]>
Commit | Line | Data |
---|---|---|
e5329c37 NB |
1 | /* |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
e736b623 | 4 | * Copyright (C) 2001-2009 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" | |
26 | ||
27 | #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN) | |
28 | #error no endian defined | |
29 | #endif | |
30 | #include "md_u.h" | |
31 | #include "md_p.h" | |
32 | ||
33 | int Grow_Add_device(char *devname, int fd, char *newdev) | |
34 | { | |
35 | /* Add a device to an active array. | |
36 | * Currently, just extend a linear array. | |
37 | * This requires writing a new superblock on the | |
38 | * new device, calling the kernel to add the device, | |
39 | * and if that succeeds, update the superblock on | |
40 | * all other devices. | |
41 | * This means that we need to *find* all other devices. | |
42 | */ | |
4b1ac34b NB |
43 | struct mdinfo info; |
44 | ||
e5329c37 NB |
45 | struct stat stb; |
46 | int nfd, fd2; | |
47 | int d, nd; | |
82d9eba6 | 48 | struct supertype *st = NULL; |
aba69144 | 49 | |
e5329c37 | 50 | |
4b1ac34b | 51 | if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) { |
e5329c37 NB |
52 | fprintf(stderr, Name ": cannot get array info for %s\n", devname); |
53 | return 1; | |
54 | } | |
55 | ||
1686dc25 | 56 | st = super_by_fd(fd); |
82d9eba6 | 57 | if (!st) { |
f9ce90ba NB |
58 | fprintf(stderr, Name ": cannot handle arrays with superblock version %d\n", info.array.major_version); |
59 | return 1; | |
60 | } | |
61 | ||
4b1ac34b | 62 | if (info.array.level != -1) { |
e5329c37 NB |
63 | fprintf(stderr, Name ": can only add devices to linear arrays\n"); |
64 | return 1; | |
65 | } | |
66 | ||
6416d527 | 67 | nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT); |
e5329c37 NB |
68 | if (nfd < 0) { |
69 | fprintf(stderr, Name ": cannot open %s\n", newdev); | |
70 | return 1; | |
71 | } | |
72 | fstat(nfd, &stb); | |
73 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
74 | fprintf(stderr, Name ": %s is not a block device!\n", newdev); | |
75 | close(nfd); | |
76 | return 1; | |
77 | } | |
78 | /* now check out all the devices and make sure we can read the superblock */ | |
4b1ac34b | 79 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
80 | mdu_disk_info_t disk; |
81 | char *dv; | |
82 | ||
83 | disk.number = d; | |
84 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) { | |
85 | fprintf(stderr, Name ": cannot get device detail for device %d\n", | |
86 | d); | |
87 | return 1; | |
88 | } | |
16c6fa80 | 89 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 NB |
90 | if (!dv) { |
91 | fprintf(stderr, Name ": cannot find device file for device %d\n", | |
92 | d); | |
93 | return 1; | |
94 | } | |
16c6fa80 | 95 | fd2 = dev_open(dv, O_RDWR); |
e5329c37 NB |
96 | if (!fd2) { |
97 | fprintf(stderr, Name ": cannot open device file %s\n", dv); | |
98 | return 1; | |
99 | } | |
3da92f27 NB |
100 | st->ss->free_super(st); |
101 | ||
102 | if (st->ss->load_super(st, fd2, NULL)) { | |
e5329c37 NB |
103 | fprintf(stderr, Name ": cannot find super block on %s\n", dv); |
104 | close(fd2); | |
105 | return 1; | |
106 | } | |
107 | close(fd2); | |
108 | } | |
109 | /* Ok, looks good. Lets update the superblock and write it out to | |
110 | * newdev. | |
111 | */ | |
aba69144 | 112 | |
4b1ac34b NB |
113 | info.disk.number = d; |
114 | info.disk.major = major(stb.st_rdev); | |
115 | info.disk.minor = minor(stb.st_rdev); | |
116 | info.disk.raid_disk = d; | |
117 | info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
3da92f27 | 118 | st->ss->update_super(st, &info, "linear-grow-new", newdev, |
f752781f | 119 | 0, 0, NULL); |
e5329c37 | 120 | |
3da92f27 | 121 | if (st->ss->store_super(st, nfd)) { |
f752781f NB |
122 | fprintf(stderr, Name ": Cannot store new superblock on %s\n", |
123 | newdev); | |
e5329c37 NB |
124 | close(nfd); |
125 | return 1; | |
126 | } | |
e5329c37 | 127 | close(nfd); |
4b1ac34b NB |
128 | |
129 | if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) { | |
e5329c37 NB |
130 | fprintf(stderr, Name ": Cannot add new disk to this array\n"); |
131 | return 1; | |
132 | } | |
133 | /* Well, that seems to have worked. | |
134 | * Now go through and update all superblocks | |
135 | */ | |
136 | ||
4b1ac34b | 137 | if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) { |
e5329c37 NB |
138 | fprintf(stderr, Name ": cannot get array info for %s\n", devname); |
139 | return 1; | |
140 | } | |
141 | ||
142 | nd = d; | |
4b1ac34b | 143 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
144 | mdu_disk_info_t disk; |
145 | char *dv; | |
146 | ||
147 | disk.number = d; | |
148 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) { | |
149 | fprintf(stderr, Name ": cannot get device detail for device %d\n", | |
150 | d); | |
151 | return 1; | |
152 | } | |
16c6fa80 | 153 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 NB |
154 | if (!dv) { |
155 | fprintf(stderr, Name ": cannot find device file for device %d\n", | |
156 | d); | |
157 | return 1; | |
158 | } | |
16c6fa80 | 159 | fd2 = dev_open(dv, O_RDWR); |
e5329c37 NB |
160 | if (fd2 < 0) { |
161 | fprintf(stderr, Name ": cannot open device file %s\n", dv); | |
162 | return 1; | |
163 | } | |
3da92f27 | 164 | if (st->ss->load_super(st, fd2, NULL)) { |
e5329c37 NB |
165 | fprintf(stderr, Name ": cannot find super block on %s\n", dv); |
166 | close(fd); | |
167 | return 1; | |
168 | } | |
4b1ac34b NB |
169 | info.array.raid_disks = nd+1; |
170 | info.array.nr_disks = nd+1; | |
171 | info.array.active_disks = nd+1; | |
172 | info.array.working_disks = nd+1; | |
f752781f | 173 | |
3da92f27 | 174 | st->ss->update_super(st, &info, "linear-grow-update", dv, |
f752781f | 175 | 0, 0, NULL); |
aba69144 | 176 | |
3da92f27 | 177 | if (st->ss->store_super(st, fd2)) { |
e5329c37 NB |
178 | fprintf(stderr, Name ": Cannot store new superblock on %s\n", dv); |
179 | close(fd2); | |
180 | return 1; | |
181 | } | |
182 | close(fd2); | |
183 | } | |
184 | ||
185 | return 0; | |
186 | } | |
f5e166fe | 187 | |
8fac0577 | 188 | int Grow_addbitmap(char *devname, int fd, char *file, int chunk, int delay, int write_behind, int force) |
f5e166fe NB |
189 | { |
190 | /* | |
191 | * First check that array doesn't have a bitmap | |
192 | * Then create the bitmap | |
193 | * Then add it | |
194 | * | |
195 | * For internal bitmaps, we need to check the version, | |
196 | * find all the active devices, and write the bitmap block | |
197 | * to all devices | |
198 | */ | |
199 | mdu_bitmap_file_t bmf; | |
200 | mdu_array_info_t array; | |
201 | struct supertype *st; | |
dcec9ee5 NB |
202 | int major = BITMAP_MAJOR_HI; |
203 | int vers = md_get_version(fd); | |
8fac0577 | 204 | unsigned long long bitmapsize, array_size; |
dcec9ee5 NB |
205 | |
206 | if (vers < 9003) { | |
207 | major = BITMAP_MAJOR_HOSTENDIAN; | |
208 | #ifdef __BIG_ENDIAN | |
209 | fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n" | |
210 | " between different architectured. Consider upgrading the Linux kernel.\n"); | |
211 | #endif | |
212 | } | |
f5e166fe NB |
213 | |
214 | if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) { | |
353632d9 | 215 | if (errno == ENOMEM) |
f5e166fe NB |
216 | fprintf(stderr, Name ": Memory allocation failure.\n"); |
217 | else | |
218 | fprintf(stderr, Name ": bitmaps not supported by this kernel.\n"); | |
219 | return 1; | |
220 | } | |
221 | if (bmf.pathname[0]) { | |
fe80f49b NB |
222 | if (strcmp(file,"none")==0) { |
223 | if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) { | |
224 | fprintf(stderr, Name ": failed to remove bitmap %s\n", | |
225 | bmf.pathname); | |
226 | return 1; | |
227 | } | |
228 | return 0; | |
229 | } | |
f5e166fe NB |
230 | fprintf(stderr, Name ": %s already has a bitmap (%s)\n", |
231 | devname, bmf.pathname); | |
232 | return 1; | |
233 | } | |
234 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
235 | fprintf(stderr, Name ": cannot get array status for %s\n", devname); | |
236 | return 1; | |
237 | } | |
238 | if (array.state & (1<<MD_SB_BITMAP_PRESENT)) { | |
fe80f49b NB |
239 | if (strcmp(file, "none")==0) { |
240 | array.state &= ~(1<<MD_SB_BITMAP_PRESENT); | |
241 | if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) { | |
242 | fprintf(stderr, Name ": failed to remove internal bitmap.\n"); | |
243 | return 1; | |
244 | } | |
245 | return 0; | |
246 | } | |
f5e166fe NB |
247 | fprintf(stderr, Name ": Internal bitmap already present on %s\n", |
248 | devname); | |
249 | return 1; | |
250 | } | |
5b28bd56 NB |
251 | if (array.level <= 0) { |
252 | fprintf(stderr, Name ": Bitmaps not meaningful with level %s\n", | |
253 | map_num(pers, array.level)?:"of this array"); | |
254 | return 1; | |
255 | } | |
8fac0577 NB |
256 | bitmapsize = array.size; |
257 | bitmapsize <<= 1; | |
beae1dfe | 258 | if (get_dev_size(fd, NULL, &array_size) && |
8fac0577 NB |
259 | array_size > (0x7fffffffULL<<9)) { |
260 | /* Array is big enough that we cannot trust array.size | |
261 | * try other approaches | |
262 | */ | |
263 | bitmapsize = get_component_size(fd); | |
264 | } | |
8fac0577 NB |
265 | if (bitmapsize == 0) { |
266 | fprintf(stderr, Name ": Cannot reliably determine size of array to create bitmap - sorry.\n"); | |
267 | return 1; | |
268 | } | |
269 | ||
f9c25f1d | 270 | if (array.level == 10) { |
8686f3ed | 271 | int ncopies = (array.layout&255)*((array.layout>>8)&255); |
f9c25f1d NB |
272 | bitmapsize = bitmapsize * array.raid_disks / ncopies; |
273 | } | |
274 | ||
1686dc25 | 275 | st = super_by_fd(fd); |
f5e166fe NB |
276 | if (!st) { |
277 | fprintf(stderr, Name ": Cannot understand version %d.%d\n", | |
278 | array.major_version, array.minor_version); | |
279 | return 1; | |
280 | } | |
fe80f49b NB |
281 | if (strcmp(file, "none") == 0) { |
282 | fprintf(stderr, Name ": no bitmap found on %s\n", devname); | |
283 | return 1; | |
284 | } else if (strcmp(file, "internal") == 0) { | |
f5e166fe | 285 | int d; |
ea329559 | 286 | for (d=0; d< st->max_devs; d++) { |
f5e166fe NB |
287 | mdu_disk_info_t disk; |
288 | char *dv; | |
289 | disk.number = d; | |
290 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
291 | continue; | |
292 | if (disk.major == 0 && | |
293 | disk.minor == 0) | |
294 | continue; | |
295 | if ((disk.state & (1<<MD_DISK_SYNC))==0) | |
296 | continue; | |
16c6fa80 | 297 | dv = map_dev(disk.major, disk.minor, 1); |
f5e166fe | 298 | if (dv) { |
16c6fa80 | 299 | int fd2 = dev_open(dv, O_RDWR); |
f5e166fe NB |
300 | if (fd2 < 0) |
301 | continue; | |
3da92f27 | 302 | if (st->ss->load_super(st, fd2, NULL)==0) { |
199171a2 | 303 | if (st->ss->add_internal_bitmap( |
3da92f27 | 304 | st, |
199171a2 NB |
305 | &chunk, delay, write_behind, |
306 | bitmapsize, 0, major) | |
307 | ) | |
3da92f27 | 308 | st->ss->write_bitmap(st, fd2); |
21e92547 NB |
309 | else { |
310 | fprintf(stderr, Name ": failed to create internal bitmap - chunksize problem.\n"); | |
311 | close(fd2); | |
312 | return 1; | |
313 | } | |
f5e166fe NB |
314 | } |
315 | close(fd2); | |
316 | } | |
317 | } | |
318 | array.state |= (1<<MD_SB_BITMAP_PRESENT); | |
319 | if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) { | |
320 | fprintf(stderr, Name ": failed to set internal bitmap.\n"); | |
321 | return 1; | |
322 | } | |
fe80f49b NB |
323 | } else { |
324 | int uuid[4]; | |
325 | int bitmap_fd; | |
326 | int d; | |
327 | int max_devs = st->max_devs; | |
fe80f49b NB |
328 | |
329 | /* try to load a superblock */ | |
330 | for (d=0; d<max_devs; d++) { | |
331 | mdu_disk_info_t disk; | |
332 | char *dv; | |
333 | int fd2; | |
334 | disk.number = d; | |
335 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
336 | continue; | |
337 | if ((disk.major==0 && disk.minor==0) || | |
338 | (disk.state & (1<<MD_DISK_REMOVED))) | |
339 | continue; | |
16c6fa80 | 340 | dv = map_dev(disk.major, disk.minor, 1); |
fe80f49b | 341 | if (!dv) continue; |
16c6fa80 | 342 | fd2 = dev_open(dv, O_RDONLY); |
fe80f49b | 343 | if (fd2 >= 0 && |
3da92f27 | 344 | st->ss->load_super(st, fd2, NULL) == 0) { |
fe80f49b | 345 | close(fd2); |
3da92f27 | 346 | st->ss->uuid_from_super(st, uuid); |
fe80f49b NB |
347 | break; |
348 | } | |
349 | close(fd2); | |
350 | } | |
351 | if (d == max_devs) { | |
352 | fprintf(stderr, Name ": cannot find UUID for array!\n"); | |
353 | return 1; | |
354 | } | |
8fac0577 | 355 | if (CreateBitmap(file, force, (char*)uuid, chunk, |
f9c25f1d | 356 | delay, write_behind, bitmapsize, major)) { |
fe80f49b NB |
357 | return 1; |
358 | } | |
359 | bitmap_fd = open(file, O_RDWR); | |
360 | if (bitmap_fd < 0) { | |
8fac0577 | 361 | fprintf(stderr, Name ": weird: %s cannot be opened\n", |
fe80f49b NB |
362 | file); |
363 | return 1; | |
364 | } | |
365 | if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
366 | fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n", | |
367 | devname, strerror(errno)); | |
368 | return 1; | |
369 | } | |
370 | } | |
f5e166fe NB |
371 | |
372 | return 0; | |
373 | } | |
374 | ||
e86c9dd6 NB |
375 | |
376 | /* | |
377 | * When reshaping an array we might need to backup some data. | |
378 | * This is written to all spares with a 'super_block' describing it. | |
379 | * The superblock goes 1K form the end of the used space on the | |
380 | * device. | |
381 | * It if written after the backup is complete. | |
382 | * It has the following structure. | |
383 | */ | |
384 | ||
385 | struct mdp_backup_super { | |
386 | char magic[16]; /* md_backup_data-1 */ | |
387 | __u8 set_uuid[16]; | |
388 | __u64 mtime; | |
389 | /* start/sizes in 512byte sectors */ | |
390 | __u64 devstart; | |
391 | __u64 arraystart; | |
392 | __u64 length; | |
393 | __u32 sb_csum; /* csum of preceeding bytes. */ | |
94a20f0c N |
394 | __u8 pad[512-68]; |
395 | } __attribute__((aligned(512))) bsb; | |
e86c9dd6 NB |
396 | |
397 | int bsb_csum(char *buf, int len) | |
398 | { | |
399 | int i; | |
400 | int csum = 0; | |
401 | for (i=0; i<len; i++) | |
402 | csum = (csum<<3) + buf[0]; | |
403 | return __cpu_to_le32(csum); | |
404 | } | |
405 | ||
06b0d786 | 406 | int Grow_reshape(char *devname, int fd, int quiet, char *backup_file, |
e86c9dd6 NB |
407 | long long size, |
408 | int level, int layout, int chunksize, int raid_disks) | |
409 | { | |
410 | /* Make some changes in the shape of an array. | |
411 | * The kernel must support the change. | |
412 | * Different reshapes have subtly different meaning for different | |
413 | * levels, so we need to check the current state of the array | |
414 | * and go from there. | |
415 | */ | |
416 | struct mdu_array_info_s array; | |
417 | char *c; | |
418 | ||
e86c9dd6 NB |
419 | struct supertype *st; |
420 | ||
421 | int nlevel, olevel; | |
422 | int nchunk, ochunk; | |
423 | int nlayout, olayout; | |
424 | int ndisks, odisks; | |
425 | int ndata, odata; | |
426 | unsigned long long nstripe, ostripe, last_block; | |
427 | int *fdlist; | |
428 | unsigned long long *offsets; | |
429 | int d, i, spares; | |
430 | int nrdisks; | |
431 | int err; | |
e86c9dd6 | 432 | |
7e0f6979 | 433 | struct mdinfo *sra; |
06c7f68e | 434 | struct mdinfo *sd; |
e86c9dd6 NB |
435 | |
436 | if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) { | |
437 | fprintf(stderr, Name ": %s is not an active md array - aborting\n", | |
438 | devname); | |
439 | return 1; | |
440 | } | |
441 | c = map_num(pers, array.level); | |
442 | if (c == NULL) c = "-unknown-"; | |
443 | switch(array.level) { | |
444 | default: /* raid0, linear, multipath cannot be reconfigured */ | |
445 | fprintf(stderr, Name ": %s array %s cannot be reshaped.\n", | |
446 | c, devname); | |
447 | return 1; | |
448 | ||
449 | case LEVEL_FAULTY: /* only 'layout' change is permitted */ | |
450 | ||
451 | if (size >= 0) { | |
452 | fprintf(stderr, Name ": %s: Cannot change size of a 'faulty' array\n", | |
453 | devname); | |
454 | return 1; | |
455 | } | |
456 | if (level != UnSet && level != LEVEL_FAULTY) { | |
457 | fprintf(stderr, Name ": %s: Cannot change RAID level of a 'faulty' array\n", | |
458 | devname); | |
459 | return 1; | |
460 | } | |
461 | if (chunksize || raid_disks) { | |
462 | fprintf(stderr, Name ": %s: Cannot change chunksize or disks of a 'faulty' array\n", | |
463 | devname); | |
464 | return 1; | |
465 | } | |
466 | if (layout == UnSet) | |
467 | return 0; /* nothing to do.... */ | |
468 | ||
469 | array.layout = layout; | |
470 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
471 | fprintf(stderr, Name ": Cannot set layout for %s: %s\n", | |
472 | devname, strerror(errno)); | |
473 | return 1; | |
474 | } | |
475 | if (!quiet) | |
476 | printf("layout for %s set to %d\n", devname, array.layout); | |
477 | return 0; | |
478 | ||
479 | case 1: /* raid_disks and size can each be changed. They are independant */ | |
480 | ||
481 | if (level != UnSet && level != 1) { | |
482 | fprintf(stderr, Name ": %s: Cannot change RAID level of a RAID1 array.\n", | |
483 | devname); | |
484 | return 1; | |
485 | } | |
486 | if (chunksize || layout != UnSet) { | |
487 | fprintf(stderr, Name ": %s: Cannot change chunk size of layout for a RAID1 array.\n", | |
488 | devname); | |
489 | return 1; | |
490 | } | |
491 | ||
492 | /* Each can trigger a resync/recovery which will block the | |
493 | * other from happening. Later we could block | |
494 | * resync for the duration via 'sync_action'... | |
495 | */ | |
9860f271 | 496 | if (raid_disks > 0) { |
e86c9dd6 | 497 | array.raid_disks = raid_disks; |
9860f271 NB |
498 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { |
499 | fprintf(stderr, Name ": Cannot set raid-devices for %s: %s\n", | |
500 | devname, strerror(errno)); | |
501 | return 1; | |
502 | } | |
503 | } | |
504 | if (size >= 0) { | |
e86c9dd6 | 505 | array.size = size; |
9860f271 NB |
506 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { |
507 | fprintf(stderr, Name ": Cannot set device size for %s: %s\n", | |
508 | devname, strerror(errno)); | |
509 | return 1; | |
510 | } | |
e86c9dd6 NB |
511 | } |
512 | return 0; | |
513 | ||
514 | case 4: | |
515 | case 5: | |
516 | case 6: | |
1686dc25 NB |
517 | st = super_by_fd(fd); |
518 | ||
758d3a8e | 519 | /* size can be changed independently. |
e86c9dd6 NB |
520 | * layout/chunksize/raid_disks/level can be changed |
521 | * though the kernel may not support it all. | |
522 | * If 'suspend_lo' is not present in devfs, then | |
523 | * these cannot be changed. | |
524 | */ | |
525 | if (size >= 0) { | |
526 | /* Cannot change other details as well.. */ | |
527 | if (layout != UnSet || | |
528 | chunksize != 0 || | |
529 | raid_disks != 0 || | |
530 | level != UnSet) { | |
531 | fprintf(stderr, Name ": %s: Cannot change shape as well as size of a %s array.\n", | |
532 | devname, c); | |
533 | return 1; | |
534 | } | |
535 | array.size = size; | |
536 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
537 | fprintf(stderr, Name ": Cannot set device size/shape for %s: %s\n", | |
538 | devname, strerror(errno)); | |
539 | return 1; | |
540 | } | |
541 | return 0; | |
542 | } | |
543 | /* Ok, just change the shape. This can be awkward. | |
544 | * There are three possibilities. | |
545 | * 1/ The array will shrink. We don't support this | |
546 | * possibility. Maybe one day... | |
547 | * 2/ The array will not change size. This is easy enough | |
548 | * to do, but not reliably. If the process is aborted | |
549 | * the array *will* be corrupted. So maybe we can allow | |
550 | * this but only if the user is really certain. e.g. | |
551 | * --really-risk-everything | |
552 | * 3/ The array will grow. This can be reliably achieved. | |
553 | * However the kernel's restripe routines will cheerfully | |
554 | * overwrite some early data before it is safe. So we | |
555 | * need to make a backup of the early parts of the array | |
556 | * and be ready to restore it if rebuild aborts very early. | |
557 | * | |
558 | * We backup data by writing it to all spares (there must be | |
559 | * at least 1, so even raid6->raid5 requires a spare to be | |
560 | * present). | |
561 | * | |
562 | * So: we enumerate the devices in the array and | |
563 | * make sure we can open all of them. | |
564 | * Then we freeze the early part of the array and | |
565 | * backup to the various spares. | |
566 | * Then we request changes and start the reshape. | |
567 | * Monitor progress until it has passed the danger zone. | |
568 | * and finally invalidate the copied data and unfreeze the | |
569 | * start of the array. | |
570 | * | |
571 | * Before we can do this we need to decide: | |
572 | * - will the array grow? Just calculate size | |
573 | * - how much needs to be saved: count stripes. | |
574 | * - where to save data... good question. | |
575 | * | |
576 | */ | |
577 | nlevel = olevel = array.level; | |
578 | nchunk = ochunk = array.chunk_size; | |
579 | nlayout = olayout = array.layout; | |
580 | ndisks = odisks = array.raid_disks; | |
581 | ||
582 | if (level != UnSet) nlevel = level; | |
583 | if (chunksize) nchunk = chunksize; | |
584 | if (layout != UnSet) nlayout = layout; | |
585 | if (raid_disks) ndisks = raid_disks; | |
586 | ||
587 | odata = odisks-1; | |
588 | if (olevel == 6) odata--; /* number of data disks */ | |
589 | ndata = ndisks-1; | |
590 | if (nlevel == 6) ndata--; | |
591 | ||
592 | if (ndata < odata) { | |
593 | fprintf(stderr, Name ": %s: Cannot reduce number of data disks (yet).\n", | |
594 | devname); | |
595 | return 1; | |
596 | } | |
597 | if (ndata == odata) { | |
598 | fprintf(stderr, Name ": %s: Cannot reshape array without increasing size (yet).\n", | |
599 | devname); | |
600 | return 1; | |
601 | } | |
602 | /* Well, it is growing... so how much do we need to backup. | |
603 | * Need to backup a full number of new-stripes, such that the | |
604 | * last one does not over-write any place that it would be read | |
605 | * from | |
606 | */ | |
607 | nstripe = ostripe = 0; | |
353632d9 | 608 | while (nstripe >= ostripe) { |
e86c9dd6 NB |
609 | nstripe += nchunk/512; |
610 | last_block = nstripe * ndata; | |
353632d9 | 611 | ostripe = last_block / odata / (ochunk/512) * (ochunk/512); |
e86c9dd6 | 612 | } |
0aa389dc NB |
613 | fprintf(stderr, Name ": Need to backup %lluK of critical " |
614 | "section..\n", last_block/2); | |
e86c9dd6 NB |
615 | |
616 | sra = sysfs_read(fd, 0, | |
758d3a8e NB |
617 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE| |
618 | GET_CACHE); | |
e86c9dd6 NB |
619 | if (!sra) { |
620 | fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n", | |
621 | devname); | |
622 | return 1; | |
623 | } | |
624 | ||
625 | if (last_block >= sra->component_size/2) { | |
626 | fprintf(stderr, Name ": %s: Something wrong - reshape aborted\n", | |
627 | devname); | |
628 | return 1; | |
629 | } | |
7e0f6979 | 630 | if (sra->array.spare_disks == 0 && backup_file == NULL) { |
06b0d786 | 631 | fprintf(stderr, Name ": %s: Cannot grow - need a spare or backup-file to backup critical section\n", |
353632d9 NB |
632 | devname); |
633 | return 1; | |
634 | } | |
e86c9dd6 | 635 | |
7e0f6979 | 636 | nrdisks = array.nr_disks + sra->array.spare_disks; |
e86c9dd6 NB |
637 | /* Now we need to open all these devices so we can read/write. |
638 | */ | |
06b0d786 NB |
639 | fdlist = malloc((1+nrdisks) * sizeof(int)); |
640 | offsets = malloc((1+nrdisks) * sizeof(offsets[0])); | |
e86c9dd6 NB |
641 | if (!fdlist || !offsets) { |
642 | fprintf(stderr, Name ": malloc failed: grow aborted\n"); | |
643 | return 1; | |
644 | } | |
06b0d786 | 645 | for (d=0; d <= nrdisks; d++) |
e86c9dd6 NB |
646 | fdlist[d] = -1; |
647 | d = array.raid_disks; | |
648 | for (sd = sra->devs; sd; sd=sd->next) { | |
06c7f68e | 649 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) |
e86c9dd6 | 650 | continue; |
06c7f68e NB |
651 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { |
652 | char *dn = map_dev(sd->disk.major, | |
653 | sd->disk.minor, 1); | |
654 | fdlist[sd->disk.raid_disk] | |
655 | = dev_open(dn, O_RDONLY); | |
656 | offsets[sd->disk.raid_disk] = sd->data_offset; | |
657 | if (fdlist[sd->disk.raid_disk] < 0) { | |
e86c9dd6 | 658 | fprintf(stderr, Name ": %s: cannot open component %s\n", |
e81cdd9f | 659 | devname, dn?dn:"-unknown-"); |
e86c9dd6 NB |
660 | goto abort; |
661 | } | |
662 | } else { | |
663 | /* spare */ | |
06c7f68e NB |
664 | char *dn = map_dev(sd->disk.major, |
665 | sd->disk.minor, 1); | |
16c6fa80 | 666 | fdlist[d] = dev_open(dn, O_RDWR); |
06c7f68e | 667 | offsets[d] = sd->data_offset; |
e86c9dd6 NB |
668 | if (fdlist[d]<0) { |
669 | fprintf(stderr, Name ": %s: cannot open component %s\n", | |
e81cdd9f | 670 | devname, dn?dn:"-unknown"); |
e86c9dd6 NB |
671 | goto abort; |
672 | } | |
673 | d++; | |
674 | } | |
675 | } | |
676 | for (i=0 ; i<array.raid_disks; i++) | |
677 | if (fdlist[i] < 0) { | |
678 | fprintf(stderr, Name ": %s: failed to find device %d. Array might be degraded.\n" | |
679 | " --grow aborted\n", devname, i); | |
680 | goto abort; | |
681 | } | |
7e0f6979 | 682 | spares = sra->array.spare_disks; |
06b0d786 | 683 | if (backup_file) { |
fb97b4d6 | 684 | fdlist[d] = open(backup_file, O_RDWR|O_CREAT|O_EXCL, S_IRUSR | S_IWUSR); |
06b0d786 NB |
685 | if (fdlist[d] < 0) { |
686 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
687 | devname, backup_file, strerror(errno)); | |
688 | goto abort; | |
689 | } | |
690 | offsets[d] = 8; | |
691 | d++; | |
692 | spares++; | |
693 | } | |
e86c9dd6 | 694 | if (fdlist[array.raid_disks] < 0) { |
06b0d786 | 695 | fprintf(stderr, Name ": %s: failed to find a spare and no backup-file given - --grow aborted\n", |
e86c9dd6 NB |
696 | devname); |
697 | goto abort; | |
698 | } | |
699 | ||
700 | /* Find a superblock */ | |
3da92f27 | 701 | if (st->ss->load_super(st, fdlist[0], NULL)) { |
e86c9dd6 NB |
702 | fprintf(stderr, Name ": %s: Cannot find a superblock\n", |
703 | devname); | |
704 | goto abort; | |
705 | } | |
706 | ||
2efedc7b NB |
707 | |
708 | memcpy(bsb.magic, "md_backup_data-1", 16); | |
3da92f27 | 709 | st->ss->uuid_from_super(st, (int*)&bsb.set_uuid); |
2efedc7b NB |
710 | bsb.mtime = __cpu_to_le64(time(0)); |
711 | bsb.arraystart = 0; | |
712 | bsb.length = __cpu_to_le64(last_block); | |
713 | ||
714 | /* Decide offset for the backup, llseek the spares, and write | |
715 | * a leading superblock 4K earlier. | |
716 | */ | |
e86c9dd6 | 717 | for (i=array.raid_disks; i<d; i++) { |
94a20f0c N |
718 | char abuf[4096+512]; |
719 | char *buf = (char*)(((unsigned long)abuf+511)& ~511); | |
06b0d786 NB |
720 | if (i==d-1 && backup_file) { |
721 | /* This is the backup file */ | |
722 | offsets[i] = 8; | |
723 | } else | |
724 | offsets[i] += sra->component_size - last_block - 8; | |
2efedc7b NB |
725 | if (lseek64(fdlist[i], (offsets[i]<<9) - 4096, 0) |
726 | != (offsets[i]<<9) - 4096) { | |
e86c9dd6 NB |
727 | fprintf(stderr, Name ": could not seek...\n"); |
728 | goto abort; | |
729 | } | |
94a20f0c | 730 | memset(buf, 0, 4096); |
2efedc7b NB |
731 | bsb.devstart = __cpu_to_le64(offsets[i]); |
732 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
733 | memcpy(buf, &bsb, sizeof(bsb)); | |
734 | if (write(fdlist[i], buf, 4096) != 4096) { | |
735 | fprintf(stderr, Name ": could not write leading superblock\n"); | |
736 | goto abort; | |
737 | } | |
e86c9dd6 NB |
738 | } |
739 | array.level = nlevel; | |
740 | array.raid_disks = ndisks; | |
741 | array.chunk_size = nchunk; | |
742 | array.layout = nlayout; | |
743 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
758d3a8e NB |
744 | if (errno == ENOSPC) { |
745 | /* stripe cache is not big enough. | |
746 | * It needs to be 4 times chunksize_size, | |
747 | * and we assume pagesize is 4K | |
748 | */ | |
749 | if (sra->cache_size < 4 * (nchunk/4096)) { | |
750 | sysfs_set_num(sra, NULL, | |
751 | "stripe_cache_size", | |
752 | 4 * (nchunk/4096) +1); | |
753 | if (ioctl(fd, SET_ARRAY_INFO, | |
754 | &array) == 0) | |
755 | goto ok; | |
756 | } | |
757 | } | |
e86c9dd6 NB |
758 | fprintf(stderr, Name ": Cannot set device size/shape for %s: %s\n", |
759 | devname, strerror(errno)); | |
760 | goto abort; | |
761 | } | |
758d3a8e | 762 | ok: ; |
e86c9dd6 NB |
763 | |
764 | /* suspend the relevant region */ | |
765 | sysfs_set_num(sra, NULL, "suspend_hi", 0); /* just in case */ | |
766 | if (sysfs_set_num(sra, NULL, "suspend_lo", 0) < 0 || | |
767 | sysfs_set_num(sra, NULL, "suspend_hi", last_block) < 0) { | |
768 | fprintf(stderr, Name ": %s: failed to suspend device.\n", | |
769 | devname); | |
770 | goto abort_resume; | |
771 | } | |
772 | ||
773 | ||
774 | err = save_stripes(fdlist, offsets, | |
775 | odisks, ochunk, olevel, olayout, | |
776 | spares, fdlist+odisks, | |
06b0d786 | 777 | 0ULL, last_block*512); |
e86c9dd6 NB |
778 | |
779 | /* abort if there was an error */ | |
780 | if (err < 0) { | |
781 | fprintf(stderr, Name ": %s: failed to save critical region\n", | |
782 | devname); | |
783 | goto abort_resume; | |
784 | } | |
2efedc7b | 785 | |
e86c9dd6 | 786 | for (i=odisks; i<d ; i++) { |
353632d9 | 787 | bsb.devstart = __cpu_to_le64(offsets[i]); |
e86c9dd6 | 788 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); |
206c5eae | 789 | if (lseek64(fdlist[i], (offsets[i]+last_block)<<9, 0) < 0 || |
06b0d786 NB |
790 | write(fdlist[i], &bsb, sizeof(bsb)) != sizeof(bsb) || |
791 | fsync(fdlist[i]) != 0) { | |
94a20f0c | 792 | fprintf(stderr, Name ": %s: failed to save metadata for critical region backups.\n", |
206c5eae NB |
793 | devname); |
794 | goto abort_resume; | |
795 | } | |
e86c9dd6 NB |
796 | } |
797 | ||
798 | /* start the reshape happening */ | |
799 | if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) { | |
800 | fprintf(stderr, Name ": %s: failed to initiate reshape\n", | |
801 | devname); | |
802 | goto abort_resume; | |
803 | } | |
804 | /* wait for reshape to pass the critical region */ | |
805 | while(1) { | |
806 | unsigned long long comp; | |
48924014 | 807 | |
206c5eae NB |
808 | if (sysfs_get_ll(sra, NULL, "sync_completed", &comp)<0) { |
809 | sleep(5); | |
e86c9dd6 | 810 | break; |
206c5eae | 811 | } |
e86c9dd6 NB |
812 | if (comp >= nstripe) |
813 | break; | |
93ecfa01 N |
814 | if (comp == 0) { |
815 | /* Maybe it finished already */ | |
816 | char action[20]; | |
817 | if (sysfs_get_str(sra, NULL, "sync_action", | |
818 | action, 20) > 0 && | |
819 | strncmp(action, "reshape", 7) != 0) | |
820 | break; | |
821 | } | |
e86c9dd6 NB |
822 | sleep(1); |
823 | } | |
aba69144 | 824 | |
e86c9dd6 NB |
825 | /* invalidate superblocks */ |
826 | memset(&bsb, 0, sizeof(bsb)); | |
827 | for (i=odisks; i<d ; i++) { | |
828 | lseek64(fdlist[i], (offsets[i]+last_block)<<9, 0); | |
9fca7d62 NB |
829 | if (write(fdlist[i], &bsb, sizeof(bsb)) < 0) { |
830 | fprintf(stderr, Name ": %s: failed to invalidate metadata for raid disk %d\n", | |
831 | devname, i); | |
832 | } | |
e86c9dd6 NB |
833 | } |
834 | ||
835 | /* unsuspend. */ | |
836 | sysfs_set_num(sra, NULL, "suspend_lo", last_block); | |
837 | ||
838 | for (i=0; i<d; i++) | |
839 | if (fdlist[i] >= 0) | |
840 | close(fdlist[i]); | |
841 | free(fdlist); | |
842 | free(offsets); | |
06b0d786 NB |
843 | if (backup_file) |
844 | unlink(backup_file); | |
e86c9dd6 | 845 | |
0aa389dc | 846 | fprintf(stderr, Name ": ... critical section passed.\n"); |
e86c9dd6 NB |
847 | break; |
848 | } | |
849 | return 0; | |
850 | ||
851 | ||
852 | abort_resume: | |
853 | sysfs_set_num(sra, NULL, "suspend_lo", last_block); | |
854 | abort: | |
855 | for (i=0; i<array.nr_disks; i++) | |
856 | if (fdlist[i] >= 0) | |
857 | close(fdlist[i]); | |
858 | free(fdlist); | |
859 | free(offsets); | |
06b0d786 NB |
860 | if (backup_file) |
861 | unlink(backup_file); | |
e86c9dd6 NB |
862 | return 1; |
863 | ||
864 | } | |
353632d9 NB |
865 | |
866 | /* | |
867 | * If any spare contains md_back_data-1 which is recent wrt mtime, | |
868 | * write that data into the array and update the super blocks with | |
869 | * the new reshape_progress | |
870 | */ | |
06b0d786 | 871 | int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt, char *backup_file) |
353632d9 NB |
872 | { |
873 | int i, j; | |
874 | int old_disks; | |
353632d9 | 875 | unsigned long long *offsets; |
6e9eac4f NB |
876 | unsigned long long nstripe, ostripe, last_block; |
877 | int ndata, odata; | |
353632d9 NB |
878 | |
879 | if (info->delta_disks < 0) | |
880 | return 1; /* cannot handle a shrink */ | |
881 | if (info->new_level != info->array.level || | |
882 | info->new_layout != info->array.layout || | |
883 | info->new_chunk != info->array.chunk_size) | |
884 | return 1; /* Can only handle change in disks */ | |
885 | ||
886 | old_disks = info->array.raid_disks - info->delta_disks; | |
887 | ||
06b0d786 | 888 | for (i=old_disks-(backup_file?1:0); i<cnt; i++) { |
353632d9 | 889 | struct mdinfo dinfo; |
2efedc7b | 890 | char buf[4096]; |
06b0d786 | 891 | int fd; |
353632d9 NB |
892 | |
893 | /* This was a spare and may have some saved data on it. | |
894 | * Load the superblock, find and load the | |
895 | * backup_super_block. | |
896 | * If either fail, go on to next device. | |
897 | * If the backup contains no new info, just return | |
206c5eae | 898 | * else restore data and update all superblocks |
353632d9 | 899 | */ |
06b0d786 NB |
900 | if (i == old_disks-1) { |
901 | fd = open(backup_file, O_RDONLY); | |
902 | if (fd<0) | |
903 | continue; | |
06b0d786 NB |
904 | } else { |
905 | fd = fdlist[i]; | |
906 | if (fd < 0) | |
907 | continue; | |
3da92f27 | 908 | if (st->ss->load_super(st, fd, NULL)) |
06b0d786 | 909 | continue; |
353632d9 | 910 | |
3da92f27 NB |
911 | st->ss->getinfo_super(st, &dinfo); |
912 | st->ss->free_super(st); | |
913 | ||
06b0d786 NB |
914 | if (lseek64(fd, |
915 | (dinfo.data_offset + dinfo.component_size - 8) <<9, | |
916 | 0) < 0) | |
917 | continue; /* Cannot seek */ | |
918 | } | |
919 | if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) | |
353632d9 NB |
920 | continue; /* Cannot read */ |
921 | if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0) | |
922 | continue; | |
923 | if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) | |
924 | continue; /* bad checksum */ | |
925 | if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) | |
926 | continue; /* Wrong uuid */ | |
927 | ||
928 | if (info->array.utime > __le64_to_cpu(bsb.mtime) + 3600 || | |
929 | info->array.utime < __le64_to_cpu(bsb.mtime)) | |
930 | continue; /* time stamp is too bad */ | |
931 | ||
932 | if (__le64_to_cpu(bsb.arraystart) != 0) | |
933 | continue; /* Can only handle backup from start of array */ | |
934 | if (__le64_to_cpu(bsb.length) < | |
935 | info->reshape_progress) | |
936 | continue; /* No new data here */ | |
937 | ||
06b0d786 | 938 | if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) |
353632d9 | 939 | continue; /* Cannot seek */ |
2efedc7b | 940 | /* There should be a duplicate backup superblock 4k before here */ |
06b0d786 NB |
941 | if (lseek64(fd, -4096, 1) < 0 || |
942 | read(fd, buf, 4096) != 4096 || | |
9860f271 | 943 | memcmp(buf, &bsb, sizeof(bsb)) != 0) |
2efedc7b NB |
944 | continue; /* Cannot find leading superblock */ |
945 | ||
353632d9 NB |
946 | /* Now need the data offsets for all devices. */ |
947 | offsets = malloc(sizeof(*offsets)*info->array.raid_disks); | |
948 | for(j=0; j<info->array.raid_disks; j++) { | |
949 | if (fdlist[j] < 0) | |
950 | continue; | |
3da92f27 | 951 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 NB |
952 | /* FIXME should be this be an error */ |
953 | continue; | |
3da92f27 NB |
954 | st->ss->getinfo_super(st, &dinfo); |
955 | st->ss->free_super(st); | |
353632d9 NB |
956 | offsets[j] = dinfo.data_offset; |
957 | } | |
958 | printf(Name ": restoring critical section\n"); | |
959 | ||
960 | if (restore_stripes(fdlist, offsets, | |
961 | info->array.raid_disks, | |
962 | info->new_chunk, | |
963 | info->new_level, | |
964 | info->new_layout, | |
06b0d786 | 965 | fd, __le64_to_cpu(bsb.devstart)*512, |
353632d9 NB |
966 | 0, __le64_to_cpu(bsb.length)*512)) { |
967 | /* didn't succeed, so giveup */ | |
2295250a | 968 | return 1; |
353632d9 NB |
969 | } |
970 | ||
971 | /* Ok, so the data is restored. Let's update those superblocks. */ | |
972 | ||
973 | for (j=0; j<info->array.raid_disks; j++) { | |
974 | if (fdlist[j] < 0) continue; | |
3da92f27 | 975 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 | 976 | continue; |
3da92f27 | 977 | st->ss->getinfo_super(st, &dinfo); |
353632d9 | 978 | dinfo.reshape_progress = __le64_to_cpu(bsb.length); |
3da92f27 | 979 | st->ss->update_super(st, &dinfo, |
68c7d6d7 NB |
980 | "_reshape_progress", |
981 | NULL,0, 0, NULL); | |
3da92f27 NB |
982 | st->ss->store_super(st, fdlist[j]); |
983 | st->ss->free_super(st); | |
353632d9 NB |
984 | } |
985 | ||
986 | /* And we are done! */ | |
987 | return 0; | |
988 | } | |
6e9eac4f NB |
989 | /* Didn't find any backup data, try to see if any |
990 | * was needed. | |
991 | */ | |
992 | nstripe = ostripe = 0; | |
993 | odata = info->array.raid_disks - info->delta_disks - 1; | |
994 | if (info->array.level == 6) odata--; /* number of data disks */ | |
995 | ndata = info->array.raid_disks - 1; | |
996 | if (info->new_level == 6) ndata--; | |
997 | last_block = 0; | |
998 | while (nstripe >= ostripe) { | |
999 | nstripe += info->new_chunk / 512; | |
1000 | last_block = nstripe * ndata; | |
1001 | ostripe = last_block / odata / (info->array.chunk_size/512) * | |
1002 | (info->array.chunk_size/512); | |
1003 | } | |
1004 | ||
1005 | if (info->reshape_progress >= last_block) | |
1006 | return 0; | |
1007 | /* needed to recover critical section! */ | |
2295250a | 1008 | return 1; |
353632d9 | 1009 | } |