sec_family = sec->anchor->orig_family_num; | |
1519 | ||
1520 | if (memcmp(first->anchor->sig, sec->anchor->sig, | |
1521 | MAX_SIGNATURE_LENGTH) != 0) | |
1522 | return 3; | |
1523 | ||
1524 | if (first_family == 0) | |
1525 | first_family = first->anchor->family_num; | |
1526 | if (sec_family == 0) | |
1527 | sec_family = sec->anchor->family_num; | |
1528 | ||
1529 | if (first_family != sec_family) | |
1530 | return 3; | |
1531 | ||
1532 | } | |
1533 | ||
1534 | ||
1535 | /* if 'first' is a spare promote it to a populated mpb with sec's | |
1536 | * family number | |
1537 | */ | |
1538 | if (first->anchor->num_raid_devs == 0 && | |
1539 | sec->anchor->num_raid_devs > 0) { | |
1540 | int i; | |
1541 | struct intel_dev *dv; | |
1542 | struct imsm_dev *dev; | |
1543 | ||
1544 | /* we need to copy raid device info from sec if an allocation | |
1545 | * fails here we don't associate the spare | |
1546 | */ | |
1547 | for (i = 0; i < sec->anchor->num_raid_devs; i++) { | |
1548 | dv = malloc(sizeof(*dv)); | |
1549 | if (!dv) | |
1550 | break; | |
1551 | dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1)); | |
1552 | if (!dev) { | |
1553 | free(dv); | |
1554 | break; | |
1555 | } | |
1556 | dv->dev = dev; | |
1557 | dv->index = i; | |
1558 | dv->next = first->devlist; | |
1559 | first->devlist = dv; | |
1560 | } | |
1561 | if (i < sec->anchor->num_raid_devs) { | |
1562 | /* allocation failure */ | |
1563 | free_devlist(first); | |
1564 | fprintf(stderr, "imsm: failed to associate spare\n"); | |
1565 | return 3; | |
1566 | } | |
1567 | first->anchor->num_raid_devs = sec->anchor->num_raid_devs; | |
1568 | first->anchor->orig_family_num = sec->anchor->orig_family_num; | |
1569 | first->anchor->family_num = sec->anchor->family_num; | |
1570 | memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH); | |
1571 | for (i = 0; i < sec->anchor->num_raid_devs; i++) | |
1572 | imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i)); | |
1573 | } | |
1574 | ||
1575 | return 0; | |
1576 | } | |
1577 | ||
1578 | static void fd2devname(int fd, char *name) | |
1579 | { | |
1580 | struct stat st; | |
1581 | char path[256]; | |
1582 | char dname[100]; | |
1583 | char *nm; | |
1584 | int rv; | |
1585 | ||
1586 | name[0] = '\0'; | |
1587 | if (fstat(fd, &st) != 0) | |
1588 | return; | |
1589 | sprintf(path, "/sys/dev/block/%d:%d", | |
1590 | major(st.st_rdev), minor(st.st_rdev)); | |
1591 | ||
1592 | rv = readlink(path, dname, sizeof(dname)); | |
1593 | if (rv <= 0) | |
1594 | return; | |
1595 | ||
1596 | dname[rv] = '\0'; | |
1597 | nm = strrchr(dname, '/'); | |
1598 | nm++; | |
1599 | snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm); | |
1600 | } | |
1601 | ||
1602 | extern int scsi_get_serial(int fd, void *buf, size_t buf_len); | |
1603 | ||
1604 | static int imsm_read_serial(int fd, char *devname, | |
1605 | __u8 serial[MAX_RAID_SERIAL_LEN]) | |
1606 | { | |
1607 | unsigned char scsi_serial[255]; | |
1608 | int rv; | |
1609 | int rsp_len; | |
1610 | int len; | |
1611 | char *dest; | |
1612 | char *src; | |
1613 | char *rsp_buf; | |
1614 | int i; | |
1615 | ||
1616 | memset(scsi_serial, 0, sizeof(scsi_serial)); | |
1617 | ||
1618 | rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial)); | |
1619 | ||
1620 | if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) { | |
1621 | memset(serial, 0, MAX_RAID_SERIAL_LEN); | |
1622 | fd2devname(fd, (char *) serial); | |
1623 | return 0; | |
1624 | } | |
1625 | ||
1626 | if (rv != 0) { | |
1627 | if (devname) | |
1628 | fprintf(stderr, | |
1629 | Name ": Failed to retrieve serial for %s\n", | |
1630 | devname); | |
1631 | return rv; | |
1632 | } | |
1633 | ||
1634 | rsp_len = scsi_serial[3]; | |
1635 | if (!rsp_len) { | |
1636 | if (devname) | |
1637 | fprintf(stderr, | |
1638 | Name ": Failed to retrieve serial for %s\n", | |
1639 | devname); | |
1640 | return 2; | |
1641 | } | |
1642 | rsp_buf = (char *) &scsi_serial[4]; | |
1643 | ||
1644 | /* trim all whitespace and non-printable characters and convert | |
1645 | * ':' to ';' | |
1646 | */ | |
1647 | for (i = 0, dest = rsp_buf; i < rsp_len; i++) { | |
1648 | src = &rsp_buf[i]; | |
1649 | if (*src > 0x20) { | |
1650 | /* ':' is reserved for use in placeholder serial | |
1651 | * numbers for missing disks | |
1652 | */ | |
1653 | if (*src == ':') | |
1654 | *dest++ = ';'; | |
1655 | else | |
1656 | *dest++ = *src; | |
1657 | } | |
1658 | } | |
1659 | len = dest - rsp_buf; | |
1660 | dest = rsp_buf; | |
1661 | ||
1662 | /* truncate leading characters */ | |
1663 | if (len > MAX_RAID_SERIAL_LEN) { | |
1664 | dest += len - MAX_RAID_SERIAL_LEN; | |
1665 | len = MAX_RAID_SERIAL_LEN; | |
1666 | } | |
1667 | ||
1668 | memset(serial, 0, MAX_RAID_SERIAL_LEN); | |
1669 | memcpy(serial, dest, len); | |
1670 | ||
1671 | return 0; | |
1672 | } | |
1673 | ||
1674 | static int serialcmp(__u8 *s1, __u8 *s2) | |
1675 | { | |
1676 | return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN); | |
1677 | } | |
1678 | ||
1679 | static void serialcpy(__u8 *dest, __u8 *src) | |
1680 | { | |
1681 | strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN); | |
1682 | } | |
1683 | ||
1684 | #ifndef MDASSEMBLE | |
1685 | static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super) | |
1686 | { | |
1687 | struct dl *dl; | |
1688 | ||
1689 | for (dl = super->disks; dl; dl = dl->next) | |
1690 | if (serialcmp(dl->serial, serial) == 0) | |
1691 | break; | |
1692 | ||
1693 | return dl; | |
1694 | } | |
1695 | #endif | |
1696 | ||
1697 | static struct imsm_disk * | |
1698 | __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx) | |
1699 | { | |
1700 | int i; | |
1701 | ||
1702 | for (i = 0; i < mpb->num_disks; i++) { | |
1703 | struct imsm_disk *disk = __get_imsm_disk(mpb, i); | |
1704 | ||
1705 | if (serialcmp(disk->serial, serial) == 0) { | |
1706 | if (idx) | |
1707 | *idx = i; | |
1708 | return disk; | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | return NULL; | |
1713 | } | |
1714 | ||
1715 | static int | |
1716 | load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd) | |
1717 | { | |
1718 | struct imsm_disk *disk; | |
1719 | struct dl *dl; | |
1720 | struct stat stb; | |
1721 | int rv; | |
1722 | char name[40]; | |
1723 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
1724 | ||
1725 | rv = imsm_read_serial(fd, devname, serial); | |
1726 | ||
1727 | if (rv != 0) | |
1728 | return 2; | |
1729 | ||
1730 | dl = calloc(1, sizeof(*dl)); | |
1731 | if (!dl) { | |
1732 | if (devname) | |
1733 | fprintf(stderr, | |
1734 | Name ": failed to allocate disk buffer for %s\n", | |
1735 | devname); | |
1736 | return 2; | |
1737 | } | |
1738 | ||
1739 | fstat(fd, &stb); | |
1740 | dl->major = major(stb.st_rdev); | |
1741 | dl->minor = minor(stb.st_rdev); | |
1742 | dl->next = super->disks; | |
1743 | dl->fd = keep_fd ? fd : -1; | |
1744 | assert(super->disks == NULL); | |
1745 | super->disks = dl; | |
1746 | serialcpy(dl->serial, serial); | |
1747 | dl->index = -2; | |
1748 | dl->e = NULL; | |
1749 | fd2devname(fd, name); | |
1750 | if (devname) | |
1751 | dl->devname = strdup(devname); | |
1752 | else | |
1753 | dl->devname = strdup(name); | |
1754 | ||
1755 | /* look up this disk's index in the current anchor */ | |
1756 | disk = __serial_to_disk(dl->serial, super->anchor, &dl->index); | |
1757 | if (disk) { | |
1758 | dl->disk = *disk; | |
1759 | /* only set index on disks that are a member of a | |
1760 | * populated contianer, i.e. one with raid_devs | |
1761 | */ | |
1762 | if (is_failed(&dl->disk)) | |
1763 | dl->index = -2; | |
1764 | else if (is_spare(&dl->disk)) | |
1765 | dl->index = -1; | |
1766 | } | |
1767 | ||
1768 | return 0; | |
1769 | } | |
1770 | ||
1771 | #ifndef MDASSEMBLE | |
1772 | /* When migrating map0 contains the 'destination' state while map1 | |
1773 | * contains the current state. When not migrating map0 contains the | |
1774 | * current state. This routine assumes that map[0].map_state is set to | |
1775 | * the current array state before being called. | |
1776 | * | |
1777 | * Migration is indicated by one of the following states | |
1778 | * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed) | |
1779 | * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal | |
1780 | * map1state=unitialized) | |
1781 | * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal | |
1782 | * map1state=normal) | |
1783 | * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal | |
1784 | * map1state=degraded) | |
1785 | */ | |
1786 | static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type) | |
1787 | { | |
1788 | struct imsm_map *dest; | |
1789 | struct imsm_map *src = get_imsm_map(dev, 0); | |
1790 | ||
1791 | dev->vol.migr_state = 1; | |
1792 | set_migr_type(dev, migr_type); | |
1793 | dev->vol.curr_migr_unit = 0; | |
1794 | dest = get_imsm_map(dev, 1); | |
1795 | ||
1796 | /* duplicate and then set the target end state in map[0] */ | |
1797 | memcpy(dest, src, sizeof_imsm_map(src)); | |
1798 | if (migr_type == MIGR_REBUILD) { | |
1799 | __u32 ord; | |
1800 | int i; | |
1801 | ||
1802 | for (i = 0; i < src->num_members; i++) { | |
1803 | ord = __le32_to_cpu(src->disk_ord_tbl[i]); | |
1804 | set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord)); | |
1805 | } | |
1806 | } | |
1807 | ||
1808 | src->map_state = to_state; | |
1809 | } | |
1810 | ||
1811 | static void end_migration(struct imsm_dev *dev, __u8 map_state) | |
1812 | { | |
1813 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1814 | struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state); | |
1815 | int i; | |
1816 | ||
1817 | /* merge any IMSM_ORD_REBUILD bits that were not successfully | |
1818 | * completed in the last migration. | |
1819 | * | |
1820 | * FIXME add support for online capacity expansion and | |
1821 | * raid-level-migration | |
1822 | */ | |
1823 | for (i = 0; i < prev->num_members; i++) | |
1824 | map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i]; | |
1825 | ||
1826 | dev->vol.migr_state = 0; | |
1827 | dev->vol.curr_migr_unit = 0; | |
1828 | map->map_state = map_state; | |
1829 | } | |
1830 | #endif | |
1831 | ||
1832 | static int parse_raid_devices(struct intel_super *super) | |
1833 | { | |
1834 | int i; | |
1835 | struct imsm_dev *dev_new; | |
1836 | size_t len, len_migr; | |
1837 | size_t space_needed = 0; | |
1838 | struct imsm_super *mpb = super->anchor; | |
1839 | ||
1840 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
1841 | struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i); | |
1842 | struct intel_dev *dv; | |
1843 | ||
1844 | len = sizeof_imsm_dev(dev_iter, 0); | |
1845 | len_migr = sizeof_imsm_dev(dev_iter, 1); | |
1846 | if (len_migr > len) | |
1847 | space_needed += len_migr - len; | |
1848 | ||
1849 | dv = malloc(sizeof(*dv)); | |
1850 | if (!dv) | |
1851 | return 1; | |
1852 | dev_new = malloc(len_migr); | |
1853 | if (!dev_new) { | |
1854 | free(dv); | |
1855 | return 1; | |
1856 | } | |
1857 | imsm_copy_dev(dev_new, dev_iter); | |
1858 | dv->dev = dev_new; | |
1859 | dv->index = i; | |
1860 | dv->next = super->devlist; | |
1861 | super->devlist = dv; | |
1862 | } | |
1863 | ||
1864 | /* ensure that super->buf is large enough when all raid devices | |
1865 | * are migrating | |
1866 | */ | |
1867 | if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) { | |
1868 | void *buf; | |
1869 | ||
1870 | len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512); | |
1871 | if (posix_memalign(&buf, 512, len) != 0) | |
1872 | return 1; | |
1873 | ||
1874 | memcpy(buf, super->buf, super->len); | |
1875 | memset(buf + super->len, 0, len - super->len); | |
1876 | free(super->buf); | |
1877 | super->buf = buf; | |
1878 | super->len = len; | |
1879 | } | |
1880 | ||
1881 | return 0; | |
1882 | } | |
1883 | ||
1884 | /* retrieve a pointer to the bbm log which starts after all raid devices */ | |
1885 | struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb) | |
1886 | { | |
1887 | void *ptr = NULL; | |
1888 | ||
1889 | if (__le32_to_cpu(mpb->bbm_log_size)) { | |
1890 | ptr = mpb; | |
1891 | ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size); | |
1892 | } | |
1893 | ||
1894 | return ptr; | |
1895 | } | |
1896 | ||
1897 | static void __free_imsm(struct intel_super *super, int free_disks); | |
1898 | ||
1899 | /* load_imsm_mpb - read matrix metadata | |
1900 | * allocates super->mpb to be freed by free_super | |
1901 | */ | |
1902 | static int load_imsm_mpb(int fd, struct intel_super *super, char *devname) | |
1903 | { | |
1904 | unsigned long long dsize; | |
1905 | unsigned long long sectors; | |
1906 | struct stat; | |
1907 | struct imsm_super *anchor; | |
1908 | __u32 check_sum; | |
1909 | ||
1910 | get_dev_size(fd, NULL, &dsize); | |
1911 | ||
1912 | if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) { | |
1913 | if (devname) | |
1914 | fprintf(stderr, | |
1915 | Name ": Cannot seek to anchor block on %s: %s\n", | |
1916 | devname, strerror(errno)); | |
1917 | return 1; | |
1918 | } | |
1919 | ||
1920 | if (posix_memalign((void**)&anchor, 512, 512) != 0) { | |
1921 | if (devname) | |
1922 | fprintf(stderr, | |
1923 | Name ": Failed to allocate imsm anchor buffer" | |
1924 | " on %s\n", devname); | |
1925 | return 1; | |
1926 | } | |
1927 | if (read(fd, anchor, 512) != 512) { | |
1928 | if (devname) | |
1929 | fprintf(stderr, | |
1930 | Name ": Cannot read anchor block on %s: %s\n", | |
1931 | devname, strerror(errno)); | |
1932 | free(anchor); | |
1933 | return 1; | |
1934 | } | |
1935 | ||
1936 | if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) { | |
1937 | if (devname) | |
1938 | fprintf(stderr, | |
1939 | Name ": no IMSM anchor on %s\n", devname); | |
1940 | free(anchor); | |
1941 | return 2; | |
1942 | } | |
1943 | ||
1944 | __free_imsm(super, 0); | |
1945 | super->len = ROUND_UP(anchor->mpb_size, 512); | |
1946 | if (posix_memalign(&super->buf, 512, super->len) != 0) { | |
1947 | if (devname) | |
1948 | fprintf(stderr, | |
1949 | Name ": unable to allocate %zu byte mpb buffer\n", | |
1950 | super->len); | |
1951 | free(anchor); | |
1952 | return 2; | |
1953 | } | |
1954 | memcpy(super->buf, anchor, 512); | |
1955 | ||
1956 | sectors = mpb_sectors(anchor) - 1; | |
1957 | free(anchor); | |
1958 | if (!sectors) { | |
1959 | check_sum = __gen_imsm_checksum(super->anchor); | |
1960 | if (check_sum != __le32_to_cpu(super->anchor->check_sum)) { | |
1961 | if (devname) | |
1962 | fprintf(stderr, | |
1963 | Name ": IMSM checksum %x != %x on %s\n", | |
1964 | check_sum, | |
1965 | __le32_to_cpu(super->anchor->check_sum), | |
1966 | devname); | |
1967 | return 2; | |
1968 | } | |
1969 | ||
1970 | return 0; | |
1971 | } | |
1972 | ||
1973 | /* read the extended mpb */ | |
1974 | if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) { | |
1975 | if (devname) | |
1976 | fprintf(stderr, | |
1977 | Name ": Cannot seek to extended mpb on %s: %s\n", | |
1978 | devname, strerror(errno)); | |
1979 | return 1; | |
1980 | } | |
1981 | ||
1982 | if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) { | |
1983 | if (devname) | |
1984 | fprintf(stderr, | |
1985 | Name ": Cannot read extended mpb on %s: %s\n", | |
1986 | devname, strerror(errno)); | |
1987 | return 2; | |
1988 | } | |
1989 | ||
1990 | check_sum = __gen_imsm_checksum(super->anchor); | |
1991 | if (check_sum != __le32_to_cpu(super->anchor->check_sum)) { | |
1992 | if (devname) | |
1993 | fprintf(stderr, | |
1994 | Name ": IMSM checksum %x != %x on %s\n", | |
1995 | check_sum, __le32_to_cpu(super->anchor->check_sum), | |
1996 | devname); | |
1997 | return 3; | |
1998 | } | |
1999 | ||
2000 | /* FIXME the BBM log is disk specific so we cannot use this global | |
2001 | * buffer for all disks. Ok for now since we only look at the global | |
2002 | * bbm_log_size parameter to gate assembly | |
2003 | */ | |
2004 | super->bbm_log = __get_imsm_bbm_log(super->anchor); | |
2005 | ||
2006 | return 0; | |
2007 | } | |
2008 | ||
2009 | static int | |
2010 | load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd) | |
2011 | { | |
2012 | int err; | |
2013 | ||
2014 | err = load_imsm_mpb(fd, super, devname); | |
2015 | if (err) | |
2016 | return err; | |
2017 | err = load_imsm_disk(fd, super, devname, keep_fd); | |
2018 | if (err) | |
2019 | return err; | |
2020 | err = parse_raid_devices(super); | |
2021 | ||
2022 | return err; | |
2023 | } | |
2024 | ||
2025 | static void __free_imsm_disk(struct dl *d) | |
2026 | { | |
2027 | if (d->fd >= 0) | |
2028 | close(d->fd); | |
2029 | if (d->devname) | |
2030 | free(d->devname); | |
2031 | if (d->e) | |
2032 | free(d->e); | |
2033 | free(d); | |
2034 | ||
2035 | } | |
2036 | static void free_imsm_disks(struct intel_super *super) | |
2037 | { | |
2038 | struct dl *d; | |
2039 | ||
2040 | while (super->disks) { | |
2041 | d = super->disks; | |
2042 | super->disks = d->next; | |
2043 | __free_imsm_disk(d); | |
2044 | } | |
2045 | while (super->missing) { | |
2046 | d = super->missing; | |
2047 | super->missing = d->next; | |
2048 | __free_imsm_disk(d); | |
2049 | } | |
2050 | ||
2051 | } | |
2052 | ||
2053 | /* free all the pieces hanging off of a super pointer */ | |
2054 | static void __free_imsm(struct intel_super *super, int free_disks) | |
2055 | { | |
2056 | if (super->buf) { | |
2057 | free(super->buf); | |
2058 | super->buf = NULL; | |
2059 | } | |
2060 | if (free_disks) | |
2061 | free_imsm_disks(super); | |
2062 | free_devlist(super); | |
2063 | if (super->hba) { | |
2064 | free((void *) super->hba); | |
2065 | super->hba = NULL; | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | static void free_imsm(struct intel_super *super) | |
2070 | { | |
2071 | __free_imsm(super, 1); | |
2072 | free(super); | |
2073 | } | |
2074 | ||
2075 | static void free_super_imsm(struct supertype *st) | |
2076 | { | |
2077 | struct intel_super *super = st->sb; | |
2078 | ||
2079 | if (!super) | |
2080 | return; | |
2081 | ||
2082 | free_imsm(super); | |
2083 | st->sb = NULL; | |
2084 | } | |
2085 | ||
2086 | static struct intel_super *alloc_super(int creating_imsm) | |
2087 | { | |
2088 | struct intel_super *super = malloc(sizeof(*super)); | |
2089 | ||
2090 | if (super) { | |
2091 | memset(super, 0, sizeof(*super)); | |
2092 | super->creating_imsm = creating_imsm; | |
2093 | super->current_vol = -1; | |
2094 | super->create_offset = ~((__u32 ) 0); | |
2095 | if (!check_env("IMSM_NO_PLATFORM")) | |
2096 | super->orom = find_imsm_orom(); | |
2097 | if (super->orom && !check_env("IMSM_TEST_OROM")) { | |
2098 | struct sys_dev *list, *ent; | |
2099 | ||
2100 | /* find the first intel ahci controller */ | |
2101 | list = find_driver_devices("pci", "ahci"); | |
2102 | for (ent = list; ent; ent = ent->next) | |
2103 | if (devpath_to_vendor(ent->path) == 0x8086) | |
2104 | break; | |
2105 | if (ent) { | |
2106 | super->hba = ent->path; | |
2107 | ent->path = NULL; | |
2108 | } | |
2109 | free_sys_dev(&list); | |
2110 | } | |
2111 | } | |
2112 | ||
2113 | return super; | |
2114 | } | |
2115 | ||
2116 | #ifndef MDASSEMBLE | |
2117 | /* find_missing - helper routine for load_super_imsm_all that identifies | |
2118 | * disks that have disappeared from the system. This routine relies on | |
2119 | * the mpb being uptodate, which it is at load time. | |
2120 | */ | |
2121 | static int find_missing(struct intel_super *super) | |
2122 | { | |
2123 | int i; | |
2124 | struct imsm_super *mpb = super->anchor; | |
2125 | struct dl *dl; | |
2126 | struct imsm_disk *disk; | |
2127 | ||
2128 | for (i = 0; i < mpb->num_disks; i++) { | |
2129 | disk = __get_imsm_disk(mpb, i); | |
2130 | dl = serial_to_dl(disk->serial, super); | |
2131 | if (dl) | |
2132 | continue; | |
2133 | ||
2134 | dl = malloc(sizeof(*dl)); | |
2135 | if (!dl) | |
2136 | return 1; | |
2137 | dl->major = 0; | |
2138 | dl->minor = 0; | |
2139 | dl->fd = -1; | |
2140 | dl->devname = strdup("missing"); | |
2141 | dl->index = i; | |
2142 | serialcpy(dl->serial, disk->serial); | |
2143 | dl->disk = *disk; | |
2144 | dl->e = NULL; | |
2145 | dl->next = super->missing; | |
2146 | super->missing = dl; | |
2147 | } | |
2148 | ||
2149 | return 0; | |
2150 | } | |
2151 | ||
2152 | static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list) | |
2153 | { | |
2154 | struct intel_disk *idisk = disk_list; | |
2155 | ||
2156 | while (idisk) { | |
2157 | if (serialcmp(idisk->disk.serial, serial) == 0) | |
2158 | break; | |
2159 | idisk = idisk->next; | |
2160 | } | |
2161 | ||
2162 | return idisk; | |
2163 | } | |
2164 | ||
2165 | static int __prep_thunderdome(struct intel_super **table, int tbl_size, | |
2166 | struct intel_super *super, | |
2167 | struct intel_disk **disk_list) | |
2168 | { | |
2169 | struct imsm_disk *d = &super->disks->disk; | |
2170 | struct imsm_super *mpb = super->anchor; | |
2171 | int i, j; | |
2172 | ||
2173 | for (i = 0; i < tbl_size; i++) { | |
2174 | struct imsm_super *tbl_mpb = table[i]->anchor; | |
2175 | struct imsm_disk *tbl_d = &table[i]->disks->disk; | |
2176 | ||
2177 | if (tbl_mpb->family_num == mpb->family_num) { | |
2178 | if (tbl_mpb->check_sum == mpb->check_sum) { | |
2179 | dprintf("%s: mpb from %d:%d matches %d:%d\n", | |
2180 | __func__, super->disks->major, | |
2181 | super->disks->minor, | |
2182 | table[i]->disks->major, | |
2183 | table[i]->disks->minor); | |
2184 | break; | |
2185 | } | |
2186 | ||
2187 | if (((is_configured(d) && !is_configured(tbl_d)) || | |
2188 | is_configured(d) == is_configured(tbl_d)) && | |
2189 | tbl_mpb->generation_num < mpb->generation_num) { | |
2190 | /* current version of the mpb is a | |
2191 | * better candidate than the one in | |
2192 | * super_table, but copy over "cross | |
2193 | * generational" status | |
2194 | */ | |
2195 | struct intel_disk *idisk; | |
2196 | ||
2197 | dprintf("%s: mpb from %d:%d replaces %d:%d\n", | |
2198 | __func__, super->disks->major, | |
2199 | super->disks->minor, | |
2200 | table[i]->disks->major, | |
2201 | table[i]->disks->minor); | |
2202 | ||
2203 | idisk = disk_list_get(tbl_d->serial, *disk_list); | |
2204 | if (idisk && is_failed(&idisk->disk)) | |
2205 | tbl_d->status |= FAILED_DISK; | |
2206 | break; | |
2207 | } else { | |
2208 | struct intel_disk *idisk; | |
2209 | struct imsm_disk *disk; | |
2210 | ||
2211 | /* tbl_mpb is more up to date, but copy | |
2212 | * over cross generational status before | |
2213 | * returning | |
2214 | */ | |
2215 | disk = __serial_to_disk(d->serial, mpb, NULL); | |
2216 | if (disk && is_failed(disk)) | |
2217 | d->status |= FAILED_DISK; | |
2218 | ||
2219 | idisk = disk_list_get(d->serial, *disk_list); | |
2220 | if (idisk) { | |
2221 | idisk->owner = i; | |
2222 | if (disk && is_configured(disk)) | |
2223 | idisk->disk.status |= CONFIGURED_DISK; | |
2224 | } | |
2225 | ||
2226 | dprintf("%s: mpb from %d:%d prefer %d:%d\n", | |
2227 | __func__, super->disks->major, | |
2228 | super->disks->minor, | |
2229 | table[i]->disks->major, | |
2230 | table[i]->disks->minor); | |
2231 | ||
2232 | return tbl_size; | |
2233 | } | |
2234 | } | |
2235 | } | |
2236 | ||
2237 | if (i >= tbl_size) | |
2238 | table[tbl_size++] = super; | |
2239 | else | |
2240 | table[i] = super; | |
2241 | ||
2242 | /* update/extend the merged list of imsm_disk records */ | |
2243 | for (j = 0; j < mpb->num_disks; j++) { | |
2244 | struct imsm_disk *disk = __get_imsm_disk(mpb, j); | |
2245 | struct intel_disk *idisk; | |
2246 | ||
2247 | idisk = disk_list_get(disk->serial, *disk_list); | |
2248 | if (idisk) { | |
2249 | idisk->disk.status |= disk->status; | |
2250 | if (is_configured(&idisk->disk) || | |
2251 | is_failed(&idisk->disk)) | |
2252 | idisk->disk.status &= ~(SPARE_DISK); | |
2253 | } else { | |
2254 | idisk = calloc(1, sizeof(*idisk)); | |
2255 | if (!idisk) | |
2256 | return -1; | |
2257 | idisk->owner = IMSM_UNKNOWN_OWNER; | |
2258 | idisk->disk = *disk; | |
2259 | idisk->next = *disk_list; | |
2260 | *disk_list = idisk; | |
2261 | } | |
2262 | ||
2263 | if (serialcmp(idisk->disk.serial, d->serial) == 0) | |
2264 | idisk->owner = i; | |
2265 | } | |
2266 | ||
2267 | return tbl_size; | |
2268 | } | |
2269 | ||
2270 | static struct intel_super * | |
2271 | validate_members(struct intel_super *super, struct intel_disk *disk_list, | |
2272 | const int owner) | |
2273 | { | |
2274 | struct imsm_super *mpb = super->anchor; | |
2275 | int ok_count = 0; | |
2276 | int i; | |
2277 | ||
2278 | for (i = 0; i < mpb->num_disks; i++) { | |
2279 | struct imsm_disk *disk = __get_imsm_disk(mpb, i); | |
2280 | struct intel_disk *idisk; | |
2281 | ||
2282 | idisk = disk_list_get(disk->serial, disk_list); | |
2283 | if (idisk) { | |
2284 | if (idisk->owner == owner || | |
2285 | idisk->owner == IMSM_UNKNOWN_OWNER) | |
2286 | ok_count++; | |
2287 | else | |
2288 | dprintf("%s: '%.16s' owner %d != %d\n", | |
2289 | __func__, disk->serial, idisk->owner, | |
2290 | owner); | |
2291 | } else { | |
2292 | dprintf("%s: unknown disk %x [%d]: %.16s\n", | |
2293 | __func__, __le32_to_cpu(mpb->family_num), i, | |
2294 | disk->serial); | |
2295 | break; | |
2296 | } | |
2297 | } | |
2298 | ||
2299 | if (ok_count == mpb->num_disks) | |
2300 | return super; | |
2301 | return NULL; | |
2302 | } | |
2303 | ||
2304 | static void show_conflicts(__u32 family_num, struct intel_super *super_list) | |
2305 | { | |
2306 | struct intel_super *s; | |
2307 | ||
2308 | for (s = super_list; s; s = s->next) { | |
2309 | if (family_num != s->anchor->family_num) | |
2310 | continue; | |
2311 | fprintf(stderr, "Conflict, offlining family %#x on '%s'\n", | |
2312 | __le32_to_cpu(family_num), s->disks->devname); | |
2313 | } | |
2314 | } | |
2315 | ||
2316 | static struct intel_super * | |
2317 | imsm_thunderdome(struct intel_super **super_list, int len) | |
2318 | { | |
2319 | struct intel_super *super_table[len]; | |
2320 | struct intel_disk *disk_list = NULL; | |
2321 | struct intel_super *champion, *spare; | |
2322 | struct intel_super *s, **del; | |
2323 | int tbl_size = 0; | |
2324 | int conflict; | |
2325 | int i; | |
2326 | ||
2327 | memset(super_table, 0, sizeof(super_table)); | |
2328 | for (s = *super_list; s; s = s->next) | |
2329 | tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list); | |
2330 | ||
2331 | for (i = 0; i < tbl_size; i++) { | |
2332 | struct imsm_disk *d; | |
2333 | struct intel_disk *idisk; | |
2334 | struct imsm_super *mpb = super_table[i]->anchor; | |
2335 | ||
2336 | s = super_table[i]; | |
2337 | d = &s->disks->disk; | |
2338 | ||
2339 | /* 'd' must appear in merged disk list for its | |
2340 | * configuration to be valid | |
2341 | */ | |
2342 | idisk = disk_list_get(d->serial, disk_list); | |
2343 | if (idisk && idisk->owner == i) | |
2344 | s = validate_members(s, disk_list, i); | |
2345 | else | |
2346 | s = NULL; | |
2347 | ||
2348 | if (!s) | |
2349 | dprintf("%s: marking family: %#x from %d:%d offline\n", | |
2350 | __func__, mpb->family_num, | |
2351 | super_table[i]->disks->major, | |
2352 | super_table[i]->disks->minor); | |
2353 | super_table[i] = s; | |
2354 | } | |
2355 | ||
2356 | /* This is where the mdadm implementation differs from the Windows | |
2357 | * driver which has no strict concept of a container. We can only | |
2358 | * assemble one family from a container, so when returning a prodigal | |
2359 | * array member to this system the code will not be able to disambiguate | |
2360 | * the container contents that should be assembled ("foreign" versus | |
2361 | * "local"). It requires user intervention to set the orig_family_num | |
2362 | * to a new value to establish a new container. The Windows driver in | |
2363 | * this situation fixes up the volume name in place and manages the | |
2364 | * foreign array as an independent entity. | |
2365 | */ | |
2366 | s = NULL; | |
2367 | spare = NULL; | |
2368 | conflict = 0; | |
2369 | for (i = 0; i < tbl_size; i++) { | |
2370 | struct intel_super *tbl_ent = super_table[i]; | |
2371 | int is_spare = 0; | |
2372 | ||
2373 | if (!tbl_ent) | |
2374 | continue; | |
2375 | ||
2376 | if (tbl_ent->anchor->num_raid_devs == 0) { | |
2377 | spare = tbl_ent; | |
2378 | is_spare = 1; | |
2379 | } | |
2380 | ||
2381 | if (s && !is_spare) { | |
2382 | show_conflicts(tbl_ent->anchor->family_num, *super_list); | |
2383 | conflict++; | |
2384 | } else if (!s && !is_spare) | |
2385 | s = tbl_ent; | |
2386 | } | |
2387 | ||
2388 | if (!s) | |
2389 | s = spare; | |
2390 | if (!s) { | |
2391 | champion = NULL; | |
2392 | goto out; | |
2393 | } | |
2394 | champion = s; | |
2395 | ||
2396 | if (conflict) | |
2397 | fprintf(stderr, "Chose family %#x on '%s', " | |
2398 | "assemble conflicts to new container with '--update=uuid'\n", | |
2399 | __le32_to_cpu(s->anchor->family_num), s->disks->devname); | |
2400 | ||
2401 | /* collect all dl's onto 'champion', and update them to | |
2402 | * champion's version of the status | |
2403 | */ | |
2404 | for (s = *super_list; s; s = s->next) { | |
2405 | struct imsm_super *mpb = champion->anchor; | |
2406 | struct dl *dl = s->disks; | |
2407 | ||
2408 | if (s == champion) | |
2409 | continue; | |
2410 | ||
2411 | for (i = 0; i < mpb->num_disks; i++) { | |
2412 | struct imsm_disk *disk; | |
2413 | ||
2414 | disk = __serial_to_disk(dl->serial, mpb, &dl->index); | |
2415 | if (disk) { | |
2416 | dl->disk = *disk; | |
2417 | /* only set index on disks that are a member of | |
2418 | * a populated contianer, i.e. one with | |
2419 | * raid_devs | |
2420 | */ | |
2421 | if (is_failed(&dl->disk)) | |
2422 | dl->index = -2; | |
2423 | else if (is_spare(&dl->disk)) | |
2424 | dl->index = -1; | |
2425 | break; | |
2426 | } | |
2427 | } | |
2428 | ||
2429 | if (i >= mpb->num_disks) { | |
2430 | struct intel_disk *idisk; | |
2431 | ||
2432 | idisk = disk_list_get(dl->serial, disk_list); | |
2433 | if (idisk && is_spare(&idisk->disk) && | |
2434 | !is_failed(&idisk->disk) && !is_configured(&idisk->disk)) | |
2435 | dl->index = -1; | |
2436 | else { | |
2437 | dl->index = -2; | |
2438 | continue; | |
2439 | } | |
2440 | } | |
2441 | ||
2442 | dl->next = champion->disks; | |
2443 | champion->disks = dl; | |
2444 | s->disks = NULL; | |
2445 | } | |
2446 | ||
2447 | /* delete 'champion' from super_list */ | |
2448 | for (del = super_list; *del; ) { | |
2449 | if (*del == champion) { | |
2450 | *del = (*del)->next; | |
2451 | break; | |
2452 | } else | |
2453 | del = &(*del)->next; | |
2454 | } | |
2455 | champion->next = NULL; | |
2456 | ||
2457 | out: | |
2458 | while (disk_list) { | |
2459 | struct intel_disk *idisk = disk_list; | |
2460 | ||
2461 | disk_list = disk_list->next; | |
2462 | free(idisk); | |
2463 | } | |
2464 | ||
2465 | return champion; | |
2466 | } | |
2467 | ||
2468 | static int load_super_imsm_all(struct supertype *st, int fd, void **sbp, | |
2469 | char *devname, int keep_fd) | |
2470 | { | |
2471 | struct mdinfo *sra; | |
2472 | struct intel_super *super_list = NULL; | |
2473 | struct intel_super *super = NULL; | |
2474 | int devnum = fd2devnum(fd); | |
2475 | struct mdinfo *sd; | |
2476 | int retry; | |
2477 | int err = 0; | |
2478 | int i; | |
2479 | enum sysfs_read_flags flags; | |
2480 | ||
2481 | flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE; | |
2482 | if (mdmon_running(devnum)) | |
2483 | flags |= SKIP_GONE_DEVS; | |
2484 | ||
2485 | /* check if 'fd' an opened container */ | |
2486 | sra = sysfs_read(fd, 0, flags); | |
2487 | if (!sra) | |
2488 | return 1; | |
2489 | ||
2490 | if (sra->array.major_version != -1 || | |
2491 | sra->array.minor_version != -2 || | |
2492 | strcmp(sra->text_version, "imsm") != 0) | |
2493 | return 1; | |
2494 | ||
2495 | /* load all mpbs */ | |
2496 | for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) { | |
2497 | struct intel_super *s = alloc_super(0); | |
2498 | char nm[20]; | |
2499 | int dfd; | |
2500 | ||
2501 | err = 1; | |
2502 | if (!s) | |
2503 | goto error; | |
2504 | s->next = super_list; | |
2505 | super_list = s; | |
2506 | ||
2507 | err = 2; | |
2508 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
2509 | dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY); | |
2510 | if (dfd < 0) | |
2511 | goto error; | |
2512 | ||
2513 | err = load_and_parse_mpb(dfd, s, NULL, keep_fd); | |
2514 | ||
2515 | /* retry the load if we might have raced against mdmon */ | |
2516 | if (err == 3 && mdmon_running(devnum)) | |
2517 | for (retry = 0; retry < 3; retry++) { | |
2518 | usleep(3000); | |
2519 | err = load_and_parse_mpb(dfd, s, NULL, keep_fd); | |
2520 | if (err != 3) | |
2521 | break; | |
2522 | } | |
2523 | if (!keep_fd) | |
2524 | close(dfd); | |
2525 | if (err) | |
2526 | goto error; | |
2527 | } | |
2528 | ||
2529 | /* all mpbs enter, maybe one leaves */ | |
2530 | super = imsm_thunderdome(&super_list, i); | |
2531 | if (!super) { | |
2532 | err = 1; | |
2533 | goto error; | |
2534 | } | |
2535 | ||
2536 | if (find_missing(super) != 0) { | |
2537 | free_imsm(super); | |
2538 | err = 2; | |
2539 | goto error; | |
2540 | } | |
2541 | ||
2542 | if (st->subarray[0]) { | |
2543 | if (atoi(st->subarray) <= super->anchor->num_raid_devs) | |
2544 | super->current_vol = atoi(st->subarray); | |
2545 | else { | |
2546 | free_imsm(super); | |
2547 | err = 1; | |
2548 | goto error; | |
2549 | } | |
2550 | } | |
2551 | err = 0; | |
2552 | ||
2553 | error: | |
2554 | while (super_list) { | |
2555 | struct intel_super *s = super_list; | |
2556 | ||
2557 | super_list = super_list->next; | |
2558 | free_imsm(s); | |
2559 | } | |
2560 | ||
2561 | if (err) | |
2562 | return err; | |
2563 | ||
2564 | *sbp = super; | |
2565 | st->container_dev = devnum; | |
2566 | if (err == 0 && st->ss == NULL) { | |
2567 | st->ss = &super_imsm; | |
2568 | st->minor_version = 0; | |
2569 | st->max_devs = IMSM_MAX_DEVICES; | |
2570 | } | |
2571 | st->loaded_container = 1; | |
2572 | ||
2573 | return 0; | |
2574 | } | |
2575 | #endif | |
2576 | ||
2577 | static int load_super_imsm(struct supertype *st, int fd, char *devname) | |
2578 | { | |
2579 | struct intel_super *super; | |
2580 | int rv; | |
2581 | ||
2582 | #ifndef MDASSEMBLE | |
2583 | if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0) | |
2584 | return 0; | |
2585 | #endif | |
2586 | ||
2587 | free_super_imsm(st); | |
2588 | ||
2589 | super = alloc_super(0); | |
2590 | if (!super) { | |
2591 | fprintf(stderr, | |
2592 | Name ": malloc of %zu failed.\n", | |
2593 | sizeof(*super)); | |
2594 | return 1; | |
2595 | } | |
2596 | ||
2597 | rv = load_and_parse_mpb(fd, super, devname, 0); | |
2598 | ||
2599 | if (rv) { | |
2600 | if (devname) | |
2601 | fprintf(stderr, | |
2602 | Name ": Failed to load all information " | |
2603 | "sections on %s\n", devname); | |
2604 | free_imsm(super); | |
2605 | return rv; | |
2606 | } | |
2607 | ||
2608 | if (st->subarray[0]) { | |
2609 | if (atoi(st->subarray) <= super->anchor->num_raid_devs) | |
2610 | super->current_vol = atoi(st->subarray); | |
2611 | else { | |
2612 | free_imsm(super); | |
2613 | return 1; | |
2614 | } | |
2615 | } | |
2616 | ||
2617 | st->sb = super; | |
2618 | if (st->ss == NULL) { | |
2619 | st->ss = &super_imsm; | |
2620 | st->minor_version = 0; | |
2621 | st->max_devs = IMSM_MAX_DEVICES; | |
2622 | } | |
2623 | st->loaded_container = 0; | |
2624 | ||
2625 | return 0; | |
2626 | } | |
2627 | ||
2628 | static __u16 info_to_blocks_per_strip(mdu_array_info_t *info) | |
2629 | { | |
2630 | if (info->level == 1) | |
2631 | return 128; | |
2632 | return info->chunk_size >> 9; | |
2633 | } | |
2634 | ||
2635 | static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains) | |
2636 | { | |
2637 | __u32 num_stripes; | |
2638 | ||
2639 | num_stripes = (info->size * 2) / info_to_blocks_per_strip(info); | |
2640 | num_stripes /= num_domains; | |
2641 | ||
2642 | return num_stripes; | |
2643 | } | |
2644 | ||
2645 | static __u32 info_to_blocks_per_member(mdu_array_info_t *info) | |
2646 | { | |
2647 | if (info->level == 1) | |
2648 | return info->size * 2; | |
2649 | else | |
2650 | return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1); | |
2651 | } | |
2652 | ||
2653 | static void imsm_update_version_info(struct intel_super *super) | |
2654 | { | |
2655 | /* update the version and attributes */ | |
2656 | struct imsm_super *mpb = super->anchor; | |
2657 | char *version; | |
2658 | struct imsm_dev *dev; | |
2659 | struct imsm_map *map; | |
2660 | int i; | |
2661 | ||
2662 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
2663 | dev = get_imsm_dev(super, i); | |
2664 | map = get_imsm_map(dev, 0); | |
2665 | if (__le32_to_cpu(dev->size_high) > 0) | |
2666 | mpb->attributes |= MPB_ATTRIB_2TB; | |
2667 | ||
2668 | /* FIXME detect when an array spans a port multiplier */ | |
2669 | #if 0 | |
2670 | mpb->attributes |= MPB_ATTRIB_PM; | |
2671 | #endif | |
2672 | ||
2673 | if (mpb->num_raid_devs > 1 || | |
2674 | mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) { | |
2675 | version = MPB_VERSION_ATTRIBS; | |
2676 | switch (get_imsm_raid_level(map)) { | |
2677 | case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break; | |
2678 | case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break; | |
2679 | case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break; | |
2680 | case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break; | |
2681 | } | |
2682 | } else { | |
2683 | if (map->num_members >= 5) | |
2684 | version = MPB_VERSION_5OR6_DISK_ARRAY; | |
2685 | else if (dev->status == DEV_CLONE_N_GO) | |
2686 | version = MPB_VERSION_CNG; | |
2687 | else if (get_imsm_raid_level(map) == 5) | |
2688 | version = MPB_VERSION_RAID5; | |
2689 | else if (map->num_members >= 3) | |
2690 | version = MPB_VERSION_3OR4_DISK_ARRAY; | |
2691 | else if (get_imsm_raid_level(map) == 1) | |
2692 | version = MPB_VERSION_RAID1; | |
2693 | else | |
2694 | version = MPB_VERSION_RAID0; | |
2695 | } | |
2696 | strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version); | |
2697 | } | |
2698 | } | |
2699 | ||
2700 | static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info, | |
2701 | unsigned long long size, char *name, | |
2702 | char *homehost, int *uuid) | |
2703 | { | |
2704 | /* We are creating a volume inside a pre-existing container. | |
2705 | * so st->sb is already set. | |
2706 | */ | |
2707 | struct intel_super *super = st->sb; | |
2708 | struct imsm_super *mpb = super->anchor; | |
2709 | struct intel_dev *dv; | |
2710 | struct imsm_dev *dev; | |
2711 | struct imsm_vol *vol; | |
2712 | struct imsm_map *map; | |
2713 | int idx = mpb->num_raid_devs; | |
2714 | int i; | |
2715 | unsigned long long array_blocks; | |
2716 | size_t size_old, size_new; | |
2717 | __u32 num_data_stripes; | |
2718 | ||
2719 | if (super->orom && mpb->num_raid_devs >= super->orom->vpa) { | |
2720 | fprintf(stderr, Name": This imsm-container already has the " | |
2721 | "maximum of %d volumes\n", super->orom->vpa); | |
2722 | return 0; | |
2723 | } | |
2724 | ||
2725 | /* ensure the mpb is large enough for the new data */ | |
2726 | size_old = __le32_to_cpu(mpb->mpb_size); | |
2727 | size_new = disks_to_mpb_size(info->nr_disks); | |
2728 | if (size_new > size_old) { | |
2729 | void *mpb_new; | |
2730 | size_t size_round = ROUND_UP(size_new, 512); | |
2731 | ||
2732 | if (posix_memalign(&mpb_new, 512, size_round) != 0) { | |
2733 | fprintf(stderr, Name": could not allocate new mpb\n"); | |
2734 | return 0; | |
2735 | } | |
2736 | memcpy(mpb_new, mpb, size_old); | |
2737 | free(mpb); | |
2738 | mpb = mpb_new; | |
2739 | super->anchor = mpb_new; | |
2740 | mpb->mpb_size = __cpu_to_le32(size_new); | |
2741 | memset(mpb_new + size_old, 0, size_round - size_old); | |
2742 | } | |
2743 | super->current_vol = idx; | |
2744 | /* when creating the first raid device in this container set num_disks | |
2745 | * to zero, i.e. delete this spare and add raid member devices in | |
2746 | * add_to_super_imsm_volume() | |
2747 | */ | |
2748 | if (super->current_vol == 0) | |
2749 | mpb->num_disks = 0; | |
2750 | ||
2751 | for (i = 0; i < super->current_vol; i++) { | |
2752 | dev = get_imsm_dev(super, i); | |
2753 | if (strncmp((char *) dev->volume, name, | |
2754 | MAX_RAID_SERIAL_LEN) == 0) { | |
2755 | fprintf(stderr, Name": '%s' is already defined for this container\n", | |
2756 | name); | |
2757 | return 0; | |
2758 | } | |
2759 | } | |
2760 | ||
2761 | sprintf(st->subarray, "%d", idx); | |
2762 | dv = malloc(sizeof(*dv)); | |
2763 | if (!dv) { | |
2764 | fprintf(stderr, Name ": failed to allocate device list entry\n"); | |
2765 | return 0; | |
2766 | } | |
2767 | dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1)); | |
2768 | if (!dev) { | |
2769 | free(dv); | |
2770 | fprintf(stderr, Name": could not allocate raid device\n"); | |
2771 | return 0; | |
2772 | } | |
2773 | strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN); | |
2774 | if (info->level == 1) | |
2775 | array_blocks = info_to_blocks_per_member(info); | |
2776 | else | |
2777 | array_blocks = calc_array_size(info->level, info->raid_disks, | |
2778 | info->layout, info->chunk_size, | |
2779 | info->size*2); | |
2780 | /* round array size down to closest MB */ | |
2781 | array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT; | |
2782 | ||
2783 | dev->size_low = __cpu_to_le32((__u32) array_blocks); | |
2784 | dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32)); | |
2785 | dev->status = __cpu_to_le32(0); | |
2786 | dev->reserved_blocks = __cpu_to_le32(0); | |
2787 | vol = &dev->vol; | |
2788 | vol->migr_state = 0; | |
2789 | set_migr_type(dev, MIGR_INIT); | |
2790 | vol->dirty = 0; | |
2791 | vol->curr_migr_unit = 0; | |
2792 | map = get_imsm_map(dev, 0); | |
2793 | map->pba_of_lba0 = __cpu_to_le32(super->create_offset); | |
2794 | map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info)); | |
2795 | map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info)); | |
2796 | map->failed_disk_num = ~0; | |
2797 | map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED : | |
2798 | IMSM_T_STATE_NORMAL; | |
2799 | map->ddf = 1; | |
2800 | ||
2801 | if (info->level == 1 && info->raid_disks > 2) { | |
2802 | fprintf(stderr, Name": imsm does not support more than 2 disks" | |
2803 | "in a raid1 volume\n"); | |
2804 | return 0; | |
2805 | } | |
2806 | ||
2807 | map->raid_level = info->level; | |
2808 | if (info->level == 10) { | |
2809 | map->raid_level = 1; | |
2810 | map->num_domains = info->raid_disks / 2; | |
2811 | } else if (info->level == 1) | |
2812 | map->num_domains = info->raid_disks; | |
2813 | else | |
2814 | map->num_domains = 1; | |
2815 | ||
2816 | num_data_stripes = info_to_num_data_stripes(info, map->num_domains); | |
2817 | map->num_data_stripes = __cpu_to_le32(num_data_stripes); | |
2818 | ||
2819 | map->num_members = info->raid_disks; | |
2820 | for (i = 0; i < map->num_members; i++) { | |
2821 | /* initialized in add_to_super */ | |
2822 | set_imsm_ord_tbl_ent(map, i, 0); | |
2823 | } | |
2824 | mpb->num_raid_devs++; | |
2825 | ||
2826 | dv->dev = dev; | |
2827 | dv->index = super->current_vol; | |
2828 | dv->next = super->devlist; | |
2829 | super->devlist = dv; | |
2830 | ||
2831 | imsm_update_version_info(super); | |
2832 | ||
2833 | return 1; | |
2834 | } | |
2835 | ||
2836 | static int init_super_imsm(struct supertype *st, mdu_array_info_t *info, | |
2837 | unsigned long long size, char *name, | |
2838 | char *homehost, int *uuid) | |
2839 | { | |
2840 | /* This is primarily called by Create when creating a new array. | |
2841 | * We will then get add_to_super called for each component, and then | |
2842 | * write_init_super called to write it out to each device. | |
2843 | * For IMSM, Create can create on fresh devices or on a pre-existing | |
2844 | * array. | |
2845 | * To create on a pre-existing array a different method will be called. | |
2846 | * This one is just for fresh drives. | |
2847 | */ | |
2848 | struct intel_super *super; | |
2849 | struct imsm_super *mpb; | |
2850 | size_t mpb_size; | |
2851 | char *version; | |
2852 | ||
2853 | if (st->sb) | |
2854 | return init_super_imsm_volume(st, info, size, name, homehost, uuid); | |
2855 | ||
2856 | if (info) | |
2857 | mpb_size = disks_to_mpb_size(info->nr_disks); | |
2858 | else | |
2859 | mpb_size = 512; | |
2860 | ||
2861 | super = alloc_super(1); | |
2862 | if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) { | |
2863 | free(super); | |
2864 | super = NULL; | |
2865 | } | |
2866 | if (!super) { | |
2867 | fprintf(stderr, Name | |
2868 | ": %s could not allocate superblock\n", __func__); | |
2869 | return 0; | |
2870 | } | |
2871 | memset(super->buf, 0, mpb_size); | |
2872 | mpb = super->buf; | |
2873 | mpb->mpb_size = __cpu_to_le32(mpb_size); | |
2874 | st->sb = super; | |
2875 | ||
2876 | if (info == NULL) { | |
2877 | /* zeroing superblock */ | |
2878 | return 0; | |
2879 | } | |
2880 | ||
2881 | mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY; | |
2882 | ||
2883 | version = (char *) mpb->sig; | |
2884 | strcpy(version, MPB_SIGNATURE); | |
2885 | version += strlen(MPB_SIGNATURE); | |
2886 | strcpy(version, MPB_VERSION_RAID0); | |
2887 | ||
2888 | return 1; | |
2889 | } | |
2890 | ||
2891 | #ifndef MDASSEMBLE | |
2892 | static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk, | |
2893 | int fd, char *devname) | |
2894 | { | |
2895 | struct intel_super *super = st->sb; | |
2896 | struct imsm_super *mpb = super->anchor; | |
2897 | struct dl *dl; | |
2898 | struct imsm_dev *dev; | |
2899 | struct imsm_map *map; | |
2900 | ||
2901 | dev = get_imsm_dev(super, super->current_vol); | |
2902 | map = get_imsm_map(dev, 0); | |
2903 | ||
2904 | if (! (dk->state & (1<<MD_DISK_SYNC))) { | |
2905 | fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n", | |
2906 | devname); | |
2907 | return 1; | |
2908 | } | |
2909 | ||
2910 | if (fd == -1) { | |
2911 | /* we're doing autolayout so grab the pre-marked (in | |
2912 | * validate_geometry) raid_disk | |
2913 | */ | |
2914 | for (dl = super->disks; dl; dl = dl->next) | |
2915 | if (dl->raiddisk == dk->raid_disk) | |
2916 | break; | |
2917 | } else { | |
2918 | for (dl = super->disks; dl ; dl = dl->next) | |
2919 | if (dl->major == dk->major && | |
2920 | dl->minor == dk->minor) | |
2921 | break; | |
2922 | } | |
2923 | ||
2924 | if (!dl) { | |
2925 | fprintf(stderr, Name ": %s is not a member of the same container\n", devname); | |
2926 | return 1; | |
2927 | } | |
2928 | ||
2929 | /* add a pristine spare to the metadata */ | |
2930 | if (dl->index < 0) { | |
2931 | dl->index = super->anchor->num_disks; | |
2932 | super->anchor->num_disks++; | |
2933 | } | |
2934 | set_imsm_ord_tbl_ent(map, dk->number, dl->index); | |
2935 | dl->disk.status = CONFIGURED_DISK; | |
2936 | ||
2937 | /* if we are creating the first raid device update the family number */ | |
2938 | if (super->current_vol == 0) { | |
2939 | __u32 sum; | |
2940 | struct imsm_dev *_dev = __get_imsm_dev(mpb, 0); | |
2941 | struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index); | |
2942 | ||
2943 | *_dev = *dev; | |
2944 | *_disk = dl->disk; | |
2945 | sum = random32(); | |
2946 | sum += __gen_imsm_checksum(mpb); | |
2947 | mpb->family_num = __cpu_to_le32(sum); | |
2948 | mpb->orig_family_num = mpb->family_num; | |
2949 | } | |
2950 | ||
2951 | return 0; | |
2952 | } | |
2953 | ||
2954 | static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk, | |
2955 | int fd, char *devname) | |
2956 | { | |
2957 | struct intel_super *super = st->sb; | |
2958 | struct dl *dd; | |
2959 | unsigned long long size; | |
2960 | __u32 id; | |
2961 | int rv; | |
2962 | struct stat stb; | |
2963 | ||
2964 | /* if we are on an RAID enabled platform check that the disk is | |
2965 | * attached to the raid controller | |
2966 | */ | |
2967 | if (super->hba && !disk_attached_to_hba(fd, super->hba)) { | |
2968 | fprintf(stderr, | |
2969 | Name ": %s is not attached to the raid controller: %s\n", | |
2970 | devname ? : "disk", super->hba); | |
2971 | return 1; | |
2972 | } | |
2973 | ||
2974 | if (super->current_vol >= 0) | |
2975 | return add_to_super_imsm_volume(st, dk, fd, devname); | |
2976 | ||
2977 | fstat(fd, &stb); | |
2978 | dd = malloc(sizeof(*dd)); | |
2979 | if (!dd) { | |
2980 | fprintf(stderr, | |
2981 | Name ": malloc failed %s:%d.\n", __func__, __LINE__); | |
2982 | return 1; | |
2983 | } | |
2984 | memset(dd, 0, sizeof(*dd)); | |
2985 | dd->major = major(stb.st_rdev); | |
2986 | dd->minor = minor(stb.st_rdev); | |
2987 | dd->index = -1; | |
2988 | dd->devname = devname ? strdup(devname) : NULL; | |
2989 | dd->fd = fd; | |
2990 | dd->e = NULL; | |
2991 | rv = imsm_read_serial(fd, devname, dd->serial); | |
2992 | if (rv) { | |
2993 | fprintf(stderr, | |
2994 | Name ": failed to retrieve scsi serial, aborting\n"); | |
2995 | free(dd); | |
2996 | abort(); | |
2997 | } | |
2998 | ||
2999 | get_dev_size(fd, NULL, &size); | |
3000 | size /= 512; | |
3001 | serialcpy(dd->disk.serial, dd->serial); | |
3002 | dd->disk.total_blocks = __cpu_to_le32(size); | |
3003 | dd->disk.status = SPARE_DISK; | |
3004 | if (sysfs_disk_to_scsi_id(fd, &id) == 0) | |
3005 | dd->disk.scsi_id = __cpu_to_le32(id); | |
3006 | else | |
3007 | dd->disk.scsi_id = __cpu_to_le32(0); | |
3008 | ||
3009 | if (st->update_tail) { | |
3010 | dd->next = super->add; | |
3011 | super->add = dd; | |
3012 | } else { | |
3013 | dd->next = super->disks; | |
3014 | super->disks = dd; | |
3015 | } | |
3016 | ||
3017 | return 0; | |
3018 | } | |
3019 | ||
3020 | static int store_imsm_mpb(int fd, struct imsm_super *mpb); | |
3021 | ||
3022 | static union { | |
3023 | char buf[512]; | |
3024 | struct imsm_super anchor; | |
3025 | } spare_record __attribute__ ((aligned(512))); | |
3026 | ||
3027 | /* spare records have their own family number and do not have any defined raid | |
3028 | * devices | |
3029 | */ | |
3030 | static int write_super_imsm_spares(struct intel_super *super, int doclose) | |
3031 | { | |
3032 | struct imsm_super *mpb = super->anchor; | |
3033 | struct imsm_super *spare = &spare_record.anchor; | |
3034 | __u32 sum; | |
3035 | struct dl *d; | |
3036 | ||
3037 | spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)), | |
3038 | spare->generation_num = __cpu_to_le32(1UL), | |
3039 | spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY; | |
3040 | spare->num_disks = 1, | |
3041 | spare->num_raid_devs = 0, | |
3042 | spare->cache_size = mpb->cache_size, | |
3043 | spare->pwr_cycle_count = __cpu_to_le32(1), | |
3044 | ||
3045 | snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH, | |
3046 | MPB_SIGNATURE MPB_VERSION_RAID0); | |
3047 | ||
3048 | for (d = super->disks; d; d = d->next) { | |
3049 | if (d->index != -1) | |
3050 | continue; | |
3051 | ||
3052 | spare->disk[0] = d->disk; | |
3053 | sum = __gen_imsm_checksum(spare); | |
3054 | spare->family_num = __cpu_to_le32(sum); | |
3055 | spare->orig_family_num = 0; | |
3056 | sum = __gen_imsm_checksum(spare); | |
3057 | spare->check_sum = __cpu_to_le32(sum); | |
3058 | ||
3059 | if (store_imsm_mpb(d->fd, spare)) { | |
3060 | fprintf(stderr, "%s: failed for device %d:%d %s\n", | |
3061 | __func__, d->major, d->minor, strerror(errno)); | |
3062 | return 1; | |
3063 | } | |
3064 | if (doclose) { | |
3065 | close(d->fd); | |
3066 | d->fd = -1; | |
3067 | } | |
3068 | } | |
3069 | ||
3070 | return 0; | |
3071 | } | |
3072 | ||
3073 | static int write_super_imsm(struct intel_super *super, int doclose) | |
3074 | { | |
3075 | struct imsm_super *mpb = super->anchor; | |
3076 | struct dl *d; | |
3077 | __u32 generation; | |
3078 | __u32 sum; | |
3079 | int spares = 0; | |
3080 | int i; | |
3081 | __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk); | |
3082 | ||
3083 | /* 'generation' is incremented everytime the metadata is written */ | |
3084 | generation = __le32_to_cpu(mpb->generation_num); | |
3085 | generation++; | |
3086 | mpb->generation_num = __cpu_to_le32(generation); | |
3087 | ||
3088 | /* fix up cases where previous mdadm releases failed to set | |
3089 | * orig_family_num | |
3090 | */ | |
3091 | if (mpb->orig_family_num == 0) | |
3092 | mpb->orig_family_num = mpb->family_num; | |
3093 | ||
3094 | mpb_size += sizeof(struct imsm_disk) * mpb->num_disks; | |
3095 | for (d = super->disks; d; d = d->next) { | |
3096 | if (d->index == -1) | |
3097 | spares++; | |
3098 | else | |
3099 | mpb->disk[d->index] = d->disk; | |
3100 | } | |
3101 | for (d = super->missing; d; d = d->next) | |
3102 | mpb->disk[d->index] = d->disk; | |
3103 | ||
3104 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
3105 | struct imsm_dev *dev = __get_imsm_dev(mpb, i); | |
3106 | ||
3107 | imsm_copy_dev(dev, get_imsm_dev(super, i)); | |
3108 | mpb_size += sizeof_imsm_dev(dev, 0); | |
3109 | } | |
3110 | mpb_size += __le32_to_cpu(mpb->bbm_log_size); | |
3111 | mpb->mpb_size = __cpu_to_le32(mpb_size); | |
3112 | ||
3113 | /* recalculate checksum */ | |
3114 | sum = __gen_imsm_checksum(mpb); | |
3115 | mpb->check_sum = __cpu_to_le32(sum); | |
3116 | ||
3117 | /* write the mpb for disks that compose raid devices */ | |
3118 | for (d = super->disks; d ; d = d->next) { | |
3119 | if (d->index < 0) | |
3120 | continue; | |
3121 | if (store_imsm_mpb(d->fd, mpb)) | |
3122 | fprintf(stderr, "%s: failed for device %d:%d %s\n", | |
3123 | __func__, d->major, d->minor, strerror(errno)); | |
3124 | if (doclose) { | |
3125 | close(d->fd); | |
3126 | d->fd = -1; | |
3127 | } | |
3128 | } | |
3129 | ||
3130 | if (spares) | |
3131 | return write_super_imsm_spares(super, doclose); | |
3132 | ||
3133 | return 0; | |
3134 | } | |
3135 | ||
3136 | ||
3137 | static int create_array(struct supertype *st, int dev_idx) | |
3138 | { | |
3139 | size_t len; | |
3140 | struct imsm_update_create_array *u; | |
3141 | struct intel_super *super = st->sb; | |
3142 | struct imsm_dev *dev = get_imsm_dev(super, dev_idx); | |
3143 | struct imsm_map *map = get_imsm_map(dev, 0); | |
3144 | struct disk_info *inf; | |
3145 | struct imsm_disk *disk; | |
3146 | int i; | |
3147 | ||
3148 | len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) + | |
3149 | sizeof(*inf) * map->num_members; | |
3150 | u = malloc(len); | |
3151 | if (!u) { | |
3152 | fprintf(stderr, "%s: failed to allocate update buffer\n", | |
3153 | __func__); | |
3154 | return 1; | |
3155 | } | |
3156 | ||
3157 | u->type = update_create_array; | |
3158 | u->dev_idx = dev_idx; | |
3159 | imsm_copy_dev(&u->dev, dev); | |
3160 | inf = get_disk_info(u); | |
3161 | for (i = 0; i < map->num_members; i++) { | |
3162 | int idx = get_imsm_disk_idx(dev, i); | |
3163 | ||
3164 | disk = get_imsm_disk(super, idx); | |
3165 | serialcpy(inf[i].serial, disk->serial); | |
3166 | } | |
3167 | append_metadata_update(st, u, len); | |
3168 | ||
3169 | return 0; | |
3170 | } | |
3171 | ||
3172 | static int _add_disk(struct supertype *st) | |
3173 | { | |
3174 | struct intel_super *super = st->sb; | |
3175 | size_t len; | |
3176 | struct imsm_update_add_disk *u; | |
3177 | ||
3178 | if (!super->add) | |
3179 | return 0; | |
3180 | ||
3181 | len = sizeof(*u); | |
3182 | u = malloc(len); | |
3183 | if (!u) { | |
3184 | fprintf(stderr, "%s: failed to allocate update buffer\n", | |
3185 | __func__); | |
3186 | return 1; | |
3187 | } | |
3188 | ||
3189 | u->type = update_add_disk; | |
3190 | append_metadata_update(st, u, len); | |
3191 | ||
3192 | return 0; | |
3193 | } | |
3194 | ||
3195 | static int write_init_super_imsm(struct supertype *st) | |
3196 | { | |
3197 | struct intel_super *super = st->sb; | |
3198 | int current_vol = super->current_vol; | |
3199 | ||
3200 | /* we are done with current_vol reset it to point st at the container */ | |
3201 | super->current_vol = -1; | |
3202 | ||
3203 | if (st->update_tail) { | |
3204 | /* queue the recently created array / added disk | |
3205 | * as a metadata update */ | |
3206 | struct dl *d; | |
3207 | int rv; | |
3208 | ||
3209 | /* determine if we are creating a volume or adding a disk */ | |
3210 | if (current_vol < 0) { | |
3211 | /* in the add disk case we are running in mdmon | |
3212 | * context, so don't close fd's | |
3213 | */ | |
3214 | return _add_disk(st); | |
3215 | } else | |
3216 | rv = create_array(st, current_vol); | |
3217 | ||
3218 | for (d = super->disks; d ; d = d->next) { | |
3219 | close(d->fd); | |
3220 | d->fd = -1; | |
3221 | } | |
3222 | ||
3223 | return rv; | |
3224 | } else | |
3225 | return write_super_imsm(st->sb, 1); | |
3226 | } | |
3227 | #endif | |
3228 | ||
3229 | static int store_super_imsm(struct supertype *st, int fd) | |
3230 | { | |
3231 | struct intel_super *super = st->sb; | |
3232 | struct imsm_super *mpb = super ? super->anchor : NULL; | |
3233 | ||
3234 | if (!mpb) | |
3235 | return 1; | |
3236 | ||
3237 | #ifndef MDASSEMBLE | |
3238 | return store_imsm_mpb(fd, mpb); | |
3239 | #else | |
3240 | return 1; | |
3241 | #endif | |
3242 | } | |
3243 | ||
3244 | static int imsm_bbm_log_size(struct imsm_super *mpb) | |
3245 | { | |
3246 | return __le32_to_cpu(mpb->bbm_log_size); | |
3247 | } | |
3248 | ||
3249 | #ifndef MDASSEMBLE | |
3250 | static int validate_geometry_imsm_container(struct supertype *st, int level, | |
3251 | int layout, int raiddisks, int chunk, | |
3252 | unsigned long long size, char *dev, | |
3253 | unsigned long long *freesize, | |
3254 | int verbose) | |
3255 | { | |
3256 | int fd; | |
3257 | unsigned long long ldsize; | |
3258 | const struct imsm_orom *orom; | |
3259 | ||
3260 | if (level != LEVEL_CONTAINER) | |
3261 | return 0; | |
3262 | if (!dev) | |
3263 | return 1; | |
3264 | ||
3265 | if (check_env("IMSM_NO_PLATFORM")) | |
3266 | orom = NULL; | |
3267 | else | |
3268 | orom = find_imsm_orom(); | |
3269 | if (orom && raiddisks > orom->tds) { | |
3270 | if (verbose) | |
3271 | fprintf(stderr, Name ": %d exceeds maximum number of" | |
3272 | " platform supported disks: %d\n", | |
3273 | raiddisks, orom->tds); | |
3274 | return 0; | |
3275 | } | |
3276 | ||
3277 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
3278 | if (fd < 0) { | |
3279 | if (verbose) | |
3280 | fprintf(stderr, Name ": imsm: Cannot open %s: %s\n", | |
3281 | dev, strerror(errno)); | |
3282 | return 0; | |
3283 | } | |
3284 | if (!get_dev_size(fd, dev, &ldsize)) { | |
3285 | close(fd); | |
3286 | return 0; | |
3287 | } | |
3288 | close(fd); | |
3289 | ||
3290 | *freesize = avail_size_imsm(st, ldsize >> 9); | |
3291 | ||
3292 | return 1; | |
3293 | } | |
3294 | ||
3295 | static unsigned long long find_size(struct extent *e, int *idx, int num_extents) | |
3296 | { | |
3297 | const unsigned long long base_start = e[*idx].start; | |
3298 | unsigned long long end = base_start + e[*idx].size; | |
3299 | int i; | |
3300 | ||
3301 | if (base_start == end) | |
3302 | return 0; | |
3303 | ||
3304 | *idx = *idx + 1; | |
3305 | for (i = *idx; i < num_extents; i++) { | |
3306 | /* extend overlapping extents */ | |
3307 | if (e[i].start >= base_start && | |
3308 | e[i].start <= end) { | |
3309 | if (e[i].size == 0) | |
3310 | return 0; | |
3311 | if (e[i].start + e[i].size > end) | |
3312 | end = e[i].start + e[i].size; | |
3313 | } else if (e[i].start > end) { | |
3314 | *idx = i; | |
3315 | break; | |
3316 | } | |
3317 | } | |
3318 | ||
3319 | return end - base_start; | |
3320 | } | |
3321 | ||
3322 | static unsigned long long merge_extents(struct intel_super *super, int sum_extents) | |
3323 | { | |
3324 | /* build a composite disk with all known extents and generate a new | |
3325 | * 'maxsize' given the "all disks in an array must share a common start | |
3326 | * offset" constraint | |
3327 | */ | |
3328 | struct extent *e = calloc(sum_extents, sizeof(*e)); | |
3329 | struct dl *dl; | |
3330 | int i, j; | |
3331 | int start_extent; | |
3332 | unsigned long long pos; | |
3333 | unsigned long long start = 0; | |
3334 | unsigned long long maxsize; | |
3335 | unsigned long reserve; | |
3336 | ||
3337 | if (!e) | |
3338 | return 0; | |
3339 | ||
3340 | /* coalesce and sort all extents. also, check to see if we need to | |
3341 | * reserve space between member arrays | |
3342 | */ | |
3343 | j = 0; | |
3344 | for (dl = super->disks; dl; dl = dl->next) { | |
3345 | if (!dl->e) | |
3346 | continue; | |
3347 | for (i = 0; i < dl->extent_cnt; i++) | |
3348 | e[j++] = dl->e[i]; | |
3349 | } | |
3350 | qsort(e, sum_extents, sizeof(*e), cmp_extent); | |
3351 | ||
3352 | /* merge extents */ | |
3353 | i = 0; | |
3354 | j = 0; | |
3355 | while (i < sum_extents) { | |
3356 | e[j].start = e[i].start; | |
3357 | e[j].size = find_size(e, &i, sum_extents); | |
3358 | j++; | |
3359 | if (e[j-1].size == 0) | |
3360 | break; | |
3361 | } | |
3362 | ||
3363 | pos = 0; | |
3364 | maxsize = 0; | |
3365 | start_extent = 0; | |
3366 | i = 0; | |
3367 | do { | |
3368 | unsigned long long esize; | |
3369 | ||
3370 | esize = e[i].start - pos; | |
3371 | if (esize >= maxsize) { | |
3372 | maxsize = esize; | |
3373 | start = pos; | |
3374 | start_extent = i; | |
3375 | } | |
3376 | pos = e[i].start + e[i].size; | |
3377 | i++; | |
3378 | } while (e[i-1].size); | |
3379 | free(e); | |
3380 | ||
3381 | if (maxsize == 0) | |
3382 | return 0; | |
3383 | ||
3384 | /* FIXME assumes volume at offset 0 is the first volume in a | |
3385 | * container | |
3386 | */ | |
3387 | if (start_extent > 0) | |
3388 | reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */ | |
3389 | else | |
3390 | reserve = 0; | |
3391 | ||
3392 | if (maxsize < reserve) | |
3393 | return 0; | |
3394 | ||
3395 | super->create_offset = ~((__u32) 0); | |
3396 | if (start + reserve > super->create_offset) | |
3397 | return 0; /* start overflows create_offset */ | |
3398 | super->create_offset = start + reserve; | |
3399 | ||
3400 | return maxsize - reserve; | |
3401 | } | |
3402 | ||
3403 | static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks) | |
3404 | { | |
3405 | if (level < 0 || level == 6 || level == 4) | |
3406 | return 0; | |
3407 | ||
3408 | /* if we have an orom prevent invalid raid levels */ | |
3409 | if (orom) | |
3410 | switch (level) { | |
3411 | case 0: return imsm_orom_has_raid0(orom); | |
3412 | case 1: | |
3413 | if (raiddisks > 2) | |
3414 | return imsm_orom_has_raid1e(orom); | |
3415 | return imsm_orom_has_raid1(orom) && raiddisks == 2; | |
3416 | case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4; | |
3417 | case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2; | |
3418 | } | |
3419 | else | |
3420 | return 1; /* not on an Intel RAID platform so anything goes */ | |
3421 | ||
3422 | return 0; | |
3423 | } | |
3424 | ||
3425 | #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg)) | |
3426 | static int | |
3427 | validate_geometry_imsm_orom(struct intel_super *super, int level, int layout, | |
3428 | int raiddisks, int chunk, int verbose) | |
3429 | { | |
3430 | if (!is_raid_level_supported(super->orom, level, raiddisks)) { | |
3431 | pr_vrb(": platform does not support raid%d with %d disk%s\n", | |
3432 | level, raiddisks, raiddisks > 1 ? "s" : ""); | |
3433 | return 0; | |
3434 | } | |
3435 | if (super->orom && level != 1 && | |
3436 | !imsm_orom_has_chunk(super->orom, chunk)) { | |
3437 | pr_vrb(": platform does not support a chunk size of: %d\n", chunk); | |
3438 | return 0; | |
3439 | } | |
3440 | if (layout != imsm_level_to_layout(level)) { | |
3441 | if (level == 5) | |
3442 | pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n"); | |
3443 | else if (level == 10) | |
3444 | pr_vrb(": imsm raid 10 only supports the n2 layout\n"); | |
3445 | else | |
3446 | pr_vrb(": imsm unknown layout %#x for this raid level %d\n", | |
3447 | layout, level); | |
3448 | return 0; | |
3449 | } | |
3450 | ||
3451 | return 1; | |
3452 | } | |
3453 | ||
3454 | /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd | |
3455 | * FIX ME add ahci details | |
3456 | */ | |
3457 | static int validate_geometry_imsm_volume(struct supertype *st, int level, | |
3458 | int layout, int raiddisks, int chunk, | |
3459 | unsigned long long size, char *dev, | |
3460 | unsigned long long *freesize, | |
3461 | int verbose) | |
3462 | { | |
3463 | struct stat stb; | |
3464 | struct intel_super *super = st->sb; | |
3465 | struct imsm_super *mpb = super->anchor; | |
3466 | struct dl *dl; | |
3467 | unsigned long long pos = 0; | |
3468 | unsigned long long maxsize; | |
3469 | struct extent *e; | |
3470 | int i; | |
3471 | ||
3472 | /* We must have the container info already read in. */ | |
3473 | if (!super) | |
3474 | return 0; | |
3475 | ||
3476 | if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose)) | |
3477 | return 0; | |
3478 | ||
3479 | if (!dev) { | |
3480 | /* General test: make sure there is space for | |
3481 | * 'raiddisks' device extents of size 'size' at a given | |
3482 | * offset | |
3483 | */ | |
3484 | unsigned long long minsize = size; | |
3485 | unsigned long long start_offset = MaxSector; | |
3486 | int dcnt = 0; | |
3487 | if (minsize == 0) | |
3488 | minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
3489 | for (dl = super->disks; dl ; dl = dl->next) { | |
3490 | int found = 0; | |
3491 | ||
3492 | pos = 0; | |
3493 | i = 0; | |
3494 | e = get_extents(super, dl); | |
3495 | if (!e) continue; | |
3496 | do { | |
3497 | unsigned long long esize; | |
3498 | esize = e[i].start - pos; | |
3499 | if (esize >= minsize) | |
3500 | found = 1; | |
3501 | if (found && start_offset == MaxSector) { | |
3502 | start_offset = pos; | |
3503 | break; | |
3504 | } else if (found && pos != start_offset) { | |
3505 | found = 0; | |
3506 | break; | |
3507 | } | |
3508 | pos = e[i].start + e[i].size; | |
3509 | i++; | |
3510 | } while (e[i-1].size); | |
3511 | if (found) | |
3512 | dcnt++; | |
3513 | free(e); | |
3514 | } | |
3515 | if (dcnt < raiddisks) { | |
3516 | if (verbose) | |
3517 | fprintf(stderr, Name ": imsm: Not enough " | |
3518 | "devices with space for this array " | |
3519 | "(%d < %d)\n", | |
3520 | dcnt, raiddisks); | |
3521 | return 0; | |
3522 | } | |
3523 | return 1; | |
3524 | } | |
3525 | ||
3526 | /* This device must be a member of the set */ | |
3527 | if (stat(dev, &stb) < 0) | |
3528 | return 0; | |
3529 | if ((S_IFMT & stb.st_mode) != S_IFBLK) | |
3530 | return 0; | |
3531 | for (dl = super->disks ; dl ; dl = dl->next) { | |
3532 | if (dl->major == major(stb.st_rdev) && | |
3533 | dl->minor == minor(stb.st_rdev)) | |
3534 | break; | |
3535 | } | |
3536 | if (!dl) { | |
3537 | if (verbose) | |
3538 | fprintf(stderr, Name ": %s is not in the " | |
3539 | "same imsm set\n", dev); | |
3540 | return 0; | |
3541 | } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) { | |
3542 | /* If a volume is present then the current creation attempt | |
3543 | * cannot incorporate new spares because the orom may not | |
3544 | * understand this configuration (all member disks must be | |
3545 | * members of each array in the container). | |
3546 | */ | |
3547 | fprintf(stderr, Name ": %s is a spare and a volume" | |
3548 | " is already defined for this container\n", dev); | |
3549 | fprintf(stderr, Name ": The option-rom requires all member" | |
3550 | " disks to be a member of all volumes\n"); | |
3551 | return 0; | |
3552 | } | |
3553 | ||
3554 | /* retrieve the largest free space block */ | |
3555 | e = get_extents(super, dl); | |
3556 | maxsize = 0; | |
3557 | i = 0; | |
3558 | if (e) { | |
3559 | do { | |
3560 | unsigned long long esize; | |
3561 | ||
3562 | esize = e[i].start - pos; | |
3563 | if (esize >= maxsize) | |
3564 | maxsize = esize; | |
3565 | pos = e[i].start + e[i].size; | |
3566 | i++; | |
3567 | } while (e[i-1].size); | |
3568 | dl->e = e; | |
3569 | dl->extent_cnt = i; | |
3570 | } else { | |
3571 | if (verbose) | |
3572 | fprintf(stderr, Name ": unable to determine free space for: %s\n", | |
3573 | dev); | |
3574 | return 0; | |
3575 | } | |
3576 | if (maxsize < size) { | |
3577 | if (verbose) | |
3578 | fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n", | |
3579 | dev, maxsize, size); | |
3580 | return 0; | |
3581 | } | |
3582 | ||
3583 | /* count total number of extents for merge */ | |
3584 | i = 0; | |
3585 | for (dl = super->disks; dl; dl = dl->next) | |
3586 | if (dl->e) | |
3587 | i += dl->extent_cnt; | |
3588 | ||
3589 | maxsize = merge_extents(super, i); | |
3590 | if (maxsize < size || maxsize == 0) { | |
3591 | if (verbose) | |
3592 | fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n", | |
3593 | maxsize, size); | |
3594 | return 0; | |
3595 | } | |
3596 | ||
3597 | *freesize = maxsize; | |
3598 | ||
3599 | return 1; | |
3600 | } | |
3601 | ||
3602 | static int reserve_space(struct supertype *st, int raiddisks, | |
3603 | unsigned long long size, int chunk, | |
3604 | unsigned long long *freesize) | |
3605 | { | |
3606 | struct intel_super *super = st->sb; | |
3607 | struct imsm_super *mpb = super->anchor; | |
3608 | struct dl *dl; | |
3609 | int i; | |
3610 | int extent_cnt; | |
3611 | struct extent *e; | |
3612 | unsigned long long maxsize; | |
3613 | unsigned long long minsize; | |
3614 | int cnt; | |
3615 | int used; | |
3616 | ||
3617 | /* find the largest common start free region of the possible disks */ | |
3618 | used = 0; | |
3619 | extent_cnt = 0; | |
3620 | cnt = 0; | |
3621 | for (dl = super->disks; dl; dl = dl->next) { | |
3622 | dl->raiddisk = -1; | |
3623 | ||
3624 | if (dl->index >= 0) | |
3625 | used++; | |
3626 | ||
3627 | /* don't activate new spares if we are orom constrained | |
3628 | * and there is already a volume active in the container | |
3629 | */ | |
3630 | if (super->orom && dl->index < 0 && mpb->num_raid_devs) | |
3631 | continue; | |
3632 | ||
3633 | e = get_extents(super, dl); | |
3634 | if (!e) | |
3635 | continue; | |
3636 | for (i = 1; e[i-1].size; i++) | |
3637 | ; | |
3638 | dl->e = e; | |
3639 | dl->extent_cnt = i; | |
3640 | extent_cnt += i; | |
3641 | cnt++; | |
3642 | } | |
3643 | ||
3644 | maxsize = merge_extents(super, extent_cnt); | |
3645 | minsize = size; | |
3646 | if (size == 0) | |
3647 | minsize = chunk; | |
3648 | ||
3649 | if (cnt < raiddisks || | |
3650 | (super->orom && used && used != raiddisks) || | |
3651 | maxsize < minsize || | |
3652 | maxsize == 0) { | |
3653 | fprintf(stderr, Name ": not enough devices with space to create array.\n"); | |
3654 | return 0; /* No enough free spaces large enough */ | |
3655 | } | |
3656 | ||
3657 | if (size == 0) { | |
3658 | size = maxsize; | |
3659 | if (chunk) { | |
3660 | size /= chunk; | |
3661 | size *= chunk; | |
3662 | } | |
3663 | } | |
3664 | ||
3665 | cnt = 0; | |
3666 | for (dl = super->disks; dl; dl = dl->next) | |
3667 | if (dl->e) | |
3668 | dl->raiddisk = cnt++; | |
3669 | ||
3670 | *freesize = size; | |
3671 | ||
3672 | return 1; | |
3673 | } | |
3674 | ||
3675 | static int validate_geometry_imsm(struct supertype *st, int level, int layout, | |
3676 | int raiddisks, int chunk, unsigned long long size, | |
3677 | char *dev, unsigned long long *freesize, | |
3678 | int verbose) | |
3679 | { | |
3680 | int fd, cfd; | |
3681 | struct mdinfo *sra; | |
3682 | ||
3683 | /* if given unused devices create a container | |
3684 | * if given given devices in a container create a member volume | |
3685 | */ | |
3686 | if (level == LEVEL_CONTAINER) { | |
3687 | /* Must be a fresh device to add to a container */ | |
3688 | return validate_geometry_imsm_container(st, level, layout, | |
3689 | raiddisks, chunk, size, | |
3690 | dev, freesize, | |
3691 | verbose); | |
3692 | } | |
3693 | ||
3694 | if (!dev) { | |
3695 | if (st->sb && freesize) { | |
3696 | /* we are being asked to automatically layout a | |
3697 | * new volume based on the current contents of | |
3698 | * the container. If the the parameters can be | |
3699 | * satisfied reserve_space will record the disks, | |
3700 | * start offset, and size of the volume to be | |
3701 | * created. add_to_super and getinfo_super | |
3702 | * detect when autolayout is in progress. | |
3703 | */ | |
3704 | if (!validate_geometry_imsm_orom(st->sb, level, layout, | |
3705 | raiddisks, chunk, | |
3706 | verbose)) | |
3707 | return 0; | |
3708 | return reserve_space(st, raiddisks, size, chunk, freesize); | |
3709 | } | |
3710 | return 1; | |
3711 | } | |
3712 | if (st->sb) { | |
3713 | /* creating in a given container */ | |
3714 | return validate_geometry_imsm_volume(st, level, layout, | |
3715 | raiddisks, chunk, size, | |
3716 | dev, freesize, verbose); | |
3717 | } | |
3718 | ||
3719 | /* limit creation to the following levels */ | |
3720 | if (!dev) | |
3721 | switch (level) { | |
3722 | case 0: | |
3723 | case 1: | |
3724 | case 10: | |
3725 | case 5: | |
3726 | return 0; | |
3727 | default: | |
3728 | if (verbose) | |
3729 | fprintf(stderr, Name | |
3730 | ": IMSM only supports levels 0,1,5,10\n"); | |
3731 | return 1; | |
3732 | } | |
3733 | ||
3734 | /* This device needs to be a device in an 'imsm' container */ | |
3735 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
3736 | if (fd >= 0) { | |
3737 | if (verbose) | |
3738 | fprintf(stderr, | |
3739 | Name ": Cannot create this array on device %s\n", | |
3740 | dev); | |
3741 | close(fd); | |
3742 | return 0; | |
3743 | } | |
3744 | if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) { | |
3745 | if (verbose) | |
3746 | fprintf(stderr, Name ": Cannot open %s: %s\n", | |
3747 | dev, strerror(errno)); | |
3748 | return 0; | |
3749 | } | |
3750 | /* Well, it is in use by someone, maybe an 'imsm' container. */ | |
3751 | cfd = open_container(fd); | |
3752 | if (cfd < 0) { | |
3753 | close(fd); | |
3754 | if (verbose) | |
3755 | fprintf(stderr, Name ": Cannot use %s: It is busy\n", | |
3756 | dev); | |
3757 | return 0; | |
3758 | } | |
3759 | sra = sysfs_read(cfd, 0, GET_VERSION); | |
3760 | close(fd); | |
3761 | if (sra && sra->array.major_version == -1 && | |
3762 | strcmp(sra->text_version, "imsm") == 0) { | |
3763 | /* This is a member of a imsm container. Load the container | |
3764 | * and try to create a volume | |
3765 | */ | |
3766 | struct intel_super *super; | |
3767 | ||
3768 | if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) { | |
3769 | st->sb = super; | |
3770 | st->container_dev = fd2devnum(cfd); | |
3771 | close(cfd); | |
3772 | return validate_geometry_imsm_volume(st, level, layout, | |
3773 | raiddisks, chunk, | |
3774 | size, dev, | |
3775 | freesize, verbose); | |
3776 | } | |
3777 | close(cfd); | |
3778 | } else /* may belong to another container */ | |
3779 | return 0; | |
3780 | ||
3781 | return 1; | |
3782 | } | |
3783 | #endif /* MDASSEMBLE */ | |
3784 | ||
3785 | static struct mdinfo *container_content_imsm(struct supertype *st) | |
3786 | { | |
3787 | /* Given a container loaded by load_super_imsm_all, | |
3788 | * extract information about all the arrays into | |
3789 | * an mdinfo tree. | |
3790 | * | |
3791 | * For each imsm_dev create an mdinfo, fill it in, | |
3792 | * then look for matching devices in super->disks | |
3793 | * and create appropriate device mdinfo. | |
3794 | */ | |
3795 | struct intel_super *super = st->sb; | |
3796 | struct imsm_super *mpb = super->anchor; | |
3797 | struct mdinfo *rest = NULL; | |
3798 | int i; | |
3799 | ||
3800 | /* do not assemble arrays that might have bad blocks */ | |
3801 | if (imsm_bbm_log_size(super->anchor)) { | |
3802 | fprintf(stderr, Name ": BBM log found in metadata. " | |
3803 | "Cannot activate array(s).\n"); | |
3804 | return NULL; | |
3805 | } | |
3806 | ||
3807 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
3808 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
3809 | struct imsm_map *map = get_imsm_map(dev, 0); | |
3810 | struct mdinfo *this; | |
3811 | int slot; | |
3812 | ||
3813 | /* do not publish arrays that are in the middle of an | |
3814 | * unsupported migration | |
3815 | */ | |
3816 | if (dev->vol.migr_state && | |
3817 | (migr_type(dev) == MIGR_GEN_MIGR || | |
3818 | migr_type(dev) == MIGR_STATE_CHANGE)) { | |
3819 | fprintf(stderr, Name ": cannot assemble volume '%.16s':" | |
3820 | " unsupported migration in progress\n", | |
3821 | dev->volume); | |
3822 | continue; | |
3823 | } | |
3824 | ||
3825 | this = malloc(sizeof(*this)); | |
3826 | memset(this, 0, sizeof(*this)); | |
3827 | this->next = rest; | |
3828 | ||
3829 | super->current_vol = i; | |
3830 | getinfo_super_imsm_volume(st, this); | |
3831 | for (slot = 0 ; slot < map->num_members; slot++) { | |
3832 | struct mdinfo *info_d; | |
3833 | struct dl *d; | |
3834 | int idx; | |
3835 | int skip; | |
3836 | __u32 ord; | |
3837 | ||
3838 | skip = 0; | |
3839 | idx = get_imsm_disk_idx(dev, slot); | |
3840 | ord = get_imsm_ord_tbl_ent(dev, slot); | |
3841 | for (d = super->disks; d ; d = d->next) | |
3842 | if (d->index == idx) | |
3843 | break; | |
3844 | ||
3845 | if (d == NULL) | |
3846 | skip = 1; | |
3847 | if (d && is_failed(&d->disk)) | |
3848 | skip = 1; | |
3849 | if (ord & IMSM_ORD_REBUILD) | |
3850 | skip = 1; | |
3851 | ||
3852 | /* | |
3853 | * if we skip some disks the array will be assmebled degraded; | |
3854 | * reset resync start to avoid a dirty-degraded situation | |
3855 | * | |
3856 | * FIXME handle dirty degraded | |
3857 | */ | |
3858 | if (skip && !dev->vol.dirty) | |
3859 | this->resync_start = MaxSector; | |
3860 | if (skip) | |
3861 | continue; | |
3862 | ||
3863 | info_d = malloc(sizeof(*info_d)); | |
3864 | if (!info_d) { | |
3865 | fprintf(stderr, Name ": failed to allocate disk" | |
3866 | " for volume %.16s\n", dev->volume); | |
3867 | free(this); | |
3868 | this = rest; | |
3869 | break; | |
3870 | } | |
3871 | memset(info_d, 0, sizeof(*info_d)); | |
3872 | info_d->next = this->devs; | |
3873 | this->devs = info_d; | |
3874 | ||
3875 | info_d->disk.number = d->index; | |
3876 | info_d->disk.major = d->major; | |
3877 | info_d->disk.minor = d->minor; | |
3878 | info_d->disk.raid_disk = slot; | |
3879 | info_d->recovery_start = MaxSector; | |
3880 | ||
3881 | this->array.working_disks++; | |
3882 | ||
3883 | info_d->events = __le32_to_cpu(mpb->generation_num); | |
3884 | info_d->data_offset = __le32_to_cpu(map->pba_of_lba0); | |
3885 | info_d->component_size = __le32_to_cpu(map->blocks_per_member); | |
3886 | if (d->devname) | |
3887 | strcpy(info_d->name, d->devname); | |
3888 | } | |
3889 | rest = this; | |
3890 | } | |
3891 | ||
3892 | return rest; | |
3893 | } | |
3894 | ||
3895 | ||
3896 | #ifndef MDASSEMBLE | |
3897 | static int imsm_open_new(struct supertype *c, struct active_array *a, | |
3898 | char *inst) | |
3899 | { | |
3900 | struct intel_super *super = c->sb; | |
3901 | struct imsm_super *mpb = super->anchor; | |
3902 | ||
3903 | if (atoi(inst) >= mpb->num_raid_devs) { | |
3904 | fprintf(stderr, "%s: subarry index %d, out of range\n", | |
3905 | __func__, atoi(inst)); | |
3906 | return -ENODEV; | |
3907 | } | |
3908 | ||
3909 | dprintf("imsm: open_new %s\n", inst); | |
3910 | a->info.container_member = atoi(inst); | |
3911 | return 0; | |
3912 | } | |
3913 | ||
3914 | static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed) | |
3915 | { | |
3916 | struct imsm_map *map = get_imsm_map(dev, 0); | |
3917 | ||
3918 | if (!failed) | |
3919 | return map->map_state == IMSM_T_STATE_UNINITIALIZED ? | |
3920 | IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL; | |
3921 | ||
3922 | switch (get_imsm_raid_level(map)) { | |
3923 | case 0: | |
3924 | return IMSM_T_STATE_FAILED; | |
3925 | break; | |
3926 | case 1: | |
3927 | if (failed < map->num_members) | |
3928 | return IMSM_T_STATE_DEGRADED; | |
3929 | else | |
3930 | return IMSM_T_STATE_FAILED; | |
3931 | break; | |
3932 | case 10: | |
3933 | { | |
3934 | /** | |
3935 | * check to see if any mirrors have failed, otherwise we | |
3936 | * are degraded. Even numbered slots are mirrored on | |
3937 | * slot+1 | |
3938 | */ | |
3939 | int i; | |
3940 | /* gcc -Os complains that this is unused */ | |
3941 | int insync = insync; | |
3942 | ||
3943 | for (i = 0; i < map->num_members; i++) { | |
3944 | __u32 ord = get_imsm_ord_tbl_ent(dev, i); | |
3945 | int idx = ord_to_idx(ord); | |
3946 | struct imsm_disk *disk; | |
3947 | ||
3948 | /* reset the potential in-sync count on even-numbered | |
3949 | * slots. num_copies is always 2 for imsm raid10 | |
3950 | */ | |
3951 | if ((i & 1) == 0) | |
3952 | insync = 2; | |
3953 | ||
3954 | disk = get_imsm_disk(super, idx); | |
3955 | if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD) | |
3956 | insync--; | |
3957 | ||
3958 | /* no in-sync disks left in this mirror the | |
3959 | * array has failed | |
3960 | */ | |
3961 | if (insync == 0) | |
3962 | return IMSM_T_STATE_FAILED; | |
3963 | } | |
3964 | ||
3965 | return IMSM_T_STATE_DEGRADED; | |
3966 | } | |
3967 | case 5: | |
3968 | if (failed < 2) | |
3969 | return IMSM_T_STATE_DEGRADED; | |
3970 | else | |
3971 | return IMSM_T_STATE_FAILED; | |
3972 | break; | |
3973 | default: | |
3974 | break; | |
3975 | } | |
3976 | ||
3977 | return map->map_state; | |
3978 | } | |
3979 | ||
3980 | static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev) | |
3981 | { | |
3982 | int i; | |
3983 | int failed = 0; | |
3984 | struct imsm_disk *disk; | |
3985 | struct imsm_map *map = get_imsm_map(dev, 0); | |
3986 | struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state); | |
3987 | __u32 ord; | |
3988 | int idx; | |
3989 | ||
3990 | /* at the beginning of migration we set IMSM_ORD_REBUILD on | |
3991 | * disks that are being rebuilt. New failures are recorded to | |
3992 | * map[0]. So we look through all the disks we started with and | |
3993 | * see if any failures are still present, or if any new ones | |
3994 | * have arrived | |
3995 | * | |
3996 | * FIXME add support for online capacity expansion and | |
3997 | * raid-level-migration | |
3998 | */ | |
3999 | for (i = 0; i < prev->num_members; i++) { | |
4000 | ord = __le32_to_cpu(prev->disk_ord_tbl[i]); | |
4001 | ord |= __le32_to_cpu(map->disk_ord_tbl[i]); | |
4002 | idx = ord_to_idx(ord); | |
4003 | ||
4004 | disk = get_imsm_disk(super, idx); | |
4005 | if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD) | |
4006 | failed++; | |
4007 | } | |
4008 | ||
4009 | return failed; | |
4010 | } | |
4011 | ||
4012 | static int is_resyncing(struct imsm_dev *dev) | |
4013 | { | |
4014 | struct imsm_map *migr_map; | |
4015 | ||
4016 | if (!dev->vol.migr_state) | |
4017 | return 0; | |
4018 | ||
4019 | if (migr_type(dev) == MIGR_INIT || | |
4020 | migr_type(dev) == MIGR_REPAIR) | |
4021 | return 1; | |
4022 | ||
4023 | migr_map = get_imsm_map(dev, 1); | |
4024 | ||
4025 | if (migr_map->map_state == IMSM_T_STATE_NORMAL) | |
4026 | return 1; | |
4027 | else | |
4028 | return 0; | |
4029 | } | |
4030 | ||
4031 | static int is_rebuilding(struct imsm_dev *dev) | |
4032 | { | |
4033 | struct imsm_map *migr_map; | |
4034 | ||
4035 | if (!dev->vol.migr_state) | |
4036 | return 0; | |
4037 | ||
4038 | if (migr_type(dev) != MIGR_REBUILD) | |
4039 | return 0; | |
4040 | ||
4041 | migr_map = get_imsm_map(dev, 1); | |
4042 | ||
4043 | if (migr_map->map_state == IMSM_T_STATE_DEGRADED) | |
4044 | return 1; | |
4045 | else | |
4046 | return 0; | |
4047 | } | |
4048 | ||
4049 | /* return true if we recorded new information */ | |
4050 | static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx) | |
4051 | { | |
4052 | __u32 ord; | |
4053 | int slot; | |
4054 | struct imsm_map *map; | |
4055 | ||
4056 | /* new failures are always set in map[0] */ | |
4057 | map = get_imsm_map(dev, 0); | |
4058 | ||
4059 | slot = get_imsm_disk_slot(map, idx); | |
4060 | if (slot < 0) | |
4061 | return 0; | |
4062 | ||
4063 | ord = __le32_to_cpu(map->disk_ord_tbl[slot]); | |
4064 | if (is_failed(disk) && (ord & IMSM_ORD_REBUILD)) | |
4065 | return 0; | |
4066 | ||
4067 | disk->status |= FAILED_DISK; | |
4068 | disk->status &= ~CONFIGURED_DISK; | |
4069 | set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD); | |
4070 | if (~map->failed_disk_num == 0) | |
4071 | map->failed_disk_num = slot; | |
4072 | return 1; | |
4073 | } | |
4074 | ||
4075 | static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx) | |
4076 | { | |
4077 | mark_failure(dev, disk, idx); | |
4078 | ||
4079 | if (disk->scsi_id == __cpu_to_le32(~(__u32)0)) | |
4080 | return; | |
4081 | ||
4082 | disk->scsi_id = __cpu_to_le32(~(__u32)0); | |
4083 | memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1); | |
4084 | } | |
4085 | ||
4086 | /* Handle dirty -> clean transititions and resync. Degraded and rebuild | |
4087 | * states are handled in imsm_set_disk() with one exception, when a | |
4088 | * resync is stopped due to a new failure this routine will set the | |
4089 | * 'degraded' state for the array. | |
4090 | */ | |
4091 | static int imsm_set_array_state(struct active_array *a, int consistent) | |
4092 | { | |
4093 | int inst = a->info.container_member; | |
4094 | struct intel_super *super = a->container->sb; | |
4095 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
4096 | struct imsm_map *map = get_imsm_map(dev, 0); | |
4097 | int failed = imsm_count_failed(super, dev); | |
4098 | __u8 map_state = imsm_check_degraded(super, dev, failed); | |
4099 | ||
4100 | /* before we activate this array handle any missing disks */ | |
4101 | if (consistent == 2 && super->missing) { | |
4102 | struct dl *dl; | |
4103 | ||
4104 | dprintf("imsm: mark missing\n"); | |
4105 | end_migration(dev, map_state); | |
4106 | for (dl = super->missing; dl; dl = dl->next) | |
4107 | mark_missing(dev, &dl->disk, dl->index); | |
4108 | super->updates_pending++; | |
4109 | } | |
4110 | ||
4111 | if (consistent == 2 && | |
4112 | (!is_resync_complete(&a->info) || | |
4113 | map_state != IMSM_T_STATE_NORMAL || | |
4114 | dev->vol.migr_state)) | |
4115 | consistent = 0; | |
4116 | ||
4117 | if (is_resync_complete(&a->info)) { | |
4118 | /* complete intialization / resync, | |
4119 | * recovery and interrupted recovery is completed in | |
4120 | * ->set_disk | |
4121 | */ | |
4122 | if (is_resyncing(dev)) { | |
4123 | dprintf("imsm: mark resync done\n"); | |
4124 | end_migration(dev, map_state); | |
4125 | super->updates_pending++; | |
4126 | } | |
4127 | } else if (!is_resyncing(dev) && !failed) { | |
4128 | /* mark the start of the init process if nothing is failed */ | |
4129 | dprintf("imsm: mark resync start\n"); | |
4130 | if (map->map_state == IMSM_T_STATE_UNINITIALIZED) | |
4131 | migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT); | |
4132 | else | |
4133 | migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REPAIR); | |
4134 | super->updates_pending++; | |
4135 | } | |
4136 | ||
4137 | /* FIXME check if we can update curr_migr_unit from resync_start */ | |
4138 | ||
4139 | /* mark dirty / clean */ | |
4140 | if (dev->vol.dirty != !consistent) { | |
4141 | dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty"); | |
4142 | if (consistent) | |
4143 | dev->vol.dirty = 0; | |
4144 | else | |
4145 | dev->vol.dirty = 1; | |
4146 | super->updates_pending++; | |
4147 | } | |
4148 | return consistent; | |
4149 | } | |
4150 | ||
4151 | static void imsm_set_disk(struct active_array *a, int n, int state) | |
4152 | { | |
4153 | int inst = a->info.container_member; | |
4154 | struct intel_super *super = a->container->sb; | |
4155 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
4156 | struct imsm_map *map = get_imsm_map(dev, 0); | |
4157 | struct imsm_disk *disk; | |
4158 | int failed; | |
4159 | __u32 ord; | |
4160 | __u8 map_state; | |
4161 | ||
4162 | if (n > map->num_members) | |
4163 | fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n", | |
4164 | n, map->num_members - 1); | |
4165 | ||
4166 | if (n < 0) | |
4167 | return; | |
4168 | ||
4169 | dprintf("imsm: set_disk %d:%x\n", n, state); | |
4170 | ||
4171 | ord = get_imsm_ord_tbl_ent(dev, n); | |
4172 | disk = get_imsm_disk(super, ord_to_idx(ord)); | |
4173 | ||
4174 | /* check for new failures */ | |
4175 | if (state & DS_FAULTY) { | |
4176 | if (mark_failure(dev, disk, ord_to_idx(ord))) | |
4177 | super->updates_pending++; | |
4178 | } | |
4179 | ||
4180 | /* check if in_sync */ | |
4181 | if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) { | |
4182 | struct imsm_map *migr_map = get_imsm_map(dev, 1); | |
4183 | ||
4184 | set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord)); | |
4185 | super->updates_pending++; | |
4186 | } | |
4187 | ||
4188 | failed = imsm_count_failed(super, dev); | |
4189 | map_state = imsm_check_degraded(super, dev, failed); | |
4190 | ||
4191 | /* check if recovery complete, newly degraded, or failed */ | |
4192 | if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) { | |
4193 | end_migration(dev, map_state); | |
4194 | map = get_imsm_map(dev, 0); | |
4195 | map->failed_disk_num = ~0; | |
4196 | super->updates_pending++; | |
4197 | } else if (map_state == IMSM_T_STATE_DEGRADED && | |
4198 | map->map_state != map_state && | |
4199 | !dev->vol.migr_state) { | |
4200 | dprintf("imsm: mark degraded\n"); | |
4201 | map->map_state = map_state; | |
4202 | super->updates_pending++; | |
4203 | } else if (map_state == IMSM_T_STATE_FAILED && | |
4204 | map->map_state != map_state) { | |
4205 | dprintf("imsm: mark failed\n"); | |
4206 | end_migration(dev, map_state); | |
4207 | super->updates_pending++; | |
4208 | } | |
4209 | } | |
4210 | ||
4211 | static int store_imsm_mpb(int fd, struct imsm_super *mpb) | |
4212 | { | |
4213 | void *buf = mpb; | |
4214 | __u32 mpb_size = __le32_to_cpu(mpb->mpb_size); | |
4215 | unsigned long long dsize; | |
4216 | unsigned long long sectors; | |
4217 | ||
4218 | get_dev_size(fd, NULL, &dsize); | |
4219 | ||
4220 | if (mpb_size > 512) { | |
4221 | /* -1 to account for anchor */ | |
4222 | sectors = mpb_sectors(mpb) - 1; | |
4223 | ||
4224 | /* write the extended mpb to the sectors preceeding the anchor */ | |
4225 | if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) | |
4226 | return 1; | |
4227 | ||
4228 | if (write(fd, buf + 512, 512 * sectors) != 512 * sectors) | |
4229 | return 1; | |
4230 | } | |
4231 | ||
4232 | /* first block is stored on second to last sector of the disk */ | |
4233 | if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) | |
4234 | return 1; | |
4235 | ||
4236 | if (write(fd, buf, 512) != 512) | |
4237 | return 1; | |
4238 | ||
4239 | return 0; | |
4240 | } | |
4241 | ||
4242 | static void imsm_sync_metadata(struct supertype *container) | |
4243 | { | |
4244 | struct intel_super *super = container->sb; | |
4245 | ||
4246 | if (!super->updates_pending) | |
4247 | return; | |
4248 | ||
4249 | write_super_imsm(super, 0); | |
4250 | ||
4251 | super->updates_pending = 0; | |
4252 | } | |
4253 | ||
4254 | static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a) | |
4255 | { | |
4256 | struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member); | |
4257 | int i = get_imsm_disk_idx(dev, idx); | |
4258 | struct dl *dl; | |
4259 | ||
4260 | for (dl = super->disks; dl; dl = dl->next) | |
4261 | if (dl->index == i) | |
4262 | break; | |
4263 | ||
4264 | if (dl && is_failed(&dl->disk)) | |
4265 | dl = NULL; | |
4266 | ||
4267 | if (dl) | |
4268 | dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor); | |
4269 | ||
4270 | return dl; | |
4271 | } | |
4272 | ||
4273 | static struct dl *imsm_add_spare(struct intel_super *super, int slot, | |
4274 | struct active_array *a, int activate_new) | |
4275 | { | |
4276 | struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member); | |
4277 | int idx = get_imsm_disk_idx(dev, slot); | |
4278 | struct imsm_super *mpb = super->anchor; | |
4279 | struct imsm_map *map; | |
4280 | unsigned long long pos; | |
4281 | struct mdinfo *d; | |
4282 | struct extent *ex; | |
4283 | int i, j; | |
4284 | int found; | |
4285 | __u32 array_start; | |
4286 | __u32 array_end; | |
4287 | struct dl *dl; | |
4288 | ||
4289 | for (dl = super->disks; dl; dl = dl->next) { | |
4290 | /* If in this array, skip */ | |
4291 | for (d = a->info.devs ; d ; d = d->next) | |
4292 | if (d->state_fd >= 0 && | |
4293 | d->disk.major == dl->major && | |
4294 | d->disk.minor == dl->minor) { | |
4295 | dprintf("%x:%x already in array\n", dl->major, dl->minor); | |
4296 | break; | |
4297 | } | |
4298 | if (d) | |
4299 | continue; | |
4300 | ||
4301 | /* skip in use or failed drives */ | |
4302 | if (is_failed(&dl->disk) || idx == dl->index || | |
4303 | dl->index == -2) { | |
4304 | dprintf("%x:%x status (failed: %d index: %d)\n", | |
4305 | dl->major, dl->minor, is_failed(&dl->disk), idx); | |
4306 | continue; | |
4307 | } | |
4308 | ||
4309 | /* skip pure spares when we are looking for partially | |
4310 | * assimilated drives | |
4311 | */ | |
4312 | if (dl->index == -1 && !activate_new) | |
4313 | continue; | |
4314 | ||
4315 | /* Does this unused device have the requisite free space? | |
4316 | * It needs to be able to cover all member volumes | |
4317 | */ | |
4318 | ex = get_extents(super, dl); | |
4319 | if (!ex) { | |
4320 | dprintf("cannot get extents\n"); | |
4321 | continue; | |
4322 | } | |
4323 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
4324 | dev = get_imsm_dev(super, i); | |
4325 | map = get_imsm_map(dev, 0); | |
4326 | ||
4327 | /* check if this disk is already a member of | |
4328 | * this array | |
4329 | */ | |
4330 | if (get_imsm_disk_slot(map, dl->index) >= 0) | |
4331 | continue; | |
4332 | ||
4333 | found = 0; | |
4334 | j = 0; | |
4335 | pos = 0; | |
4336 | array_start = __le32_to_cpu(map->pba_of_lba0); | |
4337 | array_end = array_start + | |
4338 | __le32_to_cpu(map->blocks_per_member) - 1; | |
4339 | ||
4340 | do { | |
4341 | /* check that we can start at pba_of_lba0 with | |
4342 | * blocks_per_member of space | |
4343 | */ | |
4344 | if (array_start >= pos && array_end < ex[j].start) { | |
4345 | found = 1; | |
4346 | break; | |
4347 | } | |
4348 | pos = ex[j].start + ex[j].size; | |
4349 | j++; | |
4350 | } while (ex[j-1].size); | |
4351 | ||
4352 | if (!found) | |
4353 | break; | |
4354 | } | |
4355 | ||
4356 | free(ex); | |
4357 | if (i < mpb->num_raid_devs) { | |
4358 | dprintf("%x:%x does not have %u to %u available\n", | |
4359 | dl->major, dl->minor, array_start, array_end); | |
4360 | /* No room */ | |
4361 | continue; | |
4362 | } | |
4363 | return dl; | |
4364 | } | |
4365 | ||
4366 | return dl; | |
4367 | } | |
4368 | ||
4369 | static struct mdinfo *imsm_activate_spare(struct active_array *a, | |
4370 | struct metadata_update **updates) | |
4371 | { | |
4372 | /** | |
4373 | * Find a device with unused free space and use it to replace a | |
4374 | * failed/vacant region in an array. We replace failed regions one a | |
4375 | * array at a time. The result is that a new spare disk will be added | |
4376 | * to the first failed array and after the monitor has finished | |
4377 | * propagating failures the remainder will be consumed. | |
4378 | * | |
4379 | * FIXME add a capability for mdmon to request spares from another | |
4380 | * container. | |
4381 | */ | |
4382 | ||
4383 | struct intel_super *super = a->container->sb; | |
4384 | int inst = a->info.container_member; | |
4385 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
4386 | struct imsm_map *map = get_imsm_map(dev, 0); | |
4387 | int failed = a->info.array.raid_disks; | |
4388 | struct mdinfo *rv = NULL; | |
4389 | struct mdinfo *d; | |
4390 | struct mdinfo *di; | |
4391 | struct metadata_update *mu; | |
4392 | struct dl *dl; | |
4393 | struct imsm_update_activate_spare *u; | |
4394 | int num_spares = 0; | |
4395 | int i; | |
4396 | ||
4397 | for (d = a->info.devs ; d ; d = d->next) { | |
4398 | if ((d->curr_state & DS_FAULTY) && | |
4399 | d->state_fd >= 0) | |
4400 | /* wait for Removal to happen */ | |
4401 | return NULL; | |
4402 | if (d->state_fd >= 0) | |
4403 | failed--; | |
4404 | } | |
4405 | ||
4406 | dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n", | |
4407 | inst, failed, a->info.array.raid_disks, a->info.array.level); | |
4408 | if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED) | |
4409 | return NULL; | |
4410 | ||
4411 | /* For each slot, if it is not working, find a spare */ | |
4412 | for (i = 0; i < a->info.array.raid_disks; i++) { | |
4413 | for (d = a->info.devs ; d ; d = d->next) | |
4414 | if (d->disk.raid_disk == i) | |
4415 | break; | |
4416 | dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0); | |
4417 | if (d && (d->state_fd >= 0)) | |
4418 | continue; | |
4419 | ||
4420 | /* | |
4421 | * OK, this device needs recovery. Try to re-add the | |
4422 | * previous occupant of this slot, if this fails see if | |
4423 | * we can continue the assimilation of a spare that was | |
4424 | * partially assimilated, finally try to activate a new | |
4425 | * spare. | |
4426 | */ | |
4427 | dl = imsm_readd(super, i, a); | |
4428 | if (!dl) | |
4429 | dl = imsm_add_spare(super, i, a, 0); | |
4430 | if (!dl) | |
4431 | dl = imsm_add_spare(super, i, a, 1); | |
4432 | if (!dl) | |
4433 | continue; | |
4434 | ||
4435 | /* found a usable disk with enough space */ | |
4436 | di = malloc(sizeof(*di)); | |
4437 | if (!di) | |
4438 | continue; | |
4439 | memset(di, 0, sizeof(*di)); | |
4440 | ||
4441 | /* dl->index will be -1 in the case we are activating a | |
4442 | * pristine spare. imsm_process_update() will create a | |
4443 | * new index in this case. Once a disk is found to be | |
4444 | * failed in all member arrays it is kicked from the | |
4445 | * metadata | |
4446 | */ | |
4447 | di->disk.number = dl->index; | |
4448 | ||
4449 | /* (ab)use di->devs to store a pointer to the device | |
4450 | * we chose | |
4451 | */ | |
4452 | di->devs = (struct mdinfo *) dl; | |
4453 | ||
4454 | di->disk.raid_disk = i; | |
4455 | di->disk.major = dl->major; | |
4456 | di->disk.minor = dl->minor; | |
4457 | di->disk.state = 0; | |
4458 | di->recovery_start = 0; | |
4459 | di->data_offset = __le32_to_cpu(map->pba_of_lba0); | |
4460 | di->component_size = a->info.component_size; | |
4461 | di->container_member = inst; | |
4462 | super->random = random32(); | |
4463 | di->next = rv; | |
4464 | rv = di; | |
4465 | num_spares++; | |
4466 | dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor, | |
4467 | i, di->data_offset); | |
4468 | ||
4469 | break; | |
4470 | } | |
4471 | ||
4472 | if (!rv) | |
4473 | /* No spares found */ | |
4474 | return rv; | |
4475 | /* Now 'rv' has a list of devices to return. | |
4476 | * Create a metadata_update record to update the | |
4477 | * disk_ord_tbl for the array | |
4478 | */ | |
4479 | mu = malloc(sizeof(*mu)); | |
4480 | if (mu) { | |
4481 | mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares); | |
4482 | if (mu->buf == NULL) { | |
4483 | free(mu); | |
4484 | mu = NULL; | |
4485 | } | |
4486 | } | |
4487 | if (!mu) { | |
4488 | while (rv) { | |
4489 | struct mdinfo *n = rv->next; | |
4490 | ||
4491 | free(rv); | |
4492 | rv = n; | |
4493 | } | |
4494 | return NULL; | |
4495 | } | |
4496 | ||
4497 | mu->space = NULL; | |
4498 | mu->len = sizeof(struct imsm_update_activate_spare) * num_spares; | |
4499 | mu->next = *updates; | |
4500 | u = (struct imsm_update_activate_spare *) mu->buf; | |
4501 | ||
4502 | for (di = rv ; di ; di = di->next) { | |
4503 | u->type = update_activate_spare; | |
4504 | u->dl = (struct dl *) di->devs; | |
4505 | di->devs = NULL; | |
4506 | u->slot = di->disk.raid_disk; | |
4507 | u->array = inst; | |
4508 | u->next = u + 1; | |
4509 | u++; | |
4510 | } | |
4511 | (u-1)->next = NULL; | |
4512 | *updates = mu; | |
4513 | ||
4514 | return rv; | |
4515 | } | |
4516 | ||
4517 | static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u) | |
4518 | { | |
4519 | struct imsm_dev *dev = get_imsm_dev(super, idx); | |
4520 | struct imsm_map *map = get_imsm_map(dev, 0); | |
4521 | struct imsm_map *new_map = get_imsm_map(&u->dev, 0); | |
4522 | struct disk_info *inf = get_disk_info(u); | |
4523 | struct imsm_disk *disk; | |
4524 | int i; | |
4525 | int j; | |
4526 | ||
4527 | for (i = 0; i < map->num_members; i++) { | |
4528 | disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i)); | |
4529 | for (j = 0; j < new_map->num_members; j++) | |
4530 | if (serialcmp(disk->serial, inf[j].serial) == 0) | |
4531 | return 1; | |
4532 | } | |
4533 | ||
4534 | return 0; | |
4535 | } | |
4536 | ||
4537 | static void imsm_delete(struct intel_super *super, struct dl **dlp, int index); | |
4538 | ||
4539 | static void imsm_process_update(struct supertype *st, | |
4540 | struct metadata_update *update) | |
4541 | { | |
4542 | /** | |
4543 | * crack open the metadata_update envelope to find the update record | |
4544 | * update can be one of: | |
4545 | * update_activate_spare - a spare device has replaced a failed | |
4546 | * device in an array, update the disk_ord_tbl. If this disk is | |
4547 | * present in all member arrays then also clear the SPARE_DISK | |
4548 | * flag | |
4549 | */ | |
4550 | struct intel_super *super = st->sb; | |
4551 | struct imsm_super *mpb; | |
4552 | enum imsm_update_type type = *(enum imsm_update_type *) update->buf; | |
4553 | ||
4554 | /* update requires a larger buf but the allocation failed */ | |
4555 | if (super->next_len && !super->next_buf) { | |
4556 | super->next_len = 0; | |
4557 | return; | |
4558 | } | |
4559 | ||
4560 | if (super->next_buf) { | |
4561 | memcpy(super->next_buf, super->buf, super->len); | |
4562 | free(super->buf); | |
4563 | super->len = super->next_len; | |
4564 | super->buf = super->next_buf; | |
4565 | ||
4566 | super->next_len = 0; | |
4567 | super->next_buf = NULL; | |
4568 | } | |
4569 | ||
4570 | mpb = super->anchor; | |
4571 | ||
4572 | switch (type) { | |
4573 | case update_activate_spare: { | |
4574 | struct imsm_update_activate_spare *u = (void *) update->buf; | |
4575 | struct imsm_dev *dev = get_imsm_dev(super, u->array); | |
4576 | struct imsm_map *map = get_imsm_map(dev, 0); | |
4577 | struct imsm_map *migr_map; | |
4578 | struct active_array *a; | |
4579 | struct imsm_disk *disk; | |
4580 | __u8 to_state; | |
4581 | struct dl *dl; | |
4582 | unsigned int found; | |
4583 | int failed; | |
4584 | int victim = get_imsm_disk_idx(dev, u->slot); | |
4585 | int i; | |
4586 | ||
4587 | for (dl = super->disks; dl; dl = dl->next) | |
4588 | if (dl == u->dl) | |
4589 | break; | |
4590 | ||
4591 | if (!dl) { | |
4592 | fprintf(stderr, "error: imsm_activate_spare passed " | |
4593 | "an unknown disk (index: %d)\n", | |
4594 | u->dl->index); | |
4595 | return; | |
4596 | } | |
4597 | ||
4598 | super->updates_pending++; | |
4599 | ||
4600 | /* count failures (excluding rebuilds and the victim) | |
4601 | * to determine map[0] state | |
4602 | */ | |
4603 | failed = 0; | |
4604 | for (i = 0; i < map->num_members; i++) { | |
4605 | if (i == u->slot) | |
4606 | continue; | |
4607 | disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i)); | |
4608 | if (!disk || is_failed(disk)) | |
4609 | failed++; | |
4610 | } | |
4611 | ||
4612 | /* adding a pristine spare, assign a new index */ | |
4613 | if (dl->index < 0) { | |
4614 | dl->index = super->anchor->num_disks; | |
4615 | super->anchor->num_disks++; | |
4616 | } | |
4617 | disk = &dl->disk; | |
4618 | disk->status |= CONFIGURED_DISK; | |
4619 | disk->status &= ~SPARE_DISK; | |
4620 | ||
4621 | /* mark rebuild */ | |
4622 | to_state = imsm_check_degraded(super, dev, failed); | |
4623 | map->map_state = IMSM_T_STATE_DEGRADED; | |
4624 | migrate(dev, to_state, MIGR_REBUILD); | |
4625 | migr_map = get_imsm_map(dev, 1); | |
4626 | set_imsm_ord_tbl_ent(map, u->slot, dl->index); | |
4627 | set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD); | |
4628 | ||
4629 | /* update the family_num to mark a new container | |
4630 | * generation, being careful to record the existing | |
4631 | * family_num in orig_family_num to clean up after | |
4632 | * earlier mdadm versions that neglected to set it. | |
4633 | */ | |
4634 | if (mpb->orig_family_num == 0) | |
4635 | mpb->orig_family_num = mpb->family_num; | |
4636 | mpb->family_num += super->random; | |
4637 | ||
4638 | /* count arrays using the victim in the metadata */ | |
4639 | found = 0; | |
4640 | for (a = st->arrays; a ; a = a->next) { | |
4641 | dev = get_imsm_dev(super, a->info.container_member); | |
4642 | map = get_imsm_map(dev, 0); | |
4643 | ||
4644 | if (get_imsm_disk_slot(map, victim) >= 0) | |
4645 | found++; | |
4646 | } | |
4647 | ||
4648 | /* delete the victim if it is no longer being | |
4649 | * utilized anywhere | |
4650 | */ | |
4651 | if (!found) { | |
4652 | struct dl **dlp; | |
4653 | ||
4654 | /* We know that 'manager' isn't touching anything, | |
4655 | * so it is safe to delete | |
4656 | */ | |
4657 | for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next) | |
4658 | if ((*dlp)->index == victim) | |
4659 | break; | |
4660 | ||
4661 | /* victim may be on the missing list */ | |
4662 | if (!*dlp) | |
4663 | for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next) | |
4664 | if ((*dlp)->index == victim) | |
4665 | break; | |
4666 | imsm_delete(super, dlp, victim); | |
4667 | } | |
4668 | break; | |
4669 | } | |
4670 | case update_create_array: { | |
4671 | /* someone wants to create a new array, we need to be aware of | |
4672 | * a few races/collisions: | |
4673 | * 1/ 'Create' called by two separate instances of mdadm | |
4674 | * 2/ 'Create' versus 'activate_spare': mdadm has chosen | |
4675 | * devices that have since been assimilated via | |
4676 | * activate_spare. | |
4677 | * In the event this update can not be carried out mdadm will | |
4678 | * (FIX ME) notice that its update did not take hold. | |
4679 | */ | |
4680 | struct imsm_update_create_array *u = (void *) update->buf; | |
4681 | struct intel_dev *dv; | |
4682 | struct imsm_dev *dev; | |
4683 | struct imsm_map *map, *new_map; | |
4684 | unsigned long long start, end; | |
4685 | unsigned long long new_start, new_end; | |
4686 | int i; | |
4687 | struct disk_info *inf; | |
4688 | struct dl *dl; | |
4689 | ||
4690 | /* handle racing creates: first come first serve */ | |
4691 | if (u->dev_idx < mpb->num_raid_devs) { | |
4692 | dprintf("%s: subarray %d already defined\n", | |
4693 | __func__, u->dev_idx); | |
4694 | goto create_error; | |
4695 | } | |
4696 | ||
4697 | /* check update is next in sequence */ | |
4698 | if (u->dev_idx != mpb->num_raid_devs) { | |
4699 | dprintf("%s: can not create array %d expected index %d\n", | |
4700 | __func__, u->dev_idx, mpb->num_raid_devs); | |
4701 | goto create_error; | |
4702 | } | |
4703 | ||
4704 | new_map = get_imsm_map(&u->dev, 0); | |
4705 | new_start = __le32_to_cpu(new_map->pba_of_lba0); | |
4706 | new_end = new_start + __le32_to_cpu(new_map->blocks_per_member); | |
4707 | inf = get_disk_info(u); | |
4708 | ||
4709 | /* handle activate_spare versus create race: | |
4710 | * check to make sure that overlapping arrays do not include | |
4711 | * overalpping disks | |
4712 | */ | |
4713 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
4714 | dev = get_imsm_dev(super, i); | |
4715 | map = get_imsm_map(dev, 0); | |
4716 | start = __le32_to_cpu(map->pba_of_lba0); | |
4717 | end = start + __le32_to_cpu(map->blocks_per_member); | |
4718 | if ((new_start >= start && new_start <= end) || | |
4719 | (start >= new_start && start <= new_end)) | |
4720 | /* overlap */; | |
4721 | else | |
4722 | continue; | |
4723 | ||
4724 | if (disks_overlap(super, i, u)) { | |
4725 | dprintf("%s: arrays overlap\n", __func__); | |
4726 | goto create_error; | |
4727 | } | |
4728 | } | |
4729 | ||
4730 | /* check that prepare update was successful */ | |
4731 | if (!update->space) { | |
4732 | dprintf("%s: prepare update failed\n", __func__); | |
4733 | goto create_error; | |
4734 | } | |
4735 | ||
4736 | /* check that all disks are still active before committing | |
4737 | * changes. FIXME: could we instead handle this by creating a | |
4738 | * degraded array? That's probably not what the user expects, | |
4739 | * so better to drop this update on the floor. | |
4740 | */ | |
4741 | for (i = 0; i < new_map->num_members; i++) { | |
4742 | dl = serial_to_dl(inf[i].serial, super); | |
4743 | if (!dl) { | |
4744 | dprintf("%s: disk disappeared\n", __func__); | |
4745 | goto create_error; | |
4746 | } | |
4747 | } | |
4748 | ||
4749 | super->updates_pending++; | |
4750 | ||
4751 | /* convert spares to members and fixup ord_tbl */ | |
4752 | for (i = 0; i < new_map->num_members; i++) { | |
4753 | dl = serial_to_dl(inf[i].serial, super); | |
4754 | if (dl->index == -1) { | |
4755 | dl->index = mpb->num_disks; | |
4756 | mpb->num_disks++; | |
4757 | dl->disk.status |= CONFIGURED_DISK; | |
4758 | dl->disk.status &= ~SPARE_DISK; | |
4759 | } | |
4760 | set_imsm_ord_tbl_ent(new_map, i, dl->index); | |
4761 | } | |
4762 | ||
4763 | dv = update->space; | |
4764 | dev = dv->dev; | |
4765 | update->space = NULL; | |
4766 | imsm_copy_dev(dev, &u->dev); | |
4767 | dv->index = u->dev_idx; | |
4768 | dv->next = super->devlist; | |
4769 | super->devlist = dv; | |
4770 | mpb->num_raid_devs++; | |
4771 | ||
4772 | imsm_update_version_info(super); | |
4773 | break; | |
4774 | create_error: | |
4775 | /* mdmon knows how to release update->space, but not | |
4776 | * ((struct intel_dev *) update->space)->dev | |
4777 | */ | |
4778 | if (update->space) { | |
4779 | dv = update->space; | |
4780 | free(dv->dev); | |
4781 | } | |
4782 | break; | |
4783 | } | |
4784 | case update_add_disk: | |
4785 | ||
4786 | /* we may be able to repair some arrays if disks are | |
4787 | * being added */ | |
4788 | if (super->add) { | |
4789 | struct active_array *a; | |
4790 | ||
4791 | super->updates_pending++; | |
4792 | for (a = st->arrays; a; a = a->next) | |
4793 | a->check_degraded = 1; | |
4794 | } | |
4795 | /* add some spares to the metadata */ | |
4796 | while (super->add) { | |
4797 | struct dl *al; | |
4798 | ||
4799 | al = super->add; | |
4800 | super->add = al->next; | |
4801 | al->next = super->disks; | |
4802 | super->disks = al; | |
4803 | dprintf("%s: added %x:%x\n", | |
4804 | __func__, al->major, al->minor); | |
4805 | } | |
4806 | ||
4807 | break; | |
4808 | } | |
4809 | } | |
4810 | ||
4811 | static void imsm_prepare_update(struct supertype *st, | |
4812 | struct metadata_update *update) | |
4813 | { | |
4814 | /** | |
4815 | * Allocate space to hold new disk entries, raid-device entries or a new | |
4816 | * mpb if necessary. The manager synchronously waits for updates to | |
4817 | * complete in the monitor, so new mpb buffers allocated here can be | |
4818 | * integrated by the monitor thread without worrying about live pointers | |
4819 | * in the manager thread. | |
4820 | */ | |
4821 | enum imsm_update_type type = *(enum imsm_update_type *) update->buf; | |
4822 | struct intel_super *super = st->sb; | |
4823 | struct imsm_super *mpb = super->anchor; | |
4824 | size_t buf_len; | |
4825 | size_t len = 0; | |
4826 | ||
4827 | switch (type) { | |
4828 | case update_create_array: { | |
4829 | struct imsm_update_create_array *u = (void *) update->buf; | |
4830 | struct intel_dev *dv; | |
4831 | struct imsm_dev *dev = &u->dev; | |
4832 | struct imsm_map *map = get_imsm_map(dev, 0); | |
4833 | struct dl *dl; | |
4834 | struct disk_info *inf; | |
4835 | int i; | |
4836 | int activate = 0; | |
4837 | ||
4838 | inf = get_disk_info(u); | |
4839 | len = sizeof_imsm_dev(dev, 1); | |
4840 | /* allocate a new super->devlist entry */ | |
4841 | dv = malloc(sizeof(*dv)); | |
4842 | if (dv) { | |
4843 | dv->dev = malloc(len); | |
4844 | if (dv->dev) | |
4845 | update->space = dv; | |
4846 | else { | |
4847 | free(dv); | |
4848 | update->space = NULL; | |
4849 | } | |
4850 | } | |
4851 | ||
4852 | /* count how many spares will be converted to members */ | |
4853 | for (i = 0; i < map->num_members; i++) { | |
4854 | dl = serial_to_dl(inf[i].serial, super); | |
4855 | if (!dl) { | |
4856 | /* hmm maybe it failed?, nothing we can do about | |
4857 | * it here | |
4858 | */ | |
4859 | continue; | |
4860 | } | |
4861 | if (count_memberships(dl, super) == 0) | |
4862 | activate++; | |
4863 | } | |
4864 | len += activate * sizeof(struct imsm_disk); | |
4865 | break; | |
4866 | default: | |
4867 | break; | |
4868 | } | |
4869 | } | |
4870 | ||
4871 | /* check if we need a larger metadata buffer */ | |
4872 | if (super->next_buf) | |
4873 | buf_len = super->next_len; | |
4874 | else | |
4875 | buf_len = super->len; | |
4876 | ||
4877 | if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) { | |
4878 | /* ok we need a larger buf than what is currently allocated | |
4879 | * if this allocation fails process_update will notice that | |
4880 | * ->next_len is set and ->next_buf is NULL | |
4881 | */ | |
4882 | buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512); | |
4883 | if (super->next_buf) | |
4884 | free(super->next_buf); | |
4885 | ||
4886 | super->next_len = buf_len; | |
4887 | if (posix_memalign(&super->next_buf, 512, buf_len) == 0) | |
4888 | memset(super->next_buf, 0, buf_len); | |
4889 | else | |
4890 | super->next_buf = NULL; | |
4891 | } | |
4892 | } | |
4893 | ||
4894 | /* must be called while manager is quiesced */ | |
4895 | static void imsm_delete(struct intel_super *super, struct dl **dlp, int index) | |
4896 | { | |
4897 | struct imsm_super *mpb = super->anchor; | |
4898 | struct dl *iter; | |
4899 | struct imsm_dev *dev; | |
4900 | struct imsm_map *map; | |
4901 | int i, j, num_members; | |
4902 | __u32 ord; | |
4903 | ||
4904 | dprintf("%s: deleting device[%d] from imsm_super\n", | |
4905 | __func__, index); | |
4906 | ||
4907 | /* shift all indexes down one */ | |
4908 | for (iter = super->disks; iter; iter = iter->next) | |
4909 | if (iter->index > index) | |
4910 | iter->index--; | |
4911 | for (iter = super->missing; iter; iter = iter->next) | |
4912 | if (iter->index > index) | |
4913 | iter->index--; | |
4914 | ||
4915 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
4916 | dev = get_imsm_dev(super, i); | |
4917 | map = get_imsm_map(dev, 0); | |
4918 | num_members = map->num_members; | |
4919 | for (j = 0; j < num_members; j++) { | |
4920 | /* update ord entries being careful not to propagate | |
4921 | * ord-flags to the first map | |
4922 | */ | |
4923 | ord = get_imsm_ord_tbl_ent(dev, j); | |
4924 | ||
4925 | if (ord_to_idx(ord) <= index) | |
4926 | continue; | |
4927 | ||
4928 | map = get_imsm_map(dev, 0); | |
4929 | set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1)); | |
4930 | map = get_imsm_map(dev, 1); | |
4931 | if (map) | |
4932 | set_imsm_ord_tbl_ent(map, j, ord - 1); | |
4933 | } | |
4934 | } | |
4935 | ||
4936 | mpb->num_disks--; | |
4937 | super->updates_pending++; | |
4938 | if (*dlp) { | |
4939 | struct dl *dl = *dlp; | |
4940 | ||
4941 | *dlp = (*dlp)->next; | |
4942 | __free_imsm_disk(dl); | |
4943 | } | |
4944 | } | |
4945 | #endif /* MDASSEMBLE */ | |
4946 | ||
4947 | struct superswitch super_imsm = { | |
4948 | #ifndef MDASSEMBLE | |
4949 | .examine_super = examine_super_imsm, | |
4950 | .brief_examine_super = brief_examine_super_imsm, | |
4951 | .brief_examine_subarrays = brief_examine_subarrays_imsm, | |
4952 | .export_examine_super = export_examine_super_imsm, | |
4953 | .detail_super = detail_super_imsm, | |
4954 | .brief_detail_super = brief_detail_super_imsm, | |
4955 | .write_init_super = write_init_super_imsm, | |
4956 | .validate_geometry = validate_geometry_imsm, | |
4957 | .add_to_super = add_to_super_imsm, | |
4958 | .detail_platform = detail_platform_imsm, | |
4959 | #endif | |
4960 | .match_home = match_home_imsm, | |
4961 | .uuid_from_super= uuid_from_super_imsm, | |
4962 | .getinfo_super = getinfo_super_imsm, | |
4963 | .update_super = update_super_imsm, | |
4964 | ||
4965 | .avail_size = avail_size_imsm, | |
4966 | ||
4967 | .compare_super = compare_super_imsm, | |
4968 | ||
4969 | .load_super = load_super_imsm, | |
4970 | .init_super = init_super_imsm, | |
4971 | .store_super = store_super_imsm, | |
4972 | .free_super = free_super_imsm, | |
4973 | .match_metadata_desc = match_metadata_desc_imsm, | |
4974 | .container_content = container_content_imsm, | |
4975 | .default_layout = imsm_level_to_layout, | |
4976 | ||
4977 | .external = 1, | |
4978 | .name = "imsm", | |
4979 | ||
4980 | #ifndef MDASSEMBLE | |
4981 | /* for mdmon */ | |
4982 | .open_new = imsm_open_new, | |
4983 | .load_super = load_super_imsm, | |
4984 | .set_array_state= imsm_set_array_state, | |
4985 | .set_disk = imsm_set_disk, | |
4986 | .sync_metadata = imsm_sync_metadata, | |
4987 | .activate_spare = imsm_activate_spare, | |
4988 | .process_update = imsm_process_update, | |
4989 | .prepare_update = imsm_prepare_update, | |
4990 | #endif /* MDASSEMBLE */ | |
4991 | }; |