]>
Commit | Line | Data |
---|---|---|
e5329c37 NB |
1 | /* |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2001-2004 Neil Brown <neilb@cse.unsw.edu.au> | |
5 | * | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | * Author: Neil Brown | |
22 | * Email: <neilb@cse.unsw.edu.au> | |
23 | * Paper: Neil Brown | |
24 | * School of Computer Science and Engineering | |
25 | * The University of New South Wales | |
26 | * Sydney, 2052 | |
27 | * Australia | |
28 | */ | |
29 | #include "mdadm.h" | |
30 | #include "dlink.h" | |
31 | ||
32 | #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN) | |
33 | #error no endian defined | |
34 | #endif | |
35 | #include "md_u.h" | |
36 | #include "md_p.h" | |
37 | ||
38 | int Grow_Add_device(char *devname, int fd, char *newdev) | |
39 | { | |
40 | /* Add a device to an active array. | |
41 | * Currently, just extend a linear array. | |
42 | * This requires writing a new superblock on the | |
43 | * new device, calling the kernel to add the device, | |
44 | * and if that succeeds, update the superblock on | |
45 | * all other devices. | |
46 | * This means that we need to *find* all other devices. | |
47 | */ | |
4b1ac34b NB |
48 | struct mdinfo info; |
49 | ||
50 | void *super = NULL; | |
e5329c37 NB |
51 | struct stat stb; |
52 | int nfd, fd2; | |
53 | int d, nd; | |
82d9eba6 | 54 | struct supertype *st = NULL; |
e5329c37 NB |
55 | |
56 | ||
4b1ac34b | 57 | if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) { |
e5329c37 NB |
58 | fprintf(stderr, Name ": cannot get array info for %s\n", devname); |
59 | return 1; | |
60 | } | |
61 | ||
82d9eba6 NB |
62 | st = super_by_version(info.array.major_version, info.array.minor_version); |
63 | if (!st) { | |
f9ce90ba NB |
64 | fprintf(stderr, Name ": cannot handle arrays with superblock version %d\n", info.array.major_version); |
65 | return 1; | |
66 | } | |
67 | ||
4b1ac34b | 68 | if (info.array.level != -1) { |
e5329c37 NB |
69 | fprintf(stderr, Name ": can only add devices to linear arrays\n"); |
70 | return 1; | |
71 | } | |
72 | ||
73 | nfd = open(newdev, O_RDWR|O_EXCL); | |
74 | if (nfd < 0) { | |
75 | fprintf(stderr, Name ": cannot open %s\n", newdev); | |
76 | return 1; | |
77 | } | |
78 | fstat(nfd, &stb); | |
79 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
80 | fprintf(stderr, Name ": %s is not a block device!\n", newdev); | |
81 | close(nfd); | |
82 | return 1; | |
83 | } | |
84 | /* now check out all the devices and make sure we can read the superblock */ | |
4b1ac34b | 85 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
86 | mdu_disk_info_t disk; |
87 | char *dv; | |
88 | ||
89 | disk.number = d; | |
90 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) { | |
91 | fprintf(stderr, Name ": cannot get device detail for device %d\n", | |
92 | d); | |
93 | return 1; | |
94 | } | |
16c6fa80 | 95 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 NB |
96 | if (!dv) { |
97 | fprintf(stderr, Name ": cannot find device file for device %d\n", | |
98 | d); | |
99 | return 1; | |
100 | } | |
16c6fa80 | 101 | fd2 = dev_open(dv, O_RDWR); |
e5329c37 NB |
102 | if (!fd2) { |
103 | fprintf(stderr, Name ": cannot open device file %s\n", dv); | |
104 | return 1; | |
105 | } | |
4b1ac34b NB |
106 | if (super) free(super); |
107 | super= NULL; | |
82d9eba6 | 108 | if (st->ss->load_super(st, fd2, &super, NULL)) { |
e5329c37 NB |
109 | fprintf(stderr, Name ": cannot find super block on %s\n", dv); |
110 | close(fd2); | |
111 | return 1; | |
112 | } | |
113 | close(fd2); | |
114 | } | |
115 | /* Ok, looks good. Lets update the superblock and write it out to | |
116 | * newdev. | |
117 | */ | |
118 | ||
4b1ac34b NB |
119 | info.disk.number = d; |
120 | info.disk.major = major(stb.st_rdev); | |
121 | info.disk.minor = minor(stb.st_rdev); | |
122 | info.disk.raid_disk = d; | |
123 | info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
82d9eba6 | 124 | st->ss->update_super(&info, super, "grow", newdev, 0); |
e5329c37 | 125 | |
96395475 | 126 | if (st->ss->store_super(st, nfd, super)) { |
e5329c37 NB |
127 | fprintf(stderr, Name ": Cannot store new superblock on %s\n", newdev); |
128 | close(nfd); | |
129 | return 1; | |
130 | } | |
e5329c37 | 131 | close(nfd); |
4b1ac34b NB |
132 | |
133 | if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) { | |
e5329c37 NB |
134 | fprintf(stderr, Name ": Cannot add new disk to this array\n"); |
135 | return 1; | |
136 | } | |
137 | /* Well, that seems to have worked. | |
138 | * Now go through and update all superblocks | |
139 | */ | |
140 | ||
4b1ac34b | 141 | if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) { |
e5329c37 NB |
142 | fprintf(stderr, Name ": cannot get array info for %s\n", devname); |
143 | return 1; | |
144 | } | |
145 | ||
146 | nd = d; | |
4b1ac34b | 147 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
148 | mdu_disk_info_t disk; |
149 | char *dv; | |
150 | ||
151 | disk.number = d; | |
152 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) { | |
153 | fprintf(stderr, Name ": cannot get device detail for device %d\n", | |
154 | d); | |
155 | return 1; | |
156 | } | |
16c6fa80 | 157 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 NB |
158 | if (!dv) { |
159 | fprintf(stderr, Name ": cannot find device file for device %d\n", | |
160 | d); | |
161 | return 1; | |
162 | } | |
16c6fa80 | 163 | fd2 = dev_open(dv, O_RDWR); |
e5329c37 NB |
164 | if (fd2 < 0) { |
165 | fprintf(stderr, Name ": cannot open device file %s\n", dv); | |
166 | return 1; | |
167 | } | |
82d9eba6 | 168 | if (st->ss->load_super(st, fd2, &super, NULL)) { |
e5329c37 NB |
169 | fprintf(stderr, Name ": cannot find super block on %s\n", dv); |
170 | close(fd); | |
171 | return 1; | |
172 | } | |
4b1ac34b NB |
173 | info.array.raid_disks = nd+1; |
174 | info.array.nr_disks = nd+1; | |
175 | info.array.active_disks = nd+1; | |
176 | info.array.working_disks = nd+1; | |
177 | info.disk.number = nd; | |
178 | info.disk.major = major(stb.st_rdev); | |
179 | info.disk.minor = minor(stb.st_rdev); | |
180 | info.disk.raid_disk = nd; | |
181 | info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
82d9eba6 | 182 | st->ss->update_super(&info, super, "grow", dv, 0); |
4b1ac34b | 183 | |
96395475 | 184 | if (st->ss->store_super(st, fd2, super)) { |
e5329c37 NB |
185 | fprintf(stderr, Name ": Cannot store new superblock on %s\n", dv); |
186 | close(fd2); | |
187 | return 1; | |
188 | } | |
189 | close(fd2); | |
190 | } | |
191 | ||
192 | return 0; | |
193 | } | |
f5e166fe | 194 | |
8fac0577 | 195 | int Grow_addbitmap(char *devname, int fd, char *file, int chunk, int delay, int write_behind, int force) |
f5e166fe NB |
196 | { |
197 | /* | |
198 | * First check that array doesn't have a bitmap | |
199 | * Then create the bitmap | |
200 | * Then add it | |
201 | * | |
202 | * For internal bitmaps, we need to check the version, | |
203 | * find all the active devices, and write the bitmap block | |
204 | * to all devices | |
205 | */ | |
206 | mdu_bitmap_file_t bmf; | |
207 | mdu_array_info_t array; | |
208 | struct supertype *st; | |
dcec9ee5 NB |
209 | int major = BITMAP_MAJOR_HI; |
210 | int vers = md_get_version(fd); | |
8fac0577 | 211 | unsigned long long bitmapsize, array_size; |
dcec9ee5 NB |
212 | |
213 | if (vers < 9003) { | |
214 | major = BITMAP_MAJOR_HOSTENDIAN; | |
215 | #ifdef __BIG_ENDIAN | |
216 | fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n" | |
217 | " between different architectured. Consider upgrading the Linux kernel.\n"); | |
218 | #endif | |
219 | } | |
f5e166fe NB |
220 | |
221 | if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) { | |
353632d9 | 222 | if (errno == ENOMEM) |
f5e166fe NB |
223 | fprintf(stderr, Name ": Memory allocation failure.\n"); |
224 | else | |
225 | fprintf(stderr, Name ": bitmaps not supported by this kernel.\n"); | |
226 | return 1; | |
227 | } | |
228 | if (bmf.pathname[0]) { | |
fe80f49b NB |
229 | if (strcmp(file,"none")==0) { |
230 | if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) { | |
231 | fprintf(stderr, Name ": failed to remove bitmap %s\n", | |
232 | bmf.pathname); | |
233 | return 1; | |
234 | } | |
235 | return 0; | |
236 | } | |
f5e166fe NB |
237 | fprintf(stderr, Name ": %s already has a bitmap (%s)\n", |
238 | devname, bmf.pathname); | |
239 | return 1; | |
240 | } | |
241 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
242 | fprintf(stderr, Name ": cannot get array status for %s\n", devname); | |
243 | return 1; | |
244 | } | |
245 | if (array.state & (1<<MD_SB_BITMAP_PRESENT)) { | |
fe80f49b NB |
246 | if (strcmp(file, "none")==0) { |
247 | array.state &= ~(1<<MD_SB_BITMAP_PRESENT); | |
248 | if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) { | |
249 | fprintf(stderr, Name ": failed to remove internal bitmap.\n"); | |
250 | return 1; | |
251 | } | |
252 | return 0; | |
253 | } | |
f5e166fe NB |
254 | fprintf(stderr, Name ": Internal bitmap already present on %s\n", |
255 | devname); | |
256 | return 1; | |
257 | } | |
8fac0577 NB |
258 | bitmapsize = array.size; |
259 | bitmapsize <<= 1; | |
260 | #ifdef BLKGETSIZE64 | |
261 | if (ioctl(fd, BLKGETSIZE64, &array_size) == 0 && | |
262 | array_size > (0x7fffffffULL<<9)) { | |
263 | /* Array is big enough that we cannot trust array.size | |
264 | * try other approaches | |
265 | */ | |
266 | bitmapsize = get_component_size(fd); | |
267 | } | |
268 | #endif | |
269 | if (bitmapsize == 0) { | |
270 | fprintf(stderr, Name ": Cannot reliably determine size of array to create bitmap - sorry.\n"); | |
271 | return 1; | |
272 | } | |
273 | ||
f9c25f1d NB |
274 | if (array.level == 10) { |
275 | int ncopies = (array.layout&255)*(array.layout>>8); | |
276 | bitmapsize = bitmapsize * array.raid_disks / ncopies; | |
277 | } | |
278 | ||
f5e166fe NB |
279 | st = super_by_version(array.major_version, array.minor_version); |
280 | if (!st) { | |
281 | fprintf(stderr, Name ": Cannot understand version %d.%d\n", | |
282 | array.major_version, array.minor_version); | |
283 | return 1; | |
284 | } | |
fe80f49b NB |
285 | if (strcmp(file, "none") == 0) { |
286 | fprintf(stderr, Name ": no bitmap found on %s\n", devname); | |
287 | return 1; | |
288 | } else if (strcmp(file, "internal") == 0) { | |
f5e166fe | 289 | int d; |
ea329559 | 290 | for (d=0; d< st->max_devs; d++) { |
f5e166fe NB |
291 | mdu_disk_info_t disk; |
292 | char *dv; | |
293 | disk.number = d; | |
294 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
295 | continue; | |
296 | if (disk.major == 0 && | |
297 | disk.minor == 0) | |
298 | continue; | |
299 | if ((disk.state & (1<<MD_DISK_SYNC))==0) | |
300 | continue; | |
16c6fa80 | 301 | dv = map_dev(disk.major, disk.minor, 1); |
f5e166fe NB |
302 | if (dv) { |
303 | void *super; | |
16c6fa80 | 304 | int fd2 = dev_open(dv, O_RDWR); |
f5e166fe NB |
305 | if (fd2 < 0) |
306 | continue; | |
307 | if (st->ss->load_super(st, fd2, &super, NULL)==0) { | |
21e92547 | 308 | if (st->ss->add_internal_bitmap(st, super, |
e86c9dd6 | 309 | chunk, delay, write_behind, |
21e92547 NB |
310 | bitmapsize, 0, major)) |
311 | st->ss->write_bitmap(st, fd2, super); | |
312 | else { | |
313 | fprintf(stderr, Name ": failed to create internal bitmap - chunksize problem.\n"); | |
314 | close(fd2); | |
315 | return 1; | |
316 | } | |
f5e166fe NB |
317 | } |
318 | close(fd2); | |
319 | } | |
320 | } | |
321 | array.state |= (1<<MD_SB_BITMAP_PRESENT); | |
322 | if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) { | |
323 | fprintf(stderr, Name ": failed to set internal bitmap.\n"); | |
324 | return 1; | |
325 | } | |
fe80f49b NB |
326 | } else { |
327 | int uuid[4]; | |
328 | int bitmap_fd; | |
329 | int d; | |
330 | int max_devs = st->max_devs; | |
331 | void *super = NULL; | |
332 | if (chunk == UnSet) | |
333 | chunk = DEFAULT_BITMAP_CHUNK; | |
334 | ||
335 | /* try to load a superblock */ | |
336 | for (d=0; d<max_devs; d++) { | |
337 | mdu_disk_info_t disk; | |
338 | char *dv; | |
339 | int fd2; | |
340 | disk.number = d; | |
341 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
342 | continue; | |
343 | if ((disk.major==0 && disk.minor==0) || | |
344 | (disk.state & (1<<MD_DISK_REMOVED))) | |
345 | continue; | |
16c6fa80 | 346 | dv = map_dev(disk.major, disk.minor, 1); |
fe80f49b | 347 | if (!dv) continue; |
16c6fa80 | 348 | fd2 = dev_open(dv, O_RDONLY); |
fe80f49b NB |
349 | if (fd2 >= 0 && |
350 | st->ss->load_super(st, fd2, &super, NULL) == 0) { | |
351 | close(fd2); | |
352 | st->ss->uuid_from_super(uuid, super); | |
353 | break; | |
354 | } | |
355 | close(fd2); | |
356 | } | |
357 | if (d == max_devs) { | |
358 | fprintf(stderr, Name ": cannot find UUID for array!\n"); | |
359 | return 1; | |
360 | } | |
8fac0577 | 361 | if (CreateBitmap(file, force, (char*)uuid, chunk, |
f9c25f1d | 362 | delay, write_behind, bitmapsize, major)) { |
fe80f49b NB |
363 | return 1; |
364 | } | |
365 | bitmap_fd = open(file, O_RDWR); | |
366 | if (bitmap_fd < 0) { | |
8fac0577 | 367 | fprintf(stderr, Name ": weird: %s cannot be opened\n", |
fe80f49b NB |
368 | file); |
369 | return 1; | |
370 | } | |
371 | if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
372 | fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n", | |
373 | devname, strerror(errno)); | |
374 | return 1; | |
375 | } | |
376 | } | |
f5e166fe NB |
377 | |
378 | return 0; | |
379 | } | |
380 | ||
e86c9dd6 NB |
381 | |
382 | /* | |
383 | * When reshaping an array we might need to backup some data. | |
384 | * This is written to all spares with a 'super_block' describing it. | |
385 | * The superblock goes 1K form the end of the used space on the | |
386 | * device. | |
387 | * It if written after the backup is complete. | |
388 | * It has the following structure. | |
389 | */ | |
390 | ||
391 | struct mdp_backup_super { | |
392 | char magic[16]; /* md_backup_data-1 */ | |
393 | __u8 set_uuid[16]; | |
394 | __u64 mtime; | |
395 | /* start/sizes in 512byte sectors */ | |
396 | __u64 devstart; | |
397 | __u64 arraystart; | |
398 | __u64 length; | |
399 | __u32 sb_csum; /* csum of preceeding bytes. */ | |
400 | }; | |
401 | ||
402 | int bsb_csum(char *buf, int len) | |
403 | { | |
404 | int i; | |
405 | int csum = 0; | |
406 | for (i=0; i<len; i++) | |
407 | csum = (csum<<3) + buf[0]; | |
408 | return __cpu_to_le32(csum); | |
409 | } | |
410 | ||
06b0d786 | 411 | int Grow_reshape(char *devname, int fd, int quiet, char *backup_file, |
e86c9dd6 NB |
412 | long long size, |
413 | int level, int layout, int chunksize, int raid_disks) | |
414 | { | |
415 | /* Make some changes in the shape of an array. | |
416 | * The kernel must support the change. | |
417 | * Different reshapes have subtly different meaning for different | |
418 | * levels, so we need to check the current state of the array | |
419 | * and go from there. | |
420 | */ | |
421 | struct mdu_array_info_s array; | |
422 | char *c; | |
423 | ||
424 | struct mdp_backup_super bsb; | |
425 | struct supertype *st; | |
426 | ||
427 | int nlevel, olevel; | |
428 | int nchunk, ochunk; | |
429 | int nlayout, olayout; | |
430 | int ndisks, odisks; | |
431 | int ndata, odata; | |
432 | unsigned long long nstripe, ostripe, last_block; | |
433 | int *fdlist; | |
434 | unsigned long long *offsets; | |
435 | int d, i, spares; | |
436 | int nrdisks; | |
437 | int err; | |
438 | void *super = NULL; | |
439 | ||
440 | struct sysarray *sra; | |
441 | struct sysdev *sd; | |
442 | ||
443 | if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) { | |
444 | fprintf(stderr, Name ": %s is not an active md array - aborting\n", | |
445 | devname); | |
446 | return 1; | |
447 | } | |
448 | c = map_num(pers, array.level); | |
449 | if (c == NULL) c = "-unknown-"; | |
450 | switch(array.level) { | |
451 | default: /* raid0, linear, multipath cannot be reconfigured */ | |
452 | fprintf(stderr, Name ": %s array %s cannot be reshaped.\n", | |
453 | c, devname); | |
454 | return 1; | |
455 | ||
456 | case LEVEL_FAULTY: /* only 'layout' change is permitted */ | |
457 | ||
458 | if (size >= 0) { | |
459 | fprintf(stderr, Name ": %s: Cannot change size of a 'faulty' array\n", | |
460 | devname); | |
461 | return 1; | |
462 | } | |
463 | if (level != UnSet && level != LEVEL_FAULTY) { | |
464 | fprintf(stderr, Name ": %s: Cannot change RAID level of a 'faulty' array\n", | |
465 | devname); | |
466 | return 1; | |
467 | } | |
468 | if (chunksize || raid_disks) { | |
469 | fprintf(stderr, Name ": %s: Cannot change chunksize or disks of a 'faulty' array\n", | |
470 | devname); | |
471 | return 1; | |
472 | } | |
473 | if (layout == UnSet) | |
474 | return 0; /* nothing to do.... */ | |
475 | ||
476 | array.layout = layout; | |
477 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
478 | fprintf(stderr, Name ": Cannot set layout for %s: %s\n", | |
479 | devname, strerror(errno)); | |
480 | return 1; | |
481 | } | |
482 | if (!quiet) | |
483 | printf("layout for %s set to %d\n", devname, array.layout); | |
484 | return 0; | |
485 | ||
486 | case 1: /* raid_disks and size can each be changed. They are independant */ | |
487 | ||
488 | if (level != UnSet && level != 1) { | |
489 | fprintf(stderr, Name ": %s: Cannot change RAID level of a RAID1 array.\n", | |
490 | devname); | |
491 | return 1; | |
492 | } | |
493 | if (chunksize || layout != UnSet) { | |
494 | fprintf(stderr, Name ": %s: Cannot change chunk size of layout for a RAID1 array.\n", | |
495 | devname); | |
496 | return 1; | |
497 | } | |
498 | ||
499 | /* Each can trigger a resync/recovery which will block the | |
500 | * other from happening. Later we could block | |
501 | * resync for the duration via 'sync_action'... | |
502 | */ | |
503 | if (raid_disks >= 0) | |
504 | array.raid_disks = raid_disks; | |
505 | if (size >= 0) | |
506 | array.size = size; | |
507 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
508 | fprintf(stderr, Name ": Cannot set device size/shape for %s: %s\n", | |
509 | devname, strerror(errno)); | |
510 | return 1; | |
511 | } | |
512 | return 0; | |
513 | ||
514 | case 4: | |
515 | case 5: | |
516 | case 6: | |
517 | st = super_by_version(array.major_version, | |
518 | array.minor_version); | |
519 | /* size can be changed independantly. | |
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 | } |
353632d9 | 613 | printf("mdadm: Need to backup %lluK of critical section..\n", last_block/2); |
e86c9dd6 NB |
614 | |
615 | sra = sysfs_read(fd, 0, | |
616 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE); | |
617 | if (!sra) { | |
618 | fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n", | |
619 | devname); | |
620 | return 1; | |
621 | } | |
622 | ||
623 | if (last_block >= sra->component_size/2) { | |
624 | fprintf(stderr, Name ": %s: Something wrong - reshape aborted\n", | |
625 | devname); | |
626 | return 1; | |
627 | } | |
06b0d786 NB |
628 | if (sra->spares == 0 && backup_file == NULL) { |
629 | fprintf(stderr, Name ": %s: Cannot grow - need a spare or backup-file to backup critical section\n", | |
353632d9 NB |
630 | devname); |
631 | return 1; | |
632 | } | |
e86c9dd6 NB |
633 | |
634 | nrdisks = array.nr_disks + sra->spares; | |
635 | /* Now we need to open all these devices so we can read/write. | |
636 | */ | |
06b0d786 NB |
637 | fdlist = malloc((1+nrdisks) * sizeof(int)); |
638 | offsets = malloc((1+nrdisks) * sizeof(offsets[0])); | |
e86c9dd6 NB |
639 | if (!fdlist || !offsets) { |
640 | fprintf(stderr, Name ": malloc failed: grow aborted\n"); | |
641 | return 1; | |
642 | } | |
06b0d786 | 643 | for (d=0; d <= nrdisks; d++) |
e86c9dd6 NB |
644 | fdlist[d] = -1; |
645 | d = array.raid_disks; | |
646 | for (sd = sra->devs; sd; sd=sd->next) { | |
647 | if (sd->state & (1<<MD_DISK_FAULTY)) | |
648 | continue; | |
649 | if (sd->state & (1<<MD_DISK_SYNC)) { | |
16c6fa80 NB |
650 | char *dn = map_dev(sd->major, sd->minor, 1); |
651 | fdlist[sd->role] = dev_open(dn, O_RDONLY); | |
e86c9dd6 NB |
652 | offsets[sd->role] = sd->offset; |
653 | if (fdlist[sd->role] < 0) { | |
654 | fprintf(stderr, Name ": %s: cannot open component %s\n", | |
655 | devname, dn); | |
656 | goto abort; | |
657 | } | |
658 | } else { | |
659 | /* spare */ | |
16c6fa80 NB |
660 | char *dn = map_dev(sd->major, sd->minor, 1); |
661 | fdlist[d] = dev_open(dn, O_RDWR); | |
e86c9dd6 NB |
662 | offsets[d] = sd->offset; |
663 | if (fdlist[d]<0) { | |
664 | fprintf(stderr, Name ": %s: cannot open component %s\n", | |
665 | devname, dn); | |
666 | goto abort; | |
667 | } | |
668 | d++; | |
669 | } | |
670 | } | |
671 | for (i=0 ; i<array.raid_disks; i++) | |
672 | if (fdlist[i] < 0) { | |
673 | fprintf(stderr, Name ": %s: failed to find device %d. Array might be degraded.\n" | |
674 | " --grow aborted\n", devname, i); | |
675 | goto abort; | |
676 | } | |
06b0d786 NB |
677 | spares = sra->spares; |
678 | if (backup_file) { | |
679 | fdlist[d] = open(backup_file, O_RDWR|O_CREAT|O_EXCL, 0600); | |
680 | if (fdlist[d] < 0) { | |
681 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
682 | devname, backup_file, strerror(errno)); | |
683 | goto abort; | |
684 | } | |
685 | offsets[d] = 8; | |
686 | d++; | |
687 | spares++; | |
688 | } | |
e86c9dd6 | 689 | if (fdlist[array.raid_disks] < 0) { |
06b0d786 | 690 | fprintf(stderr, Name ": %s: failed to find a spare and no backup-file given - --grow aborted\n", |
e86c9dd6 NB |
691 | devname); |
692 | goto abort; | |
693 | } | |
694 | ||
695 | /* Find a superblock */ | |
696 | if (st->ss->load_super(st, fdlist[0], &super, NULL)) { | |
697 | fprintf(stderr, Name ": %s: Cannot find a superblock\n", | |
698 | devname); | |
699 | goto abort; | |
700 | } | |
701 | ||
2efedc7b NB |
702 | |
703 | memcpy(bsb.magic, "md_backup_data-1", 16); | |
704 | st->ss->uuid_from_super((int*)&bsb.set_uuid, super); | |
705 | bsb.mtime = __cpu_to_le64(time(0)); | |
706 | bsb.arraystart = 0; | |
707 | bsb.length = __cpu_to_le64(last_block); | |
708 | ||
709 | /* Decide offset for the backup, llseek the spares, and write | |
710 | * a leading superblock 4K earlier. | |
711 | */ | |
e86c9dd6 | 712 | for (i=array.raid_disks; i<d; i++) { |
2efedc7b | 713 | char buf[4096]; |
06b0d786 NB |
714 | if (i==d-1 && backup_file) { |
715 | /* This is the backup file */ | |
716 | offsets[i] = 8; | |
717 | } else | |
718 | offsets[i] += sra->component_size - last_block - 8; | |
2efedc7b NB |
719 | if (lseek64(fdlist[i], (offsets[i]<<9) - 4096, 0) |
720 | != (offsets[i]<<9) - 4096) { | |
e86c9dd6 NB |
721 | fprintf(stderr, Name ": could not seek...\n"); |
722 | goto abort; | |
723 | } | |
2efedc7b NB |
724 | memset(buf, 0, sizeof(buf)); |
725 | bsb.devstart = __cpu_to_le64(offsets[i]); | |
726 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
727 | memcpy(buf, &bsb, sizeof(bsb)); | |
728 | if (write(fdlist[i], buf, 4096) != 4096) { | |
729 | fprintf(stderr, Name ": could not write leading superblock\n"); | |
730 | goto abort; | |
731 | } | |
e86c9dd6 NB |
732 | } |
733 | array.level = nlevel; | |
734 | array.raid_disks = ndisks; | |
735 | array.chunk_size = nchunk; | |
736 | array.layout = nlayout; | |
737 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
738 | fprintf(stderr, Name ": Cannot set device size/shape for %s: %s\n", | |
739 | devname, strerror(errno)); | |
740 | goto abort; | |
741 | } | |
742 | ||
743 | /* suspend the relevant region */ | |
744 | sysfs_set_num(sra, NULL, "suspend_hi", 0); /* just in case */ | |
745 | if (sysfs_set_num(sra, NULL, "suspend_lo", 0) < 0 || | |
746 | sysfs_set_num(sra, NULL, "suspend_hi", last_block) < 0) { | |
747 | fprintf(stderr, Name ": %s: failed to suspend device.\n", | |
748 | devname); | |
749 | goto abort_resume; | |
750 | } | |
751 | ||
752 | ||
753 | err = save_stripes(fdlist, offsets, | |
754 | odisks, ochunk, olevel, olayout, | |
755 | spares, fdlist+odisks, | |
06b0d786 | 756 | 0ULL, last_block*512); |
e86c9dd6 NB |
757 | |
758 | /* abort if there was an error */ | |
759 | if (err < 0) { | |
760 | fprintf(stderr, Name ": %s: failed to save critical region\n", | |
761 | devname); | |
762 | goto abort_resume; | |
763 | } | |
2efedc7b | 764 | |
e86c9dd6 | 765 | for (i=odisks; i<d ; i++) { |
353632d9 | 766 | bsb.devstart = __cpu_to_le64(offsets[i]); |
e86c9dd6 | 767 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); |
206c5eae | 768 | if (lseek64(fdlist[i], (offsets[i]+last_block)<<9, 0) < 0 || |
06b0d786 NB |
769 | write(fdlist[i], &bsb, sizeof(bsb)) != sizeof(bsb) || |
770 | fsync(fdlist[i]) != 0) { | |
206c5eae NB |
771 | fprintf(stderr, Name ": %s: fail to save metadata for critical region backups.\n", |
772 | devname); | |
773 | goto abort_resume; | |
774 | } | |
e86c9dd6 NB |
775 | } |
776 | ||
777 | /* start the reshape happening */ | |
778 | if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) { | |
779 | fprintf(stderr, Name ": %s: failed to initiate reshape\n", | |
780 | devname); | |
781 | goto abort_resume; | |
782 | } | |
783 | /* wait for reshape to pass the critical region */ | |
784 | while(1) { | |
785 | unsigned long long comp; | |
206c5eae NB |
786 | if (sysfs_get_ll(sra, NULL, "sync_completed", &comp)<0) { |
787 | sleep(5); | |
e86c9dd6 | 788 | break; |
206c5eae | 789 | } |
e86c9dd6 NB |
790 | if (comp >= nstripe) |
791 | break; | |
792 | sleep(1); | |
793 | } | |
f5e166fe | 794 | |
e86c9dd6 NB |
795 | /* invalidate superblocks */ |
796 | memset(&bsb, 0, sizeof(bsb)); | |
797 | for (i=odisks; i<d ; i++) { | |
798 | lseek64(fdlist[i], (offsets[i]+last_block)<<9, 0); | |
799 | write(fdlist[i], &bsb, sizeof(bsb)); | |
800 | } | |
801 | ||
802 | /* unsuspend. */ | |
803 | sysfs_set_num(sra, NULL, "suspend_lo", last_block); | |
804 | ||
805 | for (i=0; i<d; i++) | |
806 | if (fdlist[i] >= 0) | |
807 | close(fdlist[i]); | |
808 | free(fdlist); | |
809 | free(offsets); | |
06b0d786 NB |
810 | if (backup_file) |
811 | unlink(backup_file); | |
e86c9dd6 | 812 | |
206c5eae | 813 | printf(Name ": ... critical section passed.\n"); |
e86c9dd6 NB |
814 | break; |
815 | } | |
816 | return 0; | |
817 | ||
818 | ||
819 | abort_resume: | |
820 | sysfs_set_num(sra, NULL, "suspend_lo", last_block); | |
821 | abort: | |
822 | for (i=0; i<array.nr_disks; i++) | |
823 | if (fdlist[i] >= 0) | |
824 | close(fdlist[i]); | |
825 | free(fdlist); | |
826 | free(offsets); | |
06b0d786 NB |
827 | if (backup_file) |
828 | unlink(backup_file); | |
e86c9dd6 NB |
829 | return 1; |
830 | ||
831 | } | |
353632d9 NB |
832 | |
833 | /* | |
834 | * If any spare contains md_back_data-1 which is recent wrt mtime, | |
835 | * write that data into the array and update the super blocks with | |
836 | * the new reshape_progress | |
837 | */ | |
06b0d786 | 838 | int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt, char *backup_file) |
353632d9 NB |
839 | { |
840 | int i, j; | |
841 | int old_disks; | |
842 | int err = 0; | |
843 | unsigned long long *offsets; | |
844 | ||
845 | if (info->delta_disks < 0) | |
846 | return 1; /* cannot handle a shrink */ | |
847 | if (info->new_level != info->array.level || | |
848 | info->new_layout != info->array.layout || | |
849 | info->new_chunk != info->array.chunk_size) | |
850 | return 1; /* Can only handle change in disks */ | |
851 | ||
852 | old_disks = info->array.raid_disks - info->delta_disks; | |
853 | ||
06b0d786 | 854 | for (i=old_disks-(backup_file?1:0); i<cnt; i++) { |
353632d9 NB |
855 | void *super = NULL; |
856 | struct mdinfo dinfo; | |
353632d9 | 857 | struct mdp_backup_super bsb; |
2efedc7b | 858 | char buf[4096]; |
06b0d786 | 859 | int fd; |
353632d9 NB |
860 | |
861 | /* This was a spare and may have some saved data on it. | |
862 | * Load the superblock, find and load the | |
863 | * backup_super_block. | |
864 | * If either fail, go on to next device. | |
865 | * If the backup contains no new info, just return | |
206c5eae | 866 | * else restore data and update all superblocks |
353632d9 | 867 | */ |
06b0d786 NB |
868 | if (i == old_disks-1) { |
869 | fd = open(backup_file, O_RDONLY); | |
870 | if (fd<0) | |
871 | continue; | |
872 | if (lseek(fd, 4096, 0) != 4096) | |
873 | continue; | |
874 | } else { | |
875 | fd = fdlist[i]; | |
876 | if (fd < 0) | |
877 | continue; | |
878 | if (st->ss->load_super(st, fd, &super, NULL)) | |
879 | continue; | |
353632d9 | 880 | |
06b0d786 NB |
881 | st->ss->getinfo_super(&dinfo, super); |
882 | free(super); super = NULL; | |
883 | if (lseek64(fd, | |
884 | (dinfo.data_offset + dinfo.component_size - 8) <<9, | |
885 | 0) < 0) | |
886 | continue; /* Cannot seek */ | |
887 | } | |
888 | if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) | |
353632d9 NB |
889 | continue; /* Cannot read */ |
890 | if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0) | |
891 | continue; | |
892 | if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) | |
893 | continue; /* bad checksum */ | |
894 | if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) | |
895 | continue; /* Wrong uuid */ | |
896 | ||
897 | if (info->array.utime > __le64_to_cpu(bsb.mtime) + 3600 || | |
898 | info->array.utime < __le64_to_cpu(bsb.mtime)) | |
899 | continue; /* time stamp is too bad */ | |
900 | ||
901 | if (__le64_to_cpu(bsb.arraystart) != 0) | |
902 | continue; /* Can only handle backup from start of array */ | |
903 | if (__le64_to_cpu(bsb.length) < | |
904 | info->reshape_progress) | |
905 | continue; /* No new data here */ | |
906 | ||
06b0d786 | 907 | if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) |
353632d9 | 908 | continue; /* Cannot seek */ |
2efedc7b | 909 | /* There should be a duplicate backup superblock 4k before here */ |
06b0d786 NB |
910 | if (lseek64(fd, -4096, 1) < 0 || |
911 | read(fd, buf, 4096) != 4096 || | |
2efedc7b NB |
912 | memcmp(buf, &bsb, sizeof(buf)) != 0) |
913 | continue; /* Cannot find leading superblock */ | |
914 | ||
353632d9 NB |
915 | /* Now need the data offsets for all devices. */ |
916 | offsets = malloc(sizeof(*offsets)*info->array.raid_disks); | |
917 | for(j=0; j<info->array.raid_disks; j++) { | |
918 | if (fdlist[j] < 0) | |
919 | continue; | |
920 | if (st->ss->load_super(st, fdlist[j], &super, NULL)) | |
921 | /* FIXME should be this be an error */ | |
922 | continue; | |
31317663 | 923 | st->ss->getinfo_super(&dinfo, super); |
353632d9 NB |
924 | free(super); super = NULL; |
925 | offsets[j] = dinfo.data_offset; | |
926 | } | |
927 | printf(Name ": restoring critical section\n"); | |
928 | ||
929 | if (restore_stripes(fdlist, offsets, | |
930 | info->array.raid_disks, | |
931 | info->new_chunk, | |
932 | info->new_level, | |
933 | info->new_layout, | |
06b0d786 | 934 | fd, __le64_to_cpu(bsb.devstart)*512, |
353632d9 NB |
935 | 0, __le64_to_cpu(bsb.length)*512)) { |
936 | /* didn't succeed, so giveup */ | |
206c5eae | 937 | return -1; |
353632d9 NB |
938 | } |
939 | ||
940 | /* Ok, so the data is restored. Let's update those superblocks. */ | |
941 | ||
942 | for (j=0; j<info->array.raid_disks; j++) { | |
943 | if (fdlist[j] < 0) continue; | |
944 | if (st->ss->load_super(st, fdlist[j], &super, NULL)) | |
945 | continue; | |
31317663 | 946 | st->ss->getinfo_super(&dinfo, super); |
353632d9 NB |
947 | dinfo.reshape_progress = __le64_to_cpu(bsb.length); |
948 | st->ss->update_super(&dinfo, super, "_reshape_progress",NULL,0); | |
949 | st->ss->store_super(st, fdlist[j], super); | |
950 | free(super); | |
951 | } | |
952 | ||
953 | /* And we are done! */ | |
954 | return 0; | |
955 | } | |
956 | return err; | |
957 | } |