]>
Commit | Line | Data |
---|---|---|
e5329c37 NB |
1 | /* |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
6f02172d | 4 | * Copyright (C) 2001-2013 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> |
04f903b2 | 27 | #include <stddef.h> |
f9b08fec | 28 | #include <stdint.h> |
5e76dce1 | 29 | #include <sys/wait.h> |
e5329c37 NB |
30 | |
31 | #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN) | |
32 | #error no endian defined | |
33 | #endif | |
34 | #include "md_u.h" | |
35 | #include "md_p.h" | |
36 | ||
3f54bd62 AK |
37 | int restore_backup(struct supertype *st, |
38 | struct mdinfo *content, | |
39 | int working_disks, | |
40 | int next_spare, | |
54ded86f | 41 | char **backup_filep, |
3f54bd62 AK |
42 | int verbose) |
43 | { | |
44 | int i; | |
45 | int *fdlist; | |
46 | struct mdinfo *dev; | |
47 | int err; | |
48 | int disk_count = next_spare + working_disks; | |
54ded86f | 49 | char *backup_file = *backup_filep; |
3f54bd62 AK |
50 | |
51 | dprintf("Called restore_backup()\n"); | |
503975b9 N |
52 | fdlist = xmalloc(sizeof(int) * disk_count); |
53 | ||
a7dec3fd | 54 | enable_fds(next_spare); |
3f54bd62 AK |
55 | for (i = 0; i < next_spare; i++) |
56 | fdlist[i] = -1; | |
57 | for (dev = content->devs; dev; dev = dev->next) { | |
58 | char buf[22]; | |
59 | int fd; | |
358ef9bf JS |
60 | |
61 | sprintf(buf, "%d:%d", dev->disk.major, dev->disk.minor); | |
3f54bd62 AK |
62 | fd = dev_open(buf, O_RDWR); |
63 | ||
64 | if (dev->disk.raid_disk >= 0) | |
65 | fdlist[dev->disk.raid_disk] = fd; | |
66 | else | |
67 | fdlist[next_spare++] = fd; | |
68 | } | |
69 | ||
54ded86f N |
70 | if (!backup_file) { |
71 | backup_file = locate_backup(content->sys_name); | |
72 | *backup_filep = backup_file; | |
73 | } | |
74 | ||
3f54bd62 AK |
75 | if (st->ss->external && st->ss->recover_backup) |
76 | err = st->ss->recover_backup(st, content); | |
77 | else | |
78 | err = Grow_restart(st, content, fdlist, next_spare, | |
79 | backup_file, verbose > 0); | |
80 | ||
81 | while (next_spare > 0) { | |
cc7f63e5 N |
82 | next_spare--; |
83 | if (fdlist[next_spare] >= 0) | |
84 | close(fdlist[next_spare]); | |
3f54bd62 AK |
85 | } |
86 | free(fdlist); | |
87 | if (err) { | |
7a862a02 | 88 | pr_err("Failed to restore critical section for reshape - sorry.\n"); |
3f54bd62 | 89 | if (!backup_file) |
7a862a02 | 90 | pr_err("Possibly you need to specify a --backup-file\n"); |
3f54bd62 AK |
91 | return 1; |
92 | } | |
93 | ||
94 | dprintf("restore_backup() returns status OK.\n"); | |
95 | return 0; | |
96 | } | |
97 | ||
e5329c37 NB |
98 | int Grow_Add_device(char *devname, int fd, char *newdev) |
99 | { | |
100 | /* Add a device to an active array. | |
101 | * Currently, just extend a linear array. | |
102 | * This requires writing a new superblock on the | |
103 | * new device, calling the kernel to add the device, | |
104 | * and if that succeeds, update the superblock on | |
105 | * all other devices. | |
106 | * This means that we need to *find* all other devices. | |
107 | */ | |
4b1ac34b NB |
108 | struct mdinfo info; |
109 | ||
0a6bff09 | 110 | dev_t rdev; |
e5329c37 NB |
111 | int nfd, fd2; |
112 | int d, nd; | |
82d9eba6 | 113 | struct supertype *st = NULL; |
4725bc31 | 114 | char *subarray = NULL; |
e5329c37 | 115 | |
9cd39f01 | 116 | if (md_get_array_info(fd, &info.array) < 0) { |
e7b84f9d | 117 | pr_err("cannot get array info for %s\n", devname); |
e5329c37 NB |
118 | return 1; |
119 | } | |
120 | ||
4725bc31 | 121 | if (info.array.level != -1) { |
e7b84f9d | 122 | pr_err("can only add devices to linear arrays\n"); |
4725bc31 N |
123 | return 1; |
124 | } | |
125 | ||
126 | st = super_by_fd(fd, &subarray); | |
82d9eba6 | 127 | if (!st) { |
ca3b6696 N |
128 | pr_err("cannot handle arrays with superblock version %d\n", |
129 | info.array.major_version); | |
f9ce90ba NB |
130 | return 1; |
131 | } | |
132 | ||
4725bc31 | 133 | if (subarray) { |
e7b84f9d | 134 | pr_err("Cannot grow linear sub-arrays yet\n"); |
4725bc31 N |
135 | free(subarray); |
136 | free(st); | |
2641101b | 137 | return 1; |
e5329c37 NB |
138 | } |
139 | ||
6416d527 | 140 | nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT); |
e5329c37 | 141 | if (nfd < 0) { |
e7b84f9d | 142 | pr_err("cannot open %s\n", newdev); |
4725bc31 | 143 | free(st); |
e5329c37 NB |
144 | return 1; |
145 | } | |
0a6bff09 | 146 | if (!fstat_is_blkdev(nfd, newdev, &rdev)) { |
e5329c37 | 147 | close(nfd); |
4725bc31 | 148 | free(st); |
e5329c37 NB |
149 | return 1; |
150 | } | |
ca3b6696 N |
151 | /* now check out all the devices and make sure we can read the |
152 | * superblock */ | |
4b1ac34b | 153 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
154 | mdu_disk_info_t disk; |
155 | char *dv; | |
156 | ||
4725bc31 N |
157 | st->ss->free_super(st); |
158 | ||
e5329c37 | 159 | disk.number = d; |
d97572f5 | 160 | if (md_get_disk_info(fd, &disk) < 0) { |
358ef9bf | 161 | pr_err("cannot get device detail for device %d\n", d); |
4725bc31 N |
162 | close(nfd); |
163 | free(st); | |
e5329c37 NB |
164 | return 1; |
165 | } | |
16c6fa80 | 166 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 | 167 | if (!dv) { |
358ef9bf | 168 | pr_err("cannot find device file for device %d\n", d); |
4725bc31 N |
169 | close(nfd); |
170 | free(st); | |
e5329c37 NB |
171 | return 1; |
172 | } | |
16c6fa80 | 173 | fd2 = dev_open(dv, O_RDWR); |
68fe8c6e | 174 | if (fd2 < 0) { |
e7b84f9d | 175 | pr_err("cannot open device file %s\n", dv); |
4725bc31 N |
176 | close(nfd); |
177 | free(st); | |
e5329c37 NB |
178 | return 1; |
179 | } | |
3da92f27 NB |
180 | |
181 | if (st->ss->load_super(st, fd2, NULL)) { | |
e7b84f9d | 182 | pr_err("cannot find super block on %s\n", dv); |
4725bc31 | 183 | close(nfd); |
e5329c37 | 184 | close(fd2); |
4725bc31 | 185 | free(st); |
e5329c37 NB |
186 | return 1; |
187 | } | |
188 | close(fd2); | |
189 | } | |
190 | /* Ok, looks good. Lets update the superblock and write it out to | |
191 | * newdev. | |
192 | */ | |
aba69144 | 193 | |
4b1ac34b | 194 | info.disk.number = d; |
0a6bff09 ZL |
195 | info.disk.major = major(rdev); |
196 | info.disk.minor = minor(rdev); | |
4b1ac34b NB |
197 | info.disk.raid_disk = d; |
198 | info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
8818d4e7 N |
199 | if (st->ss->update_super(st, &info, "linear-grow-new", newdev, |
200 | 0, 0, NULL) != 0) { | |
201 | pr_err("Preparing new metadata failed on %s\n", newdev); | |
202 | close(nfd); | |
203 | return 1; | |
204 | } | |
e5329c37 | 205 | |
3da92f27 | 206 | if (st->ss->store_super(st, nfd)) { |
358ef9bf | 207 | pr_err("Cannot store new superblock on %s\n", 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 | ||
9cd39f01 | 221 | if (md_get_array_info(fd, &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; | |
d97572f5 | 232 | if (md_get_disk_info(fd, &disk) < 0) { |
358ef9bf | 233 | pr_err("cannot get device detail for device %d\n", d); |
e5329c37 NB |
234 | return 1; |
235 | } | |
16c6fa80 | 236 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 | 237 | if (!dv) { |
358ef9bf | 238 | pr_err("cannot find device file for device %d\n", d); |
e5329c37 NB |
239 | return 1; |
240 | } | |
16c6fa80 | 241 | fd2 = dev_open(dv, O_RDWR); |
e5329c37 | 242 | if (fd2 < 0) { |
e7b84f9d | 243 | pr_err("cannot open device file %s\n", dv); |
e5329c37 NB |
244 | return 1; |
245 | } | |
3da92f27 | 246 | if (st->ss->load_super(st, fd2, NULL)) { |
e7b84f9d | 247 | pr_err("cannot find super block on %s\n", dv); |
e5329c37 | 248 | close(fd); |
d94eb07f | 249 | close(fd2); |
e5329c37 NB |
250 | return 1; |
251 | } | |
4b1ac34b NB |
252 | info.array.raid_disks = nd+1; |
253 | info.array.nr_disks = nd+1; | |
254 | info.array.active_disks = nd+1; | |
255 | info.array.working_disks = nd+1; | |
f752781f | 256 | |
8818d4e7 N |
257 | if (st->ss->update_super(st, &info, "linear-grow-update", dv, |
258 | 0, 0, NULL) != 0) { | |
259 | pr_err("Updating metadata failed on %s\n", dv); | |
260 | close(fd2); | |
261 | return 1; | |
262 | } | |
aba69144 | 263 | |
3da92f27 | 264 | if (st->ss->store_super(st, fd2)) { |
e7b84f9d | 265 | pr_err("Cannot store new superblock on %s\n", dv); |
e5329c37 NB |
266 | close(fd2); |
267 | return 1; | |
268 | } | |
269 | close(fd2); | |
270 | } | |
271 | ||
272 | return 0; | |
273 | } | |
f5e166fe | 274 | |
50f01ba5 | 275 | int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s) |
f5e166fe NB |
276 | { |
277 | /* | |
278 | * First check that array doesn't have a bitmap | |
279 | * Then create the bitmap | |
280 | * Then add it | |
281 | * | |
282 | * For internal bitmaps, we need to check the version, | |
283 | * find all the active devices, and write the bitmap block | |
284 | * to all devices | |
285 | */ | |
286 | mdu_bitmap_file_t bmf; | |
287 | mdu_array_info_t array; | |
288 | struct supertype *st; | |
4725bc31 | 289 | char *subarray = NULL; |
dcec9ee5 | 290 | int major = BITMAP_MAJOR_HI; |
8fac0577 | 291 | unsigned long long bitmapsize, array_size; |
e97a7cd0 | 292 | struct mdinfo *mdi; |
dcec9ee5 | 293 | |
efdfcc9e JS |
294 | /* |
295 | * We only ever get called if s->bitmap_file is != NULL, so this check | |
296 | * is just here to quiet down static code checkers. | |
297 | */ | |
298 | if (!s->bitmap_file) | |
299 | return 1; | |
300 | ||
301 | if (strcmp(s->bitmap_file, "clustered") == 0) | |
6d9c7c25 GR |
302 | major = BITMAP_MAJOR_CLUSTERED; |
303 | ||
f5e166fe | 304 | if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) { |
353632d9 | 305 | if (errno == ENOMEM) |
e7b84f9d | 306 | pr_err("Memory allocation failure.\n"); |
f5e166fe | 307 | else |
e7b84f9d | 308 | pr_err("bitmaps not supported by this kernel.\n"); |
f5e166fe NB |
309 | return 1; |
310 | } | |
311 | if (bmf.pathname[0]) { | |
6ac963ce JS |
312 | if (strcmp(s->bitmap_file,"none") == 0) { |
313 | if (ioctl(fd, SET_BITMAP_FILE, -1) != 0) { | |
e7b84f9d | 314 | pr_err("failed to remove bitmap %s\n", |
fe80f49b NB |
315 | bmf.pathname); |
316 | return 1; | |
317 | } | |
318 | return 0; | |
319 | } | |
358ef9bf | 320 | pr_err("%s already has a bitmap (%s)\n", devname, bmf.pathname); |
f5e166fe NB |
321 | return 1; |
322 | } | |
9cd39f01 | 323 | if (md_get_array_info(fd, &array) != 0) { |
e7b84f9d | 324 | pr_err("cannot get array status for %s\n", devname); |
f5e166fe NB |
325 | return 1; |
326 | } | |
6ac963ce | 327 | if (array.state & (1 << MD_SB_BITMAP_PRESENT)) { |
50f01ba5 | 328 | if (strcmp(s->bitmap_file, "none")==0) { |
6ac963ce | 329 | array.state &= ~(1 << MD_SB_BITMAP_PRESENT); |
018a4882 | 330 | if (md_set_array_info(fd, &array) != 0) { |
6ac963ce | 331 | if (array.state & (1 << MD_SB_CLUSTERED)) |
37d0ca9b GJ |
332 | pr_err("failed to remove clustered bitmap.\n"); |
333 | else | |
334 | pr_err("failed to remove internal bitmap.\n"); | |
fe80f49b NB |
335 | return 1; |
336 | } | |
337 | return 0; | |
338 | } | |
37d0ca9b | 339 | pr_err("bitmap already present on %s\n", devname); |
f5e166fe NB |
340 | return 1; |
341 | } | |
4725bc31 | 342 | |
50f01ba5 | 343 | if (strcmp(s->bitmap_file, "none") == 0) { |
e7b84f9d | 344 | pr_err("no bitmap found on %s\n", devname); |
4725bc31 N |
345 | return 1; |
346 | } | |
5b28bd56 | 347 | if (array.level <= 0) { |
e7b84f9d | 348 | pr_err("Bitmaps not meaningful with level %s\n", |
5b28bd56 NB |
349 | map_num(pers, array.level)?:"of this array"); |
350 | return 1; | |
351 | } | |
8fac0577 NB |
352 | bitmapsize = array.size; |
353 | bitmapsize <<= 1; | |
beae1dfe | 354 | if (get_dev_size(fd, NULL, &array_size) && |
6ac963ce | 355 | array_size > (0x7fffffffULL << 9)) { |
8fac0577 NB |
356 | /* Array is big enough that we cannot trust array.size |
357 | * try other approaches | |
358 | */ | |
359 | bitmapsize = get_component_size(fd); | |
360 | } | |
8fac0577 | 361 | if (bitmapsize == 0) { |
e7b84f9d | 362 | pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n"); |
8fac0577 NB |
363 | return 1; |
364 | } | |
365 | ||
f9c25f1d | 366 | if (array.level == 10) { |
6ac963ce JS |
367 | int ncopies; |
368 | ||
369 | ncopies = (array.layout & 255) * ((array.layout >> 8) & 255); | |
f9c25f1d | 370 | bitmapsize = bitmapsize * array.raid_disks / ncopies; |
5339f996 GJ |
371 | |
372 | if (strcmp(s->bitmap_file, "clustered") == 0 && | |
373 | !is_near_layout_10(array.layout)) { | |
374 | pr_err("only near layout is supported with clustered raid10\n"); | |
375 | return 1; | |
376 | } | |
f9c25f1d NB |
377 | } |
378 | ||
4725bc31 | 379 | st = super_by_fd(fd, &subarray); |
f5e166fe | 380 | if (!st) { |
e7b84f9d | 381 | pr_err("Cannot understand version %d.%d\n", |
f5e166fe NB |
382 | array.major_version, array.minor_version); |
383 | return 1; | |
384 | } | |
4725bc31 | 385 | if (subarray) { |
e7b84f9d | 386 | pr_err("Cannot add bitmaps to sub-arrays yet\n"); |
4725bc31 N |
387 | free(subarray); |
388 | free(st); | |
fe80f49b | 389 | return 1; |
4725bc31 | 390 | } |
e97a7cd0 AP |
391 | |
392 | mdi = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY); | |
393 | if (mdi) { | |
394 | if (mdi->consistency_policy == CONSISTENCY_POLICY_PPL) { | |
395 | pr_err("Cannot add bitmap to array with PPL\n"); | |
396 | free(mdi); | |
397 | free(st); | |
398 | return 1; | |
399 | } | |
400 | free(mdi); | |
401 | } | |
402 | ||
7c25f4d7 GJ |
403 | if (strcmp(s->bitmap_file, "internal") == 0 || |
404 | strcmp(s->bitmap_file, "clustered") == 0) { | |
c0c1acd6 | 405 | int rv; |
f5e166fe | 406 | int d; |
c0c1acd6 | 407 | int offset_setable = 0; |
ebeb3663 | 408 | if (st->ss->add_internal_bitmap == NULL) { |
7a862a02 | 409 | pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name); |
ebeb3663 N |
410 | return 1; |
411 | } | |
7c25f4d7 GJ |
412 | st->nodes = c->nodes; |
413 | st->cluster_name = c->homecluster; | |
4dd2df09 | 414 | mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION); |
c0c1acd6 N |
415 | if (mdi) |
416 | offset_setable = 1; | |
6ac963ce | 417 | for (d = 0; d < st->max_devs; d++) { |
f5e166fe NB |
418 | mdu_disk_info_t disk; |
419 | char *dv; | |
dac1b111 JS |
420 | int fd2; |
421 | ||
f5e166fe | 422 | disk.number = d; |
d97572f5 | 423 | if (md_get_disk_info(fd, &disk) < 0) |
f5e166fe | 424 | continue; |
6ac963ce | 425 | if (disk.major == 0 && disk.minor == 0) |
f5e166fe | 426 | continue; |
6ac963ce | 427 | if ((disk.state & (1 << MD_DISK_SYNC)) == 0) |
f5e166fe | 428 | continue; |
16c6fa80 | 429 | dv = map_dev(disk.major, disk.minor, 1); |
dac1b111 JS |
430 | if (!dv) |
431 | continue; | |
a0422106 NC |
432 | if (((disk.state & (1 << MD_DISK_WRITEMOSTLY)) == 0) && |
433 | (strcmp(s->bitmap_file, "clustered") == 0)) { | |
434 | pr_err("%s disks marked write-mostly are not supported with clustered bitmap\n",devname); | |
435 | return 1; | |
436 | } | |
dac1b111 JS |
437 | fd2 = dev_open(dv, O_RDWR); |
438 | if (fd2 < 0) | |
439 | continue; | |
c152f361 JS |
440 | rv = st->ss->load_super(st, fd2, NULL); |
441 | if (!rv) { | |
4ed129ac JS |
442 | rv = st->ss->add_internal_bitmap( |
443 | st, &s->bitmap_chunk, c->delay, | |
444 | s->write_behind, bitmapsize, | |
445 | offset_setable, major); | |
446 | if (!rv) { | |
447 | st->ss->write_bitmap(st, fd2, | |
448 | NodeNumUpdate); | |
449 | } else { | |
dac1b111 | 450 | pr_err("failed to create internal bitmap - chunksize problem.\n"); |
f5e166fe | 451 | } |
c152f361 JS |
452 | } else { |
453 | pr_err("failed to load super-block.\n"); | |
f5e166fe | 454 | } |
dac1b111 | 455 | close(fd2); |
4ed129ac JS |
456 | if (rv) |
457 | return 1; | |
f5e166fe | 458 | } |
c0c1acd6 N |
459 | if (offset_setable) { |
460 | st->ss->getinfo_super(st, mdi, NULL); | |
dae13137 | 461 | if (sysfs_init(mdi, fd, NULL)) { |
ebf3be99 | 462 | pr_err("failed to initialize sysfs.\n"); |
dae13137 JS |
463 | free(mdi); |
464 | } | |
012a8641 JS |
465 | rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location", |
466 | mdi->bitmap_offset); | |
e97a7cd0 | 467 | free(mdi); |
c0c1acd6 | 468 | } else { |
7c25f4d7 | 469 | if (strcmp(s->bitmap_file, "clustered") == 0) |
6ac963ce JS |
470 | array.state |= (1 << MD_SB_CLUSTERED); |
471 | array.state |= (1 << MD_SB_BITMAP_PRESENT); | |
018a4882 | 472 | rv = md_set_array_info(fd, &array); |
c0c1acd6 N |
473 | } |
474 | if (rv < 0) { | |
ff634064 | 475 | if (errno == EBUSY) |
7a862a02 | 476 | pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n"); |
e7b84f9d | 477 | pr_err("failed to set internal bitmap.\n"); |
f5e166fe NB |
478 | return 1; |
479 | } | |
fe80f49b NB |
480 | } else { |
481 | int uuid[4]; | |
482 | int bitmap_fd; | |
483 | int d; | |
484 | int max_devs = st->max_devs; | |
fe80f49b NB |
485 | |
486 | /* try to load a superblock */ | |
ca3b6696 | 487 | for (d = 0; d < max_devs; d++) { |
fe80f49b NB |
488 | mdu_disk_info_t disk; |
489 | char *dv; | |
490 | int fd2; | |
491 | disk.number = d; | |
d97572f5 | 492 | if (md_get_disk_info(fd, &disk) < 0) |
fe80f49b | 493 | continue; |
6ac963ce JS |
494 | if ((disk.major==0 && disk.minor == 0) || |
495 | (disk.state & (1 << MD_DISK_REMOVED))) | |
fe80f49b | 496 | continue; |
16c6fa80 | 497 | dv = map_dev(disk.major, disk.minor, 1); |
ca3b6696 N |
498 | if (!dv) |
499 | continue; | |
16c6fa80 | 500 | fd2 = dev_open(dv, O_RDONLY); |
e7344e90 JS |
501 | if (fd2 >= 0) { |
502 | if (st->ss->load_super(st, fd2, NULL) == 0) { | |
503 | close(fd2); | |
504 | st->ss->uuid_from_super(st, uuid); | |
505 | break; | |
506 | } | |
fe80f49b | 507 | close(fd2); |
fe80f49b | 508 | } |
fe80f49b NB |
509 | } |
510 | if (d == max_devs) { | |
e7b84f9d | 511 | pr_err("cannot find UUID for array!\n"); |
fe80f49b NB |
512 | return 1; |
513 | } | |
6ac963ce JS |
514 | if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, |
515 | s->bitmap_chunk, c->delay, s->write_behind, | |
516 | bitmapsize, major)) { | |
fe80f49b NB |
517 | return 1; |
518 | } | |
50f01ba5 | 519 | bitmap_fd = open(s->bitmap_file, O_RDWR); |
fe80f49b | 520 | if (bitmap_fd < 0) { |
6ac963ce | 521 | pr_err("weird: %s cannot be opened\n", s->bitmap_file); |
fe80f49b NB |
522 | return 1; |
523 | } | |
524 | if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
ff634064 N |
525 | int err = errno; |
526 | if (errno == EBUSY) | |
7a862a02 | 527 | pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n"); |
e7b84f9d | 528 | pr_err("Cannot set bitmap file for %s: %s\n", |
ff634064 | 529 | devname, strerror(err)); |
fe80f49b NB |
530 | return 1; |
531 | } | |
532 | } | |
f5e166fe NB |
533 | |
534 | return 0; | |
535 | } | |
536 | ||
860f11ed AP |
537 | int Grow_consistency_policy(char *devname, int fd, struct context *c, struct shape *s) |
538 | { | |
539 | struct supertype *st; | |
540 | struct mdinfo *sra; | |
541 | struct mdinfo *sd; | |
542 | char *subarray = NULL; | |
543 | int ret = 0; | |
544 | char container_dev[PATH_MAX]; | |
922a5829 | 545 | char buf[20]; |
860f11ed AP |
546 | |
547 | if (s->consistency_policy != CONSISTENCY_POLICY_RESYNC && | |
548 | s->consistency_policy != CONSISTENCY_POLICY_PPL) { | |
549 | pr_err("Operation not supported for consistency policy %s\n", | |
5f21d674 | 550 | map_num_s(consistency_policies, s->consistency_policy)); |
860f11ed AP |
551 | return 1; |
552 | } | |
553 | ||
554 | st = super_by_fd(fd, &subarray); | |
555 | if (!st) | |
556 | return 1; | |
557 | ||
558 | sra = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY|GET_LEVEL| | |
559 | GET_DEVS|GET_STATE); | |
560 | if (!sra) { | |
561 | ret = 1; | |
562 | goto free_st; | |
563 | } | |
564 | ||
565 | if (s->consistency_policy == CONSISTENCY_POLICY_PPL && | |
566 | !st->ss->write_init_ppl) { | |
567 | pr_err("%s metadata does not support PPL\n", st->ss->name); | |
568 | ret = 1; | |
569 | goto free_info; | |
570 | } | |
571 | ||
572 | if (sra->array.level != 5) { | |
573 | pr_err("Operation not supported for array level %d\n", | |
574 | sra->array.level); | |
575 | ret = 1; | |
576 | goto free_info; | |
577 | } | |
578 | ||
579 | if (sra->consistency_policy == (unsigned)s->consistency_policy) { | |
580 | pr_err("Consistency policy is already %s\n", | |
5f21d674 | 581 | map_num_s(consistency_policies, s->consistency_policy)); |
860f11ed AP |
582 | ret = 1; |
583 | goto free_info; | |
584 | } else if (sra->consistency_policy != CONSISTENCY_POLICY_RESYNC && | |
585 | sra->consistency_policy != CONSISTENCY_POLICY_PPL) { | |
586 | pr_err("Current consistency policy is %s, cannot change to %s\n", | |
5f21d674 MT |
587 | map_num_s(consistency_policies, sra->consistency_policy), |
588 | map_num_s(consistency_policies, s->consistency_policy)); | |
860f11ed AP |
589 | ret = 1; |
590 | goto free_info; | |
591 | } | |
592 | ||
922a5829 TM |
593 | if (s->consistency_policy == CONSISTENCY_POLICY_PPL) { |
594 | if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0) { | |
595 | ret = 1; | |
596 | goto free_info; | |
597 | } else if (strcmp(buf, "reshape\n") == 0) { | |
598 | pr_err("PPL cannot be enabled when reshape is in progress\n"); | |
599 | ret = 1; | |
600 | goto free_info; | |
601 | } | |
602 | } | |
603 | ||
860f11ed AP |
604 | if (subarray) { |
605 | char *update; | |
606 | ||
607 | if (s->consistency_policy == CONSISTENCY_POLICY_PPL) | |
608 | update = "ppl"; | |
609 | else | |
610 | update = "no-ppl"; | |
611 | ||
612 | sprintf(container_dev, "/dev/%s", st->container_devnm); | |
613 | ||
614 | ret = Update_subarray(container_dev, subarray, update, NULL, | |
615 | c->verbose); | |
616 | if (ret) | |
617 | goto free_info; | |
618 | } | |
619 | ||
620 | if (s->consistency_policy == CONSISTENCY_POLICY_PPL) { | |
621 | struct mdinfo info; | |
622 | ||
623 | if (subarray) { | |
624 | struct mdinfo *mdi; | |
625 | int cfd; | |
626 | ||
627 | cfd = open(container_dev, O_RDWR|O_EXCL); | |
628 | if (cfd < 0) { | |
629 | pr_err("Failed to open %s\n", container_dev); | |
630 | ret = 1; | |
631 | goto free_info; | |
632 | } | |
633 | ||
634 | ret = st->ss->load_container(st, cfd, st->container_devnm); | |
635 | close(cfd); | |
636 | ||
637 | if (ret) { | |
638 | pr_err("Cannot read superblock for %s\n", | |
639 | container_dev); | |
640 | goto free_info; | |
641 | } | |
642 | ||
643 | mdi = st->ss->container_content(st, subarray); | |
644 | info = *mdi; | |
645 | free(mdi); | |
646 | } | |
647 | ||
648 | for (sd = sra->devs; sd; sd = sd->next) { | |
649 | int dfd; | |
650 | char *devpath; | |
651 | ||
860f11ed AP |
652 | devpath = map_dev(sd->disk.major, sd->disk.minor, 0); |
653 | dfd = dev_open(devpath, O_RDWR); | |
654 | if (dfd < 0) { | |
655 | pr_err("Failed to open %s\n", devpath); | |
656 | ret = 1; | |
657 | goto free_info; | |
658 | } | |
659 | ||
660 | if (!subarray) { | |
661 | ret = st->ss->load_super(st, dfd, NULL); | |
662 | if (ret) { | |
663 | pr_err("Failed to load super-block.\n"); | |
664 | close(dfd); | |
665 | goto free_info; | |
666 | } | |
667 | ||
358ef9bf JS |
668 | ret = st->ss->update_super(st, sra, "ppl", |
669 | devname, | |
860f11ed AP |
670 | c->verbose, 0, NULL); |
671 | if (ret) { | |
672 | close(dfd); | |
673 | st->ss->free_super(st); | |
674 | goto free_info; | |
675 | } | |
676 | st->ss->getinfo_super(st, &info, NULL); | |
677 | } | |
678 | ||
358ef9bf JS |
679 | ret |= sysfs_set_num(sra, sd, "ppl_sector", |
680 | info.ppl_sector); | |
681 | ret |= sysfs_set_num(sra, sd, "ppl_size", | |
682 | info.ppl_size); | |
860f11ed AP |
683 | |
684 | if (ret) { | |
685 | pr_err("Failed to set PPL attributes for %s\n", | |
686 | sd->sys_name); | |
687 | close(dfd); | |
688 | st->ss->free_super(st); | |
689 | goto free_info; | |
690 | } | |
691 | ||
692 | ret = st->ss->write_init_ppl(st, &info, dfd); | |
693 | if (ret) | |
694 | pr_err("Failed to write PPL\n"); | |
695 | ||
696 | close(dfd); | |
697 | ||
698 | if (!subarray) | |
699 | st->ss->free_super(st); | |
700 | ||
701 | if (ret) | |
702 | goto free_info; | |
703 | } | |
704 | } | |
705 | ||
706 | ret = sysfs_set_str(sra, NULL, "consistency_policy", | |
5f21d674 MT |
707 | map_num_s(consistency_policies, |
708 | s->consistency_policy)); | |
860f11ed AP |
709 | if (ret) |
710 | pr_err("Failed to change array consistency policy\n"); | |
711 | ||
712 | free_info: | |
713 | sysfs_free(sra); | |
714 | free_st: | |
715 | free(st); | |
716 | free(subarray); | |
717 | ||
718 | return ret; | |
719 | } | |
720 | ||
e86c9dd6 NB |
721 | /* |
722 | * When reshaping an array we might need to backup some data. | |
723 | * This is written to all spares with a 'super_block' describing it. | |
ff94fb86 | 724 | * The superblock goes 4K from the end of the used space on the |
e86c9dd6 NB |
725 | * device. |
726 | * It if written after the backup is complete. | |
727 | * It has the following structure. | |
728 | */ | |
729 | ||
5fdf37e3 | 730 | static struct mdp_backup_super { |
7236ee7a | 731 | char magic[16]; /* md_backup_data-1 or -2 */ |
e86c9dd6 NB |
732 | __u8 set_uuid[16]; |
733 | __u64 mtime; | |
734 | /* start/sizes in 512byte sectors */ | |
7236ee7a | 735 | __u64 devstart; /* address on backup device/file of data */ |
e86c9dd6 NB |
736 | __u64 arraystart; |
737 | __u64 length; | |
738 | __u32 sb_csum; /* csum of preceeding bytes. */ | |
7236ee7a N |
739 | __u32 pad1; |
740 | __u64 devstart2; /* offset in to data of second section */ | |
741 | __u64 arraystart2; | |
742 | __u64 length2; | |
743 | __u32 sb_csum2; /* csum of preceeding bytes. */ | |
744 | __u8 pad[512-68-32]; | |
5fdf37e3 | 745 | } __attribute__((aligned(512))) bsb, bsb2; |
e86c9dd6 | 746 | |
4411fb17 | 747 | static __u32 bsb_csum(char *buf, int len) |
e86c9dd6 NB |
748 | { |
749 | int i; | |
750 | int csum = 0; | |
ca3b6696 | 751 | for (i = 0; i < len; i++) |
e86c9dd6 NB |
752 | csum = (csum<<3) + buf[0]; |
753 | return __cpu_to_le32(csum); | |
754 | } | |
755 | ||
d7ca196c N |
756 | static int check_idle(struct supertype *st) |
757 | { | |
758 | /* Check that all member arrays for this container, or the | |
759 | * container of this array, are idle | |
760 | */ | |
4dd2df09 N |
761 | char *container = (st->container_devnm[0] |
762 | ? st->container_devnm : st->devnm); | |
d7ca196c N |
763 | struct mdstat_ent *ent, *e; |
764 | int is_idle = 1; | |
24daa16f | 765 | |
d7ca196c N |
766 | ent = mdstat_read(0, 0); |
767 | for (e = ent ; e; e = e->next) { | |
768 | if (!is_container_member(e, container)) | |
769 | continue; | |
84d88fd8 MT |
770 | /* frozen array is not idle*/ |
771 | if (e->percent >= 0 || e->metadata_version[9] == '-') { | |
d7ca196c N |
772 | is_idle = 0; |
773 | break; | |
774 | } | |
775 | } | |
776 | free_mdstat(ent); | |
777 | return is_idle; | |
778 | } | |
779 | ||
7f2ba464 | 780 | static int freeze_container(struct supertype *st) |
7236ee7a | 781 | { |
4dd2df09 N |
782 | char *container = (st->container_devnm[0] |
783 | ? st->container_devnm : st->devnm); | |
7f2ba464 | 784 | |
d7ca196c N |
785 | if (!check_idle(st)) |
786 | return -1; | |
24daa16f | 787 | |
7f2ba464 | 788 | if (block_monitor(container, 1)) { |
e7b84f9d | 789 | pr_err("failed to freeze container\n"); |
7f2ba464 DW |
790 | return -2; |
791 | } | |
792 | ||
793 | return 1; | |
794 | } | |
795 | ||
796 | static void unfreeze_container(struct supertype *st) | |
797 | { | |
4dd2df09 N |
798 | char *container = (st->container_devnm[0] |
799 | ? st->container_devnm : st->devnm); | |
7f2ba464 DW |
800 | |
801 | unblock_monitor(container, 1); | |
802 | } | |
803 | ||
804 | static int freeze(struct supertype *st) | |
805 | { | |
806 | /* Try to freeze resync/rebuild on this array/container. | |
7236ee7a | 807 | * Return -1 if the array is busy, |
7f2ba464 | 808 | * return -2 container cannot be frozen, |
7236ee7a N |
809 | * return 0 if this kernel doesn't support 'frozen' |
810 | * return 1 if it worked. | |
811 | */ | |
7f2ba464 DW |
812 | if (st->ss->external) |
813 | return freeze_container(st); | |
814 | else { | |
4dd2df09 | 815 | struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION); |
7f2ba464 | 816 | int err; |
815c8a7e | 817 | char buf[20]; |
7f2ba464 DW |
818 | |
819 | if (!sra) | |
820 | return -1; | |
815c8a7e N |
821 | /* Need to clear any 'read-auto' status */ |
822 | if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 && | |
823 | strncmp(buf, "read-auto", 9) == 0) | |
824 | sysfs_set_str(sra, NULL, "array_state", "clean"); | |
825 | ||
7f2ba464 DW |
826 | err = sysfs_freeze_array(sra); |
827 | sysfs_free(sra); | |
828 | return err; | |
829 | } | |
7236ee7a N |
830 | } |
831 | ||
9202b8eb | 832 | static void unfreeze(struct supertype *st) |
7236ee7a | 833 | { |
7f2ba464 DW |
834 | if (st->ss->external) |
835 | return unfreeze_container(st); | |
836 | else { | |
4dd2df09 | 837 | struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION); |
5e7be838 | 838 | char buf[20]; |
7f2ba464 | 839 | |
5e7be838 | 840 | if (sra && |
fc54fe7a JS |
841 | sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 && |
842 | strcmp(buf, "frozen\n") == 0) | |
7f2ba464 | 843 | sysfs_set_str(sra, NULL, "sync_action", "idle"); |
7f2ba464 DW |
844 | sysfs_free(sra); |
845 | } | |
7236ee7a N |
846 | } |
847 | ||
4411fb17 | 848 | static void wait_reshape(struct mdinfo *sra) |
7236ee7a N |
849 | { |
850 | int fd = sysfs_get_fd(sra, NULL, "sync_action"); | |
851 | char action[20]; | |
852 | ||
92a19f1a AK |
853 | if (fd < 0) |
854 | return; | |
855 | ||
efc67e8e N |
856 | while (sysfs_fd_get_str(fd, action, 20) > 0 && |
857 | strncmp(action, "reshape", 7) == 0) | |
858 | sysfs_wait(fd, NULL); | |
92a19f1a | 859 | close(fd); |
7236ee7a | 860 | } |
7bc71196 | 861 | |
7ec0996c LD |
862 | static int reshape_super(struct supertype *st, unsigned long long size, |
863 | int level, int layout, int chunksize, int raid_disks, | |
41784c88 | 864 | int delta_disks, char *backup_file, char *dev, |
016e00f5 | 865 | int direction, int verbose) |
7bc71196 DW |
866 | { |
867 | /* nothing extra to check in the native case */ | |
868 | if (!st->ss->external) | |
869 | return 0; | |
358ef9bf | 870 | if (!st->ss->reshape_super || !st->ss->manage_reshape) { |
e7b84f9d | 871 | pr_err("%s metadata does not support reshape\n", |
7bc71196 DW |
872 | st->ss->name); |
873 | return 1; | |
874 | } | |
875 | ||
876 | return st->ss->reshape_super(st, size, level, layout, chunksize, | |
41784c88 | 877 | raid_disks, delta_disks, backup_file, dev, |
016e00f5 | 878 | direction, verbose); |
7bc71196 DW |
879 | } |
880 | ||
881 | static void sync_metadata(struct supertype *st) | |
882 | { | |
883 | if (st->ss->external) { | |
76266030 | 884 | if (st->update_tail) { |
7bc71196 | 885 | flush_metadata_updates(st); |
76266030 N |
886 | st->update_tail = &st->updates; |
887 | } else | |
7bc71196 DW |
888 | st->ss->sync_metadata(st); |
889 | } | |
890 | } | |
891 | ||
892 | static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n) | |
893 | { | |
894 | /* when dealing with external metadata subarrays we need to be | |
895 | * prepared to handle EAGAIN. The kernel may need to wait for | |
896 | * mdmon to mark the array active so the kernel can handle | |
897 | * allocations/writeback when preparing the reshape action | |
898 | * (md_allow_write()). We temporarily disable safe_mode_delay | |
899 | * to close a race with the array_state going clean before the | |
900 | * next write to raid_disks / stripe_cache_size | |
901 | */ | |
902 | char safe[50]; | |
903 | int rc; | |
904 | ||
905 | /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */ | |
5da9ab98 N |
906 | if (!container || |
907 | (strcmp(name, "raid_disks") != 0 && | |
908 | strcmp(name, "stripe_cache_size") != 0)) | |
7bc71196 DW |
909 | return sysfs_set_num(sra, NULL, name, n); |
910 | ||
911 | rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe)); | |
912 | if (rc <= 0) | |
913 | return -1; | |
914 | sysfs_set_num(sra, NULL, "safe_mode_delay", 0); | |
915 | rc = sysfs_set_num(sra, NULL, name, n); | |
916 | if (rc < 0 && errno == EAGAIN) { | |
917 | ping_monitor(container); | |
918 | /* if we get EAGAIN here then the monitor is not active | |
919 | * so stop trying | |
920 | */ | |
921 | rc = sysfs_set_num(sra, NULL, name, n); | |
922 | } | |
923 | sysfs_set_str(sra, NULL, "safe_mode_delay", safe); | |
924 | return rc; | |
925 | } | |
926 | ||
27a1e5b5 | 927 | int start_reshape(struct mdinfo *sra, int already_running, |
78c93b00 | 928 | int before_data_disks, int data_disks, struct supertype *st) |
47eb4d5a N |
929 | { |
930 | int err; | |
6937e6d2 AK |
931 | unsigned long long sync_max_to_set; |
932 | ||
0f28668b | 933 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); |
6937e6d2 AK |
934 | err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress); |
935 | err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", | |
936 | sra->reshape_progress); | |
27a1e5b5 N |
937 | if (before_data_disks <= data_disks) |
938 | sync_max_to_set = sra->reshape_progress / data_disks; | |
939 | else | |
940 | sync_max_to_set = (sra->component_size * data_disks | |
941 | - sra->reshape_progress) / data_disks; | |
78c93b00 | 942 | |
1e971e71 | 943 | if (!already_running) |
6937e6d2 | 944 | sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set); |
78c93b00 NC |
945 | |
946 | if (st->ss->external) | |
947 | err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set); | |
948 | else | |
949 | err = err ?: sysfs_set_str(sra, NULL, "sync_max", "max"); | |
950 | ||
ada38ebb N |
951 | if (!already_running && err == 0) { |
952 | int cnt = 5; | |
953 | do { | |
358ef9bf JS |
954 | err = sysfs_set_str(sra, NULL, "sync_action", |
955 | "reshape"); | |
78c93b00 | 956 | if (err) |
239b3cc0 | 957 | sleep_for(1, 0, true); |
78c93b00 | 958 | } while (err && errno == EBUSY && cnt-- > 0); |
ada38ebb | 959 | } |
47eb4d5a N |
960 | return err; |
961 | } | |
962 | ||
963 | void abort_reshape(struct mdinfo *sra) | |
964 | { | |
965 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
bf08f6b1 JS |
966 | /* |
967 | * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and | |
968 | * suspend_hi to decrease as well as increase.") | |
969 | * you could only increase suspend_{lo,hi} unless the region they | |
970 | * covered was empty. So to reset to 0, you need to push suspend_lo | |
971 | * up past suspend_hi first. So to maximize the chance of mdadm | |
972 | * working on all kernels, we want to keep doing that. | |
973 | */ | |
47eb4d5a N |
974 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); |
975 | sysfs_set_num(sra, NULL, "suspend_hi", 0); | |
976 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
977 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
84d11e6c N |
978 | // It isn't safe to reset sync_max as we aren't monitoring. |
979 | // Array really should be stopped at this point. | |
47eb4d5a N |
980 | } |
981 | ||
dfe77a9e KW |
982 | int remove_disks_for_takeover(struct supertype *st, |
983 | struct mdinfo *sra, | |
984 | int layout) | |
62a48395 AK |
985 | { |
986 | int nr_of_copies; | |
987 | struct mdinfo *remaining; | |
988 | int slot; | |
989 | ||
ddab63c7 MD |
990 | if (st->ss->external) { |
991 | int rv = 0; | |
992 | struct mdinfo *arrays = st->ss->container_content(st, NULL); | |
993 | /* | |
994 | * containter_content returns list of arrays in container | |
995 | * If arrays->next is not NULL it means that there are | |
996 | * 2 arrays in container and operation should be blocked | |
997 | */ | |
998 | if (arrays) { | |
999 | if (arrays->next) | |
1000 | rv = 1; | |
1001 | sysfs_free(arrays); | |
1002 | if (rv) { | |
f1cc8ab9 LF |
1003 | pr_err("Error. Cannot perform operation on %s- for this operation " |
1004 | "it MUST be single array in container\n", st->devnm); | |
ddab63c7 MD |
1005 | return rv; |
1006 | } | |
1007 | } | |
1008 | } | |
1009 | ||
dfe77a9e KW |
1010 | if (sra->array.level == 10) |
1011 | nr_of_copies = layout & 0xff; | |
1012 | else if (sra->array.level == 1) | |
1013 | nr_of_copies = sra->array.raid_disks; | |
1014 | else | |
1015 | return 1; | |
62a48395 AK |
1016 | |
1017 | remaining = sra->devs; | |
1018 | sra->devs = NULL; | |
1019 | /* for each 'copy', select one device and remove from the list. */ | |
1020 | for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) { | |
1021 | struct mdinfo **diskp; | |
1022 | int found = 0; | |
1023 | ||
1024 | /* Find a working device to keep */ | |
1025 | for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) { | |
1026 | struct mdinfo *disk = *diskp; | |
1027 | ||
1028 | if (disk->disk.raid_disk < slot) | |
1029 | continue; | |
1030 | if (disk->disk.raid_disk >= slot + nr_of_copies) | |
1031 | continue; | |
1032 | if (disk->disk.state & (1<<MD_DISK_REMOVED)) | |
1033 | continue; | |
1034 | if (disk->disk.state & (1<<MD_DISK_FAULTY)) | |
1035 | continue; | |
1036 | if (!(disk->disk.state & (1<<MD_DISK_SYNC))) | |
1037 | continue; | |
1038 | ||
1039 | /* We have found a good disk to use! */ | |
1040 | *diskp = disk->next; | |
1041 | disk->next = sra->devs; | |
1042 | sra->devs = disk; | |
1043 | found = 1; | |
1044 | break; | |
1045 | } | |
1046 | if (!found) | |
1047 | break; | |
1048 | } | |
1049 | ||
1050 | if (slot < sra->array.raid_disks) { | |
1051 | /* didn't find all slots */ | |
1052 | struct mdinfo **e; | |
1053 | e = &remaining; | |
1054 | while (*e) | |
1055 | e = &(*e)->next; | |
1056 | *e = sra->devs; | |
1057 | sra->devs = remaining; | |
1058 | return 1; | |
1059 | } | |
1060 | ||
1061 | /* Remove all 'remaining' devices from the array */ | |
1062 | while (remaining) { | |
1063 | struct mdinfo *sd = remaining; | |
1064 | remaining = sd->next; | |
1065 | ||
1066 | sysfs_set_str(sra, sd, "state", "faulty"); | |
1067 | sysfs_set_str(sra, sd, "slot", "none"); | |
d5ca4a23 KW |
1068 | /* for external metadata disks should be removed in mdmon */ |
1069 | if (!st->ss->external) | |
1070 | sysfs_set_str(sra, sd, "state", "remove"); | |
62a48395 AK |
1071 | sd->disk.state |= (1<<MD_DISK_REMOVED); |
1072 | sd->disk.state &= ~(1<<MD_DISK_SYNC); | |
1073 | sd->next = sra->devs; | |
1074 | sra->devs = sd; | |
1075 | } | |
1076 | return 0; | |
1077 | } | |
1078 | ||
130994cb AK |
1079 | void reshape_free_fdlist(int *fdlist, |
1080 | unsigned long long *offsets, | |
1081 | int size) | |
1082 | { | |
1083 | int i; | |
1084 | ||
1085 | for (i = 0; i < size; i++) | |
1086 | if (fdlist[i] >= 0) | |
1087 | close(fdlist[i]); | |
1088 | ||
1089 | free(fdlist); | |
1090 | free(offsets); | |
1091 | } | |
1092 | ||
1093 | int reshape_prepare_fdlist(char *devname, | |
1094 | struct mdinfo *sra, | |
1095 | int raid_disks, | |
1096 | int nrdisks, | |
1097 | unsigned long blocks, | |
1098 | char *backup_file, | |
1099 | int *fdlist, | |
1100 | unsigned long long *offsets) | |
1101 | { | |
1102 | int d = 0; | |
1103 | struct mdinfo *sd; | |
1104 | ||
a7dec3fd | 1105 | enable_fds(nrdisks); |
130994cb AK |
1106 | for (d = 0; d <= nrdisks; d++) |
1107 | fdlist[d] = -1; | |
1108 | d = raid_disks; | |
1109 | for (sd = sra->devs; sd; sd = sd->next) { | |
1110 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
1111 | continue; | |
8e7ddc5f N |
1112 | if (sd->disk.state & (1<<MD_DISK_SYNC) && |
1113 | sd->disk.raid_disk < raid_disks) { | |
358ef9bf JS |
1114 | char *dn = map_dev(sd->disk.major, sd->disk.minor, 1); |
1115 | fdlist[sd->disk.raid_disk] = dev_open(dn, O_RDONLY); | |
130994cb AK |
1116 | offsets[sd->disk.raid_disk] = sd->data_offset*512; |
1117 | if (fdlist[sd->disk.raid_disk] < 0) { | |
e7b84f9d N |
1118 | pr_err("%s: cannot open component %s\n", |
1119 | devname, dn ? dn : "-unknown-"); | |
130994cb AK |
1120 | d = -1; |
1121 | goto release; | |
1122 | } | |
1123 | } else if (backup_file == NULL) { | |
1124 | /* spare */ | |
358ef9bf | 1125 | char *dn = map_dev(sd->disk.major, sd->disk.minor, 1); |
130994cb AK |
1126 | fdlist[d] = dev_open(dn, O_RDWR); |
1127 | offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512; | |
1128 | if (fdlist[d] < 0) { | |
e7b84f9d | 1129 | pr_err("%s: cannot open component %s\n", |
130994cb AK |
1130 | devname, dn ? dn : "-unknown-"); |
1131 | d = -1; | |
1132 | goto release; | |
1133 | } | |
1134 | d++; | |
1135 | } | |
1136 | } | |
1137 | release: | |
1138 | return d; | |
1139 | } | |
1140 | ||
e6e9d47b AK |
1141 | int reshape_open_backup_file(char *backup_file, |
1142 | int fd, | |
1143 | char *devname, | |
1144 | long blocks, | |
1145 | int *fdlist, | |
a93f87ee | 1146 | unsigned long long *offsets, |
54ded86f | 1147 | char *sys_name, |
a93f87ee | 1148 | int restart) |
e6e9d47b AK |
1149 | { |
1150 | /* Return 1 on success, 0 on any form of failure */ | |
1151 | /* need to check backup file is large enough */ | |
1152 | char buf[512]; | |
1153 | struct stat stb; | |
1154 | unsigned int dev; | |
1155 | int i; | |
1156 | ||
a93f87ee | 1157 | *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL), |
e6e9d47b AK |
1158 | S_IRUSR | S_IWUSR); |
1159 | *offsets = 8 * 512; | |
1160 | if (*fdlist < 0) { | |
e7b84f9d | 1161 | pr_err("%s: cannot create backup file %s: %s\n", |
e6e9d47b AK |
1162 | devname, backup_file, strerror(errno)); |
1163 | return 0; | |
1164 | } | |
1165 | /* Guard against backup file being on array device. | |
1166 | * If array is partitioned or if LVM etc is in the | |
1167 | * way this will not notice, but it is better than | |
1168 | * nothing. | |
1169 | */ | |
1170 | fstat(*fdlist, &stb); | |
1171 | dev = stb.st_dev; | |
1172 | fstat(fd, &stb); | |
1173 | if (stb.st_rdev == dev) { | |
7a862a02 | 1174 | pr_err("backup file must NOT be on the array being reshaped.\n"); |
e6e9d47b AK |
1175 | close(*fdlist); |
1176 | return 0; | |
1177 | } | |
1178 | ||
1179 | memset(buf, 0, 512); | |
ca4fe0bf | 1180 | for (i=0; i < blocks + 8 ; i++) { |
e6e9d47b | 1181 | if (write(*fdlist, buf, 512) != 512) { |
7a862a02 | 1182 | pr_err("%s: cannot create backup file %s: %s\n", |
e6e9d47b AK |
1183 | devname, backup_file, strerror(errno)); |
1184 | return 0; | |
1185 | } | |
1186 | } | |
1187 | if (fsync(*fdlist) != 0) { | |
e7b84f9d | 1188 | pr_err("%s: cannot create backup file %s: %s\n", |
e6e9d47b AK |
1189 | devname, backup_file, strerror(errno)); |
1190 | return 0; | |
1191 | } | |
1192 | ||
54ded86f N |
1193 | if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) { |
1194 | char *bu = make_backup(sys_name); | |
1e60caeb | 1195 | if (symlink(backup_file, bu)) |
a5a6a7d9 | 1196 | pr_err("Recording backup file in " MAP_DIR " failed: %s\n", |
1e60caeb | 1197 | strerror(errno)); |
54ded86f N |
1198 | free(bu); |
1199 | } | |
1200 | ||
e6e9d47b AK |
1201 | return 1; |
1202 | } | |
1203 | ||
1c009fc2 AK |
1204 | unsigned long compute_backup_blocks(int nchunk, int ochunk, |
1205 | unsigned int ndata, unsigned int odata) | |
1206 | { | |
1207 | unsigned long a, b, blocks; | |
1208 | /* So how much do we need to backup. | |
1209 | * We need an amount of data which is both a whole number of | |
1210 | * old stripes and a whole number of new stripes. | |
1211 | * So LCM for (chunksize*datadisks). | |
1212 | */ | |
1213 | a = (ochunk/512) * odata; | |
1214 | b = (nchunk/512) * ndata; | |
1215 | /* Find GCD */ | |
b48e2e25 | 1216 | a = GCD(a, b); |
1c009fc2 | 1217 | /* LCM == product / GCD */ |
085df422 N |
1218 | blocks = (unsigned long)(ochunk/512) * (unsigned long)(nchunk/512) * |
1219 | odata * ndata / a; | |
1c009fc2 AK |
1220 | |
1221 | return blocks; | |
1222 | } | |
1223 | ||
a73b0081 | 1224 | char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re) |
5da9ab98 N |
1225 | { |
1226 | /* Based on the current array state in info->array and | |
1227 | * the changes in info->new_* etc, determine: | |
1228 | * - whether the change is possible | |
1229 | * - Intermediate level/raid_disks/layout | |
1230 | * - whether a restriping reshape is needed | |
1231 | * - number of sectors in minimum change unit. This | |
1232 | * will cover a whole number of stripes in 'before' and | |
1233 | * 'after'. | |
1234 | * | |
1235 | * Return message if the change should be rejected | |
1236 | * NULL if the change can be achieved | |
1237 | * | |
1238 | * This can be called as part of starting a reshape, or | |
1239 | * when assembling an array that is undergoing reshape. | |
1240 | */ | |
19ceb16d | 1241 | int near, far, offset, copies; |
5da9ab98 | 1242 | int new_disks; |
19ceb16d | 1243 | int old_chunk, new_chunk; |
b545e14a N |
1244 | /* delta_parity records change in number of devices |
1245 | * caused by level change | |
1246 | */ | |
1247 | int delta_parity = 0; | |
5da9ab98 | 1248 | |
6b2fc3c1 N |
1249 | memset(re, 0, sizeof(*re)); |
1250 | ||
5da9ab98 N |
1251 | /* If a new level not explicitly given, we assume no-change */ |
1252 | if (info->new_level == UnSet) | |
1253 | info->new_level = info->array.level; | |
1254 | ||
1255 | if (info->new_chunk) | |
1256 | switch (info->new_level) { | |
1257 | case 0: | |
1258 | case 4: | |
1259 | case 5: | |
1260 | case 6: | |
1261 | case 10: | |
1262 | /* chunk size is meaningful, must divide component_size | |
1263 | * evenly | |
1264 | */ | |
a73b0081 N |
1265 | if (info->component_size % (info->new_chunk/512)) { |
1266 | unsigned long long shrink = info->component_size; | |
1267 | shrink &= ~(unsigned long long)(info->new_chunk/512-1); | |
1268 | pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n", | |
1269 | info->new_chunk/1024, info->component_size/2); | |
1270 | pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n", | |
1271 | devname, shrink/2); | |
1272 | pr_err("will shrink the array so the given chunk size would work.\n"); | |
1273 | return ""; | |
1274 | } | |
5da9ab98 N |
1275 | break; |
1276 | default: | |
1277 | return "chunk size not meaningful for this level"; | |
1278 | } | |
1279 | else | |
1280 | info->new_chunk = info->array.chunk_size; | |
1281 | ||
1282 | switch (info->array.level) { | |
90b60dfa | 1283 | default: |
46643e1a N |
1284 | return "No reshape is possibly for this RAID level"; |
1285 | case LEVEL_LINEAR: | |
1286 | if (info->delta_disks != UnSet) | |
1287 | return "Only --add is supported for LINEAR, setting --raid-disks is not needed"; | |
1288 | else | |
1289 | return "Only --add is supported for LINEAR, other --grow options are not meaningful"; | |
5da9ab98 N |
1290 | case 1: |
1291 | /* RAID1 can convert to RAID1 with different disks, or | |
dfe77a9e KW |
1292 | * raid5 with 2 disks, or |
1293 | * raid0 with 1 disk | |
5da9ab98 | 1294 | */ |
6ebf34e6 | 1295 | if (info->new_level > 1 && (info->component_size & 7)) |
7a862a02 | 1296 | return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first."; |
dfe77a9e | 1297 | if (info->new_level == 0) { |
b545e14a N |
1298 | if (info->delta_disks != UnSet && |
1299 | info->delta_disks != 0) | |
7a862a02 | 1300 | return "Cannot change number of disks with RAID1->RAID0 conversion"; |
dfe77a9e KW |
1301 | re->level = 0; |
1302 | re->before.data_disks = 1; | |
1303 | re->after.data_disks = 1; | |
dfe77a9e KW |
1304 | return NULL; |
1305 | } | |
5da9ab98 N |
1306 | if (info->new_level == 1) { |
1307 | if (info->delta_disks == UnSet) | |
1308 | /* Don't know what to do */ | |
1309 | return "no change requested for Growing RAID1"; | |
1310 | re->level = 1; | |
5da9ab98 N |
1311 | return NULL; |
1312 | } | |
6ebf34e6 | 1313 | if (info->array.raid_disks != 2 && info->new_level == 5) |
a58e0da4 | 1314 | return "Can only convert a 2-device array to RAID5"; |
6ebf34e6 | 1315 | if (info->array.raid_disks == 2 && info->new_level == 5) { |
5da9ab98 | 1316 | re->level = 5; |
5da9ab98 | 1317 | re->before.data_disks = 1; |
5652f2d9 N |
1318 | if (info->delta_disks != UnSet && |
1319 | info->delta_disks != 0) | |
1320 | re->after.data_disks = 1 + info->delta_disks; | |
1321 | else | |
1322 | re->after.data_disks = 1; | |
1323 | if (re->after.data_disks < 1) | |
1324 | return "Number of disks too small for RAID5"; | |
1325 | ||
5da9ab98 | 1326 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC; |
446d2a5a N |
1327 | info->array.chunk_size = 65536; |
1328 | break; | |
5da9ab98 N |
1329 | } |
1330 | /* Could do some multi-stage conversions, but leave that to | |
1331 | * later. | |
1332 | */ | |
1333 | return "Impossibly level change request for RAID1"; | |
1334 | ||
1335 | case 10: | |
19ceb16d N |
1336 | /* RAID10 can be converted from near mode to |
1337 | * RAID0 by removing some devices. | |
1338 | * It can also be reshaped if the kernel supports | |
1339 | * new_data_offset. | |
5da9ab98 | 1340 | */ |
19ceb16d N |
1341 | switch (info->new_level) { |
1342 | case 0: | |
1343 | if ((info->array.layout & ~0xff) != 0x100) | |
1344 | return "Cannot Grow RAID10 with far/offset layout"; | |
358ef9bf JS |
1345 | /* |
1346 | * number of devices must be multiple of | |
1347 | * number of copies | |
1348 | */ | |
1349 | if (info->array.raid_disks % | |
1350 | (info->array.layout & 0xff)) | |
19ceb16d N |
1351 | return "RAID10 layout too complex for Grow operation"; |
1352 | ||
358ef9bf JS |
1353 | new_disks = (info->array.raid_disks / |
1354 | (info->array.layout & 0xff)); | |
19ceb16d N |
1355 | if (info->delta_disks == UnSet) |
1356 | info->delta_disks = (new_disks | |
1357 | - info->array.raid_disks); | |
5da9ab98 | 1358 | |
358ef9bf JS |
1359 | if (info->delta_disks != |
1360 | new_disks - info->array.raid_disks) | |
19ceb16d N |
1361 | return "New number of raid-devices impossible for RAID10"; |
1362 | if (info->new_chunk && | |
1363 | info->new_chunk != info->array.chunk_size) | |
1364 | return "Cannot change chunk-size with RAID10 Grow"; | |
1365 | ||
1366 | /* looks good */ | |
1367 | re->level = 0; | |
19ceb16d N |
1368 | re->before.data_disks = new_disks; |
1369 | re->after.data_disks = re->before.data_disks; | |
19ceb16d N |
1370 | return NULL; |
1371 | ||
1372 | case 10: | |
1373 | near = info->array.layout & 0xff; | |
1374 | far = (info->array.layout >> 8) & 0xff; | |
1375 | offset = info->array.layout & 0x10000; | |
1376 | if (far > 1 && !offset) | |
1377 | return "Cannot reshape RAID10 in far-mode"; | |
1378 | copies = near * far; | |
1379 | ||
1380 | old_chunk = info->array.chunk_size * far; | |
1381 | ||
1382 | if (info->new_layout == UnSet) | |
1383 | info->new_layout = info->array.layout; | |
1384 | else { | |
1385 | near = info->new_layout & 0xff; | |
1386 | far = (info->new_layout >> 8) & 0xff; | |
1387 | offset = info->new_layout & 0x10000; | |
1388 | if (far > 1 && !offset) | |
1389 | return "Cannot reshape RAID10 to far-mode"; | |
1390 | if (near * far != copies) | |
7a862a02 | 1391 | return "Cannot change number of copies when reshaping RAID10"; |
19ceb16d N |
1392 | } |
1393 | if (info->delta_disks == UnSet) | |
1394 | info->delta_disks = 0; | |
1395 | new_disks = (info->array.raid_disks + | |
1396 | info->delta_disks); | |
1397 | ||
1398 | new_chunk = info->new_chunk * far; | |
1399 | ||
1400 | re->level = 10; | |
19ceb16d N |
1401 | re->before.layout = info->array.layout; |
1402 | re->before.data_disks = info->array.raid_disks; | |
1403 | re->after.layout = info->new_layout; | |
1404 | re->after.data_disks = new_disks; | |
89ecd3cf N |
1405 | /* For RAID10 we don't do backup but do allow reshape, |
1406 | * so set backup_blocks to INVALID_SECTORS rather than | |
1407 | * zero. | |
1408 | * And there is no need to synchronise stripes on both | |
19ceb16d N |
1409 | * 'old' and 'new'. So the important |
1410 | * number is the minimum data_offset difference | |
1411 | * which is the larger of (offset copies * chunk). | |
1412 | */ | |
89ecd3cf N |
1413 | re->backup_blocks = INVALID_SECTORS; |
1414 | re->min_offset_change = max(old_chunk, new_chunk) / 512; | |
ee2429e0 | 1415 | if (new_disks < re->before.data_disks && |
89ecd3cf | 1416 | info->space_after < re->min_offset_change) |
ee2429e0 N |
1417 | /* Reduce component size by one chunk */ |
1418 | re->new_size = (info->component_size - | |
89ecd3cf | 1419 | re->min_offset_change); |
ee2429e0 N |
1420 | else |
1421 | re->new_size = info->component_size; | |
1422 | re->new_size = re->new_size * new_disks / copies; | |
19ceb16d | 1423 | return NULL; |
5da9ab98 | 1424 | |
19ceb16d N |
1425 | default: |
1426 | return "RAID10 can only be changed to RAID0"; | |
1427 | } | |
5da9ab98 N |
1428 | case 0: |
1429 | /* RAID0 can be converted to RAID10, or to RAID456 */ | |
1430 | if (info->new_level == 10) { | |
6ebf34e6 JS |
1431 | if (info->new_layout == UnSet && |
1432 | info->delta_disks == UnSet) { | |
5da9ab98 N |
1433 | /* Assume near=2 layout */ |
1434 | info->new_layout = 0x102; | |
1435 | info->delta_disks = info->array.raid_disks; | |
1436 | } | |
1437 | if (info->new_layout == UnSet) { | |
1438 | int copies = 1 + (info->delta_disks | |
1439 | / info->array.raid_disks); | |
d16a7494 JS |
1440 | if (info->array.raid_disks * (copies-1) != |
1441 | info->delta_disks) | |
7a862a02 | 1442 | return "Impossible number of devices for RAID0->RAID10"; |
5da9ab98 N |
1443 | info->new_layout = 0x100 + copies; |
1444 | } | |
1445 | if (info->delta_disks == UnSet) { | |
1446 | int copies = info->new_layout & 0xff; | |
1447 | if (info->new_layout != 0x100 + copies) | |
7a862a02 | 1448 | return "New layout impossible for RAID0->RAID10";; |
5da9ab98 N |
1449 | info->delta_disks = (copies - 1) * |
1450 | info->array.raid_disks; | |
1451 | } | |
1452 | if (info->new_chunk && | |
1453 | info->new_chunk != info->array.chunk_size) | |
1454 | return "Cannot change chunk-size with RAID0->RAID10"; | |
1455 | /* looks good */ | |
1456 | re->level = 10; | |
5da9ab98 N |
1457 | re->before.data_disks = (info->array.raid_disks + |
1458 | info->delta_disks); | |
031d445c | 1459 | re->after.data_disks = re->before.data_disks; |
5da9ab98 | 1460 | re->before.layout = info->new_layout; |
5da9ab98 N |
1461 | return NULL; |
1462 | } | |
1463 | ||
1464 | /* RAID0 can also covert to RAID0/4/5/6 by first converting to | |
1465 | * a raid4 style layout of the final level. | |
1466 | */ | |
1467 | switch (info->new_level) { | |
5da9ab98 | 1468 | case 4: |
b545e14a N |
1469 | delta_parity = 1; |
1470 | case 0: | |
5da9ab98 N |
1471 | re->level = 4; |
1472 | re->before.layout = 0; | |
1473 | break; | |
1474 | case 5: | |
b545e14a | 1475 | delta_parity = 1; |
5da9ab98 N |
1476 | re->level = 5; |
1477 | re->before.layout = ALGORITHM_PARITY_N; | |
e5ba75ce N |
1478 | if (info->new_layout == UnSet) |
1479 | info->new_layout = map_name(r5layout, "default"); | |
5da9ab98 N |
1480 | break; |
1481 | case 6: | |
b545e14a | 1482 | delta_parity = 2; |
5da9ab98 N |
1483 | re->level = 6; |
1484 | re->before.layout = ALGORITHM_PARITY_N; | |
e5ba75ce N |
1485 | if (info->new_layout == UnSet) |
1486 | info->new_layout = map_name(r6layout, "default"); | |
5da9ab98 N |
1487 | break; |
1488 | default: | |
1489 | return "Impossible level change requested"; | |
1490 | } | |
1491 | re->before.data_disks = info->array.raid_disks; | |
1492 | /* determining 'after' layout happens outside this 'switch' */ | |
1493 | break; | |
1494 | ||
1495 | case 4: | |
1496 | info->array.layout = ALGORITHM_PARITY_N; | |
1497 | case 5: | |
1498 | switch (info->new_level) { | |
f0cce442 | 1499 | case 0: |
b545e14a | 1500 | delta_parity = -1; |
5da9ab98 N |
1501 | case 4: |
1502 | re->level = info->array.level; | |
1503 | re->before.data_disks = info->array.raid_disks - 1; | |
1504 | re->before.layout = info->array.layout; | |
1505 | break; | |
1506 | case 5: | |
1507 | re->level = 5; | |
1508 | re->before.data_disks = info->array.raid_disks - 1; | |
1509 | re->before.layout = info->array.layout; | |
1510 | break; | |
1511 | case 6: | |
b545e14a | 1512 | delta_parity = 1; |
5da9ab98 N |
1513 | re->level = 6; |
1514 | re->before.data_disks = info->array.raid_disks - 1; | |
1515 | switch (info->array.layout) { | |
1516 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1517 | re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
1518 | break; | |
1519 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1520 | re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
1521 | break; | |
1522 | case ALGORITHM_LEFT_SYMMETRIC: | |
1523 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
1524 | break; | |
1525 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1526 | re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
1527 | break; | |
1528 | case ALGORITHM_PARITY_0: | |
1529 | re->before.layout = ALGORITHM_PARITY_0_6; | |
1530 | break; | |
1531 | case ALGORITHM_PARITY_N: | |
1532 | re->before.layout = ALGORITHM_PARITY_N_6; | |
1533 | break; | |
1534 | default: | |
1535 | return "Cannot convert an array with this layout"; | |
1536 | } | |
1537 | break; | |
1538 | case 1: | |
1539 | if (info->array.raid_disks != 2) | |
1540 | return "Can only convert a 2-device array to RAID1"; | |
b545e14a N |
1541 | if (info->delta_disks != UnSet && |
1542 | info->delta_disks != 0) | |
7a862a02 | 1543 | return "Cannot set raid_disk when converting RAID5->RAID1"; |
5da9ab98 | 1544 | re->level = 1; |
6a67848a N |
1545 | info->new_chunk = 0; |
1546 | return NULL; | |
5da9ab98 N |
1547 | default: |
1548 | return "Impossible level change requested"; | |
1549 | } | |
1550 | break; | |
1551 | case 6: | |
1552 | switch (info->new_level) { | |
1553 | case 4: | |
1554 | case 5: | |
b545e14a | 1555 | delta_parity = -1; |
5da9ab98 N |
1556 | case 6: |
1557 | re->level = 6; | |
1558 | re->before.data_disks = info->array.raid_disks - 2; | |
1559 | re->before.layout = info->array.layout; | |
1560 | break; | |
1561 | default: | |
1562 | return "Impossible level change requested"; | |
1563 | } | |
1564 | break; | |
1565 | } | |
1566 | ||
1567 | /* If we reached here then it looks like a re-stripe is | |
1568 | * happening. We have determined the intermediate level | |
1569 | * and initial raid_disks/layout and stored these in 're'. | |
1570 | * | |
1571 | * We need to deduce the final layout that can be atomically | |
1572 | * converted to the end state. | |
1573 | */ | |
1574 | switch (info->new_level) { | |
1575 | case 0: | |
1576 | /* We can only get to RAID0 from RAID4 or RAID5 | |
1577 | * with appropriate layout and one extra device | |
1578 | */ | |
1579 | if (re->level != 4 && re->level != 5) | |
1580 | return "Cannot covert to RAID0 from this level"; | |
b545e14a | 1581 | |
5da9ab98 N |
1582 | switch (re->level) { |
1583 | case 4: | |
e339dba2 | 1584 | re->before.layout = 0; |
ca3b6696 N |
1585 | re->after.layout = 0; |
1586 | break; | |
5da9ab98 | 1587 | case 5: |
ca3b6696 N |
1588 | re->after.layout = ALGORITHM_PARITY_N; |
1589 | break; | |
5da9ab98 N |
1590 | } |
1591 | break; | |
1592 | ||
1593 | case 4: | |
1594 | /* We can only get to RAID4 from RAID5 */ | |
1595 | if (re->level != 4 && re->level != 5) | |
1596 | return "Cannot convert to RAID4 from this level"; | |
b545e14a | 1597 | |
5da9ab98 N |
1598 | switch (re->level) { |
1599 | case 4: | |
ca3b6696 N |
1600 | re->after.layout = 0; |
1601 | break; | |
5da9ab98 | 1602 | case 5: |
ca3b6696 N |
1603 | re->after.layout = ALGORITHM_PARITY_N; |
1604 | break; | |
5da9ab98 N |
1605 | } |
1606 | break; | |
1607 | ||
1608 | case 5: | |
0073a6e1 | 1609 | /* We get to RAID5 from RAID5 or RAID6 */ |
5da9ab98 N |
1610 | if (re->level != 5 && re->level != 6) |
1611 | return "Cannot convert to RAID5 from this level"; | |
b545e14a | 1612 | |
5da9ab98 N |
1613 | switch (re->level) { |
1614 | case 5: | |
1615 | if (info->new_layout == UnSet) | |
1616 | re->after.layout = re->before.layout; | |
1617 | else | |
1618 | re->after.layout = info->new_layout; | |
1619 | break; | |
1620 | case 6: | |
4a870364 N |
1621 | if (info->new_layout == UnSet) |
1622 | info->new_layout = re->before.layout; | |
1623 | ||
5da9ab98 N |
1624 | /* after.layout needs to be raid6 version of new_layout */ |
1625 | if (info->new_layout == ALGORITHM_PARITY_N) | |
1626 | re->after.layout = ALGORITHM_PARITY_N; | |
1627 | else { | |
1628 | char layout[40]; | |
1629 | char *ls = map_num(r5layout, info->new_layout); | |
1630 | int l; | |
0073a6e1 N |
1631 | if (ls) { |
1632 | /* Current RAID6 layout has a RAID5 | |
1633 | * equivalent - good | |
1634 | */ | |
1635 | strcat(strcpy(layout, ls), "-6"); | |
1636 | l = map_name(r6layout, layout); | |
1637 | if (l == UnSet) | |
7a862a02 | 1638 | return "Cannot find RAID6 layout to convert to"; |
0073a6e1 N |
1639 | } else { |
1640 | /* Current RAID6 has no equivalent. | |
1641 | * If it is already a '-6' layout we | |
1642 | * can leave it unchanged, else we must | |
1643 | * fail | |
1644 | */ | |
358ef9bf JS |
1645 | ls = map_num(r6layout, |
1646 | info->new_layout); | |
0073a6e1 N |
1647 | if (!ls || |
1648 | strcmp(ls+strlen(ls)-2, "-6") != 0) | |
1649 | return "Please specify new layout"; | |
1650 | l = info->new_layout; | |
1651 | } | |
5da9ab98 N |
1652 | re->after.layout = l; |
1653 | } | |
1654 | } | |
1655 | break; | |
1656 | ||
1657 | case 6: | |
1658 | /* We must already be at level 6 */ | |
1659 | if (re->level != 6) | |
1660 | return "Impossible level change"; | |
5da9ab98 | 1661 | if (info->new_layout == UnSet) |
4a870364 | 1662 | re->after.layout = info->array.layout; |
5da9ab98 N |
1663 | else |
1664 | re->after.layout = info->new_layout; | |
1665 | break; | |
1666 | default: | |
1667 | return "Impossible level change requested"; | |
1668 | } | |
b545e14a N |
1669 | if (info->delta_disks == UnSet) |
1670 | info->delta_disks = delta_parity; | |
1671 | ||
6ebf34e6 JS |
1672 | re->after.data_disks = |
1673 | (re->before.data_disks + info->delta_disks - delta_parity); | |
1674 | ||
5da9ab98 | 1675 | switch (re->level) { |
6ebf34e6 JS |
1676 | case 6: |
1677 | re->parity = 2; | |
ca3b6696 | 1678 | break; |
5da9ab98 | 1679 | case 4: |
6ebf34e6 JS |
1680 | case 5: |
1681 | re->parity = 1; | |
ca3b6696 | 1682 | break; |
6ebf34e6 JS |
1683 | default: |
1684 | re->parity = 0; | |
ca3b6696 | 1685 | break; |
5da9ab98 N |
1686 | } |
1687 | /* So we have a restripe operation, we need to calculate the number | |
1688 | * of blocks per reshape operation. | |
1689 | */ | |
ddbf2ebb | 1690 | re->new_size = info->component_size * re->before.data_disks; |
5da9ab98 N |
1691 | if (info->new_chunk == 0) |
1692 | info->new_chunk = info->array.chunk_size; | |
1693 | if (re->after.data_disks == re->before.data_disks && | |
1694 | re->after.layout == re->before.layout && | |
1695 | info->new_chunk == info->array.chunk_size) { | |
fdcad551 N |
1696 | /* Nothing to change, can change level immediately. */ |
1697 | re->level = info->new_level; | |
b4f8e38b | 1698 | re->backup_blocks = 0; |
5da9ab98 N |
1699 | return NULL; |
1700 | } | |
1701 | if (re->after.data_disks == 1 && re->before.data_disks == 1) { | |
446d2a5a | 1702 | /* chunk and layout changes make no difference */ |
fdcad551 | 1703 | re->level = info->new_level; |
b4f8e38b | 1704 | re->backup_blocks = 0; |
5da9ab98 N |
1705 | return NULL; |
1706 | } | |
1707 | ||
1708 | if (re->after.data_disks == re->before.data_disks && | |
1709 | get_linux_version() < 2006032) | |
1710 | return "in-place reshape is not safe before 2.6.32 - sorry."; | |
1711 | ||
1712 | if (re->after.data_disks < re->before.data_disks && | |
1713 | get_linux_version() < 2006030) | |
508ede86 | 1714 | return "reshape to fewer devices is not supported before 2.6.30 - sorry."; |
5da9ab98 | 1715 | |
b4f8e38b | 1716 | re->backup_blocks = compute_backup_blocks( |
5da9ab98 | 1717 | info->new_chunk, info->array.chunk_size, |
358ef9bf | 1718 | re->after.data_disks, re->before.data_disks); |
63c12c89 | 1719 | re->min_offset_change = re->backup_blocks / re->before.data_disks; |
5da9ab98 N |
1720 | |
1721 | re->new_size = info->component_size * re->after.data_disks; | |
1722 | return NULL; | |
1723 | } | |
1724 | ||
54397ed9 AK |
1725 | static int set_array_size(struct supertype *st, struct mdinfo *sra, |
1726 | char *text_version) | |
1727 | { | |
1728 | struct mdinfo *info; | |
1729 | char *subarray; | |
1730 | int ret_val = -1; | |
1731 | ||
1732 | if ((st == NULL) || (sra == NULL)) | |
1733 | return ret_val; | |
1734 | ||
1735 | if (text_version == NULL) | |
1736 | text_version = sra->text_version; | |
6ebf34e6 | 1737 | subarray = strchr(text_version + 1, '/')+1; |
54397ed9 AK |
1738 | info = st->ss->container_content(st, subarray); |
1739 | if (info) { | |
1740 | unsigned long long current_size = 0; | |
358ef9bf | 1741 | unsigned long long new_size = info->custom_array_size/2; |
54397ed9 AK |
1742 | |
1743 | if (sysfs_get_ll(sra, NULL, "array_size", ¤t_size) == 0 && | |
1744 | new_size > current_size) { | |
1745 | if (sysfs_set_num(sra, NULL, "array_size", new_size) | |
1746 | < 0) | |
1747 | dprintf("Error: Cannot set array size"); | |
1748 | else { | |
1749 | ret_val = 0; | |
1750 | dprintf("Array size changed"); | |
1751 | } | |
1ade5cc1 N |
1752 | dprintf_cont(" from %llu to %llu.\n", |
1753 | current_size, new_size); | |
54397ed9 AK |
1754 | } |
1755 | sysfs_free(info); | |
1756 | } else | |
1757 | dprintf("Error: set_array_size(): info pointer in NULL\n"); | |
1758 | ||
1759 | return ret_val; | |
1760 | } | |
1761 | ||
5da9ab98 N |
1762 | static int reshape_array(char *container, int fd, char *devname, |
1763 | struct supertype *st, struct mdinfo *info, | |
e2e53a2d | 1764 | int force, struct mddev_dev *devlist, |
19ceb16d | 1765 | unsigned long long data_offset, |
ba728be7 | 1766 | char *backup_file, int verbose, int forked, |
b76b30e0 | 1767 | int restart, int freeze_reshape); |
493f5dd6 | 1768 | static int reshape_container(char *container, char *devname, |
9ad6f6e6 | 1769 | int mdfd, |
24daa16f | 1770 | struct supertype *st, |
5da9ab98 N |
1771 | struct mdinfo *info, |
1772 | int force, | |
b0140ae8 N |
1773 | char *backup_file, int verbose, |
1774 | int forked, int restart, int freeze_reshape); | |
1c009fc2 | 1775 | |
32754b7d | 1776 | int Grow_reshape(char *devname, int fd, |
e2e53a2d | 1777 | struct mddev_dev *devlist, |
40c9a66a | 1778 | unsigned long long data_offset, |
32754b7d | 1779 | struct context *c, struct shape *s) |
e86c9dd6 NB |
1780 | { |
1781 | /* Make some changes in the shape of an array. | |
1782 | * The kernel must support the change. | |
7236ee7a N |
1783 | * |
1784 | * There are three different changes. Each can trigger | |
1785 | * a resync or recovery so we freeze that until we have | |
1786 | * requested everything (if kernel supports freezing - 2.6.30). | |
1787 | * The steps are: | |
1788 | * - change size (i.e. component_size) | |
1789 | * - change level | |
1790 | * - change layout/chunksize/ndisks | |
1791 | * | |
1792 | * The last can require a reshape. It is different on different | |
1793 | * levels so we need to check the level before actioning it. | |
1794 | * Some times the level change needs to be requested after the | |
1795 | * reshape (e.g. raid6->raid5, raid5->raid0) | |
1796 | * | |
e86c9dd6 | 1797 | */ |
5da9ab98 | 1798 | struct mdu_array_info_s array; |
7236ee7a | 1799 | int rv = 0; |
e86c9dd6 | 1800 | struct supertype *st; |
4725bc31 | 1801 | char *subarray = NULL; |
e86c9dd6 | 1802 | |
7236ee7a | 1803 | int frozen; |
7236ee7a | 1804 | int changed = 0; |
7bc71196 DW |
1805 | char *container = NULL; |
1806 | int cfd = -1; | |
e86c9dd6 | 1807 | |
e2e53a2d N |
1808 | struct mddev_dev *dv; |
1809 | int added_disks; | |
1810 | ||
5da9ab98 | 1811 | struct mdinfo info; |
7e0f6979 | 1812 | struct mdinfo *sra; |
e86c9dd6 | 1813 | |
9cd39f01 | 1814 | if (md_get_array_info(fd, &array) < 0) { |
ed503f89 | 1815 | pr_err("%s is not an active md array - aborting\n", |
e86c9dd6 NB |
1816 | devname); |
1817 | return 1; | |
1818 | } | |
a3b831c9 MT |
1819 | if (s->level != UnSet && s->chunk) { |
1820 | pr_err("Cannot change array level in the same operation as changing chunk size.\n"); | |
1821 | return 1; | |
1822 | } | |
1823 | ||
6ebf34e6 JS |
1824 | if (data_offset != INVALID_SECTORS && array.level != 10 && |
1825 | (array.level < 4 || array.level > 6)) { | |
19ceb16d N |
1826 | pr_err("--grow --data-offset not yet supported\n"); |
1827 | return 1; | |
1828 | } | |
24d40069 | 1829 | |
32754b7d N |
1830 | if (s->size > 0 && |
1831 | (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) { | |
7a862a02 N |
1832 | pr_err("cannot change component size at the same time as other changes.\n" |
1833 | " Change size first, then check data is intact before making other changes.\n"); | |
9ce510be N |
1834 | return 1; |
1835 | } | |
1836 | ||
6ebf34e6 JS |
1837 | if (s->raiddisks && s->raiddisks < array.raid_disks && |
1838 | array.level > 1 && get_linux_version() < 2006032 && | |
24d40069 | 1839 | !check_env("MDADM_FORCE_FEWER")) { |
e7b84f9d | 1840 | pr_err("reducing the number of devices is not safe before Linux 2.6.32\n" |
24d40069 N |
1841 | " Please use a newer kernel\n"); |
1842 | return 1; | |
1843 | } | |
7bc71196 | 1844 | |
56e1e6ac | 1845 | if (array.level > 1 && s->size > 1 && |
5d518de8 | 1846 | (unsigned long long) (array.chunk_size / 1024) > s->size) { |
4b74a905 ZL |
1847 | pr_err("component size must be larger than chunk size.\n"); |
1848 | return 1; | |
1849 | } | |
1850 | ||
7bc71196 DW |
1851 | st = super_by_fd(fd, &subarray); |
1852 | if (!st) { | |
e7b84f9d | 1853 | pr_err("Unable to determine metadata format for %s\n", devname); |
7bc71196 DW |
1854 | return 1; |
1855 | } | |
32754b7d | 1856 | if (s->raiddisks > st->max_devs) { |
7a862a02 | 1857 | pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs); |
acab7bb1 N |
1858 | return 1; |
1859 | } | |
bdbe7f81 JR |
1860 | if (s->level == 0 && (array.state & (1 << MD_SB_BITMAP_PRESENT)) && |
1861 | !(array.state & (1 << MD_SB_CLUSTERED)) && !st->ss->external) { | |
1862 | array.state &= ~(1 << MD_SB_BITMAP_PRESENT); | |
1863 | if (md_set_array_info(fd, &array) != 0) { | |
1864 | pr_err("failed to remove internal bitmap.\n"); | |
1865 | return 1; | |
1866 | } | |
1867 | } | |
7bc71196 DW |
1868 | |
1869 | /* in the external case we need to check that the requested reshape is | |
1870 | * supported, and perform an initial check that the container holds the | |
1871 | * pre-requisite spare devices (mdmon owns final validation) | |
1872 | */ | |
1873 | if (st->ss->external) { | |
49948a35 | 1874 | int retval; |
7bc71196 DW |
1875 | |
1876 | if (subarray) { | |
4dd2df09 N |
1877 | container = st->container_devnm; |
1878 | cfd = open_dev_excl(st->container_devnm); | |
7bc71196 | 1879 | } else { |
4dd2df09 | 1880 | container = st->devnm; |
7bc71196 | 1881 | close(fd); |
4dd2df09 | 1882 | cfd = open_dev_excl(st->devnm); |
7bc71196 DW |
1883 | fd = cfd; |
1884 | } | |
1885 | if (cfd < 0) { | |
358ef9bf | 1886 | pr_err("Unable to open container for %s\n", devname); |
7f2ba464 | 1887 | free(subarray); |
7bc71196 DW |
1888 | return 1; |
1889 | } | |
1890 | ||
49948a35 | 1891 | retval = st->ss->load_container(st, cfd, NULL); |
eb744788 | 1892 | |
49948a35 | 1893 | if (retval) { |
358ef9bf | 1894 | pr_err("Cannot read superblock for %s\n", devname); |
20e114e3 MK |
1895 | close(cfd); |
1896 | free(subarray); | |
1897 | return 1; | |
1898 | } | |
1899 | ||
1900 | if (s->raiddisks && subarray) { | |
1901 | pr_err("--raid-devices operation can be performed on a container only\n"); | |
1902 | close(cfd); | |
7f2ba464 | 1903 | free(subarray); |
7bc71196 DW |
1904 | return 1; |
1905 | } | |
1906 | ||
81219e70 LM |
1907 | /* check if operation is supported for metadata handler */ |
1908 | if (st->ss->container_content) { | |
1909 | struct mdinfo *cc = NULL; | |
1910 | struct mdinfo *content = NULL; | |
1911 | ||
1912 | cc = st->ss->container_content(st, subarray); | |
1913 | for (content = cc; content ; content = content->next) { | |
446894ea | 1914 | int allow_reshape = 1; |
81219e70 LM |
1915 | |
1916 | /* check if reshape is allowed based on metadata | |
1917 | * indications stored in content.array.status | |
1918 | */ | |
6ebf34e6 JS |
1919 | if (content->array.state & |
1920 | (1 << MD_SB_BLOCK_VOLUME)) | |
446894ea | 1921 | allow_reshape = 0; |
6ebf34e6 JS |
1922 | if (content->array.state & |
1923 | (1 << MD_SB_BLOCK_CONTAINER_RESHAPE)) | |
446894ea | 1924 | allow_reshape = 0; |
81219e70 | 1925 | if (!allow_reshape) { |
7a862a02 | 1926 | pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n", |
4dd2df09 | 1927 | devname, container); |
81219e70 LM |
1928 | sysfs_free(cc); |
1929 | free(subarray); | |
1930 | return 1; | |
1931 | } | |
b208f817 TM |
1932 | if (content->consistency_policy == |
1933 | CONSISTENCY_POLICY_PPL) { | |
1934 | pr_err("Operation not supported when ppl consistency policy is enabled\n"); | |
1935 | sysfs_free(cc); | |
1936 | free(subarray); | |
1937 | return 1; | |
1938 | } | |
bdbe7f81 JR |
1939 | if (content->consistency_policy == |
1940 | CONSISTENCY_POLICY_BITMAP) { | |
1941 | pr_err("Operation not supported when write-intent bitmap is enabled\n"); | |
1942 | sysfs_free(cc); | |
1943 | free(subarray); | |
1944 | return 1; | |
1945 | } | |
81219e70 LM |
1946 | } |
1947 | sysfs_free(cc); | |
1948 | } | |
4dd2df09 | 1949 | if (mdmon_running(container)) |
7bc71196 | 1950 | st->update_tail = &st->updates; |
e2e53a2d | 1951 | } |
691a36b7 | 1952 | |
e2e53a2d N |
1953 | added_disks = 0; |
1954 | for (dv = devlist; dv; dv = dv->next) | |
1955 | added_disks++; | |
32754b7d | 1956 | if (s->raiddisks > array.raid_disks && |
358ef9bf JS |
1957 | array.spare_disks + added_disks < |
1958 | (s->raiddisks - array.raid_disks) && | |
32754b7d | 1959 | !c->force) { |
7a862a02 | 1960 | pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n" |
e7b84f9d | 1961 | " Use --force to over-ride this check.\n", |
32754b7d N |
1962 | s->raiddisks - array.raid_disks, |
1963 | s->raiddisks - array.raid_disks == 1 ? "" : "s", | |
e7b84f9d | 1964 | array.spare_disks + added_disks); |
691a36b7 | 1965 | return 1; |
7bc71196 DW |
1966 | } |
1967 | ||
6ebf34e6 JS |
1968 | sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS | |
1969 | GET_STATE | GET_VERSION); | |
24daa16f | 1970 | if (sra) { |
7f2ba464 | 1971 | if (st->ss->external && subarray == NULL) { |
7bc71196 DW |
1972 | array.level = LEVEL_CONTAINER; |
1973 | sra->array.level = LEVEL_CONTAINER; | |
1974 | } | |
7bc71196 | 1975 | } else { |
e7b84f9d | 1976 | pr_err("failed to read sysfs parameters for %s\n", |
b526e52d DW |
1977 | devname); |
1978 | return 1; | |
1979 | } | |
7f2ba464 DW |
1980 | frozen = freeze(st); |
1981 | if (frozen < -1) { | |
1982 | /* freeze() already spewed the reason */ | |
8e61e0d7 | 1983 | sysfs_free(sra); |
7f2ba464 DW |
1984 | return 1; |
1985 | } else if (frozen < 0) { | |
7a862a02 | 1986 | pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname); |
8e61e0d7 | 1987 | sysfs_free(sra); |
7236ee7a N |
1988 | return 1; |
1989 | } | |
1990 | ||
1991 | /* ========= set size =============== */ | |
6ebf34e6 JS |
1992 | if (s->size > 0 && |
1993 | (s->size == MAX_SIZE || s->size != (unsigned)array.size)) { | |
d04f65f4 N |
1994 | unsigned long long orig_size = get_component_size(fd)/2; |
1995 | unsigned long long min_csize; | |
d1537ed1 | 1996 | struct mdinfo *mdi; |
44f6f181 | 1997 | int raid0_takeover = 0; |
7bc71196 | 1998 | |
85f10287 | 1999 | if (orig_size == 0) |
d04f65f4 | 2000 | orig_size = (unsigned) array.size; |
85f10287 | 2001 | |
3920235e N |
2002 | if (orig_size == 0) { |
2003 | pr_err("Cannot set device size in this type of array.\n"); | |
2004 | rv = 1; | |
2005 | goto release; | |
2006 | } | |
2007 | ||
fea026b4 MK |
2008 | if (array.level == 0) { |
2009 | pr_err("Component size change is not supported for RAID0\n"); | |
2010 | rv = 1; | |
2011 | goto release; | |
2012 | } | |
2013 | ||
32754b7d | 2014 | if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL, |
6ebf34e6 JS |
2015 | devname, APPLY_METADATA_CHANGES, |
2016 | c->verbose > 0)) { | |
7bc71196 DW |
2017 | rv = 1; |
2018 | goto release; | |
2019 | } | |
2020 | sync_metadata(st); | |
7e7e9a4d AK |
2021 | if (st->ss->external) { |
2022 | /* metadata can have size limitation | |
2023 | * update size value according to metadata information | |
2024 | */ | |
2025 | struct mdinfo *sizeinfo = | |
2026 | st->ss->container_content(st, subarray); | |
2027 | if (sizeinfo) { | |
2028 | unsigned long long new_size = | |
2029 | sizeinfo->custom_array_size/2; | |
2030 | int data_disks = get_data_disks( | |
2031 | sizeinfo->array.level, | |
2032 | sizeinfo->array.layout, | |
2033 | sizeinfo->array.raid_disks); | |
2034 | new_size /= data_disks; | |
6ebf34e6 JS |
2035 | dprintf("Metadata size correction from %llu to %llu (%llu)\n", |
2036 | orig_size, new_size, | |
7e7e9a4d | 2037 | new_size * data_disks); |
32754b7d | 2038 | s->size = new_size; |
7e7e9a4d AK |
2039 | sysfs_free(sizeinfo); |
2040 | } | |
2041 | } | |
d1537ed1 N |
2042 | |
2043 | /* Update the size of each member device in case | |
2044 | * they have been resized. This will never reduce | |
2045 | * below the current used-size. The "size" attribute | |
20a46756 | 2046 | * understands '0' to mean 'max'. |
d1537ed1 | 2047 | */ |
20a46756 N |
2048 | min_csize = 0; |
2049 | for (mdi = sra->devs; mdi; mdi = mdi->next) { | |
358ef9bf JS |
2050 | sysfs_set_num(sra, mdi, "size", |
2051 | s->size == MAX_SIZE ? 0 : s->size); | |
20a46756 N |
2052 | if (array.not_persistent == 0 && |
2053 | array.major_version == 0 && | |
2054 | get_linux_version() < 3001000) { | |
2055 | /* Dangerous to allow size to exceed 2TB */ | |
2056 | unsigned long long csize; | |
358ef9bf JS |
2057 | if (sysfs_get_ll(sra, mdi, "size", |
2058 | &csize) == 0) { | |
20a46756 N |
2059 | if (csize >= 2ULL*1024*1024*1024) |
2060 | csize = 2ULL*1024*1024*1024; | |
358ef9bf JS |
2061 | if ((min_csize == 0 || |
2062 | (min_csize > csize))) | |
20a46756 N |
2063 | min_csize = csize; |
2064 | } | |
2065 | } | |
2066 | } | |
32754b7d | 2067 | if (min_csize && s->size > min_csize) { |
7a862a02 | 2068 | pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n"); |
20a46756 | 2069 | rv = 1; |
65a9798b | 2070 | goto size_change_error; |
20a46756 | 2071 | } |
32754b7d | 2072 | if (min_csize && s->size == MAX_SIZE) { |
20a46756 N |
2073 | /* Don't let the kernel choose a size - it will get |
2074 | * it wrong | |
2075 | */ | |
7a862a02 | 2076 | pr_err("Limited v0.90 array to 2TB per device\n"); |
32754b7d | 2077 | s->size = min_csize; |
20a46756 | 2078 | } |
44f6f181 AK |
2079 | if (st->ss->external) { |
2080 | if (sra->array.level == 0) { | |
358ef9bf | 2081 | rv = sysfs_set_str(sra, NULL, "level", "raid5"); |
44f6f181 AK |
2082 | if (!rv) { |
2083 | raid0_takeover = 1; | |
2a6493cf N |
2084 | /* get array parameters after takeover |
2085 | * to change one parameter at time only | |
44f6f181 | 2086 | */ |
9cd39f01 | 2087 | rv = md_get_array_info(fd, &array); |
44f6f181 AK |
2088 | } |
2089 | } | |
2090 | /* make sure mdmon is | |
2091 | * aware of the new level */ | |
4dd2df09 N |
2092 | if (!mdmon_running(st->container_devnm)) |
2093 | start_mdmon(st->container_devnm); | |
44f6f181 | 2094 | ping_monitor(container); |
4dd2df09 | 2095 | if (mdmon_running(st->container_devnm) && |
44f6f181 AK |
2096 | st->update_tail == NULL) |
2097 | st->update_tail = &st->updates; | |
2098 | } | |
2099 | ||
0a8b92a6 N |
2100 | if (s->size == MAX_SIZE) |
2101 | s->size = 0; | |
2102 | array.size = s->size; | |
0448027b | 2103 | if (s->size & ~INT32_MAX) { |
7236ee7a N |
2104 | /* got truncated to 32bit, write to |
2105 | * component_size instead | |
2106 | */ | |
2107 | if (sra) | |
2108 | rv = sysfs_set_num(sra, NULL, | |
32754b7d | 2109 | "component_size", s->size); |
7236ee7a N |
2110 | else |
2111 | rv = -1; | |
54397ed9 | 2112 | } else { |
018a4882 | 2113 | rv = md_set_array_info(fd, &array); |
44f6f181 | 2114 | |
54397ed9 AK |
2115 | /* manage array size when it is managed externally |
2116 | */ | |
2117 | if ((rv == 0) && st->ss->external) | |
2118 | rv = set_array_size(st, sra, sra->text_version); | |
2119 | } | |
2120 | ||
44f6f181 AK |
2121 | if (raid0_takeover) { |
2122 | /* do not recync non-existing parity, | |
2123 | * we will drop it anyway | |
2124 | */ | |
b51702b8 | 2125 | sysfs_set_str(sra, NULL, "sync_action", "frozen"); |
44f6f181 AK |
2126 | /* go back to raid0, drop parity disk |
2127 | */ | |
2128 | sysfs_set_str(sra, NULL, "level", "raid0"); | |
9cd39f01 | 2129 | md_get_array_info(fd, &array); |
44f6f181 AK |
2130 | } |
2131 | ||
65a9798b | 2132 | size_change_error: |
7236ee7a | 2133 | if (rv != 0) { |
39bbb392 | 2134 | int err = errno; |
7bc71196 DW |
2135 | |
2136 | /* restore metadata */ | |
2137 | if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0, | |
016e00f5 AK |
2138 | UnSet, NULL, devname, |
2139 | ROLLBACK_METADATA_CHANGES, | |
32754b7d | 2140 | c->verbose) == 0) |
7bc71196 | 2141 | sync_metadata(st); |
e7b84f9d | 2142 | pr_err("Cannot set device size for %s: %s\n", |
39bbb392 | 2143 | devname, strerror(err)); |
24daa16f | 2144 | if (err == EBUSY && |
39bbb392 | 2145 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) |
e7b84f9d | 2146 | cont_err("Bitmap must be removed before size can be changed\n"); |
7236ee7a N |
2147 | rv = 1; |
2148 | goto release; | |
2149 | } | |
32754b7d | 2150 | if (s->assume_clean) { |
86952387 | 2151 | /* This will fail on kernels older than 3.0 unless |
ce52f92f N |
2152 | * a backport has been arranged. |
2153 | */ | |
2154 | if (sra == NULL || | |
358ef9bf JS |
2155 | sysfs_set_str(sra, NULL, "resync_start", |
2156 | "none") < 0) | |
86952387 | 2157 | pr_err("--assume-clean not supported with --grow on this kernel\n"); |
ce52f92f | 2158 | } |
9cd39f01 | 2159 | md_get_array_info(fd, &array); |
32754b7d N |
2160 | s->size = get_component_size(fd)/2; |
2161 | if (s->size == 0) | |
2162 | s->size = array.size; | |
2163 | if (c->verbose >= 0) { | |
2164 | if (s->size == orig_size) | |
7a862a02 | 2165 | pr_err("component size of %s unchanged at %lluK\n", |
32754b7d | 2166 | devname, s->size); |
85f10287 | 2167 | else |
7a862a02 | 2168 | pr_err("component size of %s has been set to %lluK\n", |
32754b7d | 2169 | devname, s->size); |
85f10287 | 2170 | } |
7236ee7a | 2171 | changed = 1; |
7bc71196 | 2172 | } else if (array.level != LEVEL_CONTAINER) { |
32754b7d N |
2173 | s->size = get_component_size(fd)/2; |
2174 | if (s->size == 0) | |
2175 | s->size = array.size; | |
7236ee7a N |
2176 | } |
2177 | ||
d515d27f | 2178 | /* See if there is anything else to do */ |
32754b7d N |
2179 | if ((s->level == UnSet || s->level == array.level) && |
2180 | (s->layout_str == NULL) && | |
2181 | (s->chunk == 0 || s->chunk == array.chunk_size) && | |
8876bf0b | 2182 | data_offset == INVALID_SECTORS && |
32754b7d | 2183 | (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) { |
d515d27f | 2184 | /* Nothing more to do */ |
32754b7d | 2185 | if (!changed && c->verbose >= 0) |
358ef9bf | 2186 | pr_err("%s: no change requested\n", devname); |
d515d27f N |
2187 | goto release; |
2188 | } | |
2189 | ||
dfe77a9e | 2190 | /* ========= check for Raid10/Raid1 -> Raid0 conversion =============== |
5da9ab98 | 2191 | * current implementation assumes that following conditions must be met: |
dfe77a9e | 2192 | * - RAID10: |
24daa16f N |
2193 | * - far_copies == 1 |
2194 | * - near_copies == 2 | |
62a48395 | 2195 | */ |
32754b7d | 2196 | if ((s->level == 0 && array.level == 10 && sra && |
24daa16f | 2197 | array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) || |
32754b7d | 2198 | (s->level == 0 && array.level == 1 && sra)) { |
62a48395 | 2199 | int err; |
358ef9bf | 2200 | |
dfe77a9e | 2201 | err = remove_disks_for_takeover(st, sra, array.layout); |
62a48395 | 2202 | if (err) { |
1ade5cc1 | 2203 | dprintf("Array cannot be reshaped\n"); |
62a48395 AK |
2204 | if (cfd > -1) |
2205 | close(cfd); | |
5da9ab98 N |
2206 | rv = 1; |
2207 | goto release; | |
7236ee7a | 2208 | } |
fde139b9 N |
2209 | /* Make sure mdmon has seen the device removal |
2210 | * and updated metadata before we continue with | |
2211 | * level change | |
2212 | */ | |
2213 | if (container) | |
2214 | ping_monitor(container); | |
7236ee7a N |
2215 | } |
2216 | ||
8ff6d094 | 2217 | memset(&info, 0, sizeof(info)); |
5da9ab98 | 2218 | info.array = array; |
dae13137 | 2219 | if (sysfs_init(&info, fd, NULL)) { |
ebf3be99 | 2220 | pr_err("failed to initialize sysfs.\n"); |
dae13137 JS |
2221 | rv = 1; |
2222 | goto release; | |
2223 | } | |
3fa436ac | 2224 | strcpy(info.text_version, sra->text_version); |
32754b7d N |
2225 | info.component_size = s->size*2; |
2226 | info.new_level = s->level; | |
2227 | info.new_chunk = s->chunk * 1024; | |
eff4954d N |
2228 | if (info.array.level == LEVEL_CONTAINER) { |
2229 | info.delta_disks = UnSet; | |
32754b7d N |
2230 | info.array.raid_disks = s->raiddisks; |
2231 | } else if (s->raiddisks) | |
2232 | info.delta_disks = s->raiddisks - info.array.raid_disks; | |
5da9ab98 N |
2233 | else |
2234 | info.delta_disks = UnSet; | |
32754b7d | 2235 | if (s->layout_str == NULL) { |
5da9ab98 N |
2236 | info.new_layout = UnSet; |
2237 | if (info.array.level == 6 && | |
2238 | (info.new_level == 6 || info.new_level == UnSet) && | |
2239 | info.array.layout >= 16) { | |
7a862a02 N |
2240 | pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname); |
2241 | cont_err("during the reshape, please specify --layout=preserve\n"); | |
2242 | cont_err("If you want to change it, specify a layout or use --layout=normalise\n"); | |
7236ee7a N |
2243 | rv = 1; |
2244 | goto release; | |
2245 | } | |
32754b7d N |
2246 | } else if (strcmp(s->layout_str, "normalise") == 0 || |
2247 | strcmp(s->layout_str, "normalize") == 0) { | |
5da9ab98 N |
2248 | /* If we have a -6 RAID6 layout, remove the '-6'. */ |
2249 | info.new_layout = UnSet; | |
2250 | if (info.array.level == 6 && info.new_level == UnSet) { | |
2251 | char l[40], *h; | |
5f21d674 | 2252 | strcpy(l, map_num_s(r6layout, info.array.layout)); |
5da9ab98 N |
2253 | h = strrchr(l, '-'); |
2254 | if (h && strcmp(h, "-6") == 0) { | |
2255 | *h = 0; | |
2256 | info.new_layout = map_name(r6layout, l); | |
7236ee7a | 2257 | } |
385167f3 | 2258 | } else { |
7a862a02 | 2259 | pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str); |
385167f3 N |
2260 | rv = 1; |
2261 | goto release; | |
7236ee7a | 2262 | } |
32754b7d | 2263 | } else if (strcmp(s->layout_str, "preserve") == 0) { |
385167f3 N |
2264 | /* This means that a non-standard RAID6 layout |
2265 | * is OK. | |
2266 | * In particular: | |
2267 | * - When reshape a RAID6 (e.g. adding a device) | |
2268 | * which is in a non-standard layout, it is OK | |
2269 | * to preserve that layout. | |
2270 | * - When converting a RAID5 to RAID6, leave it in | |
2271 | * the XXX-6 layout, don't re-layout. | |
2272 | */ | |
2273 | if (info.array.level == 6 && info.new_level == UnSet) | |
2274 | info.new_layout = info.array.layout; | |
2275 | else if (info.array.level == 5 && info.new_level == 6) { | |
2276 | char l[40]; | |
5f21d674 | 2277 | strcpy(l, map_num_s(r5layout, info.array.layout)); |
385167f3 N |
2278 | strcat(l, "-6"); |
2279 | info.new_layout = map_name(r6layout, l); | |
2280 | } else { | |
7a862a02 | 2281 | pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str); |
385167f3 N |
2282 | rv = 1; |
2283 | goto release; | |
2284 | } | |
5da9ab98 N |
2285 | } else { |
2286 | int l = info.new_level; | |
2287 | if (l == UnSet) | |
2288 | l = info.array.level; | |
2289 | switch (l) { | |
2290 | case 5: | |
32754b7d | 2291 | info.new_layout = map_name(r5layout, s->layout_str); |
5da9ab98 N |
2292 | break; |
2293 | case 6: | |
32754b7d | 2294 | info.new_layout = map_name(r6layout, s->layout_str); |
5da9ab98 N |
2295 | break; |
2296 | case 10: | |
32754b7d | 2297 | info.new_layout = parse_layout_10(s->layout_str); |
5da9ab98 N |
2298 | break; |
2299 | case LEVEL_FAULTY: | |
32754b7d | 2300 | info.new_layout = parse_layout_faulty(s->layout_str); |
5da9ab98 N |
2301 | break; |
2302 | default: | |
7a862a02 | 2303 | pr_err("layout not meaningful with this level\n"); |
7bc71196 DW |
2304 | rv = 1; |
2305 | goto release; | |
2306 | } | |
5da9ab98 | 2307 | if (info.new_layout == UnSet) { |
7a862a02 | 2308 | pr_err("layout %s not understood for this level\n", |
32754b7d | 2309 | s->layout_str); |
5da9ab98 N |
2310 | rv = 1; |
2311 | goto release; | |
7bc71196 | 2312 | } |
7236ee7a N |
2313 | } |
2314 | ||
907ea753 | 2315 | if (array.level == LEVEL_FAULTY) { |
32754b7d | 2316 | if (s->level != UnSet && s->level != array.level) { |
e7b84f9d | 2317 | pr_err("cannot change level of Faulty device\n"); |
907ea753 N |
2318 | rv =1 ; |
2319 | } | |
32754b7d | 2320 | if (s->chunk) { |
e7b84f9d | 2321 | pr_err("cannot set chunksize of Faulty device\n"); |
907ea753 N |
2322 | rv =1 ; |
2323 | } | |
32754b7d | 2324 | if (s->raiddisks && s->raiddisks != 1) { |
e7b84f9d | 2325 | pr_err("cannot set raid_disks of Faulty device\n"); |
907ea753 N |
2326 | rv =1 ; |
2327 | } | |
32754b7d | 2328 | if (s->layout_str) { |
9cd39f01 | 2329 | if (md_get_array_info(fd, &array) != 0) { |
907ea753 N |
2330 | dprintf("Cannot get array information.\n"); |
2331 | goto release; | |
2332 | } | |
2333 | array.layout = info.new_layout; | |
018a4882 | 2334 | if (md_set_array_info(fd, &array) != 0) { |
e7b84f9d | 2335 | pr_err("failed to set new layout\n"); |
907ea753 | 2336 | rv = 1; |
32754b7d | 2337 | } else if (c->verbose >= 0) |
907ea753 N |
2338 | printf("layout for %s set to %d\n", |
2339 | devname, array.layout); | |
2340 | } | |
2341 | } else if (array.level == LEVEL_CONTAINER) { | |
5da9ab98 N |
2342 | /* This change is to be applied to every array in the |
2343 | * container. This is only needed when the metadata imposes | |
2344 | * restraints of the various arrays in the container. | |
2345 | * Currently we only know that IMSM requires all arrays | |
2346 | * to have the same number of devices so changing the | |
2347 | * number of devices (On-Line Capacity Expansion) must be | |
2348 | * performed at the level of the container | |
2349 | */ | |
b71de056 | 2350 | close_fd(&fd); |
9ad6f6e6 | 2351 | rv = reshape_container(container, devname, -1, st, &info, |
358ef9bf JS |
2352 | c->force, c->backup_file, c->verbose, |
2353 | 0, 0, 0); | |
9202b8eb | 2354 | frozen = 0; |
5da9ab98 | 2355 | } else { |
08f9e34b AK |
2356 | /* get spare devices from external metadata |
2357 | */ | |
2358 | if (st->ss->external) { | |
2359 | struct mdinfo *info2; | |
2360 | ||
2361 | info2 = st->ss->container_content(st, subarray); | |
2362 | if (info2) { | |
2363 | info.array.spare_disks = | |
2364 | info2->array.spare_disks; | |
2365 | sysfs_free(info2); | |
2366 | } | |
2367 | } | |
2368 | ||
5da9ab98 N |
2369 | /* Impose these changes on a single array. First |
2370 | * check that the metadata is OK with the change. */ | |
7bc71196 | 2371 | |
7ec0996c | 2372 | if (reshape_super(st, 0, info.new_level, |
5da9ab98 | 2373 | info.new_layout, info.new_chunk, |
41784c88 | 2374 | info.array.raid_disks, info.delta_disks, |
358ef9bf JS |
2375 | c->backup_file, devname, |
2376 | APPLY_METADATA_CHANGES, c->verbose)) { | |
7bc71196 DW |
2377 | rv = 1; |
2378 | goto release; | |
2379 | } | |
5da9ab98 | 2380 | sync_metadata(st); |
32754b7d | 2381 | rv = reshape_array(container, fd, devname, st, &info, c->force, |
358ef9bf JS |
2382 | devlist, data_offset, c->backup_file, |
2383 | c->verbose, 0, 0, 0); | |
9202b8eb | 2384 | frozen = 0; |
5da9ab98 N |
2385 | } |
2386 | release: | |
8e61e0d7 | 2387 | sysfs_free(sra); |
9202b8eb N |
2388 | if (frozen > 0) |
2389 | unfreeze(st); | |
5da9ab98 N |
2390 | return rv; |
2391 | } | |
7bc71196 | 2392 | |
f93346ef AK |
2393 | /* verify_reshape_position() |
2394 | * Function checks if reshape position in metadata is not farther | |
2395 | * than position in md. | |
2396 | * Return value: | |
2397 | * 0 : not valid sysfs entry | |
2398 | * it can be caused by not started reshape, it should be started | |
2399 | * by reshape array or raid0 array is before takeover | |
2400 | * -1 : error, reshape position is obviously wrong | |
2401 | * 1 : success, reshape progress correct or updated | |
2402 | */ | |
2403 | static int verify_reshape_position(struct mdinfo *info, int level) | |
2404 | { | |
2405 | int ret_val = 0; | |
2406 | char buf[40]; | |
178950ea | 2407 | int rv; |
f93346ef AK |
2408 | |
2409 | /* read sync_max, failure can mean raid0 array */ | |
178950ea AK |
2410 | rv = sysfs_get_str(info, NULL, "sync_max", buf, 40); |
2411 | ||
2412 | if (rv > 0) { | |
f93346ef AK |
2413 | char *ep; |
2414 | unsigned long long position = strtoull(buf, &ep, 0); | |
2415 | ||
1ade5cc1 | 2416 | dprintf("Read sync_max sysfs entry is: %s\n", buf); |
f93346ef AK |
2417 | if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) { |
2418 | position *= get_data_disks(level, | |
2419 | info->new_layout, | |
2420 | info->array.raid_disks); | |
2421 | if (info->reshape_progress < position) { | |
7a862a02 | 2422 | dprintf("Corrected reshape progress (%llu) to md position (%llu)\n", |
f93346ef AK |
2423 | info->reshape_progress, position); |
2424 | info->reshape_progress = position; | |
2425 | ret_val = 1; | |
2426 | } else if (info->reshape_progress > position) { | |
7a862a02 | 2427 | pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n", |
e7b84f9d | 2428 | position, info->reshape_progress); |
f93346ef AK |
2429 | ret_val = -1; |
2430 | } else { | |
7a862a02 | 2431 | dprintf("Reshape position in md and metadata are the same;"); |
f93346ef AK |
2432 | ret_val = 1; |
2433 | } | |
2434 | } | |
178950ea AK |
2435 | } else if (rv == 0) { |
2436 | /* for valid sysfs entry, 0-length content | |
2437 | * should be indicated as error | |
2438 | */ | |
2439 | ret_val = -1; | |
f93346ef AK |
2440 | } |
2441 | ||
2442 | return ret_val; | |
2443 | } | |
2444 | ||
534f5432 N |
2445 | static unsigned long long choose_offset(unsigned long long lo, |
2446 | unsigned long long hi, | |
2447 | unsigned long long min, | |
2448 | unsigned long long max) | |
2449 | { | |
2450 | /* Choose a new offset between hi and lo. | |
2451 | * It must be between min and max, but | |
2452 | * we would prefer something near the middle of hi/lo, and also | |
2453 | * prefer to be aligned to a big power of 2. | |
2454 | * | |
2455 | * So we start with the middle, then for each bit, | |
2456 | * starting at '1' and increasing, if it is set, we either | |
2457 | * add it or subtract it if possible, preferring the option | |
2458 | * which is furthest from the boundary. | |
2459 | * | |
2460 | * We stop once we get a 1MB alignment. As units are in sectors, | |
2461 | * 1MB = 2*1024 sectors. | |
2462 | */ | |
2463 | unsigned long long choice = (lo + hi) / 2; | |
2464 | unsigned long long bit = 1; | |
2465 | ||
2466 | for (bit = 1; bit < 2*1024; bit = bit << 1) { | |
2467 | unsigned long long bigger, smaller; | |
2468 | if (! (bit & choice)) | |
2469 | continue; | |
2470 | bigger = choice + bit; | |
2471 | smaller = choice - bit; | |
2472 | if (bigger > max && smaller < min) | |
2473 | break; | |
2474 | if (bigger > max) | |
2475 | choice = smaller; | |
2476 | else if (smaller < min) | |
2477 | choice = bigger; | |
2478 | else if (hi - bigger > smaller - lo) | |
2479 | choice = bigger; | |
2480 | else | |
2481 | choice = smaller; | |
2482 | } | |
2483 | return choice; | |
2484 | } | |
2485 | ||
13bbb145 | 2486 | static int set_new_data_offset(struct mdinfo *sra, struct supertype *st, |
50a4962f | 2487 | char *devname, int delta_disks, |
13bbb145 | 2488 | unsigned long long data_offset, |
6a23fb9d N |
2489 | unsigned long long min, |
2490 | int can_fallback) | |
19ceb16d | 2491 | { |
13bbb145 | 2492 | struct mdinfo *sd; |
19ceb16d N |
2493 | int dir = 0; |
2494 | int err = 0; | |
f9b08fec | 2495 | unsigned long long before, after; |
19ceb16d | 2496 | |
f9b08fec N |
2497 | /* Need to find min space before and after so same is used |
2498 | * on all devices | |
2499 | */ | |
2500 | before = UINT64_MAX; | |
2501 | after = UINT64_MAX; | |
19ceb16d N |
2502 | for (sd = sra->devs; sd; sd = sd->next) { |
2503 | char *dn; | |
2504 | int dfd; | |
2505 | int rv; | |
2506 | struct supertype *st2; | |
2507 | struct mdinfo info2; | |
2508 | ||
2509 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2510 | continue; | |
2511 | dn = map_dev(sd->disk.major, sd->disk.minor, 0); | |
2512 | dfd = dev_open(dn, O_RDONLY); | |
2513 | if (dfd < 0) { | |
ed503f89 | 2514 | pr_err("%s: cannot open component %s\n", |
19ceb16d | 2515 | devname, dn ? dn : "-unknown-"); |
19ceb16d N |
2516 | goto release; |
2517 | } | |
2518 | st2 = dup_super(st); | |
2519 | rv = st2->ss->load_super(st2,dfd, NULL); | |
2520 | close(dfd); | |
2521 | if (rv) { | |
2522 | free(st2); | |
ed503f89 | 2523 | pr_err("%s: cannot get superblock from %s\n", |
19ceb16d N |
2524 | devname, dn); |
2525 | goto release; | |
2526 | } | |
2527 | st2->ss->getinfo_super(st2, &info2, NULL); | |
2528 | st2->ss->free_super(st2); | |
2529 | free(st2); | |
c4b26c64 N |
2530 | if (info2.space_before == 0 && |
2531 | info2.space_after == 0) { | |
2532 | /* Metadata doesn't support data_offset changes */ | |
dd243f56 N |
2533 | if (!can_fallback) |
2534 | pr_err("%s: Metadata version doesn't support data_offset changes\n", | |
2535 | devname); | |
16afb1a5 | 2536 | goto fallback; |
c4b26c64 | 2537 | } |
f9b08fec N |
2538 | if (before > info2.space_before) |
2539 | before = info2.space_before; | |
2540 | if (after > info2.space_after) | |
2541 | after = info2.space_after; | |
2542 | ||
2543 | if (data_offset != INVALID_SECTORS) { | |
2544 | if (dir == 0) { | |
2545 | if (info2.data_offset == data_offset) { | |
2546 | pr_err("%s: already has that data_offset\n", | |
2547 | dn); | |
2548 | goto release; | |
2549 | } | |
2550 | if (data_offset < info2.data_offset) | |
2551 | dir = -1; | |
2552 | else | |
2553 | dir = 1; | |
358ef9bf JS |
2554 | } else if ((data_offset <= info2.data_offset && |
2555 | dir == 1) || | |
2556 | (data_offset >= info2.data_offset && | |
2557 | dir == -1)) { | |
f9b08fec N |
2558 | pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n", |
2559 | dn); | |
2560 | goto release; | |
2561 | } | |
2562 | } | |
2563 | } | |
2564 | if (before == UINT64_MAX) | |
2565 | /* impossible really, there must be no devices */ | |
2566 | return 1; | |
2567 | ||
2568 | for (sd = sra->devs; sd; sd = sd->next) { | |
2569 | char *dn = map_dev(sd->disk.major, sd->disk.minor, 0); | |
c0f0d812 | 2570 | unsigned long long new_data_offset; |
f9b08fec | 2571 | |
b397d7f3 N |
2572 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) |
2573 | continue; | |
50a4962f | 2574 | if (delta_disks < 0) { |
19ceb16d N |
2575 | /* Don't need any space as array is shrinking |
2576 | * just move data_offset up by min | |
2577 | */ | |
5582b118 | 2578 | if (data_offset == INVALID_SECTORS) |
c0f0d812 | 2579 | new_data_offset = sd->data_offset + min; |
19ceb16d | 2580 | else { |
c0f0d812 | 2581 | if (data_offset < sd->data_offset + min) { |
ed503f89 | 2582 | pr_err("--data-offset too small for %s\n", |
19ceb16d N |
2583 | dn); |
2584 | goto release; | |
2585 | } | |
c0f0d812 | 2586 | new_data_offset = data_offset; |
19ceb16d | 2587 | } |
50a4962f | 2588 | } else if (delta_disks > 0) { |
19ceb16d | 2589 | /* need space before */ |
f9b08fec | 2590 | if (before < min) { |
6a23fb9d N |
2591 | if (can_fallback) |
2592 | goto fallback; | |
ed503f89 | 2593 | pr_err("Insufficient head-space for reshape on %s\n", |
19ceb16d N |
2594 | dn); |
2595 | goto release; | |
2596 | } | |
5582b118 | 2597 | if (data_offset == INVALID_SECTORS) |
c0f0d812 | 2598 | new_data_offset = sd->data_offset - min; |
19ceb16d | 2599 | else { |
c0f0d812 | 2600 | if (data_offset > sd->data_offset - min) { |
ed503f89 | 2601 | pr_err("--data-offset too large for %s\n", |
19ceb16d N |
2602 | dn); |
2603 | goto release; | |
2604 | } | |
c0f0d812 | 2605 | new_data_offset = data_offset; |
19ceb16d N |
2606 | } |
2607 | } else { | |
2608 | if (dir == 0) { | |
f9b08fec N |
2609 | /* can move up or down. If 'data_offset' |
2610 | * was set we would have already decided, | |
2611 | * so just choose direction with most space. | |
19ceb16d | 2612 | */ |
c0f0d812 | 2613 | if (before > after) |
19ceb16d N |
2614 | dir = -1; |
2615 | else | |
2616 | dir = 1; | |
19ceb16d | 2617 | } |
b397d7f3 N |
2618 | sysfs_set_str(sra, NULL, "reshape_direction", |
2619 | dir == 1 ? "backwards" : "forwards"); | |
c0f0d812 N |
2620 | if (dir > 0) { |
2621 | /* Increase data offset */ | |
f9b08fec | 2622 | if (after < min) { |
6a23fb9d N |
2623 | if (can_fallback) |
2624 | goto fallback; | |
ed503f89 | 2625 | pr_err("Insufficient tail-space for reshape on %s\n", |
19ceb16d N |
2626 | dn); |
2627 | goto release; | |
2628 | } | |
5582b118 | 2629 | if (data_offset != INVALID_SECTORS && |
c0f0d812 | 2630 | data_offset < sd->data_offset + min) { |
ed503f89 | 2631 | pr_err("--data-offset too small on %s\n", |
19ceb16d N |
2632 | dn); |
2633 | goto release; | |
2634 | } | |
5582b118 | 2635 | if (data_offset != INVALID_SECTORS) |
c0f0d812 | 2636 | new_data_offset = data_offset; |
534f5432 N |
2637 | else |
2638 | new_data_offset = choose_offset(sd->data_offset, | |
2639 | sd->data_offset + after, | |
2640 | sd->data_offset + min, | |
2641 | sd->data_offset + after); | |
c0f0d812 N |
2642 | } else { |
2643 | /* Decrease data offset */ | |
f9b08fec | 2644 | if (before < min) { |
6a23fb9d N |
2645 | if (can_fallback) |
2646 | goto fallback; | |
ed503f89 | 2647 | pr_err("insufficient head-room on %s\n", |
19ceb16d N |
2648 | dn); |
2649 | goto release; | |
2650 | } | |
5582b118 | 2651 | if (data_offset != INVALID_SECTORS && |
cab114c5 CH |
2652 | data_offset > sd->data_offset - min) { |
2653 | pr_err("--data-offset too large on %s\n", | |
19ceb16d N |
2654 | dn); |
2655 | goto release; | |
2656 | } | |
5582b118 | 2657 | if (data_offset != INVALID_SECTORS) |
c0f0d812 | 2658 | new_data_offset = data_offset; |
534f5432 N |
2659 | else |
2660 | new_data_offset = choose_offset(sd->data_offset - before, | |
2661 | sd->data_offset, | |
2662 | sd->data_offset - before, | |
2663 | sd->data_offset - min); | |
19ceb16d N |
2664 | } |
2665 | } | |
a6a78630 N |
2666 | err = sysfs_set_num(sra, sd, "new_offset", new_data_offset); |
2667 | if (err < 0 && errno == E2BIG) { | |
2668 | /* try again after increasing data size to max */ | |
2669 | err = sysfs_set_num(sra, sd, "size", 0); | |
2670 | if (err < 0 && errno == EINVAL && | |
2671 | !(sd->disk.state & (1<<MD_DISK_SYNC))) { | |
2672 | /* some kernels have a bug where you cannot | |
2673 | * use '0' on spare devices. */ | |
2674 | sysfs_set_num(sra, sd, "size", | |
2675 | (sra->component_size + after)/2); | |
2676 | } | |
2677 | err = sysfs_set_num(sra, sd, "new_offset", | |
2678 | new_data_offset); | |
2679 | } | |
2680 | if (err < 0) { | |
d7e1f52b | 2681 | if (errno == E2BIG && data_offset != INVALID_SECTORS) { |
358ef9bf | 2682 | pr_err("data-offset is too big for %s\n", dn); |
d7e1f52b N |
2683 | goto release; |
2684 | } | |
9ad2a640 N |
2685 | if (sd == sra->devs && |
2686 | (errno == ENOENT || errno == E2BIG)) | |
2687 | /* Early kernel, no 'new_offset' file, | |
2688 | * or kernel doesn't like us. | |
93f174b9 N |
2689 | * For RAID5/6 this is not fatal |
2690 | */ | |
2691 | return 1; | |
358ef9bf | 2692 | pr_err("Cannot set new_offset for %s\n", dn); |
19ceb16d N |
2693 | break; |
2694 | } | |
2695 | } | |
13bbb145 N |
2696 | return err; |
2697 | release: | |
2698 | return -1; | |
6a23fb9d N |
2699 | fallback: |
2700 | /* Just use a backup file */ | |
2701 | return 1; | |
13bbb145 N |
2702 | } |
2703 | ||
2704 | static int raid10_reshape(char *container, int fd, char *devname, | |
2705 | struct supertype *st, struct mdinfo *info, | |
2706 | struct reshape *reshape, | |
2707 | unsigned long long data_offset, | |
2708 | int force, int verbose) | |
2709 | { | |
2710 | /* Changing raid_disks, layout, chunksize or possibly | |
2711 | * just data_offset for a RAID10. | |
2712 | * We must always change data_offset. We change by at least | |
89ecd3cf | 2713 | * ->min_offset_change which is the largest of the old and new |
13bbb145 N |
2714 | * chunk sizes. |
2715 | * If raid_disks is increasing, then data_offset must decrease | |
2716 | * by at least this copy size. | |
2717 | * If raid_disks is unchanged, data_offset must increase or | |
89ecd3cf | 2718 | * decrease by at least min_offset_change but preferably by much more. |
13bbb145 N |
2719 | * We choose half of the available space. |
2720 | * If raid_disks is decreasing, data_offset must increase by | |
89ecd3cf | 2721 | * at least min_offset_change. To allow of this, component_size |
13bbb145 N |
2722 | * must be decreased by the same amount. |
2723 | * | |
2724 | * So we calculate the required minimum and direction, possibly | |
2725 | * reduce the component_size, then iterate through the devices | |
2726 | * and set the new_data_offset. | |
2727 | * If that all works, we set chunk_size, layout, raid_disks, and start | |
2728 | * 'reshape' | |
2729 | */ | |
2730 | struct mdinfo *sra; | |
2731 | unsigned long long min; | |
2732 | int err = 0; | |
2733 | ||
2734 | sra = sysfs_read(fd, NULL, | |
2735 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK | |
2736 | ); | |
2737 | if (!sra) { | |
358ef9bf | 2738 | pr_err("%s: Cannot get array details from sysfs\n", devname); |
13bbb145 N |
2739 | goto release; |
2740 | } | |
89ecd3cf | 2741 | min = reshape->min_offset_change; |
13bbb145 N |
2742 | |
2743 | if (info->delta_disks) | |
2744 | sysfs_set_str(sra, NULL, "reshape_direction", | |
2745 | info->delta_disks < 0 ? "backwards" : "forwards"); | |
358ef9bf | 2746 | if (info->delta_disks < 0 && info->space_after < min) { |
13bbb145 | 2747 | int rv = sysfs_set_num(sra, NULL, "component_size", |
358ef9bf | 2748 | (sra->component_size - min)/2); |
13bbb145 | 2749 | if (rv) { |
ed503f89 | 2750 | pr_err("cannot reduce component size\n"); |
13bbb145 N |
2751 | goto release; |
2752 | } | |
2753 | } | |
358ef9bf JS |
2754 | err = set_new_data_offset(sra, st, devname, info->delta_disks, |
2755 | data_offset, min, 0); | |
93f174b9 N |
2756 | if (err == 1) { |
2757 | pr_err("Cannot set new_data_offset: RAID10 reshape not\n"); | |
2758 | cont_err("supported on this kernel\n"); | |
2759 | err = -1; | |
2760 | } | |
13bbb145 N |
2761 | if (err < 0) |
2762 | goto release; | |
2763 | ||
ee2429e0 | 2764 | if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0) |
19ceb16d | 2765 | err = errno; |
358ef9bf JS |
2766 | if (!err && sysfs_set_num(sra, NULL, "layout", |
2767 | reshape->after.layout) < 0) | |
19ceb16d | 2768 | err = errno; |
358ef9bf JS |
2769 | if (!err && |
2770 | sysfs_set_num(sra, NULL, "raid_disks", | |
2771 | info->array.raid_disks + info->delta_disks) < 0) | |
19ceb16d N |
2772 | err = errno; |
2773 | if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) | |
2774 | err = errno; | |
2775 | if (err) { | |
ed503f89 | 2776 | pr_err("Cannot set array shape for %s\n", |
d33f1518 N |
2777 | devname); |
2778 | if (err == EBUSY && | |
2779 | (info->array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
7a862a02 | 2780 | cont_err(" Bitmap must be removed before shape can be changed\n"); |
d33f1518 | 2781 | goto release; |
19ceb16d N |
2782 | } |
2783 | sysfs_free(sra); | |
2784 | return 0; | |
2785 | release: | |
2786 | sysfs_free(sra); | |
2787 | return 1; | |
2788 | } | |
2789 | ||
ee2429e0 N |
2790 | static void get_space_after(int fd, struct supertype *st, struct mdinfo *info) |
2791 | { | |
2792 | struct mdinfo *sra, *sd; | |
f3f09a52 LD |
2793 | /* Initialisation to silence compiler warning */ |
2794 | unsigned long long min_space_before = 0, min_space_after = 0; | |
ee2429e0 N |
2795 | int first = 1; |
2796 | ||
4dd2df09 | 2797 | sra = sysfs_read(fd, NULL, GET_DEVS); |
ee2429e0 N |
2798 | if (!sra) |
2799 | return; | |
2800 | for (sd = sra->devs; sd; sd = sd->next) { | |
2801 | char *dn; | |
2802 | int dfd; | |
2803 | struct supertype *st2; | |
2804 | struct mdinfo info2; | |
2805 | ||
2806 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2807 | continue; | |
2808 | dn = map_dev(sd->disk.major, sd->disk.minor, 0); | |
2809 | dfd = dev_open(dn, O_RDONLY); | |
2810 | if (dfd < 0) | |
2811 | break; | |
2812 | st2 = dup_super(st); | |
2813 | if (st2->ss->load_super(st2,dfd, NULL)) { | |
2814 | close(dfd); | |
2815 | free(st2); | |
2816 | break; | |
2817 | } | |
2818 | close(dfd); | |
2819 | st2->ss->getinfo_super(st2, &info2, NULL); | |
2820 | st2->ss->free_super(st2); | |
2821 | free(st2); | |
2822 | if (first || | |
2823 | min_space_before > info2.space_before) | |
2824 | min_space_before = info2.space_before; | |
2825 | if (first || | |
2826 | min_space_after > info2.space_after) | |
2827 | min_space_after = info2.space_after; | |
2828 | first = 0; | |
2829 | } | |
2830 | if (sd == NULL && !first) { | |
2831 | info->space_after = min_space_after; | |
2832 | info->space_before = min_space_before; | |
2833 | } | |
2834 | sysfs_free(sra); | |
2835 | } | |
2836 | ||
434d167e N |
2837 | static void update_cache_size(char *container, struct mdinfo *sra, |
2838 | struct mdinfo *info, | |
2839 | int disks, unsigned long long blocks) | |
2840 | { | |
2841 | /* Check that the internal stripe cache is | |
2842 | * large enough, or it won't work. | |
2843 | * It must hold at least 4 stripes of the larger | |
2844 | * chunk size | |
2845 | */ | |
2846 | unsigned long cache; | |
2847 | cache = max(info->array.chunk_size, info->new_chunk); | |
2848 | cache *= 4; /* 4 stripes minimum */ | |
2849 | cache /= 512; /* convert to sectors */ | |
2850 | /* make sure there is room for 'blocks' with a bit to spare */ | |
2851 | if (cache < 16 + blocks / disks) | |
2852 | cache = 16 + blocks / disks; | |
caf9ac0c | 2853 | cache /= (4096/512); /* Convert from sectors to pages */ |
434d167e N |
2854 | |
2855 | if (sra->cache_size < cache) | |
2856 | subarray_set_num(container, sra, "stripe_cache_size", | |
2857 | cache+1); | |
2858 | } | |
2859 | ||
77afa056 N |
2860 | static int impose_reshape(struct mdinfo *sra, |
2861 | struct mdinfo *info, | |
2862 | struct supertype *st, | |
2863 | int fd, | |
2864 | int restart, | |
2865 | char *devname, char *container, | |
2866 | struct reshape *reshape) | |
2867 | { | |
2868 | struct mdu_array_info_s array; | |
2869 | ||
2870 | sra->new_chunk = info->new_chunk; | |
2871 | ||
2872 | if (restart) { | |
2873 | /* for external metadata checkpoint saved by mdmon can be lost | |
2874 | * or missed /due to e.g. crash/. Check if md is not during | |
2875 | * restart farther than metadata points to. | |
2876 | * If so, this means metadata information is obsolete. | |
2877 | */ | |
2878 | if (st->ss->external) | |
2879 | verify_reshape_position(info, reshape->level); | |
2880 | sra->reshape_progress = info->reshape_progress; | |
2881 | } else { | |
2882 | sra->reshape_progress = 0; | |
2883 | if (reshape->after.data_disks < reshape->before.data_disks) | |
2884 | /* start from the end of the new array */ | |
2885 | sra->reshape_progress = (sra->component_size | |
2886 | * reshape->after.data_disks); | |
2887 | } | |
2888 | ||
9cd39f01 | 2889 | md_get_array_info(fd, &array); |
77afa056 N |
2890 | if (info->array.chunk_size == info->new_chunk && |
2891 | reshape->before.layout == reshape->after.layout && | |
2892 | st->ss->external == 0) { | |
2893 | /* use SET_ARRAY_INFO but only if reshape hasn't started */ | |
2894 | array.raid_disks = reshape->after.data_disks + reshape->parity; | |
018a4882 | 2895 | if (!restart && md_set_array_info(fd, &array) != 0) { |
77afa056 N |
2896 | int err = errno; |
2897 | ||
2898 | pr_err("Cannot set device shape for %s: %s\n", | |
2899 | devname, strerror(errno)); | |
2900 | ||
2901 | if (err == EBUSY && | |
2902 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
7a862a02 | 2903 | cont_err("Bitmap must be removed before shape can be changed\n"); |
77afa056 N |
2904 | |
2905 | goto release; | |
2906 | } | |
2907 | } else if (!restart) { | |
2908 | /* set them all just in case some old 'new_*' value | |
2909 | * persists from some earlier problem. | |
2910 | */ | |
2911 | int err = 0; | |
2912 | if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0) | |
2913 | err = errno; | |
2914 | if (!err && sysfs_set_num(sra, NULL, "layout", | |
2915 | reshape->after.layout) < 0) | |
2916 | err = errno; | |
2917 | if (!err && subarray_set_num(container, sra, "raid_disks", | |
2918 | reshape->after.data_disks + | |
2919 | reshape->parity) < 0) | |
2920 | err = errno; | |
2921 | if (err) { | |
358ef9bf | 2922 | pr_err("Cannot set device shape for %s\n", devname); |
77afa056 N |
2923 | |
2924 | if (err == EBUSY && | |
2925 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
7a862a02 | 2926 | cont_err("Bitmap must be removed before shape can be changed\n"); |
77afa056 N |
2927 | goto release; |
2928 | } | |
2929 | } | |
2930 | return 0; | |
2931 | release: | |
2932 | return -1; | |
2933 | } | |
2934 | ||
97e3a6a0 N |
2935 | static int impose_level(int fd, int level, char *devname, int verbose) |
2936 | { | |
2937 | char *c; | |
2938 | struct mdu_array_info_s array; | |
2939 | struct mdinfo info; | |
dae13137 JS |
2940 | |
2941 | if (sysfs_init(&info, fd, NULL)) { | |
ebf3be99 | 2942 | pr_err("failed to initialize sysfs.\n"); |
dae13137 JS |
2943 | return 1; |
2944 | } | |
97e3a6a0 | 2945 | |
9cd39f01 | 2946 | md_get_array_info(fd, &array); |
70f1ff42 | 2947 | if (level == 0 && is_level456(array.level)) { |
97e3a6a0 N |
2948 | /* To convert to RAID0 we need to fail and |
2949 | * remove any non-data devices. */ | |
2950 | int found = 0; | |
2951 | int d; | |
2952 | int data_disks = array.raid_disks - 1; | |
2953 | if (array.level == 6) | |
2954 | data_disks -= 1; | |
358ef9bf | 2955 | if (array.level == 5 && array.layout != ALGORITHM_PARITY_N) |
97e3a6a0 | 2956 | return -1; |
358ef9bf | 2957 | if (array.level == 6 && array.layout != ALGORITHM_PARITY_N_6) |
97e3a6a0 N |
2958 | return -1; |
2959 | sysfs_set_str(&info, NULL,"sync_action", "idle"); | |
2960 | /* First remove any spares so no recovery starts */ | |
2961 | for (d = 0, found = 0; | |
358ef9bf | 2962 | d < MAX_DISKS && found < array.nr_disks; d++) { |
5e7be838 | 2963 | mdu_disk_info_t disk; |
97e3a6a0 | 2964 | disk.number = d; |
d97572f5 | 2965 | if (md_get_disk_info(fd, &disk) < 0) |
97e3a6a0 N |
2966 | continue; |
2967 | if (disk.major == 0 && disk.minor == 0) | |
2968 | continue; | |
2969 | found++; | |
fc54fe7a JS |
2970 | if ((disk.state & (1 << MD_DISK_ACTIVE)) && |
2971 | disk.raid_disk < data_disks) | |
97e3a6a0 N |
2972 | /* keep this */ |
2973 | continue; | |
2974 | ioctl(fd, HOT_REMOVE_DISK, | |
2975 | makedev(disk.major, disk.minor)); | |
2976 | } | |
2977 | /* Now fail anything left */ | |
9cd39f01 | 2978 | md_get_array_info(fd, &array); |
97e3a6a0 | 2979 | for (d = 0, found = 0; |
358ef9bf | 2980 | d < MAX_DISKS && found < array.nr_disks; d++) { |
97e3a6a0 N |
2981 | mdu_disk_info_t disk; |
2982 | disk.number = d; | |
d97572f5 | 2983 | if (md_get_disk_info(fd, &disk) < 0) |
97e3a6a0 N |
2984 | continue; |
2985 | if (disk.major == 0 && disk.minor == 0) | |
2986 | continue; | |
2987 | found++; | |
fc54fe7a JS |
2988 | if ((disk.state & (1 << MD_DISK_ACTIVE)) && |
2989 | disk.raid_disk < data_disks) | |
97e3a6a0 N |
2990 | /* keep this */ |
2991 | continue; | |
2992 | ioctl(fd, SET_DISK_FAULTY, | |
2993 | makedev(disk.major, disk.minor)); | |
1ab9ed2a | 2994 | hot_remove_disk(fd, makedev(disk.major, disk.minor), 1); |
97e3a6a0 N |
2995 | } |
2996 | } | |
2997 | c = map_num(pers, level); | |
2998 | if (c) { | |
2999 | int err = sysfs_set_str(&info, NULL, "level", c); | |
3000 | if (err) { | |
3001 | err = errno; | |
3002 | pr_err("%s: could not set level to %s\n", | |
3003 | devname, c); | |
3004 | if (err == EBUSY && | |
3005 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
7a862a02 | 3006 | cont_err("Bitmap must be removed before level can be changed\n"); |
97e3a6a0 N |
3007 | return err; |
3008 | } | |
3009 | if (verbose >= 0) | |
358ef9bf | 3010 | pr_err("level of %s changed to %s\n", devname, c); |
97e3a6a0 N |
3011 | } |
3012 | return 0; | |
3013 | } | |
3014 | ||
84d11e6c N |
3015 | int sigterm = 0; |
3016 | static void catch_term(int sig) | |
3017 | { | |
3018 | sigterm = 1; | |
3019 | } | |
3020 | ||
5da9ab98 N |
3021 | static int reshape_array(char *container, int fd, char *devname, |
3022 | struct supertype *st, struct mdinfo *info, | |
e2e53a2d | 3023 | int force, struct mddev_dev *devlist, |
19ceb16d | 3024 | unsigned long long data_offset, |
ba728be7 | 3025 | char *backup_file, int verbose, int forked, |
b76b30e0 | 3026 | int restart, int freeze_reshape) |
5da9ab98 N |
3027 | { |
3028 | struct reshape reshape; | |
3029 | int spares_needed; | |
3030 | char *msg; | |
3031 | int orig_level = UnSet; | |
434d167e | 3032 | int odisks; |
1f9b0e28 | 3033 | int delayed; |
7bc71196 | 3034 | |
5da9ab98 N |
3035 | struct mdu_array_info_s array; |
3036 | char *c; | |
e86c9dd6 | 3037 | |
e2e53a2d N |
3038 | struct mddev_dev *dv; |
3039 | int added_disks; | |
3040 | ||
d152f53e JS |
3041 | int *fdlist = NULL; |
3042 | unsigned long long *offsets = NULL; | |
5da9ab98 N |
3043 | int d; |
3044 | int nrdisks; | |
3045 | int err; | |
da8100ca | 3046 | unsigned long blocks; |
5da9ab98 N |
3047 | unsigned long long array_size; |
3048 | int done; | |
cf6ac177 | 3049 | struct mdinfo *sra = NULL; |
13ffbe89 | 3050 | char buf[20]; |
7bc71196 | 3051 | |
9468aeac N |
3052 | /* when reshaping a RAID0, the component_size might be zero. |
3053 | * So try to fix that up. | |
3054 | */ | |
9cd39f01 | 3055 | if (md_get_array_info(fd, &array) != 0) { |
9468aeac N |
3056 | dprintf("Cannot get array information.\n"); |
3057 | goto release; | |
3058 | } | |
3059 | if (array.level == 0 && info->component_size == 0) { | |
3060 | get_dev_size(fd, NULL, &array_size); | |
3061 | info->component_size = array_size / array.raid_disks; | |
3062 | } | |
3063 | ||
ee2429e0 N |
3064 | if (array.level == 10) |
3065 | /* Need space_after info */ | |
3066 | get_space_after(fd, st, info); | |
3067 | ||
3cb2aed2 N |
3068 | if (info->reshape_active) { |
3069 | int new_level = info->new_level; | |
3070 | info->new_level = UnSet; | |
ce4783d3 N |
3071 | if (info->delta_disks > 0) |
3072 | info->array.raid_disks -= info->delta_disks; | |
a73b0081 | 3073 | msg = analyse_change(devname, info, &reshape); |
3cb2aed2 | 3074 | info->new_level = new_level; |
ce4783d3 N |
3075 | if (info->delta_disks > 0) |
3076 | info->array.raid_disks += info->delta_disks; | |
fcdfb814 AK |
3077 | if (!restart) |
3078 | /* Make sure the array isn't read-only */ | |
3079 | ioctl(fd, RESTART_ARRAY_RW, 0); | |
3cb2aed2 | 3080 | } else |
a73b0081 | 3081 | msg = analyse_change(devname, info, &reshape); |
5da9ab98 | 3082 | if (msg) { |
a73b0081 N |
3083 | /* if msg == "", error has already been printed */ |
3084 | if (msg[0]) | |
3085 | pr_err("%s\n", msg); | |
631d7405 | 3086 | goto release; |
5da9ab98 | 3087 | } |
358ef9bf JS |
3088 | if (restart && (reshape.level != info->array.level || |
3089 | reshape.before.layout != info->array.layout || | |
3090 | reshape.before.data_disks + reshape.parity != | |
3091 | info->array.raid_disks - max(0, info->delta_disks))) { | |
7a862a02 | 3092 | pr_err("reshape info is not in native format - cannot continue.\n"); |
20a40eca N |
3093 | goto release; |
3094 | } | |
a93f87ee | 3095 | |
13ffbe89 | 3096 | if (st->ss->external && restart && (info->reshape_progress == 0) && |
358ef9bf JS |
3097 | !((sysfs_get_str(info, NULL, "sync_action", |
3098 | buf, sizeof(buf)) > 0) && | |
13ffbe89 | 3099 | (strncmp(buf, "reshape", 7) == 0))) { |
e1dd332a AK |
3100 | /* When reshape is restarted from '0', very begin of array |
3101 | * it is possible that for external metadata reshape and array | |
3102 | * configuration doesn't happen. | |
3103 | * Check if md has the same opinion, and reshape is restarted | |
3104 | * from 0. If so, this is regular reshape start after reshape | |
3105 | * switch in metadata to next array only. | |
3106 | */ | |
3107 | if ((verify_reshape_position(info, reshape.level) >= 0) && | |
3108 | (info->reshape_progress == 0)) | |
3109 | restart = 0; | |
3110 | } | |
a93f87ee | 3111 | if (restart) { |
358ef9bf JS |
3112 | /* |
3113 | * reshape already started. just skip to monitoring | |
3114 | * the reshape | |
3115 | */ | |
a93f87ee N |
3116 | if (reshape.backup_blocks == 0) |
3117 | return 0; | |
8ecf12b9 N |
3118 | if (restart & RESHAPE_NO_BACKUP) |
3119 | return 0; | |
dfa4d769 N |
3120 | |
3121 | /* Need 'sra' down at 'started:' */ | |
3122 | sra = sysfs_read(fd, NULL, | |
358ef9bf JS |
3123 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE| |
3124 | GET_CHUNK|GET_CACHE); | |
dfa4d769 N |
3125 | if (!sra) { |
3126 | pr_err("%s: Cannot get array details from sysfs\n", | |
3127 | devname); | |
3128 | goto release; | |
3129 | } | |
54ded86f N |
3130 | |
3131 | if (!backup_file) | |
3132 | backup_file = locate_backup(sra->sys_name); | |
3133 | ||
a93f87ee N |
3134 | goto started; |
3135 | } | |
a9c3e78f AK |
3136 | /* The container is frozen but the array may not be. |
3137 | * So freeze the array so spares don't get put to the wrong use | |
3138 | * FIXME there should probably be a cleaner separation between | |
3139 | * freeze_array and freeze_container. | |
3140 | */ | |
3141 | sysfs_freeze_array(info); | |
e06c4e59 | 3142 | /* Check we have enough spares to not be degraded */ |
e2e53a2d N |
3143 | added_disks = 0; |
3144 | for (dv = devlist; dv ; dv=dv->next) | |
3145 | added_disks++; | |
5da9ab98 | 3146 | spares_needed = max(reshape.before.data_disks, |
358ef9bf JS |
3147 | reshape.after.data_disks) + |
3148 | reshape.parity - array.raid_disks; | |
7bc71196 | 3149 | |
358ef9bf | 3150 | if (!force && info->new_level > 1 && info->array.level > 1 && |
e2e53a2d | 3151 | spares_needed > info->array.spare_disks + added_disks) { |
7a862a02 | 3152 | pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n" |
e7b84f9d N |
3153 | " Use --force to over-ride this check.\n", |
3154 | spares_needed, | |
3155 | spares_needed == 1 ? "" : "s", | |
3156 | info->array.spare_disks + added_disks); | |
631d7405 | 3157 | goto release; |
7bc71196 | 3158 | } |
e06c4e59 N |
3159 | /* Check we have enough spares to not fail */ |
3160 | spares_needed = max(reshape.before.data_disks, | |
3161 | reshape.after.data_disks) | |
3162 | - array.raid_disks; | |
3163 | if ((info->new_level > 1 || info->new_level == 0) && | |
e2e53a2d | 3164 | spares_needed > info->array.spare_disks +added_disks) { |
7a862a02 | 3165 | pr_err("Need %d spare%s to create working array, and only have %d.\n", |
358ef9bf | 3166 | spares_needed, spares_needed == 1 ? "" : "s", |
e7b84f9d | 3167 | info->array.spare_disks + added_disks); |
e06c4e59 N |
3168 | goto release; |
3169 | } | |
e86c9dd6 | 3170 | |
eae35f5c | 3171 | if (reshape.level != array.level) { |
97e3a6a0 N |
3172 | int err = impose_level(fd, reshape.level, devname, verbose); |
3173 | if (err) | |
631d7405 | 3174 | goto release; |
97e3a6a0 N |
3175 | info->new_layout = UnSet; /* after level change, |
3176 | * layout is meaningless */ | |
eae35f5c | 3177 | orig_level = array.level; |
3cd4e7c4 | 3178 | sysfs_freeze_array(info); |
e86c9dd6 | 3179 | |
16d4d84e AK |
3180 | if (reshape.level > 0 && st->ss->external) { |
3181 | /* make sure mdmon is aware of the new level */ | |
4dd2df09 | 3182 | if (mdmon_running(container)) |
78340e26 AK |
3183 | flush_mdmon(container); |
3184 | ||
4dd2df09 N |
3185 | if (!mdmon_running(container)) |
3186 | start_mdmon(container); | |
16d4d84e | 3187 | ping_monitor(container); |
358ef9bf | 3188 | if (mdmon_running(container) && st->update_tail == NULL) |
e919fb0a | 3189 | st->update_tail = &st->updates; |
16d4d84e | 3190 | } |
5da9ab98 | 3191 | } |
5da9ab98 N |
3192 | /* ->reshape_super might have chosen some spares from the |
3193 | * container that it wants to be part of the new array. | |
3194 | * We can collect them with ->container_content and give | |
3195 | * them to the kernel. | |
3196 | */ | |
3197 | if (st->ss->reshape_super && st->ss->container_content) { | |
3198 | char *subarray = strchr(info->text_version+1, '/')+1; | |
3199 | struct mdinfo *info2 = | |
3200 | st->ss->container_content(st, subarray); | |
3201 | struct mdinfo *d; | |
3202 | ||
2dd0d697 | 3203 | if (info2) { |
dae13137 | 3204 | if (sysfs_init(info2, fd, st->devnm)) { |
8e5b52cd | 3205 | pr_err("unable to initialize sysfs for %s\n", |
dae13137 JS |
3206 | st->devnm); |
3207 | free(info2); | |
3208 | goto release; | |
3209 | } | |
0d5ac3c6 N |
3210 | /* When increasing number of devices, we need to set |
3211 | * new raid_disks before adding these, or they might | |
3212 | * be rejected. | |
3213 | */ | |
3214 | if (reshape.backup_blocks && | |
358ef9bf JS |
3215 | reshape.after.data_disks > |
3216 | reshape.before.data_disks) | |
0d5ac3c6 N |
3217 | subarray_set_num(container, info2, "raid_disks", |
3218 | reshape.after.data_disks + | |
3219 | reshape.parity); | |
5da9ab98 N |
3220 | for (d = info2->devs; d; d = d->next) { |
3221 | if (d->disk.state == 0 && | |
3222 | d->disk.raid_disk >= 0) { | |
3223 | /* This is a spare that wants to | |
3224 | * be part of the array. | |
3225 | */ | |
3226 | add_disk(fd, st, info2, d); | |
3227 | } | |
9860f271 | 3228 | } |
2dd0d697 AK |
3229 | sysfs_free(info2); |
3230 | } | |
5da9ab98 | 3231 | } |
e2e53a2d N |
3232 | /* We might have been given some devices to add to the |
3233 | * array. Now that the array has been changed to the right | |
3234 | * level and frozen, we can safely add them. | |
3235 | */ | |
31dbeda7 | 3236 | if (devlist) { |
358ef9bf | 3237 | if (Manage_subdevs(devname, fd, devlist, verbose, 0, NULL, 0)) |
31dbeda7 GJ |
3238 | goto release; |
3239 | } | |
1686dc25 | 3240 | |
e09233d0 | 3241 | if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS) |
8876bf0b | 3242 | reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512; |
b4f8e38b | 3243 | if (reshape.backup_blocks == 0) { |
5da9ab98 N |
3244 | /* No restriping needed, but we might need to impose |
3245 | * some more changes: layout, raid_disks, chunk_size | |
e86c9dd6 | 3246 | */ |
24aebf3a | 3247 | /* read current array info */ |
9cd39f01 | 3248 | if (md_get_array_info(fd, &array) != 0) { |
83732c28 | 3249 | dprintf("Cannot get array information.\n"); |
24aebf3a KW |
3250 | goto release; |
3251 | } | |
3252 | /* compare current array info with new values and if | |
3253 | * it is different update them to new */ | |
5da9ab98 | 3254 | if (info->new_layout != UnSet && |
24aebf3a KW |
3255 | info->new_layout != array.layout) { |
3256 | array.layout = info->new_layout; | |
018a4882 | 3257 | if (md_set_array_info(fd, &array) != 0) { |
e7b84f9d | 3258 | pr_err("failed to set new layout\n"); |
631d7405 | 3259 | goto release; |
ba728be7 | 3260 | } else if (verbose >= 0) |
5da9ab98 | 3261 | printf("layout for %s set to %d\n", |
24aebf3a | 3262 | devname, array.layout); |
5da9ab98 | 3263 | } |
358ef9bf JS |
3264 | if (info->delta_disks != UnSet && info->delta_disks != 0 && |
3265 | array.raid_disks != | |
3266 | (info->array.raid_disks + info->delta_disks)) { | |
24aebf3a | 3267 | array.raid_disks += info->delta_disks; |
018a4882 | 3268 | if (md_set_array_info(fd, &array) != 0) { |
e7b84f9d | 3269 | pr_err("failed to set raid disks\n"); |
631d7405 | 3270 | goto release; |
ba728be7 | 3271 | } else if (verbose >= 0) { |
5da9ab98 | 3272 | printf("raid_disks for %s set to %d\n", |
24aebf3a KW |
3273 | devname, array.raid_disks); |
3274 | } | |
5da9ab98 N |
3275 | } |
3276 | if (info->new_chunk != 0 && | |
24aebf3a | 3277 | info->new_chunk != array.chunk_size) { |
5da9ab98 N |
3278 | if (sysfs_set_num(info, NULL, |
3279 | "chunk_size", info->new_chunk) != 0) { | |
e7b84f9d | 3280 | pr_err("failed to set chunk size\n"); |
631d7405 | 3281 | goto release; |
ba728be7 | 3282 | } else if (verbose >= 0) |
5da9ab98 | 3283 | printf("chunk size for %s set to %d\n", |
76d505de | 3284 | devname, info->new_chunk); |
4725bc31 | 3285 | } |
e35b189b | 3286 | unfreeze(st); |
631d7405 | 3287 | return 0; |
5da9ab98 | 3288 | } |
19678e53 | 3289 | |
5da9ab98 N |
3290 | /* |
3291 | * There are three possibilities. | |
3292 | * 1/ The array will shrink. | |
3293 | * We need to ensure the reshape will pause before reaching | |
3294 | * the 'critical section'. We also need to fork and wait for | |
24daa16f | 3295 | * that to happen. When it does we |
5da9ab98 N |
3296 | * suspend/backup/complete/unfreeze |
3297 | * | |
3298 | * 2/ The array will not change size. | |
3299 | * This requires that we keep a backup of a sliding window | |
3300 | * so that we can restore data after a crash. So we need | |
3301 | * to fork and monitor progress. | |
3302 | * In future we will allow the data_offset to change, so | |
3303 | * a sliding backup becomes unnecessary. | |
3304 | * | |
3305 | * 3/ The array will grow. This is relatively easy. | |
3306 | * However the kernel's restripe routines will cheerfully | |
3307 | * overwrite some early data before it is safe. So we | |
3308 | * need to make a backup of the early parts of the array | |
3309 | * and be ready to restore it if rebuild aborts very early. | |
3310 | * For externally managed metadata, we still need a forked | |
3311 | * child to monitor the reshape and suspend IO over the region | |
3312 | * that is being reshaped. | |
3313 | * | |
3314 | * We backup data by writing it to one spare, or to a | |
3315 | * file which was given on command line. | |
3316 | * | |
3317 | * In each case, we first make sure that storage is available | |
3318 | * for the required backup. | |
3319 | * Then we: | |
3320 | * - request the shape change. | |
3321 | * - fork to handle backup etc. | |
3322 | */ | |
5da9ab98 N |
3323 | /* Check that we can hold all the data */ |
3324 | get_dev_size(fd, NULL, &array_size); | |
3325 | if (reshape.new_size < (array_size/512)) { | |
e7b84f9d N |
3326 | pr_err("this change will reduce the size of the array.\n" |
3327 | " use --grow --array-size first to truncate array.\n" | |
3328 | " e.g. mdadm --grow %s --array-size %llu\n", | |
3329 | devname, reshape.new_size/2); | |
5da9ab98 N |
3330 | goto release; |
3331 | } | |
7236ee7a | 3332 | |
19ceb16d | 3333 | if (array.level == 10) { |
033d0929 | 3334 | /* Reshaping RAID10 does not require any data backup by |
19ceb16d N |
3335 | * user-space. Instead it requires that the data_offset |
3336 | * is changed to avoid the need for backup. | |
3337 | * So this is handled very separately | |
3338 | */ | |
3339 | if (restart) | |
3340 | /* Nothing to do. */ | |
3341 | return 0; | |
3342 | return raid10_reshape(container, fd, devname, st, info, | |
358ef9bf | 3343 | &reshape, data_offset, force, verbose); |
19ceb16d | 3344 | } |
4dd2df09 | 3345 | sra = sysfs_read(fd, NULL, |
3656edcd | 3346 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK| |
5da9ab98 | 3347 | GET_CACHE); |
5da9ab98 | 3348 | if (!sra) { |
e7b84f9d | 3349 | pr_err("%s: Cannot get array details from sysfs\n", |
5da9ab98 | 3350 | devname); |
5da9ab98 N |
3351 | goto release; |
3352 | } | |
7236ee7a | 3353 | |
8192902f | 3354 | if (!backup_file) |
199f1a1f N |
3355 | switch(set_new_data_offset(sra, st, devname, |
3356 | reshape.after.data_disks - reshape.before.data_disks, | |
8192902f | 3357 | data_offset, |
6a23fb9d | 3358 | reshape.min_offset_change, 1)) { |
63c12c89 N |
3359 | case -1: |
3360 | goto release; | |
3361 | case 0: | |
3362 | /* Updated data_offset, so it's easy now */ | |
3363 | update_cache_size(container, sra, info, | |
3364 | min(reshape.before.data_disks, | |
3365 | reshape.after.data_disks), | |
3366 | reshape.backup_blocks); | |
3367 | ||
3368 | /* Right, everything seems fine. Let's kick things off. | |
3369 | */ | |
3370 | sync_metadata(st); | |
3371 | ||
3372 | if (impose_reshape(sra, info, st, fd, restart, | |
3373 | devname, container, &reshape) < 0) | |
3374 | goto release; | |
3375 | if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) { | |
783bbc2b N |
3376 | struct mdinfo *sd; |
3377 | if (errno != EINVAL) { | |
3378 | pr_err("Failed to initiate reshape!\n"); | |
3379 | goto release; | |
3380 | } | |
3381 | /* revert data_offset and try the old way */ | |
3382 | for (sd = sra->devs; sd; sd = sd->next) { | |
3383 | sysfs_set_num(sra, sd, "new_offset", | |
3384 | sd->data_offset); | |
3385 | sysfs_set_str(sra, NULL, "reshape_direction", | |
3386 | "forwards"); | |
3387 | } | |
3388 | break; | |
63c12c89 | 3389 | } |
97e3a6a0 N |
3390 | if (info->new_level == reshape.level) |
3391 | return 0; | |
3392 | /* need to adjust level when reshape completes */ | |
3393 | switch(fork()) { | |
3394 | case -1: /* ignore error, but don't wait */ | |
3395 | return 0; | |
3396 | default: /* parent */ | |
3397 | return 0; | |
3398 | case 0: | |
ff6bb131 | 3399 | manage_fork_fds(0); |
97e3a6a0 N |
3400 | map_fork(); |
3401 | break; | |
3402 | } | |
3377ee42 | 3403 | close(fd); |
97e3a6a0 | 3404 | wait_reshape(sra); |
3377ee42 N |
3405 | fd = open_dev(sra->sys_name); |
3406 | if (fd >= 0) | |
3407 | impose_level(fd, info->new_level, devname, verbose); | |
63c12c89 N |
3408 | return 0; |
3409 | case 1: /* Couldn't set data_offset, try the old way */ | |
3410 | if (data_offset != INVALID_SECTORS) { | |
3411 | pr_err("Cannot update data_offset on this array\n"); | |
3412 | goto release; | |
3413 | } | |
3414 | break; | |
3415 | } | |
3416 | ||
8ecf12b9 | 3417 | started: |
5da9ab98 N |
3418 | /* Decide how many blocks (sectors) for a reshape |
3419 | * unit. The number we have so far is just a minimum | |
3420 | */ | |
b4f8e38b | 3421 | blocks = reshape.backup_blocks; |
24daa16f | 3422 | if (reshape.before.data_disks == |
5da9ab98 N |
3423 | reshape.after.data_disks) { |
3424 | /* Make 'blocks' bigger for better throughput, but | |
3425 | * not so big that we reject it below. | |
3426 | * Try for 16 megabytes | |
3427 | */ | |
358ef9bf | 3428 | while (blocks * 32 < sra->component_size && blocks < 16*1024*2) |
5da9ab98 N |
3429 | blocks *= 2; |
3430 | } else | |
7a862a02 | 3431 | pr_err("Need to backup %luK of critical section..\n", blocks/2); |
5da9ab98 N |
3432 | |
3433 | if (blocks >= sra->component_size/2) { | |
358ef9bf | 3434 | pr_err("%s: Something wrong - reshape aborted\n", devname); |
5da9ab98 N |
3435 | goto release; |
3436 | } | |
7236ee7a | 3437 | |
5da9ab98 N |
3438 | /* Now we need to open all these devices so we can read/write. |
3439 | */ | |
b8b286a6 N |
3440 | nrdisks = max(reshape.before.data_disks, |
3441 | reshape.after.data_disks) + reshape.parity | |
3442 | + sra->array.spare_disks; | |
503975b9 N |
3443 | fdlist = xcalloc((1+nrdisks), sizeof(int)); |
3444 | offsets = xcalloc((1+nrdisks), sizeof(offsets[0])); | |
e380d3be | 3445 | |
b8b286a6 | 3446 | odisks = reshape.before.data_disks + reshape.parity; |
358ef9bf JS |
3447 | d = reshape_prepare_fdlist(devname, sra, odisks, nrdisks, blocks, |
3448 | backup_file, fdlist, offsets); | |
8e7ddc5f | 3449 | if (d < odisks) { |
5da9ab98 N |
3450 | goto release; |
3451 | } | |
13c37ad3 AK |
3452 | if ((st->ss->manage_reshape == NULL) || |
3453 | (st->ss->recover_backup == NULL)) { | |
3454 | if (backup_file == NULL) { | |
3455 | if (reshape.after.data_disks <= | |
3456 | reshape.before.data_disks) { | |
be7c26b4 N |
3457 | pr_err("%s: Cannot grow - need backup-file\n", |
3458 | devname); | |
3459 | pr_err(" Please provide one with \"--backup=...\"\n"); | |
13c37ad3 | 3460 | goto release; |
8e7ddc5f | 3461 | } else if (d == odisks) { |
7a862a02 | 3462 | pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname); |
13c37ad3 AK |
3463 | goto release; |
3464 | } | |
3465 | } else { | |
3466 | if (!reshape_open_backup_file(backup_file, fd, devname, | |
3467 | (signed)blocks, | |
3468 | fdlist+d, offsets+d, | |
358ef9bf | 3469 | sra->sys_name, restart)) { |
13c37ad3 AK |
3470 | goto release; |
3471 | } | |
3472 | d++; | |
e86c9dd6 | 3473 | } |
5da9ab98 | 3474 | } |
7236ee7a | 3475 | |
434d167e | 3476 | update_cache_size(container, sra, info, |
358ef9bf JS |
3477 | min(reshape.before.data_disks, |
3478 | reshape.after.data_disks), blocks); | |
5da9ab98 N |
3479 | |
3480 | /* Right, everything seems fine. Let's kick things off. | |
3481 | * If only changing raid_disks, use ioctl, else use | |
3482 | * sysfs. | |
3483 | */ | |
3484 | sync_metadata(st); | |
3485 | ||
77afa056 N |
3486 | if (impose_reshape(sra, info, st, fd, restart, |
3487 | devname, container, &reshape) < 0) | |
3488 | goto release; | |
e86c9dd6 | 3489 | |
27a1e5b5 | 3490 | err = start_reshape(sra, restart, reshape.before.data_disks, |
78c93b00 | 3491 | reshape.after.data_disks, st); |
fab32c97 | 3492 | if (err) { |
e7b84f9d | 3493 | pr_err("Cannot %s reshape for %s\n", |
358ef9bf | 3494 | restart ? "continue" : "start", devname); |
fab32c97 AK |
3495 | goto release; |
3496 | } | |
a93f87ee N |
3497 | if (restart) |
3498 | sysfs_set_str(sra, NULL, "array_state", "active"); | |
b76b30e0 AK |
3499 | if (freeze_reshape) { |
3500 | free(fdlist); | |
3501 | free(offsets); | |
3502 | sysfs_free(sra); | |
7a862a02 | 3503 | pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n", |
b76b30e0 AK |
3504 | sra->reshape_progress); |
3505 | return 1; | |
3506 | } | |
5da9ab98 | 3507 | |
ff6bb131 MT |
3508 | if (!forked) |
3509 | if (continue_via_systemd(container ?: sra->sys_name, | |
3510 | GROW_SERVICE)) { | |
b0b67933 N |
3511 | free(fdlist); |
3512 | free(offsets); | |
3513 | sysfs_free(sra); | |
3514 | return 0; | |
5e76dce1 | 3515 | } |
5e76dce1 | 3516 | |
5da9ab98 N |
3517 | /* Now we just need to kick off the reshape and watch, while |
3518 | * handling backups of the data... | |
3519 | * This is all done by a forked background process. | |
3520 | */ | |
3521 | switch(forked ? 0 : fork()) { | |
6b2d630c | 3522 | case -1: |
e7b84f9d | 3523 | pr_err("Cannot run child to monitor reshape: %s\n", |
6b2d630c | 3524 | strerror(errno)); |
6b2d630c | 3525 | abort_reshape(sra); |
631d7405 | 3526 | goto release; |
6b2d630c | 3527 | default: |
d152f53e JS |
3528 | free(fdlist); |
3529 | free(offsets); | |
3530 | sysfs_free(sra); | |
6b2d630c | 3531 | return 0; |
5da9ab98 | 3532 | case 0: |
cc700db3 | 3533 | map_fork(); |
6b2d630c N |
3534 | break; |
3535 | } | |
5da9ab98 | 3536 | |
548e9b91 LG |
3537 | /* Close unused file descriptor in the forked process */ |
3538 | close_fd(&fd); | |
3539 | ||
1f9b0e28 N |
3540 | /* If another array on the same devices is busy, the |
3541 | * reshape will wait for them. This would mean that | |
3542 | * the first section that we suspend will stay suspended | |
3543 | * for a long time. So check on that possibility | |
3544 | * by looking for "DELAYED" in /proc/mdstat, and if found, | |
3545 | * wait a while | |
3546 | */ | |
3547 | do { | |
3548 | struct mdstat_ent *mds, *m; | |
3549 | delayed = 0; | |
a7a0d8a1 | 3550 | mds = mdstat_read(1, 0); |
ae0dcfbd | 3551 | for (m = mds; m; m = m->next) |
4dd2df09 | 3552 | if (strcmp(m->devnm, sra->sys_name) == 0) { |
358ef9bf | 3553 | if (m->resync && m->percent == RESYNC_DELAYED) |
1f9b0e28 N |
3554 | delayed = 1; |
3555 | if (m->resync == 0) | |
3556 | /* Haven't started the reshape thread | |
3557 | * yet, wait a bit | |
3558 | */ | |
3559 | delayed = 2; | |
3560 | break; | |
3561 | } | |
3562 | free_mdstat(mds); | |
3563 | if (delayed == 1 && get_linux_version() < 3007000) { | |
3564 | pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n" | |
3565 | " You might experience problems until other reshapes complete.\n"); | |
3566 | delayed = 0; | |
3567 | } | |
3568 | if (delayed) | |
a7a0d8a1 | 3569 | mdstat_wait(30 - (delayed-1) * 25); |
1f9b0e28 | 3570 | } while (delayed); |
a7a0d8a1 | 3571 | mdstat_close(); |
6b2d630c N |
3572 | if (check_env("MDADM_GROW_VERIFY")) |
3573 | fd = open(devname, O_RDONLY | O_DIRECT); | |
3574 | else | |
3575 | fd = -1; | |
3576 | mlockall(MCL_FUTURE); | |
5da9ab98 | 3577 | |
83a379cf LF |
3578 | if (signal_s(SIGTERM, catch_term) == SIG_ERR) |
3579 | goto release; | |
84d11e6c | 3580 | |
6b2d630c N |
3581 | if (st->ss->external) { |
3582 | /* metadata handler takes it from here */ | |
3583 | done = st->ss->manage_reshape( | |
3584 | fd, sra, &reshape, st, blocks, | |
358ef9bf JS |
3585 | fdlist, offsets, d - odisks, fdlist + odisks, |
3586 | offsets + odisks); | |
6b2d630c N |
3587 | } else |
3588 | done = child_monitor( | |
358ef9bf JS |
3589 | fd, sra, &reshape, st, blocks, fdlist, offsets, |
3590 | d - odisks, fdlist + odisks, offsets + odisks); | |
6b2d630c | 3591 | |
d152f53e JS |
3592 | free(fdlist); |
3593 | free(offsets); | |
3594 | ||
54ded86f N |
3595 | if (backup_file && done) { |
3596 | char *bul; | |
54ded86f N |
3597 | bul = make_backup(sra->sys_name); |
3598 | if (bul) { | |
5e76dce1 | 3599 | char buf[1024]; |
9eb5ce5a | 3600 | int l = readlink(bul, buf, sizeof(buf) - 1); |
5e76dce1 N |
3601 | if (l > 0) { |
3602 | buf[l]=0; | |
3603 | unlink(buf); | |
3604 | } | |
54ded86f N |
3605 | unlink(bul); |
3606 | free(bul); | |
3607 | } | |
5e76dce1 | 3608 | unlink(backup_file); |
54ded86f | 3609 | } |
6b2d630c N |
3610 | if (!done) { |
3611 | abort_reshape(sra); | |
3612 | goto out; | |
3613 | } | |
1cbc5680 N |
3614 | |
3615 | if (!st->ss->external && | |
fc54fe7a JS |
3616 | !(reshape.before.data_disks != reshape.after.data_disks && |
3617 | info->custom_array_size) && info->new_level == reshape.level && | |
1cbc5680 N |
3618 | !forked) { |
3619 | /* no need to wait for the reshape to finish as | |
3620 | * there is nothing more to do. | |
3621 | */ | |
d152f53e | 3622 | sysfs_free(sra); |
1cbc5680 N |
3623 | exit(0); |
3624 | } | |
3625 | wait_reshape(sra); | |
3626 | ||
3627 | if (st->ss->external) { | |
3628 | /* Re-load the metadata as much could have changed */ | |
4dd2df09 | 3629 | int cfd = open_dev(st->container_devnm); |
1cbc5680 | 3630 | if (cfd >= 0) { |
78340e26 | 3631 | flush_mdmon(container); |
1cbc5680 N |
3632 | st->ss->free_super(st); |
3633 | st->ss->load_container(st, cfd, container); | |
3634 | close(cfd); | |
3635 | } | |
3636 | } | |
3637 | ||
6b2d630c N |
3638 | /* set new array size if required customer_array_size is used |
3639 | * by this metadata. | |
3640 | */ | |
358ef9bf | 3641 | if (reshape.before.data_disks != reshape.after.data_disks && |
54397ed9 AK |
3642 | info->custom_array_size) |
3643 | set_array_size(st, info, info->text_version); | |
582496b2 | 3644 | |
6b2d630c | 3645 | if (info->new_level != reshape.level) { |
97e3a6a0 N |
3646 | if (fd < 0) |
3647 | fd = open(devname, O_RDONLY); | |
3648 | impose_level(fd, info->new_level, devname, verbose); | |
3649 | close(fd); | |
e919fb0a AK |
3650 | if (info->new_level == 0) |
3651 | st->update_tail = NULL; | |
7236ee7a | 3652 | } |
6b2d630c | 3653 | out: |
d152f53e | 3654 | sysfs_free(sra); |
6b2d630c N |
3655 | if (forked) |
3656 | return 0; | |
e84f2c00 | 3657 | unfreeze(st); |
6b2d630c | 3658 | exit(0); |
e86c9dd6 | 3659 | |
6b2d630c | 3660 | release: |
d152f53e JS |
3661 | free(fdlist); |
3662 | free(offsets); | |
1cbc5680 | 3663 | if (orig_level != UnSet && sra) { |
7236ee7a N |
3664 | c = map_num(pers, orig_level); |
3665 | if (c && sysfs_set_str(sra, NULL, "level", c) == 0) | |
e7b84f9d | 3666 | pr_err("aborting level change\n"); |
7236ee7a | 3667 | } |
d152f53e | 3668 | sysfs_free(sra); |
9202b8eb N |
3669 | if (!forked) |
3670 | unfreeze(st); | |
631d7405 | 3671 | return 1; |
7236ee7a | 3672 | } |
e86c9dd6 | 3673 | |
9ad6f6e6 AK |
3674 | /* mdfd handle is passed to be closed in child process (after fork). |
3675 | */ | |
493f5dd6 | 3676 | int reshape_container(char *container, char *devname, |
9ad6f6e6 | 3677 | int mdfd, |
24daa16f | 3678 | struct supertype *st, |
5da9ab98 N |
3679 | struct mdinfo *info, |
3680 | int force, | |
b0140ae8 N |
3681 | char *backup_file, int verbose, |
3682 | int forked, int restart, int freeze_reshape) | |
5da9ab98 | 3683 | { |
1bb174ba | 3684 | struct mdinfo *cc = NULL; |
bcc9e9ed | 3685 | int rv = restart; |
4dd2df09 | 3686 | char last_devnm[32] = ""; |
1bb174ba | 3687 | |
d8eb27f7 | 3688 | /* component_size is not meaningful for a container, |
7ec0996c | 3689 | * so pass '0' meaning 'no change' |
d8eb27f7 | 3690 | */ |
493f5dd6 | 3691 | if (!restart && |
7ec0996c | 3692 | reshape_super(st, 0, info->new_level, |
5da9ab98 | 3693 | info->new_layout, info->new_chunk, |
41784c88 | 3694 | info->array.raid_disks, info->delta_disks, |
016e00f5 | 3695 | backup_file, devname, APPLY_METADATA_CHANGES, |
ba728be7 | 3696 | verbose)) { |
9e325442 | 3697 | unfreeze(st); |
5da9ab98 | 3698 | return 1; |
9e325442 | 3699 | } |
5da9ab98 N |
3700 | |
3701 | sync_metadata(st); | |
3702 | ||
1bb174ba AK |
3703 | /* ping monitor to be sure that update is on disk |
3704 | */ | |
3705 | ping_monitor(container); | |
5da9ab98 | 3706 | |
ff6bb131 MT |
3707 | if (!forked && !freeze_reshape) |
3708 | if (continue_via_systemd(container, GROW_SERVICE)) | |
4e0eb0db N |
3709 | return 0; |
3710 | ||
b0140ae8 | 3711 | switch (forked ? 0 : fork()) { |
5da9ab98 N |
3712 | case -1: /* error */ |
3713 | perror("Cannot fork to complete reshape\n"); | |
9e325442 | 3714 | unfreeze(st); |
5da9ab98 N |
3715 | return 1; |
3716 | default: /* parent */ | |
b76b30e0 | 3717 | if (!freeze_reshape) |
7a862a02 | 3718 | printf("%s: multi-array reshape continues in background\n", Name); |
5da9ab98 N |
3719 | return 0; |
3720 | case 0: /* child */ | |
ff6bb131 | 3721 | manage_fork_fds(0); |
cc700db3 | 3722 | map_fork(); |
5da9ab98 N |
3723 | break; |
3724 | } | |
3725 | ||
9ad6f6e6 AK |
3726 | /* close unused handle in child process |
3727 | */ | |
3728 | if (mdfd > -1) | |
3729 | close(mdfd); | |
3730 | ||
1bb174ba AK |
3731 | while(1) { |
3732 | /* For each member array with reshape_active, | |
3733 | * we need to perform the reshape. | |
3734 | * We pick the first array that needs reshaping and | |
3735 | * reshape it. reshape_array() will re-read the metadata | |
3736 | * so the next time through a different array should be | |
3737 | * ready for reshape. | |
493f5dd6 N |
3738 | * It is possible that the 'different' array will not |
3739 | * be assembled yet. In that case we simple exit. | |
3740 | * When it is assembled, the mdadm which assembles it | |
3741 | * will take over the reshape. | |
1bb174ba AK |
3742 | */ |
3743 | struct mdinfo *content; | |
5da9ab98 N |
3744 | int fd; |
3745 | struct mdstat_ent *mdstat; | |
5da9ab98 | 3746 | char *adev; |
13db17bd | 3747 | dev_t devid; |
5da9ab98 | 3748 | |
1bb174ba | 3749 | sysfs_free(cc); |
5da9ab98 | 3750 | |
1bb174ba AK |
3751 | cc = st->ss->container_content(st, NULL); |
3752 | ||
3753 | for (content = cc; content ; content = content->next) { | |
3754 | char *subarray; | |
3755 | if (!content->reshape_active) | |
3756 | continue; | |
3757 | ||
3758 | subarray = strchr(content->text_version+1, '/')+1; | |
4dd2df09 | 3759 | mdstat = mdstat_by_subdev(subarray, container); |
1bb174ba AK |
3760 | if (!mdstat) |
3761 | continue; | |
1ca90aa6 | 3762 | if (mdstat->active == 0) { |
4dd2df09 N |
3763 | pr_err("Skipping inactive array %s.\n", |
3764 | mdstat->devnm); | |
1ca90aa6 AK |
3765 | free_mdstat(mdstat); |
3766 | mdstat = NULL; | |
3767 | continue; | |
3768 | } | |
1bb174ba AK |
3769 | break; |
3770 | } | |
3771 | if (!content) | |
3772 | break; | |
5da9ab98 | 3773 | |
4dd2df09 N |
3774 | devid = devnm2devid(mdstat->devnm); |
3775 | adev = map_dev(major(devid), minor(devid), 0); | |
5da9ab98 | 3776 | if (!adev) |
1bb174ba AK |
3777 | adev = content->text_version; |
3778 | ||
4dd2df09 | 3779 | fd = open_dev(mdstat->devnm); |
97a3490c | 3780 | if (fd < 0) { |
358ef9bf JS |
3781 | pr_err("Device %s cannot be opened for reshape.\n", |
3782 | adev); | |
97a3490c AK |
3783 | break; |
3784 | } | |
3785 | ||
4dd2df09 | 3786 | if (strcmp(last_devnm, mdstat->devnm) == 0) { |
2d04d7e5 AK |
3787 | /* Do not allow for multiple reshape_array() calls for |
3788 | * the same array. | |
3789 | * It can happen when reshape_array() returns without | |
3790 | * error, when reshape is not finished (wrong reshape | |
3791 | * starting/continuation conditions). Mdmon doesn't | |
3792 | * switch to next array in container and reentry | |
3793 | * conditions for the same array occur. | |
3794 | * This is possibly interim until the behaviour of | |
3795 | * reshape_array is resolved(). | |
3796 | */ | |
7a862a02 | 3797 | printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev); |
2d04d7e5 AK |
3798 | close(fd); |
3799 | break; | |
3800 | } | |
4dd2df09 | 3801 | strcpy(last_devnm, mdstat->devnm); |
2d04d7e5 | 3802 | |
dae13137 JS |
3803 | if (sysfs_init(content, fd, mdstat->devnm)) { |
3804 | pr_err("Unable to initialize sysfs for %s\n", | |
3805 | mdstat->devnm); | |
3806 | rv = 1; | |
3807 | break; | |
3808 | } | |
5da9ab98 | 3809 | |
4dd2df09 | 3810 | if (mdmon_running(container)) |
78340e26 AK |
3811 | flush_mdmon(container); |
3812 | ||
1bb174ba | 3813 | rv = reshape_array(container, fd, adev, st, |
ca36d707 | 3814 | content, force, NULL, INVALID_SECTORS, |
ba728be7 | 3815 | backup_file, verbose, 1, restart, |
b76b30e0 | 3816 | freeze_reshape); |
5da9ab98 | 3817 | close(fd); |
b76b30e0 AK |
3818 | |
3819 | if (freeze_reshape) { | |
3820 | sysfs_free(cc); | |
3821 | exit(0); | |
3822 | } | |
3823 | ||
493f5dd6 | 3824 | restart = 0; |
5da9ab98 N |
3825 | if (rv) |
3826 | break; | |
78340e26 | 3827 | |
4dd2df09 | 3828 | if (mdmon_running(container)) |
78340e26 | 3829 | flush_mdmon(container); |
5da9ab98 | 3830 | } |
bcc9e9ed AK |
3831 | if (!rv) |
3832 | unfreeze(st); | |
1bb174ba | 3833 | sysfs_free(cc); |
5da9ab98 N |
3834 | exit(0); |
3835 | } | |
3836 | ||
7236ee7a N |
3837 | /* |
3838 | * We run a child process in the background which performs the following | |
3839 | * steps: | |
3840 | * - wait for resync to reach a certain point | |
3841 | * - suspend io to the following section | |
3842 | * - backup that section | |
3843 | * - allow resync to proceed further | |
3844 | * - resume io | |
3845 | * - discard the backup. | |
3846 | * | |
3847 | * When are combined in slightly different ways in the three cases. | |
3848 | * Grow: | |
3849 | * - suspend/backup/allow/wait/resume/discard | |
3850 | * Shrink: | |
3851 | * - allow/wait/suspend/backup/allow/wait/resume/discard | |
3852 | * same-size: | |
3853 | * - wait/resume/discard/suspend/backup/allow | |
3854 | * | |
3855 | * suspend/backup/allow always come together | |
3856 | * wait/resume/discard do too. | |
3857 | * For the same-size case we have two backups to improve flow. | |
24daa16f | 3858 | * |
7236ee7a | 3859 | */ |
e86c9dd6 | 3860 | |
7443ee81 N |
3861 | int progress_reshape(struct mdinfo *info, struct reshape *reshape, |
3862 | unsigned long long backup_point, | |
3863 | unsigned long long wait_point, | |
3864 | unsigned long long *suspend_point, | |
a6b2d86c | 3865 | unsigned long long *reshape_completed, int *frozen) |
7443ee81 N |
3866 | { |
3867 | /* This function is called repeatedly by the reshape manager. | |
3868 | * It determines how much progress can safely be made and allows | |
3869 | * that progress. | |
3870 | * - 'info' identifies the array and particularly records in | |
3871 | * ->reshape_progress the metadata's knowledge of progress | |
3872 | * This is a sector offset from the start of the array | |
3873 | * of the next array block to be relocated. This number | |
3874 | * may increase from 0 or decrease from array_size, depending | |
3875 | * on the type of reshape that is happening. | |
3876 | * Note that in contrast, 'sync_completed' is a block count of the | |
3877 | * reshape so far. It gives the distance between the start point | |
3878 | * (head or tail of device) and the next place that data will be | |
3879 | * written. It always increases. | |
3880 | * - 'reshape' is the structure created by analyse_change | |
3881 | * - 'backup_point' shows how much the metadata manager has backed-up | |
3882 | * data. For reshapes with increasing progress, it is the next address | |
3883 | * to be backed up, previous addresses have been backed-up. For | |
3884 | * decreasing progress, it is the earliest address that has been | |
3885 | * backed up - later address are also backed up. | |
3886 | * So addresses between reshape_progress and backup_point are | |
3887 | * backed up providing those are in the 'correct' order. | |
3888 | * - 'wait_point' is an array address. When reshape_completed | |
3889 | * passes this point, progress_reshape should return. It might | |
3890 | * return earlier if it determines that ->reshape_progress needs | |
3891 | * to be updated or further backup is needed. | |
3892 | * - suspend_point is maintained by progress_reshape and the caller | |
3893 | * should not touch it except to initialise to zero. | |
3894 | * It is an array address and it only increases in 2.6.37 and earlier. | |
b6b95155 | 3895 | * This makes it difficult to handle reducing reshapes with |
7443ee81 N |
3896 | * external metadata. |
3897 | * However: it is similar to backup_point in that it records the | |
3898 | * other end of a suspended region from reshape_progress. | |
3899 | * it is moved to extend the region that is safe to backup and/or | |
3900 | * reshape | |
3901 | * - reshape_completed is read from sysfs and returned. The caller | |
3902 | * should copy this into ->reshape_progress when it has reason to | |
3903 | * believe that the metadata knows this, and any backup outside this | |
3904 | * has been erased. | |
3905 | * | |
3906 | * Return value is: | |
3907 | * 1 if more data from backup_point - but only as far as suspend_point, | |
3908 | * should be backed up | |
3909 | * 0 if things are progressing smoothly | |
9e034f70 N |
3910 | * -1 if the reshape is finished because it is all done, |
3911 | * -2 if the reshape is finished due to an error. | |
7443ee81 N |
3912 | */ |
3913 | ||
3914 | int advancing = (reshape->after.data_disks | |
3915 | >= reshape->before.data_disks); | |
38dad34a N |
3916 | unsigned long long need_backup; /* All data between start of array and |
3917 | * here will at some point need to | |
3918 | * be backed up. | |
d0ab945e | 3919 | */ |
7443ee81 | 3920 | unsigned long long read_offset, write_offset; |
b4f8e38b | 3921 | unsigned long long write_range; |
7443ee81 | 3922 | unsigned long long max_progress, target, completed; |
d0ab945e N |
3923 | unsigned long long array_size = (info->component_size |
3924 | * reshape->before.data_disks); | |
7443ee81 | 3925 | int fd; |
77a73a17 | 3926 | char buf[20]; |
7443ee81 N |
3927 | |
3928 | /* First, we unsuspend any region that is now known to be safe. | |
3929 | * If suspend_point is on the 'wrong' side of reshape_progress, then | |
3930 | * we don't have or need suspension at the moment. This is true for | |
3931 | * native metadata when we don't need to back-up. | |
3932 | */ | |
3933 | if (advancing) { | |
49cd738b | 3934 | if (info->reshape_progress <= *suspend_point) |
7443ee81 N |
3935 | sysfs_set_num(info, NULL, "suspend_lo", |
3936 | info->reshape_progress); | |
3937 | } else { | |
3938 | /* Note: this won't work in 2.6.37 and before. | |
3939 | * Something somewhere should make sure we don't need it! | |
3940 | */ | |
49cd738b | 3941 | if (info->reshape_progress >= *suspend_point) |
7443ee81 N |
3942 | sysfs_set_num(info, NULL, "suspend_hi", |
3943 | info->reshape_progress); | |
3944 | } | |
3945 | ||
3946 | /* Now work out how far it is safe to progress. | |
3947 | * If the read_offset for ->reshape_progress is less than | |
3948 | * 'blocks' beyond the write_offset, we can only progress as far | |
3949 | * as a backup. | |
3950 | * Otherwise we can progress until the write_offset for the new location | |
3951 | * reaches (within 'blocks' of) the read_offset at the current location. | |
3952 | * However that region must be suspended unless we are using native | |
3953 | * metadata. | |
3954 | * If we need to suspend more, we limit it to 128M per device, which is | |
3955 | * rather arbitrary and should be some time-based calculation. | |
3956 | */ | |
ec757320 N |
3957 | read_offset = info->reshape_progress / reshape->before.data_disks; |
3958 | write_offset = info->reshape_progress / reshape->after.data_disks; | |
b4f8e38b | 3959 | write_range = info->new_chunk/512; |
38dad34a N |
3960 | if (reshape->before.data_disks == reshape->after.data_disks) |
3961 | need_backup = array_size; | |
3962 | else | |
3963 | need_backup = reshape->backup_blocks; | |
7443ee81 | 3964 | if (advancing) { |
38dad34a | 3965 | if (read_offset < write_offset + write_range) |
7443ee81 | 3966 | max_progress = backup_point; |
ddee071d | 3967 | else |
7443ee81 | 3968 | max_progress = |
358ef9bf | 3969 | read_offset * reshape->after.data_disks; |
7443ee81 | 3970 | } else { |
38dad34a N |
3971 | if (read_offset > write_offset - write_range) |
3972 | /* Can only progress as far as has been backed up, | |
3973 | * which must be suspended */ | |
7443ee81 | 3974 | max_progress = backup_point; |
38dad34a N |
3975 | else if (info->reshape_progress <= need_backup) |
3976 | max_progress = backup_point; | |
3977 | else { | |
3978 | if (info->array.major_version >= 0) | |
3979 | /* Can progress until backup is needed */ | |
3980 | max_progress = need_backup; | |
3981 | else { | |
3982 | /* Can progress until metadata update is required */ | |
3983 | max_progress = | |
358ef9bf | 3984 | read_offset * reshape->after.data_disks; |
38dad34a N |
3985 | /* but data must be suspended */ |
3986 | if (max_progress < *suspend_point) | |
3987 | max_progress = *suspend_point; | |
3988 | } | |
7443ee81 N |
3989 | } |
3990 | } | |
3991 | ||
3992 | /* We know it is safe to progress to 'max_progress' providing | |
3993 | * it is suspended or we are using native metadata. | |
3994 | * Consider extending suspend_point 128M per device if it | |
3995 | * is less than 64M per device beyond reshape_progress. | |
3996 | * But always do a multiple of 'blocks' | |
832b2b9a N |
3997 | * FIXME this is too big - it takes to long to complete |
3998 | * this much. | |
7443ee81 N |
3999 | */ |
4000 | target = 64*1024*2 * min(reshape->before.data_disks, | |
24daa16f | 4001 | reshape->after.data_disks); |
b4f8e38b | 4002 | target /= reshape->backup_blocks; |
7443ee81 N |
4003 | if (target < 2) |
4004 | target = 2; | |
b4f8e38b | 4005 | target *= reshape->backup_blocks; |
7443ee81 N |
4006 | |
4007 | /* For externally managed metadata we always need to suspend IO to | |
4008 | * the area being reshaped so we regularly push suspend_point forward. | |
4009 | * For native metadata we only need the suspend if we are going to do | |
4010 | * a backup. | |
4011 | */ | |
4012 | if (advancing) { | |
d7be7d87 JS |
4013 | if ((need_backup > info->reshape_progress || |
4014 | info->array.major_version < 0) && | |
7443ee81 | 4015 | *suspend_point < info->reshape_progress + target) { |
d0ab945e N |
4016 | if (need_backup < *suspend_point + 2 * target) |
4017 | *suspend_point = need_backup; | |
4018 | else if (*suspend_point + 2 * target < array_size) | |
7443ee81 | 4019 | *suspend_point += 2 * target; |
d0ab945e N |
4020 | else |
4021 | *suspend_point = array_size; | |
7443ee81 | 4022 | sysfs_set_num(info, NULL, "suspend_hi", *suspend_point); |
d0ab945e N |
4023 | if (max_progress > *suspend_point) |
4024 | max_progress = *suspend_point; | |
7443ee81 N |
4025 | } |
4026 | } else { | |
38dad34a N |
4027 | if (info->array.major_version >= 0) { |
4028 | /* Only need to suspend when about to backup */ | |
4029 | if (info->reshape_progress < need_backup * 2 && | |
4030 | *suspend_point > 0) { | |
d0ab945e | 4031 | *suspend_point = 0; |
38dad34a | 4032 | sysfs_set_num(info, NULL, "suspend_lo", 0); |
358ef9bf JS |
4033 | sysfs_set_num(info, NULL, "suspend_hi", |
4034 | need_backup); | |
38dad34a N |
4035 | } |
4036 | } else { | |
4037 | /* Need to suspend continually */ | |
4038 | if (info->reshape_progress < *suspend_point) | |
4039 | *suspend_point = info->reshape_progress; | |
4040 | if (*suspend_point + target < info->reshape_progress) | |
4041 | /* No need to move suspend region yet */; | |
4042 | else { | |
4043 | if (*suspend_point >= 2 * target) | |
4044 | *suspend_point -= 2 * target; | |
4045 | else | |
4046 | *suspend_point = 0; | |
4047 | sysfs_set_num(info, NULL, "suspend_lo", | |
4048 | *suspend_point); | |
4049 | } | |
d0ab945e N |
4050 | if (max_progress < *suspend_point) |
4051 | max_progress = *suspend_point; | |
7443ee81 N |
4052 | } |
4053 | } | |
4054 | ||
4055 | /* now set sync_max to allow that progress. sync_max, like | |
4056 | * sync_completed is a count of sectors written per device, so | |
4057 | * we find the difference between max_progress and the start point, | |
4058 | * and divide that by after.data_disks to get a sync_max | |
4059 | * number. | |
4060 | * At the same time we convert wait_point to a similar number | |
4061 | * for comparing against sync_completed. | |
4062 | */ | |
92d1991f | 4063 | /* scale down max_progress to per_disk */ |
7443ee81 | 4064 | max_progress /= reshape->after.data_disks; |
358ef9bf JS |
4065 | /* |
4066 | * Round to chunk size as some kernels give an erroneously | |
4067 | * high number | |
4068 | */ | |
92d1991f N |
4069 | max_progress /= info->new_chunk/512; |
4070 | max_progress *= info->new_chunk/512; | |
7f913e9b N |
4071 | /* And round to old chunk size as the kernel wants that */ |
4072 | max_progress /= info->array.chunk_size/512; | |
4073 | max_progress *= info->array.chunk_size/512; | |
92d1991f N |
4074 | /* Limit progress to the whole device */ |
4075 | if (max_progress > info->component_size) | |
4076 | max_progress = info->component_size; | |
7443ee81 | 4077 | wait_point /= reshape->after.data_disks; |
92d1991f N |
4078 | if (!advancing) { |
4079 | /* switch from 'device offset' to 'processed block count' */ | |
4080 | max_progress = info->component_size - max_progress; | |
4081 | wait_point = info->component_size - wait_point; | |
4082 | } | |
7443ee81 | 4083 | |
a6b2d86c N |
4084 | if (!*frozen) |
4085 | sysfs_set_num(info, NULL, "sync_max", max_progress); | |
7443ee81 N |
4086 | |
4087 | /* Now wait. If we have already reached the point that we were | |
4088 | * asked to wait to, don't wait at all, else wait for any change. | |
4089 | * We need to select on 'sync_completed' as that is the place that | |
4090 | * notifications happen, but we are really interested in | |
4091 | * 'reshape_position' | |
4092 | */ | |
4093 | fd = sysfs_get_fd(info, NULL, "sync_completed"); | |
4094 | if (fd < 0) | |
77a73a17 | 4095 | goto check_progress; |
7443ee81 | 4096 | |
621ea11b | 4097 | if (sysfs_fd_get_ll(fd, &completed) < 0) |
77a73a17 | 4098 | goto check_progress; |
621ea11b | 4099 | |
7443ee81 N |
4100 | while (completed < max_progress && completed < wait_point) { |
4101 | /* Check that sync_action is still 'reshape' to avoid | |
4102 | * waiting forever on a dead array | |
4103 | */ | |
4104 | char action[20]; | |
358ef9bf | 4105 | if (sysfs_get_str(info, NULL, "sync_action", action, 20) <= 0 || |
7443ee81 N |
4106 | strncmp(action, "reshape", 7) != 0) |
4107 | break; | |
26d6e157 AK |
4108 | /* Some kernels reset 'sync_completed' to zero |
4109 | * before setting 'sync_action' to 'idle'. | |
4110 | * So we need these extra tests. | |
4111 | */ | |
fc54fe7a JS |
4112 | if (completed == 0 && advancing && |
4113 | strncmp(action, "idle", 4) == 0 && | |
4114 | info->reshape_progress > 0) | |
26d6e157 | 4115 | break; |
fc54fe7a JS |
4116 | if (completed == 0 && !advancing && |
4117 | strncmp(action, "idle", 4) == 0 && | |
358ef9bf JS |
4118 | info->reshape_progress < |
4119 | (info->component_size * reshape->after.data_disks)) | |
26d6e157 | 4120 | break; |
efc67e8e | 4121 | sysfs_wait(fd, NULL); |
621ea11b | 4122 | if (sysfs_fd_get_ll(fd, &completed) < 0) |
77a73a17 | 4123 | goto check_progress; |
7443ee81 | 4124 | } |
10d0d365 | 4125 | /* Some kernels reset 'sync_completed' to zero, |
e0184a0c | 4126 | * we need to have real point we are in md. |
e0cc1c8d | 4127 | * So in that case, read 'reshape_position' from sysfs. |
10d0d365 | 4128 | */ |
e0184a0c | 4129 | if (completed == 0) { |
e0cc1c8d | 4130 | unsigned long long reshapep; |
e0184a0c | 4131 | char action[20]; |
358ef9bf | 4132 | if (sysfs_get_str(info, NULL, "sync_action", action, 20) > 0 && |
e0cc1c8d N |
4133 | strncmp(action, "idle", 4) == 0 && |
4134 | sysfs_get_ll(info, NULL, | |
4135 | "reshape_position", &reshapep) == 0) | |
4136 | *reshape_completed = reshapep; | |
4137 | } else { | |
4138 | /* some kernels can give an incorrectly high | |
4139 | * 'completed' number, so round down */ | |
4140 | completed /= (info->new_chunk/512); | |
4141 | completed *= (info->new_chunk/512); | |
4142 | /* Convert 'completed' back in to a 'progress' number */ | |
4143 | completed *= reshape->after.data_disks; | |
4144 | if (!advancing) | |
4145 | completed = (info->component_size | |
4146 | * reshape->after.data_disks | |
4147 | - completed); | |
4148 | *reshape_completed = completed; | |
7443ee81 | 4149 | } |
24daa16f | 4150 | |
7443ee81 N |
4151 | close(fd); |
4152 | ||
4153 | /* We return the need_backup flag. Caller will decide | |
d0ab945e | 4154 | * how much - a multiple of ->backup_blocks up to *suspend_point |
7443ee81 | 4155 | */ |
38dad34a N |
4156 | if (advancing) |
4157 | return need_backup > info->reshape_progress; | |
4158 | else | |
4159 | return need_backup >= info->reshape_progress; | |
77a73a17 N |
4160 | |
4161 | check_progress: | |
4162 | /* if we couldn't read a number from sync_completed, then | |
4163 | * either the reshape did complete, or it aborted. | |
4164 | * We can tell which by checking for 'none' in reshape_position. | |
621ea11b N |
4165 | * If it did abort, then it might immediately restart if it |
4166 | * it was just a device failure that leaves us degraded but | |
4167 | * functioning. | |
77a73a17 | 4168 | */ |
d7be7d87 | 4169 | if (sysfs_get_str(info, NULL, "reshape_position", buf, |
358ef9bf | 4170 | sizeof(buf)) < 0 || strncmp(buf, "none", 4) != 0) { |
621ea11b N |
4171 | /* The abort might only be temporary. Wait up to 10 |
4172 | * seconds for fd to contain a valid number again. | |
4173 | */ | |
efc67e8e | 4174 | int wait = 10000; |
621ea11b | 4175 | int rv = -2; |
a6b2d86c | 4176 | unsigned long long new_sync_max; |
efc67e8e N |
4177 | while (fd >= 0 && rv < 0 && wait > 0) { |
4178 | if (sysfs_wait(fd, &wait) != 1) | |
621ea11b | 4179 | break; |
6560987b N |
4180 | switch (sysfs_fd_get_ll(fd, &completed)) { |
4181 | case 0: | |
621ea11b N |
4182 | /* all good again */ |
4183 | rv = 1; | |
a6b2d86c N |
4184 | /* If "sync_max" is no longer max_progress |
4185 | * we need to freeze things | |
4186 | */ | |
358ef9bf JS |
4187 | sysfs_get_ll(info, NULL, "sync_max", |
4188 | &new_sync_max); | |
a6b2d86c | 4189 | *frozen = (new_sync_max != max_progress); |
6560987b N |
4190 | break; |
4191 | case -2: /* read error - abort */ | |
efc67e8e | 4192 | wait = 0; |
6560987b N |
4193 | break; |
4194 | } | |
621ea11b N |
4195 | } |
4196 | if (fd >= 0) | |
4197 | close(fd); | |
4198 | return rv; /* abort */ | |
4199 | } else { | |
6d5316f6 | 4200 | /* Maybe racing with array shutdown - check state */ |
621ea11b N |
4201 | if (fd >= 0) |
4202 | close(fd); | |
d7be7d87 JS |
4203 | if (sysfs_get_str(info, NULL, "array_state", buf, |
4204 | sizeof(buf)) < 0 || | |
4205 | strncmp(buf, "inactive", 8) == 0 || | |
4206 | strncmp(buf, "clear",5) == 0) | |
6d5316f6 | 4207 | return -2; /* abort */ |
77a73a17 | 4208 | return -1; /* complete */ |
6d5316f6 | 4209 | } |
7443ee81 N |
4210 | } |
4211 | ||
fcf57625 | 4212 | /* FIXME return status is never checked */ |
4411fb17 | 4213 | static int grow_backup(struct mdinfo *sra, |
7236ee7a | 4214 | unsigned long long offset, /* per device */ |
7443ee81 | 4215 | unsigned long stripes, /* per device, in old chunks */ |
7236ee7a N |
4216 | int *sources, unsigned long long *offsets, |
4217 | int disks, int chunk, int level, int layout, | |
4218 | int dests, int *destfd, unsigned long long *destoffsets, | |
d4445387 | 4219 | int part, int *degraded, |
7236ee7a N |
4220 | char *buf) |
4221 | { | |
4222 | /* Backup 'blocks' sectors at 'offset' on each device of the array, | |
4223 | * to storage 'destfd' (offset 'destoffsets'), after first | |
4224 | * suspending IO. Then allow resync to continue | |
4225 | * over the suspended section. | |
4226 | * Use part 'part' of the backup-super-block. | |
4227 | */ | |
4228 | int odata = disks; | |
4229 | int rv = 0; | |
4230 | int i; | |
f21e18ca N |
4231 | unsigned long long ll; |
4232 | int new_degraded; | |
7236ee7a N |
4233 | //printf("offset %llu\n", offset); |
4234 | if (level >= 4) | |
4235 | odata--; | |
4236 | if (level == 6) | |
4237 | odata--; | |
7443ee81 | 4238 | |
d4445387 | 4239 | /* Check that array hasn't become degraded, else we might backup the wrong data */ |
2c6ac128 N |
4240 | if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0) |
4241 | return -1; /* FIXME this error is ignored */ | |
f21e18ca | 4242 | new_degraded = (int)ll; |
d4445387 N |
4243 | if (new_degraded != *degraded) { |
4244 | /* check each device to ensure it is still working */ | |
4245 | struct mdinfo *sd; | |
4246 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
4247 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
4248 | continue; | |
4249 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
cf52eff5 TM |
4250 | char sbuf[100]; |
4251 | ||
4252 | if (sysfs_get_str(sra, sd, "state", | |
4253 | sbuf, sizeof(sbuf)) < 0 || | |
d4445387 N |
4254 | strstr(sbuf, "faulty") || |
4255 | strstr(sbuf, "in_sync") == NULL) { | |
4256 | /* this device is dead */ | |
4257 | sd->disk.state = (1<<MD_DISK_FAULTY); | |
4258 | if (sd->disk.raid_disk >= 0 && | |
4259 | sources[sd->disk.raid_disk] >= 0) { | |
4260 | close(sources[sd->disk.raid_disk]); | |
4261 | sources[sd->disk.raid_disk] = -1; | |
4262 | } | |
4263 | } | |
4264 | } | |
4265 | } | |
4266 | *degraded = new_degraded; | |
4267 | } | |
7236ee7a N |
4268 | if (part) { |
4269 | bsb.arraystart2 = __cpu_to_le64(offset * odata); | |
200871ad | 4270 | bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
4271 | } else { |
4272 | bsb.arraystart = __cpu_to_le64(offset * odata); | |
200871ad | 4273 | bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
4274 | } |
4275 | if (part) | |
4276 | bsb.magic[15] = '2'; | |
4277 | for (i = 0; i < dests; i++) | |
4278 | if (part) | |
358ef9bf JS |
4279 | lseek64(destfd[i], destoffsets[i] + |
4280 | __le64_to_cpu(bsb.devstart2)*512, 0); | |
7236ee7a N |
4281 | else |
4282 | lseek64(destfd[i], destoffsets[i], 0); | |
4283 | ||
358ef9bf JS |
4284 | rv = save_stripes(sources, offsets, disks, chunk, level, layout, |
4285 | dests, destfd, offset * 512 * odata, | |
4286 | stripes * chunk * odata, buf); | |
7236ee7a N |
4287 | |
4288 | if (rv) | |
4289 | return rv; | |
97396422 | 4290 | bsb.mtime = __cpu_to_le64(time(0)); |
7236ee7a N |
4291 | for (i = 0; i < dests; i++) { |
4292 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
4293 | ||
358ef9bf JS |
4294 | bsb.sb_csum = bsb_csum((char*)&bsb, |
4295 | ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
7236ee7a N |
4296 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) |
4297 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
4298 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
4299 | ||
72044953 | 4300 | rv = -1; |
d16a7494 JS |
4301 | if ((unsigned long long)lseek64(destfd[i], |
4302 | destoffsets[i] - 4096, 0) != | |
4303 | destoffsets[i] - 4096) | |
72044953 N |
4304 | break; |
4305 | if (write(destfd[i], &bsb, 512) != 512) | |
4306 | break; | |
ff94fb86 | 4307 | if (destoffsets[i] > 4096) { |
f21e18ca | 4308 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) != |
a847575a | 4309 | destoffsets[i]+stripes*chunk*odata) |
72044953 N |
4310 | break; |
4311 | if (write(destfd[i], &bsb, 512) != 512) | |
4312 | break; | |
ff94fb86 | 4313 | } |
7236ee7a | 4314 | fsync(destfd[i]); |
72044953 | 4315 | rv = 0; |
7236ee7a | 4316 | } |
2efedc7b | 4317 | |
fcf57625 | 4318 | return rv; |
7236ee7a N |
4319 | } |
4320 | ||
4321 | /* in 2.6.30, the value reported by sync_completed can be | |
4322 | * less that it should be by one stripe. | |
4323 | * This only happens when reshape hits sync_max and pauses. | |
4324 | * So allow wait_backup to either extent sync_max further | |
4325 | * than strictly necessary, or return before the | |
4326 | * sync has got quite as far as we would really like. | |
4327 | * This is what 'blocks2' is for. | |
4328 | * The various caller give appropriate values so that | |
4329 | * every works. | |
4330 | */ | |
fcf57625 | 4331 | /* FIXME return value is often ignored */ |
24daa16f N |
4332 | static int forget_backup(int dests, int *destfd, |
4333 | unsigned long long *destoffsets, | |
4334 | int part) | |
7236ee7a | 4335 | { |
24daa16f | 4336 | /* |
7443ee81 | 4337 | * Erase backup 'part' (which is 0 or 1) |
7236ee7a | 4338 | */ |
7236ee7a | 4339 | int i; |
fcf57625 | 4340 | int rv; |
7236ee7a | 4341 | |
7236ee7a N |
4342 | if (part) { |
4343 | bsb.arraystart2 = __cpu_to_le64(0); | |
4344 | bsb.length2 = __cpu_to_le64(0); | |
4345 | } else { | |
4346 | bsb.arraystart = __cpu_to_le64(0); | |
4347 | bsb.length = __cpu_to_le64(0); | |
4348 | } | |
97396422 | 4349 | bsb.mtime = __cpu_to_le64(time(0)); |
fcf57625 | 4350 | rv = 0; |
7236ee7a N |
4351 | for (i = 0; i < dests; i++) { |
4352 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
358ef9bf JS |
4353 | bsb.sb_csum = bsb_csum((char*)&bsb, |
4354 | ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
7236ee7a N |
4355 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) |
4356 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
4357 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
f21e18ca | 4358 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) != |
a847575a | 4359 | destoffsets[i]-4096) |
72044953 | 4360 | rv = -1; |
358ef9bf | 4361 | if (rv == 0 && write(destfd[i], &bsb, 512) != 512) |
72044953 | 4362 | rv = -1; |
7236ee7a N |
4363 | fsync(destfd[i]); |
4364 | } | |
fcf57625 | 4365 | return rv; |
7236ee7a | 4366 | } |
e86c9dd6 | 4367 | |
7236ee7a N |
4368 | static void fail(char *msg) |
4369 | { | |
fcf57625 | 4370 | int rv; |
72044953 N |
4371 | rv = (write(2, msg, strlen(msg)) != (int)strlen(msg)); |
4372 | rv |= (write(2, "\n", 1) != 1); | |
fcf57625 | 4373 | exit(rv ? 1 : 2); |
7236ee7a N |
4374 | } |
4375 | ||
4376 | static char *abuf, *bbuf; | |
f21e18ca | 4377 | static unsigned long long abuflen; |
7236ee7a N |
4378 | static void validate(int afd, int bfd, unsigned long long offset) |
4379 | { | |
4380 | /* check that the data in the backup against the array. | |
4381 | * This is only used for regression testing and should not | |
4382 | * be used while the array is active | |
4383 | */ | |
7236ee7a N |
4384 | if (afd < 0) |
4385 | return; | |
4386 | lseek64(bfd, offset - 4096, 0); | |
4387 | if (read(bfd, &bsb2, 512) != 512) | |
4388 | fail("cannot read bsb"); | |
4389 | if (bsb2.sb_csum != bsb_csum((char*)&bsb2, | |
4390 | ((char*)&bsb2.sb_csum)-((char*)&bsb2))) | |
4391 | fail("first csum bad"); | |
4392 | if (memcmp(bsb2.magic, "md_backup_data", 14) != 0) | |
4393 | fail("magic is bad"); | |
4394 | if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 && | |
4395 | bsb2.sb_csum2 != bsb_csum((char*)&bsb2, | |
24daa16f | 4396 | ((char*)&bsb2.sb_csum2)-((char*)&bsb2))) |
7236ee7a N |
4397 | fail("second csum bad"); |
4398 | ||
4399 | if (__le64_to_cpu(bsb2.devstart)*512 != offset) | |
4400 | fail("devstart is wrong"); | |
4401 | ||
4402 | if (bsb2.length) { | |
4403 | unsigned long long len = __le64_to_cpu(bsb2.length)*512; | |
4404 | ||
4405 | if (abuflen < len) { | |
4406 | free(abuf); | |
4407 | free(bbuf); | |
4408 | abuflen = len; | |
fcf57625 N |
4409 | if (posix_memalign((void**)&abuf, 4096, abuflen) || |
4410 | posix_memalign((void**)&bbuf, 4096, abuflen)) { | |
4411 | abuflen = 0; | |
4412 | /* just stop validating on mem-alloc failure */ | |
4413 | return; | |
4414 | } | |
e86c9dd6 | 4415 | } |
48924014 | 4416 | |
7236ee7a | 4417 | lseek64(bfd, offset, 0); |
f21e18ca | 4418 | if ((unsigned long long)read(bfd, bbuf, len) != len) { |
080fd005 | 4419 | //printf("len %llu\n", len); |
7236ee7a N |
4420 | fail("read first backup failed"); |
4421 | } | |
4422 | lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0); | |
f21e18ca | 4423 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
4424 | fail("read first from array failed"); |
4425 | if (memcmp(bbuf, abuf, len) != 0) { | |
24daa16f | 4426 | #if 0 |
7236ee7a N |
4427 | int i; |
4428 | printf("offset=%llu len=%llu\n", | |
080fd005 | 4429 | (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len); |
7236ee7a N |
4430 | for (i=0; i<len; i++) |
4431 | if (bbuf[i] != abuf[i]) { | |
4432 | printf("first diff byte %d\n", i); | |
93ecfa01 | 4433 | break; |
7236ee7a | 4434 | } |
24daa16f | 4435 | #endif |
7236ee7a | 4436 | fail("data1 compare failed"); |
e86c9dd6 | 4437 | } |
7236ee7a N |
4438 | } |
4439 | if (bsb2.length2) { | |
4440 | unsigned long long len = __le64_to_cpu(bsb2.length2)*512; | |
4441 | ||
4442 | if (abuflen < len) { | |
4443 | free(abuf); | |
4444 | free(bbuf); | |
4445 | abuflen = len; | |
503975b9 N |
4446 | abuf = xmalloc(abuflen); |
4447 | bbuf = xmalloc(abuflen); | |
e86c9dd6 NB |
4448 | } |
4449 | ||
7236ee7a | 4450 | lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0); |
f21e18ca | 4451 | if ((unsigned long long)read(bfd, bbuf, len) != len) |
7236ee7a N |
4452 | fail("read second backup failed"); |
4453 | lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0); | |
f21e18ca | 4454 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
4455 | fail("read second from array failed"); |
4456 | if (memcmp(bbuf, abuf, len) != 0) | |
4457 | fail("data2 compare failed"); | |
e86c9dd6 | 4458 | } |
7236ee7a | 4459 | } |
e86c9dd6 | 4460 | |
999b4972 N |
4461 | int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape, |
4462 | struct supertype *st, unsigned long blocks, | |
4463 | int *fds, unsigned long long *offsets, | |
4464 | int dests, int *destfd, unsigned long long *destoffsets) | |
7236ee7a | 4465 | { |
7443ee81 N |
4466 | /* Monitor a reshape where backup is being performed using |
4467 | * 'native' mechanism - either to a backup file, or | |
4468 | * to some space in a spare. | |
4469 | */ | |
7236ee7a | 4470 | char *buf; |
7443ee81 N |
4471 | int degraded = -1; |
4472 | unsigned long long speed; | |
4473 | unsigned long long suspend_point, array_size; | |
4474 | unsigned long long backup_point, wait_point; | |
4475 | unsigned long long reshape_completed; | |
4476 | int done = 0; | |
358ef9bf JS |
4477 | int increasing = reshape->after.data_disks >= |
4478 | reshape->before.data_disks; | |
4479 | int part = 0; /* The next part of the backup area to fill. It | |
4480 | * may already be full, so we need to check */ | |
7443ee81 N |
4481 | int level = reshape->level; |
4482 | int layout = reshape->before.layout; | |
4483 | int data = reshape->before.data_disks; | |
4484 | int disks = reshape->before.data_disks + reshape->parity; | |
4485 | int chunk = sra->array.chunk_size; | |
da8100ca N |
4486 | struct mdinfo *sd; |
4487 | unsigned long stripes; | |
3b7e9d0c | 4488 | int uuid[4]; |
a6b2d86c | 4489 | int frozen = 0; |
da8100ca N |
4490 | |
4491 | /* set up the backup-super-block. This requires the | |
4492 | * uuid from the array. | |
4493 | */ | |
4494 | /* Find a superblock */ | |
4495 | for (sd = sra->devs; sd; sd = sd->next) { | |
4496 | char *dn; | |
4497 | int devfd; | |
4498 | int ok; | |
4499 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
4500 | continue; | |
4501 | dn = map_dev(sd->disk.major, sd->disk.minor, 1); | |
4502 | devfd = dev_open(dn, O_RDONLY); | |
4503 | if (devfd < 0) | |
4504 | continue; | |
4505 | ok = st->ss->load_super(st, devfd, NULL); | |
4506 | close(devfd); | |
77f8d358 | 4507 | if (ok == 0) |
da8100ca N |
4508 | break; |
4509 | } | |
4510 | if (!sd) { | |
e7b84f9d | 4511 | pr_err("Cannot find a superblock\n"); |
da8100ca N |
4512 | return 0; |
4513 | } | |
4514 | ||
4515 | memset(&bsb, 0, 512); | |
4516 | memcpy(bsb.magic, "md_backup_data-1", 16); | |
3b7e9d0c LB |
4517 | st->ss->uuid_from_super(st, uuid); |
4518 | memcpy(bsb.set_uuid, uuid, 16); | |
da8100ca N |
4519 | bsb.mtime = __cpu_to_le64(time(0)); |
4520 | bsb.devstart2 = blocks; | |
4521 | ||
4522 | stripes = blocks / (sra->array.chunk_size/512) / | |
4523 | reshape->before.data_disks; | |
e86c9dd6 | 4524 | |
fcf57625 N |
4525 | if (posix_memalign((void**)&buf, 4096, disks * chunk)) |
4526 | /* Don't start the 'reshape' */ | |
4527 | return 0; | |
7443ee81 N |
4528 | if (reshape->before.data_disks == reshape->after.data_disks) { |
4529 | sysfs_get_ll(sra, NULL, "sync_speed_min", &speed); | |
4530 | sysfs_set_num(sra, NULL, "sync_speed_min", 200000); | |
4531 | } | |
e86c9dd6 | 4532 | |
7443ee81 | 4533 | if (increasing) { |
96533072 | 4534 | array_size = sra->component_size * reshape->after.data_disks; |
7443ee81 N |
4535 | backup_point = sra->reshape_progress; |
4536 | suspend_point = 0; | |
4537 | } else { | |
96533072 | 4538 | array_size = sra->component_size * reshape->before.data_disks; |
25da62d9 | 4539 | backup_point = reshape->backup_blocks; |
7443ee81 N |
4540 | suspend_point = array_size; |
4541 | } | |
7236ee7a | 4542 | |
7443ee81 N |
4543 | while (!done) { |
4544 | int rv; | |
7236ee7a | 4545 | |
7443ee81 N |
4546 | /* Want to return as soon the oldest backup slot can |
4547 | * be released as that allows us to start backing up | |
4548 | * some more, providing suspend_point has been | |
b6b95155 | 4549 | * advanced, which it should have. |
7443ee81 N |
4550 | */ |
4551 | if (increasing) { | |
4552 | wait_point = array_size; | |
4553 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
4554 | wait_point = (__le64_to_cpu(bsb.arraystart) + | |
4555 | __le64_to_cpu(bsb.length)); | |
4556 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
4557 | wait_point = (__le64_to_cpu(bsb.arraystart2) + | |
4558 | __le64_to_cpu(bsb.length2)); | |
4559 | } else { | |
4560 | wait_point = 0; | |
4561 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
4562 | wait_point = __le64_to_cpu(bsb.arraystart); | |
4563 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
4564 | wait_point = __le64_to_cpu(bsb.arraystart2); | |
4565 | } | |
4566 | ||
737f8574 | 4567 | reshape_completed = sra->reshape_progress; |
7443ee81 N |
4568 | rv = progress_reshape(sra, reshape, |
4569 | backup_point, wait_point, | |
a6b2d86c N |
4570 | &suspend_point, &reshape_completed, |
4571 | &frozen); | |
7443ee81 N |
4572 | /* external metadata would need to ping_monitor here */ |
4573 | sra->reshape_progress = reshape_completed; | |
7236ee7a | 4574 | |
7443ee81 N |
4575 | /* Clear any backup region that is before 'here' */ |
4576 | if (increasing) { | |
b3cf095a N |
4577 | if (__le64_to_cpu(bsb.length) > 0 && |
4578 | reshape_completed >= (__le64_to_cpu(bsb.arraystart) + | |
7443ee81 N |
4579 | __le64_to_cpu(bsb.length))) |
4580 | forget_backup(dests, destfd, | |
4581 | destoffsets, 0); | |
b3cf095a N |
4582 | if (__le64_to_cpu(bsb.length2) > 0 && |
4583 | reshape_completed >= (__le64_to_cpu(bsb.arraystart2) + | |
7443ee81 N |
4584 | __le64_to_cpu(bsb.length2))) |
4585 | forget_backup(dests, destfd, | |
4586 | destoffsets, 1); | |
4587 | } else { | |
b3cf095a N |
4588 | if (__le64_to_cpu(bsb.length) > 0 && |
4589 | reshape_completed <= (__le64_to_cpu(bsb.arraystart))) | |
7443ee81 N |
4590 | forget_backup(dests, destfd, |
4591 | destoffsets, 0); | |
b3cf095a N |
4592 | if (__le64_to_cpu(bsb.length2) > 0 && |
4593 | reshape_completed <= (__le64_to_cpu(bsb.arraystart2))) | |
7443ee81 N |
4594 | forget_backup(dests, destfd, |
4595 | destoffsets, 1); | |
4596 | } | |
84d11e6c N |
4597 | if (sigterm) |
4598 | rv = -2; | |
223c5c99 | 4599 | if (rv < 0) { |
77a73a17 N |
4600 | if (rv == -1) |
4601 | done = 1; | |
223c5c99 N |
4602 | break; |
4603 | } | |
2d4de5f9 N |
4604 | if (rv == 0 && increasing && !st->ss->external) { |
4605 | /* No longer need to monitor this reshape */ | |
2cdd5ce0 | 4606 | sysfs_set_str(sra, NULL, "sync_max", "max"); |
2d4de5f9 N |
4607 | done = 1; |
4608 | break; | |
4609 | } | |
223c5c99 | 4610 | |
60204e4e | 4611 | while (rv) { |
7443ee81 | 4612 | unsigned long long offset; |
e97ca002 | 4613 | unsigned long actual_stripes; |
60204e4e N |
4614 | /* Need to backup some data. |
4615 | * If 'part' is not used and the desired | |
4616 | * backup size is suspended, do a backup, | |
4617 | * then consider the next part. | |
4618 | */ | |
7443ee81 N |
4619 | /* Check that 'part' is unused */ |
4620 | if (part == 0 && __le64_to_cpu(bsb.length) != 0) | |
60204e4e | 4621 | break; |
7443ee81 | 4622 | if (part == 1 && __le64_to_cpu(bsb.length2) != 0) |
60204e4e | 4623 | break; |
7443ee81 N |
4624 | |
4625 | offset = backup_point / data; | |
e97ca002 | 4626 | actual_stripes = stripes; |
60204e4e | 4627 | if (increasing) { |
e97ca002 N |
4628 | if (offset + actual_stripes * (chunk/512) > |
4629 | sra->component_size) | |
4630 | actual_stripes = ((sra->component_size - offset) | |
4631 | / (chunk/512)); | |
4632 | if (offset + actual_stripes * (chunk/512) > | |
60204e4e N |
4633 | suspend_point/data) |
4634 | break; | |
4635 | } else { | |
e97ca002 N |
4636 | if (offset < actual_stripes * (chunk/512)) |
4637 | actual_stripes = offset / (chunk/512); | |
4638 | offset -= actual_stripes * (chunk/512); | |
4639 | if (offset < suspend_point/data) | |
60204e4e N |
4640 | break; |
4641 | } | |
e4b11073 N |
4642 | if (actual_stripes == 0) |
4643 | break; | |
358ef9bf JS |
4644 | grow_backup(sra, offset, actual_stripes, fds, offsets, |
4645 | disks, chunk, level, layout, dests, destfd, | |
4646 | destoffsets, part, °raded, buf); | |
7443ee81 N |
4647 | validate(afd, destfd[0], destoffsets[0]); |
4648 | /* record where 'part' is up to */ | |
4649 | part = !part; | |
4650 | if (increasing) | |
e97ca002 | 4651 | backup_point += actual_stripes * (chunk/512) * data; |
7443ee81 | 4652 | else |
e97ca002 | 4653 | backup_point -= actual_stripes * (chunk/512) * data; |
7443ee81 N |
4654 | } |
4655 | } | |
4656 | ||
223c5c99 | 4657 | /* FIXME maybe call progress_reshape one more time instead */ |
5e7be838 N |
4658 | /* remove any remaining suspension */ |
4659 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); | |
4660 | sysfs_set_num(sra, NULL, "suspend_hi", 0); | |
4661 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
4662 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
4663 | ||
7443ee81 N |
4664 | if (reshape->before.data_disks == reshape->after.data_disks) |
4665 | sysfs_set_num(sra, NULL, "sync_speed_min", speed); | |
7236ee7a | 4666 | free(buf); |
77a73a17 | 4667 | return done; |
e86c9dd6 | 4668 | } |
353632d9 NB |
4669 | |
4670 | /* | |
4671 | * If any spare contains md_back_data-1 which is recent wrt mtime, | |
4672 | * write that data into the array and update the super blocks with | |
4673 | * the new reshape_progress | |
4674 | */ | |
358ef9bf JS |
4675 | int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, |
4676 | int cnt, char *backup_file, int verbose) | |
353632d9 NB |
4677 | { |
4678 | int i, j; | |
4679 | int old_disks; | |
353632d9 | 4680 | unsigned long long *offsets; |
82f2d6ab | 4681 | unsigned long long nstripe, ostripe; |
6e9eac4f | 4682 | int ndata, odata; |
353632d9 | 4683 | |
e9e43ec3 | 4684 | odata = info->array.raid_disks - info->delta_disks - 1; |
358ef9bf JS |
4685 | if (info->array.level == 6) |
4686 | odata--; /* number of data disks */ | |
e9e43ec3 | 4687 | ndata = info->array.raid_disks - 1; |
358ef9bf JS |
4688 | if (info->new_level == 6) |
4689 | ndata--; | |
353632d9 NB |
4690 | |
4691 | old_disks = info->array.raid_disks - info->delta_disks; | |
4692 | ||
e9e43ec3 N |
4693 | if (info->delta_disks <= 0) |
4694 | /* Didn't grow, so the backup file must have | |
4695 | * been used | |
4696 | */ | |
4697 | old_disks = cnt; | |
06b0d786 | 4698 | for (i=old_disks-(backup_file?1:0); i<cnt; i++) { |
353632d9 | 4699 | struct mdinfo dinfo; |
06b0d786 | 4700 | int fd; |
e9e43ec3 | 4701 | int bsbsize; |
ea0ebe96 | 4702 | char *devname, namebuf[20]; |
10af14c4 | 4703 | unsigned long long lo, hi; |
353632d9 NB |
4704 | |
4705 | /* This was a spare and may have some saved data on it. | |
4706 | * Load the superblock, find and load the | |
4707 | * backup_super_block. | |
4708 | * If either fail, go on to next device. | |
4709 | * If the backup contains no new info, just return | |
206c5eae | 4710 | * else restore data and update all superblocks |
353632d9 | 4711 | */ |
06b0d786 NB |
4712 | if (i == old_disks-1) { |
4713 | fd = open(backup_file, O_RDONLY); | |
e9e43ec3 | 4714 | if (fd<0) { |
e7b84f9d | 4715 | pr_err("backup file %s inaccessible: %s\n", |
e9e43ec3 | 4716 | backup_file, strerror(errno)); |
06b0d786 | 4717 | continue; |
e9e43ec3 | 4718 | } |
ea0ebe96 | 4719 | devname = backup_file; |
06b0d786 NB |
4720 | } else { |
4721 | fd = fdlist[i]; | |
4722 | if (fd < 0) | |
4723 | continue; | |
3da92f27 | 4724 | if (st->ss->load_super(st, fd, NULL)) |
06b0d786 | 4725 | continue; |
353632d9 | 4726 | |
a5d85af7 | 4727 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 NB |
4728 | st->ss->free_super(st); |
4729 | ||
06b0d786 NB |
4730 | if (lseek64(fd, |
4731 | (dinfo.data_offset + dinfo.component_size - 8) <<9, | |
ea0ebe96 | 4732 | 0) < 0) { |
e7b84f9d | 4733 | pr_err("Cannot seek on device %d\n", i); |
06b0d786 | 4734 | continue; /* Cannot seek */ |
ea0ebe96 N |
4735 | } |
4736 | sprintf(namebuf, "device-%d", i); | |
4737 | devname = namebuf; | |
06b0d786 | 4738 | } |
ea0ebe96 N |
4739 | if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) { |
4740 | if (verbose) | |
e7b84f9d | 4741 | pr_err("Cannot read from %s\n", devname); |
353632d9 | 4742 | continue; /* Cannot read */ |
ea0ebe96 | 4743 | } |
e9e43ec3 | 4744 | if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 && |
ea0ebe96 N |
4745 | memcmp(bsb.magic, "md_backup_data-2", 16) != 0) { |
4746 | if (verbose) | |
e7b84f9d | 4747 | pr_err("No backup metadata on %s\n", devname); |
353632d9 | 4748 | continue; |
ea0ebe96 N |
4749 | } |
4750 | if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) { | |
4751 | if (verbose) | |
358ef9bf JS |
4752 | pr_err("Bad backup-metadata checksum on %s\n", |
4753 | devname); | |
353632d9 | 4754 | continue; /* bad checksum */ |
ea0ebe96 | 4755 | } |
e9e43ec3 | 4756 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 && |
ea0ebe96 N |
4757 | bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) { |
4758 | if (verbose) | |
358ef9bf JS |
4759 | pr_err("Bad backup-metadata checksum2 on %s\n", |
4760 | devname); | |
22e30516 | 4761 | continue; /* Bad second checksum */ |
ea0ebe96 N |
4762 | } |
4763 | if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) { | |
4764 | if (verbose) | |
358ef9bf JS |
4765 | pr_err("Wrong uuid on backup-metadata on %s\n", |
4766 | devname); | |
353632d9 | 4767 | continue; /* Wrong uuid */ |
ea0ebe96 | 4768 | } |
353632d9 | 4769 | |
358ef9bf JS |
4770 | /* |
4771 | * array utime and backup-mtime should be updated at | |
4772 | * much the same time, but it seems that sometimes | |
4773 | * they aren't... So allow considerable flexability in | |
4774 | * matching, and allow this test to be overridden by | |
4775 | * an environment variable. | |
097075b6 | 4776 | */ |
26714713 DD |
4777 | if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) || |
4778 | time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) { | |
097075b6 | 4779 | if (check_env("MDADM_GROW_ALLOW_OLD")) { |
7a862a02 | 4780 | pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n", |
097075b6 N |
4781 | (unsigned long)__le64_to_cpu(bsb.mtime), |
4782 | (unsigned long)info->array.utime); | |
4783 | } else { | |
676ab312 N |
4784 | pr_err("too-old timestamp on backup-metadata on %s\n", devname); |
4785 | pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n"); | |
097075b6 N |
4786 | continue; /* time stamp is too bad */ |
4787 | } | |
ea0ebe96 | 4788 | } |
353632d9 | 4789 | |
e9e43ec3 | 4790 | if (bsb.magic[15] == '1') { |
10af14c4 N |
4791 | if (bsb.length == 0) |
4792 | continue; | |
e7a71c6b N |
4793 | if (info->delta_disks >= 0) { |
4794 | /* reshape_progress is increasing */ | |
4795 | if (__le64_to_cpu(bsb.arraystart) | |
4796 | + __le64_to_cpu(bsb.length) | |
4797 | < info->reshape_progress) { | |
4798 | nonew: | |
4799 | if (verbose) | |
e7b84f9d | 4800 | pr_err("backup-metadata found on %s but is not needed\n", devname); |
e7a71c6b N |
4801 | continue; /* No new data here */ |
4802 | } | |
4803 | } else { | |
4804 | /* reshape_progress is decreasing */ | |
4805 | if (__le64_to_cpu(bsb.arraystart) >= | |
4806 | info->reshape_progress) | |
4807 | goto nonew; /* No new data here */ | |
ea0ebe96 | 4808 | } |
e9e43ec3 | 4809 | } else { |
10af14c4 N |
4810 | if (bsb.length == 0 && bsb.length2 == 0) |
4811 | continue; | |
e7a71c6b N |
4812 | if (info->delta_disks >= 0) { |
4813 | /* reshape_progress is increasing */ | |
4814 | if ((__le64_to_cpu(bsb.arraystart) | |
4815 | + __le64_to_cpu(bsb.length) | |
fc54fe7a | 4816 | < info->reshape_progress) && |
e7a71c6b N |
4817 | (__le64_to_cpu(bsb.arraystart2) |
4818 | + __le64_to_cpu(bsb.length2) | |
4819 | < info->reshape_progress)) | |
4820 | goto nonew; /* No new data here */ | |
4821 | } else { | |
4822 | /* reshape_progress is decreasing */ | |
4823 | if (__le64_to_cpu(bsb.arraystart) >= | |
4824 | info->reshape_progress && | |
4825 | __le64_to_cpu(bsb.arraystart2) >= | |
4826 | info->reshape_progress) | |
4827 | goto nonew; /* No new data here */ | |
4828 | } | |
e9e43ec3 | 4829 | } |
ea0ebe96 N |
4830 | if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) { |
4831 | second_fail: | |
4832 | if (verbose) | |
e7b84f9d N |
4833 | pr_err("Failed to verify secondary backup-metadata block on %s\n", |
4834 | devname); | |
353632d9 | 4835 | continue; /* Cannot seek */ |
ea0ebe96 | 4836 | } |
2efedc7b | 4837 | /* There should be a duplicate backup superblock 4k before here */ |
06b0d786 | 4838 | if (lseek64(fd, -4096, 1) < 0 || |
0155af90 | 4839 | read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2)) |
ea0ebe96 | 4840 | goto second_fail; /* Cannot find leading superblock */ |
e9e43ec3 N |
4841 | if (bsb.magic[15] == '1') |
4842 | bsbsize = offsetof(struct mdp_backup_super, pad1); | |
4843 | else | |
4844 | bsbsize = offsetof(struct mdp_backup_super, pad); | |
ff94fb86 | 4845 | if (memcmp(&bsb2, &bsb, bsbsize) != 0) |
ea0ebe96 | 4846 | goto second_fail; /* Cannot find leading superblock */ |
2efedc7b | 4847 | |
353632d9 | 4848 | /* Now need the data offsets for all devices. */ |
503975b9 | 4849 | offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks); |
353632d9 NB |
4850 | for(j=0; j<info->array.raid_disks; j++) { |
4851 | if (fdlist[j] < 0) | |
4852 | continue; | |
3da92f27 | 4853 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 NB |
4854 | /* FIXME should be this be an error */ |
4855 | continue; | |
a5d85af7 | 4856 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 | 4857 | st->ss->free_super(st); |
14e5b4d7 | 4858 | offsets[j] = dinfo.data_offset * 512; |
353632d9 | 4859 | } |
d56dd607 | 4860 | printf("%s: restoring critical section\n", Name); |
353632d9 | 4861 | |
358ef9bf JS |
4862 | if (restore_stripes(fdlist, offsets, info->array.raid_disks, |
4863 | info->new_chunk, info->new_level, | |
4864 | info->new_layout, fd, | |
4865 | __le64_to_cpu(bsb.devstart)*512, | |
92dcdf7c | 4866 | __le64_to_cpu(bsb.arraystart)*512, |
2fcb75ae | 4867 | __le64_to_cpu(bsb.length)*512, NULL)) { |
e9e43ec3 | 4868 | /* didn't succeed, so giveup */ |
ea0ebe96 | 4869 | if (verbose) |
e7b84f9d | 4870 | pr_err("Error restoring backup from %s\n", |
ea0ebe96 | 4871 | devname); |
730ae51f | 4872 | free(offsets); |
e9e43ec3 N |
4873 | return 1; |
4874 | } | |
24daa16f | 4875 | |
e9e43ec3 | 4876 | if (bsb.magic[15] == '2' && |
358ef9bf JS |
4877 | restore_stripes(fdlist, offsets, info->array.raid_disks, |
4878 | info->new_chunk, info->new_level, | |
4879 | info->new_layout, fd, | |
4880 | __le64_to_cpu(bsb.devstart)*512 + | |
e9e43ec3 | 4881 | __le64_to_cpu(bsb.devstart2)*512, |
92dcdf7c | 4882 | __le64_to_cpu(bsb.arraystart2)*512, |
2fcb75ae | 4883 | __le64_to_cpu(bsb.length2)*512, NULL)) { |
353632d9 | 4884 | /* didn't succeed, so giveup */ |
ea0ebe96 | 4885 | if (verbose) |
e7b84f9d | 4886 | pr_err("Error restoring second backup from %s\n", |
ea0ebe96 | 4887 | devname); |
730ae51f | 4888 | free(offsets); |
2295250a | 4889 | return 1; |
353632d9 NB |
4890 | } |
4891 | ||
730ae51f | 4892 | free(offsets); |
e9e43ec3 | 4893 | |
353632d9 NB |
4894 | /* Ok, so the data is restored. Let's update those superblocks. */ |
4895 | ||
10af14c4 N |
4896 | lo = hi = 0; |
4897 | if (bsb.length) { | |
4898 | lo = __le64_to_cpu(bsb.arraystart); | |
4899 | hi = lo + __le64_to_cpu(bsb.length); | |
4900 | } | |
4901 | if (bsb.magic[15] == '2' && bsb.length2) { | |
4902 | unsigned long long lo1, hi1; | |
4903 | lo1 = __le64_to_cpu(bsb.arraystart2); | |
4904 | hi1 = lo1 + __le64_to_cpu(bsb.length2); | |
4905 | if (lo == hi) { | |
4906 | lo = lo1; | |
4907 | hi = hi1; | |
4908 | } else if (lo < lo1) | |
4909 | hi = hi1; | |
4910 | else | |
4911 | lo = lo1; | |
4912 | } | |
358ef9bf JS |
4913 | if (lo < hi && (info->reshape_progress < lo || |
4914 | info->reshape_progress > hi)) | |
10af14c4 N |
4915 | /* backup does not affect reshape_progress*/ ; |
4916 | else if (info->delta_disks >= 0) { | |
e9e43ec3 N |
4917 | info->reshape_progress = __le64_to_cpu(bsb.arraystart) + |
4918 | __le64_to_cpu(bsb.length); | |
4919 | if (bsb.magic[15] == '2') { | |
358ef9bf JS |
4920 | unsigned long long p2; |
4921 | ||
4922 | p2 = __le64_to_cpu(bsb.arraystart2) + | |
e9e43ec3 N |
4923 | __le64_to_cpu(bsb.length2); |
4924 | if (p2 > info->reshape_progress) | |
4925 | info->reshape_progress = p2; | |
4926 | } | |
4927 | } else { | |
4928 | info->reshape_progress = __le64_to_cpu(bsb.arraystart); | |
4929 | if (bsb.magic[15] == '2') { | |
358ef9bf JS |
4930 | unsigned long long p2; |
4931 | ||
4932 | p2 = __le64_to_cpu(bsb.arraystart2); | |
e9e43ec3 N |
4933 | if (p2 < info->reshape_progress) |
4934 | info->reshape_progress = p2; | |
4935 | } | |
4936 | } | |
353632d9 | 4937 | for (j=0; j<info->array.raid_disks; j++) { |
ca3b6696 N |
4938 | if (fdlist[j] < 0) |
4939 | continue; | |
3da92f27 | 4940 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 | 4941 | continue; |
a5d85af7 | 4942 | st->ss->getinfo_super(st, &dinfo, NULL); |
e9e43ec3 | 4943 | dinfo.reshape_progress = info->reshape_progress; |
358ef9bf | 4944 | st->ss->update_super(st, &dinfo, "_reshape_progress", |
68c7d6d7 | 4945 | NULL,0, 0, NULL); |
3da92f27 NB |
4946 | st->ss->store_super(st, fdlist[j]); |
4947 | st->ss->free_super(st); | |
353632d9 | 4948 | } |
353632d9 NB |
4949 | return 0; |
4950 | } | |
6e9eac4f NB |
4951 | /* Didn't find any backup data, try to see if any |
4952 | * was needed. | |
4953 | */ | |
82f2d6ab N |
4954 | if (info->delta_disks < 0) { |
4955 | /* When shrinking, the critical section is at the end. | |
4956 | * So see if we are before the critical section. | |
4957 | */ | |
4958 | unsigned long long first_block; | |
4959 | nstripe = ostripe = 0; | |
4960 | first_block = 0; | |
4961 | while (ostripe >= nstripe) { | |
4962 | ostripe += info->array.chunk_size / 512; | |
4963 | first_block = ostripe * odata; | |
4964 | nstripe = first_block / ndata / (info->new_chunk/512) * | |
4965 | (info->new_chunk/512); | |
4966 | } | |
4967 | ||
4968 | if (info->reshape_progress >= first_block) | |
4969 | return 0; | |
6e9eac4f | 4970 | } |
82f2d6ab N |
4971 | if (info->delta_disks > 0) { |
4972 | /* See if we are beyond the critical section. */ | |
4973 | unsigned long long last_block; | |
4974 | nstripe = ostripe = 0; | |
4975 | last_block = 0; | |
4976 | while (nstripe >= ostripe) { | |
4977 | nstripe += info->new_chunk / 512; | |
4978 | last_block = nstripe * ndata; | |
4979 | ostripe = last_block / odata / (info->array.chunk_size/512) * | |
4980 | (info->array.chunk_size/512); | |
4981 | } | |
6e9eac4f | 4982 | |
82f2d6ab N |
4983 | if (info->reshape_progress >= last_block) |
4984 | return 0; | |
4985 | } | |
6e9eac4f | 4986 | /* needed to recover critical section! */ |
ea0ebe96 | 4987 | if (verbose) |
e7b84f9d | 4988 | pr_err("Failed to find backup of critical section\n"); |
2295250a | 4989 | return 1; |
353632d9 | 4990 | } |
e9e43ec3 | 4991 | |
2dddadb0 AK |
4992 | int Grow_continue_command(char *devname, int fd, |
4993 | char *backup_file, int verbose) | |
4994 | { | |
4995 | int ret_val = 0; | |
4996 | struct supertype *st = NULL; | |
4997 | struct mdinfo *content = NULL; | |
4998 | struct mdinfo array; | |
4999 | char *subarray = NULL; | |
5000 | struct mdinfo *cc = NULL; | |
5001 | struct mdstat_ent *mdstat = NULL; | |
2dddadb0 | 5002 | int cfd = -1; |
12add445 | 5003 | int fd2; |
2dddadb0 | 5004 | |
358ef9bf | 5005 | dprintf("Grow continue from command line called for %s\n", devname); |
2dddadb0 AK |
5006 | |
5007 | st = super_by_fd(fd, &subarray); | |
5008 | if (!st || !st->ss) { | |
358ef9bf | 5009 | pr_err("Unable to determine metadata format for %s\n", devname); |
2dddadb0 AK |
5010 | return 1; |
5011 | } | |
5012 | dprintf("Grow continue is run for "); | |
5013 | if (st->ss->external == 0) { | |
e0ab37a3 | 5014 | int d; |
8800f853 | 5015 | int cnt = 5; |
1ade5cc1 | 5016 | dprintf_cont("native array (%s)\n", devname); |
9cd39f01 JS |
5017 | if (md_get_array_info(fd, &array.array) < 0) { |
5018 | pr_err("%s is not an active md array - aborting\n", | |
5019 | devname); | |
2dddadb0 AK |
5020 | ret_val = 1; |
5021 | goto Grow_continue_command_exit; | |
5022 | } | |
5023 | content = &array; | |
a250ce24 | 5024 | sysfs_init(content, fd, NULL); |
e0ab37a3 N |
5025 | /* Need to load a superblock. |
5026 | * FIXME we should really get what we need from | |
5027 | * sysfs | |
5028 | */ | |
8800f853 XN |
5029 | do { |
5030 | for (d = 0; d < MAX_DISKS; d++) { | |
5031 | mdu_disk_info_t disk; | |
5032 | char *dv; | |
5033 | int err; | |
5034 | disk.number = d; | |
d97572f5 | 5035 | if (md_get_disk_info(fd, &disk) < 0) |
8800f853 XN |
5036 | continue; |
5037 | if (disk.major == 0 && disk.minor == 0) | |
5038 | continue; | |
5039 | if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0) | |
5040 | continue; | |
5041 | dv = map_dev(disk.major, disk.minor, 1); | |
5042 | if (!dv) | |
5043 | continue; | |
5044 | fd2 = dev_open(dv, O_RDONLY); | |
5045 | if (fd2 < 0) | |
5046 | continue; | |
5047 | err = st->ss->load_super(st, fd2, NULL); | |
5048 | close(fd2); | |
5049 | if (err) | |
5050 | continue; | |
5051 | break; | |
5052 | } | |
5053 | if (d == MAX_DISKS) { | |
5054 | pr_err("Unable to load metadata for %s\n", | |
5055 | devname); | |
5056 | ret_val = 1; | |
5057 | goto Grow_continue_command_exit; | |
5058 | } | |
5059 | st->ss->getinfo_super(st, content, NULL); | |
5060 | if (!content->reshape_active) | |
239b3cc0 | 5061 | sleep_for(3, 0, true); |
8800f853 XN |
5062 | else |
5063 | break; | |
5064 | } while (cnt-- > 0); | |
2dddadb0 | 5065 | } else { |
4dd2df09 | 5066 | char *container; |
2dddadb0 AK |
5067 | |
5068 | if (subarray) { | |
1ade5cc1 | 5069 | dprintf_cont("subarray (%s)\n", subarray); |
4dd2df09 N |
5070 | container = st->container_devnm; |
5071 | cfd = open_dev_excl(st->container_devnm); | |
2dddadb0 | 5072 | } else { |
4dd2df09 | 5073 | container = st->devnm; |
2dddadb0 | 5074 | close(fd); |
4dd2df09 | 5075 | cfd = open_dev_excl(st->devnm); |
1ade5cc1 | 5076 | dprintf_cont("container (%s)\n", container); |
2dddadb0 AK |
5077 | fd = cfd; |
5078 | } | |
5079 | if (cfd < 0) { | |
7a862a02 | 5080 | pr_err("Unable to open container for %s\n", devname); |
2dddadb0 AK |
5081 | ret_val = 1; |
5082 | goto Grow_continue_command_exit; | |
5083 | } | |
2dddadb0 AK |
5084 | |
5085 | /* find in container array under reshape | |
5086 | */ | |
5087 | ret_val = st->ss->load_container(st, cfd, NULL); | |
5088 | if (ret_val) { | |
358ef9bf | 5089 | pr_err("Cannot read superblock for %s\n", devname); |
2dddadb0 AK |
5090 | ret_val = 1; |
5091 | goto Grow_continue_command_exit; | |
5092 | } | |
5093 | ||
446894ea | 5094 | cc = st->ss->container_content(st, subarray); |
2dddadb0 | 5095 | for (content = cc; content ; content = content->next) { |
9e4524df | 5096 | char *array_name; |
446894ea | 5097 | int allow_reshape = 1; |
2dddadb0 AK |
5098 | |
5099 | if (content->reshape_active == 0) | |
5100 | continue; | |
81219e70 LM |
5101 | /* The decision about array or container wide |
5102 | * reshape is taken in Grow_continue based | |
5103 | * content->reshape_active state, therefore we | |
5104 | * need to check_reshape based on | |
5105 | * reshape_active and subarray name | |
446894ea N |
5106 | */ |
5107 | if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) | |
5108 | allow_reshape = 0; | |
5109 | if (content->reshape_active == CONTAINER_RESHAPE && | |
5110 | (content->array.state | |
5111 | & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))) | |
5112 | allow_reshape = 0; | |
5113 | ||
81219e70 | 5114 | if (!allow_reshape) { |
7a862a02 | 5115 | pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n", |
4dd2df09 | 5116 | devname, container); |
81219e70 LM |
5117 | ret_val = 1; |
5118 | goto Grow_continue_command_exit; | |
5119 | } | |
2dddadb0 | 5120 | |
9e4524df JS |
5121 | array_name = strchr(content->text_version+1, '/')+1; |
5122 | mdstat = mdstat_by_subdev(array_name, container); | |
2dddadb0 AK |
5123 | if (!mdstat) |
5124 | continue; | |
1ca90aa6 | 5125 | if (mdstat->active == 0) { |
4dd2df09 N |
5126 | pr_err("Skipping inactive array %s.\n", |
5127 | mdstat->devnm); | |
1ca90aa6 AK |
5128 | free_mdstat(mdstat); |
5129 | mdstat = NULL; | |
5130 | continue; | |
5131 | } | |
2dddadb0 AK |
5132 | break; |
5133 | } | |
5134 | if (!content) { | |
7a862a02 | 5135 | pr_err("Unable to determine reshaped array for %s\n", devname); |
2dddadb0 AK |
5136 | ret_val = 1; |
5137 | goto Grow_continue_command_exit; | |
5138 | } | |
4dd2df09 | 5139 | fd2 = open_dev(mdstat->devnm); |
2dddadb0 | 5140 | if (fd2 < 0) { |
4dd2df09 | 5141 | pr_err("cannot open (%s)\n", mdstat->devnm); |
2dddadb0 AK |
5142 | ret_val = 1; |
5143 | goto Grow_continue_command_exit; | |
5144 | } | |
5145 | ||
dae13137 | 5146 | if (sysfs_init(content, fd2, mdstat->devnm)) { |
8e5b52cd | 5147 | pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n", |
dae13137 JS |
5148 | mdstat->devnm); |
5149 | ret_val = 1; | |
5150 | close(fd2); | |
5151 | goto Grow_continue_command_exit; | |
5152 | } | |
2dddadb0 | 5153 | |
10df72a0 | 5154 | close(fd2); |
10df72a0 | 5155 | |
2dddadb0 AK |
5156 | /* start mdmon in case it is not running |
5157 | */ | |
4dd2df09 N |
5158 | if (!mdmon_running(container)) |
5159 | start_mdmon(container); | |
5160 | ping_monitor(container); | |
2dddadb0 | 5161 | |
4dd2df09 | 5162 | if (mdmon_running(container)) |
2dddadb0 AK |
5163 | st->update_tail = &st->updates; |
5164 | else { | |
7a862a02 | 5165 | pr_err("No mdmon found. Grow cannot continue.\n"); |
2dddadb0 AK |
5166 | ret_val = 1; |
5167 | goto Grow_continue_command_exit; | |
5168 | } | |
5169 | } | |
5170 | ||
f1fe496b AK |
5171 | /* verify that array under reshape is started from |
5172 | * correct position | |
5173 | */ | |
e0ab37a3 | 5174 | if (verify_reshape_position(content, content->array.level) < 0) { |
f1fe496b AK |
5175 | ret_val = 1; |
5176 | goto Grow_continue_command_exit; | |
5177 | } | |
5178 | ||
2dddadb0 AK |
5179 | /* continue reshape |
5180 | */ | |
06e293d0 | 5181 | ret_val = Grow_continue(fd, st, content, backup_file, 1, 0); |
2dddadb0 AK |
5182 | |
5183 | Grow_continue_command_exit: | |
2dddadb0 AK |
5184 | if (cfd > -1) |
5185 | close(cfd); | |
5186 | st->ss->free_super(st); | |
5187 | free_mdstat(mdstat); | |
5188 | sysfs_free(cc); | |
5189 | free(subarray); | |
5190 | ||
5191 | return ret_val; | |
5192 | } | |
5193 | ||
e9e43ec3 | 5194 | int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info, |
06e293d0 | 5195 | char *backup_file, int forked, int freeze_reshape) |
e9e43ec3 | 5196 | { |
3bd58dc6 AK |
5197 | int ret_val = 2; |
5198 | ||
5199 | if (!info->reshape_active) | |
5200 | return ret_val; | |
864a004f | 5201 | |
20a40eca | 5202 | if (st->ss->external) { |
4dd2df09 | 5203 | int cfd = open_dev(st->container_devnm); |
3db2fdd8 | 5204 | |
3bd58dc6 AK |
5205 | if (cfd < 0) |
5206 | return 1; | |
f362d22b | 5207 | |
4dd2df09 | 5208 | st->ss->load_container(st, cfd, st->container_devnm); |
3bd58dc6 | 5209 | close(cfd); |
4dd2df09 | 5210 | ret_val = reshape_container(st->container_devnm, NULL, mdfd, |
358ef9bf JS |
5211 | st, info, 0, backup_file, 0, |
5212 | forked, 1 | info->reshape_active, | |
8ecf12b9 | 5213 | freeze_reshape); |
3bd58dc6 AK |
5214 | } else |
5215 | ret_val = reshape_array(NULL, mdfd, "array", st, info, 1, | |
358ef9bf JS |
5216 | NULL, INVALID_SECTORS, backup_file, |
5217 | 0, forked, 1 | info->reshape_active, | |
3bd58dc6 | 5218 | freeze_reshape); |
f362d22b | 5219 | |
3bd58dc6 | 5220 | return ret_val; |
e9e43ec3 | 5221 | } |
54ded86f N |
5222 | |
5223 | char *make_backup(char *name) | |
5224 | { | |
5225 | char *base = "backup_file-"; | |
5226 | int len; | |
5227 | char *fname; | |
5228 | ||
5229 | len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1; | |
5230 | fname = xmalloc(len); | |
5231 | sprintf(fname, "%s/%s%s", MAP_DIR, base, name); | |
5232 | return fname; | |
5233 | } | |
5234 | ||
5235 | char *locate_backup(char *name) | |
5236 | { | |
5237 | char *fl = make_backup(name); | |
5238 | struct stat stb; | |
5239 | ||
358ef9bf | 5240 | if (stat(fl, &stb) == 0 && S_ISREG(stb.st_mode)) |
54ded86f N |
5241 | return fl; |
5242 | ||
5243 | free(fl); | |
5244 | return NULL; | |
5245 | } |