]>
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" | |
7236ee7a | 26 | #include <sys/mman.h> |
e5329c37 NB |
27 | |
28 | #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN) | |
29 | #error no endian defined | |
30 | #endif | |
31 | #include "md_u.h" | |
32 | #include "md_p.h" | |
33 | ||
e9e43ec3 N |
34 | #ifndef offsetof |
35 | #define offsetof(t,f) ((size_t)&(((t*)0)->f)) | |
36 | #endif | |
37 | ||
e5329c37 NB |
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 | ||
e5329c37 NB |
50 | struct stat stb; |
51 | int nfd, fd2; | |
52 | int d, nd; | |
82d9eba6 | 53 | struct supertype *st = NULL; |
4725bc31 | 54 | char *subarray = NULL; |
e5329c37 | 55 | |
4b1ac34b | 56 | if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) { |
e5329c37 NB |
57 | fprintf(stderr, Name ": cannot get array info for %s\n", devname); |
58 | return 1; | |
59 | } | |
60 | ||
4725bc31 N |
61 | if (info.array.level != -1) { |
62 | fprintf(stderr, Name ": can only add devices to linear arrays\n"); | |
63 | return 1; | |
64 | } | |
65 | ||
66 | st = super_by_fd(fd, &subarray); | |
82d9eba6 | 67 | if (!st) { |
f9ce90ba NB |
68 | fprintf(stderr, Name ": cannot handle arrays with superblock version %d\n", info.array.major_version); |
69 | return 1; | |
70 | } | |
71 | ||
4725bc31 N |
72 | if (subarray) { |
73 | fprintf(stderr, Name ": Cannot grow linear sub-arrays yet\n"); | |
74 | free(subarray); | |
75 | free(st); | |
e5329c37 NB |
76 | } |
77 | ||
6416d527 | 78 | nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT); |
e5329c37 NB |
79 | if (nfd < 0) { |
80 | fprintf(stderr, Name ": cannot open %s\n", newdev); | |
4725bc31 | 81 | free(st); |
e5329c37 NB |
82 | return 1; |
83 | } | |
84 | fstat(nfd, &stb); | |
85 | if ((stb.st_mode & S_IFMT) != S_IFBLK) { | |
86 | fprintf(stderr, Name ": %s is not a block device!\n", newdev); | |
87 | close(nfd); | |
4725bc31 | 88 | free(st); |
e5329c37 NB |
89 | return 1; |
90 | } | |
91 | /* now check out all the devices and make sure we can read the superblock */ | |
4b1ac34b | 92 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
93 | mdu_disk_info_t disk; |
94 | char *dv; | |
95 | ||
4725bc31 N |
96 | st->ss->free_super(st); |
97 | ||
e5329c37 NB |
98 | disk.number = d; |
99 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) { | |
100 | fprintf(stderr, Name ": cannot get device detail for device %d\n", | |
101 | d); | |
4725bc31 N |
102 | close(nfd); |
103 | free(st); | |
e5329c37 NB |
104 | return 1; |
105 | } | |
16c6fa80 | 106 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 NB |
107 | if (!dv) { |
108 | fprintf(stderr, Name ": cannot find device file for device %d\n", | |
109 | d); | |
4725bc31 N |
110 | close(nfd); |
111 | free(st); | |
e5329c37 NB |
112 | return 1; |
113 | } | |
16c6fa80 | 114 | fd2 = dev_open(dv, O_RDWR); |
e5329c37 NB |
115 | if (!fd2) { |
116 | fprintf(stderr, Name ": cannot open device file %s\n", dv); | |
4725bc31 N |
117 | close(nfd); |
118 | free(st); | |
e5329c37 NB |
119 | return 1; |
120 | } | |
3da92f27 NB |
121 | |
122 | if (st->ss->load_super(st, fd2, NULL)) { | |
e5329c37 | 123 | fprintf(stderr, Name ": cannot find super block on %s\n", dv); |
4725bc31 | 124 | close(nfd); |
e5329c37 | 125 | close(fd2); |
4725bc31 | 126 | free(st); |
e5329c37 NB |
127 | return 1; |
128 | } | |
129 | close(fd2); | |
130 | } | |
131 | /* Ok, looks good. Lets update the superblock and write it out to | |
132 | * newdev. | |
133 | */ | |
aba69144 | 134 | |
4b1ac34b NB |
135 | info.disk.number = d; |
136 | info.disk.major = major(stb.st_rdev); | |
137 | info.disk.minor = minor(stb.st_rdev); | |
138 | info.disk.raid_disk = d; | |
139 | info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
3da92f27 | 140 | st->ss->update_super(st, &info, "linear-grow-new", newdev, |
f752781f | 141 | 0, 0, NULL); |
e5329c37 | 142 | |
3da92f27 | 143 | if (st->ss->store_super(st, nfd)) { |
f752781f NB |
144 | fprintf(stderr, Name ": Cannot store new superblock on %s\n", |
145 | newdev); | |
e5329c37 NB |
146 | close(nfd); |
147 | return 1; | |
148 | } | |
e5329c37 | 149 | close(nfd); |
4b1ac34b NB |
150 | |
151 | if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) { | |
e5329c37 NB |
152 | fprintf(stderr, Name ": Cannot add new disk to this array\n"); |
153 | return 1; | |
154 | } | |
155 | /* Well, that seems to have worked. | |
156 | * Now go through and update all superblocks | |
157 | */ | |
158 | ||
4b1ac34b | 159 | if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) { |
e5329c37 NB |
160 | fprintf(stderr, Name ": cannot get array info for %s\n", devname); |
161 | return 1; | |
162 | } | |
163 | ||
164 | nd = d; | |
4b1ac34b | 165 | for (d=0 ; d < info.array.raid_disks ; d++) { |
e5329c37 NB |
166 | mdu_disk_info_t disk; |
167 | char *dv; | |
168 | ||
169 | disk.number = d; | |
170 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) { | |
171 | fprintf(stderr, Name ": cannot get device detail for device %d\n", | |
172 | d); | |
173 | return 1; | |
174 | } | |
16c6fa80 | 175 | dv = map_dev(disk.major, disk.minor, 1); |
e5329c37 NB |
176 | if (!dv) { |
177 | fprintf(stderr, Name ": cannot find device file for device %d\n", | |
178 | d); | |
179 | return 1; | |
180 | } | |
16c6fa80 | 181 | fd2 = dev_open(dv, O_RDWR); |
e5329c37 NB |
182 | if (fd2 < 0) { |
183 | fprintf(stderr, Name ": cannot open device file %s\n", dv); | |
184 | return 1; | |
185 | } | |
3da92f27 | 186 | if (st->ss->load_super(st, fd2, NULL)) { |
e5329c37 NB |
187 | fprintf(stderr, Name ": cannot find super block on %s\n", dv); |
188 | close(fd); | |
189 | return 1; | |
190 | } | |
4b1ac34b NB |
191 | info.array.raid_disks = nd+1; |
192 | info.array.nr_disks = nd+1; | |
193 | info.array.active_disks = nd+1; | |
194 | info.array.working_disks = nd+1; | |
f752781f | 195 | |
3da92f27 | 196 | st->ss->update_super(st, &info, "linear-grow-update", dv, |
f752781f | 197 | 0, 0, NULL); |
aba69144 | 198 | |
3da92f27 | 199 | if (st->ss->store_super(st, fd2)) { |
e5329c37 NB |
200 | fprintf(stderr, Name ": Cannot store new superblock on %s\n", dv); |
201 | close(fd2); | |
202 | return 1; | |
203 | } | |
204 | close(fd2); | |
205 | } | |
206 | ||
207 | return 0; | |
208 | } | |
f5e166fe | 209 | |
8fac0577 | 210 | int Grow_addbitmap(char *devname, int fd, char *file, int chunk, int delay, int write_behind, int force) |
f5e166fe NB |
211 | { |
212 | /* | |
213 | * First check that array doesn't have a bitmap | |
214 | * Then create the bitmap | |
215 | * Then add it | |
216 | * | |
217 | * For internal bitmaps, we need to check the version, | |
218 | * find all the active devices, and write the bitmap block | |
219 | * to all devices | |
220 | */ | |
221 | mdu_bitmap_file_t bmf; | |
222 | mdu_array_info_t array; | |
223 | struct supertype *st; | |
4725bc31 | 224 | char *subarray = NULL; |
dcec9ee5 NB |
225 | int major = BITMAP_MAJOR_HI; |
226 | int vers = md_get_version(fd); | |
8fac0577 | 227 | unsigned long long bitmapsize, array_size; |
dcec9ee5 NB |
228 | |
229 | if (vers < 9003) { | |
230 | major = BITMAP_MAJOR_HOSTENDIAN; | |
b3bd581b N |
231 | fprintf(stderr, Name ": Warning - bitmaps created on this kernel" |
232 | " are not portable\n" | |
233 | " between different architectures. Consider upgrading" | |
234 | " the Linux kernel.\n"); | |
dcec9ee5 | 235 | } |
f5e166fe NB |
236 | |
237 | if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) { | |
353632d9 | 238 | if (errno == ENOMEM) |
f5e166fe NB |
239 | fprintf(stderr, Name ": Memory allocation failure.\n"); |
240 | else | |
241 | fprintf(stderr, Name ": bitmaps not supported by this kernel.\n"); | |
242 | return 1; | |
243 | } | |
244 | if (bmf.pathname[0]) { | |
fe80f49b NB |
245 | if (strcmp(file,"none")==0) { |
246 | if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) { | |
247 | fprintf(stderr, Name ": failed to remove bitmap %s\n", | |
248 | bmf.pathname); | |
249 | return 1; | |
250 | } | |
251 | return 0; | |
252 | } | |
f5e166fe NB |
253 | fprintf(stderr, Name ": %s already has a bitmap (%s)\n", |
254 | devname, bmf.pathname); | |
255 | return 1; | |
256 | } | |
257 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
258 | fprintf(stderr, Name ": cannot get array status for %s\n", devname); | |
259 | return 1; | |
260 | } | |
261 | if (array.state & (1<<MD_SB_BITMAP_PRESENT)) { | |
fe80f49b NB |
262 | if (strcmp(file, "none")==0) { |
263 | array.state &= ~(1<<MD_SB_BITMAP_PRESENT); | |
264 | if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) { | |
265 | fprintf(stderr, Name ": failed to remove internal bitmap.\n"); | |
266 | return 1; | |
267 | } | |
268 | return 0; | |
269 | } | |
f5e166fe NB |
270 | fprintf(stderr, Name ": Internal bitmap already present on %s\n", |
271 | devname); | |
272 | return 1; | |
273 | } | |
4725bc31 N |
274 | |
275 | if (strcmp(file, "none") == 0) { | |
276 | fprintf(stderr, Name ": no bitmap found on %s\n", devname); | |
277 | return 1; | |
278 | } | |
5b28bd56 NB |
279 | if (array.level <= 0) { |
280 | fprintf(stderr, Name ": Bitmaps not meaningful with level %s\n", | |
281 | map_num(pers, array.level)?:"of this array"); | |
282 | return 1; | |
283 | } | |
8fac0577 NB |
284 | bitmapsize = array.size; |
285 | bitmapsize <<= 1; | |
beae1dfe | 286 | if (get_dev_size(fd, NULL, &array_size) && |
8fac0577 NB |
287 | array_size > (0x7fffffffULL<<9)) { |
288 | /* Array is big enough that we cannot trust array.size | |
289 | * try other approaches | |
290 | */ | |
291 | bitmapsize = get_component_size(fd); | |
292 | } | |
8fac0577 NB |
293 | if (bitmapsize == 0) { |
294 | fprintf(stderr, Name ": Cannot reliably determine size of array to create bitmap - sorry.\n"); | |
295 | return 1; | |
296 | } | |
297 | ||
f9c25f1d | 298 | if (array.level == 10) { |
8686f3ed | 299 | int ncopies = (array.layout&255)*((array.layout>>8)&255); |
f9c25f1d NB |
300 | bitmapsize = bitmapsize * array.raid_disks / ncopies; |
301 | } | |
302 | ||
4725bc31 | 303 | st = super_by_fd(fd, &subarray); |
f5e166fe NB |
304 | if (!st) { |
305 | fprintf(stderr, Name ": Cannot understand version %d.%d\n", | |
306 | array.major_version, array.minor_version); | |
307 | return 1; | |
308 | } | |
4725bc31 N |
309 | if (subarray) { |
310 | fprintf(stderr, Name ": Cannot add bitmaps to sub-arrays yet\n"); | |
311 | free(subarray); | |
312 | free(st); | |
fe80f49b | 313 | return 1; |
4725bc31 N |
314 | } |
315 | if (strcmp(file, "internal") == 0) { | |
f5e166fe | 316 | int d; |
ebeb3663 N |
317 | if (st->ss->add_internal_bitmap == NULL) { |
318 | fprintf(stderr, Name ": Internal bitmaps not supported " | |
319 | "with %s metadata\n", st->ss->name); | |
320 | return 1; | |
321 | } | |
ea329559 | 322 | for (d=0; d< st->max_devs; d++) { |
f5e166fe NB |
323 | mdu_disk_info_t disk; |
324 | char *dv; | |
325 | disk.number = d; | |
326 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
327 | continue; | |
328 | if (disk.major == 0 && | |
329 | disk.minor == 0) | |
330 | continue; | |
331 | if ((disk.state & (1<<MD_DISK_SYNC))==0) | |
332 | continue; | |
16c6fa80 | 333 | dv = map_dev(disk.major, disk.minor, 1); |
f5e166fe | 334 | if (dv) { |
16c6fa80 | 335 | int fd2 = dev_open(dv, O_RDWR); |
f5e166fe NB |
336 | if (fd2 < 0) |
337 | continue; | |
3da92f27 | 338 | if (st->ss->load_super(st, fd2, NULL)==0) { |
199171a2 | 339 | if (st->ss->add_internal_bitmap( |
3da92f27 | 340 | st, |
199171a2 NB |
341 | &chunk, delay, write_behind, |
342 | bitmapsize, 0, major) | |
343 | ) | |
3da92f27 | 344 | st->ss->write_bitmap(st, fd2); |
21e92547 NB |
345 | else { |
346 | fprintf(stderr, Name ": failed to create internal bitmap - chunksize problem.\n"); | |
347 | close(fd2); | |
348 | return 1; | |
349 | } | |
f5e166fe NB |
350 | } |
351 | close(fd2); | |
352 | } | |
353 | } | |
354 | array.state |= (1<<MD_SB_BITMAP_PRESENT); | |
355 | if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) { | |
ff634064 N |
356 | if (errno == EBUSY) |
357 | fprintf(stderr, Name | |
358 | ": Cannot add bitmap while array is" | |
359 | " resyncing or reshaping etc.\n"); | |
f5e166fe NB |
360 | fprintf(stderr, Name ": failed to set internal bitmap.\n"); |
361 | return 1; | |
362 | } | |
fe80f49b NB |
363 | } else { |
364 | int uuid[4]; | |
365 | int bitmap_fd; | |
366 | int d; | |
367 | int max_devs = st->max_devs; | |
fe80f49b NB |
368 | |
369 | /* try to load a superblock */ | |
370 | for (d=0; d<max_devs; d++) { | |
371 | mdu_disk_info_t disk; | |
372 | char *dv; | |
373 | int fd2; | |
374 | disk.number = d; | |
375 | if (ioctl(fd, GET_DISK_INFO, &disk) < 0) | |
376 | continue; | |
377 | if ((disk.major==0 && disk.minor==0) || | |
378 | (disk.state & (1<<MD_DISK_REMOVED))) | |
379 | continue; | |
16c6fa80 | 380 | dv = map_dev(disk.major, disk.minor, 1); |
fe80f49b | 381 | if (!dv) continue; |
16c6fa80 | 382 | fd2 = dev_open(dv, O_RDONLY); |
fe80f49b | 383 | if (fd2 >= 0 && |
3da92f27 | 384 | st->ss->load_super(st, fd2, NULL) == 0) { |
fe80f49b | 385 | close(fd2); |
3da92f27 | 386 | st->ss->uuid_from_super(st, uuid); |
fe80f49b NB |
387 | break; |
388 | } | |
389 | close(fd2); | |
390 | } | |
391 | if (d == max_devs) { | |
392 | fprintf(stderr, Name ": cannot find UUID for array!\n"); | |
393 | return 1; | |
394 | } | |
8fac0577 | 395 | if (CreateBitmap(file, force, (char*)uuid, chunk, |
f9c25f1d | 396 | delay, write_behind, bitmapsize, major)) { |
fe80f49b NB |
397 | return 1; |
398 | } | |
399 | bitmap_fd = open(file, O_RDWR); | |
400 | if (bitmap_fd < 0) { | |
8fac0577 | 401 | fprintf(stderr, Name ": weird: %s cannot be opened\n", |
fe80f49b NB |
402 | file); |
403 | return 1; | |
404 | } | |
405 | if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) { | |
ff634064 N |
406 | int err = errno; |
407 | if (errno == EBUSY) | |
408 | fprintf(stderr, Name | |
409 | ": Cannot add bitmap while array is" | |
410 | " resyncing or reshaping etc.\n"); | |
fe80f49b | 411 | fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n", |
ff634064 | 412 | devname, strerror(err)); |
fe80f49b NB |
413 | return 1; |
414 | } | |
415 | } | |
f5e166fe NB |
416 | |
417 | return 0; | |
418 | } | |
419 | ||
e86c9dd6 NB |
420 | |
421 | /* | |
422 | * When reshaping an array we might need to backup some data. | |
423 | * This is written to all spares with a 'super_block' describing it. | |
ff94fb86 | 424 | * The superblock goes 4K from the end of the used space on the |
e86c9dd6 NB |
425 | * device. |
426 | * It if written after the backup is complete. | |
427 | * It has the following structure. | |
428 | */ | |
429 | ||
5fdf37e3 | 430 | static struct mdp_backup_super { |
7236ee7a | 431 | char magic[16]; /* md_backup_data-1 or -2 */ |
e86c9dd6 NB |
432 | __u8 set_uuid[16]; |
433 | __u64 mtime; | |
434 | /* start/sizes in 512byte sectors */ | |
7236ee7a | 435 | __u64 devstart; /* address on backup device/file of data */ |
e86c9dd6 NB |
436 | __u64 arraystart; |
437 | __u64 length; | |
438 | __u32 sb_csum; /* csum of preceeding bytes. */ | |
7236ee7a N |
439 | __u32 pad1; |
440 | __u64 devstart2; /* offset in to data of second section */ | |
441 | __u64 arraystart2; | |
442 | __u64 length2; | |
443 | __u32 sb_csum2; /* csum of preceeding bytes. */ | |
444 | __u8 pad[512-68-32]; | |
5fdf37e3 | 445 | } __attribute__((aligned(512))) bsb, bsb2; |
e86c9dd6 | 446 | |
4411fb17 | 447 | static __u32 bsb_csum(char *buf, int len) |
e86c9dd6 NB |
448 | { |
449 | int i; | |
450 | int csum = 0; | |
451 | for (i=0; i<len; i++) | |
452 | csum = (csum<<3) + buf[0]; | |
453 | return __cpu_to_le32(csum); | |
454 | } | |
455 | ||
7236ee7a N |
456 | static int child_grow(int afd, struct mdinfo *sra, unsigned long blocks, |
457 | int *fds, unsigned long long *offsets, | |
458 | int disks, int chunk, int level, int layout, int data, | |
459 | int dests, int *destfd, unsigned long long *destoffsets); | |
460 | static int child_shrink(int afd, struct mdinfo *sra, unsigned long blocks, | |
461 | int *fds, unsigned long long *offsets, | |
462 | int disks, int chunk, int level, int layout, int data, | |
463 | int dests, int *destfd, unsigned long long *destoffsets); | |
464 | static int child_same_size(int afd, struct mdinfo *sra, unsigned long blocks, | |
465 | int *fds, unsigned long long *offsets, | |
e9e43ec3 | 466 | unsigned long long start, |
7236ee7a N |
467 | int disks, int chunk, int level, int layout, int data, |
468 | int dests, int *destfd, unsigned long long *destoffsets); | |
469 | ||
7f2ba464 | 470 | static int freeze_container(struct supertype *st) |
7236ee7a | 471 | { |
7f2ba464 DW |
472 | int container_dev = (st->container_dev != NoMdDev |
473 | ? st->container_dev : st->devnum); | |
474 | char *container = devnum2devname(container_dev); | |
475 | ||
476 | if (!container) { | |
477 | fprintf(stderr, Name | |
478 | ": could not determine container name, freeze aborted\n"); | |
479 | return -2; | |
480 | } | |
481 | ||
482 | if (block_monitor(container, 1)) { | |
483 | fprintf(stderr, Name ": failed to freeze container\n"); | |
484 | return -2; | |
485 | } | |
486 | ||
487 | return 1; | |
488 | } | |
489 | ||
490 | static void unfreeze_container(struct supertype *st) | |
491 | { | |
492 | int container_dev = (st->container_dev != NoMdDev | |
493 | ? st->container_dev : st->devnum); | |
494 | char *container = devnum2devname(container_dev); | |
495 | ||
496 | if (!container) { | |
497 | fprintf(stderr, Name | |
498 | ": could not determine container name, unfreeze aborted\n"); | |
499 | return; | |
500 | } | |
501 | ||
502 | unblock_monitor(container, 1); | |
503 | } | |
504 | ||
505 | static int freeze(struct supertype *st) | |
506 | { | |
507 | /* Try to freeze resync/rebuild on this array/container. | |
7236ee7a | 508 | * Return -1 if the array is busy, |
7f2ba464 | 509 | * return -2 container cannot be frozen, |
7236ee7a N |
510 | * return 0 if this kernel doesn't support 'frozen' |
511 | * return 1 if it worked. | |
512 | */ | |
7f2ba464 DW |
513 | if (st->ss->external) |
514 | return freeze_container(st); | |
515 | else { | |
516 | struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION); | |
517 | int err; | |
518 | ||
519 | if (!sra) | |
520 | return -1; | |
521 | err = sysfs_freeze_array(sra); | |
522 | sysfs_free(sra); | |
523 | return err; | |
524 | } | |
7236ee7a N |
525 | } |
526 | ||
7f2ba464 | 527 | static void unfreeze(struct supertype *st, int frozen) |
7236ee7a N |
528 | { |
529 | /* If 'frozen' is 1, unfreeze the array */ | |
7f2ba464 DW |
530 | if (frozen <= 0) |
531 | return; | |
532 | ||
533 | if (st->ss->external) | |
534 | return unfreeze_container(st); | |
535 | else { | |
536 | struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION); | |
537 | ||
538 | if (sra) | |
539 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
540 | else | |
541 | fprintf(stderr, Name ": failed to unfreeze array\n"); | |
542 | sysfs_free(sra); | |
543 | } | |
7236ee7a N |
544 | } |
545 | ||
4411fb17 | 546 | static void wait_reshape(struct mdinfo *sra) |
7236ee7a N |
547 | { |
548 | int fd = sysfs_get_fd(sra, NULL, "sync_action"); | |
549 | char action[20]; | |
550 | ||
551 | do { | |
552 | fd_set rfds; | |
553 | FD_ZERO(&rfds); | |
554 | FD_SET(fd, &rfds); | |
555 | select(fd+1, NULL, NULL, &rfds, NULL); | |
556 | ||
557 | if (sysfs_fd_get_str(fd, action, 20) < 0) { | |
558 | close(fd); | |
559 | return; | |
560 | } | |
561 | } while (strncmp(action, "reshape", 7) == 0); | |
562 | } | |
7bc71196 DW |
563 | |
564 | static int reshape_super(struct supertype *st, long long size, int level, | |
565 | int layout, int chunksize, int raid_disks, | |
566 | char *backup_file, char *dev, int verbose) | |
567 | { | |
568 | /* nothing extra to check in the native case */ | |
569 | if (!st->ss->external) | |
570 | return 0; | |
571 | if (!st->ss->reshape_super || | |
572 | !st->ss->manage_reshape) { | |
573 | fprintf(stderr, Name ": %s metadata does not support reshape\n", | |
574 | st->ss->name); | |
575 | return 1; | |
576 | } | |
577 | ||
578 | return st->ss->reshape_super(st, size, level, layout, chunksize, | |
579 | raid_disks, backup_file, dev, verbose); | |
580 | } | |
581 | ||
582 | static void sync_metadata(struct supertype *st) | |
583 | { | |
584 | if (st->ss->external) { | |
585 | if (st->update_tail) | |
586 | flush_metadata_updates(st); | |
587 | else | |
588 | st->ss->sync_metadata(st); | |
589 | } | |
590 | } | |
591 | ||
592 | static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n) | |
593 | { | |
594 | /* when dealing with external metadata subarrays we need to be | |
595 | * prepared to handle EAGAIN. The kernel may need to wait for | |
596 | * mdmon to mark the array active so the kernel can handle | |
597 | * allocations/writeback when preparing the reshape action | |
598 | * (md_allow_write()). We temporarily disable safe_mode_delay | |
599 | * to close a race with the array_state going clean before the | |
600 | * next write to raid_disks / stripe_cache_size | |
601 | */ | |
602 | char safe[50]; | |
603 | int rc; | |
604 | ||
605 | /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */ | |
606 | if (strcmp(name, "raid_disks") != 0 && | |
607 | strcmp(name, "stripe_cache_size") != 0) | |
608 | return sysfs_set_num(sra, NULL, name, n); | |
609 | ||
610 | rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe)); | |
611 | if (rc <= 0) | |
612 | return -1; | |
613 | sysfs_set_num(sra, NULL, "safe_mode_delay", 0); | |
614 | rc = sysfs_set_num(sra, NULL, name, n); | |
615 | if (rc < 0 && errno == EAGAIN) { | |
616 | ping_monitor(container); | |
617 | /* if we get EAGAIN here then the monitor is not active | |
618 | * so stop trying | |
619 | */ | |
620 | rc = sysfs_set_num(sra, NULL, name, n); | |
621 | } | |
622 | sysfs_set_str(sra, NULL, "safe_mode_delay", safe); | |
623 | return rc; | |
624 | } | |
625 | ||
626 | static int reshape_container_raid_disks(char *container, int raid_disks) | |
627 | { | |
628 | /* for each subarray switch to a raid level that can | |
629 | * support the reshape, and set raid disks | |
630 | */ | |
631 | struct mdstat_ent *ent, *e; | |
632 | int changed = 0, rv = 0, err = 0; | |
633 | ||
634 | ent = mdstat_read(1, 0); | |
635 | if (!ent) { | |
636 | fprintf(stderr, Name ": unable to read /proc/mdstat\n"); | |
637 | return -1; | |
638 | } | |
639 | ||
640 | changed = 0; | |
641 | for (e = ent; e; e = e->next) { | |
642 | struct mdinfo *sub; | |
643 | unsigned int cache; | |
644 | int level, takeover_delta = 0; | |
645 | ||
646 | if (!is_container_member(e, container)) | |
647 | continue; | |
648 | ||
649 | level = map_name(pers, e->level); | |
650 | if (level == 0) { | |
651 | sub = sysfs_read(-1, e->devnum, GET_VERSION); | |
652 | if (!sub) | |
653 | break; | |
654 | /* metadata records 'orig_level' */ | |
655 | rv = sysfs_set_num(sub, NULL, "level", 4); | |
656 | if (rv < 0) { | |
657 | err = errno; | |
658 | break; | |
659 | } | |
660 | /* we want spares to be used for capacity | |
661 | * expansion, not rebuild | |
662 | */ | |
663 | takeover_delta = 1; | |
664 | ||
665 | sysfs_free(sub); | |
666 | level = 4; | |
667 | } | |
668 | ||
669 | sub = NULL; | |
670 | switch (level) { | |
671 | default: | |
672 | rv = -1; | |
673 | break; | |
674 | case 4: | |
675 | case 5: | |
676 | case 6: | |
677 | sub = sysfs_read(-1, e->devnum, GET_CHUNK|GET_CACHE); | |
678 | if (!sub) | |
679 | break; | |
680 | cache = (sub->array.chunk_size / 4096) * 4; | |
681 | if (cache > sub->cache_size) | |
682 | rv = subarray_set_num(container, sub, | |
683 | "stripe_cache_size", cache); | |
684 | if (rv) { | |
685 | err = errno; | |
686 | break; | |
687 | } | |
688 | /* fall through */ | |
689 | case 1: | |
690 | if (!sub) | |
691 | sub = sysfs_read(-1, e->devnum, GET_VERSION); | |
692 | if (!sub) | |
693 | break; | |
694 | ||
695 | rv = subarray_set_num(container, sub, "raid_disks", | |
696 | raid_disks + takeover_delta); | |
697 | if (rv) | |
698 | err = errno; | |
699 | else | |
700 | changed++; | |
701 | break; | |
702 | } | |
703 | sysfs_free(sub); | |
704 | if (rv) | |
705 | break; | |
706 | } | |
707 | free_mdstat(ent); | |
708 | if (rv) { | |
709 | fprintf(stderr, Name | |
710 | ": failed to initiate container reshape%s%s\n", | |
711 | err ? ": " : "", err ? strerror(err) : ""); | |
712 | return rv; | |
713 | } | |
714 | ||
715 | return changed; | |
716 | } | |
717 | ||
718 | static void revert_container_raid_disks(struct supertype *st, int fd, char *container) | |
719 | { | |
720 | /* we failed to prepare all subarrays in the container for | |
721 | * reshape, so cancel the changes and restore the nominal raid | |
722 | * level | |
723 | */ | |
724 | struct mdstat_ent *ent, *e; | |
725 | ||
726 | ent = mdstat_read(0, 0); | |
727 | if (!ent) { | |
728 | fprintf(stderr, Name | |
729 | ": failed to read /proc/mdstat while aborting reshape\n"); | |
730 | return; | |
731 | } | |
732 | ||
7f2ba464 DW |
733 | if (st->ss->load_container(st, fd, NULL)) { |
734 | fprintf(stderr, Name | |
735 | ": failed read metadata while aborting reshape\n"); | |
736 | return ; | |
737 | } | |
738 | ||
739 | ||
7bc71196 DW |
740 | for (e = ent; e; e = e->next) { |
741 | int level_fixed = 0, disks_fixed = 0; | |
7f2ba464 DW |
742 | struct mdinfo *sub, *prev; |
743 | char *subarray; | |
7bc71196 DW |
744 | |
745 | if (!is_container_member(e, container)) | |
746 | continue; | |
747 | ||
7f2ba464 DW |
748 | subarray = to_subarray(e, container); |
749 | prev = st->ss->container_content(st, subarray); | |
7bc71196 DW |
750 | |
751 | /* changing level might change raid_disks so we do it | |
752 | * first and then check if raid_disks still needs fixing | |
753 | */ | |
7f2ba464 | 754 | if (map_name(pers, e->level) != prev->array.level) { |
7bc71196 DW |
755 | sub = sysfs_read(-1, e->devnum, GET_VERSION); |
756 | if (sub && | |
7f2ba464 | 757 | !sysfs_set_num(sub, NULL, "level", prev->array.level)) |
7bc71196 DW |
758 | level_fixed = 1; |
759 | sysfs_free(sub); | |
760 | } else | |
761 | level_fixed = 1; | |
762 | ||
763 | sub = sysfs_read(-1, e->devnum, GET_DISKS); | |
7f2ba464 | 764 | if (sub && sub->array.raid_disks != prev->array.raid_disks) { |
7bc71196 | 765 | if (!subarray_set_num(container, sub, "raid_disks", |
7f2ba464 | 766 | prev->array.raid_disks)) |
7bc71196 DW |
767 | disks_fixed = 1; |
768 | } else if (sub) | |
769 | disks_fixed = 1; | |
770 | sysfs_free(sub); | |
771 | ||
772 | if (!disks_fixed || !level_fixed) | |
773 | fprintf(stderr, Name | |
774 | ": failed to restore %s to a %d-disk %s array\n", | |
7f2ba464 DW |
775 | e->dev, prev->array.raid_disks, |
776 | map_num(pers, prev->array.level)); | |
777 | free(prev); | |
7bc71196 | 778 | } |
7f2ba464 | 779 | st->ss->free_super(st); |
7bc71196 DW |
780 | free_mdstat(ent); |
781 | } | |
782 | ||
62a48395 AK |
783 | int remove_disks_on_raid10_to_raid0_takeover(struct supertype *st, |
784 | struct mdinfo *sra, | |
785 | int layout) | |
786 | { | |
787 | int nr_of_copies; | |
788 | struct mdinfo *remaining; | |
789 | int slot; | |
790 | ||
791 | nr_of_copies = layout & 0xff; | |
792 | ||
793 | remaining = sra->devs; | |
794 | sra->devs = NULL; | |
795 | /* for each 'copy', select one device and remove from the list. */ | |
796 | for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) { | |
797 | struct mdinfo **diskp; | |
798 | int found = 0; | |
799 | ||
800 | /* Find a working device to keep */ | |
801 | for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) { | |
802 | struct mdinfo *disk = *diskp; | |
803 | ||
804 | if (disk->disk.raid_disk < slot) | |
805 | continue; | |
806 | if (disk->disk.raid_disk >= slot + nr_of_copies) | |
807 | continue; | |
808 | if (disk->disk.state & (1<<MD_DISK_REMOVED)) | |
809 | continue; | |
810 | if (disk->disk.state & (1<<MD_DISK_FAULTY)) | |
811 | continue; | |
812 | if (!(disk->disk.state & (1<<MD_DISK_SYNC))) | |
813 | continue; | |
814 | ||
815 | /* We have found a good disk to use! */ | |
816 | *diskp = disk->next; | |
817 | disk->next = sra->devs; | |
818 | sra->devs = disk; | |
819 | found = 1; | |
820 | break; | |
821 | } | |
822 | if (!found) | |
823 | break; | |
824 | } | |
825 | ||
826 | if (slot < sra->array.raid_disks) { | |
827 | /* didn't find all slots */ | |
828 | struct mdinfo **e; | |
829 | e = &remaining; | |
830 | while (*e) | |
831 | e = &(*e)->next; | |
832 | *e = sra->devs; | |
833 | sra->devs = remaining; | |
834 | return 1; | |
835 | } | |
836 | ||
837 | /* Remove all 'remaining' devices from the array */ | |
838 | while (remaining) { | |
839 | struct mdinfo *sd = remaining; | |
840 | remaining = sd->next; | |
841 | ||
842 | sysfs_set_str(sra, sd, "state", "faulty"); | |
843 | sysfs_set_str(sra, sd, "slot", "none"); | |
844 | sysfs_set_str(sra, sd, "state", "remove"); | |
845 | sd->disk.state |= (1<<MD_DISK_REMOVED); | |
846 | sd->disk.state &= ~(1<<MD_DISK_SYNC); | |
847 | sd->next = sra->devs; | |
848 | sra->devs = sd; | |
849 | } | |
850 | return 0; | |
851 | } | |
852 | ||
130994cb AK |
853 | void reshape_free_fdlist(int *fdlist, |
854 | unsigned long long *offsets, | |
855 | int size) | |
856 | { | |
857 | int i; | |
858 | ||
859 | for (i = 0; i < size; i++) | |
860 | if (fdlist[i] >= 0) | |
861 | close(fdlist[i]); | |
862 | ||
863 | free(fdlist); | |
864 | free(offsets); | |
865 | } | |
866 | ||
867 | int reshape_prepare_fdlist(char *devname, | |
868 | struct mdinfo *sra, | |
869 | int raid_disks, | |
870 | int nrdisks, | |
871 | unsigned long blocks, | |
872 | char *backup_file, | |
873 | int *fdlist, | |
874 | unsigned long long *offsets) | |
875 | { | |
876 | int d = 0; | |
877 | struct mdinfo *sd; | |
878 | ||
879 | for (d = 0; d <= nrdisks; d++) | |
880 | fdlist[d] = -1; | |
881 | d = raid_disks; | |
882 | for (sd = sra->devs; sd; sd = sd->next) { | |
883 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
884 | continue; | |
885 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
886 | char *dn = map_dev(sd->disk.major, | |
887 | sd->disk.minor, 1); | |
888 | fdlist[sd->disk.raid_disk] | |
889 | = dev_open(dn, O_RDONLY); | |
890 | offsets[sd->disk.raid_disk] = sd->data_offset*512; | |
891 | if (fdlist[sd->disk.raid_disk] < 0) { | |
892 | fprintf(stderr, | |
893 | Name ": %s: cannot open component %s\n", | |
894 | devname, dn ? dn : "-unknown-"); | |
895 | d = -1; | |
896 | goto release; | |
897 | } | |
898 | } else if (backup_file == NULL) { | |
899 | /* spare */ | |
900 | char *dn = map_dev(sd->disk.major, | |
901 | sd->disk.minor, 1); | |
902 | fdlist[d] = dev_open(dn, O_RDWR); | |
903 | offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512; | |
904 | if (fdlist[d] < 0) { | |
905 | fprintf(stderr, Name ": %s: cannot open component %s\n", | |
906 | devname, dn ? dn : "-unknown-"); | |
907 | d = -1; | |
908 | goto release; | |
909 | } | |
910 | d++; | |
911 | } | |
912 | } | |
913 | release: | |
914 | return d; | |
915 | } | |
916 | ||
1c009fc2 AK |
917 | unsigned long compute_backup_blocks(int nchunk, int ochunk, |
918 | unsigned int ndata, unsigned int odata) | |
919 | { | |
920 | unsigned long a, b, blocks; | |
921 | /* So how much do we need to backup. | |
922 | * We need an amount of data which is both a whole number of | |
923 | * old stripes and a whole number of new stripes. | |
924 | * So LCM for (chunksize*datadisks). | |
925 | */ | |
926 | a = (ochunk/512) * odata; | |
927 | b = (nchunk/512) * ndata; | |
928 | /* Find GCD */ | |
929 | while (a != b) { | |
930 | if (a < b) | |
931 | b -= a; | |
932 | if (b < a) | |
933 | a -= b; | |
934 | } | |
935 | /* LCM == product / GCD */ | |
936 | blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a; | |
937 | ||
938 | return blocks; | |
939 | } | |
940 | ||
941 | ||
06b0d786 | 942 | int Grow_reshape(char *devname, int fd, int quiet, char *backup_file, |
e86c9dd6 | 943 | long long size, |
19678e53 | 944 | int level, char *layout_str, int chunksize, int raid_disks) |
e86c9dd6 NB |
945 | { |
946 | /* Make some changes in the shape of an array. | |
947 | * The kernel must support the change. | |
7236ee7a N |
948 | * |
949 | * There are three different changes. Each can trigger | |
950 | * a resync or recovery so we freeze that until we have | |
951 | * requested everything (if kernel supports freezing - 2.6.30). | |
952 | * The steps are: | |
953 | * - change size (i.e. component_size) | |
954 | * - change level | |
955 | * - change layout/chunksize/ndisks | |
956 | * | |
957 | * The last can require a reshape. It is different on different | |
958 | * levels so we need to check the level before actioning it. | |
959 | * Some times the level change needs to be requested after the | |
960 | * reshape (e.g. raid6->raid5, raid5->raid0) | |
961 | * | |
e86c9dd6 | 962 | */ |
7236ee7a | 963 | struct mdu_array_info_s array, orig; |
e86c9dd6 | 964 | char *c; |
7236ee7a | 965 | int rv = 0; |
e86c9dd6 | 966 | struct supertype *st; |
4725bc31 | 967 | char *subarray = NULL; |
e86c9dd6 | 968 | |
e86c9dd6 NB |
969 | int nchunk, ochunk; |
970 | int nlayout, olayout; | |
971 | int ndisks, odisks; | |
f21e18ca | 972 | unsigned int ndata, odata; |
7236ee7a N |
973 | int orig_level = UnSet; |
974 | char alt_layout[40]; | |
e86c9dd6 NB |
975 | int *fdlist; |
976 | unsigned long long *offsets; | |
7236ee7a | 977 | int d, i; |
e86c9dd6 NB |
978 | int nrdisks; |
979 | int err; | |
7236ee7a | 980 | int frozen; |
1c009fc2 | 981 | unsigned long blocks, stripes; |
f21e18ca | 982 | unsigned long cache; |
7236ee7a N |
983 | unsigned long long array_size; |
984 | int changed = 0; | |
7bc71196 DW |
985 | char *container = NULL; |
986 | int cfd = -1; | |
7236ee7a | 987 | int done; |
e86c9dd6 | 988 | |
7e0f6979 | 989 | struct mdinfo *sra; |
06c7f68e | 990 | struct mdinfo *sd; |
e86c9dd6 NB |
991 | |
992 | if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) { | |
993 | fprintf(stderr, Name ": %s is not an active md array - aborting\n", | |
994 | devname); | |
995 | return 1; | |
996 | } | |
24d40069 | 997 | |
9ce510be N |
998 | if (size >= 0 && |
999 | (chunksize || level!= UnSet || layout_str || raid_disks)) { | |
1000 | fprintf(stderr, Name ": cannot change component size at the same time " | |
1001 | "as other changes.\n" | |
1002 | " Change size first, then check data is intact before " | |
1003 | "making other changes.\n"); | |
1004 | return 1; | |
1005 | } | |
1006 | ||
24d40069 N |
1007 | if (raid_disks && raid_disks < array.raid_disks && array.level > 1 && |
1008 | get_linux_version() < 2006032 && | |
1009 | !check_env("MDADM_FORCE_FEWER")) { | |
1010 | fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n" | |
1011 | " Please use a newer kernel\n"); | |
1012 | return 1; | |
1013 | } | |
7bc71196 DW |
1014 | |
1015 | st = super_by_fd(fd, &subarray); | |
1016 | if (!st) { | |
1017 | fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname); | |
1018 | return 1; | |
1019 | } | |
1020 | ||
1021 | /* in the external case we need to check that the requested reshape is | |
1022 | * supported, and perform an initial check that the container holds the | |
1023 | * pre-requisite spare devices (mdmon owns final validation) | |
1024 | */ | |
1025 | if (st->ss->external) { | |
1026 | int container_dev; | |
7f2ba464 | 1027 | int rv; |
7bc71196 DW |
1028 | |
1029 | if (subarray) { | |
1030 | container_dev = st->container_dev; | |
1031 | cfd = open_dev_excl(st->container_dev); | |
1032 | } else if (size >= 0 || layout_str != NULL || chunksize != 0 || | |
1033 | level != UnSet) { | |
1034 | fprintf(stderr, | |
1035 | Name ": %s is a container, only 'raid-devices' can be changed\n", | |
1036 | devname); | |
1037 | return 1; | |
1038 | } else { | |
1039 | container_dev = st->devnum; | |
1040 | close(fd); | |
1041 | cfd = open_dev_excl(st->devnum); | |
1042 | fd = cfd; | |
1043 | } | |
1044 | if (cfd < 0) { | |
1045 | fprintf(stderr, Name ": Unable to open container for %s\n", | |
1046 | devname); | |
7f2ba464 | 1047 | free(subarray); |
7bc71196 DW |
1048 | return 1; |
1049 | } | |
1050 | ||
1051 | container = devnum2devname(st->devnum); | |
1052 | if (!container) { | |
1053 | fprintf(stderr, Name ": Could not determine container name\n"); | |
7f2ba464 | 1054 | free(subarray); |
7bc71196 DW |
1055 | return 1; |
1056 | } | |
1057 | ||
7f2ba464 DW |
1058 | if (subarray) |
1059 | rv = st->ss->load_container(st, cfd, NULL); | |
1060 | else | |
1061 | rv = st->ss->load_super(st, cfd, NULL); | |
1062 | if (rv) { | |
7bc71196 DW |
1063 | fprintf(stderr, Name ": Cannot read superblock for %s\n", |
1064 | devname); | |
7f2ba464 | 1065 | free(subarray); |
7bc71196 DW |
1066 | return 1; |
1067 | } | |
1068 | ||
1069 | if (mdmon_running(container_dev)) | |
1070 | st->update_tail = &st->updates; | |
1071 | } | |
1072 | ||
62a48395 | 1073 | sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS | GET_STATE); |
7bc71196 | 1074 | if (sra) { |
7f2ba464 | 1075 | if (st->ss->external && subarray == NULL) { |
7bc71196 DW |
1076 | array.level = LEVEL_CONTAINER; |
1077 | sra->array.level = LEVEL_CONTAINER; | |
1078 | } | |
7bc71196 | 1079 | } else { |
b526e52d DW |
1080 | fprintf(stderr, Name ": failed to read sysfs parameters for %s\n", |
1081 | devname); | |
1082 | return 1; | |
1083 | } | |
7f2ba464 DW |
1084 | frozen = freeze(st); |
1085 | if (frozen < -1) { | |
1086 | /* freeze() already spewed the reason */ | |
1087 | return 1; | |
1088 | } else if (frozen < 0) { | |
7236ee7a N |
1089 | fprintf(stderr, Name ": %s is performing resync/recovery and cannot" |
1090 | " be reshaped\n", devname); | |
1091 | return 1; | |
1092 | } | |
1093 | ||
1094 | /* ========= set size =============== */ | |
1095 | if (size >= 0 && (size == 0 || size != array.size)) { | |
7bc71196 DW |
1096 | long long orig_size = array.size; |
1097 | ||
1098 | if (reshape_super(st, size, UnSet, UnSet, 0, 0, NULL, devname, !quiet)) { | |
1099 | rv = 1; | |
1100 | goto release; | |
1101 | } | |
1102 | sync_metadata(st); | |
7236ee7a N |
1103 | array.size = size; |
1104 | if (array.size != size) { | |
1105 | /* got truncated to 32bit, write to | |
1106 | * component_size instead | |
1107 | */ | |
1108 | if (sra) | |
1109 | rv = sysfs_set_num(sra, NULL, | |
1110 | "component_size", size); | |
1111 | else | |
1112 | rv = -1; | |
1113 | } else | |
1114 | rv = ioctl(fd, SET_ARRAY_INFO, &array); | |
1115 | if (rv != 0) { | |
39bbb392 | 1116 | int err = errno; |
7bc71196 DW |
1117 | |
1118 | /* restore metadata */ | |
1119 | if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0, | |
1120 | NULL, devname, !quiet) == 0) | |
1121 | sync_metadata(st); | |
7236ee7a | 1122 | fprintf(stderr, Name ": Cannot set device size for %s: %s\n", |
39bbb392 N |
1123 | devname, strerror(err)); |
1124 | if (err == EBUSY && | |
1125 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1126 | fprintf(stderr, " Bitmap must be removed before size can be changed\n"); | |
7236ee7a N |
1127 | rv = 1; |
1128 | goto release; | |
1129 | } | |
1130 | ioctl(fd, GET_ARRAY_INFO, &array); | |
be1cabbd | 1131 | size = get_component_size(fd)/2; |
f98841b3 N |
1132 | if (size == 0) |
1133 | size = array.size; | |
7236ee7a | 1134 | if (!quiet) |
f98841b3 N |
1135 | fprintf(stderr, Name ": component size of %s has been set to %lluK\n", |
1136 | devname, size); | |
7236ee7a | 1137 | changed = 1; |
7bc71196 | 1138 | } else if (array.level != LEVEL_CONTAINER) { |
be1cabbd | 1139 | size = get_component_size(fd)/2; |
f98841b3 N |
1140 | if (size == 0) |
1141 | size = array.size; | |
7236ee7a N |
1142 | } |
1143 | ||
62a48395 AK |
1144 | /* ========= check for Raid10 -> Raid0 conversion =============== |
1145 | * current implemenation assumes that following conditions must be met: | |
1146 | * - far_copies == 1 | |
1147 | * - near_copies == 2 | |
1148 | */ | |
1149 | if (level == 0 && array.level == 10 && | |
1150 | array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) { | |
1151 | int err; | |
1152 | err = remove_disks_on_raid10_to_raid0_takeover(st, sra, array.layout); | |
1153 | if (err) { | |
1154 | dprintf(Name": Array cannot be reshaped\n"); | |
1155 | if (container) | |
1156 | free(container); | |
1157 | if (cfd > -1) | |
1158 | close(cfd); | |
1159 | return 1; | |
1160 | } | |
1161 | } | |
1162 | ||
7236ee7a N |
1163 | /* ======= set level =========== */ |
1164 | if (level != UnSet && level != array.level) { | |
1165 | /* Trying to change the level. | |
1166 | * We might need to change layout first and schedule a | |
1167 | * level change for later. | |
1168 | * Level changes that can happen immediately are: | |
1169 | * 0->4,5,6 1->5 4->5,6 5->1,6 | |
1170 | * Level changes that need a layout change first are: | |
1171 | * 6->5,4,0 : need a -6 layout, or parity-last | |
1172 | * 5->4,0 : need parity-last | |
1173 | */ | |
1174 | if ((array.level == 6 || array.level == 5) && | |
1175 | (level == 5 || level == 4 || level == 0)) { | |
1176 | /* Don't change level yet, but choose intermediate | |
1177 | * layout | |
1178 | */ | |
1179 | if (level == 5) { | |
1180 | if (layout_str == NULL) | |
1181 | switch (array.layout) { | |
1182 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1183 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1184 | case ALGORITHM_ROTATING_N_RESTART: | |
1185 | layout_str = "left-asymmetric-6"; | |
1186 | break; | |
1187 | case ALGORITHM_LEFT_SYMMETRIC: | |
1188 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1189 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1190 | layout_str = "left-symmetric-6"; | |
1191 | break; | |
1192 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1193 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1194 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1195 | layout_str = "right-asymmetric-6"; | |
1196 | break; | |
1197 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1198 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1199 | layout_str = "right-symmetric-6"; | |
1200 | break; | |
1201 | case ALGORITHM_PARITY_0: | |
1202 | case ALGORITHM_PARITY_0_6: | |
1203 | layout_str = "parity-first-6"; | |
1204 | break; | |
1205 | case ALGORITHM_PARITY_N: | |
1206 | layout_str = "parity-last"; | |
1207 | break; | |
1208 | default: | |
1209 | fprintf(stderr, Name ": %s: cannot" | |
1210 | "convert layout to RAID5 equivalent\n", | |
1211 | devname); | |
1212 | rv = 1; | |
1213 | goto release; | |
1214 | } | |
1215 | else { | |
1216 | int l = map_name(r5layout, layout_str); | |
1217 | if (l == UnSet) { | |
1218 | fprintf(stderr, Name ": %s: layout '%s' not recognised\n", | |
1219 | devname, layout_str); | |
1220 | rv = 1; | |
1221 | goto release; | |
1222 | } | |
1223 | if (l != ALGORITHM_PARITY_N) { | |
1224 | /* need the -6 version */ | |
1225 | char *ls = map_num(r5layout, l); | |
1226 | strcat(strcpy(alt_layout, ls), | |
1227 | "-6"); | |
1228 | layout_str = alt_layout; | |
1229 | } | |
1230 | } | |
1231 | if (raid_disks) | |
24d40069 | 1232 | /* The final raid6->raid5 conversion |
7236ee7a N |
1233 | * will reduce the number of disks, |
1234 | * so now we need to aim higher | |
1235 | */ | |
1236 | raid_disks++; | |
1237 | } else | |
1238 | layout_str = "parity-last"; | |
1239 | } else { | |
7bc71196 DW |
1240 | /* Level change is a simple takeover. In the external |
1241 | * case we don't check with the metadata handler until | |
1242 | * we establish what the final layout will be. If the | |
1243 | * level change is disallowed we will revert to | |
1244 | * orig_level without disturbing the metadata, otherwise | |
1245 | * we will send an update. | |
1246 | */ | |
7236ee7a | 1247 | c = map_num(pers, level); |
b5ea446a N |
1248 | if (c == NULL) { |
1249 | rv = 1;/* not possible */ | |
1250 | goto release; | |
1251 | } | |
7236ee7a N |
1252 | err = sysfs_set_str(sra, NULL, "level", c); |
1253 | if (err) { | |
39bbb392 | 1254 | err = errno; |
7236ee7a N |
1255 | fprintf(stderr, Name ": %s: could not set level to %s\n", |
1256 | devname, c); | |
39bbb392 N |
1257 | if (err == EBUSY && |
1258 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1259 | fprintf(stderr, " Bitmap must be removed before level can be changed\n"); | |
7236ee7a N |
1260 | rv = 1; |
1261 | goto release; | |
1262 | } | |
1263 | orig = array; | |
1264 | orig_level = orig.level; | |
1265 | ioctl(fd, GET_ARRAY_INFO, &array); | |
1266 | if (layout_str == NULL && | |
1267 | orig.level == 5 && level == 6 && | |
1268 | array.layout != orig.layout) | |
1269 | layout_str = map_num(r5layout, orig.layout); | |
1270 | if (!quiet) | |
1271 | fprintf(stderr, Name " level of %s changed to %s\n", | |
1272 | devname, c); | |
1273 | changed = 1; | |
1274 | } | |
1275 | } | |
1276 | ||
1277 | /* ========= set shape (chunk_size / layout / ndisks) ============== */ | |
1278 | /* Check if layout change is a no-op */ | |
7bc71196 | 1279 | switch (array.level) { |
7236ee7a | 1280 | case 5: |
72e4a378 | 1281 | if (layout_str && array.layout == map_name(r5layout, layout_str)) |
7236ee7a N |
1282 | layout_str = NULL; |
1283 | break; | |
1284 | case 6: | |
1285 | if (layout_str == NULL && | |
1286 | ((chunksize && chunksize * 1024 != array.chunk_size) || | |
1287 | (raid_disks && raid_disks != array.raid_disks)) && | |
1288 | array.layout >= 16) { | |
1289 | fprintf(stderr, Name | |
1290 | ": %s has a non-standard layout. If you wish to preserve this\n" | |
1291 | " during the reshape, please specify --layout=preserve\n" | |
1292 | " If you want to change it, specify a layout or use --layout=normalise\n", | |
1293 | devname); | |
1294 | rv = 1; | |
1295 | goto release; | |
1296 | } | |
72e4a378 DW |
1297 | if (layout_str && |
1298 | (strcmp(layout_str, "normalise") == 0 || | |
1299 | strcmp(layout_str, "normalize") == 0)) { | |
7236ee7a N |
1300 | char *hyphen; |
1301 | strcpy(alt_layout, map_num(r6layout, array.layout)); | |
1302 | hyphen = strrchr(alt_layout, '-'); | |
1303 | if (hyphen && strcmp(hyphen, "-6") == 0) { | |
1304 | *hyphen = 0; | |
1305 | layout_str = alt_layout; | |
1306 | } | |
1307 | } | |
1308 | ||
72e4a378 | 1309 | if (layout_str && array.layout == map_name(r6layout, layout_str)) |
7236ee7a N |
1310 | layout_str = NULL; |
1311 | if (layout_str && strcmp(layout_str, "preserve") == 0) | |
1312 | layout_str = NULL; | |
1313 | break; | |
1314 | } | |
1315 | if (layout_str == NULL | |
1316 | && (chunksize == 0 || chunksize*1024 == array.chunk_size) | |
1317 | && (raid_disks == 0 || raid_disks == array.raid_disks)) { | |
7bc71196 DW |
1318 | if (reshape_super(st, -1, level, UnSet, 0, 0, NULL, devname, !quiet)) { |
1319 | rv = 1; | |
1320 | goto release; | |
1321 | } | |
1322 | sync_metadata(st); | |
7236ee7a N |
1323 | rv = 0; |
1324 | if (level != UnSet && level != array.level) { | |
1325 | /* Looks like this level change doesn't need | |
1326 | * a reshape after all. | |
1327 | */ | |
1328 | c = map_num(pers, level); | |
1329 | if (c) { | |
1330 | rv = sysfs_set_str(sra, NULL, "level", c); | |
39bbb392 N |
1331 | if (rv) { |
1332 | int err = errno; | |
7236ee7a N |
1333 | fprintf(stderr, Name ": %s: could not set level to %s\n", |
1334 | devname, c); | |
39bbb392 N |
1335 | if (err == EBUSY && |
1336 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1337 | fprintf(stderr, " Bitmap must be removed before level can be changed\n"); | |
b7e734fc | 1338 | rv = 1; |
39bbb392 | 1339 | } |
7236ee7a N |
1340 | } |
1341 | } else if (!changed && !quiet) | |
1342 | fprintf(stderr, Name ": %s: no change requested\n", | |
1343 | devname); | |
7bc71196 DW |
1344 | |
1345 | if (st->ss->external && !mdmon_running(st->container_dev) && | |
1346 | level > 0) { | |
1347 | start_mdmon(st->container_dev); | |
1348 | ping_monitor(container); | |
1349 | } | |
7236ee7a N |
1350 | goto release; |
1351 | } | |
1352 | ||
e86c9dd6 NB |
1353 | c = map_num(pers, array.level); |
1354 | if (c == NULL) c = "-unknown-"; | |
7bc71196 | 1355 | switch (array.level) { |
e86c9dd6 NB |
1356 | default: /* raid0, linear, multipath cannot be reconfigured */ |
1357 | fprintf(stderr, Name ": %s array %s cannot be reshaped.\n", | |
1358 | c, devname); | |
7bc71196 | 1359 | /* TODO raid0 raiddisks can be reshaped via raid4 */ |
7236ee7a N |
1360 | rv = 1; |
1361 | break; | |
7bc71196 DW |
1362 | case LEVEL_CONTAINER: { |
1363 | int count; | |
1364 | ||
1365 | /* double check that we are not changing anything but raid_disks */ | |
1366 | if (size >= 0 || layout_str != NULL || chunksize != 0 || level != UnSet) { | |
1367 | fprintf(stderr, | |
1368 | Name ": %s is a container, only 'raid-devices' can be changed\n", | |
1369 | devname); | |
1370 | rv = 1; | |
1371 | goto release; | |
1372 | } | |
1373 | ||
1374 | st->update_tail = &st->updates; | |
1375 | if (reshape_super(st, -1, UnSet, UnSet, 0, raid_disks, | |
1376 | backup_file, devname, !quiet)) { | |
1377 | rv = 1; | |
1378 | goto release; | |
1379 | } | |
1380 | ||
1381 | count = reshape_container_raid_disks(container, raid_disks); | |
1382 | if (count < 0) { | |
1383 | revert_container_raid_disks(st, fd, container); | |
1384 | rv = 1; | |
1385 | goto release; | |
1386 | } else if (count == 0) { | |
1387 | if (!quiet) | |
1388 | fprintf(stderr, Name | |
1389 | ": no active subarrays to reshape\n"); | |
1390 | goto release; | |
1391 | } | |
e86c9dd6 | 1392 | |
7bc71196 DW |
1393 | if (!mdmon_running(st->devnum)) { |
1394 | start_mdmon(st->devnum); | |
1395 | ping_monitor(container); | |
1396 | } | |
1397 | sync_metadata(st); | |
1398 | ||
1399 | /* give mdmon a chance to allocate spares */ | |
1400 | ping_manager(container); | |
1401 | ||
1402 | /* manage_reshape takes care of releasing the array(s) */ | |
1403 | st->ss->manage_reshape(st, backup_file); | |
1404 | frozen = 0; | |
1405 | goto release; | |
1406 | } | |
e86c9dd6 NB |
1407 | case LEVEL_FAULTY: /* only 'layout' change is permitted */ |
1408 | ||
e86c9dd6 NB |
1409 | if (chunksize || raid_disks) { |
1410 | fprintf(stderr, Name ": %s: Cannot change chunksize or disks of a 'faulty' array\n", | |
1411 | devname); | |
7236ee7a N |
1412 | rv = 1; |
1413 | break; | |
e86c9dd6 | 1414 | } |
19678e53 | 1415 | if (layout_str == NULL) |
7236ee7a | 1416 | break; /* nothing to do.... */ |
e86c9dd6 | 1417 | |
19678e53 N |
1418 | array.layout = parse_layout_faulty(layout_str); |
1419 | if (array.layout < 0) { | |
1420 | fprintf(stderr, Name ": %s: layout %s not understood for 'faulty' array\n", | |
1421 | devname, layout_str); | |
7236ee7a N |
1422 | rv = 1; |
1423 | break; | |
19678e53 | 1424 | } |
e86c9dd6 NB |
1425 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { |
1426 | fprintf(stderr, Name ": Cannot set layout for %s: %s\n", | |
1427 | devname, strerror(errno)); | |
7236ee7a N |
1428 | rv = 1; |
1429 | } else if (!quiet) | |
e86c9dd6 | 1430 | printf("layout for %s set to %d\n", devname, array.layout); |
7236ee7a | 1431 | break; |
e86c9dd6 | 1432 | |
7236ee7a | 1433 | case 1: /* only raid_disks can each be changed. */ |
e86c9dd6 | 1434 | |
19678e53 | 1435 | if (chunksize || layout_str != NULL) { |
7236ee7a | 1436 | fprintf(stderr, Name ": %s: Cannot change chunk size or layout for a RAID1 array.\n", |
e86c9dd6 | 1437 | devname); |
7236ee7a N |
1438 | rv = 1; |
1439 | break; | |
e86c9dd6 | 1440 | } |
9860f271 | 1441 | if (raid_disks > 0) { |
7bc71196 DW |
1442 | if (reshape_super(st, -1, UnSet, UnSet, 0, raid_disks, |
1443 | NULL, devname, !quiet)) { | |
1444 | rv = 1; | |
1445 | goto release; | |
1446 | } | |
1447 | sync_metadata(st); | |
e86c9dd6 | 1448 | array.raid_disks = raid_disks; |
9860f271 NB |
1449 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { |
1450 | fprintf(stderr, Name ": Cannot set raid-devices for %s: %s\n", | |
1451 | devname, strerror(errno)); | |
7236ee7a | 1452 | rv = 1; |
9860f271 | 1453 | } |
e86c9dd6 | 1454 | } |
7236ee7a | 1455 | break; |
e86c9dd6 NB |
1456 | |
1457 | case 4: | |
1458 | case 5: | |
1459 | case 6: | |
1686dc25 | 1460 | |
7236ee7a N |
1461 | /* |
1462 | * layout/chunksize/raid_disks can be changed | |
e86c9dd6 | 1463 | * though the kernel may not support it all. |
e86c9dd6 | 1464 | */ |
4725bc31 N |
1465 | if (subarray) { |
1466 | fprintf(stderr, Name ": Cannot reshape subarrays yet\n"); | |
1467 | break; | |
1468 | } | |
7236ee7a N |
1469 | |
1470 | /* | |
1471 | * There are three possibilities. | |
1472 | * 1/ The array will shrink. | |
1473 | * We need to ensure the reshape will pause before reaching | |
1474 | * the 'critical section'. We also need to fork and wait for | |
1475 | * that to happen. When it does we | |
1476 | * suspend/backup/complete/unfreeze | |
1477 | * | |
1478 | * 2/ The array will not change size. | |
1479 | * This requires that we keep a backup of a sliding window | |
1480 | * so that we can restore data after a crash. So we need | |
1481 | * to fork and monitor progress. | |
1482 | * | |
1483 | * 3/ The array will grow. This is relatively easy. | |
e86c9dd6 NB |
1484 | * However the kernel's restripe routines will cheerfully |
1485 | * overwrite some early data before it is safe. So we | |
1486 | * need to make a backup of the early parts of the array | |
1487 | * and be ready to restore it if rebuild aborts very early. | |
1488 | * | |
7236ee7a N |
1489 | * We backup data by writing it to one spare, or to a |
1490 | * file which was given on command line. | |
e86c9dd6 | 1491 | * |
7236ee7a | 1492 | * [FOLLOWING IS OLD AND PARTLY WRONG] |
e86c9dd6 NB |
1493 | * So: we enumerate the devices in the array and |
1494 | * make sure we can open all of them. | |
1495 | * Then we freeze the early part of the array and | |
1496 | * backup to the various spares. | |
1497 | * Then we request changes and start the reshape. | |
1498 | * Monitor progress until it has passed the danger zone. | |
1499 | * and finally invalidate the copied data and unfreeze the | |
1500 | * start of the array. | |
1501 | * | |
7236ee7a N |
1502 | * In each case, we first make sure that storage is available |
1503 | * for the required backup. | |
1504 | * Then we: | |
1505 | * - request the shape change. | |
1506 | * - for to handle backup etc. | |
e86c9dd6 | 1507 | */ |
e86c9dd6 NB |
1508 | nchunk = ochunk = array.chunk_size; |
1509 | nlayout = olayout = array.layout; | |
1510 | ndisks = odisks = array.raid_disks; | |
1511 | ||
7236ee7a N |
1512 | if (chunksize) { |
1513 | nchunk = chunksize * 1024; | |
f98841b3 N |
1514 | if (size % chunksize) { |
1515 | fprintf(stderr, Name ": component size %lluK is not" | |
7236ee7a | 1516 | " a multiple of chunksize %dK\n", |
f98841b3 | 1517 | size, chunksize); |
7236ee7a N |
1518 | break; |
1519 | } | |
1520 | } | |
19678e53 | 1521 | if (layout_str != NULL) |
7236ee7a | 1522 | switch(array.level) { |
19678e53 N |
1523 | case 4: /* ignore layout */ |
1524 | break; | |
1525 | case 5: | |
1526 | nlayout = map_name(r5layout, layout_str); | |
1527 | if (nlayout == UnSet) { | |
1528 | fprintf(stderr, Name ": layout %s not understood for raid5.\n", | |
1529 | layout_str); | |
b5ea446a N |
1530 | rv = 1; |
1531 | goto release; | |
19678e53 N |
1532 | } |
1533 | break; | |
1534 | ||
1535 | case 6: | |
1536 | nlayout = map_name(r6layout, layout_str); | |
1537 | if (nlayout == UnSet) { | |
1538 | fprintf(stderr, Name ": layout %s not understood for raid6.\n", | |
1539 | layout_str); | |
b5ea446a N |
1540 | rv = 1; |
1541 | goto release; | |
19678e53 N |
1542 | } |
1543 | break; | |
1544 | } | |
e86c9dd6 NB |
1545 | if (raid_disks) ndisks = raid_disks; |
1546 | ||
1547 | odata = odisks-1; | |
e86c9dd6 | 1548 | ndata = ndisks-1; |
7236ee7a N |
1549 | if (array.level == 6) { |
1550 | odata--; /* number of data disks */ | |
1551 | ndata--; | |
e86c9dd6 | 1552 | } |
7236ee7a | 1553 | |
d2505cff N |
1554 | if (odata == ndata && |
1555 | get_linux_version() < 2006032) { | |
1556 | fprintf(stderr, Name ": in-place reshape is not safe before 2.6.32, sorry.\n"); | |
1557 | break; | |
1558 | } | |
1559 | ||
7236ee7a | 1560 | /* Check that we can hold all the data */ |
7236ee7a | 1561 | get_dev_size(fd, NULL, &array_size); |
f21e18ca | 1562 | if (ndata * (unsigned long long)size < (array_size/1024)) { |
7236ee7a N |
1563 | fprintf(stderr, Name ": this change will reduce the size of the array.\n" |
1564 | " use --grow --array-size first to truncate array.\n" | |
1565 | " e.g. mdadm --grow %s --array-size %llu\n", | |
f98841b3 | 1566 | devname, ndata * size); |
7236ee7a N |
1567 | rv = 1; |
1568 | break; | |
e86c9dd6 | 1569 | } |
7236ee7a | 1570 | |
1c009fc2 | 1571 | blocks = compute_backup_blocks(nchunk, ochunk, ndata, odata); |
7236ee7a | 1572 | |
e380d3be N |
1573 | sysfs_free(sra); |
1574 | sra = sysfs_read(fd, 0, | |
1575 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE| | |
1576 | GET_CACHE); | |
1577 | ||
c03ef02d N |
1578 | if (!sra) { |
1579 | fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n", | |
1580 | devname); | |
1581 | rv = 1; | |
1582 | break; | |
1583 | } | |
1584 | ||
eba71529 N |
1585 | if (ndata == odata) { |
1586 | /* Make 'blocks' bigger for better throughput, but | |
1587 | * not so big that we reject it below. | |
1b13faf7 | 1588 | * Try for 16 megabytes |
eba71529 | 1589 | */ |
1b13faf7 N |
1590 | while (blocks * 32 < sra->component_size && |
1591 | blocks < 16*1024*2) | |
1592 | blocks *= 2; | |
eba71529 | 1593 | } else |
7236ee7a N |
1594 | fprintf(stderr, Name ": Need to backup %luK of critical " |
1595 | "section..\n", blocks/2); | |
e86c9dd6 | 1596 | |
7236ee7a | 1597 | if (blocks >= sra->component_size/2) { |
e86c9dd6 NB |
1598 | fprintf(stderr, Name ": %s: Something wrong - reshape aborted\n", |
1599 | devname); | |
7236ee7a N |
1600 | rv = 1; |
1601 | break; | |
353632d9 | 1602 | } |
925211e3 | 1603 | nrdisks = array.raid_disks + sra->array.spare_disks; |
e86c9dd6 NB |
1604 | /* Now we need to open all these devices so we can read/write. |
1605 | */ | |
06b0d786 NB |
1606 | fdlist = malloc((1+nrdisks) * sizeof(int)); |
1607 | offsets = malloc((1+nrdisks) * sizeof(offsets[0])); | |
e86c9dd6 NB |
1608 | if (!fdlist || !offsets) { |
1609 | fprintf(stderr, Name ": malloc failed: grow aborted\n"); | |
7236ee7a N |
1610 | rv = 1; |
1611 | break; | |
e86c9dd6 | 1612 | } |
130994cb AK |
1613 | |
1614 | d = reshape_prepare_fdlist(devname, sra, array.raid_disks, | |
1615 | nrdisks, blocks, backup_file, | |
1616 | fdlist, offsets); | |
1617 | if (d < 0) { | |
1618 | rv = 1; | |
1619 | goto release; | |
e86c9dd6 | 1620 | } |
7236ee7a | 1621 | if (backup_file == NULL) { |
7bc71196 DW |
1622 | if (st->ss->external && !st->ss->manage_reshape) { |
1623 | fprintf(stderr, Name ": %s Grow operation not supported by %s metadata\n", | |
1624 | devname, st->ss->name); | |
1625 | rv = 1; | |
1626 | break; | |
1627 | } | |
7236ee7a N |
1628 | if (ndata <= odata) { |
1629 | fprintf(stderr, Name ": %s: Cannot grow - need backup-file\n", | |
1630 | devname); | |
1631 | rv = 1; | |
1632 | break; | |
1633 | } else if (sra->array.spare_disks == 0) { | |
1634 | fprintf(stderr, Name ": %s: Cannot grow - need a spare or " | |
1635 | "backup-file to backup critical section\n", | |
1636 | devname); | |
1637 | rv = 1; | |
1638 | break; | |
1639 | } | |
1640 | if (d == array.raid_disks) { | |
1641 | fprintf(stderr, Name ": %s: No spare device for backup\n", | |
1642 | devname); | |
1643 | rv = 1; | |
1644 | break; | |
e86c9dd6 | 1645 | } |
7236ee7a N |
1646 | } else { |
1647 | /* need to check backup file is large enough */ | |
1648 | char buf[512]; | |
c82afc17 N |
1649 | struct stat stb; |
1650 | unsigned int dev; | |
7236ee7a N |
1651 | fdlist[d] = open(backup_file, O_RDWR|O_CREAT|O_EXCL, |
1652 | S_IRUSR | S_IWUSR); | |
1653 | offsets[d] = 8 * 512; | |
06b0d786 NB |
1654 | if (fdlist[d] < 0) { |
1655 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
1656 | devname, backup_file, strerror(errno)); | |
7236ee7a N |
1657 | rv = 1; |
1658 | break; | |
1659 | } | |
c82afc17 N |
1660 | /* Guard against backup file being on array device. |
1661 | * If array is partitioned or if LVM etc is in the | |
1662 | * way this will not notice, but it is better than | |
1663 | * nothing. | |
1664 | */ | |
1665 | fstat(fdlist[d], &stb); | |
1666 | dev = stb.st_dev; | |
1667 | fstat(fd, &stb); | |
1668 | if (stb.st_rdev == dev) { | |
1669 | fprintf(stderr, Name ": backup file must NOT be" | |
1670 | " on the array being reshaped.\n"); | |
1671 | rv = 1; | |
1672 | close(fdlist[d]); | |
1673 | break; | |
1674 | } | |
1675 | ||
7236ee7a | 1676 | memset(buf, 0, 512); |
f21e18ca | 1677 | for (i=0; i < (signed)blocks + 1 ; i++) { |
7236ee7a N |
1678 | if (write(fdlist[d], buf, 512) != 512) { |
1679 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
1680 | devname, backup_file, strerror(errno)); | |
1681 | rv = 1; | |
1682 | break; | |
1683 | } | |
1684 | } | |
1685 | if (fsync(fdlist[d]) != 0) { | |
1686 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
1687 | devname, backup_file, strerror(errno)); | |
1688 | rv = 1; | |
1689 | break; | |
06b0d786 | 1690 | } |
06b0d786 | 1691 | d++; |
06b0d786 | 1692 | } |
7236ee7a | 1693 | |
7bc71196 DW |
1694 | /* check that the operation is supported by the metadata */ |
1695 | if (reshape_super(st, -1, level, nlayout, nchunk, ndisks, | |
1696 | backup_file, devname, !quiet)) { | |
1697 | rv = 1; | |
1698 | break; | |
1699 | } | |
1700 | ||
7236ee7a N |
1701 | /* lastly, check that the internal stripe cache is |
1702 | * large enough, or it won't work. | |
1703 | */ | |
1704 | ||
1705 | cache = (nchunk < ochunk) ? ochunk : nchunk; | |
1706 | cache = cache * 4 / 4096; | |
1b13faf7 N |
1707 | if (cache < blocks / 8 / odisks + 16) |
1708 | /* Make it big enough to hold 'blocks' */ | |
1709 | cache = blocks / 8 / odisks + 16; | |
7236ee7a N |
1710 | if (sra->cache_size < cache) |
1711 | sysfs_set_num(sra, NULL, "stripe_cache_size", | |
1712 | cache+1); | |
1713 | /* Right, everything seems fine. Let's kick things off. | |
1714 | * If only changing raid_disks, use ioctl, else use | |
1715 | * sysfs. | |
1716 | */ | |
7bc71196 | 1717 | sync_metadata(st); |
7236ee7a N |
1718 | if (ochunk == nchunk && olayout == nlayout) { |
1719 | array.raid_disks = ndisks; | |
1720 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
39bbb392 | 1721 | int err = errno; |
7236ee7a N |
1722 | rv = 1; |
1723 | fprintf(stderr, Name ": Cannot set device shape for %s: %s\n", | |
1724 | devname, strerror(errno)); | |
1725 | if (ndisks < odisks && | |
1726 | get_linux_version() < 2006030) | |
1727 | fprintf(stderr, Name ": linux 2.6.30 or later required\n"); | |
39bbb392 N |
1728 | if (err == EBUSY && |
1729 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1730 | fprintf(stderr, " Bitmap must be removed before shape can be changed\n"); | |
7236ee7a N |
1731 | |
1732 | break; | |
1733 | } | |
1734 | } else { | |
1735 | /* set them all just in case some old 'new_*' value | |
1736 | * persists from some earlier problem | |
1737 | */ | |
53f50353 N |
1738 | int err = err; /* only used if rv==1, and always set if |
1739 | * rv==1, so initialisation not needed, | |
1740 | * despite gcc warning | |
1741 | */ | |
7236ee7a | 1742 | if (sysfs_set_num(sra, NULL, "chunk_size", nchunk) < 0) |
39bbb392 N |
1743 | rv = 1, err = errno; |
1744 | if (!rv && sysfs_set_num(sra, NULL, "layout", nlayout) < 0) | |
1745 | rv = 1, err = errno; | |
1746 | if (!rv && sysfs_set_num(sra, NULL, "raid_disks", ndisks) < 0) | |
1747 | rv = 1, err = errno; | |
7236ee7a N |
1748 | if (rv) { |
1749 | fprintf(stderr, Name ": Cannot set device shape for %s\n", | |
1750 | devname); | |
1751 | if (get_linux_version() < 2006030) | |
1752 | fprintf(stderr, Name ": linux 2.6.30 or later required\n"); | |
39bbb392 N |
1753 | if (err == EBUSY && |
1754 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1755 | fprintf(stderr, " Bitmap must be removed before shape can be changed\n"); | |
7236ee7a N |
1756 | break; |
1757 | } | |
e86c9dd6 NB |
1758 | } |
1759 | ||
7236ee7a N |
1760 | if (ndisks == 2 && odisks == 2) { |
1761 | /* No reshape is needed in this trivial case */ | |
1762 | rv = 0; | |
1763 | break; | |
1764 | } | |
1765 | ||
7bc71196 DW |
1766 | if (st->ss->external) { |
1767 | /* metadata handler takes it from here */ | |
1768 | ping_manager(container); | |
1769 | st->ss->manage_reshape(st, backup_file); | |
1770 | frozen = 0; | |
1771 | break; | |
1772 | } | |
1773 | ||
7236ee7a N |
1774 | /* set up the backup-super-block. This requires the |
1775 | * uuid from the array. | |
1776 | */ | |
e86c9dd6 | 1777 | /* Find a superblock */ |
7236ee7a N |
1778 | for (sd = sra->devs; sd; sd = sd->next) { |
1779 | char *dn; | |
1780 | int devfd; | |
1781 | int ok; | |
1782 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
1783 | continue; | |
1784 | dn = map_dev(sd->disk.major, sd->disk.minor, 1); | |
1785 | devfd = dev_open(dn, O_RDONLY); | |
1786 | if (devfd < 0) | |
1787 | continue; | |
1788 | ok = st->ss->load_super(st, devfd, NULL); | |
1789 | close(devfd); | |
1790 | if (ok >= 0) | |
1791 | break; | |
1792 | } | |
1793 | if (!sd) { | |
e86c9dd6 NB |
1794 | fprintf(stderr, Name ": %s: Cannot find a superblock\n", |
1795 | devname); | |
7236ee7a N |
1796 | rv = 1; |
1797 | break; | |
e86c9dd6 NB |
1798 | } |
1799 | ||
7236ee7a | 1800 | memset(&bsb, 0, 512); |
2efedc7b | 1801 | memcpy(bsb.magic, "md_backup_data-1", 16); |
3da92f27 | 1802 | st->ss->uuid_from_super(st, (int*)&bsb.set_uuid); |
2efedc7b | 1803 | bsb.mtime = __cpu_to_le64(time(0)); |
7236ee7a N |
1804 | bsb.devstart2 = blocks; |
1805 | stripes = blocks / (ochunk/512) / odata; | |
1806 | /* Now we just need to kick off the reshape and watch, while | |
1807 | * handling backups of the data... | |
1808 | * This is all done by a forked background process. | |
2efedc7b | 1809 | */ |
7236ee7a N |
1810 | switch(fork()) { |
1811 | case 0: | |
1812 | close(fd); | |
1813 | if (check_env("MDADM_GROW_VERIFY")) | |
1814 | fd = open(devname, O_RDONLY | O_DIRECT); | |
1815 | else | |
1816 | fd = -1; | |
1817 | mlockall(MCL_FUTURE); | |
1818 | ||
1819 | if (odata < ndata) | |
1820 | done = child_grow(fd, sra, stripes, | |
1821 | fdlist, offsets, | |
1822 | odisks, ochunk, array.level, olayout, odata, | |
1823 | d - odisks, fdlist+odisks, offsets+odisks); | |
1824 | else if (odata > ndata) | |
1825 | done = child_shrink(fd, sra, stripes, | |
1826 | fdlist, offsets, | |
1827 | odisks, ochunk, array.level, olayout, odata, | |
1828 | d - odisks, fdlist+odisks, offsets+odisks); | |
1829 | else | |
1830 | done = child_same_size(fd, sra, stripes, | |
1831 | fdlist, offsets, | |
e9e43ec3 | 1832 | 0, |
7236ee7a N |
1833 | odisks, ochunk, array.level, olayout, odata, |
1834 | d - odisks, fdlist+odisks, offsets+odisks); | |
1835 | if (backup_file && done) | |
1836 | unlink(backup_file); | |
1837 | if (level != UnSet && level != array.level) { | |
1838 | /* We need to wait for the reshape to finish | |
1839 | * (which will have happened unless odata < ndata) | |
1840 | * and then set the level | |
758d3a8e | 1841 | */ |
7236ee7a N |
1842 | |
1843 | c = map_num(pers, level); | |
1844 | if (c == NULL) | |
1845 | exit(0);/* not possible */ | |
1846 | ||
1847 | if (odata < ndata) | |
1848 | wait_reshape(sra); | |
1849 | err = sysfs_set_str(sra, NULL, "level", c); | |
1850 | if (err) | |
1851 | fprintf(stderr, Name ": %s: could not set level to %s\n", | |
1852 | devname, c); | |
758d3a8e | 1853 | } |
7236ee7a N |
1854 | exit(0); |
1855 | case -1: | |
1856 | fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n", | |
1857 | strerror(errno)); | |
1858 | rv = 1; | |
1859 | break; | |
1860 | default: | |
1861 | /* The child will take care of unfreezing the array */ | |
1862 | frozen = 0; | |
1863 | break; | |
e86c9dd6 | 1864 | } |
7236ee7a | 1865 | break; |
e86c9dd6 | 1866 | |
7236ee7a | 1867 | } |
e86c9dd6 | 1868 | |
7236ee7a N |
1869 | release: |
1870 | if (rv && orig_level != UnSet && sra) { | |
1871 | c = map_num(pers, orig_level); | |
1872 | if (c && sysfs_set_str(sra, NULL, "level", c) == 0) | |
1873 | fprintf(stderr, Name ": aborting level change\n"); | |
1874 | } | |
7f2ba464 | 1875 | unfreeze(st, frozen); |
7236ee7a N |
1876 | return rv; |
1877 | } | |
e86c9dd6 | 1878 | |
7236ee7a N |
1879 | /* |
1880 | * We run a child process in the background which performs the following | |
1881 | * steps: | |
1882 | * - wait for resync to reach a certain point | |
1883 | * - suspend io to the following section | |
1884 | * - backup that section | |
1885 | * - allow resync to proceed further | |
1886 | * - resume io | |
1887 | * - discard the backup. | |
1888 | * | |
1889 | * When are combined in slightly different ways in the three cases. | |
1890 | * Grow: | |
1891 | * - suspend/backup/allow/wait/resume/discard | |
1892 | * Shrink: | |
1893 | * - allow/wait/suspend/backup/allow/wait/resume/discard | |
1894 | * same-size: | |
1895 | * - wait/resume/discard/suspend/backup/allow | |
1896 | * | |
1897 | * suspend/backup/allow always come together | |
1898 | * wait/resume/discard do too. | |
1899 | * For the same-size case we have two backups to improve flow. | |
1900 | * | |
1901 | */ | |
e86c9dd6 | 1902 | |
fcf57625 | 1903 | /* FIXME return status is never checked */ |
4411fb17 | 1904 | static int grow_backup(struct mdinfo *sra, |
7236ee7a N |
1905 | unsigned long long offset, /* per device */ |
1906 | unsigned long stripes, /* per device */ | |
1907 | int *sources, unsigned long long *offsets, | |
1908 | int disks, int chunk, int level, int layout, | |
1909 | int dests, int *destfd, unsigned long long *destoffsets, | |
d4445387 | 1910 | int part, int *degraded, |
7236ee7a N |
1911 | char *buf) |
1912 | { | |
1913 | /* Backup 'blocks' sectors at 'offset' on each device of the array, | |
1914 | * to storage 'destfd' (offset 'destoffsets'), after first | |
1915 | * suspending IO. Then allow resync to continue | |
1916 | * over the suspended section. | |
1917 | * Use part 'part' of the backup-super-block. | |
1918 | */ | |
1919 | int odata = disks; | |
1920 | int rv = 0; | |
1921 | int i; | |
f21e18ca N |
1922 | unsigned long long ll; |
1923 | int new_degraded; | |
7236ee7a N |
1924 | //printf("offset %llu\n", offset); |
1925 | if (level >= 4) | |
1926 | odata--; | |
1927 | if (level == 6) | |
1928 | odata--; | |
200871ad | 1929 | sysfs_set_num(sra, NULL, "suspend_hi", (offset + stripes * (chunk/512)) * odata); |
d4445387 | 1930 | /* Check that array hasn't become degraded, else we might backup the wrong data */ |
f21e18ca N |
1931 | sysfs_get_ll(sra, NULL, "degraded", &ll); |
1932 | new_degraded = (int)ll; | |
d4445387 N |
1933 | if (new_degraded != *degraded) { |
1934 | /* check each device to ensure it is still working */ | |
1935 | struct mdinfo *sd; | |
1936 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
1937 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
1938 | continue; | |
1939 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
1940 | char sbuf[20]; | |
1941 | if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 || | |
1942 | strstr(sbuf, "faulty") || | |
1943 | strstr(sbuf, "in_sync") == NULL) { | |
1944 | /* this device is dead */ | |
1945 | sd->disk.state = (1<<MD_DISK_FAULTY); | |
1946 | if (sd->disk.raid_disk >= 0 && | |
1947 | sources[sd->disk.raid_disk] >= 0) { | |
1948 | close(sources[sd->disk.raid_disk]); | |
1949 | sources[sd->disk.raid_disk] = -1; | |
1950 | } | |
1951 | } | |
1952 | } | |
1953 | } | |
1954 | *degraded = new_degraded; | |
1955 | } | |
7236ee7a N |
1956 | if (part) { |
1957 | bsb.arraystart2 = __cpu_to_le64(offset * odata); | |
200871ad | 1958 | bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
1959 | } else { |
1960 | bsb.arraystart = __cpu_to_le64(offset * odata); | |
200871ad | 1961 | bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
1962 | } |
1963 | if (part) | |
1964 | bsb.magic[15] = '2'; | |
1965 | for (i = 0; i < dests; i++) | |
1966 | if (part) | |
1967 | lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0); | |
1968 | else | |
1969 | lseek64(destfd[i], destoffsets[i], 0); | |
1970 | ||
1971 | rv = save_stripes(sources, offsets, | |
1972 | disks, chunk, level, layout, | |
1973 | dests, destfd, | |
1974 | offset*512*odata, stripes * chunk * odata, | |
1975 | buf); | |
1976 | ||
1977 | if (rv) | |
1978 | return rv; | |
97396422 | 1979 | bsb.mtime = __cpu_to_le64(time(0)); |
7236ee7a N |
1980 | for (i = 0; i < dests; i++) { |
1981 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
1982 | ||
1983 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
1984 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
1985 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
1986 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
1987 | ||
72044953 | 1988 | rv = -1; |
f21e18ca N |
1989 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0) |
1990 | != destoffsets[i] - 4096) | |
72044953 N |
1991 | break; |
1992 | if (write(destfd[i], &bsb, 512) != 512) | |
1993 | break; | |
ff94fb86 | 1994 | if (destoffsets[i] > 4096) { |
f21e18ca | 1995 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) != |
a847575a | 1996 | destoffsets[i]+stripes*chunk*odata) |
72044953 N |
1997 | break; |
1998 | if (write(destfd[i], &bsb, 512) != 512) | |
1999 | break; | |
ff94fb86 | 2000 | } |
7236ee7a | 2001 | fsync(destfd[i]); |
72044953 | 2002 | rv = 0; |
7236ee7a | 2003 | } |
2efedc7b | 2004 | |
fcf57625 | 2005 | return rv; |
7236ee7a N |
2006 | } |
2007 | ||
2008 | /* in 2.6.30, the value reported by sync_completed can be | |
2009 | * less that it should be by one stripe. | |
2010 | * This only happens when reshape hits sync_max and pauses. | |
2011 | * So allow wait_backup to either extent sync_max further | |
2012 | * than strictly necessary, or return before the | |
2013 | * sync has got quite as far as we would really like. | |
2014 | * This is what 'blocks2' is for. | |
2015 | * The various caller give appropriate values so that | |
2016 | * every works. | |
2017 | */ | |
fcf57625 | 2018 | /* FIXME return value is often ignored */ |
4411fb17 | 2019 | static int wait_backup(struct mdinfo *sra, |
7236ee7a N |
2020 | unsigned long long offset, /* per device */ |
2021 | unsigned long long blocks, /* per device */ | |
2022 | unsigned long long blocks2, /* per device - hack */ | |
2023 | int dests, int *destfd, unsigned long long *destoffsets, | |
2024 | int part) | |
2025 | { | |
2026 | /* Wait for resync to pass the section that was backed up | |
2027 | * then erase the backup and allow IO | |
2028 | */ | |
2029 | int fd = sysfs_get_fd(sra, NULL, "sync_completed"); | |
2030 | unsigned long long completed; | |
2031 | int i; | |
fcf57625 | 2032 | int rv; |
7236ee7a N |
2033 | |
2034 | if (fd < 0) | |
2035 | return -1; | |
2036 | sysfs_set_num(sra, NULL, "sync_max", offset + blocks + blocks2); | |
2037 | if (offset == 0) | |
2038 | sysfs_set_str(sra, NULL, "sync_action", "reshape"); | |
2039 | do { | |
2040 | char action[20]; | |
2041 | fd_set rfds; | |
2042 | FD_ZERO(&rfds); | |
2043 | FD_SET(fd, &rfds); | |
2044 | select(fd+1, NULL, NULL, &rfds, NULL); | |
2045 | if (sysfs_fd_get_ll(fd, &completed) < 0) { | |
2046 | close(fd); | |
2047 | return -1; | |
e86c9dd6 | 2048 | } |
7236ee7a N |
2049 | if (sysfs_get_str(sra, NULL, "sync_action", |
2050 | action, 20) > 0 && | |
2051 | strncmp(action, "reshape", 7) != 0) | |
2052 | break; | |
2053 | } while (completed < offset + blocks); | |
2054 | close(fd); | |
2055 | ||
2056 | if (part) { | |
2057 | bsb.arraystart2 = __cpu_to_le64(0); | |
2058 | bsb.length2 = __cpu_to_le64(0); | |
2059 | } else { | |
2060 | bsb.arraystart = __cpu_to_le64(0); | |
2061 | bsb.length = __cpu_to_le64(0); | |
2062 | } | |
97396422 | 2063 | bsb.mtime = __cpu_to_le64(time(0)); |
fcf57625 | 2064 | rv = 0; |
7236ee7a N |
2065 | for (i = 0; i < dests; i++) { |
2066 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
2067 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
2068 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
2069 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
2070 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
f21e18ca | 2071 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) != |
a847575a | 2072 | destoffsets[i]-4096) |
72044953 N |
2073 | rv = -1; |
2074 | if (rv == 0 && | |
2075 | write(destfd[i], &bsb, 512) != 512) | |
2076 | rv = -1; | |
7236ee7a N |
2077 | fsync(destfd[i]); |
2078 | } | |
fcf57625 | 2079 | return rv; |
7236ee7a | 2080 | } |
e86c9dd6 | 2081 | |
7236ee7a N |
2082 | static void fail(char *msg) |
2083 | { | |
fcf57625 | 2084 | int rv; |
72044953 N |
2085 | rv = (write(2, msg, strlen(msg)) != (int)strlen(msg)); |
2086 | rv |= (write(2, "\n", 1) != 1); | |
fcf57625 | 2087 | exit(rv ? 1 : 2); |
7236ee7a N |
2088 | } |
2089 | ||
2090 | static char *abuf, *bbuf; | |
f21e18ca | 2091 | static unsigned long long abuflen; |
7236ee7a N |
2092 | static void validate(int afd, int bfd, unsigned long long offset) |
2093 | { | |
2094 | /* check that the data in the backup against the array. | |
2095 | * This is only used for regression testing and should not | |
2096 | * be used while the array is active | |
2097 | */ | |
7236ee7a N |
2098 | if (afd < 0) |
2099 | return; | |
2100 | lseek64(bfd, offset - 4096, 0); | |
2101 | if (read(bfd, &bsb2, 512) != 512) | |
2102 | fail("cannot read bsb"); | |
2103 | if (bsb2.sb_csum != bsb_csum((char*)&bsb2, | |
2104 | ((char*)&bsb2.sb_csum)-((char*)&bsb2))) | |
2105 | fail("first csum bad"); | |
2106 | if (memcmp(bsb2.magic, "md_backup_data", 14) != 0) | |
2107 | fail("magic is bad"); | |
2108 | if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 && | |
2109 | bsb2.sb_csum2 != bsb_csum((char*)&bsb2, | |
2110 | ((char*)&bsb2.sb_csum2)-((char*)&bsb2))) | |
2111 | fail("second csum bad"); | |
2112 | ||
2113 | if (__le64_to_cpu(bsb2.devstart)*512 != offset) | |
2114 | fail("devstart is wrong"); | |
2115 | ||
2116 | if (bsb2.length) { | |
2117 | unsigned long long len = __le64_to_cpu(bsb2.length)*512; | |
2118 | ||
2119 | if (abuflen < len) { | |
2120 | free(abuf); | |
2121 | free(bbuf); | |
2122 | abuflen = len; | |
fcf57625 N |
2123 | if (posix_memalign((void**)&abuf, 4096, abuflen) || |
2124 | posix_memalign((void**)&bbuf, 4096, abuflen)) { | |
2125 | abuflen = 0; | |
2126 | /* just stop validating on mem-alloc failure */ | |
2127 | return; | |
2128 | } | |
e86c9dd6 | 2129 | } |
48924014 | 2130 | |
7236ee7a | 2131 | lseek64(bfd, offset, 0); |
f21e18ca | 2132 | if ((unsigned long long)read(bfd, bbuf, len) != len) { |
080fd005 | 2133 | //printf("len %llu\n", len); |
7236ee7a N |
2134 | fail("read first backup failed"); |
2135 | } | |
2136 | lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0); | |
f21e18ca | 2137 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
2138 | fail("read first from array failed"); |
2139 | if (memcmp(bbuf, abuf, len) != 0) { | |
080fd005 | 2140 | #if 0 |
7236ee7a N |
2141 | int i; |
2142 | printf("offset=%llu len=%llu\n", | |
080fd005 | 2143 | (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len); |
7236ee7a N |
2144 | for (i=0; i<len; i++) |
2145 | if (bbuf[i] != abuf[i]) { | |
2146 | printf("first diff byte %d\n", i); | |
93ecfa01 | 2147 | break; |
7236ee7a | 2148 | } |
080fd005 | 2149 | #endif |
7236ee7a | 2150 | fail("data1 compare failed"); |
e86c9dd6 | 2151 | } |
7236ee7a N |
2152 | } |
2153 | if (bsb2.length2) { | |
2154 | unsigned long long len = __le64_to_cpu(bsb2.length2)*512; | |
2155 | ||
2156 | if (abuflen < len) { | |
2157 | free(abuf); | |
2158 | free(bbuf); | |
2159 | abuflen = len; | |
2160 | abuf = malloc(abuflen); | |
2161 | bbuf = malloc(abuflen); | |
e86c9dd6 NB |
2162 | } |
2163 | ||
7236ee7a | 2164 | lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0); |
f21e18ca | 2165 | if ((unsigned long long)read(bfd, bbuf, len) != len) |
7236ee7a N |
2166 | fail("read second backup failed"); |
2167 | lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0); | |
f21e18ca | 2168 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
2169 | fail("read second from array failed"); |
2170 | if (memcmp(bbuf, abuf, len) != 0) | |
2171 | fail("data2 compare failed"); | |
e86c9dd6 | 2172 | } |
7236ee7a | 2173 | } |
e86c9dd6 | 2174 | |
7236ee7a N |
2175 | static int child_grow(int afd, struct mdinfo *sra, unsigned long stripes, |
2176 | int *fds, unsigned long long *offsets, | |
2177 | int disks, int chunk, int level, int layout, int data, | |
2178 | int dests, int *destfd, unsigned long long *destoffsets) | |
2179 | { | |
2180 | char *buf; | |
d4445387 | 2181 | int degraded = 0; |
e86c9dd6 | 2182 | |
fcf57625 N |
2183 | if (posix_memalign((void**)&buf, 4096, disks * chunk)) |
2184 | /* Don't start the 'reshape' */ | |
2185 | return 0; | |
7236ee7a N |
2186 | sysfs_set_num(sra, NULL, "suspend_hi", 0); |
2187 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
2188 | grow_backup(sra, 0, stripes, | |
2189 | fds, offsets, disks, chunk, level, layout, | |
2190 | dests, destfd, destoffsets, | |
d4445387 | 2191 | 0, °raded, buf); |
7236ee7a | 2192 | validate(afd, destfd[0], destoffsets[0]); |
200871ad | 2193 | wait_backup(sra, 0, stripes * (chunk / 512), stripes * (chunk / 512), |
725cac4c N |
2194 | dests, destfd, destoffsets, |
2195 | 0); | |
200871ad | 2196 | sysfs_set_num(sra, NULL, "suspend_lo", (stripes * (chunk/512)) * data); |
7236ee7a N |
2197 | free(buf); |
2198 | /* FIXME this should probably be numeric */ | |
2199 | sysfs_set_str(sra, NULL, "sync_max", "max"); | |
e86c9dd6 | 2200 | return 1; |
7236ee7a | 2201 | } |
e86c9dd6 | 2202 | |
7236ee7a N |
2203 | static int child_shrink(int afd, struct mdinfo *sra, unsigned long stripes, |
2204 | int *fds, unsigned long long *offsets, | |
2205 | int disks, int chunk, int level, int layout, int data, | |
2206 | int dests, int *destfd, unsigned long long *destoffsets) | |
2207 | { | |
2208 | char *buf; | |
2209 | unsigned long long start; | |
2210 | int rv; | |
d4445387 | 2211 | int degraded = 0; |
7236ee7a | 2212 | |
fcf57625 N |
2213 | if (posix_memalign((void**)&buf, 4096, disks * chunk)) |
2214 | return 0; | |
200871ad | 2215 | start = sra->component_size - stripes * (chunk/512); |
7236ee7a N |
2216 | sysfs_set_num(sra, NULL, "sync_max", start); |
2217 | sysfs_set_str(sra, NULL, "sync_action", "reshape"); | |
2218 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
2219 | sysfs_set_num(sra, NULL, "suspend_hi", 0); | |
200871ad | 2220 | rv = wait_backup(sra, 0, start - stripes * (chunk/512), stripes * (chunk/512), |
7236ee7a N |
2221 | dests, destfd, destoffsets, 0); |
2222 | if (rv < 0) | |
2223 | return 0; | |
2224 | grow_backup(sra, 0, stripes, | |
2225 | fds, offsets, | |
2226 | disks, chunk, level, layout, | |
2227 | dests, destfd, destoffsets, | |
d4445387 | 2228 | 0, °raded, buf); |
7236ee7a | 2229 | validate(afd, destfd[0], destoffsets[0]); |
200871ad | 2230 | wait_backup(sra, start, stripes*(chunk/512), 0, |
725cac4c | 2231 | dests, destfd, destoffsets, 0); |
200871ad | 2232 | sysfs_set_num(sra, NULL, "suspend_lo", (stripes * (chunk/512)) * data); |
7236ee7a N |
2233 | free(buf); |
2234 | /* FIXME this should probably be numeric */ | |
2235 | sysfs_set_str(sra, NULL, "sync_max", "max"); | |
2236 | return 1; | |
2237 | } | |
2238 | ||
2239 | static int child_same_size(int afd, struct mdinfo *sra, unsigned long stripes, | |
2240 | int *fds, unsigned long long *offsets, | |
e9e43ec3 | 2241 | unsigned long long start, |
7236ee7a N |
2242 | int disks, int chunk, int level, int layout, int data, |
2243 | int dests, int *destfd, unsigned long long *destoffsets) | |
2244 | { | |
e9e43ec3 | 2245 | unsigned long long size; |
7236ee7a N |
2246 | unsigned long tailstripes = stripes; |
2247 | int part; | |
2248 | char *buf; | |
2249 | unsigned long long speed; | |
d4445387 | 2250 | int degraded = 0; |
7236ee7a N |
2251 | |
2252 | ||
fcf57625 N |
2253 | if (posix_memalign((void**)&buf, 4096, disks * chunk)) |
2254 | return 0; | |
7236ee7a N |
2255 | |
2256 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
2257 | sysfs_set_num(sra, NULL, "suspend_hi", 0); | |
2258 | ||
2259 | sysfs_get_ll(sra, NULL, "sync_speed_min", &speed); | |
2260 | sysfs_set_num(sra, NULL, "sync_speed_min", 200000); | |
2261 | ||
e9e43ec3 | 2262 | grow_backup(sra, start, stripes, |
7236ee7a N |
2263 | fds, offsets, |
2264 | disks, chunk, level, layout, | |
2265 | dests, destfd, destoffsets, | |
d4445387 | 2266 | 0, °raded, buf); |
200871ad | 2267 | grow_backup(sra, (start + stripes) * (chunk/512), stripes, |
7236ee7a N |
2268 | fds, offsets, |
2269 | disks, chunk, level, layout, | |
2270 | dests, destfd, destoffsets, | |
d4445387 | 2271 | 1, °raded, buf); |
7236ee7a N |
2272 | validate(afd, destfd[0], destoffsets[0]); |
2273 | part = 0; | |
e9e43ec3 | 2274 | start += stripes * 2; /* where to read next */ |
7236ee7a N |
2275 | size = sra->component_size / (chunk/512); |
2276 | while (start < size) { | |
200871ad N |
2277 | if (wait_backup(sra, (start-stripes*2)*(chunk/512), |
2278 | stripes*(chunk/512), 0, | |
7236ee7a N |
2279 | dests, destfd, destoffsets, |
2280 | part) < 0) | |
2281 | return 0; | |
200871ad | 2282 | sysfs_set_num(sra, NULL, "suspend_lo", start*(chunk/512) * data); |
7236ee7a N |
2283 | if (start + stripes > size) |
2284 | tailstripes = (size - start); | |
2285 | ||
200871ad | 2286 | grow_backup(sra, start*(chunk/512), tailstripes, |
7236ee7a N |
2287 | fds, offsets, |
2288 | disks, chunk, level, layout, | |
2289 | dests, destfd, destoffsets, | |
d4445387 | 2290 | part, °raded, buf); |
7236ee7a N |
2291 | start += stripes; |
2292 | part = 1 - part; | |
2293 | validate(afd, destfd[0], destoffsets[0]); | |
2294 | } | |
200871ad | 2295 | if (wait_backup(sra, (start-stripes*2) * (chunk/512), stripes * (chunk/512), 0, |
7236ee7a N |
2296 | dests, destfd, destoffsets, |
2297 | part) < 0) | |
2298 | return 0; | |
200871ad N |
2299 | sysfs_set_num(sra, NULL, "suspend_lo", ((start-stripes)*(chunk/512)) * data); |
2300 | wait_backup(sra, (start-stripes) * (chunk/512), tailstripes * (chunk/512), 0, | |
725cac4c N |
2301 | dests, destfd, destoffsets, |
2302 | 1-part); | |
200871ad | 2303 | sysfs_set_num(sra, NULL, "suspend_lo", (size*(chunk/512)) * data); |
7236ee7a N |
2304 | sysfs_set_num(sra, NULL, "sync_speed_min", speed); |
2305 | free(buf); | |
2306 | return 1; | |
e86c9dd6 | 2307 | } |
353632d9 NB |
2308 | |
2309 | /* | |
2310 | * If any spare contains md_back_data-1 which is recent wrt mtime, | |
2311 | * write that data into the array and update the super blocks with | |
2312 | * the new reshape_progress | |
2313 | */ | |
ea0ebe96 N |
2314 | int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt, |
2315 | char *backup_file, int verbose) | |
353632d9 NB |
2316 | { |
2317 | int i, j; | |
2318 | int old_disks; | |
353632d9 | 2319 | unsigned long long *offsets; |
82f2d6ab | 2320 | unsigned long long nstripe, ostripe; |
6e9eac4f | 2321 | int ndata, odata; |
353632d9 | 2322 | |
e9e43ec3 N |
2323 | if (info->new_level != info->array.level) |
2324 | return 1; /* Cannot handle level changes (they are instantaneous) */ | |
2325 | ||
2326 | odata = info->array.raid_disks - info->delta_disks - 1; | |
2327 | if (info->array.level == 6) odata--; /* number of data disks */ | |
2328 | ndata = info->array.raid_disks - 1; | |
2329 | if (info->new_level == 6) ndata--; | |
353632d9 NB |
2330 | |
2331 | old_disks = info->array.raid_disks - info->delta_disks; | |
2332 | ||
e9e43ec3 N |
2333 | if (info->delta_disks <= 0) |
2334 | /* Didn't grow, so the backup file must have | |
2335 | * been used | |
2336 | */ | |
2337 | old_disks = cnt; | |
06b0d786 | 2338 | for (i=old_disks-(backup_file?1:0); i<cnt; i++) { |
353632d9 | 2339 | struct mdinfo dinfo; |
06b0d786 | 2340 | int fd; |
e9e43ec3 | 2341 | int bsbsize; |
ea0ebe96 | 2342 | char *devname, namebuf[20]; |
353632d9 NB |
2343 | |
2344 | /* This was a spare and may have some saved data on it. | |
2345 | * Load the superblock, find and load the | |
2346 | * backup_super_block. | |
2347 | * If either fail, go on to next device. | |
2348 | * If the backup contains no new info, just return | |
206c5eae | 2349 | * else restore data and update all superblocks |
353632d9 | 2350 | */ |
06b0d786 NB |
2351 | if (i == old_disks-1) { |
2352 | fd = open(backup_file, O_RDONLY); | |
e9e43ec3 N |
2353 | if (fd<0) { |
2354 | fprintf(stderr, Name ": backup file %s inaccessible: %s\n", | |
2355 | backup_file, strerror(errno)); | |
06b0d786 | 2356 | continue; |
e9e43ec3 | 2357 | } |
ea0ebe96 | 2358 | devname = backup_file; |
06b0d786 NB |
2359 | } else { |
2360 | fd = fdlist[i]; | |
2361 | if (fd < 0) | |
2362 | continue; | |
3da92f27 | 2363 | if (st->ss->load_super(st, fd, NULL)) |
06b0d786 | 2364 | continue; |
353632d9 | 2365 | |
a5d85af7 | 2366 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 NB |
2367 | st->ss->free_super(st); |
2368 | ||
06b0d786 NB |
2369 | if (lseek64(fd, |
2370 | (dinfo.data_offset + dinfo.component_size - 8) <<9, | |
ea0ebe96 N |
2371 | 0) < 0) { |
2372 | fprintf(stderr, Name ": Cannot seek on device %d\n", i); | |
06b0d786 | 2373 | continue; /* Cannot seek */ |
ea0ebe96 N |
2374 | } |
2375 | sprintf(namebuf, "device-%d", i); | |
2376 | devname = namebuf; | |
06b0d786 | 2377 | } |
ea0ebe96 N |
2378 | if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) { |
2379 | if (verbose) | |
2380 | fprintf(stderr, Name ": Cannot read from %s\n", devname); | |
353632d9 | 2381 | continue; /* Cannot read */ |
ea0ebe96 | 2382 | } |
e9e43ec3 | 2383 | if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 && |
ea0ebe96 N |
2384 | memcmp(bsb.magic, "md_backup_data-2", 16) != 0) { |
2385 | if (verbose) | |
2386 | fprintf(stderr, Name ": No backup metadata on %s\n", devname); | |
353632d9 | 2387 | continue; |
ea0ebe96 N |
2388 | } |
2389 | if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) { | |
2390 | if (verbose) | |
2391 | fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname); | |
353632d9 | 2392 | continue; /* bad checksum */ |
ea0ebe96 | 2393 | } |
e9e43ec3 | 2394 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 && |
ea0ebe96 N |
2395 | bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) { |
2396 | if (verbose) | |
2397 | fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname); | |
22e30516 | 2398 | continue; /* Bad second checksum */ |
ea0ebe96 N |
2399 | } |
2400 | if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) { | |
2401 | if (verbose) | |
2402 | fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname); | |
353632d9 | 2403 | continue; /* Wrong uuid */ |
ea0ebe96 | 2404 | } |
353632d9 | 2405 | |
097075b6 N |
2406 | /* array utime and backup-mtime should be updated at much the same time, but it seems that |
2407 | * sometimes they aren't... So allow considerable flexability in matching, and allow | |
2408 | * this test to be overridden by an environment variable. | |
2409 | */ | |
f21e18ca N |
2410 | if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 || |
2411 | info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) { | |
097075b6 N |
2412 | if (check_env("MDADM_GROW_ALLOW_OLD")) { |
2413 | fprintf(stderr, Name ": accepting backup with timestamp %lu " | |
2414 | "for array with timestamp %lu\n", | |
2415 | (unsigned long)__le64_to_cpu(bsb.mtime), | |
2416 | (unsigned long)info->array.utime); | |
2417 | } else { | |
2418 | if (verbose) | |
2419 | fprintf(stderr, Name ": too-old timestamp on " | |
2420 | "backup-metadata on %s\n", devname); | |
2421 | continue; /* time stamp is too bad */ | |
2422 | } | |
ea0ebe96 | 2423 | } |
353632d9 | 2424 | |
e9e43ec3 N |
2425 | if (bsb.magic[15] == '1') { |
2426 | if (info->delta_disks >= 0) { | |
2427 | /* reshape_progress is increasing */ | |
2428 | if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) < | |
ea0ebe96 N |
2429 | info->reshape_progress) { |
2430 | nonew: | |
2431 | if (verbose) | |
2432 | fprintf(stderr, Name ": backup-metadata found on %s but is not needed\n", devname); | |
e9e43ec3 | 2433 | continue; /* No new data here */ |
ea0ebe96 | 2434 | } |
e9e43ec3 N |
2435 | } else { |
2436 | /* reshape_progress is decreasing */ | |
2437 | if (__le64_to_cpu(bsb.arraystart) >= | |
2438 | info->reshape_progress) | |
ea0ebe96 | 2439 | goto nonew; /* No new data here */ |
e9e43ec3 N |
2440 | } |
2441 | } else { | |
2442 | if (info->delta_disks >= 0) { | |
2443 | /* reshape_progress is increasing */ | |
2444 | if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) < | |
2445 | info->reshape_progress && | |
2446 | __le64_to_cpu(bsb.arraystart2) + __le64_to_cpu(bsb.length2) < | |
2447 | info->reshape_progress) | |
ea0ebe96 | 2448 | goto nonew; /* No new data here */ |
e9e43ec3 N |
2449 | } else { |
2450 | /* reshape_progress is decreasing */ | |
2451 | if (__le64_to_cpu(bsb.arraystart) >= | |
2452 | info->reshape_progress && | |
2453 | __le64_to_cpu(bsb.arraystart2) >= | |
2454 | info->reshape_progress) | |
ea0ebe96 | 2455 | goto nonew; /* No new data here */ |
e9e43ec3 N |
2456 | } |
2457 | } | |
ea0ebe96 N |
2458 | if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) { |
2459 | second_fail: | |
2460 | if (verbose) | |
2461 | fprintf(stderr, Name ": Failed to verify secondary backup-metadata block on %s\n", | |
2462 | devname); | |
353632d9 | 2463 | continue; /* Cannot seek */ |
ea0ebe96 | 2464 | } |
2efedc7b | 2465 | /* There should be a duplicate backup superblock 4k before here */ |
06b0d786 | 2466 | if (lseek64(fd, -4096, 1) < 0 || |
0155af90 | 2467 | read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2)) |
ea0ebe96 | 2468 | goto second_fail; /* Cannot find leading superblock */ |
e9e43ec3 N |
2469 | if (bsb.magic[15] == '1') |
2470 | bsbsize = offsetof(struct mdp_backup_super, pad1); | |
2471 | else | |
2472 | bsbsize = offsetof(struct mdp_backup_super, pad); | |
ff94fb86 | 2473 | if (memcmp(&bsb2, &bsb, bsbsize) != 0) |
ea0ebe96 | 2474 | goto second_fail; /* Cannot find leading superblock */ |
2efedc7b | 2475 | |
353632d9 NB |
2476 | /* Now need the data offsets for all devices. */ |
2477 | offsets = malloc(sizeof(*offsets)*info->array.raid_disks); | |
2478 | for(j=0; j<info->array.raid_disks; j++) { | |
2479 | if (fdlist[j] < 0) | |
2480 | continue; | |
3da92f27 | 2481 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 NB |
2482 | /* FIXME should be this be an error */ |
2483 | continue; | |
a5d85af7 | 2484 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 | 2485 | st->ss->free_super(st); |
14e5b4d7 | 2486 | offsets[j] = dinfo.data_offset * 512; |
353632d9 NB |
2487 | } |
2488 | printf(Name ": restoring critical section\n"); | |
2489 | ||
2490 | if (restore_stripes(fdlist, offsets, | |
2491 | info->array.raid_disks, | |
2492 | info->new_chunk, | |
2493 | info->new_level, | |
2494 | info->new_layout, | |
06b0d786 | 2495 | fd, __le64_to_cpu(bsb.devstart)*512, |
92dcdf7c | 2496 | __le64_to_cpu(bsb.arraystart)*512, |
e9e43ec3 N |
2497 | __le64_to_cpu(bsb.length)*512)) { |
2498 | /* didn't succeed, so giveup */ | |
ea0ebe96 N |
2499 | if (verbose) |
2500 | fprintf(stderr, Name ": Error restoring backup from %s\n", | |
2501 | devname); | |
e9e43ec3 N |
2502 | return 1; |
2503 | } | |
2504 | ||
2505 | if (bsb.magic[15] == '2' && | |
2506 | restore_stripes(fdlist, offsets, | |
2507 | info->array.raid_disks, | |
2508 | info->new_chunk, | |
2509 | info->new_level, | |
2510 | info->new_layout, | |
2511 | fd, __le64_to_cpu(bsb.devstart)*512 + | |
2512 | __le64_to_cpu(bsb.devstart2)*512, | |
92dcdf7c | 2513 | __le64_to_cpu(bsb.arraystart2)*512, |
e9e43ec3 | 2514 | __le64_to_cpu(bsb.length2)*512)) { |
353632d9 | 2515 | /* didn't succeed, so giveup */ |
ea0ebe96 N |
2516 | if (verbose) |
2517 | fprintf(stderr, Name ": Error restoring second backup from %s\n", | |
2518 | devname); | |
2295250a | 2519 | return 1; |
353632d9 NB |
2520 | } |
2521 | ||
e9e43ec3 | 2522 | |
353632d9 NB |
2523 | /* Ok, so the data is restored. Let's update those superblocks. */ |
2524 | ||
e9e43ec3 N |
2525 | if (info->delta_disks >= 0) { |
2526 | info->reshape_progress = __le64_to_cpu(bsb.arraystart) + | |
2527 | __le64_to_cpu(bsb.length); | |
2528 | if (bsb.magic[15] == '2') { | |
2529 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) + | |
2530 | __le64_to_cpu(bsb.length2); | |
2531 | if (p2 > info->reshape_progress) | |
2532 | info->reshape_progress = p2; | |
2533 | } | |
2534 | } else { | |
2535 | info->reshape_progress = __le64_to_cpu(bsb.arraystart); | |
2536 | if (bsb.magic[15] == '2') { | |
2537 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2); | |
2538 | if (p2 < info->reshape_progress) | |
2539 | info->reshape_progress = p2; | |
2540 | } | |
2541 | } | |
353632d9 NB |
2542 | for (j=0; j<info->array.raid_disks; j++) { |
2543 | if (fdlist[j] < 0) continue; | |
3da92f27 | 2544 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 | 2545 | continue; |
a5d85af7 | 2546 | st->ss->getinfo_super(st, &dinfo, NULL); |
e9e43ec3 | 2547 | dinfo.reshape_progress = info->reshape_progress; |
3da92f27 | 2548 | st->ss->update_super(st, &dinfo, |
68c7d6d7 NB |
2549 | "_reshape_progress", |
2550 | NULL,0, 0, NULL); | |
3da92f27 NB |
2551 | st->ss->store_super(st, fdlist[j]); |
2552 | st->ss->free_super(st); | |
353632d9 | 2553 | } |
353632d9 NB |
2554 | return 0; |
2555 | } | |
6e9eac4f NB |
2556 | /* Didn't find any backup data, try to see if any |
2557 | * was needed. | |
2558 | */ | |
82f2d6ab N |
2559 | if (info->delta_disks < 0) { |
2560 | /* When shrinking, the critical section is at the end. | |
2561 | * So see if we are before the critical section. | |
2562 | */ | |
2563 | unsigned long long first_block; | |
2564 | nstripe = ostripe = 0; | |
2565 | first_block = 0; | |
2566 | while (ostripe >= nstripe) { | |
2567 | ostripe += info->array.chunk_size / 512; | |
2568 | first_block = ostripe * odata; | |
2569 | nstripe = first_block / ndata / (info->new_chunk/512) * | |
2570 | (info->new_chunk/512); | |
2571 | } | |
2572 | ||
2573 | if (info->reshape_progress >= first_block) | |
2574 | return 0; | |
6e9eac4f | 2575 | } |
82f2d6ab N |
2576 | if (info->delta_disks > 0) { |
2577 | /* See if we are beyond the critical section. */ | |
2578 | unsigned long long last_block; | |
2579 | nstripe = ostripe = 0; | |
2580 | last_block = 0; | |
2581 | while (nstripe >= ostripe) { | |
2582 | nstripe += info->new_chunk / 512; | |
2583 | last_block = nstripe * ndata; | |
2584 | ostripe = last_block / odata / (info->array.chunk_size/512) * | |
2585 | (info->array.chunk_size/512); | |
2586 | } | |
6e9eac4f | 2587 | |
82f2d6ab N |
2588 | if (info->reshape_progress >= last_block) |
2589 | return 0; | |
2590 | } | |
6e9eac4f | 2591 | /* needed to recover critical section! */ |
ea0ebe96 N |
2592 | if (verbose) |
2593 | fprintf(stderr, Name ": Failed to find backup of critical section\n"); | |
2295250a | 2594 | return 1; |
353632d9 | 2595 | } |
e9e43ec3 N |
2596 | |
2597 | int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info, | |
2598 | char *backup_file) | |
2599 | { | |
2600 | /* Array is assembled and ready to be started, but | |
2601 | * monitoring is probably required. | |
2602 | * So: | |
2603 | * - start read-only | |
2604 | * - set upper bound for resync | |
2605 | * - initialise the 'suspend' boundaries | |
2606 | * - switch to read-write | |
2607 | * - fork and continue monitoring | |
2608 | */ | |
2609 | int err; | |
2610 | int backup_list[1]; | |
2611 | unsigned long long backup_offsets[1]; | |
2612 | int odisks, ndisks, ochunk, nchunk,odata,ndata; | |
2613 | unsigned long a,b,blocks,stripes; | |
2614 | int backup_fd; | |
2615 | int *fds; | |
2616 | unsigned long long *offsets; | |
2617 | int d; | |
2618 | struct mdinfo *sra, *sd; | |
2619 | int rv; | |
f21e18ca | 2620 | unsigned long cache; |
e9e43ec3 N |
2621 | int done = 0; |
2622 | ||
2623 | err = sysfs_set_str(info, NULL, "array_state", "readonly"); | |
2624 | if (err) | |
2625 | return err; | |
2626 | ||
2627 | /* make sure reshape doesn't progress until we are ready */ | |
2628 | sysfs_set_str(info, NULL, "sync_max", "0"); | |
2629 | sysfs_set_str(info, NULL, "array_state", "active"); /* FIXME or clean */ | |
5f6ca90a N |
2630 | |
2631 | sra = sysfs_read(-1, devname2devnum(info->sys_name), | |
2632 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE| | |
2633 | GET_CACHE); | |
2634 | if (!sra) | |
2635 | return 1; | |
2636 | ||
e9e43ec3 N |
2637 | /* ndisks is not growing, so raid_disks is old and +delta is new */ |
2638 | odisks = info->array.raid_disks; | |
2639 | ndisks = odisks + info->delta_disks; | |
2640 | odata = odisks - 1; | |
2641 | ndata = ndisks - 1; | |
2642 | if (info->array.level == 6) { | |
2643 | odata--; | |
2644 | ndata--; | |
2645 | } | |
2646 | ochunk = info->array.chunk_size; | |
2647 | nchunk = info->new_chunk; | |
2648 | ||
200871ad N |
2649 | a = (ochunk/512) * odata; |
2650 | b = (nchunk/512) * ndata; | |
e9e43ec3 N |
2651 | /* Find GCD */ |
2652 | while (a != b) { | |
2653 | if (a < b) | |
2654 | b -= a; | |
2655 | if (b < a) | |
2656 | a -= b; | |
2657 | } | |
2658 | /* LCM == product / GCD */ | |
200871ad | 2659 | blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a; |
e9e43ec3 N |
2660 | |
2661 | if (ndata == odata) | |
1b13faf7 N |
2662 | while (blocks * 32 < sra->component_size && |
2663 | blocks < 16*1024*2) | |
2664 | blocks *= 2; | |
e9e43ec3 N |
2665 | stripes = blocks / (info->array.chunk_size/512) / odata; |
2666 | ||
1b13faf7 N |
2667 | /* check that the internal stripe cache is |
2668 | * large enough, or it won't work. | |
2669 | */ | |
2670 | cache = (nchunk < ochunk) ? ochunk : nchunk; | |
2671 | cache = cache * 4 / 4096; | |
2672 | if (cache < blocks / 8 / odisks + 16) | |
2673 | /* Make it big enough to hold 'blocks' */ | |
2674 | cache = blocks / 8 / odisks + 16; | |
2675 | if (sra->cache_size < cache) | |
2676 | sysfs_set_num(sra, NULL, "stripe_cache_size", | |
2677 | cache+1); | |
e9e43ec3 N |
2678 | |
2679 | memset(&bsb, 0, 512); | |
2680 | memcpy(bsb.magic, "md_backup_data-1", 16); | |
2681 | memcpy(&bsb.set_uuid, info->uuid, 16); | |
2682 | bsb.mtime = __cpu_to_le64(time(0)); | |
2683 | bsb.devstart2 = blocks; | |
2684 | ||
2685 | backup_fd = open(backup_file, O_RDWR|O_CREAT, S_IRUSR | S_IWUSR); | |
87f26d14 N |
2686 | if (backup_fd < 0) { |
2687 | fprintf(stderr, Name ": Cannot open backup file %s\n", | |
2688 | backup_file ?: "- no backup-file given"); | |
2689 | return 1; | |
2690 | } | |
e9e43ec3 N |
2691 | backup_list[0] = backup_fd; |
2692 | backup_offsets[0] = 8 * 512; | |
2693 | fds = malloc(odisks * sizeof(fds[0])); | |
2694 | offsets = malloc(odisks * sizeof(offsets[0])); | |
2695 | for (d=0; d<odisks; d++) | |
2696 | fds[d] = -1; | |
2697 | ||
e9e43ec3 N |
2698 | for (sd = sra->devs; sd; sd = sd->next) { |
2699 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2700 | continue; | |
2701 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
2702 | char *dn = map_dev(sd->disk.major, | |
2703 | sd->disk.minor, 1); | |
2704 | fds[sd->disk.raid_disk] | |
2705 | = dev_open(dn, O_RDONLY); | |
2706 | offsets[sd->disk.raid_disk] = sd->data_offset*512; | |
2707 | if (fds[sd->disk.raid_disk] < 0) { | |
2708 | fprintf(stderr, Name ": %s: cannot open component %s\n", | |
2709 | info->sys_name, dn?dn:"-unknown-"); | |
2710 | rv = 1; | |
2711 | goto release; | |
2712 | } | |
2713 | free(dn); | |
2714 | } | |
2715 | } | |
2716 | ||
2717 | switch(fork()) { | |
2718 | case 0: | |
2719 | close(mdfd); | |
2720 | mlockall(MCL_FUTURE); | |
2721 | if (info->delta_disks < 0) | |
2722 | done = child_shrink(-1, info, stripes, | |
2723 | fds, offsets, | |
2724 | info->array.raid_disks, | |
2725 | info->array.chunk_size, | |
2726 | info->array.level, info->array.layout, | |
2727 | odata, | |
2728 | 1, backup_list, backup_offsets); | |
2729 | else if (info->delta_disks == 0) { | |
2730 | /* The 'start' is a per-device stripe number. | |
2731 | * reshape_progress is a per-array sector number. | |
2732 | * So divide by ndata * chunk_size | |
2733 | */ | |
2734 | unsigned long long start = info->reshape_progress / ndata; | |
2735 | start /= (info->array.chunk_size/512); | |
2736 | done = child_same_size(-1, info, stripes, | |
2737 | fds, offsets, | |
2738 | start, | |
2739 | info->array.raid_disks, | |
2740 | info->array.chunk_size, | |
2741 | info->array.level, info->array.layout, | |
2742 | odata, | |
2743 | 1, backup_list, backup_offsets); | |
2744 | } | |
2745 | if (backup_file && done) | |
2746 | unlink(backup_file); | |
2747 | /* FIXME should I intuit a level change */ | |
2748 | exit(0); | |
2749 | case -1: | |
2750 | fprintf(stderr, Name ": Cannot run child to continue monitoring reshape: %s\n", | |
2751 | strerror(errno)); | |
2752 | return 1; | |
2753 | default: | |
2754 | break; | |
2755 | } | |
2756 | release: | |
2757 | return 0; | |
2758 | } | |
2759 | ||
2760 |