]>
Commit | Line | Data |
---|---|---|
e5329c37 NB |
1 | /* |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
ca3b6696 | 4 | * Copyright (C) 2001-2012 Neil Brown <neilb@suse.de> |
e5329c37 NB |
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 | |
e736b623 | 22 | * Email: <neilb@suse.de> |
e5329c37 NB |
23 | */ |
24 | #include "mdadm.h" | |
25 | #include "dlink.h" | |
7236ee7a | 26 | #include <sys/mman.h> |
f9b08fec | 27 | #include <stdint.h> |
e5329c37 NB |
28 | |
29 | #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN) | |
30 | #error no endian defined | |
31 | #endif | |
32 | #include "md_u.h" | |
33 | #include "md_p.h" | |
34 | ||
e9e43ec3 N |
35 | #ifndef offsetof |
36 | #define offsetof(t,f) ((size_t)&(((t*)0)->f)) | |
37 | #endif | |
38 | ||
3f54bd62 AK |
39 | int restore_backup(struct supertype *st, |
40 | struct mdinfo *content, | |
41 | int working_disks, | |
42 | int next_spare, | |
43 | char *backup_file, | |
44 | int verbose) | |
45 | { | |
46 | int i; | |
47 | int *fdlist; | |
48 | struct mdinfo *dev; | |
49 | int err; | |
50 | int disk_count = next_spare + working_disks; | |
51 | ||
52 | dprintf("Called restore_backup()\n"); | |
503975b9 N |
53 | fdlist = xmalloc(sizeof(int) * disk_count); |
54 | ||
a7dec3fd | 55 | enable_fds(next_spare); |
3f54bd62 AK |
56 | for (i = 0; i < next_spare; i++) |
57 | fdlist[i] = -1; | |
58 | for (dev = content->devs; dev; dev = dev->next) { | |
59 | char buf[22]; | |
60 | int fd; | |
61 | sprintf(buf, "%d:%d", | |
62 | dev->disk.major, | |
63 | dev->disk.minor); | |
64 | fd = dev_open(buf, O_RDWR); | |
65 | ||
66 | if (dev->disk.raid_disk >= 0) | |
67 | fdlist[dev->disk.raid_disk] = fd; | |
68 | else | |
69 | fdlist[next_spare++] = fd; | |
70 | } | |
71 | ||
72 | if (st->ss->external && st->ss->recover_backup) | |
73 | err = st->ss->recover_backup(st, content); | |
74 | else | |
75 | err = Grow_restart(st, content, fdlist, next_spare, | |
76 | backup_file, verbose > 0); | |
77 | ||
78 | while (next_spare > 0) { | |
cc7f63e5 N |
79 | next_spare--; |
80 | if (fdlist[next_spare] >= 0) | |
81 | close(fdlist[next_spare]); | |
3f54bd62 AK |
82 | } |
83 | free(fdlist); | |
84 | if (err) { | |
e7b84f9d N |
85 | pr_err("Failed to restore critical" |
86 | " section for reshape - sorry.\n"); | |
3f54bd62 | 87 | if (!backup_file) |
e7b84f9d | 88 | pr_err("Possibly you need" |
3f54bd62 AK |
89 | " to specify a --backup-file\n"); |
90 | return 1; | |
91 | } | |
92 | ||
93 | dprintf("restore_backup() returns status OK.\n"); | |
94 | return 0; | |
95 | } | |
96 | ||
e5329c37 NB |
97 | int Grow_Add_device(char *devname, int fd, char *newdev) |
98 | { | |
99 | /* Add a device to an active array. | |
100 | * Currently, just extend a linear array. | |
101 | * This requires writing a new superblock on the | |
102 | * new device, calling the kernel to add the device, | |
103 | * and if that succeeds, update the superblock on | |
104 | * all other devices. | |
105 | * This means that we need to *find* all other devices. | |
106 | */ | |
4b1ac34b NB |
107 | struct mdinfo info; |
108 | ||
e5329c37 NB |
109 | struct stat stb; |
110 | int nfd, fd2; | |
111 | int d, nd; | |
82d9eba6 | 112 | struct supertype *st = NULL; |
4725bc31 | 113 | char *subarray = NULL; |
e5329c37 | 114 | |
4b1ac34b | 115 | if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) { |
e7b84f9d | 116 | pr_err("cannot get array info for %s\n", devname); |
e5329c37 NB |
117 | return 1; |
118 | } | |
119 | ||
4725bc31 | 120 | if (info.array.level != -1) { |
e7b84f9d | 121 | pr_err("can only add devices to linear arrays\n"); |
4725bc31 N |
122 | return 1; |
123 | } | |
124 | ||
125 | st = super_by_fd(fd, &subarray); | |
82d9eba6 | 126 | if (!st) { |
ca3b6696 N |
127 | pr_err("cannot handle arrays with superblock version %d\n", |
128 | info.array.major_version); | |
f9ce90ba NB |
129 | return 1; |
130 | } | |
131 | ||
4725bc31 | 132 | if (subarray) { |
e7b84f9d | 133 | pr_err("Cannot grow linear sub-arrays yet\n"); |
4725bc31 N |
134 | free(subarray); |
135 | free(st); | |
2641101b | 136 | return 1; |
e5329c37 NB |
137 | } |
138 | ||
6416d527 | 139 | nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT); |
e5329c37 | 140 | if (nfd < 0) { |
e7b84f9d | 141 | pr_err("cannot open %s\n", newdev); |
4725bc31 | 142 | free(st); |
e5329c37 NB |
143 | return 1; |
144 | } | |
145 | fstat(nfd, &stb); | |
146 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
e7b84f9d | 147 | pr_err("%s is not a block device!\n", newdev); |
e5329c37 | 148 | close(nfd); |
4725bc31 | 149 | free(st); |
e5329c37 NB |
150 | return 1; |
151 | } | |
ca3b6696 N |
152 | /* now check out all the devices and make sure we can read the |
153 | * superblock */ | |
4b1ac34b | 154 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
155 | mdu_disk_info_t disk; |
156 | char *dv; | |
157 | ||
4725bc31 N |
158 | st->ss->free_super(st); |
159 | ||
e5329c37 NB |
160 | disk.number = d; |
161 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) { | |
e7b84f9d | 162 | pr_err("cannot get device detail for device %d\n", |
e5329c37 | 163 | d); |
4725bc31 N |
164 | close(nfd); |
165 | free(st); | |
e5329c37 NB |
166 | return 1; |
167 | } | |
16c6fa80 | 168 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 | 169 | if (!dv) { |
e7b84f9d | 170 | pr_err("cannot find device file for device %d\n", |
e5329c37 | 171 | d); |
4725bc31 N |
172 | close(nfd); |
173 | free(st); | |
e5329c37 NB |
174 | return 1; |
175 | } | |
16c6fa80 | 176 | fd2 = dev_open(dv, O_RDWR); |
68fe8c6e | 177 | if (fd2 < 0) { |
e7b84f9d | 178 | pr_err("cannot open device file %s\n", dv); |
4725bc31 N |
179 | close(nfd); |
180 | free(st); | |
e5329c37 NB |
181 | return 1; |
182 | } | |
3da92f27 NB |
183 | |
184 | if (st->ss->load_super(st, fd2, NULL)) { | |
e7b84f9d | 185 | pr_err("cannot find super block on %s\n", dv); |
4725bc31 | 186 | close(nfd); |
e5329c37 | 187 | close(fd2); |
4725bc31 | 188 | free(st); |
e5329c37 NB |
189 | return 1; |
190 | } | |
191 | close(fd2); | |
192 | } | |
193 | /* Ok, looks good. Lets update the superblock and write it out to | |
194 | * newdev. | |
195 | */ | |
aba69144 | 196 | |
4b1ac34b NB |
197 | info.disk.number = d; |
198 | info.disk.major = major(stb.st_rdev); | |
199 | info.disk.minor = minor(stb.st_rdev); | |
200 | info.disk.raid_disk = d; | |
201 | info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
3da92f27 | 202 | st->ss->update_super(st, &info, "linear-grow-new", newdev, |
f752781f | 203 | 0, 0, NULL); |
e5329c37 | 204 | |
3da92f27 | 205 | if (st->ss->store_super(st, nfd)) { |
e7b84f9d | 206 | pr_err("Cannot store new superblock on %s\n", |
f752781f | 207 | newdev); |
e5329c37 NB |
208 | close(nfd); |
209 | return 1; | |
210 | } | |
e5329c37 | 211 | close(nfd); |
4b1ac34b NB |
212 | |
213 | if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) { | |
e7b84f9d | 214 | pr_err("Cannot add new disk to this array\n"); |
e5329c37 NB |
215 | return 1; |
216 | } | |
217 | /* Well, that seems to have worked. | |
218 | * Now go through and update all superblocks | |
219 | */ | |
220 | ||
4b1ac34b | 221 | if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) { |
e7b84f9d | 222 | pr_err("cannot get array info for %s\n", devname); |
e5329c37 NB |
223 | return 1; |
224 | } | |
225 | ||
226 | nd = d; | |
4b1ac34b | 227 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
228 | mdu_disk_info_t disk; |
229 | char *dv; | |
230 | ||
231 | disk.number = d; | |
232 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) { | |
e7b84f9d | 233 | pr_err("cannot get device detail for device %d\n", |
e5329c37 NB |
234 | d); |
235 | return 1; | |
236 | } | |
16c6fa80 | 237 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 | 238 | if (!dv) { |
e7b84f9d | 239 | pr_err("cannot find device file for device %d\n", |
e5329c37 NB |
240 | d); |
241 | return 1; | |
242 | } | |
16c6fa80 | 243 | fd2 = dev_open(dv, O_RDWR); |
e5329c37 | 244 | if (fd2 < 0) { |
e7b84f9d | 245 | pr_err("cannot open device file %s\n", dv); |
e5329c37 NB |
246 | return 1; |
247 | } | |
3da92f27 | 248 | if (st->ss->load_super(st, fd2, NULL)) { |
e7b84f9d | 249 | pr_err("cannot find super block on %s\n", dv); |
e5329c37 NB |
250 | close(fd); |
251 | return 1; | |
252 | } | |
4b1ac34b NB |
253 | info.array.raid_disks = nd+1; |
254 | info.array.nr_disks = nd+1; | |
255 | info.array.active_disks = nd+1; | |
256 | info.array.working_disks = nd+1; | |
f752781f | 257 | |
3da92f27 | 258 | st->ss->update_super(st, &info, "linear-grow-update", dv, |
f752781f | 259 | 0, 0, NULL); |
aba69144 | 260 | |
3da92f27 | 261 | if (st->ss->store_super(st, fd2)) { |
e7b84f9d | 262 | pr_err("Cannot store new superblock on %s\n", dv); |
e5329c37 NB |
263 | close(fd2); |
264 | return 1; | |
265 | } | |
266 | close(fd2); | |
267 | } | |
268 | ||
269 | return 0; | |
270 | } | |
f5e166fe | 271 | |
50f01ba5 | 272 | int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s) |
f5e166fe NB |
273 | { |
274 | /* | |
275 | * First check that array doesn't have a bitmap | |
276 | * Then create the bitmap | |
277 | * Then add it | |
278 | * | |
279 | * For internal bitmaps, we need to check the version, | |
280 | * find all the active devices, and write the bitmap block | |
281 | * to all devices | |
282 | */ | |
283 | mdu_bitmap_file_t bmf; | |
284 | mdu_array_info_t array; | |
285 | struct supertype *st; | |
4725bc31 | 286 | char *subarray = NULL; |
dcec9ee5 NB |
287 | int major = BITMAP_MAJOR_HI; |
288 | int vers = md_get_version(fd); | |
8fac0577 | 289 | unsigned long long bitmapsize, array_size; |
dcec9ee5 NB |
290 | |
291 | if (vers < 9003) { | |
292 | major = BITMAP_MAJOR_HOSTENDIAN; | |
e7b84f9d | 293 | pr_err("Warning - bitmaps created on this kernel" |
b3bd581b N |
294 | " are not portable\n" |
295 | " between different architectures. Consider upgrading" | |
296 | " the Linux kernel.\n"); | |
dcec9ee5 | 297 | } |
f5e166fe NB |
298 | |
299 | if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) { | |
353632d9 | 300 | if (errno == ENOMEM) |
e7b84f9d | 301 | pr_err("Memory allocation failure.\n"); |
f5e166fe | 302 | else |
e7b84f9d | 303 | pr_err("bitmaps not supported by this kernel.\n"); |
f5e166fe NB |
304 | return 1; |
305 | } | |
306 | if (bmf.pathname[0]) { | |
50f01ba5 | 307 | if (strcmp(s->bitmap_file,"none")==0) { |
fe80f49b | 308 | if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) { |
e7b84f9d | 309 | pr_err("failed to remove bitmap %s\n", |
fe80f49b NB |
310 | bmf.pathname); |
311 | return 1; | |
312 | } | |
313 | return 0; | |
314 | } | |
e7b84f9d | 315 | pr_err("%s already has a bitmap (%s)\n", |
f5e166fe NB |
316 | devname, bmf.pathname); |
317 | return 1; | |
318 | } | |
319 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
e7b84f9d | 320 | pr_err("cannot get array status for %s\n", devname); |
f5e166fe NB |
321 | return 1; |
322 | } | |
323 | if (array.state & (1<<MD_SB_BITMAP_PRESENT)) { | |
50f01ba5 | 324 | if (strcmp(s->bitmap_file, "none")==0) { |
fe80f49b NB |
325 | array.state &= ~(1<<MD_SB_BITMAP_PRESENT); |
326 | if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) { | |
e7b84f9d | 327 | pr_err("failed to remove internal bitmap.\n"); |
fe80f49b NB |
328 | return 1; |
329 | } | |
330 | return 0; | |
331 | } | |
e7b84f9d | 332 | pr_err("Internal bitmap already present on %s\n", |
f5e166fe NB |
333 | devname); |
334 | return 1; | |
335 | } | |
4725bc31 | 336 | |
50f01ba5 | 337 | if (strcmp(s->bitmap_file, "none") == 0) { |
e7b84f9d | 338 | pr_err("no bitmap found on %s\n", devname); |
4725bc31 N |
339 | return 1; |
340 | } | |
5b28bd56 | 341 | if (array.level <= 0) { |
e7b84f9d | 342 | pr_err("Bitmaps not meaningful with level %s\n", |
5b28bd56 NB |
343 | map_num(pers, array.level)?:"of this array"); |
344 | return 1; | |
345 | } | |
8fac0577 NB |
346 | bitmapsize = array.size; |
347 | bitmapsize <<= 1; | |
beae1dfe | 348 | if (get_dev_size(fd, NULL, &array_size) && |
8fac0577 NB |
349 | array_size > (0x7fffffffULL<<9)) { |
350 | /* Array is big enough that we cannot trust array.size | |
351 | * try other approaches | |
352 | */ | |
353 | bitmapsize = get_component_size(fd); | |
354 | } | |
8fac0577 | 355 | if (bitmapsize == 0) { |
e7b84f9d | 356 | pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n"); |
8fac0577 NB |
357 | return 1; |
358 | } | |
359 | ||
f9c25f1d | 360 | if (array.level == 10) { |
8686f3ed | 361 | int ncopies = (array.layout&255)*((array.layout>>8)&255); |
f9c25f1d NB |
362 | bitmapsize = bitmapsize * array.raid_disks / ncopies; |
363 | } | |
364 | ||
4725bc31 | 365 | st = super_by_fd(fd, &subarray); |
f5e166fe | 366 | if (!st) { |
e7b84f9d | 367 | pr_err("Cannot understand version %d.%d\n", |
f5e166fe NB |
368 | array.major_version, array.minor_version); |
369 | return 1; | |
370 | } | |
4725bc31 | 371 | if (subarray) { |
e7b84f9d | 372 | pr_err("Cannot add bitmaps to sub-arrays yet\n"); |
4725bc31 N |
373 | free(subarray); |
374 | free(st); | |
fe80f49b | 375 | return 1; |
4725bc31 | 376 | } |
50f01ba5 | 377 | if (strcmp(s->bitmap_file, "internal") == 0) { |
c0c1acd6 | 378 | int rv; |
f5e166fe | 379 | int d; |
c0c1acd6 N |
380 | int offset_setable = 0; |
381 | struct mdinfo *mdi; | |
ebeb3663 | 382 | if (st->ss->add_internal_bitmap == NULL) { |
e7b84f9d | 383 | pr_err("Internal bitmaps not supported " |
ebeb3663 N |
384 | "with %s metadata\n", st->ss->name); |
385 | return 1; | |
386 | } | |
4dd2df09 | 387 | mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION); |
c0c1acd6 N |
388 | if (mdi) |
389 | offset_setable = 1; | |
ea329559 | 390 | for (d=0; d< st->max_devs; d++) { |
f5e166fe NB |
391 | mdu_disk_info_t disk; |
392 | char *dv; | |
393 | disk.number = d; | |
394 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
395 | continue; | |
396 | if (disk.major == 0 && | |
397 | disk.minor == 0) | |
398 | continue; | |
399 | if ((disk.state & (1<<MD_DISK_SYNC))==0) | |
400 | continue; | |
16c6fa80 | 401 | dv = map_dev(disk.major, disk.minor, 1); |
f5e166fe | 402 | if (dv) { |
16c6fa80 | 403 | int fd2 = dev_open(dv, O_RDWR); |
f5e166fe NB |
404 | if (fd2 < 0) |
405 | continue; | |
3da92f27 | 406 | if (st->ss->load_super(st, fd2, NULL)==0) { |
199171a2 | 407 | if (st->ss->add_internal_bitmap( |
3da92f27 | 408 | st, |
50f01ba5 | 409 | &s->bitmap_chunk, c->delay, s->write_behind, |
c0c1acd6 N |
410 | bitmapsize, offset_setable, |
411 | major) | |
199171a2 | 412 | ) |
3da92f27 | 413 | st->ss->write_bitmap(st, fd2); |
21e92547 | 414 | else { |
ca3b6696 N |
415 | pr_err("failed to create internal bitmap" |
416 | " - chunksize problem.\n"); | |
21e92547 NB |
417 | close(fd2); |
418 | return 1; | |
419 | } | |
f5e166fe NB |
420 | } |
421 | close(fd2); | |
422 | } | |
423 | } | |
c0c1acd6 N |
424 | if (offset_setable) { |
425 | st->ss->getinfo_super(st, mdi, NULL); | |
4dd2df09 | 426 | sysfs_init(mdi, fd, NULL); |
012a8641 JS |
427 | rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location", |
428 | mdi->bitmap_offset); | |
c0c1acd6 N |
429 | } else { |
430 | array.state |= (1<<MD_SB_BITMAP_PRESENT); | |
431 | rv = ioctl(fd, SET_ARRAY_INFO, &array); | |
432 | } | |
433 | if (rv < 0) { | |
ff634064 | 434 | if (errno == EBUSY) |
e7b84f9d N |
435 | pr_err("Cannot add bitmap while array is" |
436 | " resyncing or reshaping etc.\n"); | |
437 | pr_err("failed to set internal bitmap.\n"); | |
f5e166fe NB |
438 | return 1; |
439 | } | |
fe80f49b NB |
440 | } else { |
441 | int uuid[4]; | |
442 | int bitmap_fd; | |
443 | int d; | |
444 | int max_devs = st->max_devs; | |
fe80f49b NB |
445 | |
446 | /* try to load a superblock */ | |
ca3b6696 | 447 | for (d = 0; d < max_devs; d++) { |
fe80f49b NB |
448 | mdu_disk_info_t disk; |
449 | char *dv; | |
450 | int fd2; | |
451 | disk.number = d; | |
452 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
453 | continue; | |
454 | if ((disk.major==0 && disk.minor==0) || | |
455 | (disk.state & (1<<MD_DISK_REMOVED))) | |
456 | continue; | |
16c6fa80 | 457 | dv = map_dev(disk.major, disk.minor, 1); |
ca3b6696 N |
458 | if (!dv) |
459 | continue; | |
16c6fa80 | 460 | fd2 = dev_open(dv, O_RDONLY); |
e7344e90 JS |
461 | if (fd2 >= 0) { |
462 | if (st->ss->load_super(st, fd2, NULL) == 0) { | |
463 | close(fd2); | |
464 | st->ss->uuid_from_super(st, uuid); | |
465 | break; | |
466 | } | |
fe80f49b | 467 | close(fd2); |
fe80f49b | 468 | } |
fe80f49b NB |
469 | } |
470 | if (d == max_devs) { | |
e7b84f9d | 471 | pr_err("cannot find UUID for array!\n"); |
fe80f49b NB |
472 | return 1; |
473 | } | |
50f01ba5 N |
474 | if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk, |
475 | c->delay, s->write_behind, bitmapsize, major)) { | |
fe80f49b NB |
476 | return 1; |
477 | } | |
50f01ba5 | 478 | bitmap_fd = open(s->bitmap_file, O_RDWR); |
fe80f49b | 479 | if (bitmap_fd < 0) { |
e7b84f9d | 480 | pr_err("weird: %s cannot be opened\n", |
50f01ba5 | 481 | s->bitmap_file); |
fe80f49b NB |
482 | return 1; |
483 | } | |
484 | if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
ff634064 N |
485 | int err = errno; |
486 | if (errno == EBUSY) | |
e7b84f9d N |
487 | pr_err("Cannot add bitmap while array is" |
488 | " resyncing or reshaping etc.\n"); | |
489 | pr_err("Cannot set bitmap file for %s: %s\n", | |
ff634064 | 490 | devname, strerror(err)); |
fe80f49b NB |
491 | return 1; |
492 | } | |
493 | } | |
f5e166fe NB |
494 | |
495 | return 0; | |
496 | } | |
497 | ||
e86c9dd6 NB |
498 | /* |
499 | * When reshaping an array we might need to backup some data. | |
500 | * This is written to all spares with a 'super_block' describing it. | |
ff94fb86 | 501 | * The superblock goes 4K from the end of the used space on the |
e86c9dd6 NB |
502 | * device. |
503 | * It if written after the backup is complete. | |
504 | * It has the following structure. | |
505 | */ | |
506 | ||
5fdf37e3 | 507 | static struct mdp_backup_super { |
7236ee7a | 508 | char magic[16]; /* md_backup_data-1 or -2 */ |
e86c9dd6 NB |
509 | __u8 set_uuid[16]; |
510 | __u64 mtime; | |
511 | /* start/sizes in 512byte sectors */ | |
7236ee7a | 512 | __u64 devstart; /* address on backup device/file of data */ |
e86c9dd6 NB |
513 | __u64 arraystart; |
514 | __u64 length; | |
515 | __u32 sb_csum; /* csum of preceeding bytes. */ | |
7236ee7a N |
516 | __u32 pad1; |
517 | __u64 devstart2; /* offset in to data of second section */ | |
518 | __u64 arraystart2; | |
519 | __u64 length2; | |
520 | __u32 sb_csum2; /* csum of preceeding bytes. */ | |
521 | __u8 pad[512-68-32]; | |
5fdf37e3 | 522 | } __attribute__((aligned(512))) bsb, bsb2; |
e86c9dd6 | 523 | |
4411fb17 | 524 | static __u32 bsb_csum(char *buf, int len) |
e86c9dd6 NB |
525 | { |
526 | int i; | |
527 | int csum = 0; | |
ca3b6696 | 528 | for (i = 0; i < len; i++) |
e86c9dd6 NB |
529 | csum = (csum<<3) + buf[0]; |
530 | return __cpu_to_le32(csum); | |
531 | } | |
532 | ||
d7ca196c N |
533 | static int check_idle(struct supertype *st) |
534 | { | |
535 | /* Check that all member arrays for this container, or the | |
536 | * container of this array, are idle | |
537 | */ | |
4dd2df09 N |
538 | char *container = (st->container_devnm[0] |
539 | ? st->container_devnm : st->devnm); | |
d7ca196c N |
540 | struct mdstat_ent *ent, *e; |
541 | int is_idle = 1; | |
24daa16f | 542 | |
d7ca196c N |
543 | ent = mdstat_read(0, 0); |
544 | for (e = ent ; e; e = e->next) { | |
545 | if (!is_container_member(e, container)) | |
546 | continue; | |
547 | if (e->percent >= 0) { | |
548 | is_idle = 0; | |
549 | break; | |
550 | } | |
551 | } | |
552 | free_mdstat(ent); | |
553 | return is_idle; | |
554 | } | |
555 | ||
7f2ba464 | 556 | static int freeze_container(struct supertype *st) |
7236ee7a | 557 | { |
4dd2df09 N |
558 | char *container = (st->container_devnm[0] |
559 | ? st->container_devnm : st->devnm); | |
7f2ba464 | 560 | |
d7ca196c N |
561 | if (!check_idle(st)) |
562 | return -1; | |
24daa16f | 563 | |
7f2ba464 | 564 | if (block_monitor(container, 1)) { |
e7b84f9d | 565 | pr_err("failed to freeze container\n"); |
7f2ba464 DW |
566 | return -2; |
567 | } | |
568 | ||
569 | return 1; | |
570 | } | |
571 | ||
572 | static void unfreeze_container(struct supertype *st) | |
573 | { | |
4dd2df09 N |
574 | char *container = (st->container_devnm[0] |
575 | ? st->container_devnm : st->devnm); | |
7f2ba464 DW |
576 | |
577 | unblock_monitor(container, 1); | |
578 | } | |
579 | ||
580 | static int freeze(struct supertype *st) | |
581 | { | |
582 | /* Try to freeze resync/rebuild on this array/container. | |
7236ee7a | 583 | * Return -1 if the array is busy, |
7f2ba464 | 584 | * return -2 container cannot be frozen, |
7236ee7a N |
585 | * return 0 if this kernel doesn't support 'frozen' |
586 | * return 1 if it worked. | |
587 | */ | |
7f2ba464 DW |
588 | if (st->ss->external) |
589 | return freeze_container(st); | |
590 | else { | |
4dd2df09 | 591 | struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION); |
7f2ba464 | 592 | int err; |
815c8a7e | 593 | char buf[20]; |
7f2ba464 DW |
594 | |
595 | if (!sra) | |
596 | return -1; | |
815c8a7e N |
597 | /* Need to clear any 'read-auto' status */ |
598 | if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 && | |
599 | strncmp(buf, "read-auto", 9) == 0) | |
600 | sysfs_set_str(sra, NULL, "array_state", "clean"); | |
601 | ||
7f2ba464 DW |
602 | err = sysfs_freeze_array(sra); |
603 | sysfs_free(sra); | |
604 | return err; | |
605 | } | |
7236ee7a N |
606 | } |
607 | ||
9202b8eb | 608 | static void unfreeze(struct supertype *st) |
7236ee7a | 609 | { |
7f2ba464 DW |
610 | if (st->ss->external) |
611 | return unfreeze_container(st); | |
612 | else { | |
4dd2df09 | 613 | struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION); |
7f2ba464 DW |
614 | |
615 | if (sra) | |
616 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
7f2ba464 DW |
617 | sysfs_free(sra); |
618 | } | |
7236ee7a N |
619 | } |
620 | ||
4411fb17 | 621 | static void wait_reshape(struct mdinfo *sra) |
7236ee7a N |
622 | { |
623 | int fd = sysfs_get_fd(sra, NULL, "sync_action"); | |
624 | char action[20]; | |
625 | ||
92a19f1a AK |
626 | if (fd < 0) |
627 | return; | |
628 | ||
efc67e8e N |
629 | while (sysfs_fd_get_str(fd, action, 20) > 0 && |
630 | strncmp(action, "reshape", 7) == 0) | |
631 | sysfs_wait(fd, NULL); | |
92a19f1a | 632 | close(fd); |
7236ee7a | 633 | } |
7bc71196 | 634 | |
7ec0996c LD |
635 | static int reshape_super(struct supertype *st, unsigned long long size, |
636 | int level, int layout, int chunksize, int raid_disks, | |
41784c88 | 637 | int delta_disks, char *backup_file, char *dev, |
016e00f5 | 638 | int direction, int verbose) |
7bc71196 DW |
639 | { |
640 | /* nothing extra to check in the native case */ | |
641 | if (!st->ss->external) | |
642 | return 0; | |
643 | if (!st->ss->reshape_super || | |
644 | !st->ss->manage_reshape) { | |
e7b84f9d | 645 | pr_err("%s metadata does not support reshape\n", |
7bc71196 DW |
646 | st->ss->name); |
647 | return 1; | |
648 | } | |
649 | ||
650 | return st->ss->reshape_super(st, size, level, layout, chunksize, | |
41784c88 | 651 | raid_disks, delta_disks, backup_file, dev, |
016e00f5 | 652 | direction, verbose); |
7bc71196 DW |
653 | } |
654 | ||
655 | static void sync_metadata(struct supertype *st) | |
656 | { | |
657 | if (st->ss->external) { | |
76266030 | 658 | if (st->update_tail) { |
7bc71196 | 659 | flush_metadata_updates(st); |
76266030 N |
660 | st->update_tail = &st->updates; |
661 | } else | |
7bc71196 DW |
662 | st->ss->sync_metadata(st); |
663 | } | |
664 | } | |
665 | ||
666 | static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n) | |
667 | { | |
668 | /* when dealing with external metadata subarrays we need to be | |
669 | * prepared to handle EAGAIN. The kernel may need to wait for | |
670 | * mdmon to mark the array active so the kernel can handle | |
671 | * allocations/writeback when preparing the reshape action | |
672 | * (md_allow_write()). We temporarily disable safe_mode_delay | |
673 | * to close a race with the array_state going clean before the | |
674 | * next write to raid_disks / stripe_cache_size | |
675 | */ | |
676 | char safe[50]; | |
677 | int rc; | |
678 | ||
679 | /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */ | |
5da9ab98 N |
680 | if (!container || |
681 | (strcmp(name, "raid_disks") != 0 && | |
682 | strcmp(name, "stripe_cache_size") != 0)) | |
7bc71196 DW |
683 | return sysfs_set_num(sra, NULL, name, n); |
684 | ||
685 | rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe)); | |
686 | if (rc <= 0) | |
687 | return -1; | |
688 | sysfs_set_num(sra, NULL, "safe_mode_delay", 0); | |
689 | rc = sysfs_set_num(sra, NULL, name, n); | |
690 | if (rc < 0 && errno == EAGAIN) { | |
691 | ping_monitor(container); | |
692 | /* if we get EAGAIN here then the monitor is not active | |
693 | * so stop trying | |
694 | */ | |
695 | rc = sysfs_set_num(sra, NULL, name, n); | |
696 | } | |
697 | sysfs_set_str(sra, NULL, "safe_mode_delay", safe); | |
698 | return rc; | |
699 | } | |
700 | ||
27a1e5b5 N |
701 | int start_reshape(struct mdinfo *sra, int already_running, |
702 | int before_data_disks, int data_disks) | |
47eb4d5a N |
703 | { |
704 | int err; | |
6937e6d2 AK |
705 | unsigned long long sync_max_to_set; |
706 | ||
0f28668b | 707 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); |
6937e6d2 AK |
708 | err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress); |
709 | err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", | |
710 | sra->reshape_progress); | |
27a1e5b5 N |
711 | if (before_data_disks <= data_disks) |
712 | sync_max_to_set = sra->reshape_progress / data_disks; | |
713 | else | |
714 | sync_max_to_set = (sra->component_size * data_disks | |
715 | - sra->reshape_progress) / data_disks; | |
1e971e71 | 716 | if (!already_running) |
6937e6d2 AK |
717 | sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set); |
718 | err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set); | |
1e971e71 N |
719 | if (!already_running) |
720 | err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape"); | |
47eb4d5a N |
721 | |
722 | return err; | |
723 | } | |
724 | ||
725 | void abort_reshape(struct mdinfo *sra) | |
726 | { | |
727 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
728 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); | |
729 | sysfs_set_num(sra, NULL, "suspend_hi", 0); | |
730 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
731 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
732 | sysfs_set_str(sra, NULL, "sync_max", "max"); | |
733 | } | |
734 | ||
dfe77a9e KW |
735 | int remove_disks_for_takeover(struct supertype *st, |
736 | struct mdinfo *sra, | |
737 | int layout) | |
62a48395 AK |
738 | { |
739 | int nr_of_copies; | |
740 | struct mdinfo *remaining; | |
741 | int slot; | |
742 | ||
dfe77a9e KW |
743 | if (sra->array.level == 10) |
744 | nr_of_copies = layout & 0xff; | |
745 | else if (sra->array.level == 1) | |
746 | nr_of_copies = sra->array.raid_disks; | |
747 | else | |
748 | return 1; | |
62a48395 AK |
749 | |
750 | remaining = sra->devs; | |
751 | sra->devs = NULL; | |
752 | /* for each 'copy', select one device and remove from the list. */ | |
753 | for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) { | |
754 | struct mdinfo **diskp; | |
755 | int found = 0; | |
756 | ||
757 | /* Find a working device to keep */ | |
758 | for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) { | |
759 | struct mdinfo *disk = *diskp; | |
760 | ||
761 | if (disk->disk.raid_disk < slot) | |
762 | continue; | |
763 | if (disk->disk.raid_disk >= slot + nr_of_copies) | |
764 | continue; | |
765 | if (disk->disk.state & (1<<MD_DISK_REMOVED)) | |
766 | continue; | |
767 | if (disk->disk.state & (1<<MD_DISK_FAULTY)) | |
768 | continue; | |
769 | if (!(disk->disk.state & (1<<MD_DISK_SYNC))) | |
770 | continue; | |
771 | ||
772 | /* We have found a good disk to use! */ | |
773 | *diskp = disk->next; | |
774 | disk->next = sra->devs; | |
775 | sra->devs = disk; | |
776 | found = 1; | |
777 | break; | |
778 | } | |
779 | if (!found) | |
780 | break; | |
781 | } | |
782 | ||
783 | if (slot < sra->array.raid_disks) { | |
784 | /* didn't find all slots */ | |
785 | struct mdinfo **e; | |
786 | e = &remaining; | |
787 | while (*e) | |
788 | e = &(*e)->next; | |
789 | *e = sra->devs; | |
790 | sra->devs = remaining; | |
791 | return 1; | |
792 | } | |
793 | ||
794 | /* Remove all 'remaining' devices from the array */ | |
795 | while (remaining) { | |
796 | struct mdinfo *sd = remaining; | |
797 | remaining = sd->next; | |
798 | ||
799 | sysfs_set_str(sra, sd, "state", "faulty"); | |
800 | sysfs_set_str(sra, sd, "slot", "none"); | |
d5ca4a23 KW |
801 | /* for external metadata disks should be removed in mdmon */ |
802 | if (!st->ss->external) | |
803 | sysfs_set_str(sra, sd, "state", "remove"); | |
62a48395 AK |
804 | sd->disk.state |= (1<<MD_DISK_REMOVED); |
805 | sd->disk.state &= ~(1<<MD_DISK_SYNC); | |
806 | sd->next = sra->devs; | |
807 | sra->devs = sd; | |
808 | } | |
809 | return 0; | |
810 | } | |
811 | ||
130994cb AK |
812 | void reshape_free_fdlist(int *fdlist, |
813 | unsigned long long *offsets, | |
814 | int size) | |
815 | { | |
816 | int i; | |
817 | ||
818 | for (i = 0; i < size; i++) | |
819 | if (fdlist[i] >= 0) | |
820 | close(fdlist[i]); | |
821 | ||
822 | free(fdlist); | |
823 | free(offsets); | |
824 | } | |
825 | ||
826 | int reshape_prepare_fdlist(char *devname, | |
827 | struct mdinfo *sra, | |
828 | int raid_disks, | |
829 | int nrdisks, | |
830 | unsigned long blocks, | |
831 | char *backup_file, | |
832 | int *fdlist, | |
833 | unsigned long long *offsets) | |
834 | { | |
835 | int d = 0; | |
836 | struct mdinfo *sd; | |
837 | ||
a7dec3fd | 838 | enable_fds(nrdisks); |
130994cb AK |
839 | for (d = 0; d <= nrdisks; d++) |
840 | fdlist[d] = -1; | |
841 | d = raid_disks; | |
842 | for (sd = sra->devs; sd; sd = sd->next) { | |
843 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
844 | continue; | |
845 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
846 | char *dn = map_dev(sd->disk.major, | |
847 | sd->disk.minor, 1); | |
848 | fdlist[sd->disk.raid_disk] | |
849 | = dev_open(dn, O_RDONLY); | |
850 | offsets[sd->disk.raid_disk] = sd->data_offset*512; | |
851 | if (fdlist[sd->disk.raid_disk] < 0) { | |
e7b84f9d N |
852 | pr_err("%s: cannot open component %s\n", |
853 | devname, dn ? dn : "-unknown-"); | |
130994cb AK |
854 | d = -1; |
855 | goto release; | |
856 | } | |
857 | } else if (backup_file == NULL) { | |
858 | /* spare */ | |
859 | char *dn = map_dev(sd->disk.major, | |
860 | sd->disk.minor, 1); | |
861 | fdlist[d] = dev_open(dn, O_RDWR); | |
862 | offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512; | |
863 | if (fdlist[d] < 0) { | |
e7b84f9d | 864 | pr_err("%s: cannot open component %s\n", |
130994cb AK |
865 | devname, dn ? dn : "-unknown-"); |
866 | d = -1; | |
867 | goto release; | |
868 | } | |
869 | d++; | |
870 | } | |
871 | } | |
872 | release: | |
873 | return d; | |
874 | } | |
875 | ||
e6e9d47b AK |
876 | int reshape_open_backup_file(char *backup_file, |
877 | int fd, | |
878 | char *devname, | |
879 | long blocks, | |
880 | int *fdlist, | |
a93f87ee N |
881 | unsigned long long *offsets, |
882 | int restart) | |
e6e9d47b AK |
883 | { |
884 | /* Return 1 on success, 0 on any form of failure */ | |
885 | /* need to check backup file is large enough */ | |
886 | char buf[512]; | |
887 | struct stat stb; | |
888 | unsigned int dev; | |
889 | int i; | |
890 | ||
a93f87ee | 891 | *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL), |
e6e9d47b AK |
892 | S_IRUSR | S_IWUSR); |
893 | *offsets = 8 * 512; | |
894 | if (*fdlist < 0) { | |
e7b84f9d | 895 | pr_err("%s: cannot create backup file %s: %s\n", |
e6e9d47b AK |
896 | devname, backup_file, strerror(errno)); |
897 | return 0; | |
898 | } | |
899 | /* Guard against backup file being on array device. | |
900 | * If array is partitioned or if LVM etc is in the | |
901 | * way this will not notice, but it is better than | |
902 | * nothing. | |
903 | */ | |
904 | fstat(*fdlist, &stb); | |
905 | dev = stb.st_dev; | |
906 | fstat(fd, &stb); | |
907 | if (stb.st_rdev == dev) { | |
e7b84f9d | 908 | pr_err("backup file must NOT be" |
e6e9d47b AK |
909 | " on the array being reshaped.\n"); |
910 | close(*fdlist); | |
911 | return 0; | |
912 | } | |
913 | ||
914 | memset(buf, 0, 512); | |
ca4fe0bf | 915 | for (i=0; i < blocks + 8 ; i++) { |
e6e9d47b | 916 | if (write(*fdlist, buf, 512) != 512) { |
e7b84f9d | 917 | pr_err("%s: cannot create" |
e6e9d47b AK |
918 | " backup file %s: %s\n", |
919 | devname, backup_file, strerror(errno)); | |
920 | return 0; | |
921 | } | |
922 | } | |
923 | if (fsync(*fdlist) != 0) { | |
e7b84f9d | 924 | pr_err("%s: cannot create backup file %s: %s\n", |
e6e9d47b AK |
925 | devname, backup_file, strerror(errno)); |
926 | return 0; | |
927 | } | |
928 | ||
929 | return 1; | |
930 | } | |
931 | ||
1c009fc2 AK |
932 | unsigned long compute_backup_blocks(int nchunk, int ochunk, |
933 | unsigned int ndata, unsigned int odata) | |
934 | { | |
935 | unsigned long a, b, blocks; | |
936 | /* So how much do we need to backup. | |
937 | * We need an amount of data which is both a whole number of | |
938 | * old stripes and a whole number of new stripes. | |
939 | * So LCM for (chunksize*datadisks). | |
940 | */ | |
941 | a = (ochunk/512) * odata; | |
942 | b = (nchunk/512) * ndata; | |
943 | /* Find GCD */ | |
b48e2e25 | 944 | a = GCD(a, b); |
1c009fc2 AK |
945 | /* LCM == product / GCD */ |
946 | blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a; | |
947 | ||
948 | return blocks; | |
949 | } | |
950 | ||
a73b0081 | 951 | char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re) |
5da9ab98 N |
952 | { |
953 | /* Based on the current array state in info->array and | |
954 | * the changes in info->new_* etc, determine: | |
955 | * - whether the change is possible | |
956 | * - Intermediate level/raid_disks/layout | |
957 | * - whether a restriping reshape is needed | |
958 | * - number of sectors in minimum change unit. This | |
959 | * will cover a whole number of stripes in 'before' and | |
960 | * 'after'. | |
961 | * | |
962 | * Return message if the change should be rejected | |
963 | * NULL if the change can be achieved | |
964 | * | |
965 | * This can be called as part of starting a reshape, or | |
966 | * when assembling an array that is undergoing reshape. | |
967 | */ | |
19ceb16d | 968 | int near, far, offset, copies; |
5da9ab98 | 969 | int new_disks; |
19ceb16d | 970 | int old_chunk, new_chunk; |
b545e14a N |
971 | /* delta_parity records change in number of devices |
972 | * caused by level change | |
973 | */ | |
974 | int delta_parity = 0; | |
5da9ab98 | 975 | |
6b2fc3c1 N |
976 | memset(re, 0, sizeof(*re)); |
977 | ||
5da9ab98 N |
978 | /* If a new level not explicitly given, we assume no-change */ |
979 | if (info->new_level == UnSet) | |
980 | info->new_level = info->array.level; | |
981 | ||
982 | if (info->new_chunk) | |
983 | switch (info->new_level) { | |
984 | case 0: | |
985 | case 4: | |
986 | case 5: | |
987 | case 6: | |
988 | case 10: | |
989 | /* chunk size is meaningful, must divide component_size | |
990 | * evenly | |
991 | */ | |
a73b0081 N |
992 | if (info->component_size % (info->new_chunk/512)) { |
993 | unsigned long long shrink = info->component_size; | |
994 | shrink &= ~(unsigned long long)(info->new_chunk/512-1); | |
995 | pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n", | |
996 | info->new_chunk/1024, info->component_size/2); | |
997 | pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n", | |
998 | devname, shrink/2); | |
999 | pr_err("will shrink the array so the given chunk size would work.\n"); | |
1000 | return ""; | |
1001 | } | |
5da9ab98 N |
1002 | break; |
1003 | default: | |
1004 | return "chunk size not meaningful for this level"; | |
1005 | } | |
1006 | else | |
1007 | info->new_chunk = info->array.chunk_size; | |
1008 | ||
1009 | switch (info->array.level) { | |
90b60dfa N |
1010 | default: |
1011 | return "Cannot understand this RAID level"; | |
5da9ab98 N |
1012 | case 1: |
1013 | /* RAID1 can convert to RAID1 with different disks, or | |
dfe77a9e KW |
1014 | * raid5 with 2 disks, or |
1015 | * raid0 with 1 disk | |
5da9ab98 | 1016 | */ |
5ca3a902 N |
1017 | if (info->new_level > 1 && |
1018 | (info->component_size & 7)) | |
1019 | return "Cannot convert RAID1 of this size - " | |
1020 | "reduce size to multiple of 4K first."; | |
dfe77a9e | 1021 | if (info->new_level == 0) { |
b545e14a N |
1022 | if (info->delta_disks != UnSet && |
1023 | info->delta_disks != 0) | |
1024 | return "Cannot change number of disks " | |
1025 | "with RAID1->RAID0 conversion"; | |
dfe77a9e KW |
1026 | re->level = 0; |
1027 | re->before.data_disks = 1; | |
1028 | re->after.data_disks = 1; | |
dfe77a9e KW |
1029 | return NULL; |
1030 | } | |
5da9ab98 N |
1031 | if (info->new_level == 1) { |
1032 | if (info->delta_disks == UnSet) | |
1033 | /* Don't know what to do */ | |
1034 | return "no change requested for Growing RAID1"; | |
1035 | re->level = 1; | |
5da9ab98 N |
1036 | return NULL; |
1037 | } | |
1038 | if (info->array.raid_disks == 2 && | |
446d2a5a | 1039 | info->new_level == 5) { |
5652f2d9 | 1040 | |
5da9ab98 | 1041 | re->level = 5; |
5da9ab98 | 1042 | re->before.data_disks = 1; |
5652f2d9 N |
1043 | if (info->delta_disks != UnSet && |
1044 | info->delta_disks != 0) | |
1045 | re->after.data_disks = 1 + info->delta_disks; | |
1046 | else | |
1047 | re->after.data_disks = 1; | |
1048 | if (re->after.data_disks < 1) | |
1049 | return "Number of disks too small for RAID5"; | |
1050 | ||
5da9ab98 | 1051 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC; |
446d2a5a N |
1052 | info->array.chunk_size = 65536; |
1053 | break; | |
5da9ab98 N |
1054 | } |
1055 | /* Could do some multi-stage conversions, but leave that to | |
1056 | * later. | |
1057 | */ | |
1058 | return "Impossibly level change request for RAID1"; | |
1059 | ||
1060 | case 10: | |
19ceb16d N |
1061 | /* RAID10 can be converted from near mode to |
1062 | * RAID0 by removing some devices. | |
1063 | * It can also be reshaped if the kernel supports | |
1064 | * new_data_offset. | |
5da9ab98 | 1065 | */ |
19ceb16d N |
1066 | switch (info->new_level) { |
1067 | case 0: | |
1068 | if ((info->array.layout & ~0xff) != 0x100) | |
1069 | return "Cannot Grow RAID10 with far/offset layout"; | |
1070 | /* number of devices must be multiple of number of copies */ | |
1071 | if (info->array.raid_disks % (info->array.layout & 0xff)) | |
1072 | return "RAID10 layout too complex for Grow operation"; | |
1073 | ||
1074 | new_disks = (info->array.raid_disks | |
1075 | / (info->array.layout & 0xff)); | |
1076 | if (info->delta_disks == UnSet) | |
1077 | info->delta_disks = (new_disks | |
1078 | - info->array.raid_disks); | |
5da9ab98 | 1079 | |
19ceb16d N |
1080 | if (info->delta_disks != new_disks - info->array.raid_disks) |
1081 | return "New number of raid-devices impossible for RAID10"; | |
1082 | if (info->new_chunk && | |
1083 | info->new_chunk != info->array.chunk_size) | |
1084 | return "Cannot change chunk-size with RAID10 Grow"; | |
1085 | ||
1086 | /* looks good */ | |
1087 | re->level = 0; | |
19ceb16d N |
1088 | re->before.data_disks = new_disks; |
1089 | re->after.data_disks = re->before.data_disks; | |
19ceb16d N |
1090 | return NULL; |
1091 | ||
1092 | case 10: | |
1093 | near = info->array.layout & 0xff; | |
1094 | far = (info->array.layout >> 8) & 0xff; | |
1095 | offset = info->array.layout & 0x10000; | |
1096 | if (far > 1 && !offset) | |
1097 | return "Cannot reshape RAID10 in far-mode"; | |
1098 | copies = near * far; | |
1099 | ||
1100 | old_chunk = info->array.chunk_size * far; | |
1101 | ||
1102 | if (info->new_layout == UnSet) | |
1103 | info->new_layout = info->array.layout; | |
1104 | else { | |
1105 | near = info->new_layout & 0xff; | |
1106 | far = (info->new_layout >> 8) & 0xff; | |
1107 | offset = info->new_layout & 0x10000; | |
1108 | if (far > 1 && !offset) | |
1109 | return "Cannot reshape RAID10 to far-mode"; | |
1110 | if (near * far != copies) | |
1111 | return "Cannot change number of copies" | |
1112 | " when reshaping RAID10"; | |
1113 | } | |
1114 | if (info->delta_disks == UnSet) | |
1115 | info->delta_disks = 0; | |
1116 | new_disks = (info->array.raid_disks + | |
1117 | info->delta_disks); | |
1118 | ||
1119 | new_chunk = info->new_chunk * far; | |
1120 | ||
1121 | re->level = 10; | |
19ceb16d N |
1122 | re->before.layout = info->array.layout; |
1123 | re->before.data_disks = info->array.raid_disks; | |
1124 | re->after.layout = info->new_layout; | |
1125 | re->after.data_disks = new_disks; | |
89ecd3cf N |
1126 | /* For RAID10 we don't do backup but do allow reshape, |
1127 | * so set backup_blocks to INVALID_SECTORS rather than | |
1128 | * zero. | |
1129 | * And there is no need to synchronise stripes on both | |
19ceb16d N |
1130 | * 'old' and 'new'. So the important |
1131 | * number is the minimum data_offset difference | |
1132 | * which is the larger of (offset copies * chunk). | |
1133 | */ | |
89ecd3cf N |
1134 | re->backup_blocks = INVALID_SECTORS; |
1135 | re->min_offset_change = max(old_chunk, new_chunk) / 512; | |
ee2429e0 | 1136 | if (new_disks < re->before.data_disks && |
89ecd3cf | 1137 | info->space_after < re->min_offset_change) |
ee2429e0 N |
1138 | /* Reduce component size by one chunk */ |
1139 | re->new_size = (info->component_size - | |
89ecd3cf | 1140 | re->min_offset_change); |
ee2429e0 N |
1141 | else |
1142 | re->new_size = info->component_size; | |
1143 | re->new_size = re->new_size * new_disks / copies; | |
19ceb16d | 1144 | return NULL; |
5da9ab98 | 1145 | |
19ceb16d N |
1146 | default: |
1147 | return "RAID10 can only be changed to RAID0"; | |
1148 | } | |
5da9ab98 N |
1149 | case 0: |
1150 | /* RAID0 can be converted to RAID10, or to RAID456 */ | |
1151 | if (info->new_level == 10) { | |
1152 | if (info->new_layout == UnSet && info->delta_disks == UnSet) { | |
1153 | /* Assume near=2 layout */ | |
1154 | info->new_layout = 0x102; | |
1155 | info->delta_disks = info->array.raid_disks; | |
1156 | } | |
1157 | if (info->new_layout == UnSet) { | |
1158 | int copies = 1 + (info->delta_disks | |
1159 | / info->array.raid_disks); | |
1160 | if (info->array.raid_disks * (copies-1) | |
1161 | != info->delta_disks) | |
1162 | return "Impossible number of devices" | |
1163 | " for RAID0->RAID10"; | |
1164 | info->new_layout = 0x100 + copies; | |
1165 | } | |
1166 | if (info->delta_disks == UnSet) { | |
1167 | int copies = info->new_layout & 0xff; | |
1168 | if (info->new_layout != 0x100 + copies) | |
1169 | return "New layout impossible" | |
1170 | " for RAID0->RAID10";; | |
1171 | info->delta_disks = (copies - 1) * | |
1172 | info->array.raid_disks; | |
1173 | } | |
1174 | if (info->new_chunk && | |
1175 | info->new_chunk != info->array.chunk_size) | |
1176 | return "Cannot change chunk-size with RAID0->RAID10"; | |
1177 | /* looks good */ | |
1178 | re->level = 10; | |
5da9ab98 N |
1179 | re->before.data_disks = (info->array.raid_disks + |
1180 | info->delta_disks); | |
031d445c | 1181 | re->after.data_disks = re->before.data_disks; |
5da9ab98 | 1182 | re->before.layout = info->new_layout; |
5da9ab98 N |
1183 | return NULL; |
1184 | } | |
1185 | ||
1186 | /* RAID0 can also covert to RAID0/4/5/6 by first converting to | |
1187 | * a raid4 style layout of the final level. | |
1188 | */ | |
1189 | switch (info->new_level) { | |
5da9ab98 | 1190 | case 4: |
b545e14a N |
1191 | delta_parity = 1; |
1192 | case 0: | |
5da9ab98 N |
1193 | re->level = 4; |
1194 | re->before.layout = 0; | |
1195 | break; | |
1196 | case 5: | |
b545e14a | 1197 | delta_parity = 1; |
5da9ab98 N |
1198 | re->level = 5; |
1199 | re->before.layout = ALGORITHM_PARITY_N; | |
e5ba75ce N |
1200 | if (info->new_layout == UnSet) |
1201 | info->new_layout = map_name(r5layout, "default"); | |
5da9ab98 N |
1202 | break; |
1203 | case 6: | |
b545e14a | 1204 | delta_parity = 2; |
5da9ab98 N |
1205 | re->level = 6; |
1206 | re->before.layout = ALGORITHM_PARITY_N; | |
e5ba75ce N |
1207 | if (info->new_layout == UnSet) |
1208 | info->new_layout = map_name(r6layout, "default"); | |
5da9ab98 N |
1209 | break; |
1210 | default: | |
1211 | return "Impossible level change requested"; | |
1212 | } | |
1213 | re->before.data_disks = info->array.raid_disks; | |
1214 | /* determining 'after' layout happens outside this 'switch' */ | |
1215 | break; | |
1216 | ||
1217 | case 4: | |
1218 | info->array.layout = ALGORITHM_PARITY_N; | |
1219 | case 5: | |
1220 | switch (info->new_level) { | |
f0cce442 | 1221 | case 0: |
b545e14a | 1222 | delta_parity = -1; |
5da9ab98 N |
1223 | case 4: |
1224 | re->level = info->array.level; | |
1225 | re->before.data_disks = info->array.raid_disks - 1; | |
1226 | re->before.layout = info->array.layout; | |
1227 | break; | |
1228 | case 5: | |
1229 | re->level = 5; | |
1230 | re->before.data_disks = info->array.raid_disks - 1; | |
1231 | re->before.layout = info->array.layout; | |
1232 | break; | |
1233 | case 6: | |
b545e14a | 1234 | delta_parity = 1; |
5da9ab98 N |
1235 | re->level = 6; |
1236 | re->before.data_disks = info->array.raid_disks - 1; | |
1237 | switch (info->array.layout) { | |
1238 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1239 | re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
1240 | break; | |
1241 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1242 | re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
1243 | break; | |
1244 | case ALGORITHM_LEFT_SYMMETRIC: | |
1245 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
1246 | break; | |
1247 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1248 | re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
1249 | break; | |
1250 | case ALGORITHM_PARITY_0: | |
1251 | re->before.layout = ALGORITHM_PARITY_0_6; | |
1252 | break; | |
1253 | case ALGORITHM_PARITY_N: | |
1254 | re->before.layout = ALGORITHM_PARITY_N_6; | |
1255 | break; | |
1256 | default: | |
1257 | return "Cannot convert an array with this layout"; | |
1258 | } | |
1259 | break; | |
1260 | case 1: | |
1261 | if (info->array.raid_disks != 2) | |
1262 | return "Can only convert a 2-device array to RAID1"; | |
b545e14a N |
1263 | if (info->delta_disks != UnSet && |
1264 | info->delta_disks != 0) | |
1265 | return "Cannot set raid_disk when " | |
1266 | "converting RAID5->RAID1"; | |
5da9ab98 | 1267 | re->level = 1; |
6a67848a N |
1268 | info->new_chunk = 0; |
1269 | return NULL; | |
5da9ab98 N |
1270 | default: |
1271 | return "Impossible level change requested"; | |
1272 | } | |
1273 | break; | |
1274 | case 6: | |
1275 | switch (info->new_level) { | |
1276 | case 4: | |
1277 | case 5: | |
b545e14a | 1278 | delta_parity = -1; |
5da9ab98 N |
1279 | case 6: |
1280 | re->level = 6; | |
1281 | re->before.data_disks = info->array.raid_disks - 2; | |
1282 | re->before.layout = info->array.layout; | |
1283 | break; | |
1284 | default: | |
1285 | return "Impossible level change requested"; | |
1286 | } | |
1287 | break; | |
1288 | } | |
1289 | ||
1290 | /* If we reached here then it looks like a re-stripe is | |
1291 | * happening. We have determined the intermediate level | |
1292 | * and initial raid_disks/layout and stored these in 're'. | |
1293 | * | |
1294 | * We need to deduce the final layout that can be atomically | |
1295 | * converted to the end state. | |
1296 | */ | |
1297 | switch (info->new_level) { | |
1298 | case 0: | |
1299 | /* We can only get to RAID0 from RAID4 or RAID5 | |
1300 | * with appropriate layout and one extra device | |
1301 | */ | |
1302 | if (re->level != 4 && re->level != 5) | |
1303 | return "Cannot covert to RAID0 from this level"; | |
b545e14a | 1304 | |
5da9ab98 N |
1305 | switch (re->level) { |
1306 | case 4: | |
fdcad551 | 1307 | re->before.layout = 0; |
ca3b6696 N |
1308 | re->after.layout = 0; |
1309 | break; | |
5da9ab98 | 1310 | case 5: |
ca3b6696 N |
1311 | re->after.layout = ALGORITHM_PARITY_N; |
1312 | break; | |
5da9ab98 N |
1313 | } |
1314 | break; | |
1315 | ||
1316 | case 4: | |
1317 | /* We can only get to RAID4 from RAID5 */ | |
1318 | if (re->level != 4 && re->level != 5) | |
1319 | return "Cannot convert to RAID4 from this level"; | |
b545e14a | 1320 | |
5da9ab98 N |
1321 | switch (re->level) { |
1322 | case 4: | |
fdcad551 | 1323 | re->before.layout = 0; |
ca3b6696 N |
1324 | re->after.layout = 0; |
1325 | break; | |
5da9ab98 | 1326 | case 5: |
ca3b6696 N |
1327 | re->after.layout = ALGORITHM_PARITY_N; |
1328 | break; | |
5da9ab98 N |
1329 | } |
1330 | break; | |
1331 | ||
1332 | case 5: | |
0073a6e1 | 1333 | /* We get to RAID5 from RAID5 or RAID6 */ |
5da9ab98 N |
1334 | if (re->level != 5 && re->level != 6) |
1335 | return "Cannot convert to RAID5 from this level"; | |
b545e14a | 1336 | |
5da9ab98 N |
1337 | switch (re->level) { |
1338 | case 5: | |
1339 | if (info->new_layout == UnSet) | |
1340 | re->after.layout = re->before.layout; | |
1341 | else | |
1342 | re->after.layout = info->new_layout; | |
1343 | break; | |
1344 | case 6: | |
4a870364 N |
1345 | if (info->new_layout == UnSet) |
1346 | info->new_layout = re->before.layout; | |
1347 | ||
5da9ab98 N |
1348 | /* after.layout needs to be raid6 version of new_layout */ |
1349 | if (info->new_layout == ALGORITHM_PARITY_N) | |
1350 | re->after.layout = ALGORITHM_PARITY_N; | |
1351 | else { | |
1352 | char layout[40]; | |
1353 | char *ls = map_num(r5layout, info->new_layout); | |
1354 | int l; | |
0073a6e1 N |
1355 | if (ls) { |
1356 | /* Current RAID6 layout has a RAID5 | |
1357 | * equivalent - good | |
1358 | */ | |
1359 | strcat(strcpy(layout, ls), "-6"); | |
1360 | l = map_name(r6layout, layout); | |
1361 | if (l == UnSet) | |
1362 | return "Cannot find RAID6 layout" | |
1363 | " to convert to"; | |
1364 | } else { | |
1365 | /* Current RAID6 has no equivalent. | |
1366 | * If it is already a '-6' layout we | |
1367 | * can leave it unchanged, else we must | |
1368 | * fail | |
1369 | */ | |
1370 | ls = map_num(r6layout, info->new_layout); | |
1371 | if (!ls || | |
1372 | strcmp(ls+strlen(ls)-2, "-6") != 0) | |
1373 | return "Please specify new layout"; | |
1374 | l = info->new_layout; | |
1375 | } | |
5da9ab98 N |
1376 | re->after.layout = l; |
1377 | } | |
1378 | } | |
1379 | break; | |
1380 | ||
1381 | case 6: | |
1382 | /* We must already be at level 6 */ | |
1383 | if (re->level != 6) | |
1384 | return "Impossible level change"; | |
5da9ab98 | 1385 | if (info->new_layout == UnSet) |
4a870364 | 1386 | re->after.layout = info->array.layout; |
5da9ab98 N |
1387 | else |
1388 | re->after.layout = info->new_layout; | |
1389 | break; | |
1390 | default: | |
1391 | return "Impossible level change requested"; | |
1392 | } | |
b545e14a N |
1393 | if (info->delta_disks == UnSet) |
1394 | info->delta_disks = delta_parity; | |
1395 | ||
1396 | re->after.data_disks = (re->before.data_disks | |
1397 | + info->delta_disks | |
1398 | - delta_parity); | |
5da9ab98 | 1399 | switch (re->level) { |
ca3b6696 N |
1400 | case 6: re->parity = 2; |
1401 | break; | |
5da9ab98 | 1402 | case 4: |
ca3b6696 N |
1403 | case 5: re->parity = 1; |
1404 | break; | |
1405 | default: re->parity = 0; | |
1406 | break; | |
5da9ab98 N |
1407 | } |
1408 | /* So we have a restripe operation, we need to calculate the number | |
1409 | * of blocks per reshape operation. | |
1410 | */ | |
ddbf2ebb | 1411 | re->new_size = info->component_size * re->before.data_disks; |
5da9ab98 N |
1412 | if (info->new_chunk == 0) |
1413 | info->new_chunk = info->array.chunk_size; | |
1414 | if (re->after.data_disks == re->before.data_disks && | |
1415 | re->after.layout == re->before.layout && | |
1416 | info->new_chunk == info->array.chunk_size) { | |
fdcad551 N |
1417 | /* Nothing to change, can change level immediately. */ |
1418 | re->level = info->new_level; | |
b4f8e38b | 1419 | re->backup_blocks = 0; |
5da9ab98 N |
1420 | return NULL; |
1421 | } | |
1422 | if (re->after.data_disks == 1 && re->before.data_disks == 1) { | |
446d2a5a | 1423 | /* chunk and layout changes make no difference */ |
fdcad551 | 1424 | re->level = info->new_level; |
b4f8e38b | 1425 | re->backup_blocks = 0; |
5da9ab98 N |
1426 | return NULL; |
1427 | } | |
1428 | ||
1429 | if (re->after.data_disks == re->before.data_disks && | |
1430 | get_linux_version() < 2006032) | |
1431 | return "in-place reshape is not safe before 2.6.32 - sorry."; | |
1432 | ||
1433 | if (re->after.data_disks < re->before.data_disks && | |
1434 | get_linux_version() < 2006030) | |
508ede86 | 1435 | return "reshape to fewer devices is not supported before 2.6.30 - sorry."; |
5da9ab98 | 1436 | |
b4f8e38b | 1437 | re->backup_blocks = compute_backup_blocks( |
5da9ab98 N |
1438 | info->new_chunk, info->array.chunk_size, |
1439 | re->after.data_disks, | |
1440 | re->before.data_disks); | |
63c12c89 | 1441 | re->min_offset_change = re->backup_blocks / re->before.data_disks; |
5da9ab98 N |
1442 | |
1443 | re->new_size = info->component_size * re->after.data_disks; | |
1444 | return NULL; | |
1445 | } | |
1446 | ||
54397ed9 AK |
1447 | static int set_array_size(struct supertype *st, struct mdinfo *sra, |
1448 | char *text_version) | |
1449 | { | |
1450 | struct mdinfo *info; | |
1451 | char *subarray; | |
1452 | int ret_val = -1; | |
1453 | ||
1454 | if ((st == NULL) || (sra == NULL)) | |
1455 | return ret_val; | |
1456 | ||
1457 | if (text_version == NULL) | |
1458 | text_version = sra->text_version; | |
1459 | subarray = strchr(text_version+1, '/')+1; | |
1460 | info = st->ss->container_content(st, subarray); | |
1461 | if (info) { | |
1462 | unsigned long long current_size = 0; | |
1463 | unsigned long long new_size = | |
1464 | info->custom_array_size/2; | |
1465 | ||
1466 | if (sysfs_get_ll(sra, NULL, "array_size", ¤t_size) == 0 && | |
1467 | new_size > current_size) { | |
1468 | if (sysfs_set_num(sra, NULL, "array_size", new_size) | |
1469 | < 0) | |
1470 | dprintf("Error: Cannot set array size"); | |
1471 | else { | |
1472 | ret_val = 0; | |
1473 | dprintf("Array size changed"); | |
1474 | } | |
1475 | dprintf(" from %llu to %llu.\n", | |
1476 | current_size, new_size); | |
1477 | } | |
1478 | sysfs_free(info); | |
1479 | } else | |
1480 | dprintf("Error: set_array_size(): info pointer in NULL\n"); | |
1481 | ||
1482 | return ret_val; | |
1483 | } | |
1484 | ||
5da9ab98 N |
1485 | static int reshape_array(char *container, int fd, char *devname, |
1486 | struct supertype *st, struct mdinfo *info, | |
e2e53a2d | 1487 | int force, struct mddev_dev *devlist, |
19ceb16d | 1488 | unsigned long long data_offset, |
ba728be7 | 1489 | char *backup_file, int verbose, int forked, |
b76b30e0 | 1490 | int restart, int freeze_reshape); |
493f5dd6 | 1491 | static int reshape_container(char *container, char *devname, |
9ad6f6e6 | 1492 | int mdfd, |
24daa16f | 1493 | struct supertype *st, |
5da9ab98 N |
1494 | struct mdinfo *info, |
1495 | int force, | |
1496 | char *backup_file, | |
ba728be7 | 1497 | int verbose, int restart, int freeze_reshape); |
1c009fc2 | 1498 | |
32754b7d | 1499 | int Grow_reshape(char *devname, int fd, |
e2e53a2d | 1500 | struct mddev_dev *devlist, |
40c9a66a | 1501 | unsigned long long data_offset, |
32754b7d | 1502 | struct context *c, struct shape *s) |
e86c9dd6 NB |
1503 | { |
1504 | /* Make some changes in the shape of an array. | |
1505 | * The kernel must support the change. | |
7236ee7a N |
1506 | * |
1507 | * There are three different changes. Each can trigger | |
1508 | * a resync or recovery so we freeze that until we have | |
1509 | * requested everything (if kernel supports freezing - 2.6.30). | |
1510 | * The steps are: | |
1511 | * - change size (i.e. component_size) | |
1512 | * - change level | |
1513 | * - change layout/chunksize/ndisks | |
1514 | * | |
1515 | * The last can require a reshape. It is different on different | |
1516 | * levels so we need to check the level before actioning it. | |
1517 | * Some times the level change needs to be requested after the | |
1518 | * reshape (e.g. raid6->raid5, raid5->raid0) | |
1519 | * | |
e86c9dd6 | 1520 | */ |
5da9ab98 | 1521 | struct mdu_array_info_s array; |
7236ee7a | 1522 | int rv = 0; |
e86c9dd6 | 1523 | struct supertype *st; |
4725bc31 | 1524 | char *subarray = NULL; |
e86c9dd6 | 1525 | |
7236ee7a | 1526 | int frozen; |
7236ee7a | 1527 | int changed = 0; |
7bc71196 DW |
1528 | char *container = NULL; |
1529 | int cfd = -1; | |
e86c9dd6 | 1530 | |
e2e53a2d N |
1531 | struct mddev_dev *dv; |
1532 | int added_disks; | |
1533 | ||
5da9ab98 | 1534 | struct mdinfo info; |
7e0f6979 | 1535 | struct mdinfo *sra; |
e86c9dd6 NB |
1536 | |
1537 | if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) { | |
ed503f89 | 1538 | pr_err("%s is not an active md array - aborting\n", |
e86c9dd6 NB |
1539 | devname); |
1540 | return 1; | |
1541 | } | |
4abcbc21 N |
1542 | if (data_offset != INVALID_SECTORS && array.level != 10 |
1543 | && (array.level < 4 || array.level > 6)) { | |
19ceb16d N |
1544 | pr_err("--grow --data-offset not yet supported\n"); |
1545 | return 1; | |
1546 | } | |
24d40069 | 1547 | |
32754b7d N |
1548 | if (s->size > 0 && |
1549 | (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) { | |
e7b84f9d | 1550 | pr_err("cannot change component size at the same time " |
9ce510be N |
1551 | "as other changes.\n" |
1552 | " Change size first, then check data is intact before " | |
1553 | "making other changes.\n"); | |
1554 | return 1; | |
1555 | } | |
1556 | ||
32754b7d | 1557 | if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 && |
24d40069 N |
1558 | get_linux_version() < 2006032 && |
1559 | !check_env("MDADM_FORCE_FEWER")) { | |
e7b84f9d | 1560 | pr_err("reducing the number of devices is not safe before Linux 2.6.32\n" |
24d40069 N |
1561 | " Please use a newer kernel\n"); |
1562 | return 1; | |
1563 | } | |
7bc71196 DW |
1564 | |
1565 | st = super_by_fd(fd, &subarray); | |
1566 | if (!st) { | |
e7b84f9d | 1567 | pr_err("Unable to determine metadata format for %s\n", devname); |
7bc71196 DW |
1568 | return 1; |
1569 | } | |
32754b7d | 1570 | if (s->raiddisks > st->max_devs) { |
e7b84f9d | 1571 | pr_err("Cannot increase raid-disks on this array" |
acab7bb1 N |
1572 | " beyond %d\n", st->max_devs); |
1573 | return 1; | |
1574 | } | |
7bc71196 DW |
1575 | |
1576 | /* in the external case we need to check that the requested reshape is | |
1577 | * supported, and perform an initial check that the container holds the | |
1578 | * pre-requisite spare devices (mdmon owns final validation) | |
1579 | */ | |
1580 | if (st->ss->external) { | |
7f2ba464 | 1581 | int rv; |
7bc71196 DW |
1582 | |
1583 | if (subarray) { | |
4dd2df09 N |
1584 | container = st->container_devnm; |
1585 | cfd = open_dev_excl(st->container_devnm); | |
7bc71196 | 1586 | } else { |
4dd2df09 | 1587 | container = st->devnm; |
7bc71196 | 1588 | close(fd); |
4dd2df09 | 1589 | cfd = open_dev_excl(st->devnm); |
7bc71196 DW |
1590 | fd = cfd; |
1591 | } | |
1592 | if (cfd < 0) { | |
e7b84f9d | 1593 | pr_err("Unable to open container for %s\n", |
7bc71196 | 1594 | devname); |
7f2ba464 | 1595 | free(subarray); |
7bc71196 DW |
1596 | return 1; |
1597 | } | |
1598 | ||
eb744788 AK |
1599 | rv = st->ss->load_container(st, cfd, NULL); |
1600 | ||
7f2ba464 | 1601 | if (rv) { |
e7b84f9d | 1602 | pr_err("Cannot read superblock for %s\n", |
7bc71196 | 1603 | devname); |
7f2ba464 | 1604 | free(subarray); |
7bc71196 DW |
1605 | return 1; |
1606 | } | |
1607 | ||
81219e70 LM |
1608 | /* check if operation is supported for metadata handler */ |
1609 | if (st->ss->container_content) { | |
1610 | struct mdinfo *cc = NULL; | |
1611 | struct mdinfo *content = NULL; | |
1612 | ||
1613 | cc = st->ss->container_content(st, subarray); | |
1614 | for (content = cc; content ; content = content->next) { | |
446894ea | 1615 | int allow_reshape = 1; |
81219e70 LM |
1616 | |
1617 | /* check if reshape is allowed based on metadata | |
1618 | * indications stored in content.array.status | |
1619 | */ | |
446894ea N |
1620 | if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) |
1621 | allow_reshape = 0; | |
1622 | if (content->array.state | |
1623 | & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)) | |
1624 | allow_reshape = 0; | |
81219e70 | 1625 | if (!allow_reshape) { |
e7b84f9d N |
1626 | pr_err("cannot reshape arrays in" |
1627 | " container with unsupported" | |
1628 | " metadata: %s(%s)\n", | |
4dd2df09 | 1629 | devname, container); |
81219e70 LM |
1630 | sysfs_free(cc); |
1631 | free(subarray); | |
1632 | return 1; | |
1633 | } | |
1634 | } | |
1635 | sysfs_free(cc); | |
1636 | } | |
4dd2df09 | 1637 | if (mdmon_running(container)) |
7bc71196 | 1638 | st->update_tail = &st->updates; |
e2e53a2d | 1639 | } |
691a36b7 | 1640 | |
e2e53a2d N |
1641 | added_disks = 0; |
1642 | for (dv = devlist; dv; dv = dv->next) | |
1643 | added_disks++; | |
32754b7d N |
1644 | if (s->raiddisks > array.raid_disks && |
1645 | array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) && | |
1646 | !c->force) { | |
e7b84f9d N |
1647 | pr_err("Need %d spare%s to avoid degraded array," |
1648 | " and only have %d.\n" | |
1649 | " Use --force to over-ride this check.\n", | |
32754b7d N |
1650 | s->raiddisks - array.raid_disks, |
1651 | s->raiddisks - array.raid_disks == 1 ? "" : "s", | |
e7b84f9d | 1652 | array.spare_disks + added_disks); |
691a36b7 | 1653 | return 1; |
7bc71196 DW |
1654 | } |
1655 | ||
4dd2df09 | 1656 | sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS |
3fa436ac | 1657 | | GET_STATE | GET_VERSION); |
24daa16f | 1658 | if (sra) { |
7f2ba464 | 1659 | if (st->ss->external && subarray == NULL) { |
7bc71196 DW |
1660 | array.level = LEVEL_CONTAINER; |
1661 | sra->array.level = LEVEL_CONTAINER; | |
1662 | } | |
7bc71196 | 1663 | } else { |
e7b84f9d | 1664 | pr_err("failed to read sysfs parameters for %s\n", |
b526e52d DW |
1665 | devname); |
1666 | return 1; | |
1667 | } | |
7f2ba464 DW |
1668 | frozen = freeze(st); |
1669 | if (frozen < -1) { | |
1670 | /* freeze() already spewed the reason */ | |
8e61e0d7 | 1671 | sysfs_free(sra); |
7f2ba464 DW |
1672 | return 1; |
1673 | } else if (frozen < 0) { | |
e7b84f9d | 1674 | pr_err("%s is performing resync/recovery and cannot" |
7236ee7a | 1675 | " be reshaped\n", devname); |
8e61e0d7 | 1676 | sysfs_free(sra); |
7236ee7a N |
1677 | return 1; |
1678 | } | |
1679 | ||
1680 | /* ========= set size =============== */ | |
887a7a9e | 1681 | if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) { |
d04f65f4 N |
1682 | unsigned long long orig_size = get_component_size(fd)/2; |
1683 | unsigned long long min_csize; | |
d1537ed1 | 1684 | struct mdinfo *mdi; |
44f6f181 | 1685 | int raid0_takeover = 0; |
7bc71196 | 1686 | |
85f10287 | 1687 | if (orig_size == 0) |
d04f65f4 | 1688 | orig_size = (unsigned) array.size; |
85f10287 | 1689 | |
3920235e N |
1690 | if (orig_size == 0) { |
1691 | pr_err("Cannot set device size in this type of array.\n"); | |
1692 | rv = 1; | |
1693 | goto release; | |
1694 | } | |
1695 | ||
32754b7d N |
1696 | if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL, |
1697 | devname, APPLY_METADATA_CHANGES, c->verbose > 0)) { | |
7bc71196 DW |
1698 | rv = 1; |
1699 | goto release; | |
1700 | } | |
1701 | sync_metadata(st); | |
7e7e9a4d AK |
1702 | if (st->ss->external) { |
1703 | /* metadata can have size limitation | |
1704 | * update size value according to metadata information | |
1705 | */ | |
1706 | struct mdinfo *sizeinfo = | |
1707 | st->ss->container_content(st, subarray); | |
1708 | if (sizeinfo) { | |
1709 | unsigned long long new_size = | |
1710 | sizeinfo->custom_array_size/2; | |
1711 | int data_disks = get_data_disks( | |
1712 | sizeinfo->array.level, | |
1713 | sizeinfo->array.layout, | |
1714 | sizeinfo->array.raid_disks); | |
1715 | new_size /= data_disks; | |
1716 | dprintf("Metadata size correction from %llu to " | |
1717 | "%llu (%llu)\n", orig_size, new_size, | |
1718 | new_size * data_disks); | |
32754b7d | 1719 | s->size = new_size; |
7e7e9a4d AK |
1720 | sysfs_free(sizeinfo); |
1721 | } | |
1722 | } | |
d1537ed1 N |
1723 | |
1724 | /* Update the size of each member device in case | |
1725 | * they have been resized. This will never reduce | |
1726 | * below the current used-size. The "size" attribute | |
20a46756 | 1727 | * understands '0' to mean 'max'. |
d1537ed1 | 1728 | */ |
20a46756 | 1729 | min_csize = 0; |
65a9798b | 1730 | rv = 0; |
20a46756 | 1731 | for (mdi = sra->devs; mdi; mdi = mdi->next) { |
d04f65f4 | 1732 | if (sysfs_set_num(sra, mdi, "size", |
32754b7d | 1733 | s->size == MAX_SIZE ? 0 : s->size) < 0) { |
b0a658ff N |
1734 | /* Probably kernel refusing to let us |
1735 | * reduce the size - not an error. | |
1736 | */ | |
20a46756 | 1737 | break; |
65a9798b | 1738 | } |
20a46756 N |
1739 | if (array.not_persistent == 0 && |
1740 | array.major_version == 0 && | |
1741 | get_linux_version() < 3001000) { | |
1742 | /* Dangerous to allow size to exceed 2TB */ | |
1743 | unsigned long long csize; | |
1744 | if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) { | |
1745 | if (csize >= 2ULL*1024*1024*1024) | |
1746 | csize = 2ULL*1024*1024*1024; | |
1747 | if ((min_csize == 0 || (min_csize | |
d04f65f4 | 1748 | > csize))) |
20a46756 N |
1749 | min_csize = csize; |
1750 | } | |
1751 | } | |
1752 | } | |
65a9798b | 1753 | if (rv) { |
e7b84f9d | 1754 | pr_err("Cannot set size on " |
65a9798b AK |
1755 | "array members.\n"); |
1756 | goto size_change_error; | |
1757 | } | |
32754b7d | 1758 | if (min_csize && s->size > min_csize) { |
e7b84f9d | 1759 | pr_err("Cannot safely make this array " |
20a46756 N |
1760 | "use more than 2TB per device on this kernel.\n"); |
1761 | rv = 1; | |
65a9798b | 1762 | goto size_change_error; |
20a46756 | 1763 | } |
32754b7d | 1764 | if (min_csize && s->size == MAX_SIZE) { |
20a46756 N |
1765 | /* Don't let the kernel choose a size - it will get |
1766 | * it wrong | |
1767 | */ | |
e7b84f9d N |
1768 | pr_err("Limited v0.90 array to " |
1769 | "2TB per device\n"); | |
32754b7d | 1770 | s->size = min_csize; |
20a46756 | 1771 | } |
44f6f181 AK |
1772 | if (st->ss->external) { |
1773 | if (sra->array.level == 0) { | |
1774 | rv = sysfs_set_str(sra, NULL, "level", | |
1775 | "raid5"); | |
1776 | if (!rv) { | |
1777 | raid0_takeover = 1; | |
1778 | /* get array parametes after takeover | |
1779 | * to chane one parameter at time only | |
1780 | */ | |
1781 | rv = ioctl(fd, GET_ARRAY_INFO, &array); | |
1782 | } | |
1783 | } | |
1784 | /* make sure mdmon is | |
1785 | * aware of the new level */ | |
4dd2df09 N |
1786 | if (!mdmon_running(st->container_devnm)) |
1787 | start_mdmon(st->container_devnm); | |
44f6f181 | 1788 | ping_monitor(container); |
4dd2df09 | 1789 | if (mdmon_running(st->container_devnm) && |
44f6f181 AK |
1790 | st->update_tail == NULL) |
1791 | st->update_tail = &st->updates; | |
1792 | } | |
1793 | ||
0a8b92a6 N |
1794 | if (s->size == MAX_SIZE) |
1795 | s->size = 0; | |
1796 | array.size = s->size; | |
32754b7d | 1797 | if ((unsigned)array.size != s->size) { |
7236ee7a N |
1798 | /* got truncated to 32bit, write to |
1799 | * component_size instead | |
1800 | */ | |
1801 | if (sra) | |
1802 | rv = sysfs_set_num(sra, NULL, | |
32754b7d | 1803 | "component_size", s->size); |
7236ee7a N |
1804 | else |
1805 | rv = -1; | |
54397ed9 | 1806 | } else { |
7236ee7a | 1807 | rv = ioctl(fd, SET_ARRAY_INFO, &array); |
44f6f181 | 1808 | |
54397ed9 AK |
1809 | /* manage array size when it is managed externally |
1810 | */ | |
1811 | if ((rv == 0) && st->ss->external) | |
1812 | rv = set_array_size(st, sra, sra->text_version); | |
1813 | } | |
1814 | ||
44f6f181 AK |
1815 | if (raid0_takeover) { |
1816 | /* do not recync non-existing parity, | |
1817 | * we will drop it anyway | |
1818 | */ | |
b51702b8 | 1819 | sysfs_set_str(sra, NULL, "sync_action", "frozen"); |
44f6f181 AK |
1820 | /* go back to raid0, drop parity disk |
1821 | */ | |
1822 | sysfs_set_str(sra, NULL, "level", "raid0"); | |
1823 | ioctl(fd, GET_ARRAY_INFO, &array); | |
1824 | } | |
1825 | ||
65a9798b | 1826 | size_change_error: |
7236ee7a | 1827 | if (rv != 0) { |
39bbb392 | 1828 | int err = errno; |
7bc71196 DW |
1829 | |
1830 | /* restore metadata */ | |
1831 | if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0, | |
016e00f5 AK |
1832 | UnSet, NULL, devname, |
1833 | ROLLBACK_METADATA_CHANGES, | |
32754b7d | 1834 | c->verbose) == 0) |
7bc71196 | 1835 | sync_metadata(st); |
e7b84f9d | 1836 | pr_err("Cannot set device size for %s: %s\n", |
39bbb392 | 1837 | devname, strerror(err)); |
24daa16f | 1838 | if (err == EBUSY && |
39bbb392 | 1839 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) |
e7b84f9d | 1840 | cont_err("Bitmap must be removed before size can be changed\n"); |
7236ee7a N |
1841 | rv = 1; |
1842 | goto release; | |
1843 | } | |
32754b7d | 1844 | if (s->assume_clean) { |
86952387 | 1845 | /* This will fail on kernels older than 3.0 unless |
ce52f92f N |
1846 | * a backport has been arranged. |
1847 | */ | |
1848 | if (sra == NULL || | |
1849 | sysfs_set_str(sra, NULL, "resync_start", "none") < 0) | |
86952387 | 1850 | pr_err("--assume-clean not supported with --grow on this kernel\n"); |
ce52f92f | 1851 | } |
7236ee7a | 1852 | ioctl(fd, GET_ARRAY_INFO, &array); |
32754b7d N |
1853 | s->size = get_component_size(fd)/2; |
1854 | if (s->size == 0) | |
1855 | s->size = array.size; | |
1856 | if (c->verbose >= 0) { | |
1857 | if (s->size == orig_size) | |
e7b84f9d | 1858 | pr_err("component size of %s " |
85f10287 | 1859 | "unchanged at %lluK\n", |
32754b7d | 1860 | devname, s->size); |
85f10287 | 1861 | else |
e7b84f9d | 1862 | pr_err("component size of %s " |
85f10287 | 1863 | "has been set to %lluK\n", |
32754b7d | 1864 | devname, s->size); |
85f10287 | 1865 | } |
7236ee7a | 1866 | changed = 1; |
7bc71196 | 1867 | } else if (array.level != LEVEL_CONTAINER) { |
32754b7d N |
1868 | s->size = get_component_size(fd)/2; |
1869 | if (s->size == 0) | |
1870 | s->size = array.size; | |
7236ee7a N |
1871 | } |
1872 | ||
d515d27f | 1873 | /* See if there is anything else to do */ |
32754b7d N |
1874 | if ((s->level == UnSet || s->level == array.level) && |
1875 | (s->layout_str == NULL) && | |
1876 | (s->chunk == 0 || s->chunk == array.chunk_size) && | |
8876bf0b | 1877 | data_offset == INVALID_SECTORS && |
32754b7d | 1878 | (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) { |
d515d27f | 1879 | /* Nothing more to do */ |
32754b7d | 1880 | if (!changed && c->verbose >= 0) |
e7b84f9d | 1881 | pr_err("%s: no change requested\n", |
d515d27f N |
1882 | devname); |
1883 | goto release; | |
1884 | } | |
1885 | ||
dfe77a9e | 1886 | /* ========= check for Raid10/Raid1 -> Raid0 conversion =============== |
5da9ab98 | 1887 | * current implementation assumes that following conditions must be met: |
dfe77a9e | 1888 | * - RAID10: |
24daa16f N |
1889 | * - far_copies == 1 |
1890 | * - near_copies == 2 | |
62a48395 | 1891 | */ |
32754b7d | 1892 | if ((s->level == 0 && array.level == 10 && sra && |
24daa16f | 1893 | array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) || |
32754b7d | 1894 | (s->level == 0 && array.level == 1 && sra)) { |
62a48395 | 1895 | int err; |
dfe77a9e | 1896 | err = remove_disks_for_takeover(st, sra, array.layout); |
62a48395 AK |
1897 | if (err) { |
1898 | dprintf(Name": Array cannot be reshaped\n"); | |
62a48395 AK |
1899 | if (cfd > -1) |
1900 | close(cfd); | |
5da9ab98 N |
1901 | rv = 1; |
1902 | goto release; | |
7236ee7a | 1903 | } |
fde139b9 N |
1904 | /* Make sure mdmon has seen the device removal |
1905 | * and updated metadata before we continue with | |
1906 | * level change | |
1907 | */ | |
1908 | if (container) | |
1909 | ping_monitor(container); | |
7236ee7a N |
1910 | } |
1911 | ||
8ff6d094 | 1912 | memset(&info, 0, sizeof(info)); |
5da9ab98 | 1913 | info.array = array; |
4dd2df09 | 1914 | sysfs_init(&info, fd, NULL); |
3fa436ac | 1915 | strcpy(info.text_version, sra->text_version); |
32754b7d N |
1916 | info.component_size = s->size*2; |
1917 | info.new_level = s->level; | |
1918 | info.new_chunk = s->chunk * 1024; | |
eff4954d N |
1919 | if (info.array.level == LEVEL_CONTAINER) { |
1920 | info.delta_disks = UnSet; | |
32754b7d N |
1921 | info.array.raid_disks = s->raiddisks; |
1922 | } else if (s->raiddisks) | |
1923 | info.delta_disks = s->raiddisks - info.array.raid_disks; | |
5da9ab98 N |
1924 | else |
1925 | info.delta_disks = UnSet; | |
32754b7d | 1926 | if (s->layout_str == NULL) { |
5da9ab98 N |
1927 | info.new_layout = UnSet; |
1928 | if (info.array.level == 6 && | |
1929 | (info.new_level == 6 || info.new_level == UnSet) && | |
1930 | info.array.layout >= 16) { | |
e7b84f9d N |
1931 | pr_err("%s has a non-standard layout. If you" |
1932 | " wish to preserve this\n", devname); | |
1933 | cont_err("during the reshape, please specify" | |
1934 | " --layout=preserve\n"); | |
1935 | cont_err("If you want to change it, specify a" | |
1936 | " layout or use --layout=normalise\n"); | |
7236ee7a N |
1937 | rv = 1; |
1938 | goto release; | |
1939 | } | |
32754b7d N |
1940 | } else if (strcmp(s->layout_str, "normalise") == 0 || |
1941 | strcmp(s->layout_str, "normalize") == 0) { | |
5da9ab98 N |
1942 | /* If we have a -6 RAID6 layout, remove the '-6'. */ |
1943 | info.new_layout = UnSet; | |
1944 | if (info.array.level == 6 && info.new_level == UnSet) { | |
1945 | char l[40], *h; | |
1946 | strcpy(l, map_num(r6layout, info.array.layout)); | |
1947 | h = strrchr(l, '-'); | |
1948 | if (h && strcmp(h, "-6") == 0) { | |
1949 | *h = 0; | |
1950 | info.new_layout = map_name(r6layout, l); | |
7236ee7a | 1951 | } |
385167f3 | 1952 | } else { |
e7b84f9d | 1953 | pr_err("%s is only meaningful when reshaping" |
32754b7d | 1954 | " a RAID6 array.\n", s->layout_str); |
385167f3 N |
1955 | rv = 1; |
1956 | goto release; | |
7236ee7a | 1957 | } |
32754b7d | 1958 | } else if (strcmp(s->layout_str, "preserve") == 0) { |
385167f3 N |
1959 | /* This means that a non-standard RAID6 layout |
1960 | * is OK. | |
1961 | * In particular: | |
1962 | * - When reshape a RAID6 (e.g. adding a device) | |
1963 | * which is in a non-standard layout, it is OK | |
1964 | * to preserve that layout. | |
1965 | * - When converting a RAID5 to RAID6, leave it in | |
1966 | * the XXX-6 layout, don't re-layout. | |
1967 | */ | |
1968 | if (info.array.level == 6 && info.new_level == UnSet) | |
1969 | info.new_layout = info.array.layout; | |
1970 | else if (info.array.level == 5 && info.new_level == 6) { | |
1971 | char l[40]; | |
1972 | strcpy(l, map_num(r5layout, info.array.layout)); | |
1973 | strcat(l, "-6"); | |
1974 | info.new_layout = map_name(r6layout, l); | |
1975 | } else { | |
e7b84f9d | 1976 | pr_err("%s in only meaningful when reshaping" |
32754b7d | 1977 | " to RAID6\n", s->layout_str); |
385167f3 N |
1978 | rv = 1; |
1979 | goto release; | |
1980 | } | |
5da9ab98 N |
1981 | } else { |
1982 | int l = info.new_level; | |
1983 | if (l == UnSet) | |
1984 | l = info.array.level; | |
1985 | switch (l) { | |
1986 | case 5: | |
32754b7d | 1987 | info.new_layout = map_name(r5layout, s->layout_str); |
5da9ab98 N |
1988 | break; |
1989 | case 6: | |
32754b7d | 1990 | info.new_layout = map_name(r6layout, s->layout_str); |
5da9ab98 N |
1991 | break; |
1992 | case 10: | |
32754b7d | 1993 | info.new_layout = parse_layout_10(s->layout_str); |
5da9ab98 N |
1994 | break; |
1995 | case LEVEL_FAULTY: | |
32754b7d | 1996 | info.new_layout = parse_layout_faulty(s->layout_str); |
5da9ab98 N |
1997 | break; |
1998 | default: | |
e7b84f9d | 1999 | pr_err("layout not meaningful" |
5da9ab98 | 2000 | " with this level\n"); |
7bc71196 DW |
2001 | rv = 1; |
2002 | goto release; | |
2003 | } | |
5da9ab98 | 2004 | if (info.new_layout == UnSet) { |
e7b84f9d | 2005 | pr_err("layout %s not understood" |
5da9ab98 | 2006 | " for this level\n", |
32754b7d | 2007 | s->layout_str); |
5da9ab98 N |
2008 | rv = 1; |
2009 | goto release; | |
7bc71196 | 2010 | } |
7236ee7a N |
2011 | } |
2012 | ||
907ea753 | 2013 | if (array.level == LEVEL_FAULTY) { |
32754b7d | 2014 | if (s->level != UnSet && s->level != array.level) { |
e7b84f9d | 2015 | pr_err("cannot change level of Faulty device\n"); |
907ea753 N |
2016 | rv =1 ; |
2017 | } | |
32754b7d | 2018 | if (s->chunk) { |
e7b84f9d | 2019 | pr_err("cannot set chunksize of Faulty device\n"); |
907ea753 N |
2020 | rv =1 ; |
2021 | } | |
32754b7d | 2022 | if (s->raiddisks && s->raiddisks != 1) { |
e7b84f9d | 2023 | pr_err("cannot set raid_disks of Faulty device\n"); |
907ea753 N |
2024 | rv =1 ; |
2025 | } | |
32754b7d | 2026 | if (s->layout_str) { |
907ea753 N |
2027 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { |
2028 | dprintf("Cannot get array information.\n"); | |
2029 | goto release; | |
2030 | } | |
2031 | array.layout = info.new_layout; | |
2032 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
e7b84f9d | 2033 | pr_err("failed to set new layout\n"); |
907ea753 | 2034 | rv = 1; |
32754b7d | 2035 | } else if (c->verbose >= 0) |
907ea753 N |
2036 | printf("layout for %s set to %d\n", |
2037 | devname, array.layout); | |
2038 | } | |
2039 | } else if (array.level == LEVEL_CONTAINER) { | |
5da9ab98 N |
2040 | /* This change is to be applied to every array in the |
2041 | * container. This is only needed when the metadata imposes | |
2042 | * restraints of the various arrays in the container. | |
2043 | * Currently we only know that IMSM requires all arrays | |
2044 | * to have the same number of devices so changing the | |
2045 | * number of devices (On-Line Capacity Expansion) must be | |
2046 | * performed at the level of the container | |
2047 | */ | |
9ad6f6e6 | 2048 | rv = reshape_container(container, devname, -1, st, &info, |
32754b7d | 2049 | c->force, c->backup_file, c->verbose, 0, 0); |
9202b8eb | 2050 | frozen = 0; |
5da9ab98 | 2051 | } else { |
08f9e34b AK |
2052 | /* get spare devices from external metadata |
2053 | */ | |
2054 | if (st->ss->external) { | |
2055 | struct mdinfo *info2; | |
2056 | ||
2057 | info2 = st->ss->container_content(st, subarray); | |
2058 | if (info2) { | |
2059 | info.array.spare_disks = | |
2060 | info2->array.spare_disks; | |
2061 | sysfs_free(info2); | |
2062 | } | |
2063 | } | |
2064 | ||
5da9ab98 N |
2065 | /* Impose these changes on a single array. First |
2066 | * check that the metadata is OK with the change. */ | |
7bc71196 | 2067 | |
7ec0996c | 2068 | if (reshape_super(st, 0, info.new_level, |
5da9ab98 | 2069 | info.new_layout, info.new_chunk, |
41784c88 | 2070 | info.array.raid_disks, info.delta_disks, |
32754b7d N |
2071 | c->backup_file, devname, APPLY_METADATA_CHANGES, |
2072 | c->verbose)) { | |
7bc71196 DW |
2073 | rv = 1; |
2074 | goto release; | |
2075 | } | |
5da9ab98 | 2076 | sync_metadata(st); |
32754b7d | 2077 | rv = reshape_array(container, fd, devname, st, &info, c->force, |
19ceb16d N |
2078 | devlist, data_offset, c->backup_file, c->verbose, |
2079 | 0, 0, 0); | |
9202b8eb | 2080 | frozen = 0; |
5da9ab98 N |
2081 | } |
2082 | release: | |
8e61e0d7 | 2083 | sysfs_free(sra); |
9202b8eb N |
2084 | if (frozen > 0) |
2085 | unfreeze(st); | |
5da9ab98 N |
2086 | return rv; |
2087 | } | |
7bc71196 | 2088 | |
f93346ef AK |
2089 | /* verify_reshape_position() |
2090 | * Function checks if reshape position in metadata is not farther | |
2091 | * than position in md. | |
2092 | * Return value: | |
2093 | * 0 : not valid sysfs entry | |
2094 | * it can be caused by not started reshape, it should be started | |
2095 | * by reshape array or raid0 array is before takeover | |
2096 | * -1 : error, reshape position is obviously wrong | |
2097 | * 1 : success, reshape progress correct or updated | |
2098 | */ | |
2099 | static int verify_reshape_position(struct mdinfo *info, int level) | |
2100 | { | |
2101 | int ret_val = 0; | |
2102 | char buf[40]; | |
178950ea | 2103 | int rv; |
f93346ef AK |
2104 | |
2105 | /* read sync_max, failure can mean raid0 array */ | |
178950ea AK |
2106 | rv = sysfs_get_str(info, NULL, "sync_max", buf, 40); |
2107 | ||
2108 | if (rv > 0) { | |
f93346ef AK |
2109 | char *ep; |
2110 | unsigned long long position = strtoull(buf, &ep, 0); | |
2111 | ||
2112 | dprintf(Name": Read sync_max sysfs entry is: %s\n", buf); | |
2113 | if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) { | |
2114 | position *= get_data_disks(level, | |
2115 | info->new_layout, | |
2116 | info->array.raid_disks); | |
2117 | if (info->reshape_progress < position) { | |
2118 | dprintf("Corrected reshape progress (%llu) to " | |
2119 | "md position (%llu)\n", | |
2120 | info->reshape_progress, position); | |
2121 | info->reshape_progress = position; | |
2122 | ret_val = 1; | |
2123 | } else if (info->reshape_progress > position) { | |
e7b84f9d N |
2124 | pr_err("Fatal error: array " |
2125 | "reshape was not properly frozen " | |
2126 | "(expected reshape position is %llu, " | |
2127 | "but reshape progress is %llu.\n", | |
2128 | position, info->reshape_progress); | |
f93346ef AK |
2129 | ret_val = -1; |
2130 | } else { | |
2131 | dprintf("Reshape position in md and metadata " | |
2132 | "are the same;"); | |
2133 | ret_val = 1; | |
2134 | } | |
2135 | } | |
178950ea AK |
2136 | } else if (rv == 0) { |
2137 | /* for valid sysfs entry, 0-length content | |
2138 | * should be indicated as error | |
2139 | */ | |
2140 | ret_val = -1; | |
f93346ef AK |
2141 | } |
2142 | ||
2143 | return ret_val; | |
2144 | } | |
2145 | ||
534f5432 N |
2146 | static unsigned long long choose_offset(unsigned long long lo, |
2147 | unsigned long long hi, | |
2148 | unsigned long long min, | |
2149 | unsigned long long max) | |
2150 | { | |
2151 | /* Choose a new offset between hi and lo. | |
2152 | * It must be between min and max, but | |
2153 | * we would prefer something near the middle of hi/lo, and also | |
2154 | * prefer to be aligned to a big power of 2. | |
2155 | * | |
2156 | * So we start with the middle, then for each bit, | |
2157 | * starting at '1' and increasing, if it is set, we either | |
2158 | * add it or subtract it if possible, preferring the option | |
2159 | * which is furthest from the boundary. | |
2160 | * | |
2161 | * We stop once we get a 1MB alignment. As units are in sectors, | |
2162 | * 1MB = 2*1024 sectors. | |
2163 | */ | |
2164 | unsigned long long choice = (lo + hi) / 2; | |
2165 | unsigned long long bit = 1; | |
2166 | ||
2167 | for (bit = 1; bit < 2*1024; bit = bit << 1) { | |
2168 | unsigned long long bigger, smaller; | |
2169 | if (! (bit & choice)) | |
2170 | continue; | |
2171 | bigger = choice + bit; | |
2172 | smaller = choice - bit; | |
2173 | if (bigger > max && smaller < min) | |
2174 | break; | |
2175 | if (bigger > max) | |
2176 | choice = smaller; | |
2177 | else if (smaller < min) | |
2178 | choice = bigger; | |
2179 | else if (hi - bigger > smaller - lo) | |
2180 | choice = bigger; | |
2181 | else | |
2182 | choice = smaller; | |
2183 | } | |
2184 | return choice; | |
2185 | } | |
2186 | ||
13bbb145 | 2187 | static int set_new_data_offset(struct mdinfo *sra, struct supertype *st, |
50a4962f | 2188 | char *devname, int delta_disks, |
13bbb145 | 2189 | unsigned long long data_offset, |
6a23fb9d N |
2190 | unsigned long long min, |
2191 | int can_fallback) | |
19ceb16d | 2192 | { |
13bbb145 | 2193 | struct mdinfo *sd; |
19ceb16d N |
2194 | int dir = 0; |
2195 | int err = 0; | |
f9b08fec | 2196 | unsigned long long before, after; |
19ceb16d | 2197 | |
f9b08fec N |
2198 | /* Need to find min space before and after so same is used |
2199 | * on all devices | |
2200 | */ | |
2201 | before = UINT64_MAX; | |
2202 | after = UINT64_MAX; | |
19ceb16d N |
2203 | for (sd = sra->devs; sd; sd = sd->next) { |
2204 | char *dn; | |
2205 | int dfd; | |
2206 | int rv; | |
2207 | struct supertype *st2; | |
2208 | struct mdinfo info2; | |
2209 | ||
2210 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2211 | continue; | |
2212 | dn = map_dev(sd->disk.major, sd->disk.minor, 0); | |
2213 | dfd = dev_open(dn, O_RDONLY); | |
2214 | if (dfd < 0) { | |
ed503f89 | 2215 | pr_err("%s: cannot open component %s\n", |
19ceb16d | 2216 | devname, dn ? dn : "-unknown-"); |
19ceb16d N |
2217 | goto release; |
2218 | } | |
2219 | st2 = dup_super(st); | |
2220 | rv = st2->ss->load_super(st2,dfd, NULL); | |
2221 | close(dfd); | |
2222 | if (rv) { | |
2223 | free(st2); | |
ed503f89 | 2224 | pr_err("%s: cannot get superblock from %s\n", |
19ceb16d N |
2225 | devname, dn); |
2226 | goto release; | |
2227 | } | |
2228 | st2->ss->getinfo_super(st2, &info2, NULL); | |
2229 | st2->ss->free_super(st2); | |
2230 | free(st2); | |
c4b26c64 N |
2231 | if (info2.space_before == 0 && |
2232 | info2.space_after == 0) { | |
2233 | /* Metadata doesn't support data_offset changes */ | |
2234 | return 1; | |
2235 | } | |
f9b08fec N |
2236 | if (before > info2.space_before) |
2237 | before = info2.space_before; | |
2238 | if (after > info2.space_after) | |
2239 | after = info2.space_after; | |
2240 | ||
2241 | if (data_offset != INVALID_SECTORS) { | |
2242 | if (dir == 0) { | |
2243 | if (info2.data_offset == data_offset) { | |
2244 | pr_err("%s: already has that data_offset\n", | |
2245 | dn); | |
2246 | goto release; | |
2247 | } | |
2248 | if (data_offset < info2.data_offset) | |
2249 | dir = -1; | |
2250 | else | |
2251 | dir = 1; | |
2252 | } else if ((data_offset <= info2.data_offset && dir == 1) || | |
2253 | (data_offset >= info2.data_offset && dir == -1)) { | |
2254 | pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n", | |
2255 | dn); | |
2256 | goto release; | |
2257 | } | |
2258 | } | |
2259 | } | |
2260 | if (before == UINT64_MAX) | |
2261 | /* impossible really, there must be no devices */ | |
2262 | return 1; | |
2263 | ||
2264 | for (sd = sra->devs; sd; sd = sd->next) { | |
2265 | char *dn = map_dev(sd->disk.major, sd->disk.minor, 0); | |
c0f0d812 | 2266 | unsigned long long new_data_offset; |
f9b08fec | 2267 | |
b397d7f3 N |
2268 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) |
2269 | continue; | |
50a4962f | 2270 | if (delta_disks < 0) { |
19ceb16d N |
2271 | /* Don't need any space as array is shrinking |
2272 | * just move data_offset up by min | |
2273 | */ | |
5582b118 | 2274 | if (data_offset == INVALID_SECTORS) |
c0f0d812 | 2275 | new_data_offset = sd->data_offset + min; |
19ceb16d | 2276 | else { |
c0f0d812 | 2277 | if (data_offset < sd->data_offset + min) { |
ed503f89 | 2278 | pr_err("--data-offset too small for %s\n", |
19ceb16d N |
2279 | dn); |
2280 | goto release; | |
2281 | } | |
c0f0d812 | 2282 | new_data_offset = data_offset; |
19ceb16d | 2283 | } |
50a4962f | 2284 | } else if (delta_disks > 0) { |
19ceb16d | 2285 | /* need space before */ |
f9b08fec | 2286 | if (before < min) { |
6a23fb9d N |
2287 | if (can_fallback) |
2288 | goto fallback; | |
ed503f89 | 2289 | pr_err("Insufficient head-space for reshape on %s\n", |
19ceb16d N |
2290 | dn); |
2291 | goto release; | |
2292 | } | |
5582b118 | 2293 | if (data_offset == INVALID_SECTORS) |
c0f0d812 | 2294 | new_data_offset = sd->data_offset - min; |
19ceb16d | 2295 | else { |
c0f0d812 | 2296 | if (data_offset > sd->data_offset - min) { |
ed503f89 | 2297 | pr_err("--data-offset too large for %s\n", |
19ceb16d N |
2298 | dn); |
2299 | goto release; | |
2300 | } | |
c0f0d812 | 2301 | new_data_offset = data_offset; |
19ceb16d N |
2302 | } |
2303 | } else { | |
2304 | if (dir == 0) { | |
f9b08fec N |
2305 | /* can move up or down. If 'data_offset' |
2306 | * was set we would have already decided, | |
2307 | * so just choose direction with most space. | |
19ceb16d | 2308 | */ |
c0f0d812 | 2309 | if (before > after) |
19ceb16d N |
2310 | dir = -1; |
2311 | else | |
2312 | dir = 1; | |
19ceb16d | 2313 | } |
b397d7f3 N |
2314 | sysfs_set_str(sra, NULL, "reshape_direction", |
2315 | dir == 1 ? "backwards" : "forwards"); | |
c0f0d812 N |
2316 | if (dir > 0) { |
2317 | /* Increase data offset */ | |
f9b08fec | 2318 | if (after < min) { |
6a23fb9d N |
2319 | if (can_fallback) |
2320 | goto fallback; | |
ed503f89 | 2321 | pr_err("Insufficient tail-space for reshape on %s\n", |
19ceb16d N |
2322 | dn); |
2323 | goto release; | |
2324 | } | |
5582b118 | 2325 | if (data_offset != INVALID_SECTORS && |
c0f0d812 | 2326 | data_offset < sd->data_offset + min) { |
ed503f89 | 2327 | pr_err("--data-offset too small on %s\n", |
19ceb16d N |
2328 | dn); |
2329 | goto release; | |
2330 | } | |
5582b118 | 2331 | if (data_offset != INVALID_SECTORS) |
c0f0d812 | 2332 | new_data_offset = data_offset; |
534f5432 N |
2333 | else |
2334 | new_data_offset = choose_offset(sd->data_offset, | |
2335 | sd->data_offset + after, | |
2336 | sd->data_offset + min, | |
2337 | sd->data_offset + after); | |
c0f0d812 N |
2338 | } else { |
2339 | /* Decrease data offset */ | |
f9b08fec | 2340 | if (before < min) { |
6a23fb9d N |
2341 | if (can_fallback) |
2342 | goto fallback; | |
ed503f89 | 2343 | pr_err("insufficient head-room on %s\n", |
19ceb16d N |
2344 | dn); |
2345 | goto release; | |
2346 | } | |
5582b118 | 2347 | if (data_offset != INVALID_SECTORS && |
c0f0d812 | 2348 | data_offset < sd->data_offset - min) { |
ed503f89 | 2349 | pr_err("--data-offset too small on %s\n", |
19ceb16d N |
2350 | dn); |
2351 | goto release; | |
2352 | } | |
5582b118 | 2353 | if (data_offset != INVALID_SECTORS) |
c0f0d812 | 2354 | new_data_offset = data_offset; |
534f5432 N |
2355 | else |
2356 | new_data_offset = choose_offset(sd->data_offset - before, | |
2357 | sd->data_offset, | |
2358 | sd->data_offset - before, | |
2359 | sd->data_offset - min); | |
19ceb16d N |
2360 | } |
2361 | } | |
a6a78630 N |
2362 | err = sysfs_set_num(sra, sd, "new_offset", new_data_offset); |
2363 | if (err < 0 && errno == E2BIG) { | |
2364 | /* try again after increasing data size to max */ | |
2365 | err = sysfs_set_num(sra, sd, "size", 0); | |
2366 | if (err < 0 && errno == EINVAL && | |
2367 | !(sd->disk.state & (1<<MD_DISK_SYNC))) { | |
2368 | /* some kernels have a bug where you cannot | |
2369 | * use '0' on spare devices. */ | |
2370 | sysfs_set_num(sra, sd, "size", | |
2371 | (sra->component_size + after)/2); | |
2372 | } | |
2373 | err = sysfs_set_num(sra, sd, "new_offset", | |
2374 | new_data_offset); | |
2375 | } | |
2376 | if (err < 0) { | |
d7e1f52b N |
2377 | if (errno == E2BIG && data_offset != INVALID_SECTORS) { |
2378 | pr_err("data-offset is too big for %s\n", | |
2379 | dn); | |
2380 | goto release; | |
2381 | } | |
9ad2a640 N |
2382 | if (sd == sra->devs && |
2383 | (errno == ENOENT || errno == E2BIG)) | |
2384 | /* Early kernel, no 'new_offset' file, | |
2385 | * or kernel doesn't like us. | |
93f174b9 N |
2386 | * For RAID5/6 this is not fatal |
2387 | */ | |
2388 | return 1; | |
ed503f89 | 2389 | pr_err("Cannot set new_offset for %s\n", |
19ceb16d N |
2390 | dn); |
2391 | break; | |
2392 | } | |
2393 | } | |
13bbb145 N |
2394 | return err; |
2395 | release: | |
2396 | return -1; | |
6a23fb9d N |
2397 | fallback: |
2398 | /* Just use a backup file */ | |
2399 | return 1; | |
13bbb145 N |
2400 | } |
2401 | ||
2402 | static int raid10_reshape(char *container, int fd, char *devname, | |
2403 | struct supertype *st, struct mdinfo *info, | |
2404 | struct reshape *reshape, | |
2405 | unsigned long long data_offset, | |
2406 | int force, int verbose) | |
2407 | { | |
2408 | /* Changing raid_disks, layout, chunksize or possibly | |
2409 | * just data_offset for a RAID10. | |
2410 | * We must always change data_offset. We change by at least | |
89ecd3cf | 2411 | * ->min_offset_change which is the largest of the old and new |
13bbb145 N |
2412 | * chunk sizes. |
2413 | * If raid_disks is increasing, then data_offset must decrease | |
2414 | * by at least this copy size. | |
2415 | * If raid_disks is unchanged, data_offset must increase or | |
89ecd3cf | 2416 | * decrease by at least min_offset_change but preferably by much more. |
13bbb145 N |
2417 | * We choose half of the available space. |
2418 | * If raid_disks is decreasing, data_offset must increase by | |
89ecd3cf | 2419 | * at least min_offset_change. To allow of this, component_size |
13bbb145 N |
2420 | * must be decreased by the same amount. |
2421 | * | |
2422 | * So we calculate the required minimum and direction, possibly | |
2423 | * reduce the component_size, then iterate through the devices | |
2424 | * and set the new_data_offset. | |
2425 | * If that all works, we set chunk_size, layout, raid_disks, and start | |
2426 | * 'reshape' | |
2427 | */ | |
2428 | struct mdinfo *sra; | |
2429 | unsigned long long min; | |
2430 | int err = 0; | |
2431 | ||
2432 | sra = sysfs_read(fd, NULL, | |
2433 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK | |
2434 | ); | |
2435 | if (!sra) { | |
ed503f89 | 2436 | pr_err("%s: Cannot get array details from sysfs\n", |
13bbb145 N |
2437 | devname); |
2438 | goto release; | |
2439 | } | |
89ecd3cf | 2440 | min = reshape->min_offset_change; |
13bbb145 N |
2441 | |
2442 | if (info->delta_disks) | |
2443 | sysfs_set_str(sra, NULL, "reshape_direction", | |
2444 | info->delta_disks < 0 ? "backwards" : "forwards"); | |
2445 | if (info->delta_disks < 0 && | |
89ecd3cf | 2446 | info->space_after < min) { |
13bbb145 N |
2447 | int rv = sysfs_set_num(sra, NULL, "component_size", |
2448 | (sra->component_size - | |
89ecd3cf | 2449 | min)/2); |
13bbb145 | 2450 | if (rv) { |
ed503f89 | 2451 | pr_err("cannot reduce component size\n"); |
13bbb145 N |
2452 | goto release; |
2453 | } | |
2454 | } | |
50a4962f | 2455 | err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset, |
6a23fb9d | 2456 | min, 0); |
93f174b9 N |
2457 | if (err == 1) { |
2458 | pr_err("Cannot set new_data_offset: RAID10 reshape not\n"); | |
2459 | cont_err("supported on this kernel\n"); | |
2460 | err = -1; | |
2461 | } | |
13bbb145 N |
2462 | if (err < 0) |
2463 | goto release; | |
2464 | ||
ee2429e0 | 2465 | if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0) |
19ceb16d N |
2466 | err = errno; |
2467 | if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0) | |
2468 | err = errno; | |
2469 | if (!err && sysfs_set_num(sra, NULL, "raid_disks", | |
2470 | info->array.raid_disks + info->delta_disks) < 0) | |
2471 | err = errno; | |
2472 | if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) | |
2473 | err = errno; | |
2474 | if (err) { | |
ed503f89 | 2475 | pr_err("Cannot set array shape for %s\n", |
d33f1518 N |
2476 | devname); |
2477 | if (err == EBUSY && | |
2478 | (info->array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
2479 | cont_err(" Bitmap must be removed before" | |
2480 | " shape can be changed\n"); | |
2481 | goto release; | |
19ceb16d N |
2482 | } |
2483 | sysfs_free(sra); | |
2484 | return 0; | |
2485 | release: | |
2486 | sysfs_free(sra); | |
2487 | return 1; | |
2488 | } | |
2489 | ||
ee2429e0 N |
2490 | static void get_space_after(int fd, struct supertype *st, struct mdinfo *info) |
2491 | { | |
2492 | struct mdinfo *sra, *sd; | |
f3f09a52 LD |
2493 | /* Initialisation to silence compiler warning */ |
2494 | unsigned long long min_space_before = 0, min_space_after = 0; | |
ee2429e0 N |
2495 | int first = 1; |
2496 | ||
4dd2df09 | 2497 | sra = sysfs_read(fd, NULL, GET_DEVS); |
ee2429e0 N |
2498 | if (!sra) |
2499 | return; | |
2500 | for (sd = sra->devs; sd; sd = sd->next) { | |
2501 | char *dn; | |
2502 | int dfd; | |
2503 | struct supertype *st2; | |
2504 | struct mdinfo info2; | |
2505 | ||
2506 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2507 | continue; | |
2508 | dn = map_dev(sd->disk.major, sd->disk.minor, 0); | |
2509 | dfd = dev_open(dn, O_RDONLY); | |
2510 | if (dfd < 0) | |
2511 | break; | |
2512 | st2 = dup_super(st); | |
2513 | if (st2->ss->load_super(st2,dfd, NULL)) { | |
2514 | close(dfd); | |
2515 | free(st2); | |
2516 | break; | |
2517 | } | |
2518 | close(dfd); | |
2519 | st2->ss->getinfo_super(st2, &info2, NULL); | |
2520 | st2->ss->free_super(st2); | |
2521 | free(st2); | |
2522 | if (first || | |
2523 | min_space_before > info2.space_before) | |
2524 | min_space_before = info2.space_before; | |
2525 | if (first || | |
2526 | min_space_after > info2.space_after) | |
2527 | min_space_after = info2.space_after; | |
2528 | first = 0; | |
2529 | } | |
2530 | if (sd == NULL && !first) { | |
2531 | info->space_after = min_space_after; | |
2532 | info->space_before = min_space_before; | |
2533 | } | |
2534 | sysfs_free(sra); | |
2535 | } | |
2536 | ||
434d167e N |
2537 | static void update_cache_size(char *container, struct mdinfo *sra, |
2538 | struct mdinfo *info, | |
2539 | int disks, unsigned long long blocks) | |
2540 | { | |
2541 | /* Check that the internal stripe cache is | |
2542 | * large enough, or it won't work. | |
2543 | * It must hold at least 4 stripes of the larger | |
2544 | * chunk size | |
2545 | */ | |
2546 | unsigned long cache; | |
2547 | cache = max(info->array.chunk_size, info->new_chunk); | |
2548 | cache *= 4; /* 4 stripes minimum */ | |
2549 | cache /= 512; /* convert to sectors */ | |
2550 | /* make sure there is room for 'blocks' with a bit to spare */ | |
2551 | if (cache < 16 + blocks / disks) | |
2552 | cache = 16 + blocks / disks; | |
2553 | cache /= (4096/512); /* Covert from sectors to pages */ | |
2554 | ||
2555 | if (sra->cache_size < cache) | |
2556 | subarray_set_num(container, sra, "stripe_cache_size", | |
2557 | cache+1); | |
2558 | } | |
2559 | ||
77afa056 N |
2560 | static int impose_reshape(struct mdinfo *sra, |
2561 | struct mdinfo *info, | |
2562 | struct supertype *st, | |
2563 | int fd, | |
2564 | int restart, | |
2565 | char *devname, char *container, | |
2566 | struct reshape *reshape) | |
2567 | { | |
2568 | struct mdu_array_info_s array; | |
2569 | ||
2570 | sra->new_chunk = info->new_chunk; | |
2571 | ||
2572 | if (restart) { | |
2573 | /* for external metadata checkpoint saved by mdmon can be lost | |
2574 | * or missed /due to e.g. crash/. Check if md is not during | |
2575 | * restart farther than metadata points to. | |
2576 | * If so, this means metadata information is obsolete. | |
2577 | */ | |
2578 | if (st->ss->external) | |
2579 | verify_reshape_position(info, reshape->level); | |
2580 | sra->reshape_progress = info->reshape_progress; | |
2581 | } else { | |
2582 | sra->reshape_progress = 0; | |
2583 | if (reshape->after.data_disks < reshape->before.data_disks) | |
2584 | /* start from the end of the new array */ | |
2585 | sra->reshape_progress = (sra->component_size | |
2586 | * reshape->after.data_disks); | |
2587 | } | |
2588 | ||
2589 | ioctl(fd, GET_ARRAY_INFO, &array); | |
2590 | if (info->array.chunk_size == info->new_chunk && | |
2591 | reshape->before.layout == reshape->after.layout && | |
2592 | st->ss->external == 0) { | |
2593 | /* use SET_ARRAY_INFO but only if reshape hasn't started */ | |
2594 | array.raid_disks = reshape->after.data_disks + reshape->parity; | |
2595 | if (!restart && | |
2596 | ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
2597 | int err = errno; | |
2598 | ||
2599 | pr_err("Cannot set device shape for %s: %s\n", | |
2600 | devname, strerror(errno)); | |
2601 | ||
2602 | if (err == EBUSY && | |
2603 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
2604 | cont_err("Bitmap must be removed before" | |
2605 | " shape can be changed\n"); | |
2606 | ||
2607 | goto release; | |
2608 | } | |
2609 | } else if (!restart) { | |
2610 | /* set them all just in case some old 'new_*' value | |
2611 | * persists from some earlier problem. | |
2612 | */ | |
2613 | int err = 0; | |
2614 | if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0) | |
2615 | err = errno; | |
2616 | if (!err && sysfs_set_num(sra, NULL, "layout", | |
2617 | reshape->after.layout) < 0) | |
2618 | err = errno; | |
2619 | if (!err && subarray_set_num(container, sra, "raid_disks", | |
2620 | reshape->after.data_disks + | |
2621 | reshape->parity) < 0) | |
2622 | err = errno; | |
2623 | if (err) { | |
2624 | pr_err("Cannot set device shape for %s\n", | |
2625 | devname); | |
2626 | ||
2627 | if (err == EBUSY && | |
2628 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
2629 | cont_err("Bitmap must be removed before" | |
2630 | " shape can be changed\n"); | |
2631 | goto release; | |
2632 | } | |
2633 | } | |
2634 | return 0; | |
2635 | release: | |
2636 | return -1; | |
2637 | } | |
2638 | ||
97e3a6a0 N |
2639 | static int impose_level(int fd, int level, char *devname, int verbose) |
2640 | { | |
2641 | char *c; | |
2642 | struct mdu_array_info_s array; | |
2643 | struct mdinfo info; | |
2644 | sysfs_init(&info, fd, NULL); | |
2645 | ||
2646 | ioctl(fd, GET_ARRAY_INFO, &array); | |
2647 | if (level == 0 && | |
2648 | (array.level >= 4 && array.level <= 6)) { | |
2649 | /* To convert to RAID0 we need to fail and | |
2650 | * remove any non-data devices. */ | |
2651 | int found = 0; | |
2652 | int d; | |
2653 | int data_disks = array.raid_disks - 1; | |
2654 | if (array.level == 6) | |
2655 | data_disks -= 1; | |
2656 | if (array.level == 5 && | |
2657 | array.layout != ALGORITHM_PARITY_N) | |
2658 | return -1; | |
2659 | if (array.level == 6 && | |
2660 | array.layout != ALGORITHM_PARITY_N_6) | |
2661 | return -1; | |
2662 | sysfs_set_str(&info, NULL,"sync_action", "idle"); | |
2663 | /* First remove any spares so no recovery starts */ | |
2664 | for (d = 0, found = 0; | |
2665 | d < MAX_DISKS && found < array.nr_disks; | |
2666 | d++) { | |
2667 | mdu_disk_info_t disk; | |
2668 | disk.number = d; | |
2669 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
2670 | continue; | |
2671 | if (disk.major == 0 && disk.minor == 0) | |
2672 | continue; | |
2673 | found++; | |
2674 | if ((disk.state & (1 << MD_DISK_ACTIVE)) | |
2675 | && disk.raid_disk < data_disks) | |
2676 | /* keep this */ | |
2677 | continue; | |
2678 | ioctl(fd, HOT_REMOVE_DISK, | |
2679 | makedev(disk.major, disk.minor)); | |
2680 | } | |
2681 | /* Now fail anything left */ | |
2682 | ioctl(fd, GET_ARRAY_INFO, &array); | |
2683 | for (d = 0, found = 0; | |
2684 | d < MAX_DISKS && found < array.nr_disks; | |
2685 | d++) { | |
2686 | int cnt; | |
2687 | mdu_disk_info_t disk; | |
2688 | disk.number = d; | |
2689 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
2690 | continue; | |
2691 | if (disk.major == 0 && disk.minor == 0) | |
2692 | continue; | |
2693 | found++; | |
2694 | if ((disk.state & (1 << MD_DISK_ACTIVE)) | |
2695 | && disk.raid_disk < data_disks) | |
2696 | /* keep this */ | |
2697 | continue; | |
2698 | ioctl(fd, SET_DISK_FAULTY, | |
2699 | makedev(disk.major, disk.minor)); | |
2700 | cnt = 5; | |
2701 | while (ioctl(fd, HOT_REMOVE_DISK, | |
2702 | makedev(disk.major, disk.minor)) < 0 | |
2703 | && errno == EBUSY | |
2704 | && cnt--) { | |
2705 | usleep(10000); | |
2706 | } | |
2707 | } | |
2708 | } | |
2709 | c = map_num(pers, level); | |
2710 | if (c) { | |
2711 | int err = sysfs_set_str(&info, NULL, "level", c); | |
2712 | if (err) { | |
2713 | err = errno; | |
2714 | pr_err("%s: could not set level to %s\n", | |
2715 | devname, c); | |
2716 | if (err == EBUSY && | |
2717 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
2718 | cont_err("Bitmap must be removed" | |
2719 | " before level can be changed\n"); | |
2720 | return err; | |
2721 | } | |
2722 | if (verbose >= 0) | |
2723 | pr_err("level of %s changed to %s\n", | |
2724 | devname, c); | |
2725 | } | |
2726 | return 0; | |
2727 | } | |
2728 | ||
5da9ab98 N |
2729 | static int reshape_array(char *container, int fd, char *devname, |
2730 | struct supertype *st, struct mdinfo *info, | |
e2e53a2d | 2731 | int force, struct mddev_dev *devlist, |
19ceb16d | 2732 | unsigned long long data_offset, |
ba728be7 | 2733 | char *backup_file, int verbose, int forked, |
b76b30e0 | 2734 | int restart, int freeze_reshape) |
5da9ab98 N |
2735 | { |
2736 | struct reshape reshape; | |
2737 | int spares_needed; | |
2738 | char *msg; | |
2739 | int orig_level = UnSet; | |
434d167e | 2740 | int odisks; |
1f9b0e28 | 2741 | int delayed; |
7bc71196 | 2742 | |
5da9ab98 N |
2743 | struct mdu_array_info_s array; |
2744 | char *c; | |
e86c9dd6 | 2745 | |
e2e53a2d N |
2746 | struct mddev_dev *dv; |
2747 | int added_disks; | |
2748 | ||
d152f53e JS |
2749 | int *fdlist = NULL; |
2750 | unsigned long long *offsets = NULL; | |
5da9ab98 N |
2751 | int d; |
2752 | int nrdisks; | |
2753 | int err; | |
da8100ca | 2754 | unsigned long blocks; |
5da9ab98 N |
2755 | unsigned long long array_size; |
2756 | int done; | |
cf6ac177 | 2757 | struct mdinfo *sra = NULL; |
7bc71196 | 2758 | |
9468aeac N |
2759 | /* when reshaping a RAID0, the component_size might be zero. |
2760 | * So try to fix that up. | |
2761 | */ | |
2762 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
2763 | dprintf("Cannot get array information.\n"); | |
2764 | goto release; | |
2765 | } | |
2766 | if (array.level == 0 && info->component_size == 0) { | |
2767 | get_dev_size(fd, NULL, &array_size); | |
2768 | info->component_size = array_size / array.raid_disks; | |
2769 | } | |
2770 | ||
ee2429e0 N |
2771 | if (array.level == 10) |
2772 | /* Need space_after info */ | |
2773 | get_space_after(fd, st, info); | |
2774 | ||
3cb2aed2 N |
2775 | if (info->reshape_active) { |
2776 | int new_level = info->new_level; | |
2777 | info->new_level = UnSet; | |
ce4783d3 N |
2778 | if (info->delta_disks > 0) |
2779 | info->array.raid_disks -= info->delta_disks; | |
a73b0081 | 2780 | msg = analyse_change(devname, info, &reshape); |
3cb2aed2 | 2781 | info->new_level = new_level; |
ce4783d3 N |
2782 | if (info->delta_disks > 0) |
2783 | info->array.raid_disks += info->delta_disks; | |
fcdfb814 AK |
2784 | if (!restart) |
2785 | /* Make sure the array isn't read-only */ | |
2786 | ioctl(fd, RESTART_ARRAY_RW, 0); | |
3cb2aed2 | 2787 | } else |
a73b0081 | 2788 | msg = analyse_change(devname, info, &reshape); |
5da9ab98 | 2789 | if (msg) { |
a73b0081 N |
2790 | /* if msg == "", error has already been printed */ |
2791 | if (msg[0]) | |
2792 | pr_err("%s\n", msg); | |
631d7405 | 2793 | goto release; |
5da9ab98 | 2794 | } |
817ed7d6 N |
2795 | if (restart && |
2796 | (reshape.level != info->array.level || | |
2797 | reshape.before.layout != info->array.layout || | |
384e9be1 | 2798 | reshape.before.data_disks + reshape.parity |
ce4783d3 | 2799 | != info->array.raid_disks - max(0, info->delta_disks))) { |
e7b84f9d | 2800 | pr_err("reshape info is not in native format -" |
20a40eca N |
2801 | " cannot continue.\n"); |
2802 | goto release; | |
2803 | } | |
a93f87ee | 2804 | |
e1dd332a AK |
2805 | if (st->ss->external && restart && (info->reshape_progress == 0)) { |
2806 | /* When reshape is restarted from '0', very begin of array | |
2807 | * it is possible that for external metadata reshape and array | |
2808 | * configuration doesn't happen. | |
2809 | * Check if md has the same opinion, and reshape is restarted | |
2810 | * from 0. If so, this is regular reshape start after reshape | |
2811 | * switch in metadata to next array only. | |
2812 | */ | |
2813 | if ((verify_reshape_position(info, reshape.level) >= 0) && | |
2814 | (info->reshape_progress == 0)) | |
2815 | restart = 0; | |
2816 | } | |
a93f87ee N |
2817 | if (restart) { |
2818 | /* reshape already started. just skip to monitoring the reshape */ | |
2819 | if (reshape.backup_blocks == 0) | |
2820 | return 0; | |
8ecf12b9 N |
2821 | if (restart & RESHAPE_NO_BACKUP) |
2822 | return 0; | |
dfa4d769 N |
2823 | |
2824 | /* Need 'sra' down at 'started:' */ | |
2825 | sra = sysfs_read(fd, NULL, | |
2826 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK| | |
2827 | GET_CACHE); | |
2828 | if (!sra) { | |
2829 | pr_err("%s: Cannot get array details from sysfs\n", | |
2830 | devname); | |
2831 | goto release; | |
2832 | } | |
a93f87ee N |
2833 | goto started; |
2834 | } | |
a9c3e78f AK |
2835 | /* The container is frozen but the array may not be. |
2836 | * So freeze the array so spares don't get put to the wrong use | |
2837 | * FIXME there should probably be a cleaner separation between | |
2838 | * freeze_array and freeze_container. | |
2839 | */ | |
2840 | sysfs_freeze_array(info); | |
e06c4e59 | 2841 | /* Check we have enough spares to not be degraded */ |
e2e53a2d N |
2842 | added_disks = 0; |
2843 | for (dv = devlist; dv ; dv=dv->next) | |
2844 | added_disks++; | |
5da9ab98 N |
2845 | spares_needed = max(reshape.before.data_disks, |
2846 | reshape.after.data_disks) | |
2847 | + reshape.parity - array.raid_disks; | |
7bc71196 | 2848 | |
88c1a083 | 2849 | if (!force && |
e2e53a2d N |
2850 | info->new_level > 1 && info->array.level > 1 && |
2851 | spares_needed > info->array.spare_disks + added_disks) { | |
e7b84f9d N |
2852 | pr_err("Need %d spare%s to avoid degraded array," |
2853 | " and only have %d.\n" | |
2854 | " Use --force to over-ride this check.\n", | |
2855 | spares_needed, | |
2856 | spares_needed == 1 ? "" : "s", | |
2857 | info->array.spare_disks + added_disks); | |
631d7405 | 2858 | goto release; |
7bc71196 | 2859 | } |
e06c4e59 N |
2860 | /* Check we have enough spares to not fail */ |
2861 | spares_needed = max(reshape.before.data_disks, | |
2862 | reshape.after.data_disks) | |
2863 | - array.raid_disks; | |
2864 | if ((info->new_level > 1 || info->new_level == 0) && | |
e2e53a2d | 2865 | spares_needed > info->array.spare_disks +added_disks) { |
e7b84f9d N |
2866 | pr_err("Need %d spare%s to create working array," |
2867 | " and only have %d.\n", | |
2868 | spares_needed, | |
2869 | spares_needed == 1 ? "" : "s", | |
2870 | info->array.spare_disks + added_disks); | |
e06c4e59 N |
2871 | goto release; |
2872 | } | |
e86c9dd6 | 2873 | |
eae35f5c | 2874 | if (reshape.level != array.level) { |
97e3a6a0 N |
2875 | int err = impose_level(fd, reshape.level, devname, verbose); |
2876 | if (err) | |
631d7405 | 2877 | goto release; |
97e3a6a0 N |
2878 | info->new_layout = UnSet; /* after level change, |
2879 | * layout is meaningless */ | |
eae35f5c | 2880 | orig_level = array.level; |
3cd4e7c4 | 2881 | sysfs_freeze_array(info); |
e86c9dd6 | 2882 | |
16d4d84e AK |
2883 | if (reshape.level > 0 && st->ss->external) { |
2884 | /* make sure mdmon is aware of the new level */ | |
4dd2df09 | 2885 | if (mdmon_running(container)) |
78340e26 AK |
2886 | flush_mdmon(container); |
2887 | ||
4dd2df09 N |
2888 | if (!mdmon_running(container)) |
2889 | start_mdmon(container); | |
16d4d84e | 2890 | ping_monitor(container); |
4dd2df09 | 2891 | if (mdmon_running(container) && |
e919fb0a AK |
2892 | st->update_tail == NULL) |
2893 | st->update_tail = &st->updates; | |
16d4d84e | 2894 | } |
5da9ab98 | 2895 | } |
5da9ab98 N |
2896 | /* ->reshape_super might have chosen some spares from the |
2897 | * container that it wants to be part of the new array. | |
2898 | * We can collect them with ->container_content and give | |
2899 | * them to the kernel. | |
2900 | */ | |
2901 | if (st->ss->reshape_super && st->ss->container_content) { | |
2902 | char *subarray = strchr(info->text_version+1, '/')+1; | |
2903 | struct mdinfo *info2 = | |
2904 | st->ss->container_content(st, subarray); | |
2905 | struct mdinfo *d; | |
2906 | ||
2dd0d697 | 2907 | if (info2) { |
4dd2df09 | 2908 | sysfs_init(info2, fd, st->devnm); |
0d5ac3c6 N |
2909 | /* When increasing number of devices, we need to set |
2910 | * new raid_disks before adding these, or they might | |
2911 | * be rejected. | |
2912 | */ | |
2913 | if (reshape.backup_blocks && | |
2914 | reshape.after.data_disks > reshape.before.data_disks) | |
2915 | subarray_set_num(container, info2, "raid_disks", | |
2916 | reshape.after.data_disks + | |
2917 | reshape.parity); | |
5da9ab98 N |
2918 | for (d = info2->devs; d; d = d->next) { |
2919 | if (d->disk.state == 0 && | |
2920 | d->disk.raid_disk >= 0) { | |
2921 | /* This is a spare that wants to | |
2922 | * be part of the array. | |
2923 | */ | |
2924 | add_disk(fd, st, info2, d); | |
2925 | } | |
9860f271 | 2926 | } |
2dd0d697 AK |
2927 | sysfs_free(info2); |
2928 | } | |
5da9ab98 | 2929 | } |
e2e53a2d N |
2930 | /* We might have been given some devices to add to the |
2931 | * array. Now that the array has been changed to the right | |
2932 | * level and frozen, we can safely add them. | |
2933 | */ | |
2934 | if (devlist) | |
ba728be7 | 2935 | Manage_subdevs(devname, fd, devlist, verbose, |
11b391ec | 2936 | 0,NULL, 0); |
1686dc25 | 2937 | |
e09233d0 | 2938 | if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS) |
8876bf0b | 2939 | reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512; |
b4f8e38b | 2940 | if (reshape.backup_blocks == 0) { |
5da9ab98 N |
2941 | /* No restriping needed, but we might need to impose |
2942 | * some more changes: layout, raid_disks, chunk_size | |
e86c9dd6 | 2943 | */ |
24aebf3a KW |
2944 | /* read current array info */ |
2945 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
83732c28 | 2946 | dprintf("Cannot get array information.\n"); |
24aebf3a KW |
2947 | goto release; |
2948 | } | |
2949 | /* compare current array info with new values and if | |
2950 | * it is different update them to new */ | |
5da9ab98 | 2951 | if (info->new_layout != UnSet && |
24aebf3a KW |
2952 | info->new_layout != array.layout) { |
2953 | array.layout = info->new_layout; | |
2954 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
e7b84f9d | 2955 | pr_err("failed to set new layout\n"); |
631d7405 | 2956 | goto release; |
ba728be7 | 2957 | } else if (verbose >= 0) |
5da9ab98 | 2958 | printf("layout for %s set to %d\n", |
24aebf3a | 2959 | devname, array.layout); |
5da9ab98 N |
2960 | } |
2961 | if (info->delta_disks != UnSet && | |
24aebf3a KW |
2962 | info->delta_disks != 0 && |
2963 | array.raid_disks != (info->array.raid_disks + info->delta_disks)) { | |
2964 | array.raid_disks += info->delta_disks; | |
2965 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
e7b84f9d | 2966 | pr_err("failed to set raid disks\n"); |
631d7405 | 2967 | goto release; |
ba728be7 | 2968 | } else if (verbose >= 0) { |
5da9ab98 | 2969 | printf("raid_disks for %s set to %d\n", |
24aebf3a KW |
2970 | devname, array.raid_disks); |
2971 | } | |
5da9ab98 N |
2972 | } |
2973 | if (info->new_chunk != 0 && | |
24aebf3a | 2974 | info->new_chunk != array.chunk_size) { |
5da9ab98 N |
2975 | if (sysfs_set_num(info, NULL, |
2976 | "chunk_size", info->new_chunk) != 0) { | |
e7b84f9d | 2977 | pr_err("failed to set chunk size\n"); |
631d7405 | 2978 | goto release; |
ba728be7 | 2979 | } else if (verbose >= 0) |
5da9ab98 | 2980 | printf("chunk size for %s set to %d\n", |
24aebf3a | 2981 | devname, array.chunk_size); |
4725bc31 | 2982 | } |
e35b189b | 2983 | unfreeze(st); |
631d7405 | 2984 | return 0; |
5da9ab98 | 2985 | } |
19678e53 | 2986 | |
5da9ab98 N |
2987 | /* |
2988 | * There are three possibilities. | |
2989 | * 1/ The array will shrink. | |
2990 | * We need to ensure the reshape will pause before reaching | |
2991 | * the 'critical section'. We also need to fork and wait for | |
24daa16f | 2992 | * that to happen. When it does we |
5da9ab98 N |
2993 | * suspend/backup/complete/unfreeze |
2994 | * | |
2995 | * 2/ The array will not change size. | |
2996 | * This requires that we keep a backup of a sliding window | |
2997 | * so that we can restore data after a crash. So we need | |
2998 | * to fork and monitor progress. | |
2999 | * In future we will allow the data_offset to change, so | |
3000 | * a sliding backup becomes unnecessary. | |
3001 | * | |
3002 | * 3/ The array will grow. This is relatively easy. | |
3003 | * However the kernel's restripe routines will cheerfully | |
3004 | * overwrite some early data before it is safe. So we | |
3005 | * need to make a backup of the early parts of the array | |
3006 | * and be ready to restore it if rebuild aborts very early. | |
3007 | * For externally managed metadata, we still need a forked | |
3008 | * child to monitor the reshape and suspend IO over the region | |
3009 | * that is being reshaped. | |
3010 | * | |
3011 | * We backup data by writing it to one spare, or to a | |
3012 | * file which was given on command line. | |
3013 | * | |
3014 | * In each case, we first make sure that storage is available | |
3015 | * for the required backup. | |
3016 | * Then we: | |
3017 | * - request the shape change. | |
3018 | * - fork to handle backup etc. | |
3019 | */ | |
5da9ab98 N |
3020 | /* Check that we can hold all the data */ |
3021 | get_dev_size(fd, NULL, &array_size); | |
3022 | if (reshape.new_size < (array_size/512)) { | |
e7b84f9d N |
3023 | pr_err("this change will reduce the size of the array.\n" |
3024 | " use --grow --array-size first to truncate array.\n" | |
3025 | " e.g. mdadm --grow %s --array-size %llu\n", | |
3026 | devname, reshape.new_size/2); | |
5da9ab98 N |
3027 | goto release; |
3028 | } | |
7236ee7a | 3029 | |
19ceb16d | 3030 | if (array.level == 10) { |
033d0929 | 3031 | /* Reshaping RAID10 does not require any data backup by |
19ceb16d N |
3032 | * user-space. Instead it requires that the data_offset |
3033 | * is changed to avoid the need for backup. | |
3034 | * So this is handled very separately | |
3035 | */ | |
3036 | if (restart) | |
3037 | /* Nothing to do. */ | |
3038 | return 0; | |
3039 | return raid10_reshape(container, fd, devname, st, info, | |
3040 | &reshape, data_offset, | |
3041 | force, verbose); | |
3042 | } | |
4dd2df09 | 3043 | sra = sysfs_read(fd, NULL, |
3656edcd | 3044 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK| |
5da9ab98 | 3045 | GET_CACHE); |
5da9ab98 | 3046 | if (!sra) { |
e7b84f9d | 3047 | pr_err("%s: Cannot get array details from sysfs\n", |
5da9ab98 | 3048 | devname); |
5da9ab98 N |
3049 | goto release; |
3050 | } | |
7236ee7a | 3051 | |
8192902f | 3052 | if (!backup_file) |
199f1a1f N |
3053 | switch(set_new_data_offset(sra, st, devname, |
3054 | reshape.after.data_disks - reshape.before.data_disks, | |
8192902f | 3055 | data_offset, |
6a23fb9d | 3056 | reshape.min_offset_change, 1)) { |
63c12c89 N |
3057 | case -1: |
3058 | goto release; | |
3059 | case 0: | |
3060 | /* Updated data_offset, so it's easy now */ | |
3061 | update_cache_size(container, sra, info, | |
3062 | min(reshape.before.data_disks, | |
3063 | reshape.after.data_disks), | |
3064 | reshape.backup_blocks); | |
3065 | ||
3066 | /* Right, everything seems fine. Let's kick things off. | |
3067 | */ | |
3068 | sync_metadata(st); | |
3069 | ||
3070 | if (impose_reshape(sra, info, st, fd, restart, | |
3071 | devname, container, &reshape) < 0) | |
3072 | goto release; | |
3073 | if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) { | |
3074 | pr_err("Failed to initiate reshape!\n"); | |
3075 | goto release; | |
3076 | } | |
97e3a6a0 N |
3077 | if (info->new_level == reshape.level) |
3078 | return 0; | |
3079 | /* need to adjust level when reshape completes */ | |
3080 | switch(fork()) { | |
3081 | case -1: /* ignore error, but don't wait */ | |
3082 | return 0; | |
3083 | default: /* parent */ | |
3084 | return 0; | |
3085 | case 0: | |
3086 | map_fork(); | |
3087 | break; | |
3088 | } | |
3089 | wait_reshape(sra); | |
3090 | impose_level(fd, info->new_level, devname, verbose); | |
63c12c89 N |
3091 | |
3092 | return 0; | |
3093 | case 1: /* Couldn't set data_offset, try the old way */ | |
3094 | if (data_offset != INVALID_SECTORS) { | |
3095 | pr_err("Cannot update data_offset on this array\n"); | |
3096 | goto release; | |
3097 | } | |
3098 | break; | |
3099 | } | |
3100 | ||
8ecf12b9 | 3101 | started: |
5da9ab98 N |
3102 | /* Decide how many blocks (sectors) for a reshape |
3103 | * unit. The number we have so far is just a minimum | |
3104 | */ | |
b4f8e38b | 3105 | blocks = reshape.backup_blocks; |
24daa16f | 3106 | if (reshape.before.data_disks == |
5da9ab98 N |
3107 | reshape.after.data_disks) { |
3108 | /* Make 'blocks' bigger for better throughput, but | |
3109 | * not so big that we reject it below. | |
3110 | * Try for 16 megabytes | |
3111 | */ | |
3112 | while (blocks * 32 < sra->component_size && | |
3113 | blocks < 16*1024*2) | |
3114 | blocks *= 2; | |
3115 | } else | |
e7b84f9d | 3116 | pr_err("Need to backup %luK of critical " |
5da9ab98 N |
3117 | "section..\n", blocks/2); |
3118 | ||
3119 | if (blocks >= sra->component_size/2) { | |
e7b84f9d | 3120 | pr_err("%s: Something wrong" |
5da9ab98 N |
3121 | " - reshape aborted\n", |
3122 | devname); | |
5da9ab98 N |
3123 | goto release; |
3124 | } | |
7236ee7a | 3125 | |
5da9ab98 N |
3126 | /* Now we need to open all these devices so we can read/write. |
3127 | */ | |
b8b286a6 N |
3128 | nrdisks = max(reshape.before.data_disks, |
3129 | reshape.after.data_disks) + reshape.parity | |
3130 | + sra->array.spare_disks; | |
503975b9 N |
3131 | fdlist = xcalloc((1+nrdisks), sizeof(int)); |
3132 | offsets = xcalloc((1+nrdisks), sizeof(offsets[0])); | |
e380d3be | 3133 | |
b8b286a6 N |
3134 | odisks = reshape.before.data_disks + reshape.parity; |
3135 | d = reshape_prepare_fdlist(devname, sra, odisks, | |
5da9ab98 N |
3136 | nrdisks, blocks, backup_file, |
3137 | fdlist, offsets); | |
3138 | if (d < 0) { | |
5da9ab98 N |
3139 | goto release; |
3140 | } | |
13c37ad3 AK |
3141 | if ((st->ss->manage_reshape == NULL) || |
3142 | (st->ss->recover_backup == NULL)) { | |
3143 | if (backup_file == NULL) { | |
3144 | if (reshape.after.data_disks <= | |
3145 | reshape.before.data_disks) { | |
e7b84f9d | 3146 | pr_err("%s: Cannot grow - " |
13c37ad3 AK |
3147 | "need backup-file\n", devname); |
3148 | goto release; | |
3149 | } else if (sra->array.spare_disks == 0) { | |
e7b84f9d | 3150 | pr_err("%s: Cannot grow - " |
13c37ad3 AK |
3151 | "need a spare or backup-file to backup " |
3152 | "critical section\n", devname); | |
3153 | goto release; | |
3154 | } | |
3155 | } else { | |
3156 | if (!reshape_open_backup_file(backup_file, fd, devname, | |
3157 | (signed)blocks, | |
3158 | fdlist+d, offsets+d, | |
3159 | restart)) { | |
3160 | goto release; | |
3161 | } | |
3162 | d++; | |
e86c9dd6 | 3163 | } |
5da9ab98 | 3164 | } |
7236ee7a | 3165 | |
434d167e N |
3166 | update_cache_size(container, sra, info, |
3167 | min(reshape.before.data_disks, reshape.after.data_disks), | |
3168 | blocks); | |
5da9ab98 N |
3169 | |
3170 | /* Right, everything seems fine. Let's kick things off. | |
3171 | * If only changing raid_disks, use ioctl, else use | |
3172 | * sysfs. | |
3173 | */ | |
3174 | sync_metadata(st); | |
3175 | ||
77afa056 N |
3176 | if (impose_reshape(sra, info, st, fd, restart, |
3177 | devname, container, &reshape) < 0) | |
3178 | goto release; | |
e86c9dd6 | 3179 | |
27a1e5b5 N |
3180 | err = start_reshape(sra, restart, reshape.before.data_disks, |
3181 | reshape.after.data_disks); | |
fab32c97 | 3182 | if (err) { |
e7b84f9d N |
3183 | pr_err("Cannot %s reshape for %s\n", |
3184 | restart ? "continue" : "start", | |
3185 | devname); | |
fab32c97 AK |
3186 | goto release; |
3187 | } | |
a93f87ee N |
3188 | if (restart) |
3189 | sysfs_set_str(sra, NULL, "array_state", "active"); | |
b76b30e0 AK |
3190 | if (freeze_reshape) { |
3191 | free(fdlist); | |
3192 | free(offsets); | |
3193 | sysfs_free(sra); | |
e7b84f9d | 3194 | pr_err("Reshape has to be continued from" |
59ab9f54 | 3195 | " location %llu when root filesystem has been mounted.\n", |
b76b30e0 AK |
3196 | sra->reshape_progress); |
3197 | return 1; | |
3198 | } | |
5da9ab98 N |
3199 | |
3200 | /* Now we just need to kick off the reshape and watch, while | |
3201 | * handling backups of the data... | |
3202 | * This is all done by a forked background process. | |
3203 | */ | |
3204 | switch(forked ? 0 : fork()) { | |
6b2d630c | 3205 | case -1: |
e7b84f9d | 3206 | pr_err("Cannot run child to monitor reshape: %s\n", |
6b2d630c | 3207 | strerror(errno)); |
6b2d630c | 3208 | abort_reshape(sra); |
631d7405 | 3209 | goto release; |
6b2d630c | 3210 | default: |
d152f53e JS |
3211 | free(fdlist); |
3212 | free(offsets); | |
3213 | sysfs_free(sra); | |
6b2d630c | 3214 | return 0; |
5da9ab98 | 3215 | case 0: |
cc700db3 | 3216 | map_fork(); |
6b2d630c N |
3217 | break; |
3218 | } | |
5da9ab98 | 3219 | |
1f9b0e28 N |
3220 | /* If another array on the same devices is busy, the |
3221 | * reshape will wait for them. This would mean that | |
3222 | * the first section that we suspend will stay suspended | |
3223 | * for a long time. So check on that possibility | |
3224 | * by looking for "DELAYED" in /proc/mdstat, and if found, | |
3225 | * wait a while | |
3226 | */ | |
3227 | do { | |
3228 | struct mdstat_ent *mds, *m; | |
3229 | delayed = 0; | |
3230 | mds = mdstat_read(0, 0); | |
ae0dcfbd | 3231 | for (m = mds; m; m = m->next) |
4dd2df09 | 3232 | if (strcmp(m->devnm, sra->sys_name) == 0) { |
1f9b0e28 N |
3233 | if (m->resync && |
3234 | m->percent == RESYNC_DELAYED) | |
3235 | delayed = 1; | |
3236 | if (m->resync == 0) | |
3237 | /* Haven't started the reshape thread | |
3238 | * yet, wait a bit | |
3239 | */ | |
3240 | delayed = 2; | |
3241 | break; | |
3242 | } | |
3243 | free_mdstat(mds); | |
3244 | if (delayed == 1 && get_linux_version() < 3007000) { | |
3245 | pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n" | |
3246 | " You might experience problems until other reshapes complete.\n"); | |
3247 | delayed = 0; | |
3248 | } | |
3249 | if (delayed) | |
3250 | sleep(30 - (delayed-1) * 25); | |
3251 | } while (delayed); | |
3252 | ||
6b2d630c N |
3253 | close(fd); |
3254 | if (check_env("MDADM_GROW_VERIFY")) | |
3255 | fd = open(devname, O_RDONLY | O_DIRECT); | |
3256 | else | |
3257 | fd = -1; | |
3258 | mlockall(MCL_FUTURE); | |
5da9ab98 | 3259 | |
6b2d630c N |
3260 | if (st->ss->external) { |
3261 | /* metadata handler takes it from here */ | |
3262 | done = st->ss->manage_reshape( | |
3263 | fd, sra, &reshape, st, blocks, | |
3264 | fdlist, offsets, | |
3265 | d - odisks, fdlist+odisks, | |
3266 | offsets+odisks); | |
3267 | } else | |
3268 | done = child_monitor( | |
3269 | fd, sra, &reshape, st, blocks, | |
3270 | fdlist, offsets, | |
3271 | d - odisks, fdlist+odisks, | |
3272 | offsets+odisks); | |
3273 | ||
d152f53e JS |
3274 | free(fdlist); |
3275 | free(offsets); | |
3276 | ||
6b2d630c N |
3277 | if (backup_file && done) |
3278 | unlink(backup_file); | |
3279 | if (!done) { | |
3280 | abort_reshape(sra); | |
3281 | goto out; | |
3282 | } | |
1cbc5680 N |
3283 | |
3284 | if (!st->ss->external && | |
3285 | !(reshape.before.data_disks != reshape.after.data_disks | |
3286 | && info->custom_array_size) && | |
3287 | info->new_level == reshape.level && | |
3288 | !forked) { | |
3289 | /* no need to wait for the reshape to finish as | |
3290 | * there is nothing more to do. | |
3291 | */ | |
d152f53e | 3292 | sysfs_free(sra); |
1cbc5680 N |
3293 | exit(0); |
3294 | } | |
3295 | wait_reshape(sra); | |
3296 | ||
3297 | if (st->ss->external) { | |
3298 | /* Re-load the metadata as much could have changed */ | |
4dd2df09 | 3299 | int cfd = open_dev(st->container_devnm); |
1cbc5680 | 3300 | if (cfd >= 0) { |
78340e26 | 3301 | flush_mdmon(container); |
1cbc5680 N |
3302 | st->ss->free_super(st); |
3303 | st->ss->load_container(st, cfd, container); | |
3304 | close(cfd); | |
3305 | } | |
3306 | } | |
3307 | ||
6b2d630c N |
3308 | /* set new array size if required customer_array_size is used |
3309 | * by this metadata. | |
3310 | */ | |
3311 | if (reshape.before.data_disks != | |
3312 | reshape.after.data_disks && | |
54397ed9 AK |
3313 | info->custom_array_size) |
3314 | set_array_size(st, info, info->text_version); | |
582496b2 | 3315 | |
6b2d630c | 3316 | if (info->new_level != reshape.level) { |
97e3a6a0 N |
3317 | if (fd < 0) |
3318 | fd = open(devname, O_RDONLY); | |
3319 | impose_level(fd, info->new_level, devname, verbose); | |
3320 | close(fd); | |
e919fb0a AK |
3321 | if (info->new_level == 0) |
3322 | st->update_tail = NULL; | |
7236ee7a | 3323 | } |
6b2d630c | 3324 | out: |
d152f53e | 3325 | sysfs_free(sra); |
6b2d630c N |
3326 | if (forked) |
3327 | return 0; | |
e84f2c00 | 3328 | unfreeze(st); |
6b2d630c | 3329 | exit(0); |
e86c9dd6 | 3330 | |
6b2d630c | 3331 | release: |
d152f53e JS |
3332 | free(fdlist); |
3333 | free(offsets); | |
1cbc5680 | 3334 | if (orig_level != UnSet && sra) { |
7236ee7a N |
3335 | c = map_num(pers, orig_level); |
3336 | if (c && sysfs_set_str(sra, NULL, "level", c) == 0) | |
e7b84f9d | 3337 | pr_err("aborting level change\n"); |
7236ee7a | 3338 | } |
d152f53e | 3339 | sysfs_free(sra); |
9202b8eb N |
3340 | if (!forked) |
3341 | unfreeze(st); | |
631d7405 | 3342 | return 1; |
7236ee7a | 3343 | } |
e86c9dd6 | 3344 | |
9ad6f6e6 AK |
3345 | /* mdfd handle is passed to be closed in child process (after fork). |
3346 | */ | |
493f5dd6 | 3347 | int reshape_container(char *container, char *devname, |
9ad6f6e6 | 3348 | int mdfd, |
24daa16f | 3349 | struct supertype *st, |
5da9ab98 N |
3350 | struct mdinfo *info, |
3351 | int force, | |
3352 | char *backup_file, | |
ba728be7 | 3353 | int verbose, int restart, int freeze_reshape) |
5da9ab98 | 3354 | { |
1bb174ba | 3355 | struct mdinfo *cc = NULL; |
bcc9e9ed | 3356 | int rv = restart; |
4dd2df09 | 3357 | char last_devnm[32] = ""; |
1bb174ba | 3358 | |
d8eb27f7 | 3359 | /* component_size is not meaningful for a container, |
7ec0996c | 3360 | * so pass '0' meaning 'no change' |
d8eb27f7 | 3361 | */ |
493f5dd6 | 3362 | if (!restart && |
7ec0996c | 3363 | reshape_super(st, 0, info->new_level, |
5da9ab98 | 3364 | info->new_layout, info->new_chunk, |
41784c88 | 3365 | info->array.raid_disks, info->delta_disks, |
016e00f5 | 3366 | backup_file, devname, APPLY_METADATA_CHANGES, |
ba728be7 | 3367 | verbose)) { |
9e325442 | 3368 | unfreeze(st); |
5da9ab98 | 3369 | return 1; |
9e325442 | 3370 | } |
5da9ab98 N |
3371 | |
3372 | sync_metadata(st); | |
3373 | ||
1bb174ba AK |
3374 | /* ping monitor to be sure that update is on disk |
3375 | */ | |
3376 | ping_monitor(container); | |
5da9ab98 N |
3377 | |
3378 | switch (fork()) { | |
3379 | case -1: /* error */ | |
3380 | perror("Cannot fork to complete reshape\n"); | |
9e325442 | 3381 | unfreeze(st); |
5da9ab98 N |
3382 | return 1; |
3383 | default: /* parent */ | |
b76b30e0 AK |
3384 | if (!freeze_reshape) |
3385 | printf(Name ": multi-array reshape continues" | |
3386 | " in background\n"); | |
5da9ab98 N |
3387 | return 0; |
3388 | case 0: /* child */ | |
cc700db3 | 3389 | map_fork(); |
5da9ab98 N |
3390 | break; |
3391 | } | |
3392 | ||
9ad6f6e6 AK |
3393 | /* close unused handle in child process |
3394 | */ | |
3395 | if (mdfd > -1) | |
3396 | close(mdfd); | |
3397 | ||
1bb174ba AK |
3398 | while(1) { |
3399 | /* For each member array with reshape_active, | |
3400 | * we need to perform the reshape. | |
3401 | * We pick the first array that needs reshaping and | |
3402 | * reshape it. reshape_array() will re-read the metadata | |
3403 | * so the next time through a different array should be | |
3404 | * ready for reshape. | |
493f5dd6 N |
3405 | * It is possible that the 'different' array will not |
3406 | * be assembled yet. In that case we simple exit. | |
3407 | * When it is assembled, the mdadm which assembles it | |
3408 | * will take over the reshape. | |
1bb174ba AK |
3409 | */ |
3410 | struct mdinfo *content; | |
5da9ab98 N |
3411 | int fd; |
3412 | struct mdstat_ent *mdstat; | |
5da9ab98 | 3413 | char *adev; |
4dd2df09 | 3414 | int devid; |
5da9ab98 | 3415 | |
1bb174ba | 3416 | sysfs_free(cc); |
5da9ab98 | 3417 | |
1bb174ba AK |
3418 | cc = st->ss->container_content(st, NULL); |
3419 | ||
3420 | for (content = cc; content ; content = content->next) { | |
3421 | char *subarray; | |
3422 | if (!content->reshape_active) | |
3423 | continue; | |
3424 | ||
3425 | subarray = strchr(content->text_version+1, '/')+1; | |
4dd2df09 | 3426 | mdstat = mdstat_by_subdev(subarray, container); |
1bb174ba AK |
3427 | if (!mdstat) |
3428 | continue; | |
1ca90aa6 | 3429 | if (mdstat->active == 0) { |
4dd2df09 N |
3430 | pr_err("Skipping inactive array %s.\n", |
3431 | mdstat->devnm); | |
1ca90aa6 AK |
3432 | free_mdstat(mdstat); |
3433 | mdstat = NULL; | |
3434 | continue; | |
3435 | } | |
1bb174ba AK |
3436 | break; |
3437 | } | |
3438 | if (!content) | |
3439 | break; | |
5da9ab98 | 3440 | |
4dd2df09 N |
3441 | devid = devnm2devid(mdstat->devnm); |
3442 | adev = map_dev(major(devid), minor(devid), 0); | |
5da9ab98 | 3443 | if (!adev) |
1bb174ba AK |
3444 | adev = content->text_version; |
3445 | ||
4dd2df09 | 3446 | fd = open_dev(mdstat->devnm); |
97a3490c AK |
3447 | if (fd < 0) { |
3448 | printf(Name ": Device %s cannot be opened for reshape.", | |
3449 | adev); | |
3450 | break; | |
3451 | } | |
3452 | ||
4dd2df09 | 3453 | if (strcmp(last_devnm, mdstat->devnm) == 0) { |
2d04d7e5 AK |
3454 | /* Do not allow for multiple reshape_array() calls for |
3455 | * the same array. | |
3456 | * It can happen when reshape_array() returns without | |
3457 | * error, when reshape is not finished (wrong reshape | |
3458 | * starting/continuation conditions). Mdmon doesn't | |
3459 | * switch to next array in container and reentry | |
3460 | * conditions for the same array occur. | |
3461 | * This is possibly interim until the behaviour of | |
3462 | * reshape_array is resolved(). | |
3463 | */ | |
3464 | printf(Name ": Multiple reshape execution detected for " | |
3465 | "device %s.", adev); | |
3466 | close(fd); | |
3467 | break; | |
3468 | } | |
4dd2df09 | 3469 | strcpy(last_devnm, mdstat->devnm); |
2d04d7e5 | 3470 | |
4dd2df09 | 3471 | sysfs_init(content, fd, mdstat->devnm); |
5da9ab98 | 3472 | |
4dd2df09 | 3473 | if (mdmon_running(container)) |
78340e26 AK |
3474 | flush_mdmon(container); |
3475 | ||
1bb174ba | 3476 | rv = reshape_array(container, fd, adev, st, |
19ceb16d | 3477 | content, force, NULL, 0ULL, |
ba728be7 | 3478 | backup_file, verbose, 1, restart, |
b76b30e0 | 3479 | freeze_reshape); |
5da9ab98 | 3480 | close(fd); |
b76b30e0 AK |
3481 | |
3482 | if (freeze_reshape) { | |
3483 | sysfs_free(cc); | |
3484 | exit(0); | |
3485 | } | |
3486 | ||
493f5dd6 | 3487 | restart = 0; |
5da9ab98 N |
3488 | if (rv) |
3489 | break; | |
78340e26 | 3490 | |
4dd2df09 | 3491 | if (mdmon_running(container)) |
78340e26 | 3492 | flush_mdmon(container); |
5da9ab98 | 3493 | } |
bcc9e9ed AK |
3494 | if (!rv) |
3495 | unfreeze(st); | |
1bb174ba | 3496 | sysfs_free(cc); |
5da9ab98 N |
3497 | exit(0); |
3498 | } | |
3499 | ||
7236ee7a N |
3500 | /* |
3501 | * We run a child process in the background which performs the following | |
3502 | * steps: | |
3503 | * - wait for resync to reach a certain point | |
3504 | * - suspend io to the following section | |
3505 | * - backup that section | |
3506 | * - allow resync to proceed further | |
3507 | * - resume io | |
3508 | * - discard the backup. | |
3509 | * | |
3510 | * When are combined in slightly different ways in the three cases. | |
3511 | * Grow: | |
3512 | * - suspend/backup/allow/wait/resume/discard | |
3513 | * Shrink: | |
3514 | * - allow/wait/suspend/backup/allow/wait/resume/discard | |
3515 | * same-size: | |
3516 | * - wait/resume/discard/suspend/backup/allow | |
3517 | * | |
3518 | * suspend/backup/allow always come together | |
3519 | * wait/resume/discard do too. | |
3520 | * For the same-size case we have two backups to improve flow. | |
24daa16f | 3521 | * |
7236ee7a | 3522 | */ |
e86c9dd6 | 3523 | |
7443ee81 N |
3524 | int progress_reshape(struct mdinfo *info, struct reshape *reshape, |
3525 | unsigned long long backup_point, | |
3526 | unsigned long long wait_point, | |
3527 | unsigned long long *suspend_point, | |
3528 | unsigned long long *reshape_completed) | |
3529 | { | |
3530 | /* This function is called repeatedly by the reshape manager. | |
3531 | * It determines how much progress can safely be made and allows | |
3532 | * that progress. | |
3533 | * - 'info' identifies the array and particularly records in | |
3534 | * ->reshape_progress the metadata's knowledge of progress | |
3535 | * This is a sector offset from the start of the array | |
3536 | * of the next array block to be relocated. This number | |
3537 | * may increase from 0 or decrease from array_size, depending | |
3538 | * on the type of reshape that is happening. | |
3539 | * Note that in contrast, 'sync_completed' is a block count of the | |
3540 | * reshape so far. It gives the distance between the start point | |
3541 | * (head or tail of device) and the next place that data will be | |
3542 | * written. It always increases. | |
3543 | * - 'reshape' is the structure created by analyse_change | |
3544 | * - 'backup_point' shows how much the metadata manager has backed-up | |
3545 | * data. For reshapes with increasing progress, it is the next address | |
3546 | * to be backed up, previous addresses have been backed-up. For | |
3547 | * decreasing progress, it is the earliest address that has been | |
3548 | * backed up - later address are also backed up. | |
3549 | * So addresses between reshape_progress and backup_point are | |
3550 | * backed up providing those are in the 'correct' order. | |
3551 | * - 'wait_point' is an array address. When reshape_completed | |
3552 | * passes this point, progress_reshape should return. It might | |
3553 | * return earlier if it determines that ->reshape_progress needs | |
3554 | * to be updated or further backup is needed. | |
3555 | * - suspend_point is maintained by progress_reshape and the caller | |
3556 | * should not touch it except to initialise to zero. | |
3557 | * It is an array address and it only increases in 2.6.37 and earlier. | |
b6b95155 | 3558 | * This makes it difficult to handle reducing reshapes with |
7443ee81 N |
3559 | * external metadata. |
3560 | * However: it is similar to backup_point in that it records the | |
3561 | * other end of a suspended region from reshape_progress. | |
3562 | * it is moved to extend the region that is safe to backup and/or | |
3563 | * reshape | |
3564 | * - reshape_completed is read from sysfs and returned. The caller | |
3565 | * should copy this into ->reshape_progress when it has reason to | |
3566 | * believe that the metadata knows this, and any backup outside this | |
3567 | * has been erased. | |
3568 | * | |
3569 | * Return value is: | |
3570 | * 1 if more data from backup_point - but only as far as suspend_point, | |
3571 | * should be backed up | |
3572 | * 0 if things are progressing smoothly | |
9e034f70 N |
3573 | * -1 if the reshape is finished because it is all done, |
3574 | * -2 if the reshape is finished due to an error. | |
7443ee81 N |
3575 | */ |
3576 | ||
3577 | int advancing = (reshape->after.data_disks | |
3578 | >= reshape->before.data_disks); | |
38dad34a N |
3579 | unsigned long long need_backup; /* All data between start of array and |
3580 | * here will at some point need to | |
3581 | * be backed up. | |
d0ab945e | 3582 | */ |
7443ee81 | 3583 | unsigned long long read_offset, write_offset; |
b4f8e38b | 3584 | unsigned long long write_range; |
7443ee81 | 3585 | unsigned long long max_progress, target, completed; |
d0ab945e N |
3586 | unsigned long long array_size = (info->component_size |
3587 | * reshape->before.data_disks); | |
7443ee81 | 3588 | int fd; |
77a73a17 | 3589 | char buf[20]; |
7443ee81 N |
3590 | |
3591 | /* First, we unsuspend any region that is now known to be safe. | |
3592 | * If suspend_point is on the 'wrong' side of reshape_progress, then | |
3593 | * we don't have or need suspension at the moment. This is true for | |
3594 | * native metadata when we don't need to back-up. | |
3595 | */ | |
3596 | if (advancing) { | |
49cd738b | 3597 | if (info->reshape_progress <= *suspend_point) |
7443ee81 N |
3598 | sysfs_set_num(info, NULL, "suspend_lo", |
3599 | info->reshape_progress); | |
3600 | } else { | |
3601 | /* Note: this won't work in 2.6.37 and before. | |
3602 | * Something somewhere should make sure we don't need it! | |
3603 | */ | |
49cd738b | 3604 | if (info->reshape_progress >= *suspend_point) |
7443ee81 N |
3605 | sysfs_set_num(info, NULL, "suspend_hi", |
3606 | info->reshape_progress); | |
3607 | } | |
3608 | ||
3609 | /* Now work out how far it is safe to progress. | |
3610 | * If the read_offset for ->reshape_progress is less than | |
3611 | * 'blocks' beyond the write_offset, we can only progress as far | |
3612 | * as a backup. | |
3613 | * Otherwise we can progress until the write_offset for the new location | |
3614 | * reaches (within 'blocks' of) the read_offset at the current location. | |
3615 | * However that region must be suspended unless we are using native | |
3616 | * metadata. | |
3617 | * If we need to suspend more, we limit it to 128M per device, which is | |
3618 | * rather arbitrary and should be some time-based calculation. | |
3619 | */ | |
ec757320 N |
3620 | read_offset = info->reshape_progress / reshape->before.data_disks; |
3621 | write_offset = info->reshape_progress / reshape->after.data_disks; | |
b4f8e38b | 3622 | write_range = info->new_chunk/512; |
38dad34a N |
3623 | if (reshape->before.data_disks == reshape->after.data_disks) |
3624 | need_backup = array_size; | |
3625 | else | |
3626 | need_backup = reshape->backup_blocks; | |
7443ee81 | 3627 | if (advancing) { |
38dad34a | 3628 | if (read_offset < write_offset + write_range) |
7443ee81 | 3629 | max_progress = backup_point; |
ddee071d | 3630 | else |
7443ee81 | 3631 | max_progress = |
2f3dd5e4 N |
3632 | read_offset * |
3633 | reshape->after.data_disks; | |
7443ee81 | 3634 | } else { |
38dad34a N |
3635 | if (read_offset > write_offset - write_range) |
3636 | /* Can only progress as far as has been backed up, | |
3637 | * which must be suspended */ | |
7443ee81 | 3638 | max_progress = backup_point; |
38dad34a N |
3639 | else if (info->reshape_progress <= need_backup) |
3640 | max_progress = backup_point; | |
3641 | else { | |
3642 | if (info->array.major_version >= 0) | |
3643 | /* Can progress until backup is needed */ | |
3644 | max_progress = need_backup; | |
3645 | else { | |
3646 | /* Can progress until metadata update is required */ | |
3647 | max_progress = | |
3648 | read_offset * | |
3649 | reshape->after.data_disks; | |
3650 | /* but data must be suspended */ | |
3651 | if (max_progress < *suspend_point) | |
3652 | max_progress = *suspend_point; | |
3653 | } | |
7443ee81 N |
3654 | } |
3655 | } | |
3656 | ||
3657 | /* We know it is safe to progress to 'max_progress' providing | |
3658 | * it is suspended or we are using native metadata. | |
3659 | * Consider extending suspend_point 128M per device if it | |
3660 | * is less than 64M per device beyond reshape_progress. | |
3661 | * But always do a multiple of 'blocks' | |
832b2b9a N |
3662 | * FIXME this is too big - it takes to long to complete |
3663 | * this much. | |
7443ee81 N |
3664 | */ |
3665 | target = 64*1024*2 * min(reshape->before.data_disks, | |
24daa16f | 3666 | reshape->after.data_disks); |
b4f8e38b | 3667 | target /= reshape->backup_blocks; |
7443ee81 N |
3668 | if (target < 2) |
3669 | target = 2; | |
b4f8e38b | 3670 | target *= reshape->backup_blocks; |
7443ee81 N |
3671 | |
3672 | /* For externally managed metadata we always need to suspend IO to | |
3673 | * the area being reshaped so we regularly push suspend_point forward. | |
3674 | * For native metadata we only need the suspend if we are going to do | |
3675 | * a backup. | |
3676 | */ | |
3677 | if (advancing) { | |
d0ab945e N |
3678 | if ((need_backup > info->reshape_progress |
3679 | || info->array.major_version < 0) && | |
7443ee81 | 3680 | *suspend_point < info->reshape_progress + target) { |
d0ab945e N |
3681 | if (need_backup < *suspend_point + 2 * target) |
3682 | *suspend_point = need_backup; | |
3683 | else if (*suspend_point + 2 * target < array_size) | |
7443ee81 | 3684 | *suspend_point += 2 * target; |
d0ab945e N |
3685 | else |
3686 | *suspend_point = array_size; | |
7443ee81 | 3687 | sysfs_set_num(info, NULL, "suspend_hi", *suspend_point); |
d0ab945e N |
3688 | if (max_progress > *suspend_point) |
3689 | max_progress = *suspend_point; | |
7443ee81 N |
3690 | } |
3691 | } else { | |
38dad34a N |
3692 | if (info->array.major_version >= 0) { |
3693 | /* Only need to suspend when about to backup */ | |
3694 | if (info->reshape_progress < need_backup * 2 && | |
3695 | *suspend_point > 0) { | |
d0ab945e | 3696 | *suspend_point = 0; |
38dad34a N |
3697 | sysfs_set_num(info, NULL, "suspend_lo", 0); |
3698 | sysfs_set_num(info, NULL, "suspend_hi", need_backup); | |
3699 | } | |
3700 | } else { | |
3701 | /* Need to suspend continually */ | |
3702 | if (info->reshape_progress < *suspend_point) | |
3703 | *suspend_point = info->reshape_progress; | |
3704 | if (*suspend_point + target < info->reshape_progress) | |
3705 | /* No need to move suspend region yet */; | |
3706 | else { | |
3707 | if (*suspend_point >= 2 * target) | |
3708 | *suspend_point -= 2 * target; | |
3709 | else | |
3710 | *suspend_point = 0; | |
3711 | sysfs_set_num(info, NULL, "suspend_lo", | |
3712 | *suspend_point); | |
3713 | } | |
d0ab945e N |
3714 | if (max_progress < *suspend_point) |
3715 | max_progress = *suspend_point; | |
7443ee81 N |
3716 | } |
3717 | } | |
3718 | ||
3719 | /* now set sync_max to allow that progress. sync_max, like | |
3720 | * sync_completed is a count of sectors written per device, so | |
3721 | * we find the difference between max_progress and the start point, | |
3722 | * and divide that by after.data_disks to get a sync_max | |
3723 | * number. | |
3724 | * At the same time we convert wait_point to a similar number | |
3725 | * for comparing against sync_completed. | |
3726 | */ | |
92d1991f | 3727 | /* scale down max_progress to per_disk */ |
7443ee81 | 3728 | max_progress /= reshape->after.data_disks; |
92d1991f N |
3729 | /* Round to chunk size as some kernels give an erroneously high number */ |
3730 | max_progress /= info->new_chunk/512; | |
3731 | max_progress *= info->new_chunk/512; | |
7f913e9b N |
3732 | /* And round to old chunk size as the kernel wants that */ |
3733 | max_progress /= info->array.chunk_size/512; | |
3734 | max_progress *= info->array.chunk_size/512; | |
92d1991f N |
3735 | /* Limit progress to the whole device */ |
3736 | if (max_progress > info->component_size) | |
3737 | max_progress = info->component_size; | |
7443ee81 | 3738 | wait_point /= reshape->after.data_disks; |
92d1991f N |
3739 | if (!advancing) { |
3740 | /* switch from 'device offset' to 'processed block count' */ | |
3741 | max_progress = info->component_size - max_progress; | |
3742 | wait_point = info->component_size - wait_point; | |
3743 | } | |
7443ee81 N |
3744 | |
3745 | sysfs_set_num(info, NULL, "sync_max", max_progress); | |
3746 | ||
3747 | /* Now wait. If we have already reached the point that we were | |
3748 | * asked to wait to, don't wait at all, else wait for any change. | |
3749 | * We need to select on 'sync_completed' as that is the place that | |
3750 | * notifications happen, but we are really interested in | |
3751 | * 'reshape_position' | |
3752 | */ | |
3753 | fd = sysfs_get_fd(info, NULL, "sync_completed"); | |
3754 | if (fd < 0) | |
77a73a17 | 3755 | goto check_progress; |
7443ee81 | 3756 | |
621ea11b | 3757 | if (sysfs_fd_get_ll(fd, &completed) < 0) |
77a73a17 | 3758 | goto check_progress; |
621ea11b | 3759 | |
7443ee81 N |
3760 | while (completed < max_progress && completed < wait_point) { |
3761 | /* Check that sync_action is still 'reshape' to avoid | |
3762 | * waiting forever on a dead array | |
3763 | */ | |
3764 | char action[20]; | |
7443ee81 N |
3765 | if (sysfs_get_str(info, NULL, "sync_action", |
3766 | action, 20) <= 0 || | |
3767 | strncmp(action, "reshape", 7) != 0) | |
3768 | break; | |
26d6e157 AK |
3769 | /* Some kernels reset 'sync_completed' to zero |
3770 | * before setting 'sync_action' to 'idle'. | |
3771 | * So we need these extra tests. | |
3772 | */ | |
3773 | if (completed == 0 && advancing | |
3774 | && info->reshape_progress > 0) | |
3775 | break; | |
3776 | if (completed == 0 && !advancing | |
3777 | && info->reshape_progress < (info->component_size | |
3778 | * reshape->after.data_disks)) | |
3779 | break; | |
efc67e8e | 3780 | sysfs_wait(fd, NULL); |
621ea11b | 3781 | if (sysfs_fd_get_ll(fd, &completed) < 0) |
77a73a17 | 3782 | goto check_progress; |
7443ee81 | 3783 | } |
10d0d365 AK |
3784 | /* Some kernels reset 'sync_completed' to zero, |
3785 | * we need to have real point we are in md | |
3786 | */ | |
3787 | if (completed == 0) | |
3788 | completed = max_progress; | |
3789 | ||
92d1991f N |
3790 | /* some kernels can give an incorrectly high 'completed' number */ |
3791 | completed /= (info->new_chunk/512); | |
3792 | completed *= (info->new_chunk/512); | |
7443ee81 N |
3793 | /* Convert 'completed' back in to a 'progress' number */ |
3794 | completed *= reshape->after.data_disks; | |
3795 | if (!advancing) { | |
3796 | completed = info->component_size * reshape->after.data_disks | |
3797 | - completed; | |
3798 | } | |
3799 | *reshape_completed = completed; | |
24daa16f | 3800 | |
7443ee81 N |
3801 | close(fd); |
3802 | ||
3803 | /* We return the need_backup flag. Caller will decide | |
d0ab945e | 3804 | * how much - a multiple of ->backup_blocks up to *suspend_point |
7443ee81 | 3805 | */ |
38dad34a N |
3806 | if (advancing) |
3807 | return need_backup > info->reshape_progress; | |
3808 | else | |
3809 | return need_backup >= info->reshape_progress; | |
77a73a17 N |
3810 | |
3811 | check_progress: | |
3812 | /* if we couldn't read a number from sync_completed, then | |
3813 | * either the reshape did complete, or it aborted. | |
3814 | * We can tell which by checking for 'none' in reshape_position. | |
621ea11b N |
3815 | * If it did abort, then it might immediately restart if it |
3816 | * it was just a device failure that leaves us degraded but | |
3817 | * functioning. | |
77a73a17 N |
3818 | */ |
3819 | strcpy(buf, "hi"); | |
3820 | if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0 | |
621ea11b N |
3821 | || strncmp(buf, "none", 4) != 0) { |
3822 | /* The abort might only be temporary. Wait up to 10 | |
3823 | * seconds for fd to contain a valid number again. | |
3824 | */ | |
efc67e8e | 3825 | int wait = 10000; |
621ea11b | 3826 | int rv = -2; |
efc67e8e N |
3827 | while (fd >= 0 && rv < 0 && wait > 0) { |
3828 | if (sysfs_wait(fd, &wait) != 1) | |
621ea11b | 3829 | break; |
6560987b N |
3830 | switch (sysfs_fd_get_ll(fd, &completed)) { |
3831 | case 0: | |
621ea11b N |
3832 | /* all good again */ |
3833 | rv = 1; | |
6560987b N |
3834 | break; |
3835 | case -2: /* read error - abort */ | |
efc67e8e | 3836 | wait = 0; |
6560987b N |
3837 | break; |
3838 | } | |
621ea11b N |
3839 | } |
3840 | if (fd >= 0) | |
3841 | close(fd); | |
3842 | return rv; /* abort */ | |
3843 | } else { | |
6d5316f6 | 3844 | /* Maybe racing with array shutdown - check state */ |
621ea11b N |
3845 | if (fd >= 0) |
3846 | close(fd); | |
6d5316f6 N |
3847 | if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0 |
3848 | || strncmp(buf, "inactive", 8) == 0 | |
3849 | || strncmp(buf, "clear",5) == 0) | |
3850 | return -2; /* abort */ | |
77a73a17 | 3851 | return -1; /* complete */ |
6d5316f6 | 3852 | } |
7443ee81 N |
3853 | } |
3854 | ||
fcf57625 | 3855 | /* FIXME return status is never checked */ |
4411fb17 | 3856 | static int grow_backup(struct mdinfo *sra, |
7236ee7a | 3857 | unsigned long long offset, /* per device */ |
7443ee81 | 3858 | unsigned long stripes, /* per device, in old chunks */ |
7236ee7a N |
3859 | int *sources, unsigned long long *offsets, |
3860 | int disks, int chunk, int level, int layout, | |
3861 | int dests, int *destfd, unsigned long long *destoffsets, | |
d4445387 | 3862 | int part, int *degraded, |
7236ee7a N |
3863 | char *buf) |
3864 | { | |
3865 | /* Backup 'blocks' sectors at 'offset' on each device of the array, | |
3866 | * to storage 'destfd' (offset 'destoffsets'), after first | |
3867 | * suspending IO. Then allow resync to continue | |
3868 | * over the suspended section. | |
3869 | * Use part 'part' of the backup-super-block. | |
3870 | */ | |
3871 | int odata = disks; | |
3872 | int rv = 0; | |
3873 | int i; | |
f21e18ca N |
3874 | unsigned long long ll; |
3875 | int new_degraded; | |
7236ee7a N |
3876 | //printf("offset %llu\n", offset); |
3877 | if (level >= 4) | |
3878 | odata--; | |
3879 | if (level == 6) | |
3880 | odata--; | |
7443ee81 | 3881 | |
d4445387 | 3882 | /* Check that array hasn't become degraded, else we might backup the wrong data */ |
2c6ac128 N |
3883 | if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0) |
3884 | return -1; /* FIXME this error is ignored */ | |
f21e18ca | 3885 | new_degraded = (int)ll; |
d4445387 N |
3886 | if (new_degraded != *degraded) { |
3887 | /* check each device to ensure it is still working */ | |
3888 | struct mdinfo *sd; | |
3889 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
3890 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
3891 | continue; | |
3892 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
3893 | char sbuf[20]; | |
3894 | if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 || | |
3895 | strstr(sbuf, "faulty") || | |
3896 | strstr(sbuf, "in_sync") == NULL) { | |
3897 | /* this device is dead */ | |
3898 | sd->disk.state = (1<<MD_DISK_FAULTY); | |
3899 | if (sd->disk.raid_disk >= 0 && | |
3900 | sources[sd->disk.raid_disk] >= 0) { | |
3901 | close(sources[sd->disk.raid_disk]); | |
3902 | sources[sd->disk.raid_disk] = -1; | |
3903 | } | |
3904 | } | |
3905 | } | |
3906 | } | |
3907 | *degraded = new_degraded; | |
3908 | } | |
7236ee7a N |
3909 | if (part) { |
3910 | bsb.arraystart2 = __cpu_to_le64(offset * odata); | |
200871ad | 3911 | bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
3912 | } else { |
3913 | bsb.arraystart = __cpu_to_le64(offset * odata); | |
200871ad | 3914 | bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
3915 | } |
3916 | if (part) | |
3917 | bsb.magic[15] = '2'; | |
3918 | for (i = 0; i < dests; i++) | |
3919 | if (part) | |
3920 | lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0); | |
3921 | else | |
3922 | lseek64(destfd[i], destoffsets[i], 0); | |
3923 | ||
24daa16f | 3924 | rv = save_stripes(sources, offsets, |
7236ee7a N |
3925 | disks, chunk, level, layout, |
3926 | dests, destfd, | |
3927 | offset*512*odata, stripes * chunk * odata, | |
3928 | buf); | |
3929 | ||
3930 | if (rv) | |
3931 | return rv; | |
97396422 | 3932 | bsb.mtime = __cpu_to_le64(time(0)); |
7236ee7a N |
3933 | for (i = 0; i < dests; i++) { |
3934 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
3935 | ||
3936 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
3937 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
3938 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
3939 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
3940 | ||
72044953 | 3941 | rv = -1; |
f21e18ca N |
3942 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0) |
3943 | != destoffsets[i] - 4096) | |
72044953 N |
3944 | break; |
3945 | if (write(destfd[i], &bsb, 512) != 512) | |
3946 | break; | |
ff94fb86 | 3947 | if (destoffsets[i] > 4096) { |
f21e18ca | 3948 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) != |
a847575a | 3949 | destoffsets[i]+stripes*chunk*odata) |
72044953 N |
3950 | break; |
3951 | if (write(destfd[i], &bsb, 512) != 512) | |
3952 | break; | |
ff94fb86 | 3953 | } |
7236ee7a | 3954 | fsync(destfd[i]); |
72044953 | 3955 | rv = 0; |
7236ee7a | 3956 | } |
2efedc7b | 3957 | |
fcf57625 | 3958 | return rv; |
7236ee7a N |
3959 | } |
3960 | ||
3961 | /* in 2.6.30, the value reported by sync_completed can be | |
3962 | * less that it should be by one stripe. | |
3963 | * This only happens when reshape hits sync_max and pauses. | |
3964 | * So allow wait_backup to either extent sync_max further | |
3965 | * than strictly necessary, or return before the | |
3966 | * sync has got quite as far as we would really like. | |
3967 | * This is what 'blocks2' is for. | |
3968 | * The various caller give appropriate values so that | |
3969 | * every works. | |
3970 | */ | |
fcf57625 | 3971 | /* FIXME return value is often ignored */ |
24daa16f N |
3972 | static int forget_backup(int dests, int *destfd, |
3973 | unsigned long long *destoffsets, | |
3974 | int part) | |
7236ee7a | 3975 | { |
24daa16f | 3976 | /* |
7443ee81 | 3977 | * Erase backup 'part' (which is 0 or 1) |
7236ee7a | 3978 | */ |
7236ee7a | 3979 | int i; |
fcf57625 | 3980 | int rv; |
7236ee7a | 3981 | |
7236ee7a N |
3982 | if (part) { |
3983 | bsb.arraystart2 = __cpu_to_le64(0); | |
3984 | bsb.length2 = __cpu_to_le64(0); | |
3985 | } else { | |
3986 | bsb.arraystart = __cpu_to_le64(0); | |
3987 | bsb.length = __cpu_to_le64(0); | |
3988 | } | |
97396422 | 3989 | bsb.mtime = __cpu_to_le64(time(0)); |
fcf57625 | 3990 | rv = 0; |
7236ee7a N |
3991 | for (i = 0; i < dests; i++) { |
3992 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
3993 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
3994 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
3995 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
3996 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
f21e18ca | 3997 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) != |
a847575a | 3998 | destoffsets[i]-4096) |
72044953 | 3999 | rv = -1; |
24daa16f | 4000 | if (rv == 0 && |
72044953 N |
4001 | write(destfd[i], &bsb, 512) != 512) |
4002 | rv = -1; | |
7236ee7a N |
4003 | fsync(destfd[i]); |
4004 | } | |
fcf57625 | 4005 | return rv; |
7236ee7a | 4006 | } |
e86c9dd6 | 4007 | |
7236ee7a N |
4008 | static void fail(char *msg) |
4009 | { | |
fcf57625 | 4010 | int rv; |
72044953 N |
4011 | rv = (write(2, msg, strlen(msg)) != (int)strlen(msg)); |
4012 | rv |= (write(2, "\n", 1) != 1); | |
fcf57625 | 4013 | exit(rv ? 1 : 2); |
7236ee7a N |
4014 | } |
4015 | ||
4016 | static char *abuf, *bbuf; | |
f21e18ca | 4017 | static unsigned long long abuflen; |
7236ee7a N |
4018 | static void validate(int afd, int bfd, unsigned long long offset) |
4019 | { | |
4020 | /* check that the data in the backup against the array. | |
4021 | * This is only used for regression testing and should not | |
4022 | * be used while the array is active | |
4023 | */ | |
7236ee7a N |
4024 | if (afd < 0) |
4025 | return; | |
4026 | lseek64(bfd, offset - 4096, 0); | |
4027 | if (read(bfd, &bsb2, 512) != 512) | |
4028 | fail("cannot read bsb"); | |
4029 | if (bsb2.sb_csum != bsb_csum((char*)&bsb2, | |
4030 | ((char*)&bsb2.sb_csum)-((char*)&bsb2))) | |
4031 | fail("first csum bad"); | |
4032 | if (memcmp(bsb2.magic, "md_backup_data", 14) != 0) | |
4033 | fail("magic is bad"); | |
4034 | if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 && | |
4035 | bsb2.sb_csum2 != bsb_csum((char*)&bsb2, | |
24daa16f | 4036 | ((char*)&bsb2.sb_csum2)-((char*)&bsb2))) |
7236ee7a N |
4037 | fail("second csum bad"); |
4038 | ||
4039 | if (__le64_to_cpu(bsb2.devstart)*512 != offset) | |
4040 | fail("devstart is wrong"); | |
4041 | ||
4042 | if (bsb2.length) { | |
4043 | unsigned long long len = __le64_to_cpu(bsb2.length)*512; | |
4044 | ||
4045 | if (abuflen < len) { | |
4046 | free(abuf); | |
4047 | free(bbuf); | |
4048 | abuflen = len; | |
fcf57625 N |
4049 | if (posix_memalign((void**)&abuf, 4096, abuflen) || |
4050 | posix_memalign((void**)&bbuf, 4096, abuflen)) { | |
4051 | abuflen = 0; | |
4052 | /* just stop validating on mem-alloc failure */ | |
4053 | return; | |
4054 | } | |
e86c9dd6 | 4055 | } |
48924014 | 4056 | |
7236ee7a | 4057 | lseek64(bfd, offset, 0); |
f21e18ca | 4058 | if ((unsigned long long)read(bfd, bbuf, len) != len) { |
080fd005 | 4059 | //printf("len %llu\n", len); |
7236ee7a N |
4060 | fail("read first backup failed"); |
4061 | } | |
4062 | lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0); | |
f21e18ca | 4063 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
4064 | fail("read first from array failed"); |
4065 | if (memcmp(bbuf, abuf, len) != 0) { | |
24daa16f | 4066 | #if 0 |
7236ee7a N |
4067 | int i; |
4068 | printf("offset=%llu len=%llu\n", | |
080fd005 | 4069 | (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len); |
7236ee7a N |
4070 | for (i=0; i<len; i++) |
4071 | if (bbuf[i] != abuf[i]) { | |
4072 | printf("first diff byte %d\n", i); | |
93ecfa01 | 4073 | break; |
7236ee7a | 4074 | } |
24daa16f | 4075 | #endif |
7236ee7a | 4076 | fail("data1 compare failed"); |
e86c9dd6 | 4077 | } |
7236ee7a N |
4078 | } |
4079 | if (bsb2.length2) { | |
4080 | unsigned long long len = __le64_to_cpu(bsb2.length2)*512; | |
4081 | ||
4082 | if (abuflen < len) { | |
4083 | free(abuf); | |
4084 | free(bbuf); | |
4085 | abuflen = len; | |
503975b9 N |
4086 | abuf = xmalloc(abuflen); |
4087 | bbuf = xmalloc(abuflen); | |
e86c9dd6 NB |
4088 | } |
4089 | ||
7236ee7a | 4090 | lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0); |
f21e18ca | 4091 | if ((unsigned long long)read(bfd, bbuf, len) != len) |
7236ee7a N |
4092 | fail("read second backup failed"); |
4093 | lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0); | |
f21e18ca | 4094 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
4095 | fail("read second from array failed"); |
4096 | if (memcmp(bbuf, abuf, len) != 0) | |
4097 | fail("data2 compare failed"); | |
e86c9dd6 | 4098 | } |
7236ee7a | 4099 | } |
e86c9dd6 | 4100 | |
999b4972 N |
4101 | int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape, |
4102 | struct supertype *st, unsigned long blocks, | |
4103 | int *fds, unsigned long long *offsets, | |
4104 | int dests, int *destfd, unsigned long long *destoffsets) | |
7236ee7a | 4105 | { |
7443ee81 N |
4106 | /* Monitor a reshape where backup is being performed using |
4107 | * 'native' mechanism - either to a backup file, or | |
4108 | * to some space in a spare. | |
4109 | */ | |
7236ee7a | 4110 | char *buf; |
7443ee81 N |
4111 | int degraded = -1; |
4112 | unsigned long long speed; | |
4113 | unsigned long long suspend_point, array_size; | |
4114 | unsigned long long backup_point, wait_point; | |
4115 | unsigned long long reshape_completed; | |
4116 | int done = 0; | |
4117 | int increasing = reshape->after.data_disks >= reshape->before.data_disks; | |
4118 | int part = 0; /* The next part of the backup area to fill. It may already | |
4119 | * be full, so we need to check */ | |
4120 | int level = reshape->level; | |
4121 | int layout = reshape->before.layout; | |
4122 | int data = reshape->before.data_disks; | |
4123 | int disks = reshape->before.data_disks + reshape->parity; | |
4124 | int chunk = sra->array.chunk_size; | |
da8100ca N |
4125 | struct mdinfo *sd; |
4126 | unsigned long stripes; | |
3b7e9d0c | 4127 | int uuid[4]; |
da8100ca N |
4128 | |
4129 | /* set up the backup-super-block. This requires the | |
4130 | * uuid from the array. | |
4131 | */ | |
4132 | /* Find a superblock */ | |
4133 | for (sd = sra->devs; sd; sd = sd->next) { | |
4134 | char *dn; | |
4135 | int devfd; | |
4136 | int ok; | |
4137 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
4138 | continue; | |
4139 | dn = map_dev(sd->disk.major, sd->disk.minor, 1); | |
4140 | devfd = dev_open(dn, O_RDONLY); | |
4141 | if (devfd < 0) | |
4142 | continue; | |
4143 | ok = st->ss->load_super(st, devfd, NULL); | |
4144 | close(devfd); | |
77f8d358 | 4145 | if (ok == 0) |
da8100ca N |
4146 | break; |
4147 | } | |
4148 | if (!sd) { | |
e7b84f9d | 4149 | pr_err("Cannot find a superblock\n"); |
da8100ca N |
4150 | return 0; |
4151 | } | |
4152 | ||
4153 | memset(&bsb, 0, 512); | |
4154 | memcpy(bsb.magic, "md_backup_data-1", 16); | |
3b7e9d0c LB |
4155 | st->ss->uuid_from_super(st, uuid); |
4156 | memcpy(bsb.set_uuid, uuid, 16); | |
da8100ca N |
4157 | bsb.mtime = __cpu_to_le64(time(0)); |
4158 | bsb.devstart2 = blocks; | |
4159 | ||
4160 | stripes = blocks / (sra->array.chunk_size/512) / | |
4161 | reshape->before.data_disks; | |
e86c9dd6 | 4162 | |
fcf57625 N |
4163 | if (posix_memalign((void**)&buf, 4096, disks * chunk)) |
4164 | /* Don't start the 'reshape' */ | |
4165 | return 0; | |
7443ee81 N |
4166 | if (reshape->before.data_disks == reshape->after.data_disks) { |
4167 | sysfs_get_ll(sra, NULL, "sync_speed_min", &speed); | |
4168 | sysfs_set_num(sra, NULL, "sync_speed_min", 200000); | |
4169 | } | |
e86c9dd6 | 4170 | |
7443ee81 | 4171 | if (increasing) { |
96533072 | 4172 | array_size = sra->component_size * reshape->after.data_disks; |
7443ee81 N |
4173 | backup_point = sra->reshape_progress; |
4174 | suspend_point = 0; | |
4175 | } else { | |
96533072 | 4176 | array_size = sra->component_size * reshape->before.data_disks; |
25da62d9 | 4177 | backup_point = reshape->backup_blocks; |
7443ee81 N |
4178 | suspend_point = array_size; |
4179 | } | |
7236ee7a | 4180 | |
7443ee81 N |
4181 | while (!done) { |
4182 | int rv; | |
7236ee7a | 4183 | |
7443ee81 N |
4184 | /* Want to return as soon the oldest backup slot can |
4185 | * be released as that allows us to start backing up | |
4186 | * some more, providing suspend_point has been | |
b6b95155 | 4187 | * advanced, which it should have. |
7443ee81 N |
4188 | */ |
4189 | if (increasing) { | |
4190 | wait_point = array_size; | |
4191 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
4192 | wait_point = (__le64_to_cpu(bsb.arraystart) + | |
4193 | __le64_to_cpu(bsb.length)); | |
4194 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
4195 | wait_point = (__le64_to_cpu(bsb.arraystart2) + | |
4196 | __le64_to_cpu(bsb.length2)); | |
4197 | } else { | |
4198 | wait_point = 0; | |
4199 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
4200 | wait_point = __le64_to_cpu(bsb.arraystart); | |
4201 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
4202 | wait_point = __le64_to_cpu(bsb.arraystart2); | |
4203 | } | |
4204 | ||
4205 | rv = progress_reshape(sra, reshape, | |
4206 | backup_point, wait_point, | |
4207 | &suspend_point, &reshape_completed); | |
7443ee81 N |
4208 | /* external metadata would need to ping_monitor here */ |
4209 | sra->reshape_progress = reshape_completed; | |
7236ee7a | 4210 | |
7443ee81 N |
4211 | /* Clear any backup region that is before 'here' */ |
4212 | if (increasing) { | |
b3cf095a N |
4213 | if (__le64_to_cpu(bsb.length) > 0 && |
4214 | reshape_completed >= (__le64_to_cpu(bsb.arraystart) + | |
7443ee81 N |
4215 | __le64_to_cpu(bsb.length))) |
4216 | forget_backup(dests, destfd, | |
4217 | destoffsets, 0); | |
b3cf095a N |
4218 | if (__le64_to_cpu(bsb.length2) > 0 && |
4219 | reshape_completed >= (__le64_to_cpu(bsb.arraystart2) + | |
7443ee81 N |
4220 | __le64_to_cpu(bsb.length2))) |
4221 | forget_backup(dests, destfd, | |
4222 | destoffsets, 1); | |
4223 | } else { | |
b3cf095a N |
4224 | if (__le64_to_cpu(bsb.length) > 0 && |
4225 | reshape_completed <= (__le64_to_cpu(bsb.arraystart))) | |
7443ee81 N |
4226 | forget_backup(dests, destfd, |
4227 | destoffsets, 0); | |
b3cf095a N |
4228 | if (__le64_to_cpu(bsb.length2) > 0 && |
4229 | reshape_completed <= (__le64_to_cpu(bsb.arraystart2))) | |
7443ee81 N |
4230 | forget_backup(dests, destfd, |
4231 | destoffsets, 1); | |
4232 | } | |
7236ee7a | 4233 | |
223c5c99 | 4234 | if (rv < 0) { |
77a73a17 N |
4235 | if (rv == -1) |
4236 | done = 1; | |
223c5c99 N |
4237 | break; |
4238 | } | |
2d4de5f9 N |
4239 | if (rv == 0 && increasing && !st->ss->external) { |
4240 | /* No longer need to monitor this reshape */ | |
4241 | done = 1; | |
4242 | break; | |
4243 | } | |
223c5c99 | 4244 | |
60204e4e | 4245 | while (rv) { |
7443ee81 | 4246 | unsigned long long offset; |
e97ca002 | 4247 | unsigned long actual_stripes; |
60204e4e N |
4248 | /* Need to backup some data. |
4249 | * If 'part' is not used and the desired | |
4250 | * backup size is suspended, do a backup, | |
4251 | * then consider the next part. | |
4252 | */ | |
7443ee81 N |
4253 | /* Check that 'part' is unused */ |
4254 | if (part == 0 && __le64_to_cpu(bsb.length) != 0) | |
60204e4e | 4255 | break; |
7443ee81 | 4256 | if (part == 1 && __le64_to_cpu(bsb.length2) != 0) |
60204e4e | 4257 | break; |
7443ee81 N |
4258 | |
4259 | offset = backup_point / data; | |
e97ca002 | 4260 | actual_stripes = stripes; |
60204e4e | 4261 | if (increasing) { |
e97ca002 N |
4262 | if (offset + actual_stripes * (chunk/512) > |
4263 | sra->component_size) | |
4264 | actual_stripes = ((sra->component_size - offset) | |
4265 | / (chunk/512)); | |
4266 | if (offset + actual_stripes * (chunk/512) > | |
60204e4e N |
4267 | suspend_point/data) |
4268 | break; | |
4269 | } else { | |
e97ca002 N |
4270 | if (offset < actual_stripes * (chunk/512)) |
4271 | actual_stripes = offset / (chunk/512); | |
4272 | offset -= actual_stripes * (chunk/512); | |
4273 | if (offset < suspend_point/data) | |
60204e4e N |
4274 | break; |
4275 | } | |
e4b11073 N |
4276 | if (actual_stripes == 0) |
4277 | break; | |
e97ca002 | 4278 | grow_backup(sra, offset, actual_stripes, |
7443ee81 N |
4279 | fds, offsets, |
4280 | disks, chunk, level, layout, | |
4281 | dests, destfd, destoffsets, | |
4282 | part, °raded, buf); | |
4283 | validate(afd, destfd[0], destoffsets[0]); | |
4284 | /* record where 'part' is up to */ | |
4285 | part = !part; | |
4286 | if (increasing) | |
e97ca002 | 4287 | backup_point += actual_stripes * (chunk/512) * data; |
7443ee81 | 4288 | else |
e97ca002 | 4289 | backup_point -= actual_stripes * (chunk/512) * data; |
7443ee81 N |
4290 | } |
4291 | } | |
4292 | ||
223c5c99 N |
4293 | /* FIXME maybe call progress_reshape one more time instead */ |
4294 | abort_reshape(sra); /* remove any remaining suspension */ | |
7443ee81 N |
4295 | if (reshape->before.data_disks == reshape->after.data_disks) |
4296 | sysfs_set_num(sra, NULL, "sync_speed_min", speed); | |
7236ee7a | 4297 | free(buf); |
77a73a17 | 4298 | return done; |
e86c9dd6 | 4299 | } |
353632d9 NB |
4300 | |
4301 | /* | |
4302 | * If any spare contains md_back_data-1 which is recent wrt mtime, | |
4303 | * write that data into the array and update the super blocks with | |
4304 | * the new reshape_progress | |
4305 | */ | |
ea0ebe96 N |
4306 | int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt, |
4307 | char *backup_file, int verbose) | |
353632d9 NB |
4308 | { |
4309 | int i, j; | |
4310 | int old_disks; | |
353632d9 | 4311 | unsigned long long *offsets; |
82f2d6ab | 4312 | unsigned long long nstripe, ostripe; |
6e9eac4f | 4313 | int ndata, odata; |
353632d9 | 4314 | |
e9e43ec3 N |
4315 | odata = info->array.raid_disks - info->delta_disks - 1; |
4316 | if (info->array.level == 6) odata--; /* number of data disks */ | |
4317 | ndata = info->array.raid_disks - 1; | |
4318 | if (info->new_level == 6) ndata--; | |
353632d9 NB |
4319 | |
4320 | old_disks = info->array.raid_disks - info->delta_disks; | |
4321 | ||
e9e43ec3 N |
4322 | if (info->delta_disks <= 0) |
4323 | /* Didn't grow, so the backup file must have | |
4324 | * been used | |
4325 | */ | |
4326 | old_disks = cnt; | |
06b0d786 | 4327 | for (i=old_disks-(backup_file?1:0); i<cnt; i++) { |
353632d9 | 4328 | struct mdinfo dinfo; |
06b0d786 | 4329 | int fd; |
e9e43ec3 | 4330 | int bsbsize; |
ea0ebe96 | 4331 | char *devname, namebuf[20]; |
10af14c4 | 4332 | unsigned long long lo, hi; |
353632d9 NB |
4333 | |
4334 | /* This was a spare and may have some saved data on it. | |
4335 | * Load the superblock, find and load the | |
4336 | * backup_super_block. | |
4337 | * If either fail, go on to next device. | |
4338 | * If the backup contains no new info, just return | |
206c5eae | 4339 | * else restore data and update all superblocks |
353632d9 | 4340 | */ |
06b0d786 NB |
4341 | if (i == old_disks-1) { |
4342 | fd = open(backup_file, O_RDONLY); | |
e9e43ec3 | 4343 | if (fd<0) { |
e7b84f9d | 4344 | pr_err("backup file %s inaccessible: %s\n", |
e9e43ec3 | 4345 | backup_file, strerror(errno)); |
06b0d786 | 4346 | continue; |
e9e43ec3 | 4347 | } |
ea0ebe96 | 4348 | devname = backup_file; |
06b0d786 NB |
4349 | } else { |
4350 | fd = fdlist[i]; | |
4351 | if (fd < 0) | |
4352 | continue; | |
3da92f27 | 4353 | if (st->ss->load_super(st, fd, NULL)) |
06b0d786 | 4354 | continue; |
353632d9 | 4355 | |
a5d85af7 | 4356 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 NB |
4357 | st->ss->free_super(st); |
4358 | ||
06b0d786 NB |
4359 | if (lseek64(fd, |
4360 | (dinfo.data_offset + dinfo.component_size - 8) <<9, | |
ea0ebe96 | 4361 | 0) < 0) { |
e7b84f9d | 4362 | pr_err("Cannot seek on device %d\n", i); |
06b0d786 | 4363 | continue; /* Cannot seek */ |
ea0ebe96 N |
4364 | } |
4365 | sprintf(namebuf, "device-%d", i); | |
4366 | devname = namebuf; | |
06b0d786 | 4367 | } |
ea0ebe96 N |
4368 | if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) { |
4369 | if (verbose) | |
e7b84f9d | 4370 | pr_err("Cannot read from %s\n", devname); |
353632d9 | 4371 | continue; /* Cannot read */ |
ea0ebe96 | 4372 | } |
e9e43ec3 | 4373 | if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 && |
ea0ebe96 N |
4374 | memcmp(bsb.magic, "md_backup_data-2", 16) != 0) { |
4375 | if (verbose) | |
e7b84f9d | 4376 | pr_err("No backup metadata on %s\n", devname); |
353632d9 | 4377 | continue; |
ea0ebe96 N |
4378 | } |
4379 | if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) { | |
4380 | if (verbose) | |
e7b84f9d | 4381 | pr_err("Bad backup-metadata checksum on %s\n", devname); |
353632d9 | 4382 | continue; /* bad checksum */ |
ea0ebe96 | 4383 | } |
e9e43ec3 | 4384 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 && |
ea0ebe96 N |
4385 | bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) { |
4386 | if (verbose) | |
e7b84f9d | 4387 | pr_err("Bad backup-metadata checksum2 on %s\n", devname); |
22e30516 | 4388 | continue; /* Bad second checksum */ |
ea0ebe96 N |
4389 | } |
4390 | if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) { | |
4391 | if (verbose) | |
e7b84f9d | 4392 | pr_err("Wrong uuid on backup-metadata on %s\n", devname); |
353632d9 | 4393 | continue; /* Wrong uuid */ |
ea0ebe96 | 4394 | } |
353632d9 | 4395 | |
097075b6 N |
4396 | /* array utime and backup-mtime should be updated at much the same time, but it seems that |
4397 | * sometimes they aren't... So allow considerable flexability in matching, and allow | |
4398 | * this test to be overridden by an environment variable. | |
4399 | */ | |
f21e18ca N |
4400 | if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 || |
4401 | info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) { | |
097075b6 | 4402 | if (check_env("MDADM_GROW_ALLOW_OLD")) { |
e7b84f9d | 4403 | pr_err("accepting backup with timestamp %lu " |
097075b6 N |
4404 | "for array with timestamp %lu\n", |
4405 | (unsigned long)__le64_to_cpu(bsb.mtime), | |
4406 | (unsigned long)info->array.utime); | |
4407 | } else { | |
676ab312 N |
4408 | pr_err("too-old timestamp on backup-metadata on %s\n", devname); |
4409 | pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n"); | |
097075b6 N |
4410 | continue; /* time stamp is too bad */ |
4411 | } | |
ea0ebe96 | 4412 | } |
353632d9 | 4413 | |
e9e43ec3 | 4414 | if (bsb.magic[15] == '1') { |
10af14c4 N |
4415 | if (bsb.length == 0) |
4416 | continue; | |
e7a71c6b N |
4417 | if (info->delta_disks >= 0) { |
4418 | /* reshape_progress is increasing */ | |
4419 | if (__le64_to_cpu(bsb.arraystart) | |
4420 | + __le64_to_cpu(bsb.length) | |
4421 | < info->reshape_progress) { | |
4422 | nonew: | |
4423 | if (verbose) | |
e7b84f9d | 4424 | pr_err("backup-metadata found on %s but is not needed\n", devname); |
e7a71c6b N |
4425 | continue; /* No new data here */ |
4426 | } | |
4427 | } else { | |
4428 | /* reshape_progress is decreasing */ | |
4429 | if (__le64_to_cpu(bsb.arraystart) >= | |
4430 | info->reshape_progress) | |
4431 | goto nonew; /* No new data here */ | |
ea0ebe96 | 4432 | } |
e9e43ec3 | 4433 | } else { |
10af14c4 N |
4434 | if (bsb.length == 0 && bsb.length2 == 0) |
4435 | continue; | |
e7a71c6b N |
4436 | if (info->delta_disks >= 0) { |
4437 | /* reshape_progress is increasing */ | |
4438 | if ((__le64_to_cpu(bsb.arraystart) | |
4439 | + __le64_to_cpu(bsb.length) | |
4440 | < info->reshape_progress) | |
4441 | && | |
4442 | (__le64_to_cpu(bsb.arraystart2) | |
4443 | + __le64_to_cpu(bsb.length2) | |
4444 | < info->reshape_progress)) | |
4445 | goto nonew; /* No new data here */ | |
4446 | } else { | |
4447 | /* reshape_progress is decreasing */ | |
4448 | if (__le64_to_cpu(bsb.arraystart) >= | |
4449 | info->reshape_progress && | |
4450 | __le64_to_cpu(bsb.arraystart2) >= | |
4451 | info->reshape_progress) | |
4452 | goto nonew; /* No new data here */ | |
4453 | } | |
e9e43ec3 | 4454 | } |
ea0ebe96 N |
4455 | if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) { |
4456 | second_fail: | |
4457 | if (verbose) | |
e7b84f9d N |
4458 | pr_err("Failed to verify secondary backup-metadata block on %s\n", |
4459 | devname); | |
353632d9 | 4460 | continue; /* Cannot seek */ |
ea0ebe96 | 4461 | } |
2efedc7b | 4462 | /* There should be a duplicate backup superblock 4k before here */ |
06b0d786 | 4463 | if (lseek64(fd, -4096, 1) < 0 || |
0155af90 | 4464 | read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2)) |
ea0ebe96 | 4465 | goto second_fail; /* Cannot find leading superblock */ |
e9e43ec3 N |
4466 | if (bsb.magic[15] == '1') |
4467 | bsbsize = offsetof(struct mdp_backup_super, pad1); | |
4468 | else | |
4469 | bsbsize = offsetof(struct mdp_backup_super, pad); | |
ff94fb86 | 4470 | if (memcmp(&bsb2, &bsb, bsbsize) != 0) |
ea0ebe96 | 4471 | goto second_fail; /* Cannot find leading superblock */ |
2efedc7b | 4472 | |
353632d9 | 4473 | /* Now need the data offsets for all devices. */ |
503975b9 | 4474 | offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks); |
353632d9 NB |
4475 | for(j=0; j<info->array.raid_disks; j++) { |
4476 | if (fdlist[j] < 0) | |
4477 | continue; | |
3da92f27 | 4478 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 NB |
4479 | /* FIXME should be this be an error */ |
4480 | continue; | |
a5d85af7 | 4481 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 | 4482 | st->ss->free_super(st); |
14e5b4d7 | 4483 | offsets[j] = dinfo.data_offset * 512; |
353632d9 NB |
4484 | } |
4485 | printf(Name ": restoring critical section\n"); | |
4486 | ||
4487 | if (restore_stripes(fdlist, offsets, | |
4488 | info->array.raid_disks, | |
4489 | info->new_chunk, | |
4490 | info->new_level, | |
4491 | info->new_layout, | |
06b0d786 | 4492 | fd, __le64_to_cpu(bsb.devstart)*512, |
92dcdf7c | 4493 | __le64_to_cpu(bsb.arraystart)*512, |
2fcb75ae | 4494 | __le64_to_cpu(bsb.length)*512, NULL)) { |
e9e43ec3 | 4495 | /* didn't succeed, so giveup */ |
ea0ebe96 | 4496 | if (verbose) |
e7b84f9d | 4497 | pr_err("Error restoring backup from %s\n", |
ea0ebe96 | 4498 | devname); |
730ae51f | 4499 | free(offsets); |
e9e43ec3 N |
4500 | return 1; |
4501 | } | |
24daa16f | 4502 | |
e9e43ec3 N |
4503 | if (bsb.magic[15] == '2' && |
4504 | restore_stripes(fdlist, offsets, | |
4505 | info->array.raid_disks, | |
4506 | info->new_chunk, | |
4507 | info->new_level, | |
4508 | info->new_layout, | |
4509 | fd, __le64_to_cpu(bsb.devstart)*512 + | |
4510 | __le64_to_cpu(bsb.devstart2)*512, | |
92dcdf7c | 4511 | __le64_to_cpu(bsb.arraystart2)*512, |
2fcb75ae | 4512 | __le64_to_cpu(bsb.length2)*512, NULL)) { |
353632d9 | 4513 | /* didn't succeed, so giveup */ |
ea0ebe96 | 4514 | if (verbose) |
e7b84f9d | 4515 | pr_err("Error restoring second backup from %s\n", |
ea0ebe96 | 4516 | devname); |
730ae51f | 4517 | free(offsets); |
2295250a | 4518 | return 1; |
353632d9 NB |
4519 | } |
4520 | ||
730ae51f | 4521 | free(offsets); |
e9e43ec3 | 4522 | |
353632d9 NB |
4523 | /* Ok, so the data is restored. Let's update those superblocks. */ |
4524 | ||
10af14c4 N |
4525 | lo = hi = 0; |
4526 | if (bsb.length) { | |
4527 | lo = __le64_to_cpu(bsb.arraystart); | |
4528 | hi = lo + __le64_to_cpu(bsb.length); | |
4529 | } | |
4530 | if (bsb.magic[15] == '2' && bsb.length2) { | |
4531 | unsigned long long lo1, hi1; | |
4532 | lo1 = __le64_to_cpu(bsb.arraystart2); | |
4533 | hi1 = lo1 + __le64_to_cpu(bsb.length2); | |
4534 | if (lo == hi) { | |
4535 | lo = lo1; | |
4536 | hi = hi1; | |
4537 | } else if (lo < lo1) | |
4538 | hi = hi1; | |
4539 | else | |
4540 | lo = lo1; | |
4541 | } | |
4542 | if (lo < hi && | |
4543 | (info->reshape_progress < lo || | |
4544 | info->reshape_progress > hi)) | |
4545 | /* backup does not affect reshape_progress*/ ; | |
4546 | else if (info->delta_disks >= 0) { | |
e9e43ec3 N |
4547 | info->reshape_progress = __le64_to_cpu(bsb.arraystart) + |
4548 | __le64_to_cpu(bsb.length); | |
4549 | if (bsb.magic[15] == '2') { | |
4550 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) + | |
4551 | __le64_to_cpu(bsb.length2); | |
4552 | if (p2 > info->reshape_progress) | |
4553 | info->reshape_progress = p2; | |
4554 | } | |
4555 | } else { | |
4556 | info->reshape_progress = __le64_to_cpu(bsb.arraystart); | |
4557 | if (bsb.magic[15] == '2') { | |
4558 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2); | |
4559 | if (p2 < info->reshape_progress) | |
4560 | info->reshape_progress = p2; | |
4561 | } | |
4562 | } | |
353632d9 | 4563 | for (j=0; j<info->array.raid_disks; j++) { |
ca3b6696 N |
4564 | if (fdlist[j] < 0) |
4565 | continue; | |
3da92f27 | 4566 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 | 4567 | continue; |
a5d85af7 | 4568 | st->ss->getinfo_super(st, &dinfo, NULL); |
e9e43ec3 | 4569 | dinfo.reshape_progress = info->reshape_progress; |
3da92f27 | 4570 | st->ss->update_super(st, &dinfo, |
68c7d6d7 NB |
4571 | "_reshape_progress", |
4572 | NULL,0, 0, NULL); | |
3da92f27 NB |
4573 | st->ss->store_super(st, fdlist[j]); |
4574 | st->ss->free_super(st); | |
353632d9 | 4575 | } |
353632d9 NB |
4576 | return 0; |
4577 | } | |
6e9eac4f NB |
4578 | /* Didn't find any backup data, try to see if any |
4579 | * was needed. | |
4580 | */ | |
82f2d6ab N |
4581 | if (info->delta_disks < 0) { |
4582 | /* When shrinking, the critical section is at the end. | |
4583 | * So see if we are before the critical section. | |
4584 | */ | |
4585 | unsigned long long first_block; | |
4586 | nstripe = ostripe = 0; | |
4587 | first_block = 0; | |
4588 | while (ostripe >= nstripe) { | |
4589 | ostripe += info->array.chunk_size / 512; | |
4590 | first_block = ostripe * odata; | |
4591 | nstripe = first_block / ndata / (info->new_chunk/512) * | |
4592 | (info->new_chunk/512); | |
4593 | } | |
4594 | ||
4595 | if (info->reshape_progress >= first_block) | |
4596 | return 0; | |
6e9eac4f | 4597 | } |
82f2d6ab N |
4598 | if (info->delta_disks > 0) { |
4599 | /* See if we are beyond the critical section. */ | |
4600 | unsigned long long last_block; | |
4601 | nstripe = ostripe = 0; | |
4602 | last_block = 0; | |
4603 | while (nstripe >= ostripe) { | |
4604 | nstripe += info->new_chunk / 512; | |
4605 | last_block = nstripe * ndata; | |
4606 | ostripe = last_block / odata / (info->array.chunk_size/512) * | |
4607 | (info->array.chunk_size/512); | |
4608 | } | |
6e9eac4f | 4609 | |
82f2d6ab N |
4610 | if (info->reshape_progress >= last_block) |
4611 | return 0; | |
4612 | } | |
6e9eac4f | 4613 | /* needed to recover critical section! */ |
ea0ebe96 | 4614 | if (verbose) |
e7b84f9d | 4615 | pr_err("Failed to find backup of critical section\n"); |
2295250a | 4616 | return 1; |
353632d9 | 4617 | } |
e9e43ec3 | 4618 | |
2dddadb0 AK |
4619 | int Grow_continue_command(char *devname, int fd, |
4620 | char *backup_file, int verbose) | |
4621 | { | |
4622 | int ret_val = 0; | |
4623 | struct supertype *st = NULL; | |
4624 | struct mdinfo *content = NULL; | |
4625 | struct mdinfo array; | |
4626 | char *subarray = NULL; | |
4627 | struct mdinfo *cc = NULL; | |
4628 | struct mdstat_ent *mdstat = NULL; | |
2dddadb0 AK |
4629 | int cfd = -1; |
4630 | int fd2 = -1; | |
4631 | ||
4632 | dprintf("Grow continue from command line called for %s\n", | |
4633 | devname); | |
4634 | ||
4635 | st = super_by_fd(fd, &subarray); | |
4636 | if (!st || !st->ss) { | |
e7b84f9d N |
4637 | pr_err("Unable to determine metadata format for %s\n", |
4638 | devname); | |
2dddadb0 AK |
4639 | return 1; |
4640 | } | |
4641 | dprintf("Grow continue is run for "); | |
4642 | if (st->ss->external == 0) { | |
e0ab37a3 | 4643 | int d; |
2dddadb0 | 4644 | dprintf("native array (%s)\n", devname); |
e0ab37a3 | 4645 | if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) { |
e7b84f9d | 4646 | pr_err("%s is not an active md array -" |
2dddadb0 AK |
4647 | " aborting\n", devname); |
4648 | ret_val = 1; | |
4649 | goto Grow_continue_command_exit; | |
4650 | } | |
4651 | content = &array; | |
e0ab37a3 N |
4652 | /* Need to load a superblock. |
4653 | * FIXME we should really get what we need from | |
4654 | * sysfs | |
4655 | */ | |
4656 | for (d = 0; d < MAX_DISKS; d++) { | |
4657 | mdu_disk_info_t disk; | |
4658 | char *dv; | |
4659 | int err; | |
4660 | disk.number = d; | |
4661 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
4662 | continue; | |
4663 | if (disk.major == 0 && disk.minor == 0) | |
4664 | continue; | |
4665 | if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0) | |
4666 | continue; | |
4667 | dv = map_dev(disk.major, disk.minor, 1); | |
4668 | if (!dv) | |
4669 | continue; | |
4670 | fd2 = dev_open(dv, O_RDONLY); | |
4671 | if (fd2 < 0) | |
4672 | continue; | |
4673 | err = st->ss->load_super(st, fd2, NULL); | |
4674 | close(fd2); | |
4675 | if (err) | |
4676 | continue; | |
4677 | break; | |
4678 | } | |
4679 | if (d == MAX_DISKS) { | |
4680 | pr_err("Unable to load metadata for %s\n", | |
4681 | devname); | |
4682 | ret_val = 1; | |
4683 | goto Grow_continue_command_exit; | |
4684 | } | |
4685 | st->ss->getinfo_super(st, content, NULL); | |
2dddadb0 | 4686 | } else { |
4dd2df09 | 4687 | char *container; |
2dddadb0 AK |
4688 | |
4689 | if (subarray) { | |
4690 | dprintf("subarray (%s)\n", subarray); | |
4dd2df09 N |
4691 | container = st->container_devnm; |
4692 | cfd = open_dev_excl(st->container_devnm); | |
2dddadb0 | 4693 | } else { |
4dd2df09 | 4694 | container = st->devnm; |
2dddadb0 | 4695 | close(fd); |
4dd2df09 N |
4696 | cfd = open_dev_excl(st->devnm); |
4697 | dprintf("container (%s)\n", container); | |
2dddadb0 AK |
4698 | fd = cfd; |
4699 | } | |
4700 | if (cfd < 0) { | |
e7b84f9d | 4701 | pr_err("Unable to open container " |
2dddadb0 AK |
4702 | "for %s\n", devname); |
4703 | ret_val = 1; | |
4704 | goto Grow_continue_command_exit; | |
4705 | } | |
2dddadb0 AK |
4706 | |
4707 | /* find in container array under reshape | |
4708 | */ | |
4709 | ret_val = st->ss->load_container(st, cfd, NULL); | |
4710 | if (ret_val) { | |
e7b84f9d N |
4711 | pr_err("Cannot read superblock for %s\n", |
4712 | devname); | |
2dddadb0 AK |
4713 | ret_val = 1; |
4714 | goto Grow_continue_command_exit; | |
4715 | } | |
4716 | ||
446894ea | 4717 | cc = st->ss->container_content(st, subarray); |
2dddadb0 AK |
4718 | for (content = cc; content ; content = content->next) { |
4719 | char *array; | |
446894ea | 4720 | int allow_reshape = 1; |
2dddadb0 AK |
4721 | |
4722 | if (content->reshape_active == 0) | |
4723 | continue; | |
81219e70 LM |
4724 | /* The decision about array or container wide |
4725 | * reshape is taken in Grow_continue based | |
4726 | * content->reshape_active state, therefore we | |
4727 | * need to check_reshape based on | |
4728 | * reshape_active and subarray name | |
446894ea N |
4729 | */ |
4730 | if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) | |
4731 | allow_reshape = 0; | |
4732 | if (content->reshape_active == CONTAINER_RESHAPE && | |
4733 | (content->array.state | |
4734 | & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))) | |
4735 | allow_reshape = 0; | |
4736 | ||
81219e70 | 4737 | if (!allow_reshape) { |
e7b84f9d N |
4738 | pr_err("cannot continue reshape of an array" |
4739 | " in container with unsupported" | |
4740 | " metadata: %s(%s)\n", | |
4dd2df09 | 4741 | devname, container); |
81219e70 LM |
4742 | ret_val = 1; |
4743 | goto Grow_continue_command_exit; | |
4744 | } | |
2dddadb0 AK |
4745 | |
4746 | array = strchr(content->text_version+1, '/')+1; | |
4dd2df09 | 4747 | mdstat = mdstat_by_subdev(array, container); |
2dddadb0 AK |
4748 | if (!mdstat) |
4749 | continue; | |
1ca90aa6 | 4750 | if (mdstat->active == 0) { |
4dd2df09 N |
4751 | pr_err("Skipping inactive array %s.\n", |
4752 | mdstat->devnm); | |
1ca90aa6 AK |
4753 | free_mdstat(mdstat); |
4754 | mdstat = NULL; | |
4755 | continue; | |
4756 | } | |
2dddadb0 AK |
4757 | break; |
4758 | } | |
4759 | if (!content) { | |
e7b84f9d N |
4760 | pr_err("Unable to determine reshaped " |
4761 | "array for %s\n", devname); | |
2dddadb0 AK |
4762 | ret_val = 1; |
4763 | goto Grow_continue_command_exit; | |
4764 | } | |
4dd2df09 | 4765 | fd2 = open_dev(mdstat->devnm); |
2dddadb0 | 4766 | if (fd2 < 0) { |
4dd2df09 | 4767 | pr_err("cannot open (%s)\n", mdstat->devnm); |
2dddadb0 AK |
4768 | ret_val = 1; |
4769 | goto Grow_continue_command_exit; | |
4770 | } | |
4771 | ||
4dd2df09 | 4772 | sysfs_init(content, fd2, mdstat->devnm); |
2dddadb0 AK |
4773 | |
4774 | /* start mdmon in case it is not running | |
4775 | */ | |
4dd2df09 N |
4776 | if (!mdmon_running(container)) |
4777 | start_mdmon(container); | |
4778 | ping_monitor(container); | |
2dddadb0 | 4779 | |
4dd2df09 | 4780 | if (mdmon_running(container)) |
2dddadb0 AK |
4781 | st->update_tail = &st->updates; |
4782 | else { | |
e7b84f9d | 4783 | pr_err("No mdmon found. " |
2dddadb0 AK |
4784 | "Grow cannot continue.\n"); |
4785 | ret_val = 1; | |
4786 | goto Grow_continue_command_exit; | |
4787 | } | |
4788 | } | |
4789 | ||
f1fe496b AK |
4790 | /* verify that array under reshape is started from |
4791 | * correct position | |
4792 | */ | |
e0ab37a3 | 4793 | if (verify_reshape_position(content, content->array.level) < 0) { |
f1fe496b AK |
4794 | ret_val = 1; |
4795 | goto Grow_continue_command_exit; | |
4796 | } | |
4797 | ||
2dddadb0 AK |
4798 | /* continue reshape |
4799 | */ | |
4800 | ret_val = Grow_continue(fd, st, content, backup_file, 0); | |
4801 | ||
4802 | Grow_continue_command_exit: | |
4803 | if (fd2 > -1) | |
4804 | close(fd2); | |
4805 | if (cfd > -1) | |
4806 | close(cfd); | |
4807 | st->ss->free_super(st); | |
4808 | free_mdstat(mdstat); | |
4809 | sysfs_free(cc); | |
4810 | free(subarray); | |
4811 | ||
4812 | return ret_val; | |
4813 | } | |
4814 | ||
e9e43ec3 | 4815 | int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info, |
b76b30e0 | 4816 | char *backup_file, int freeze_reshape) |
e9e43ec3 | 4817 | { |
3bd58dc6 AK |
4818 | int ret_val = 2; |
4819 | ||
4820 | if (!info->reshape_active) | |
4821 | return ret_val; | |
864a004f | 4822 | |
20a40eca | 4823 | if (st->ss->external) { |
4dd2df09 | 4824 | int cfd = open_dev(st->container_devnm); |
3db2fdd8 | 4825 | |
3bd58dc6 AK |
4826 | if (cfd < 0) |
4827 | return 1; | |
f362d22b | 4828 | |
4dd2df09 | 4829 | st->ss->load_container(st, cfd, st->container_devnm); |
3bd58dc6 | 4830 | close(cfd); |
4dd2df09 | 4831 | ret_val = reshape_container(st->container_devnm, NULL, mdfd, |
3bd58dc6 | 4832 | st, info, 0, backup_file, |
8ecf12b9 N |
4833 | 0, |
4834 | 1 | info->reshape_active, | |
4835 | freeze_reshape); | |
3bd58dc6 AK |
4836 | } else |
4837 | ret_val = reshape_array(NULL, mdfd, "array", st, info, 1, | |
8ecf12b9 N |
4838 | NULL, 0ULL, backup_file, 0, 0, |
4839 | 1 | info->reshape_active, | |
3bd58dc6 | 4840 | freeze_reshape); |
f362d22b | 4841 | |
3bd58dc6 | 4842 | return ret_val; |
e9e43ec3 | 4843 | } |