]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de> | |
5 | * | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | * Author: Neil Brown | |
22 | * Email: <neilb@suse.de> | |
23 | */ | |
24 | ||
25 | #include <stddef.h> | |
26 | #include "mdadm.h" | |
27 | /* | |
28 | * The version-1 superblock : | |
29 | * All numeric fields are little-endian. | |
30 | * | |
31 | * total size: 256 bytes plus 2 per device. | |
32 | * 1K allows 384 devices. | |
33 | */ | |
34 | struct mdp_superblock_1 { | |
35 | /* constant array information - 128 bytes */ | |
36 | __u32 magic; /* MD_SB_MAGIC: 0xa92b4efc - little endian */ | |
37 | __u32 major_version; /* 1 */ | |
38 | __u32 feature_map; /* 0 for now */ | |
39 | __u32 pad0; /* always set to 0 when writing */ | |
40 | ||
41 | __u8 set_uuid[16]; /* user-space generated. */ | |
42 | char set_name[32]; /* set and interpreted by user-space */ | |
43 | ||
44 | __u64 ctime; /* lo 40 bits are seconds, top 24 are microseconds or 0*/ | |
45 | __u32 level; /* -4 (multipath), -1 (linear), 0,1,4,5 */ | |
46 | __u32 layout; /* only for raid5 currently */ | |
47 | __u64 size; /* used size of component devices, in 512byte sectors */ | |
48 | ||
49 | __u32 chunksize; /* in 512byte sectors */ | |
50 | __u32 raid_disks; | |
51 | __u32 bitmap_offset; /* sectors after start of superblock that bitmap starts | |
52 | * NOTE: signed, so bitmap can be before superblock | |
53 | * only meaningful of feature_map[0] is set. | |
54 | */ | |
55 | ||
56 | /* These are only valid with feature bit '4' */ | |
57 | __u32 new_level; /* new level we are reshaping to */ | |
58 | __u64 reshape_position; /* next address in array-space for reshape */ | |
59 | __u32 delta_disks; /* change in number of raid_disks */ | |
60 | __u32 new_layout; /* new layout */ | |
61 | __u32 new_chunk; /* new chunk size (sectors) */ | |
62 | __u32 new_offset; /* signed number to add to data_offset in new | |
63 | * layout. 0 == no-change. This can be | |
64 | * different on each device in the array. | |
65 | */ | |
66 | ||
67 | /* constant this-device information - 64 bytes */ | |
68 | __u64 data_offset; /* sector start of data, often 0 */ | |
69 | __u64 data_size; /* sectors in this device that can be used for data */ | |
70 | __u64 super_offset; /* sector start of this superblock */ | |
71 | __u64 recovery_offset;/* sectors before this offset (from data_offset) have been recovered */ | |
72 | __u32 dev_number; /* permanent identifier of this device - not role in raid */ | |
73 | __u32 cnt_corrected_read; /* number of read errors that were corrected by re-writing */ | |
74 | __u8 device_uuid[16]; /* user-space setable, ignored by kernel */ | |
75 | __u8 devflags; /* per-device flags. Only one defined...*/ | |
76 | #define WriteMostly1 1 /* mask for writemostly flag in above */ | |
77 | /* bad block log. If there are any bad blocks the feature flag is set. | |
78 | * if offset and size are non-zero, that space is reserved and available. | |
79 | */ | |
80 | __u8 bblog_shift; /* shift from sectors to block size for badblocklist */ | |
81 | __u16 bblog_size; /* number of sectors reserved for badblocklist */ | |
82 | __u32 bblog_offset; /* sector offset from superblock to bblog, signed */ | |
83 | ||
84 | /* array state information - 64 bytes */ | |
85 | __u64 utime; /* 40 bits second, 24 btes microseconds */ | |
86 | __u64 events; /* incremented when superblock updated */ | |
87 | __u64 resync_offset; /* data before this offset (from data_offset) known to be in sync */ | |
88 | __u32 sb_csum; /* checksum upto dev_roles[max_dev] */ | |
89 | __u32 max_dev; /* size of dev_roles[] array to consider */ | |
90 | __u8 pad3[64-32]; /* set to 0 when writing */ | |
91 | ||
92 | /* device state information. Indexed by dev_number. | |
93 | * 2 bytes per device | |
94 | * Note there are no per-device state flags. State information is rolled | |
95 | * into the 'roles' value. If a device is spare or faulty, then it doesn't | |
96 | * have a meaningful role. | |
97 | */ | |
98 | __u16 dev_roles[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */ | |
99 | }; | |
100 | ||
101 | #define MAX_SB_SIZE 4096 | |
102 | /* bitmap super size is 256, but we round up to a sector for alignment */ | |
103 | #define BM_SUPER_SIZE 512 | |
104 | #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2) | |
105 | #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \ | |
106 | + sizeof(struct misc_dev_info)) | |
107 | ||
108 | struct misc_dev_info { | |
109 | __u64 device_size; | |
110 | }; | |
111 | ||
112 | /* feature_map bits */ | |
113 | #define MD_FEATURE_BITMAP_OFFSET 1 | |
114 | #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and | |
115 | * must be honoured | |
116 | */ | |
117 | #define MD_FEATURE_RESHAPE_ACTIVE 4 | |
118 | #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */ | |
119 | #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an | |
120 | * active device with same 'role'. | |
121 | * 'recovery_offset' is also set. | |
122 | */ | |
123 | #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number | |
124 | * of devices, but is going | |
125 | * backwards anyway. | |
126 | */ | |
127 | #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */ | |
128 | #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \ | |
129 | |MD_FEATURE_RECOVERY_OFFSET \ | |
130 | |MD_FEATURE_RESHAPE_ACTIVE \ | |
131 | |MD_FEATURE_BAD_BLOCKS \ | |
132 | |MD_FEATURE_REPLACEMENT \ | |
133 | |MD_FEATURE_RESHAPE_BACKWARDS \ | |
134 | |MD_FEATURE_NEW_OFFSET \ | |
135 | ) | |
136 | ||
137 | static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb) | |
138 | { | |
139 | unsigned int disk_csum, csum; | |
140 | unsigned long long newcsum; | |
141 | int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2; | |
142 | unsigned int *isuper = (unsigned int*)sb; | |
143 | ||
144 | /* make sure I can count... */ | |
145 | if (offsetof(struct mdp_superblock_1,data_offset) != 128 || | |
146 | offsetof(struct mdp_superblock_1, utime) != 192 || | |
147 | sizeof(struct mdp_superblock_1) != 256) { | |
148 | fprintf(stderr, "WARNING - superblock isn't sized correctly\n"); | |
149 | } | |
150 | ||
151 | disk_csum = sb->sb_csum; | |
152 | sb->sb_csum = 0; | |
153 | newcsum = 0; | |
154 | for (; size>=4; size -= 4 ) { | |
155 | newcsum += __le32_to_cpu(*isuper); | |
156 | isuper++; | |
157 | } | |
158 | ||
159 | if (size == 2) | |
160 | newcsum += __le16_to_cpu(*(unsigned short*) isuper); | |
161 | ||
162 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
163 | sb->sb_csum = disk_csum; | |
164 | return __cpu_to_le32(csum); | |
165 | } | |
166 | ||
167 | /* | |
168 | * Information related to file descriptor used for aligned reads/writes. | |
169 | * Cache the block size. | |
170 | */ | |
171 | struct align_fd { | |
172 | int fd; | |
173 | int blk_sz; | |
174 | }; | |
175 | ||
176 | static void init_afd(struct align_fd *afd, int fd) | |
177 | { | |
178 | afd->fd = fd; | |
179 | ||
180 | if (ioctl(afd->fd, BLKSSZGET, &afd->blk_sz) != 0) | |
181 | afd->blk_sz = 512; | |
182 | } | |
183 | ||
184 | static char abuf[4096+4096]; | |
185 | static int aread(struct align_fd *afd, void *buf, int len) | |
186 | { | |
187 | /* aligned read. | |
188 | * On devices with a 4K sector size, we need to read | |
189 | * the full sector and copy relevant bits into | |
190 | * the buffer | |
191 | */ | |
192 | int bsize, iosize; | |
193 | char *b; | |
194 | int n; | |
195 | ||
196 | bsize = afd->blk_sz; | |
197 | ||
198 | if (!bsize || bsize > 4096 || len > 4096) { | |
199 | if (!bsize) | |
200 | fprintf(stderr, "WARNING - aread() called with invalid block size\n"); | |
201 | return -1; | |
202 | } | |
203 | b = ROUND_UP_PTR((char *)abuf, 4096); | |
204 | ||
205 | for (iosize = 0; iosize < len; iosize += bsize) | |
206 | ; | |
207 | n = read(afd->fd, b, iosize); | |
208 | if (n <= 0) | |
209 | return n; | |
210 | lseek(afd->fd, len - n, 1); | |
211 | if (n > len) | |
212 | n = len; | |
213 | memcpy(buf, b, n); | |
214 | return n; | |
215 | } | |
216 | ||
217 | static int awrite(struct align_fd *afd, void *buf, int len) | |
218 | { | |
219 | /* aligned write. | |
220 | * On devices with a 4K sector size, we need to write | |
221 | * the full sector. We pre-read if the sector is larger | |
222 | * than the write. | |
223 | * The address must be sector-aligned. | |
224 | */ | |
225 | int bsize, iosize; | |
226 | char *b; | |
227 | int n; | |
228 | ||
229 | bsize = afd->blk_sz; | |
230 | if (!bsize || bsize > 4096 || len > 4096) { | |
231 | if (!bsize) | |
232 | fprintf(stderr, "WARNING - awrite() called with invalid block size\n"); | |
233 | return -1; | |
234 | } | |
235 | b = ROUND_UP_PTR((char *)abuf, 4096); | |
236 | ||
237 | for (iosize = 0; iosize < len ; iosize += bsize) | |
238 | ; | |
239 | ||
240 | if (len != iosize) { | |
241 | n = read(afd->fd, b, iosize); | |
242 | if (n <= 0) | |
243 | return n; | |
244 | lseek(afd->fd, -n, 1); | |
245 | } | |
246 | ||
247 | memcpy(b, buf, len); | |
248 | n = write(afd->fd, b, iosize); | |
249 | if (n <= 0) | |
250 | return n; | |
251 | lseek(afd->fd, len - n, 1); | |
252 | return len; | |
253 | } | |
254 | ||
255 | #ifndef MDASSEMBLE | |
256 | static void examine_super1(struct supertype *st, char *homehost) | |
257 | { | |
258 | struct mdp_superblock_1 *sb = st->sb; | |
259 | time_t atime; | |
260 | unsigned int d; | |
261 | int role; | |
262 | int delta_extra = 0; | |
263 | int i; | |
264 | char *c; | |
265 | int l = homehost ? strlen(homehost) : 0; | |
266 | int layout; | |
267 | unsigned long long sb_offset; | |
268 | struct mdinfo info; | |
269 | ||
270 | printf(" Magic : %08x\n", __le32_to_cpu(sb->magic)); | |
271 | printf(" Version : 1"); | |
272 | sb_offset = __le64_to_cpu(sb->super_offset); | |
273 | if (sb_offset <= 4) | |
274 | printf(".1\n"); | |
275 | else if (sb_offset <= 8) | |
276 | printf(".2\n"); | |
277 | else | |
278 | printf(".0\n"); | |
279 | printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb->feature_map)); | |
280 | printf(" Array UUID : "); | |
281 | for (i=0; i<16; i++) { | |
282 | if ((i&3)==0 && i != 0) printf(":"); | |
283 | printf("%02x", sb->set_uuid[i]); | |
284 | } | |
285 | printf("\n"); | |
286 | printf(" Name : %.32s", sb->set_name); | |
287 | if (l > 0 && l < 32 && | |
288 | sb->set_name[l] == ':' && | |
289 | strncmp(sb->set_name, homehost, l) == 0) | |
290 | printf(" (local to host %s)", homehost); | |
291 | printf("\n"); | |
292 | atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL; | |
293 | printf(" Creation Time : %.24s\n", ctime(&atime)); | |
294 | c=map_num(pers, __le32_to_cpu(sb->level)); | |
295 | printf(" Raid Level : %s\n", c?c:"-unknown-"); | |
296 | printf(" Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks)); | |
297 | printf("\n"); | |
298 | printf(" Avail Dev Size : %llu%s\n", | |
299 | (unsigned long long)__le64_to_cpu(sb->data_size), | |
300 | human_size(__le64_to_cpu(sb->data_size)<<9)); | |
301 | if (__le32_to_cpu(sb->level) > 0) { | |
302 | int ddsks = 0, ddsks_denom = 1; | |
303 | switch(__le32_to_cpu(sb->level)) { | |
304 | case 1: ddsks=1;break; | |
305 | case 4: | |
306 | case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break; | |
307 | case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break; | |
308 | case 10: | |
309 | layout = __le32_to_cpu(sb->layout); | |
310 | ddsks = __le32_to_cpu(sb->raid_disks); | |
311 | ddsks_denom = (layout&255) * ((layout>>8)&255); | |
312 | } | |
313 | if (ddsks) { | |
314 | long long asize = __le64_to_cpu(sb->size); | |
315 | asize = (asize << 9) * ddsks / ddsks_denom; | |
316 | printf(" Array Size : %llu%s\n", | |
317 | asize >> 10, human_size(asize)); | |
318 | } | |
319 | if (sb->size != sb->data_size) | |
320 | printf(" Used Dev Size : %llu%s\n", | |
321 | (unsigned long long)__le64_to_cpu(sb->size), | |
322 | human_size(__le64_to_cpu(sb->size)<<9)); | |
323 | } | |
324 | if (sb->data_offset) | |
325 | printf(" Data Offset : %llu sectors\n", | |
326 | (unsigned long long)__le64_to_cpu(sb->data_offset)); | |
327 | if (sb->new_offset && | |
328 | (__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) { | |
329 | unsigned long long offset = __le64_to_cpu(sb->data_offset); | |
330 | offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset); | |
331 | printf(" New Offset : %llu sectors\n", offset); | |
332 | } | |
333 | printf(" Super Offset : %llu sectors\n", | |
334 | (unsigned long long)__le64_to_cpu(sb->super_offset)); | |
335 | if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET) | |
336 | printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb->recovery_offset)); | |
337 | ||
338 | st->ss->getinfo_super(st, &info, NULL); | |
339 | if (info.space_after != 1 && | |
340 | !(__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
341 | printf(" Unused Space : before=%llu sectors, after=%llu sectors\n", | |
342 | info.space_before, info.space_after); | |
343 | ||
344 | printf(" State : %s\n", (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean"); | |
345 | printf(" Device UUID : "); | |
346 | for (i=0; i<16; i++) { | |
347 | if ((i&3)==0 && i != 0) printf(":"); | |
348 | printf("%02x", sb->device_uuid[i]); | |
349 | } | |
350 | printf("\n"); | |
351 | printf("\n"); | |
352 | if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) { | |
353 | printf("Internal Bitmap : %ld sectors from superblock\n", | |
354 | (long)(int32_t)__le32_to_cpu(sb->bitmap_offset)); | |
355 | } | |
356 | if (sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)) { | |
357 | printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb->reshape_position)/2, | |
358 | human_size(__le64_to_cpu(sb->reshape_position)<<9)); | |
359 | if (__le32_to_cpu(sb->delta_disks)) { | |
360 | printf(" Delta Devices : %d", __le32_to_cpu(sb->delta_disks)); | |
361 | printf(" (%d->%d)\n", | |
362 | __le32_to_cpu(sb->raid_disks)-__le32_to_cpu(sb->delta_disks), | |
363 | __le32_to_cpu(sb->raid_disks)); | |
364 | if ((int)__le32_to_cpu(sb->delta_disks) < 0) | |
365 | delta_extra = -__le32_to_cpu(sb->delta_disks); | |
366 | } | |
367 | if (__le32_to_cpu(sb->new_level) != __le32_to_cpu(sb->level)) { | |
368 | c = map_num(pers, __le32_to_cpu(sb->new_level)); | |
369 | printf(" New Level : %s\n", c?c:"-unknown-"); | |
370 | } | |
371 | if (__le32_to_cpu(sb->new_layout) != __le32_to_cpu(sb->layout)) { | |
372 | if (__le32_to_cpu(sb->level) == 5) { | |
373 | c = map_num(r5layout, __le32_to_cpu(sb->new_layout)); | |
374 | printf(" New Layout : %s\n", c?c:"-unknown-"); | |
375 | } | |
376 | if (__le32_to_cpu(sb->level) == 6) { | |
377 | c = map_num(r6layout, __le32_to_cpu(sb->new_layout)); | |
378 | printf(" New Layout : %s\n", c?c:"-unknown-"); | |
379 | } | |
380 | if (__le32_to_cpu(sb->level) == 10) { | |
381 | printf(" New Layout :"); | |
382 | print_r10_layout(__le32_to_cpu(sb->new_layout)); | |
383 | printf("\n"); | |
384 | } | |
385 | } | |
386 | if (__le32_to_cpu(sb->new_chunk) != __le32_to_cpu(sb->chunksize)) | |
387 | printf(" New Chunksize : %dK\n", __le32_to_cpu(sb->new_chunk)/2); | |
388 | printf("\n"); | |
389 | } | |
390 | if (sb->devflags) { | |
391 | printf(" Flags :"); | |
392 | if (sb->devflags & WriteMostly1) | |
393 | printf(" write-mostly"); | |
394 | printf("\n"); | |
395 | } | |
396 | ||
397 | atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL; | |
398 | printf(" Update Time : %.24s\n", ctime(&atime)); | |
399 | ||
400 | if (sb->bblog_size && sb->bblog_offset) { | |
401 | printf(" Bad Block Log : %d entries available at offset %ld sectors", | |
402 | __le16_to_cpu(sb->bblog_size)*512/8, | |
403 | (long)(int32_t)__le32_to_cpu(sb->bblog_offset)); | |
404 | if (sb->feature_map & | |
405 | __cpu_to_le32(MD_FEATURE_BAD_BLOCKS)) | |
406 | printf(" - bad blocks present."); | |
407 | printf("\n"); | |
408 | } | |
409 | ||
410 | if (calc_sb_1_csum(sb) == sb->sb_csum) | |
411 | printf(" Checksum : %x - correct\n", __le32_to_cpu(sb->sb_csum)); | |
412 | else | |
413 | printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb->sb_csum), | |
414 | __le32_to_cpu(calc_sb_1_csum(sb))); | |
415 | printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb->events)); | |
416 | printf("\n"); | |
417 | if (__le32_to_cpu(sb->level) == 5) { | |
418 | c = map_num(r5layout, __le32_to_cpu(sb->layout)); | |
419 | printf(" Layout : %s\n", c?c:"-unknown-"); | |
420 | } | |
421 | if (__le32_to_cpu(sb->level) == 6) { | |
422 | c = map_num(r6layout, __le32_to_cpu(sb->layout)); | |
423 | printf(" Layout : %s\n", c?c:"-unknown-"); | |
424 | } | |
425 | if (__le32_to_cpu(sb->level) == 10) { | |
426 | int lo = __le32_to_cpu(sb->layout); | |
427 | printf(" Layout :"); | |
428 | print_r10_layout(lo); | |
429 | printf("\n"); | |
430 | } | |
431 | switch(__le32_to_cpu(sb->level)) { | |
432 | case 0: | |
433 | case 4: | |
434 | case 5: | |
435 | case 6: | |
436 | case 10: | |
437 | printf(" Chunk Size : %dK\n", __le32_to_cpu(sb->chunksize)/2); | |
438 | break; | |
439 | case -1: | |
440 | printf(" Rounding : %dK\n", __le32_to_cpu(sb->chunksize)/2); | |
441 | break; | |
442 | default: break; | |
443 | } | |
444 | printf("\n"); | |
445 | #if 0 | |
446 | /* This turns out to just be confusing */ | |
447 | printf(" Array Slot : %d (", __le32_to_cpu(sb->dev_number)); | |
448 | for (i= __le32_to_cpu(sb->max_dev); i> 0 ; i--) | |
449 | if (__le16_to_cpu(sb->dev_roles[i-1]) != 0xffff) | |
450 | break; | |
451 | for (d=0; d < i; d++) { | |
452 | int role = __le16_to_cpu(sb->dev_roles[d]); | |
453 | if (d) printf(", "); | |
454 | if (role == 0xffff) printf("empty"); | |
455 | else if(role == 0xfffe) printf("failed"); | |
456 | else printf("%d", role); | |
457 | } | |
458 | printf(")\n"); | |
459 | #endif | |
460 | printf(" Device Role : "); | |
461 | d = __le32_to_cpu(sb->dev_number); | |
462 | if (d < __le32_to_cpu(sb->max_dev)) | |
463 | role = __le16_to_cpu(sb->dev_roles[d]); | |
464 | else | |
465 | role = 0xFFFF; | |
466 | if (role >= 0xFFFE) | |
467 | printf("spare\n"); | |
468 | else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_REPLACEMENT)) | |
469 | printf("Replacement device %d\n", role); | |
470 | else | |
471 | printf("Active device %d\n", role); | |
472 | ||
473 | printf(" Array State : "); | |
474 | for (d=0; d<__le32_to_cpu(sb->raid_disks) + delta_extra; d++) { | |
475 | int cnt = 0; | |
476 | unsigned int i; | |
477 | for (i=0; i< __le32_to_cpu(sb->max_dev); i++) { | |
478 | unsigned int role = __le16_to_cpu(sb->dev_roles[i]); | |
479 | if (role == d) | |
480 | cnt++; | |
481 | } | |
482 | if (cnt == 2) | |
483 | printf("R"); | |
484 | else if (cnt == 1) | |
485 | printf("A"); | |
486 | else if (cnt == 0) | |
487 | printf("."); | |
488 | else | |
489 | printf("?"); | |
490 | } | |
491 | #if 0 | |
492 | /* This is confusing too */ | |
493 | faulty = 0; | |
494 | for (i=0; i< __le32_to_cpu(sb->max_dev); i++) { | |
495 | int role = __le16_to_cpu(sb->dev_roles[i]); | |
496 | if (role == 0xFFFE) | |
497 | faulty++; | |
498 | } | |
499 | if (faulty) printf(" %d failed", faulty); | |
500 | #endif | |
501 | printf(" ('A' == active, '.' == missing, 'R' == replacing)"); | |
502 | printf("\n"); | |
503 | } | |
504 | ||
505 | static void brief_examine_super1(struct supertype *st, int verbose) | |
506 | { | |
507 | struct mdp_superblock_1 *sb = st->sb; | |
508 | int i; | |
509 | unsigned long long sb_offset; | |
510 | char *nm; | |
511 | char *c=map_num(pers, __le32_to_cpu(sb->level)); | |
512 | ||
513 | nm = strchr(sb->set_name, ':'); | |
514 | if (nm) | |
515 | nm++; | |
516 | else if (sb->set_name[0]) | |
517 | nm = sb->set_name; | |
518 | else | |
519 | nm = NULL; | |
520 | ||
521 | printf("ARRAY "); | |
522 | if (nm) { | |
523 | printf("/dev/md/"); | |
524 | print_escape(nm); | |
525 | putchar(' '); | |
526 | } | |
527 | if (verbose && c) | |
528 | printf(" level=%s", c); | |
529 | sb_offset = __le64_to_cpu(sb->super_offset); | |
530 | if (sb_offset <= 4) | |
531 | printf(" metadata=1.1 "); | |
532 | else if (sb_offset <= 8) | |
533 | printf(" metadata=1.2 "); | |
534 | else | |
535 | printf(" metadata=1.0 "); | |
536 | if (verbose) | |
537 | printf("num-devices=%d ", __le32_to_cpu(sb->raid_disks)); | |
538 | printf("UUID="); | |
539 | for (i=0; i<16; i++) { | |
540 | if ((i&3)==0 && i != 0) printf(":"); | |
541 | printf("%02x", sb->set_uuid[i]); | |
542 | } | |
543 | if (sb->set_name[0]) { | |
544 | printf(" name="); | |
545 | print_quoted(sb->set_name); | |
546 | } | |
547 | printf("\n"); | |
548 | } | |
549 | ||
550 | static void export_examine_super1(struct supertype *st) | |
551 | { | |
552 | struct mdp_superblock_1 *sb = st->sb; | |
553 | int i; | |
554 | int len = 32; | |
555 | int layout; | |
556 | ||
557 | printf("MD_LEVEL=%s\n", map_num(pers, __le32_to_cpu(sb->level))); | |
558 | printf("MD_DEVICES=%d\n", __le32_to_cpu(sb->raid_disks)); | |
559 | for (i=0; i<32; i++) | |
560 | if (sb->set_name[i] == '\n' || | |
561 | sb->set_name[i] == '\0') { | |
562 | len = i; | |
563 | break; | |
564 | } | |
565 | if (len) | |
566 | printf("MD_NAME=%.*s\n", len, sb->set_name); | |
567 | if (__le32_to_cpu(sb->level) > 0) { | |
568 | int ddsks = 0, ddsks_denom = 1; | |
569 | switch(__le32_to_cpu(sb->level)) { | |
570 | case 1: ddsks=1;break; | |
571 | case 4: | |
572 | case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break; | |
573 | case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break; | |
574 | case 10: | |
575 | layout = __le32_to_cpu(sb->layout); | |
576 | ddsks = __le32_to_cpu(sb->raid_disks); | |
577 | ddsks_denom = (layout&255) * ((layout>>8)&255); | |
578 | } | |
579 | if (ddsks) { | |
580 | long long asize = __le64_to_cpu(sb->size); | |
581 | asize = (asize << 9) * ddsks / ddsks_denom; | |
582 | printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize,JEDEC)); | |
583 | } | |
584 | } | |
585 | printf("MD_UUID="); | |
586 | for (i=0; i<16; i++) { | |
587 | if ((i&3)==0 && i != 0) printf(":"); | |
588 | printf("%02x", sb->set_uuid[i]); | |
589 | } | |
590 | printf("\n"); | |
591 | printf("MD_UPDATE_TIME=%llu\n", | |
592 | __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL); | |
593 | printf("MD_DEV_UUID="); | |
594 | for (i=0; i<16; i++) { | |
595 | if ((i&3)==0 && i != 0) printf(":"); | |
596 | printf("%02x", sb->device_uuid[i]); | |
597 | } | |
598 | printf("\n"); | |
599 | printf("MD_EVENTS=%llu\n", | |
600 | (unsigned long long)__le64_to_cpu(sb->events)); | |
601 | } | |
602 | ||
603 | static int copy_metadata1(struct supertype *st, int from, int to) | |
604 | { | |
605 | /* Read superblock. If it looks good, write it out. | |
606 | * Then if a bitmap is present, copy that. | |
607 | * And if a bad-block-list is present, copy that too. | |
608 | */ | |
609 | void *buf; | |
610 | unsigned long long dsize, sb_offset; | |
611 | const int bufsize = 4*1024; | |
612 | struct mdp_superblock_1 super, *sb; | |
613 | ||
614 | if (posix_memalign(&buf, 4096, bufsize) != 0) | |
615 | return 1; | |
616 | ||
617 | if (!get_dev_size(from, NULL, &dsize)) | |
618 | goto err; | |
619 | ||
620 | dsize >>= 9; | |
621 | if (dsize < 24) | |
622 | goto err; | |
623 | switch(st->minor_version) { | |
624 | case 0: | |
625 | sb_offset = dsize; | |
626 | sb_offset -= 8*2; | |
627 | sb_offset &= ~(4*2-1); | |
628 | break; | |
629 | case 1: | |
630 | sb_offset = 0; | |
631 | break; | |
632 | case 2: | |
633 | sb_offset = 4*2; | |
634 | break; | |
635 | default: | |
636 | goto err; | |
637 | } | |
638 | ||
639 | if (lseek64(from, sb_offset << 9, 0) < 0LL) | |
640 | goto err; | |
641 | if (read(from, buf, bufsize) != bufsize) | |
642 | goto err; | |
643 | ||
644 | sb = buf; | |
645 | super = *sb; // save most of sb for when we reuse buf | |
646 | ||
647 | if (__le32_to_cpu(super.magic) != MD_SB_MAGIC || | |
648 | __le32_to_cpu(super.major_version) != 1 || | |
649 | __le64_to_cpu(super.super_offset) != sb_offset || | |
650 | calc_sb_1_csum(sb) != super.sb_csum) | |
651 | goto err; | |
652 | ||
653 | if (lseek64(to, sb_offset << 9, 0) < 0LL) | |
654 | goto err; | |
655 | if (write(to, buf, bufsize) != bufsize) | |
656 | goto err; | |
657 | ||
658 | if (super.feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) { | |
659 | unsigned long long bitmap_offset = sb_offset; | |
660 | int bytes = 4096; // just an estimate. | |
661 | int written = 0; | |
662 | struct align_fd afrom, ato; | |
663 | ||
664 | init_afd(&afrom, from); | |
665 | init_afd(&ato, to); | |
666 | ||
667 | bitmap_offset += (int32_t)__le32_to_cpu(super.bitmap_offset); | |
668 | ||
669 | if (lseek64(from, bitmap_offset<<9, 0) < 0) | |
670 | goto err; | |
671 | if (lseek64(to, bitmap_offset<<9, 0) < 0) | |
672 | goto err; | |
673 | ||
674 | for (written = 0; written < bytes ; ) { | |
675 | int n = bytes - written; | |
676 | if (n > 4096) | |
677 | n = 4096; | |
678 | if (aread(&afrom, buf, n) != n) | |
679 | goto err; | |
680 | if (written == 0) { | |
681 | /* have the header, can calculate | |
682 | * correct bitmap bytes */ | |
683 | bitmap_super_t *bms; | |
684 | int bits; | |
685 | bms = (void*)buf; | |
686 | bits = __le64_to_cpu(bms->sync_size) / (__le32_to_cpu(bms->chunksize)>>9); | |
687 | bytes = (bits+7) >> 3; | |
688 | bytes += sizeof(bitmap_super_t); | |
689 | bytes = ROUND_UP(bytes, 512); | |
690 | if (n > bytes) | |
691 | n = bytes; | |
692 | } | |
693 | if (awrite(&ato, buf, n) != n) | |
694 | goto err; | |
695 | written += n; | |
696 | } | |
697 | } | |
698 | ||
699 | if (super.bblog_size != 0 && | |
700 | __le32_to_cpu(super.bblog_size) <= 100 && | |
701 | super.bblog_offset != 0 && | |
702 | (super.feature_map & __le32_to_cpu(MD_FEATURE_BAD_BLOCKS))) { | |
703 | /* There is a bad block log */ | |
704 | unsigned long long bb_offset = sb_offset; | |
705 | int bytes = __le32_to_cpu(super.bblog_size) * 512; | |
706 | int written = 0; | |
707 | struct align_fd afrom, ato; | |
708 | ||
709 | init_afd(&afrom, from); | |
710 | init_afd(&ato, to); | |
711 | ||
712 | bb_offset += (int32_t)__le32_to_cpu(super.bblog_offset); | |
713 | ||
714 | if (lseek64(from, bb_offset<<9, 0) < 0) | |
715 | goto err; | |
716 | if (lseek64(to, bb_offset<<9, 0) < 0) | |
717 | goto err; | |
718 | ||
719 | for (written = 0; written < bytes ; ) { | |
720 | int n = bytes - written; | |
721 | if (n > 4096) | |
722 | n = 4096; | |
723 | if (aread(&afrom, buf, n) != n) | |
724 | goto err; | |
725 | ||
726 | if (awrite(&ato, buf, n) != n) | |
727 | goto err; | |
728 | written += n; | |
729 | } | |
730 | } | |
731 | ||
732 | free(buf); | |
733 | return 0; | |
734 | ||
735 | err: | |
736 | free(buf); | |
737 | return 1; | |
738 | } | |
739 | ||
740 | static void detail_super1(struct supertype *st, char *homehost) | |
741 | { | |
742 | struct mdp_superblock_1 *sb = st->sb; | |
743 | int i; | |
744 | int l = homehost ? strlen(homehost) : 0; | |
745 | ||
746 | printf(" Name : %.32s", sb->set_name); | |
747 | if (l > 0 && l < 32 && | |
748 | sb->set_name[l] == ':' && | |
749 | strncmp(sb->set_name, homehost, l) == 0) | |
750 | printf(" (local to host %s)", homehost); | |
751 | printf("\n UUID : "); | |
752 | for (i=0; i<16; i++) { | |
753 | if ((i&3)==0 && i != 0) printf(":"); | |
754 | printf("%02x", sb->set_uuid[i]); | |
755 | } | |
756 | printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb->events)); | |
757 | } | |
758 | ||
759 | static void brief_detail_super1(struct supertype *st) | |
760 | { | |
761 | struct mdp_superblock_1 *sb = st->sb; | |
762 | int i; | |
763 | ||
764 | if (sb->set_name[0]) { | |
765 | printf(" name="); | |
766 | print_quoted(sb->set_name); | |
767 | } | |
768 | printf(" UUID="); | |
769 | for (i=0; i<16; i++) { | |
770 | if ((i&3)==0 && i != 0) printf(":"); | |
771 | printf("%02x", sb->set_uuid[i]); | |
772 | } | |
773 | } | |
774 | ||
775 | static void export_detail_super1(struct supertype *st) | |
776 | { | |
777 | struct mdp_superblock_1 *sb = st->sb; | |
778 | int i; | |
779 | int len = 32; | |
780 | ||
781 | for (i=0; i<32; i++) | |
782 | if (sb->set_name[i] == '\n' || | |
783 | sb->set_name[i] == '\0') { | |
784 | len = i; | |
785 | break; | |
786 | } | |
787 | if (len) | |
788 | printf("MD_NAME=%.*s\n", len, sb->set_name); | |
789 | } | |
790 | ||
791 | static int examine_badblocks_super1(struct supertype *st, int fd, char *devname) | |
792 | { | |
793 | struct mdp_superblock_1 *sb = st->sb; | |
794 | unsigned long long offset; | |
795 | int size; | |
796 | __u64 *bbl, *bbp; | |
797 | int i; | |
798 | ||
799 | if (!sb->bblog_size || __le32_to_cpu(sb->bblog_size) > 100 | |
800 | || !sb->bblog_offset){ | |
801 | printf("No bad-blocks list configured on %s\n", devname); | |
802 | return 0; | |
803 | } | |
804 | if ((sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS)) | |
805 | == 0) { | |
806 | printf("Bad-blocks list is empty in %s\n", devname); | |
807 | return 0; | |
808 | } | |
809 | ||
810 | size = __le32_to_cpu(sb->bblog_size)* 512; | |
811 | if (posix_memalign((void**)&bbl, 4096, size) != 0) { | |
812 | pr_err("could not allocate badblocks list\n"); | |
813 | return 0; | |
814 | } | |
815 | offset = __le64_to_cpu(sb->super_offset) + | |
816 | (int)__le32_to_cpu(sb->bblog_offset); | |
817 | offset <<= 9; | |
818 | if (lseek64(fd, offset, 0) < 0) { | |
819 | pr_err("Cannot seek to bad-blocks list\n"); | |
820 | return 1; | |
821 | } | |
822 | if (read(fd, bbl, size) != size) { | |
823 | pr_err("Cannot read bad-blocks list\n"); | |
824 | return 1; | |
825 | } | |
826 | /* 64bits per entry. 10 bits is block-count, 54 bits is block | |
827 | * offset. Blocks are sectors unless bblog->shift makes them bigger | |
828 | */ | |
829 | bbp = (__u64*)bbl; | |
830 | printf("Bad-blocks on %s:\n", devname); | |
831 | for (i = 0; i < size/8; i++, bbp++) { | |
832 | __u64 bb = __le64_to_cpu(*bbp); | |
833 | int count = bb & 0x3ff; | |
834 | unsigned long long sector = bb >> 10; | |
835 | ||
836 | if (bb + 1 == 0) | |
837 | break; | |
838 | ||
839 | sector <<= sb->bblog_shift; | |
840 | count <<= sb->bblog_shift; | |
841 | ||
842 | printf("%20llu for %d sectors\n", sector, count); | |
843 | } | |
844 | return 0; | |
845 | } | |
846 | ||
847 | #endif | |
848 | ||
849 | static int match_home1(struct supertype *st, char *homehost) | |
850 | { | |
851 | struct mdp_superblock_1 *sb = st->sb; | |
852 | int l = homehost ? strlen(homehost) : 0; | |
853 | ||
854 | return (l > 0 && l < 32 && | |
855 | sb->set_name[l] == ':' && | |
856 | strncmp(sb->set_name, homehost, l) == 0); | |
857 | } | |
858 | ||
859 | static void uuid_from_super1(struct supertype *st, int uuid[4]) | |
860 | { | |
861 | struct mdp_superblock_1 *super = st->sb; | |
862 | char *cuuid = (char*)uuid; | |
863 | int i; | |
864 | for (i=0; i<16; i++) | |
865 | cuuid[i] = super->set_uuid[i]; | |
866 | } | |
867 | ||
868 | static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map) | |
869 | { | |
870 | struct mdp_superblock_1 *sb = st->sb; | |
871 | struct bitmap_super_s *bsb = (void*)(((char*)sb)+MAX_SB_SIZE); | |
872 | struct misc_dev_info *misc = (void*)(((char*)sb)+MAX_SB_SIZE+BM_SUPER_SIZE); | |
873 | int working = 0; | |
874 | unsigned int i; | |
875 | unsigned int role; | |
876 | unsigned int map_disks = info->array.raid_disks; | |
877 | unsigned long long super_offset; | |
878 | unsigned long long data_size; | |
879 | ||
880 | memset(info, 0, sizeof(*info)); | |
881 | info->array.major_version = 1; | |
882 | info->array.minor_version = st->minor_version; | |
883 | info->array.patch_version = 0; | |
884 | info->array.raid_disks = __le32_to_cpu(sb->raid_disks); | |
885 | info->array.level = __le32_to_cpu(sb->level); | |
886 | info->array.layout = __le32_to_cpu(sb->layout); | |
887 | info->array.md_minor = -1; | |
888 | info->array.ctime = __le64_to_cpu(sb->ctime); | |
889 | info->array.utime = __le64_to_cpu(sb->utime); | |
890 | info->array.chunk_size = __le32_to_cpu(sb->chunksize)*512; | |
891 | info->array.state = | |
892 | (__le64_to_cpu(sb->resync_offset) == MaxSector) | |
893 | ? 1 : 0; | |
894 | ||
895 | info->data_offset = __le64_to_cpu(sb->data_offset); | |
896 | info->component_size = __le64_to_cpu(sb->size); | |
897 | if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) | |
898 | info->bitmap_offset = (int32_t)__le32_to_cpu(sb->bitmap_offset); | |
899 | ||
900 | info->disk.major = 0; | |
901 | info->disk.minor = 0; | |
902 | info->disk.number = __le32_to_cpu(sb->dev_number); | |
903 | if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) || | |
904 | __le32_to_cpu(sb->dev_number) >= MAX_DEVS) | |
905 | role = 0xfffe; | |
906 | else | |
907 | role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]); | |
908 | ||
909 | super_offset = __le64_to_cpu(sb->super_offset); | |
910 | if (info->array.level <= 0) | |
911 | data_size = __le64_to_cpu(sb->data_size); | |
912 | else | |
913 | data_size = __le64_to_cpu(sb->size); | |
914 | if (info->data_offset < super_offset) { | |
915 | unsigned long long end; | |
916 | info->space_before = info->data_offset; | |
917 | end = super_offset; | |
918 | ||
919 | if (sb->bblog_offset && sb->bblog_size) { | |
920 | unsigned long long bboffset = super_offset; | |
921 | bboffset += (int32_t)__le32_to_cpu(sb->bblog_offset); | |
922 | if (bboffset < end) | |
923 | end = bboffset; | |
924 | } | |
925 | ||
926 | if (super_offset + info->bitmap_offset < end) | |
927 | end = super_offset + info->bitmap_offset; | |
928 | ||
929 | if (info->data_offset + data_size < end) | |
930 | info->space_after = end - data_size - info->data_offset; | |
931 | else | |
932 | info->space_after = 0; | |
933 | } else { | |
934 | unsigned long long earliest; | |
935 | earliest = super_offset + (32+4)*2; /* match kernel */ | |
936 | if (info->bitmap_offset > 0) { | |
937 | unsigned long long bmend = info->bitmap_offset; | |
938 | unsigned long long size = __le64_to_cpu(bsb->sync_size); | |
939 | size /= __le32_to_cpu(bsb->chunksize) >> 9; | |
940 | size = (size + 7) >> 3; | |
941 | size += sizeof(bitmap_super_t); | |
942 | size = ROUND_UP(size, 4096); | |
943 | size /= 512; | |
944 | bmend += size; | |
945 | if (bmend > earliest) | |
946 | bmend = earliest; | |
947 | } | |
948 | if (sb->bblog_offset && sb->bblog_size) { | |
949 | unsigned long long bbend = super_offset; | |
950 | bbend += (int32_t)__le32_to_cpu(sb->bblog_offset); | |
951 | bbend += __le32_to_cpu(sb->bblog_size); | |
952 | if (bbend > earliest) | |
953 | earliest = bbend; | |
954 | } | |
955 | if (earliest < info->data_offset) | |
956 | info->space_before = info->data_offset - earliest; | |
957 | else | |
958 | info->space_before = 0; | |
959 | info->space_after = misc->device_size - data_size - info->data_offset; | |
960 | } | |
961 | if (info->space_before == 0 && info->space_after == 0) { | |
962 | /* It will look like we don't support data_offset changes, | |
963 | * be we do - it's just that there is no room. | |
964 | * A change that reduced the number of devices should | |
965 | * still be allowed, so set the otherwise useless value of '1' | |
966 | */ | |
967 | info->space_after = 1; | |
968 | } | |
969 | ||
970 | info->disk.raid_disk = -1; | |
971 | switch(role) { | |
972 | case 0xFFFF: | |
973 | info->disk.state = 0; /* spare: not active, not sync, not faulty */ | |
974 | break; | |
975 | case 0xFFFE: | |
976 | info->disk.state = 1; /* faulty */ | |
977 | break; | |
978 | default: | |
979 | info->disk.state = 6; /* active and in sync */ | |
980 | info->disk.raid_disk = role; | |
981 | } | |
982 | if (sb->devflags & WriteMostly1) | |
983 | info->disk.state |= (1 << MD_DISK_WRITEMOSTLY); | |
984 | info->events = __le64_to_cpu(sb->events); | |
985 | sprintf(info->text_version, "1.%d", st->minor_version); | |
986 | info->safe_mode_delay = 200; | |
987 | ||
988 | memcpy(info->uuid, sb->set_uuid, 16); | |
989 | ||
990 | strncpy(info->name, sb->set_name, 32); | |
991 | info->name[32] = 0; | |
992 | ||
993 | if ((__le32_to_cpu(sb->feature_map)&MD_FEATURE_REPLACEMENT)) { | |
994 | info->disk.state &= ~(1 << MD_DISK_SYNC); | |
995 | info->disk.state |= 1 << MD_DISK_REPLACEMENT; | |
996 | } | |
997 | ||
998 | if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET)) | |
999 | info->recovery_start = __le32_to_cpu(sb->recovery_offset); | |
1000 | else | |
1001 | info->recovery_start = MaxSector; | |
1002 | ||
1003 | if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) { | |
1004 | info->reshape_active = 1; | |
1005 | if ((sb->feature_map & __le32_to_cpu(MD_FEATURE_NEW_OFFSET)) && | |
1006 | sb->new_offset != 0) | |
1007 | info->reshape_active |= RESHAPE_NO_BACKUP; | |
1008 | info->reshape_progress = __le64_to_cpu(sb->reshape_position); | |
1009 | info->new_level = __le32_to_cpu(sb->new_level); | |
1010 | info->delta_disks = __le32_to_cpu(sb->delta_disks); | |
1011 | info->new_layout = __le32_to_cpu(sb->new_layout); | |
1012 | info->new_chunk = __le32_to_cpu(sb->new_chunk)<<9; | |
1013 | if (info->delta_disks < 0) | |
1014 | info->array.raid_disks -= info->delta_disks; | |
1015 | } else | |
1016 | info->reshape_active = 0; | |
1017 | ||
1018 | info->recovery_blocked = info->reshape_active; | |
1019 | ||
1020 | if (map) | |
1021 | for (i=0; i<map_disks; i++) | |
1022 | map[i] = 0; | |
1023 | for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) { | |
1024 | role = __le16_to_cpu(sb->dev_roles[i]); | |
1025 | if (/*role == 0xFFFF || */role < (unsigned) info->array.raid_disks) { | |
1026 | working++; | |
1027 | if (map && role < map_disks) | |
1028 | map[role] = 1; | |
1029 | } | |
1030 | } | |
1031 | ||
1032 | info->array.working_disks = working; | |
1033 | } | |
1034 | ||
1035 | static struct mdinfo *container_content1(struct supertype *st, char *subarray) | |
1036 | { | |
1037 | struct mdinfo *info; | |
1038 | ||
1039 | if (subarray) | |
1040 | return NULL; | |
1041 | ||
1042 | info = xmalloc(sizeof(*info)); | |
1043 | getinfo_super1(st, info, NULL); | |
1044 | return info; | |
1045 | } | |
1046 | ||
1047 | static int update_super1(struct supertype *st, struct mdinfo *info, | |
1048 | char *update, | |
1049 | char *devname, int verbose, | |
1050 | int uuid_set, char *homehost) | |
1051 | { | |
1052 | /* NOTE: for 'assemble' and 'force' we need to return non-zero | |
1053 | * if any change was made. For others, the return value is | |
1054 | * ignored. | |
1055 | */ | |
1056 | int rv = 0; | |
1057 | struct mdp_superblock_1 *sb = st->sb; | |
1058 | ||
1059 | if (strcmp(update, "homehost") == 0 && | |
1060 | homehost) { | |
1061 | /* Note that 'homehost' is special as it is really | |
1062 | * a "name" update. | |
1063 | */ | |
1064 | char *c; | |
1065 | update = "name"; | |
1066 | c = strchr(sb->set_name, ':'); | |
1067 | if (c) | |
1068 | strncpy(info->name, c+1, 31 - (c-sb->set_name)); | |
1069 | else | |
1070 | strncpy(info->name, sb->set_name, 32); | |
1071 | info->name[32] = 0; | |
1072 | } | |
1073 | ||
1074 | if (strcmp(update, "force-one")==0) { | |
1075 | /* Not enough devices for a working array, | |
1076 | * so bring this one up-to-date | |
1077 | */ | |
1078 | if (sb->events != __cpu_to_le64(info->events)) | |
1079 | rv = 1; | |
1080 | sb->events = __cpu_to_le64(info->events); | |
1081 | } else if (strcmp(update, "force-array")==0) { | |
1082 | /* Degraded array and 'force' requests to | |
1083 | * maybe need to mark it 'clean'. | |
1084 | */ | |
1085 | switch(__le32_to_cpu(sb->level)) { | |
1086 | case 5: case 4: case 6: | |
1087 | /* need to force clean */ | |
1088 | if (sb->resync_offset != MaxSector) | |
1089 | rv = 1; | |
1090 | sb->resync_offset = MaxSector; | |
1091 | } | |
1092 | } else if (strcmp(update, "assemble")==0) { | |
1093 | int d = info->disk.number; | |
1094 | int want; | |
1095 | if (info->disk.state & (1<<MD_DISK_ACTIVE)) | |
1096 | want = info->disk.raid_disk; | |
1097 | else | |
1098 | want = 0xFFFF; | |
1099 | if (sb->dev_roles[d] != __cpu_to_le16(want)) { | |
1100 | sb->dev_roles[d] = __cpu_to_le16(want); | |
1101 | rv = 1; | |
1102 | } | |
1103 | if (info->reshape_active && | |
1104 | sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) && | |
1105 | info->delta_disks >= 0 && | |
1106 | info->reshape_progress < __le64_to_cpu(sb->reshape_position)) { | |
1107 | sb->reshape_position = __cpu_to_le64(info->reshape_progress); | |
1108 | rv = 1; | |
1109 | } | |
1110 | if (info->reshape_active && | |
1111 | sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) && | |
1112 | info->delta_disks < 0 && | |
1113 | info->reshape_progress > __le64_to_cpu(sb->reshape_position)) { | |
1114 | sb->reshape_position = __cpu_to_le64(info->reshape_progress); | |
1115 | rv = 1; | |
1116 | } | |
1117 | } else if (strcmp(update, "linear-grow-new") == 0) { | |
1118 | unsigned int i; | |
1119 | int rfd, fd; | |
1120 | unsigned int max = __le32_to_cpu(sb->max_dev); | |
1121 | ||
1122 | for (i=0 ; i < max ; i++) | |
1123 | if (__le16_to_cpu(sb->dev_roles[i]) >= 0xfffe) | |
1124 | break; | |
1125 | sb->dev_number = __cpu_to_le32(i); | |
1126 | info->disk.number = i; | |
1127 | if (max >= __le32_to_cpu(sb->max_dev)) | |
1128 | sb->max_dev = __cpu_to_le32(max+1); | |
1129 | ||
1130 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
1131 | read(rfd, sb->device_uuid, 16) != 16) { | |
1132 | __u32 r[4] = {random(), random(), random(), random()}; | |
1133 | memcpy(sb->device_uuid, r, 16); | |
1134 | } | |
1135 | if (rfd >= 0) | |
1136 | close(rfd); | |
1137 | ||
1138 | sb->dev_roles[i] = | |
1139 | __cpu_to_le16(info->disk.raid_disk); | |
1140 | ||
1141 | fd = open(devname, O_RDONLY); | |
1142 | if (fd >= 0) { | |
1143 | unsigned long long ds; | |
1144 | get_dev_size(fd, devname, &ds); | |
1145 | close(fd); | |
1146 | ds >>= 9; | |
1147 | if (__le64_to_cpu(sb->super_offset) < | |
1148 | __le64_to_cpu(sb->data_offset)) { | |
1149 | sb->data_size = __cpu_to_le64( | |
1150 | ds - __le64_to_cpu(sb->data_offset)); | |
1151 | } else { | |
1152 | ds -= 8*2; | |
1153 | ds &= ~(unsigned long long)(4*2-1); | |
1154 | sb->super_offset = __cpu_to_le64(ds); | |
1155 | sb->data_size = __cpu_to_le64( | |
1156 | ds - __le64_to_cpu(sb->data_offset)); | |
1157 | } | |
1158 | } | |
1159 | } else if (strcmp(update, "linear-grow-update") == 0) { | |
1160 | sb->raid_disks = __cpu_to_le32(info->array.raid_disks); | |
1161 | sb->dev_roles[info->disk.number] = | |
1162 | __cpu_to_le16(info->disk.raid_disk); | |
1163 | } else if (strcmp(update, "resync") == 0) { | |
1164 | /* make sure resync happens */ | |
1165 | sb->resync_offset = 0ULL; | |
1166 | } else if (strcmp(update, "uuid") == 0) { | |
1167 | copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid); | |
1168 | ||
1169 | if (__le32_to_cpu(sb->feature_map)&MD_FEATURE_BITMAP_OFFSET) { | |
1170 | struct bitmap_super_s *bm; | |
1171 | bm = (struct bitmap_super_s*)(st->sb+MAX_SB_SIZE); | |
1172 | memcpy(bm->uuid, sb->set_uuid, 16); | |
1173 | } | |
1174 | } else if (strcmp(update, "no-bitmap") == 0) { | |
1175 | sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); | |
1176 | } else if (strcmp(update, "bbl") == 0) { | |
1177 | /* only possible if there is room after the bitmap, or if | |
1178 | * there is no bitmap | |
1179 | */ | |
1180 | unsigned long long sb_offset = __le64_to_cpu(sb->super_offset); | |
1181 | unsigned long long data_offset = __le64_to_cpu(sb->data_offset); | |
1182 | long bitmap_offset = (long)(int32_t)__le32_to_cpu(sb->bitmap_offset); | |
1183 | long bm_sectors = 0; | |
1184 | long space; | |
1185 | ||
1186 | #ifndef MDASSEMBLE | |
1187 | if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) { | |
1188 | struct bitmap_super_s *bsb; | |
1189 | bsb = (struct bitmap_super_s *)(((char*)sb)+MAX_SB_SIZE); | |
1190 | bm_sectors = bitmap_sectors(bsb); | |
1191 | } | |
1192 | #endif | |
1193 | if (sb_offset < data_offset) { | |
1194 | /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K | |
1195 | */ | |
1196 | long bb_offset; | |
1197 | bb_offset = sb_offset + 8; | |
1198 | if (bm_sectors && bitmap_offset > 0) | |
1199 | bb_offset = bitmap_offset + bm_sectors; | |
1200 | while (bb_offset < (long)sb_offset + 8 + 32*2 | |
1201 | && bb_offset + 8+8 <= (long)data_offset) | |
1202 | /* too close to bitmap, and room to grow */ | |
1203 | bb_offset += 8; | |
1204 | if (bb_offset + 8 <= (long)data_offset) { | |
1205 | sb->bblog_size = __cpu_to_le16(8); | |
1206 | sb->bblog_offset = __cpu_to_le32(bb_offset); | |
1207 | } | |
1208 | } else { | |
1209 | /* 1.0 - Put bbl just before super block */ | |
1210 | if (bm_sectors && bitmap_offset < 0) | |
1211 | space = -bitmap_offset - bm_sectors; | |
1212 | else | |
1213 | space = sb_offset - data_offset - | |
1214 | __le64_to_cpu(sb->data_size); | |
1215 | if (space >= 8) { | |
1216 | sb->bblog_size = __cpu_to_le16(8); | |
1217 | sb->bblog_offset = __cpu_to_le32((unsigned)-8); | |
1218 | } | |
1219 | } | |
1220 | } else if (strcmp(update, "no-bbl") == 0) { | |
1221 | if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS)) | |
1222 | pr_err("Cannot remove active bbl from %s\n",devname); | |
1223 | else { | |
1224 | sb->bblog_size = 0; | |
1225 | sb->bblog_shift = 0; | |
1226 | sb->bblog_offset = 0; | |
1227 | } | |
1228 | } else if (strcmp(update, "name") == 0) { | |
1229 | if (info->name[0] == 0) | |
1230 | sprintf(info->name, "%d", info->array.md_minor); | |
1231 | memset(sb->set_name, 0, sizeof(sb->set_name)); | |
1232 | if (homehost && | |
1233 | strchr(info->name, ':') == NULL && | |
1234 | strlen(homehost)+1+strlen(info->name) < 32) { | |
1235 | strcpy(sb->set_name, homehost); | |
1236 | strcat(sb->set_name, ":"); | |
1237 | strcat(sb->set_name, info->name); | |
1238 | } else | |
1239 | strcpy(sb->set_name, info->name); | |
1240 | } else if (strcmp(update, "devicesize") == 0 && | |
1241 | __le64_to_cpu(sb->super_offset) < | |
1242 | __le64_to_cpu(sb->data_offset)) { | |
1243 | /* set data_size to device size less data_offset */ | |
1244 | struct misc_dev_info *misc = (struct misc_dev_info*) | |
1245 | (st->sb + MAX_SB_SIZE + BM_SUPER_SIZE); | |
1246 | sb->data_size = __cpu_to_le64( | |
1247 | misc->device_size - __le64_to_cpu(sb->data_offset)); | |
1248 | } else if (strcmp(update, "revert-reshape") == 0) { | |
1249 | rv = -2; | |
1250 | if (!(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE))) | |
1251 | pr_err("No active reshape to revert on %s\n", | |
1252 | devname); | |
1253 | else { | |
1254 | __u32 temp; | |
1255 | unsigned long long reshape_sectors; | |
1256 | long reshape_chunk; | |
1257 | rv = 0; | |
1258 | /* reshape_position is a little messy. | |
1259 | * Its value must be a multiple of the larger | |
1260 | * chunk size, and of the "after" data disks. | |
1261 | * So when reverting we need to change it to | |
1262 | * be a multiple of the new "after" data disks, | |
1263 | * which is the old "before". | |
1264 | * If it isn't already a multiple of 'before', | |
1265 | * the only thing we could do would be | |
1266 | * copy some block around on the disks, which | |
1267 | * is easy to get wrong. | |
1268 | * So we reject a revert-reshape unless the | |
1269 | * alignment is good. | |
1270 | */ | |
1271 | if (__le32_to_cpu(sb->level) >= 4 && | |
1272 | __le32_to_cpu(sb->level) <= 6) { | |
1273 | reshape_sectors = __le64_to_cpu(sb->reshape_position); | |
1274 | reshape_chunk = __le32_to_cpu(sb->new_chunk); | |
1275 | reshape_chunk *= __le32_to_cpu(sb->raid_disks) - __le32_to_cpu(sb->delta_disks) - | |
1276 | (__le32_to_cpu(sb->level)==6 ? 2 : 1); | |
1277 | if (reshape_sectors % reshape_chunk) { | |
1278 | pr_err("Reshape position is not suitably aligned.\n"); | |
1279 | pr_err("Try normal assembly and stop again\n"); | |
1280 | return -2; | |
1281 | } | |
1282 | } | |
1283 | sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) - | |
1284 | __le32_to_cpu(sb->delta_disks)); | |
1285 | if (sb->delta_disks == 0) | |
1286 | sb->feature_map ^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
1287 | else | |
1288 | sb->delta_disks = __cpu_to_le32(-__le32_to_cpu(sb->delta_disks)); | |
1289 | ||
1290 | temp = sb->new_layout; | |
1291 | sb->new_layout = sb->layout; | |
1292 | sb->layout = temp; | |
1293 | ||
1294 | temp = sb->new_chunk; | |
1295 | sb->new_chunk = sb->chunksize; | |
1296 | sb->chunksize = temp; | |
1297 | ||
1298 | if (sb->feature_map & __cpu_to_le32(MD_FEATURE_NEW_OFFSET)) { | |
1299 | long offset_delta = (int32_t)__le32_to_cpu(sb->new_offset); | |
1300 | sb->data_offset = __cpu_to_le64(__le64_to_cpu(sb->data_offset) + offset_delta); | |
1301 | sb->new_offset = __cpu_to_le32(-offset_delta); | |
1302 | sb->data_size = __cpu_to_le64(__le64_to_cpu(sb->data_size) - offset_delta); | |
1303 | } | |
1304 | } | |
1305 | } else if (strcmp(update, "_reshape_progress")==0) | |
1306 | sb->reshape_position = __cpu_to_le64(info->reshape_progress); | |
1307 | else if (strcmp(update, "writemostly")==0) | |
1308 | sb->devflags |= WriteMostly1; | |
1309 | else if (strcmp(update, "readwrite")==0) | |
1310 | sb->devflags &= ~WriteMostly1; | |
1311 | else | |
1312 | rv = -1; | |
1313 | ||
1314 | sb->sb_csum = calc_sb_1_csum(sb); | |
1315 | return rv; | |
1316 | } | |
1317 | ||
1318 | static int init_super1(struct supertype *st, mdu_array_info_t *info, | |
1319 | unsigned long long size, char *name, char *homehost, | |
1320 | int *uuid, unsigned long long data_offset) | |
1321 | { | |
1322 | struct mdp_superblock_1 *sb; | |
1323 | int spares; | |
1324 | int rfd; | |
1325 | char defname[10]; | |
1326 | int sbsize; | |
1327 | ||
1328 | if (posix_memalign((void**)&sb, 4096, SUPER1_SIZE) != 0) { | |
1329 | pr_err("could not allocate superblock\n"); | |
1330 | return 0; | |
1331 | } | |
1332 | memset(sb, 0, SUPER1_SIZE); | |
1333 | ||
1334 | st->sb = sb; | |
1335 | if (info == NULL) { | |
1336 | /* zeroing superblock */ | |
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | spares = info->working_disks - info->active_disks; | |
1341 | if (info->raid_disks + spares > MAX_DEVS) { | |
1342 | pr_err("too many devices requested: %d+%d > %d\n", | |
1343 | info->raid_disks , spares, MAX_DEVS); | |
1344 | return 0; | |
1345 | } | |
1346 | ||
1347 | sb->magic = __cpu_to_le32(MD_SB_MAGIC); | |
1348 | sb->major_version = __cpu_to_le32(1); | |
1349 | sb->feature_map = 0; | |
1350 | sb->pad0 = 0; | |
1351 | ||
1352 | if (uuid) | |
1353 | copy_uuid(sb->set_uuid, uuid, super1.swapuuid); | |
1354 | else { | |
1355 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
1356 | read(rfd, sb->set_uuid, 16) != 16) { | |
1357 | __u32 r[4] = {random(), random(), random(), random()}; | |
1358 | memcpy(sb->set_uuid, r, 16); | |
1359 | } | |
1360 | if (rfd >= 0) close(rfd); | |
1361 | } | |
1362 | ||
1363 | if (name == NULL || *name == 0) { | |
1364 | sprintf(defname, "%d", info->md_minor); | |
1365 | name = defname; | |
1366 | } | |
1367 | if (homehost && | |
1368 | strchr(name, ':')== NULL && | |
1369 | strlen(homehost)+1+strlen(name) < 32) { | |
1370 | strcpy(sb->set_name, homehost); | |
1371 | strcat(sb->set_name, ":"); | |
1372 | strcat(sb->set_name, name); | |
1373 | } else | |
1374 | strcpy(sb->set_name, name); | |
1375 | ||
1376 | sb->ctime = __cpu_to_le64((unsigned long long)time(0)); | |
1377 | sb->level = __cpu_to_le32(info->level); | |
1378 | sb->layout = __cpu_to_le32(info->layout); | |
1379 | sb->size = __cpu_to_le64(size*2ULL); | |
1380 | sb->chunksize = __cpu_to_le32(info->chunk_size>>9); | |
1381 | sb->raid_disks = __cpu_to_le32(info->raid_disks); | |
1382 | ||
1383 | sb->data_offset = __cpu_to_le64(data_offset); | |
1384 | sb->data_size = __cpu_to_le64(0); | |
1385 | sb->super_offset = __cpu_to_le64(0); | |
1386 | sb->recovery_offset = __cpu_to_le64(0); | |
1387 | ||
1388 | sb->utime = sb->ctime; | |
1389 | sb->events = __cpu_to_le64(1); | |
1390 | if (info->state & (1<<MD_SB_CLEAN)) | |
1391 | sb->resync_offset = MaxSector; | |
1392 | else | |
1393 | sb->resync_offset = 0; | |
1394 | sbsize = sizeof(struct mdp_superblock_1) + 2 * (info->raid_disks + spares); | |
1395 | sbsize = ROUND_UP(sbsize, 512); | |
1396 | sb->max_dev = __cpu_to_le32((sbsize - sizeof(struct mdp_superblock_1)) / 2); | |
1397 | ||
1398 | memset(sb->dev_roles, 0xff, MAX_SB_SIZE - sizeof(struct mdp_superblock_1)); | |
1399 | ||
1400 | return 1; | |
1401 | } | |
1402 | ||
1403 | struct devinfo { | |
1404 | int fd; | |
1405 | char *devname; | |
1406 | long long data_offset; | |
1407 | mdu_disk_info_t disk; | |
1408 | struct devinfo *next; | |
1409 | }; | |
1410 | #ifndef MDASSEMBLE | |
1411 | /* Add a device to the superblock being created */ | |
1412 | static int add_to_super1(struct supertype *st, mdu_disk_info_t *dk, | |
1413 | int fd, char *devname, unsigned long long data_offset) | |
1414 | { | |
1415 | struct mdp_superblock_1 *sb = st->sb; | |
1416 | __u16 *rp = sb->dev_roles + dk->number; | |
1417 | struct devinfo *di, **dip; | |
1418 | ||
1419 | if ((dk->state & 6) == 6) /* active, sync */ | |
1420 | *rp = __cpu_to_le16(dk->raid_disk); | |
1421 | else if ((dk->state & ~2) == 0) /* active or idle -> spare */ | |
1422 | *rp = 0xffff; | |
1423 | else | |
1424 | *rp = 0xfffe; | |
1425 | ||
1426 | if (dk->number >= (int)__le32_to_cpu(sb->max_dev) && | |
1427 | __le32_to_cpu(sb->max_dev) < MAX_DEVS) | |
1428 | sb->max_dev = __cpu_to_le32(dk->number+1); | |
1429 | ||
1430 | sb->dev_number = __cpu_to_le32(dk->number); | |
1431 | sb->devflags = 0; /* don't copy another disks flags */ | |
1432 | sb->sb_csum = calc_sb_1_csum(sb); | |
1433 | ||
1434 | dip = (struct devinfo **)&st->info; | |
1435 | while (*dip) | |
1436 | dip = &(*dip)->next; | |
1437 | di = xmalloc(sizeof(struct devinfo)); | |
1438 | di->fd = fd; | |
1439 | di->devname = devname; | |
1440 | di->disk = *dk; | |
1441 | di->data_offset = data_offset; | |
1442 | di->next = NULL; | |
1443 | *dip = di; | |
1444 | ||
1445 | return 0; | |
1446 | } | |
1447 | #endif | |
1448 | ||
1449 | static void locate_bitmap1(struct supertype *st, int fd); | |
1450 | ||
1451 | static int store_super1(struct supertype *st, int fd) | |
1452 | { | |
1453 | struct mdp_superblock_1 *sb = st->sb; | |
1454 | unsigned long long sb_offset; | |
1455 | struct align_fd afd; | |
1456 | int sbsize; | |
1457 | unsigned long long dsize; | |
1458 | ||
1459 | if (!get_dev_size(fd, NULL, &dsize)) | |
1460 | return 1; | |
1461 | ||
1462 | dsize >>= 9; | |
1463 | ||
1464 | if (dsize < 24) | |
1465 | return 2; | |
1466 | ||
1467 | init_afd(&afd, fd); | |
1468 | ||
1469 | /* | |
1470 | * Calculate the position of the superblock. | |
1471 | * It is always aligned to a 4K boundary and | |
1472 | * depending on minor_version, it can be: | |
1473 | * 0: At least 8K, but less than 12K, from end of device | |
1474 | * 1: At start of device | |
1475 | * 2: 4K from start of device. | |
1476 | */ | |
1477 | switch(st->minor_version) { | |
1478 | case 0: | |
1479 | sb_offset = dsize; | |
1480 | sb_offset -= 8*2; | |
1481 | sb_offset &= ~(4*2-1); | |
1482 | break; | |
1483 | case 1: | |
1484 | sb_offset = 0; | |
1485 | break; | |
1486 | case 2: | |
1487 | sb_offset = 4*2; | |
1488 | break; | |
1489 | default: | |
1490 | return -EINVAL; | |
1491 | } | |
1492 | ||
1493 | if (sb_offset != __le64_to_cpu(sb->super_offset) && | |
1494 | 0 != __le64_to_cpu(sb->super_offset) | |
1495 | ) { | |
1496 | pr_err("internal error - sb_offset is wrong\n"); | |
1497 | abort(); | |
1498 | } | |
1499 | ||
1500 | if (lseek64(fd, sb_offset << 9, 0)< 0LL) | |
1501 | return 3; | |
1502 | ||
1503 | sbsize = ROUND_UP(sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev), 512); | |
1504 | ||
1505 | if (awrite(&afd, sb, sbsize) != sbsize) | |
1506 | return 4; | |
1507 | ||
1508 | if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) { | |
1509 | struct bitmap_super_s *bm = (struct bitmap_super_s*) | |
1510 | (((char*)sb)+MAX_SB_SIZE); | |
1511 | if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC) { | |
1512 | locate_bitmap1(st, fd); | |
1513 | if (awrite(&afd, bm, sizeof(*bm)) != sizeof(*bm)) | |
1514 | return 5; | |
1515 | } | |
1516 | } | |
1517 | fsync(fd); | |
1518 | return 0; | |
1519 | } | |
1520 | ||
1521 | static int load_super1(struct supertype *st, int fd, char *devname); | |
1522 | ||
1523 | static unsigned long choose_bm_space(unsigned long devsize) | |
1524 | { | |
1525 | /* if the device is bigger than 8Gig, save 64k for bitmap usage, | |
1526 | * if bigger than 200Gig, save 128k | |
1527 | * NOTE: result must be multiple of 4K else bad things happen | |
1528 | * on 4K-sector devices. | |
1529 | */ | |
1530 | if (devsize < 64*2) return 0; | |
1531 | if (devsize - 64*2 >= 200*1024*1024*2) | |
1532 | return 128*2; | |
1533 | if (devsize - 4*2 > 8*1024*1024*2) | |
1534 | return 64*2; | |
1535 | return 4*2; | |
1536 | } | |
1537 | ||
1538 | static void free_super1(struct supertype *st); | |
1539 | ||
1540 | #ifndef MDASSEMBLE | |
1541 | static int write_init_super1(struct supertype *st) | |
1542 | { | |
1543 | struct mdp_superblock_1 *sb = st->sb; | |
1544 | struct supertype *refst; | |
1545 | int rfd; | |
1546 | int rv = 0; | |
1547 | unsigned long long bm_space; | |
1548 | struct devinfo *di; | |
1549 | unsigned long long dsize, array_size; | |
1550 | unsigned long long sb_offset; | |
1551 | unsigned long long data_offset; | |
1552 | ||
1553 | for (di = st->info; di; di = di->next) { | |
1554 | if (di->disk.state & (1 << MD_DISK_FAULTY)) | |
1555 | continue; | |
1556 | if (di->fd < 0) | |
1557 | continue; | |
1558 | ||
1559 | while (Kill(di->devname, NULL, 0, -1, 1) == 0) | |
1560 | ; | |
1561 | ||
1562 | sb->dev_number = __cpu_to_le32(di->disk.number); | |
1563 | if (di->disk.state & (1<<MD_DISK_WRITEMOSTLY)) | |
1564 | sb->devflags |= WriteMostly1; | |
1565 | else | |
1566 | sb->devflags &= ~WriteMostly1; | |
1567 | ||
1568 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
1569 | read(rfd, sb->device_uuid, 16) != 16) { | |
1570 | __u32 r[4] = {random(), random(), random(), random()}; | |
1571 | memcpy(sb->device_uuid, r, 16); | |
1572 | } | |
1573 | if (rfd >= 0) | |
1574 | close(rfd); | |
1575 | ||
1576 | sb->events = 0; | |
1577 | ||
1578 | refst = dup_super(st); | |
1579 | if (load_super1(refst, di->fd, NULL)==0) { | |
1580 | struct mdp_superblock_1 *refsb = refst->sb; | |
1581 | ||
1582 | memcpy(sb->device_uuid, refsb->device_uuid, 16); | |
1583 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) { | |
1584 | /* same array, so preserve events and | |
1585 | * dev_number */ | |
1586 | sb->events = refsb->events; | |
1587 | /* bugs in 2.6.17 and earlier mean the | |
1588 | * dev_number chosen in Manage must be preserved | |
1589 | */ | |
1590 | if (get_linux_version() >= 2006018) | |
1591 | sb->dev_number = refsb->dev_number; | |
1592 | } | |
1593 | free_super1(refst); | |
1594 | } | |
1595 | free(refst); | |
1596 | ||
1597 | if (!get_dev_size(di->fd, NULL, &dsize)) { | |
1598 | rv = 1; | |
1599 | goto error_out; | |
1600 | } | |
1601 | dsize >>= 9; | |
1602 | ||
1603 | if (dsize < 24) { | |
1604 | close(di->fd); | |
1605 | rv = 2; | |
1606 | goto error_out; | |
1607 | } | |
1608 | ||
1609 | /* | |
1610 | * Calculate the position of the superblock. | |
1611 | * It is always aligned to a 4K boundary and | |
1612 | * depending on minor_version, it can be: | |
1613 | * 0: At least 8K, but less than 12K, from end of device | |
1614 | * 1: At start of device | |
1615 | * 2: 4K from start of device. | |
1616 | * data_offset has already been set. | |
1617 | */ | |
1618 | array_size = __le64_to_cpu(sb->size); | |
1619 | /* work out how much space we left for a bitmap, | |
1620 | * Add 8 sectors for bad block log */ | |
1621 | bm_space = choose_bm_space(array_size) + 8; | |
1622 | ||
1623 | data_offset = di->data_offset; | |
1624 | if (data_offset == INVALID_SECTORS) | |
1625 | data_offset = st->data_offset; | |
1626 | switch(st->minor_version) { | |
1627 | case 0: | |
1628 | if (data_offset == INVALID_SECTORS) | |
1629 | data_offset = 0; | |
1630 | sb_offset = dsize; | |
1631 | sb_offset -= 8*2; | |
1632 | sb_offset &= ~(4*2-1); | |
1633 | sb->data_offset = __cpu_to_le64(data_offset); | |
1634 | sb->super_offset = __cpu_to_le64(sb_offset); | |
1635 | if (sb_offset < array_size + bm_space) | |
1636 | bm_space = sb_offset - array_size; | |
1637 | sb->data_size = __cpu_to_le64(sb_offset - bm_space); | |
1638 | if (bm_space >= 8) { | |
1639 | sb->bblog_size = __cpu_to_le16(8); | |
1640 | sb->bblog_offset = __cpu_to_le32((unsigned)-8); | |
1641 | } | |
1642 | break; | |
1643 | case 1: | |
1644 | sb->super_offset = __cpu_to_le64(0); | |
1645 | if (data_offset == INVALID_SECTORS) | |
1646 | data_offset = 16; | |
1647 | ||
1648 | sb->data_offset = __cpu_to_le64(data_offset); | |
1649 | sb->data_size = __cpu_to_le64(dsize - data_offset); | |
1650 | if (data_offset >= 8 + 32*2 + 8) { | |
1651 | sb->bblog_size = __cpu_to_le16(8); | |
1652 | sb->bblog_offset = __cpu_to_le32(8 + 32*2); | |
1653 | } else if (data_offset >= 16) { | |
1654 | sb->bblog_size = __cpu_to_le16(8); | |
1655 | sb->bblog_offset = __cpu_to_le32(data_offset-8); | |
1656 | } | |
1657 | break; | |
1658 | case 2: | |
1659 | sb_offset = 4*2; | |
1660 | sb->super_offset = __cpu_to_le64(sb_offset); | |
1661 | if (data_offset == INVALID_SECTORS) | |
1662 | data_offset = 24; | |
1663 | ||
1664 | sb->data_offset = __cpu_to_le64(data_offset); | |
1665 | sb->data_size = __cpu_to_le64(dsize - data_offset); | |
1666 | if (data_offset >= 16 + 32*2 + 8) { | |
1667 | sb->bblog_size = __cpu_to_le16(8); | |
1668 | sb->bblog_offset = __cpu_to_le32(8 + 32*2); | |
1669 | } else if (data_offset >= 16+16) { | |
1670 | sb->bblog_size = __cpu_to_le16(8); | |
1671 | /* '8' sectors for the bblog, and another '8' | |
1672 | * because we want offset from superblock, not | |
1673 | * start of device. | |
1674 | */ | |
1675 | sb->bblog_offset = __cpu_to_le32(data_offset-8-8); | |
1676 | } | |
1677 | break; | |
1678 | default: | |
1679 | pr_err("Failed to write invalid metadata format 1.%i to %s\n", | |
1680 | st->minor_version, di->devname); | |
1681 | rv = -EINVAL; | |
1682 | goto out; | |
1683 | } | |
1684 | if (conf_get_create_info()->bblist == 0) { | |
1685 | sb->bblog_size = 0; | |
1686 | sb->bblog_offset = 0; | |
1687 | } | |
1688 | ||
1689 | sb->sb_csum = calc_sb_1_csum(sb); | |
1690 | rv = store_super1(st, di->fd); | |
1691 | if (rv == 0 && (__le32_to_cpu(sb->feature_map) & 1)) | |
1692 | rv = st->ss->write_bitmap(st, di->fd); | |
1693 | close(di->fd); | |
1694 | di->fd = -1; | |
1695 | if (rv) | |
1696 | goto error_out; | |
1697 | } | |
1698 | error_out: | |
1699 | if (rv) | |
1700 | pr_err("Failed to write metadata to %s\n", | |
1701 | di->devname); | |
1702 | out: | |
1703 | return rv; | |
1704 | } | |
1705 | #endif | |
1706 | ||
1707 | static int compare_super1(struct supertype *st, struct supertype *tst) | |
1708 | { | |
1709 | /* | |
1710 | * return: | |
1711 | * 0 same, or first was empty, and second was copied | |
1712 | * 1 second had wrong number | |
1713 | * 2 wrong uuid | |
1714 | * 3 wrong other info | |
1715 | */ | |
1716 | struct mdp_superblock_1 *first = st->sb; | |
1717 | struct mdp_superblock_1 *second = tst->sb; | |
1718 | ||
1719 | if (second->magic != __cpu_to_le32(MD_SB_MAGIC)) | |
1720 | return 1; | |
1721 | if (second->major_version != __cpu_to_le32(1)) | |
1722 | return 1; | |
1723 | ||
1724 | if (!first) { | |
1725 | if (posix_memalign((void**)&first, 4096, SUPER1_SIZE) != 0) { | |
1726 | pr_err("could not allocate superblock\n"); | |
1727 | return 1; | |
1728 | } | |
1729 | memcpy(first, second, SUPER1_SIZE); | |
1730 | st->sb = first; | |
1731 | return 0; | |
1732 | } | |
1733 | if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0) | |
1734 | return 2; | |
1735 | ||
1736 | if (first->ctime != second->ctime || | |
1737 | first->level != second->level || | |
1738 | first->layout != second->layout || | |
1739 | first->size != second->size || | |
1740 | first->chunksize != second->chunksize || | |
1741 | first->raid_disks != second->raid_disks) | |
1742 | return 3; | |
1743 | return 0; | |
1744 | } | |
1745 | ||
1746 | static int load_super1(struct supertype *st, int fd, char *devname) | |
1747 | { | |
1748 | unsigned long long dsize; | |
1749 | unsigned long long sb_offset; | |
1750 | struct mdp_superblock_1 *super; | |
1751 | int uuid[4]; | |
1752 | struct bitmap_super_s *bsb; | |
1753 | struct misc_dev_info *misc; | |
1754 | struct align_fd afd; | |
1755 | ||
1756 | free_super1(st); | |
1757 | ||
1758 | init_afd(&afd, fd); | |
1759 | ||
1760 | if (st->ss == NULL || st->minor_version == -1) { | |
1761 | int bestvers = -1; | |
1762 | struct supertype tst; | |
1763 | __u64 bestctime = 0; | |
1764 | /* guess... choose latest ctime */ | |
1765 | memset(&tst, 0, sizeof(tst)); | |
1766 | tst.ss = &super1; | |
1767 | for (tst.minor_version = 0; tst.minor_version <= 2 ; tst.minor_version++) { | |
1768 | switch(load_super1(&tst, fd, devname)) { | |
1769 | case 0: super = tst.sb; | |
1770 | if (bestvers == -1 || | |
1771 | bestctime < __le64_to_cpu(super->ctime)) { | |
1772 | bestvers = tst.minor_version; | |
1773 | bestctime = __le64_to_cpu(super->ctime); | |
1774 | } | |
1775 | free(super); | |
1776 | tst.sb = NULL; | |
1777 | break; | |
1778 | case 1: return 1; /*bad device */ | |
1779 | case 2: break; /* bad, try next */ | |
1780 | } | |
1781 | } | |
1782 | if (bestvers != -1) { | |
1783 | int rv; | |
1784 | tst.minor_version = bestvers; | |
1785 | tst.ss = &super1; | |
1786 | tst.max_devs = MAX_DEVS; | |
1787 | rv = load_super1(&tst, fd, devname); | |
1788 | if (rv == 0) | |
1789 | *st = tst; | |
1790 | return rv; | |
1791 | } | |
1792 | return 2; | |
1793 | } | |
1794 | if (!get_dev_size(fd, devname, &dsize)) | |
1795 | return 1; | |
1796 | dsize >>= 9; | |
1797 | ||
1798 | if (dsize < 24) { | |
1799 | if (devname) | |
1800 | pr_err("%s is too small for md: size is %llu sectors.\n", | |
1801 | devname, dsize); | |
1802 | return 1; | |
1803 | } | |
1804 | ||
1805 | /* | |
1806 | * Calculate the position of the superblock. | |
1807 | * It is always aligned to a 4K boundary and | |
1808 | * depending on minor_version, it can be: | |
1809 | * 0: At least 8K, but less than 12K, from end of device | |
1810 | * 1: At start of device | |
1811 | * 2: 4K from start of device. | |
1812 | */ | |
1813 | switch(st->minor_version) { | |
1814 | case 0: | |
1815 | sb_offset = dsize; | |
1816 | sb_offset -= 8*2; | |
1817 | sb_offset &= ~(4*2-1); | |
1818 | break; | |
1819 | case 1: | |
1820 | sb_offset = 0; | |
1821 | break; | |
1822 | case 2: | |
1823 | sb_offset = 4*2; | |
1824 | break; | |
1825 | default: | |
1826 | return -EINVAL; | |
1827 | } | |
1828 | ||
1829 | if (lseek64(fd, sb_offset << 9, 0)< 0LL) { | |
1830 | if (devname) | |
1831 | pr_err("Cannot seek to superblock on %s: %s\n", | |
1832 | devname, strerror(errno)); | |
1833 | return 1; | |
1834 | } | |
1835 | ||
1836 | if (posix_memalign((void**)&super, 4096, SUPER1_SIZE) != 0) { | |
1837 | pr_err("could not allocate superblock\n"); | |
1838 | return 1; | |
1839 | } | |
1840 | ||
1841 | if (aread(&afd, super, MAX_SB_SIZE) != MAX_SB_SIZE) { | |
1842 | if (devname) | |
1843 | pr_err("Cannot read superblock on %s\n", | |
1844 | devname); | |
1845 | free(super); | |
1846 | return 1; | |
1847 | } | |
1848 | ||
1849 | if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) { | |
1850 | if (devname) | |
1851 | pr_err("No super block found on %s (Expected magic %08x, got %08x)\n", | |
1852 | devname, MD_SB_MAGIC, __le32_to_cpu(super->magic)); | |
1853 | free(super); | |
1854 | return 2; | |
1855 | } | |
1856 | ||
1857 | if (__le32_to_cpu(super->major_version) != 1) { | |
1858 | if (devname) | |
1859 | pr_err("Cannot interpret superblock on %s - version is %d\n", | |
1860 | devname, __le32_to_cpu(super->major_version)); | |
1861 | free(super); | |
1862 | return 2; | |
1863 | } | |
1864 | if (__le64_to_cpu(super->super_offset) != sb_offset) { | |
1865 | if (devname) | |
1866 | pr_err("No superblock found on %s (super_offset is wrong)\n", | |
1867 | devname); | |
1868 | free(super); | |
1869 | return 2; | |
1870 | } | |
1871 | st->sb = super; | |
1872 | ||
1873 | bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE); | |
1874 | ||
1875 | misc = (struct misc_dev_info*) (((char*)super)+MAX_SB_SIZE+BM_SUPER_SIZE); | |
1876 | misc->device_size = dsize; | |
1877 | if (st->data_offset == INVALID_SECTORS) | |
1878 | st->data_offset = __le64_to_cpu(super->data_offset); | |
1879 | ||
1880 | /* Now check on the bitmap superblock */ | |
1881 | if ((__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) == 0) | |
1882 | return 0; | |
1883 | /* Read the bitmap superblock and make sure it looks | |
1884 | * valid. If it doesn't clear the bit. An --assemble --force | |
1885 | * should get that written out. | |
1886 | */ | |
1887 | locate_bitmap1(st, fd); | |
1888 | if (aread(&afd, bsb, 512) != 512) | |
1889 | goto no_bitmap; | |
1890 | ||
1891 | uuid_from_super1(st, uuid); | |
1892 | if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC || | |
1893 | memcmp(bsb->uuid, uuid, 16) != 0) | |
1894 | goto no_bitmap; | |
1895 | return 0; | |
1896 | ||
1897 | no_bitmap: | |
1898 | super->feature_map = __cpu_to_le32(__le32_to_cpu(super->feature_map) | |
1899 | & ~MD_FEATURE_BITMAP_OFFSET); | |
1900 | return 0; | |
1901 | } | |
1902 | ||
1903 | static struct supertype *match_metadata_desc1(char *arg) | |
1904 | { | |
1905 | struct supertype *st = xcalloc(1, sizeof(*st)); | |
1906 | ||
1907 | st->container_devnm[0] = 0; | |
1908 | st->ss = &super1; | |
1909 | st->max_devs = MAX_DEVS; | |
1910 | st->sb = NULL; | |
1911 | st->data_offset = INVALID_SECTORS; | |
1912 | /* leading zeros can be safely ignored. --detail generates them. */ | |
1913 | while (*arg == '0') | |
1914 | arg++; | |
1915 | if (strcmp(arg, "1.0") == 0 || | |
1916 | strcmp(arg, "1.00") == 0) { | |
1917 | st->minor_version = 0; | |
1918 | return st; | |
1919 | } | |
1920 | if (strcmp(arg, "1.1") == 0 || | |
1921 | strcmp(arg, "1.01") == 0 | |
1922 | ) { | |
1923 | st->minor_version = 1; | |
1924 | return st; | |
1925 | } | |
1926 | if (strcmp(arg, "1.2") == 0 || | |
1927 | #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */ | |
1928 | strcmp(arg, "default") == 0 || | |
1929 | #endif /* DEFAULT_OLD_METADATA */ | |
1930 | strcmp(arg, "1.02") == 0) { | |
1931 | st->minor_version = 2; | |
1932 | return st; | |
1933 | } | |
1934 | if (strcmp(arg, "1") == 0 || | |
1935 | strcmp(arg, "default") == 0) { | |
1936 | st->minor_version = -1; | |
1937 | return st; | |
1938 | } | |
1939 | ||
1940 | free(st); | |
1941 | return NULL; | |
1942 | } | |
1943 | ||
1944 | /* find available size on device with this devsize, using | |
1945 | * superblock type st, and reserving 'reserve' sectors for | |
1946 | * a possible bitmap | |
1947 | */ | |
1948 | static __u64 avail_size1(struct supertype *st, __u64 devsize, | |
1949 | unsigned long long data_offset) | |
1950 | { | |
1951 | struct mdp_superblock_1 *super = st->sb; | |
1952 | int bmspace = 0; | |
1953 | int bbspace = 0; | |
1954 | if (devsize < 24) | |
1955 | return 0; | |
1956 | ||
1957 | #ifndef MDASSEMBLE | |
1958 | if (__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) { | |
1959 | /* hot-add. allow for actual size of bitmap */ | |
1960 | struct bitmap_super_s *bsb; | |
1961 | bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE); | |
1962 | bmspace = bitmap_sectors(bsb); | |
1963 | } | |
1964 | #endif | |
1965 | /* Allow space for bad block log */ | |
1966 | if (super->bblog_size) | |
1967 | bbspace = __le16_to_cpu(super->bblog_size); | |
1968 | ||
1969 | if (st->minor_version < 0) | |
1970 | /* not specified, so time to set default */ | |
1971 | st->minor_version = 2; | |
1972 | ||
1973 | if (data_offset == INVALID_SECTORS) | |
1974 | data_offset = st->data_offset; | |
1975 | ||
1976 | if (data_offset != INVALID_SECTORS) | |
1977 | switch(st->minor_version) { | |
1978 | case 0: | |
1979 | return devsize - data_offset - 8*2 - bbspace; | |
1980 | case 1: | |
1981 | case 2: | |
1982 | return devsize - data_offset; | |
1983 | default: | |
1984 | return 0; | |
1985 | } | |
1986 | ||
1987 | devsize -= bmspace; | |
1988 | ||
1989 | switch(st->minor_version) { | |
1990 | case 0: | |
1991 | /* at end */ | |
1992 | return ((devsize - 8*2 - bbspace ) & ~(4*2-1)); | |
1993 | case 1: | |
1994 | /* at start, 4K for superblock and possible bitmap */ | |
1995 | return devsize - 4*2 - bbspace; | |
1996 | case 2: | |
1997 | /* 4k from start, 4K for superblock and possible bitmap */ | |
1998 | return devsize - (4+4)*2 - bbspace; | |
1999 | } | |
2000 | return 0; | |
2001 | } | |
2002 | ||
2003 | static int | |
2004 | add_internal_bitmap1(struct supertype *st, | |
2005 | int *chunkp, int delay, int write_behind, | |
2006 | unsigned long long size, | |
2007 | int may_change, int major) | |
2008 | { | |
2009 | /* | |
2010 | * If not may_change, then this is a 'Grow' without sysfs support for | |
2011 | * bitmaps, and the bitmap must fit after the superblock at 1K offset. | |
2012 | * If may_change, then this is create or a Grow with sysfs syupport, | |
2013 | * and we can put the bitmap wherever we like. | |
2014 | * | |
2015 | * size is in sectors, chunk is in bytes !!! | |
2016 | */ | |
2017 | ||
2018 | unsigned long long bits; | |
2019 | unsigned long long max_bits; | |
2020 | unsigned long long min_chunk; | |
2021 | long offset; | |
2022 | long bbl_offset, bbl_size; | |
2023 | unsigned long long chunk = *chunkp; | |
2024 | int room = 0; | |
2025 | int creating = 0; | |
2026 | struct mdp_superblock_1 *sb = st->sb; | |
2027 | bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE); | |
2028 | int uuid[4]; | |
2029 | ||
2030 | if (__le64_to_cpu(sb->data_size) == 0) | |
2031 | /* Must be creating the array, else data_size would be non-zero */ | |
2032 | creating = 1; | |
2033 | switch(st->minor_version) { | |
2034 | case 0: | |
2035 | /* either 3K after the superblock (when hot-add), | |
2036 | * or some amount of space before. | |
2037 | */ | |
2038 | if (creating) { | |
2039 | /* We are creating array, so we *know* how much room has | |
2040 | * been left. | |
2041 | */ | |
2042 | offset = 0; | |
2043 | bbl_size = 8; | |
2044 | room = choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size; | |
2045 | } else { | |
2046 | room = __le64_to_cpu(sb->super_offset) | |
2047 | - __le64_to_cpu(sb->data_offset) | |
2048 | - __le64_to_cpu(sb->data_size); | |
2049 | bbl_size = __le16_to_cpu(sb->bblog_size); | |
2050 | if (bbl_size < 8) | |
2051 | bbl_size = 8; | |
2052 | bbl_offset = (__s32)__le32_to_cpu(sb->bblog_offset); | |
2053 | if (bbl_size < -bbl_offset) | |
2054 | bbl_size = -bbl_offset; | |
2055 | ||
2056 | if (!may_change || (room < 3*2 && | |
2057 | __le32_to_cpu(sb->max_dev) <= 384)) { | |
2058 | room = 3*2; | |
2059 | offset = 1*2; | |
2060 | bbl_size = 0; | |
2061 | } else { | |
2062 | offset = 0; /* means movable offset */ | |
2063 | } | |
2064 | } | |
2065 | break; | |
2066 | case 1: | |
2067 | case 2: /* between superblock and data */ | |
2068 | if (creating) { | |
2069 | offset = 4*2; | |
2070 | bbl_size = 8; | |
2071 | room = choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size; | |
2072 | } else { | |
2073 | room = __le64_to_cpu(sb->data_offset) | |
2074 | - __le64_to_cpu(sb->super_offset); | |
2075 | bbl_size = __le16_to_cpu(sb->bblog_size); | |
2076 | if (bbl_size) | |
2077 | room = __le32_to_cpu(sb->bblog_offset) + bbl_size; | |
2078 | else | |
2079 | bbl_size = 8; | |
2080 | ||
2081 | if (!may_change) { | |
2082 | room -= 2; /* Leave 1K for superblock */ | |
2083 | offset = 2; | |
2084 | bbl_size = 0; | |
2085 | } else { | |
2086 | room -= 4*2; /* leave 4K for superblock */ | |
2087 | offset = 4*2; | |
2088 | } | |
2089 | } | |
2090 | break; | |
2091 | default: | |
2092 | return 0; | |
2093 | } | |
2094 | ||
2095 | room -= bbl_size; | |
2096 | if (chunk == UnSet && room > 128*2) | |
2097 | /* Limit to 128K of bitmap when chunk size not requested */ | |
2098 | room = 128*2; | |
2099 | ||
2100 | if (room <= 1) | |
2101 | /* No room for a bitmap */ | |
2102 | return 0; | |
2103 | ||
2104 | max_bits = (room * 512 - sizeof(bitmap_super_t)) * 8; | |
2105 | ||
2106 | min_chunk = 4096; /* sub-page chunks don't work yet.. */ | |
2107 | bits = (size*512)/min_chunk +1; | |
2108 | while (bits > max_bits) { | |
2109 | min_chunk *= 2; | |
2110 | bits = (bits+1)/2; | |
2111 | } | |
2112 | if (chunk == UnSet) { | |
2113 | /* For practical purpose, 64Meg is a good | |
2114 | * default chunk size for internal bitmaps. | |
2115 | */ | |
2116 | chunk = min_chunk; | |
2117 | if (chunk < 64*1024*1024) | |
2118 | chunk = 64*1024*1024; | |
2119 | } else if (chunk < min_chunk) | |
2120 | return 0; /* chunk size too small */ | |
2121 | if (chunk == 0) /* rounding problem */ | |
2122 | return 0; | |
2123 | ||
2124 | if (offset == 0) { | |
2125 | /* start bitmap on a 4K boundary with enough space for | |
2126 | * the bitmap | |
2127 | */ | |
2128 | bits = (size*512) / chunk + 1; | |
2129 | room = ((bits+7)/8 + sizeof(bitmap_super_t) +4095)/4096; | |
2130 | room *= 8; /* convert 4K blocks to sectors */ | |
2131 | offset = -room - bbl_size; | |
2132 | } | |
2133 | ||
2134 | sb->bitmap_offset = (int32_t)__cpu_to_le32(offset); | |
2135 | ||
2136 | sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map) | |
2137 | | MD_FEATURE_BITMAP_OFFSET); | |
2138 | memset(bms, 0, sizeof(*bms)); | |
2139 | bms->magic = __cpu_to_le32(BITMAP_MAGIC); | |
2140 | bms->version = __cpu_to_le32(major); | |
2141 | uuid_from_super1(st, uuid); | |
2142 | memcpy(bms->uuid, uuid, 16); | |
2143 | bms->chunksize = __cpu_to_le32(chunk); | |
2144 | bms->daemon_sleep = __cpu_to_le32(delay); | |
2145 | bms->sync_size = __cpu_to_le64(size); | |
2146 | bms->write_behind = __cpu_to_le32(write_behind); | |
2147 | ||
2148 | *chunkp = chunk; | |
2149 | return 1; | |
2150 | } | |
2151 | ||
2152 | static void locate_bitmap1(struct supertype *st, int fd) | |
2153 | { | |
2154 | unsigned long long offset; | |
2155 | struct mdp_superblock_1 *sb; | |
2156 | int mustfree = 0; | |
2157 | ||
2158 | if (!st->sb) { | |
2159 | if (st->ss->load_super(st, fd, NULL)) | |
2160 | return; /* no error I hope... */ | |
2161 | mustfree = 1; | |
2162 | } | |
2163 | sb = st->sb; | |
2164 | ||
2165 | offset = __le64_to_cpu(sb->super_offset); | |
2166 | offset += (int32_t) __le32_to_cpu(sb->bitmap_offset); | |
2167 | if (mustfree) | |
2168 | free(sb); | |
2169 | lseek64(fd, offset<<9, 0); | |
2170 | } | |
2171 | ||
2172 | static int write_bitmap1(struct supertype *st, int fd) | |
2173 | { | |
2174 | struct mdp_superblock_1 *sb = st->sb; | |
2175 | bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE); | |
2176 | int rv = 0; | |
2177 | void *buf; | |
2178 | int towrite, n; | |
2179 | struct align_fd afd; | |
2180 | ||
2181 | init_afd(&afd, fd); | |
2182 | ||
2183 | locate_bitmap1(st, fd); | |
2184 | ||
2185 | if (posix_memalign(&buf, 4096, 4096)) | |
2186 | return -ENOMEM; | |
2187 | ||
2188 | memset(buf, 0xff, 4096); | |
2189 | memcpy(buf, (char *)bms, sizeof(bitmap_super_t)); | |
2190 | ||
2191 | towrite = __le64_to_cpu(bms->sync_size) / (__le32_to_cpu(bms->chunksize)>>9); | |
2192 | towrite = (towrite+7) >> 3; /* bits to bytes */ | |
2193 | towrite += sizeof(bitmap_super_t); | |
2194 | towrite = ROUND_UP(towrite, 512); | |
2195 | while (towrite > 0) { | |
2196 | n = towrite; | |
2197 | if (n > 4096) | |
2198 | n = 4096; | |
2199 | n = awrite(&afd, buf, n); | |
2200 | if (n > 0) | |
2201 | towrite -= n; | |
2202 | else | |
2203 | break; | |
2204 | memset(buf, 0xff, 4096); | |
2205 | } | |
2206 | fsync(fd); | |
2207 | if (towrite) | |
2208 | rv = -2; | |
2209 | ||
2210 | free(buf); | |
2211 | return rv; | |
2212 | } | |
2213 | ||
2214 | static void free_super1(struct supertype *st) | |
2215 | { | |
2216 | if (st->sb) | |
2217 | free(st->sb); | |
2218 | while (st->info) { | |
2219 | struct devinfo *di = st->info; | |
2220 | st->info = di->next; | |
2221 | if (di->fd >= 0) | |
2222 | close(di->fd); | |
2223 | free(di); | |
2224 | } | |
2225 | st->sb = NULL; | |
2226 | } | |
2227 | ||
2228 | #ifndef MDASSEMBLE | |
2229 | static int validate_geometry1(struct supertype *st, int level, | |
2230 | int layout, int raiddisks, | |
2231 | int *chunk, unsigned long long size, | |
2232 | unsigned long long data_offset, | |
2233 | char *subdev, unsigned long long *freesize, | |
2234 | int verbose) | |
2235 | { | |
2236 | unsigned long long ldsize, devsize; | |
2237 | int bmspace; | |
2238 | unsigned long long headroom; | |
2239 | int fd; | |
2240 | ||
2241 | if (level == LEVEL_CONTAINER) { | |
2242 | if (verbose) | |
2243 | pr_err("1.x metadata does not support containers\n"); | |
2244 | return 0; | |
2245 | } | |
2246 | if (*chunk == UnSet) | |
2247 | *chunk = DEFAULT_CHUNK; | |
2248 | ||
2249 | if (!subdev) | |
2250 | return 1; | |
2251 | ||
2252 | if (st->minor_version < 0) | |
2253 | /* not specified, so time to set default */ | |
2254 | st->minor_version = 2; | |
2255 | ||
2256 | fd = open(subdev, O_RDONLY|O_EXCL, 0); | |
2257 | if (fd < 0) { | |
2258 | if (verbose) | |
2259 | pr_err("super1.x cannot open %s: %s\n", | |
2260 | subdev, strerror(errno)); | |
2261 | return 0; | |
2262 | } | |
2263 | ||
2264 | if (!get_dev_size(fd, subdev, &ldsize)) { | |
2265 | close(fd); | |
2266 | return 0; | |
2267 | } | |
2268 | close(fd); | |
2269 | ||
2270 | devsize = ldsize >> 9; | |
2271 | if (devsize < 24) { | |
2272 | *freesize = 0; | |
2273 | return 0; | |
2274 | } | |
2275 | ||
2276 | /* creating: allow suitable space for bitmap */ | |
2277 | bmspace = choose_bm_space(devsize); | |
2278 | ||
2279 | if (data_offset == INVALID_SECTORS) | |
2280 | data_offset = st->data_offset; | |
2281 | if (data_offset == INVALID_SECTORS) | |
2282 | switch (st->minor_version) { | |
2283 | case 0: | |
2284 | data_offset = 0; | |
2285 | break; | |
2286 | case 1: | |
2287 | case 2: | |
2288 | /* Choose data offset appropriate for this device | |
2289 | * and use as default for whole array. | |
2290 | * The data_offset must allow for bitmap space | |
2291 | * and base metadata, should allow for some headroom | |
2292 | * for reshape, and should be rounded to multiple | |
2293 | * of 1M. | |
2294 | * Headroom is limited to 128M, but aim for about 0.1% | |
2295 | */ | |
2296 | headroom = 128*1024*2; | |
2297 | while ((headroom << 10) > devsize && | |
2298 | (*chunk == 0 || | |
2299 | headroom / 2 >= ((unsigned)(*chunk)*2)*2)) | |
2300 | headroom >>= 1; | |
2301 | data_offset = 12*2 + bmspace + headroom; | |
2302 | #define ONE_MEG (2*1024) | |
2303 | if (data_offset > ONE_MEG) | |
2304 | data_offset = (data_offset / ONE_MEG) * ONE_MEG; | |
2305 | break; | |
2306 | } | |
2307 | if (st->data_offset == INVALID_SECTORS) | |
2308 | st->data_offset = data_offset; | |
2309 | switch(st->minor_version) { | |
2310 | case 0: /* metadata at end. Round down and subtract space to reserve */ | |
2311 | devsize = (devsize & ~(4ULL*2-1)); | |
2312 | /* space for metadata, bblog, bitmap */ | |
2313 | devsize -= 8*2 + 8 + bmspace; | |
2314 | break; | |
2315 | case 1: | |
2316 | case 2: | |
2317 | devsize -= data_offset; | |
2318 | break; | |
2319 | } | |
2320 | *freesize = devsize; | |
2321 | return 1; | |
2322 | } | |
2323 | #endif /* MDASSEMBLE */ | |
2324 | ||
2325 | void *super1_make_v0(struct supertype *st, struct mdinfo *info, mdp_super_t *sb0) | |
2326 | { | |
2327 | /* Create a v1.0 superblock based on 'info'*/ | |
2328 | void *ret; | |
2329 | struct mdp_superblock_1 *sb; | |
2330 | int i; | |
2331 | int rfd; | |
2332 | unsigned long long offset; | |
2333 | ||
2334 | if (posix_memalign(&ret, 4096, 1024) != 0) | |
2335 | return NULL; | |
2336 | sb = ret; | |
2337 | memset(ret, 0, 1024); | |
2338 | sb->magic = __cpu_to_le32(MD_SB_MAGIC); | |
2339 | sb->major_version = __cpu_to_le32(1); | |
2340 | ||
2341 | copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid); | |
2342 | sprintf(sb->set_name, "%d", sb0->md_minor); | |
2343 | sb->ctime = __cpu_to_le32(info->array.ctime+1); | |
2344 | sb->level = __cpu_to_le32(info->array.level); | |
2345 | sb->layout = __cpu_to_le32(info->array.layout); | |
2346 | sb->size = __cpu_to_le64(info->component_size); | |
2347 | sb->chunksize = __cpu_to_le32(info->array.chunk_size/512); | |
2348 | sb->raid_disks = __cpu_to_le32(info->array.raid_disks); | |
2349 | if (info->array.level > 0) | |
2350 | sb->data_size = sb->size; | |
2351 | else | |
2352 | sb->data_size = st->ss->avail_size(st, st->devsize/512, 0); | |
2353 | sb->resync_offset = MaxSector; | |
2354 | sb->max_dev = __cpu_to_le32(MD_SB_DISKS); | |
2355 | sb->dev_number = __cpu_to_le32(info->disk.number); | |
2356 | sb->utime = __cpu_to_le64(info->array.utime); | |
2357 | ||
2358 | offset = st->devsize/512 - 8*2; | |
2359 | offset &= ~(4*2-1); | |
2360 | sb->super_offset = __cpu_to_le64(offset); | |
2361 | //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset); | |
2362 | ||
2363 | if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 || | |
2364 | read(rfd, sb->device_uuid, 16) != 16) { | |
2365 | __u32 r[4] = {random(), random(), random(), random()}; | |
2366 | memcpy(sb->device_uuid, r, 16); | |
2367 | } | |
2368 | if (rfd >= 0) | |
2369 | close(rfd); | |
2370 | ||
2371 | for (i = 0; i < MD_SB_DISKS; i++) { | |
2372 | int state = sb0->disks[i].state; | |
2373 | sb->dev_roles[i] = 0xFFFF; | |
2374 | if ((state & (1<<MD_DISK_SYNC)) && | |
2375 | !(state & (1<<MD_DISK_FAULTY))) | |
2376 | sb->dev_roles[i] = __cpu_to_le16(sb0->disks[i].raid_disk); | |
2377 | } | |
2378 | sb->sb_csum = calc_sb_1_csum(sb); | |
2379 | return ret; | |
2380 | } | |
2381 | ||
2382 | struct superswitch super1 = { | |
2383 | #ifndef MDASSEMBLE | |
2384 | .examine_super = examine_super1, | |
2385 | .brief_examine_super = brief_examine_super1, | |
2386 | .export_examine_super = export_examine_super1, | |
2387 | .detail_super = detail_super1, | |
2388 | .brief_detail_super = brief_detail_super1, | |
2389 | .export_detail_super = export_detail_super1, | |
2390 | .write_init_super = write_init_super1, | |
2391 | .validate_geometry = validate_geometry1, | |
2392 | .add_to_super = add_to_super1, | |
2393 | .examine_badblocks = examine_badblocks_super1, | |
2394 | .copy_metadata = copy_metadata1, | |
2395 | #endif | |
2396 | .match_home = match_home1, | |
2397 | .uuid_from_super = uuid_from_super1, | |
2398 | .getinfo_super = getinfo_super1, | |
2399 | .container_content = container_content1, | |
2400 | .update_super = update_super1, | |
2401 | .init_super = init_super1, | |
2402 | .store_super = store_super1, | |
2403 | .compare_super = compare_super1, | |
2404 | .load_super = load_super1, | |
2405 | .match_metadata_desc = match_metadata_desc1, | |
2406 | .avail_size = avail_size1, | |
2407 | .add_internal_bitmap = add_internal_bitmap1, | |
2408 | .locate_bitmap = locate_bitmap1, | |
2409 | .write_bitmap = write_bitmap1, | |
2410 | .free_super = free_super1, | |
2411 | #if __BYTE_ORDER == BIG_ENDIAN | |
2412 | .swapuuid = 0, | |
2413 | #else | |
2414 | .swapuuid = 1, | |
2415 | #endif | |
2416 | .name = "1.x", | |
2417 | }; |