]>
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 | ||
d7ca196c N |
456 | static int check_idle(struct supertype *st) |
457 | { | |
458 | /* Check that all member arrays for this container, or the | |
459 | * container of this array, are idle | |
460 | */ | |
461 | int container_dev = (st->container_dev != NoMdDev | |
462 | ? st->container_dev : st->devnum); | |
463 | char container[40]; | |
464 | struct mdstat_ent *ent, *e; | |
465 | int is_idle = 1; | |
466 | ||
467 | fmt_devname(container, container_dev); | |
468 | ent = mdstat_read(0, 0); | |
469 | for (e = ent ; e; e = e->next) { | |
470 | if (!is_container_member(e, container)) | |
471 | continue; | |
472 | if (e->percent >= 0) { | |
473 | is_idle = 0; | |
474 | break; | |
475 | } | |
476 | } | |
477 | free_mdstat(ent); | |
478 | return is_idle; | |
479 | } | |
480 | ||
7f2ba464 | 481 | static int freeze_container(struct supertype *st) |
7236ee7a | 482 | { |
7f2ba464 DW |
483 | int container_dev = (st->container_dev != NoMdDev |
484 | ? st->container_dev : st->devnum); | |
d7ca196c | 485 | char container[40]; |
7f2ba464 | 486 | |
d7ca196c N |
487 | if (!check_idle(st)) |
488 | return -1; | |
489 | ||
490 | fmt_devname(container, container_dev); | |
7f2ba464 DW |
491 | |
492 | if (block_monitor(container, 1)) { | |
493 | fprintf(stderr, Name ": failed to freeze container\n"); | |
494 | return -2; | |
495 | } | |
496 | ||
497 | return 1; | |
498 | } | |
499 | ||
500 | static void unfreeze_container(struct supertype *st) | |
501 | { | |
502 | int container_dev = (st->container_dev != NoMdDev | |
503 | ? st->container_dev : st->devnum); | |
d7ca196c N |
504 | char container[40]; |
505 | ||
506 | fmt_devname(container, container_dev); | |
7f2ba464 DW |
507 | |
508 | unblock_monitor(container, 1); | |
509 | } | |
510 | ||
511 | static int freeze(struct supertype *st) | |
512 | { | |
513 | /* Try to freeze resync/rebuild on this array/container. | |
7236ee7a | 514 | * Return -1 if the array is busy, |
7f2ba464 | 515 | * return -2 container cannot be frozen, |
7236ee7a N |
516 | * return 0 if this kernel doesn't support 'frozen' |
517 | * return 1 if it worked. | |
518 | */ | |
7f2ba464 DW |
519 | if (st->ss->external) |
520 | return freeze_container(st); | |
521 | else { | |
522 | struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION); | |
523 | int err; | |
815c8a7e | 524 | char buf[20]; |
7f2ba464 DW |
525 | |
526 | if (!sra) | |
527 | return -1; | |
815c8a7e N |
528 | /* Need to clear any 'read-auto' status */ |
529 | if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 && | |
530 | strncmp(buf, "read-auto", 9) == 0) | |
531 | sysfs_set_str(sra, NULL, "array_state", "clean"); | |
532 | ||
7f2ba464 DW |
533 | err = sysfs_freeze_array(sra); |
534 | sysfs_free(sra); | |
535 | return err; | |
536 | } | |
7236ee7a N |
537 | } |
538 | ||
9202b8eb | 539 | static void unfreeze(struct supertype *st) |
7236ee7a | 540 | { |
7f2ba464 DW |
541 | if (st->ss->external) |
542 | return unfreeze_container(st); | |
543 | else { | |
544 | struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION); | |
545 | ||
546 | if (sra) | |
547 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
548 | else | |
549 | fprintf(stderr, Name ": failed to unfreeze array\n"); | |
550 | sysfs_free(sra); | |
551 | } | |
7236ee7a N |
552 | } |
553 | ||
4411fb17 | 554 | static void wait_reshape(struct mdinfo *sra) |
7236ee7a N |
555 | { |
556 | int fd = sysfs_get_fd(sra, NULL, "sync_action"); | |
557 | char action[20]; | |
558 | ||
92a19f1a AK |
559 | if (fd < 0) |
560 | return; | |
561 | ||
562 | while (sysfs_fd_get_str(fd, action, 20) > 0 && | |
563 | strncmp(action, "reshape", 7) == 0) { | |
7236ee7a N |
564 | fd_set rfds; |
565 | FD_ZERO(&rfds); | |
566 | FD_SET(fd, &rfds); | |
567 | select(fd+1, NULL, NULL, &rfds, NULL); | |
92a19f1a AK |
568 | } |
569 | close(fd); | |
7236ee7a | 570 | } |
7bc71196 DW |
571 | |
572 | static int reshape_super(struct supertype *st, long long size, int level, | |
573 | int layout, int chunksize, int raid_disks, | |
41784c88 AK |
574 | int delta_disks, char *backup_file, char *dev, |
575 | int verbose) | |
7bc71196 DW |
576 | { |
577 | /* nothing extra to check in the native case */ | |
578 | if (!st->ss->external) | |
579 | return 0; | |
580 | if (!st->ss->reshape_super || | |
581 | !st->ss->manage_reshape) { | |
582 | fprintf(stderr, Name ": %s metadata does not support reshape\n", | |
583 | st->ss->name); | |
584 | return 1; | |
585 | } | |
586 | ||
587 | return st->ss->reshape_super(st, size, level, layout, chunksize, | |
41784c88 AK |
588 | raid_disks, delta_disks, backup_file, dev, |
589 | verbose); | |
7bc71196 DW |
590 | } |
591 | ||
592 | static void sync_metadata(struct supertype *st) | |
593 | { | |
594 | if (st->ss->external) { | |
76266030 | 595 | if (st->update_tail) { |
7bc71196 | 596 | flush_metadata_updates(st); |
76266030 N |
597 | st->update_tail = &st->updates; |
598 | } else | |
7bc71196 DW |
599 | st->ss->sync_metadata(st); |
600 | } | |
601 | } | |
602 | ||
603 | static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n) | |
604 | { | |
605 | /* when dealing with external metadata subarrays we need to be | |
606 | * prepared to handle EAGAIN. The kernel may need to wait for | |
607 | * mdmon to mark the array active so the kernel can handle | |
608 | * allocations/writeback when preparing the reshape action | |
609 | * (md_allow_write()). We temporarily disable safe_mode_delay | |
610 | * to close a race with the array_state going clean before the | |
611 | * next write to raid_disks / stripe_cache_size | |
612 | */ | |
613 | char safe[50]; | |
614 | int rc; | |
615 | ||
616 | /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */ | |
5da9ab98 N |
617 | if (!container || |
618 | (strcmp(name, "raid_disks") != 0 && | |
619 | strcmp(name, "stripe_cache_size") != 0)) | |
7bc71196 DW |
620 | return sysfs_set_num(sra, NULL, name, n); |
621 | ||
622 | rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe)); | |
623 | if (rc <= 0) | |
624 | return -1; | |
625 | sysfs_set_num(sra, NULL, "safe_mode_delay", 0); | |
626 | rc = sysfs_set_num(sra, NULL, name, n); | |
627 | if (rc < 0 && errno == EAGAIN) { | |
628 | ping_monitor(container); | |
629 | /* if we get EAGAIN here then the monitor is not active | |
630 | * so stop trying | |
631 | */ | |
632 | rc = sysfs_set_num(sra, NULL, name, n); | |
633 | } | |
634 | sysfs_set_str(sra, NULL, "safe_mode_delay", safe); | |
635 | return rc; | |
636 | } | |
637 | ||
1e971e71 | 638 | int start_reshape(struct mdinfo *sra, int already_running) |
47eb4d5a N |
639 | { |
640 | int err; | |
0f28668b | 641 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); |
47eb4d5a N |
642 | err = sysfs_set_num(sra, NULL, "suspend_hi", 0); |
643 | err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
1e971e71 N |
644 | if (!already_running) |
645 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
47eb4d5a | 646 | err = err ?: sysfs_set_num(sra, NULL, "sync_max", 0); |
1e971e71 N |
647 | if (!already_running) |
648 | err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape"); | |
47eb4d5a N |
649 | |
650 | return err; | |
651 | } | |
652 | ||
653 | void abort_reshape(struct mdinfo *sra) | |
654 | { | |
655 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
656 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); | |
657 | sysfs_set_num(sra, NULL, "suspend_hi", 0); | |
658 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
659 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
660 | sysfs_set_str(sra, NULL, "sync_max", "max"); | |
661 | } | |
662 | ||
dfe77a9e KW |
663 | int remove_disks_for_takeover(struct supertype *st, |
664 | struct mdinfo *sra, | |
665 | int layout) | |
62a48395 AK |
666 | { |
667 | int nr_of_copies; | |
668 | struct mdinfo *remaining; | |
669 | int slot; | |
670 | ||
dfe77a9e KW |
671 | if (sra->array.level == 10) |
672 | nr_of_copies = layout & 0xff; | |
673 | else if (sra->array.level == 1) | |
674 | nr_of_copies = sra->array.raid_disks; | |
675 | else | |
676 | return 1; | |
62a48395 AK |
677 | |
678 | remaining = sra->devs; | |
679 | sra->devs = NULL; | |
680 | /* for each 'copy', select one device and remove from the list. */ | |
681 | for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) { | |
682 | struct mdinfo **diskp; | |
683 | int found = 0; | |
684 | ||
685 | /* Find a working device to keep */ | |
686 | for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) { | |
687 | struct mdinfo *disk = *diskp; | |
688 | ||
689 | if (disk->disk.raid_disk < slot) | |
690 | continue; | |
691 | if (disk->disk.raid_disk >= slot + nr_of_copies) | |
692 | continue; | |
693 | if (disk->disk.state & (1<<MD_DISK_REMOVED)) | |
694 | continue; | |
695 | if (disk->disk.state & (1<<MD_DISK_FAULTY)) | |
696 | continue; | |
697 | if (!(disk->disk.state & (1<<MD_DISK_SYNC))) | |
698 | continue; | |
699 | ||
700 | /* We have found a good disk to use! */ | |
701 | *diskp = disk->next; | |
702 | disk->next = sra->devs; | |
703 | sra->devs = disk; | |
704 | found = 1; | |
705 | break; | |
706 | } | |
707 | if (!found) | |
708 | break; | |
709 | } | |
710 | ||
711 | if (slot < sra->array.raid_disks) { | |
712 | /* didn't find all slots */ | |
713 | struct mdinfo **e; | |
714 | e = &remaining; | |
715 | while (*e) | |
716 | e = &(*e)->next; | |
717 | *e = sra->devs; | |
718 | sra->devs = remaining; | |
719 | return 1; | |
720 | } | |
721 | ||
722 | /* Remove all 'remaining' devices from the array */ | |
723 | while (remaining) { | |
724 | struct mdinfo *sd = remaining; | |
725 | remaining = sd->next; | |
726 | ||
727 | sysfs_set_str(sra, sd, "state", "faulty"); | |
728 | sysfs_set_str(sra, sd, "slot", "none"); | |
d5ca4a23 KW |
729 | /* for external metadata disks should be removed in mdmon */ |
730 | if (!st->ss->external) | |
731 | sysfs_set_str(sra, sd, "state", "remove"); | |
62a48395 AK |
732 | sd->disk.state |= (1<<MD_DISK_REMOVED); |
733 | sd->disk.state &= ~(1<<MD_DISK_SYNC); | |
734 | sd->next = sra->devs; | |
735 | sra->devs = sd; | |
736 | } | |
737 | return 0; | |
738 | } | |
739 | ||
130994cb AK |
740 | void reshape_free_fdlist(int *fdlist, |
741 | unsigned long long *offsets, | |
742 | int size) | |
743 | { | |
744 | int i; | |
745 | ||
746 | for (i = 0; i < size; i++) | |
747 | if (fdlist[i] >= 0) | |
748 | close(fdlist[i]); | |
749 | ||
750 | free(fdlist); | |
751 | free(offsets); | |
752 | } | |
753 | ||
754 | int reshape_prepare_fdlist(char *devname, | |
755 | struct mdinfo *sra, | |
756 | int raid_disks, | |
757 | int nrdisks, | |
758 | unsigned long blocks, | |
759 | char *backup_file, | |
760 | int *fdlist, | |
761 | unsigned long long *offsets) | |
762 | { | |
763 | int d = 0; | |
764 | struct mdinfo *sd; | |
765 | ||
766 | for (d = 0; d <= nrdisks; d++) | |
767 | fdlist[d] = -1; | |
768 | d = raid_disks; | |
769 | for (sd = sra->devs; sd; sd = sd->next) { | |
770 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
771 | continue; | |
772 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
773 | char *dn = map_dev(sd->disk.major, | |
774 | sd->disk.minor, 1); | |
775 | fdlist[sd->disk.raid_disk] | |
776 | = dev_open(dn, O_RDONLY); | |
777 | offsets[sd->disk.raid_disk] = sd->data_offset*512; | |
778 | if (fdlist[sd->disk.raid_disk] < 0) { | |
779 | fprintf(stderr, | |
780 | Name ": %s: cannot open component %s\n", | |
781 | devname, dn ? dn : "-unknown-"); | |
782 | d = -1; | |
783 | goto release; | |
784 | } | |
785 | } else if (backup_file == NULL) { | |
786 | /* spare */ | |
787 | char *dn = map_dev(sd->disk.major, | |
788 | sd->disk.minor, 1); | |
789 | fdlist[d] = dev_open(dn, O_RDWR); | |
790 | offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512; | |
791 | if (fdlist[d] < 0) { | |
792 | fprintf(stderr, Name ": %s: cannot open component %s\n", | |
793 | devname, dn ? dn : "-unknown-"); | |
794 | d = -1; | |
795 | goto release; | |
796 | } | |
797 | d++; | |
798 | } | |
799 | } | |
800 | release: | |
801 | return d; | |
802 | } | |
803 | ||
e6e9d47b AK |
804 | int reshape_open_backup_file(char *backup_file, |
805 | int fd, | |
806 | char *devname, | |
807 | long blocks, | |
808 | int *fdlist, | |
a93f87ee N |
809 | unsigned long long *offsets, |
810 | int restart) | |
e6e9d47b AK |
811 | { |
812 | /* Return 1 on success, 0 on any form of failure */ | |
813 | /* need to check backup file is large enough */ | |
814 | char buf[512]; | |
815 | struct stat stb; | |
816 | unsigned int dev; | |
817 | int i; | |
818 | ||
a93f87ee | 819 | *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL), |
e6e9d47b AK |
820 | S_IRUSR | S_IWUSR); |
821 | *offsets = 8 * 512; | |
822 | if (*fdlist < 0) { | |
823 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
824 | devname, backup_file, strerror(errno)); | |
825 | return 0; | |
826 | } | |
827 | /* Guard against backup file being on array device. | |
828 | * If array is partitioned or if LVM etc is in the | |
829 | * way this will not notice, but it is better than | |
830 | * nothing. | |
831 | */ | |
832 | fstat(*fdlist, &stb); | |
833 | dev = stb.st_dev; | |
834 | fstat(fd, &stb); | |
835 | if (stb.st_rdev == dev) { | |
836 | fprintf(stderr, Name ": backup file must NOT be" | |
837 | " on the array being reshaped.\n"); | |
838 | close(*fdlist); | |
839 | return 0; | |
840 | } | |
841 | ||
842 | memset(buf, 0, 512); | |
ca4fe0bf | 843 | for (i=0; i < blocks + 8 ; i++) { |
e6e9d47b AK |
844 | if (write(*fdlist, buf, 512) != 512) { |
845 | fprintf(stderr, Name ": %s: cannot create" | |
846 | " backup file %s: %s\n", | |
847 | devname, backup_file, strerror(errno)); | |
848 | return 0; | |
849 | } | |
850 | } | |
851 | if (fsync(*fdlist) != 0) { | |
852 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
853 | devname, backup_file, strerror(errno)); | |
854 | return 0; | |
855 | } | |
856 | ||
857 | return 1; | |
858 | } | |
859 | ||
1c009fc2 AK |
860 | unsigned long compute_backup_blocks(int nchunk, int ochunk, |
861 | unsigned int ndata, unsigned int odata) | |
862 | { | |
863 | unsigned long a, b, blocks; | |
864 | /* So how much do we need to backup. | |
865 | * We need an amount of data which is both a whole number of | |
866 | * old stripes and a whole number of new stripes. | |
867 | * So LCM for (chunksize*datadisks). | |
868 | */ | |
869 | a = (ochunk/512) * odata; | |
870 | b = (nchunk/512) * ndata; | |
871 | /* Find GCD */ | |
872 | while (a != b) { | |
873 | if (a < b) | |
874 | b -= a; | |
875 | if (b < a) | |
876 | a -= b; | |
877 | } | |
878 | /* LCM == product / GCD */ | |
879 | blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a; | |
880 | ||
881 | return blocks; | |
882 | } | |
883 | ||
5da9ab98 N |
884 | char *analyse_change(struct mdinfo *info, struct reshape *re) |
885 | { | |
886 | /* Based on the current array state in info->array and | |
887 | * the changes in info->new_* etc, determine: | |
888 | * - whether the change is possible | |
889 | * - Intermediate level/raid_disks/layout | |
890 | * - whether a restriping reshape is needed | |
891 | * - number of sectors in minimum change unit. This | |
892 | * will cover a whole number of stripes in 'before' and | |
893 | * 'after'. | |
894 | * | |
895 | * Return message if the change should be rejected | |
896 | * NULL if the change can be achieved | |
897 | * | |
898 | * This can be called as part of starting a reshape, or | |
899 | * when assembling an array that is undergoing reshape. | |
900 | */ | |
901 | int new_disks; | |
b545e14a N |
902 | /* delta_parity records change in number of devices |
903 | * caused by level change | |
904 | */ | |
905 | int delta_parity = 0; | |
5da9ab98 N |
906 | |
907 | /* If a new level not explicitly given, we assume no-change */ | |
908 | if (info->new_level == UnSet) | |
909 | info->new_level = info->array.level; | |
910 | ||
911 | if (info->new_chunk) | |
912 | switch (info->new_level) { | |
913 | case 0: | |
914 | case 4: | |
915 | case 5: | |
916 | case 6: | |
917 | case 10: | |
918 | /* chunk size is meaningful, must divide component_size | |
919 | * evenly | |
920 | */ | |
921 | if (info->component_size % (info->new_chunk/512)) | |
922 | return "New chunk size does not" | |
923 | " divide component size"; | |
924 | break; | |
925 | default: | |
926 | return "chunk size not meaningful for this level"; | |
927 | } | |
928 | else | |
929 | info->new_chunk = info->array.chunk_size; | |
930 | ||
931 | switch (info->array.level) { | |
90b60dfa N |
932 | default: |
933 | return "Cannot understand this RAID level"; | |
5da9ab98 N |
934 | case 1: |
935 | /* RAID1 can convert to RAID1 with different disks, or | |
dfe77a9e KW |
936 | * raid5 with 2 disks, or |
937 | * raid0 with 1 disk | |
5da9ab98 | 938 | */ |
dfe77a9e | 939 | if (info->new_level == 0) { |
b545e14a N |
940 | if (info->delta_disks != UnSet && |
941 | info->delta_disks != 0) | |
942 | return "Cannot change number of disks " | |
943 | "with RAID1->RAID0 conversion"; | |
dfe77a9e KW |
944 | re->level = 0; |
945 | re->before.data_disks = 1; | |
946 | re->after.data_disks = 1; | |
947 | re->before.layout = 0; | |
948 | re->backup_blocks = 0; | |
949 | re->parity = 0; | |
950 | return NULL; | |
951 | } | |
5da9ab98 N |
952 | if (info->new_level == 1) { |
953 | if (info->delta_disks == UnSet) | |
954 | /* Don't know what to do */ | |
955 | return "no change requested for Growing RAID1"; | |
956 | re->level = 1; | |
b4f8e38b | 957 | re->backup_blocks = 0; |
5da9ab98 N |
958 | re->parity = 0; |
959 | return NULL; | |
960 | } | |
961 | if (info->array.raid_disks == 2 && | |
446d2a5a | 962 | info->new_level == 5) { |
5652f2d9 | 963 | |
5da9ab98 | 964 | re->level = 5; |
5da9ab98 | 965 | re->before.data_disks = 1; |
5652f2d9 N |
966 | if (info->delta_disks != UnSet && |
967 | info->delta_disks != 0) | |
968 | re->after.data_disks = 1 + info->delta_disks; | |
969 | else | |
970 | re->after.data_disks = 1; | |
971 | if (re->after.data_disks < 1) | |
972 | return "Number of disks too small for RAID5"; | |
973 | ||
5da9ab98 | 974 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC; |
446d2a5a N |
975 | info->array.chunk_size = 65536; |
976 | break; | |
5da9ab98 N |
977 | } |
978 | /* Could do some multi-stage conversions, but leave that to | |
979 | * later. | |
980 | */ | |
981 | return "Impossibly level change request for RAID1"; | |
982 | ||
983 | case 10: | |
984 | /* RAID10 can only be converted from near mode to | |
985 | * RAID0 by removing some devices | |
986 | */ | |
987 | if ((info->array.layout & ~0xff) != 0x100) | |
988 | return "Cannot Grow RAID10 with far/offset layout"; | |
989 | /* number of devices must be multiple of number of copies */ | |
990 | if (info->array.raid_disks % (info->array.layout & 0xff)) | |
991 | return "RAID10 layout too complex for Grow operation"; | |
992 | ||
993 | if (info->new_level != 0) | |
994 | return "RAID10 can only be changed to RAID0"; | |
995 | new_disks = (info->array.raid_disks | |
996 | / (info->array.layout & 0xff)); | |
b545e14a | 997 | if (info->delta_disks == UnSet) |
5da9ab98 N |
998 | info->delta_disks = (new_disks |
999 | - info->array.raid_disks); | |
b545e14a | 1000 | |
5da9ab98 N |
1001 | if (info->delta_disks != new_disks - info->array.raid_disks) |
1002 | return "New number of raid-devices impossible for RAID10"; | |
1003 | if (info->new_chunk && | |
1004 | info->new_chunk != info->array.chunk_size) | |
1005 | return "Cannot change chunk-size with RAID10 Grow"; | |
1006 | ||
1007 | /* looks good */ | |
1008 | re->level = 0; | |
1009 | re->parity = 0; | |
1010 | re->before.data_disks = new_disks; | |
031d445c | 1011 | re->after.data_disks = re->before.data_disks; |
5da9ab98 | 1012 | re->before.layout = 0; |
b4f8e38b | 1013 | re->backup_blocks = 0; |
5da9ab98 N |
1014 | return NULL; |
1015 | ||
1016 | case 0: | |
1017 | /* RAID0 can be converted to RAID10, or to RAID456 */ | |
1018 | if (info->new_level == 10) { | |
1019 | if (info->new_layout == UnSet && info->delta_disks == UnSet) { | |
1020 | /* Assume near=2 layout */ | |
1021 | info->new_layout = 0x102; | |
1022 | info->delta_disks = info->array.raid_disks; | |
1023 | } | |
1024 | if (info->new_layout == UnSet) { | |
1025 | int copies = 1 + (info->delta_disks | |
1026 | / info->array.raid_disks); | |
1027 | if (info->array.raid_disks * (copies-1) | |
1028 | != info->delta_disks) | |
1029 | return "Impossible number of devices" | |
1030 | " for RAID0->RAID10"; | |
1031 | info->new_layout = 0x100 + copies; | |
1032 | } | |
1033 | if (info->delta_disks == UnSet) { | |
1034 | int copies = info->new_layout & 0xff; | |
1035 | if (info->new_layout != 0x100 + copies) | |
1036 | return "New layout impossible" | |
1037 | " for RAID0->RAID10";; | |
1038 | info->delta_disks = (copies - 1) * | |
1039 | info->array.raid_disks; | |
1040 | } | |
1041 | if (info->new_chunk && | |
1042 | info->new_chunk != info->array.chunk_size) | |
1043 | return "Cannot change chunk-size with RAID0->RAID10"; | |
1044 | /* looks good */ | |
1045 | re->level = 10; | |
1046 | re->parity = 0; | |
1047 | re->before.data_disks = (info->array.raid_disks + | |
1048 | info->delta_disks); | |
031d445c | 1049 | re->after.data_disks = re->before.data_disks; |
5da9ab98 | 1050 | re->before.layout = info->new_layout; |
b4f8e38b | 1051 | re->backup_blocks = 0; |
5da9ab98 N |
1052 | return NULL; |
1053 | } | |
1054 | ||
1055 | /* RAID0 can also covert to RAID0/4/5/6 by first converting to | |
1056 | * a raid4 style layout of the final level. | |
1057 | */ | |
1058 | switch (info->new_level) { | |
5da9ab98 | 1059 | case 4: |
b545e14a N |
1060 | delta_parity = 1; |
1061 | case 0: | |
5da9ab98 N |
1062 | re->level = 4; |
1063 | re->before.layout = 0; | |
1064 | break; | |
1065 | case 5: | |
b545e14a | 1066 | delta_parity = 1; |
5da9ab98 N |
1067 | re->level = 5; |
1068 | re->before.layout = ALGORITHM_PARITY_N; | |
1069 | break; | |
1070 | case 6: | |
b545e14a | 1071 | delta_parity = 2; |
5da9ab98 N |
1072 | re->level = 6; |
1073 | re->before.layout = ALGORITHM_PARITY_N; | |
1074 | break; | |
1075 | default: | |
1076 | return "Impossible level change requested"; | |
1077 | } | |
1078 | re->before.data_disks = info->array.raid_disks; | |
1079 | /* determining 'after' layout happens outside this 'switch' */ | |
1080 | break; | |
1081 | ||
1082 | case 4: | |
1083 | info->array.layout = ALGORITHM_PARITY_N; | |
1084 | case 5: | |
1085 | switch (info->new_level) { | |
f0cce442 | 1086 | case 0: |
b545e14a | 1087 | delta_parity = -1; |
5da9ab98 N |
1088 | case 4: |
1089 | re->level = info->array.level; | |
1090 | re->before.data_disks = info->array.raid_disks - 1; | |
1091 | re->before.layout = info->array.layout; | |
1092 | break; | |
1093 | case 5: | |
1094 | re->level = 5; | |
1095 | re->before.data_disks = info->array.raid_disks - 1; | |
1096 | re->before.layout = info->array.layout; | |
1097 | break; | |
1098 | case 6: | |
b545e14a | 1099 | delta_parity = 1; |
5da9ab98 N |
1100 | re->level = 6; |
1101 | re->before.data_disks = info->array.raid_disks - 1; | |
1102 | switch (info->array.layout) { | |
1103 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1104 | re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
1105 | break; | |
1106 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1107 | re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
1108 | break; | |
1109 | case ALGORITHM_LEFT_SYMMETRIC: | |
1110 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
1111 | break; | |
1112 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1113 | re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
1114 | break; | |
1115 | case ALGORITHM_PARITY_0: | |
1116 | re->before.layout = ALGORITHM_PARITY_0_6; | |
1117 | break; | |
1118 | case ALGORITHM_PARITY_N: | |
1119 | re->before.layout = ALGORITHM_PARITY_N_6; | |
1120 | break; | |
1121 | default: | |
1122 | return "Cannot convert an array with this layout"; | |
1123 | } | |
1124 | break; | |
1125 | case 1: | |
1126 | if (info->array.raid_disks != 2) | |
1127 | return "Can only convert a 2-device array to RAID1"; | |
b545e14a N |
1128 | if (info->delta_disks != UnSet && |
1129 | info->delta_disks != 0) | |
1130 | return "Cannot set raid_disk when " | |
1131 | "converting RAID5->RAID1"; | |
5da9ab98 | 1132 | re->level = 1; |
5da9ab98 N |
1133 | break; |
1134 | default: | |
1135 | return "Impossible level change requested"; | |
1136 | } | |
1137 | break; | |
1138 | case 6: | |
1139 | switch (info->new_level) { | |
1140 | case 4: | |
1141 | case 5: | |
b545e14a | 1142 | delta_parity = -1; |
5da9ab98 N |
1143 | case 6: |
1144 | re->level = 6; | |
1145 | re->before.data_disks = info->array.raid_disks - 2; | |
1146 | re->before.layout = info->array.layout; | |
1147 | break; | |
1148 | default: | |
1149 | return "Impossible level change requested"; | |
1150 | } | |
1151 | break; | |
1152 | } | |
1153 | ||
1154 | /* If we reached here then it looks like a re-stripe is | |
1155 | * happening. We have determined the intermediate level | |
1156 | * and initial raid_disks/layout and stored these in 're'. | |
1157 | * | |
1158 | * We need to deduce the final layout that can be atomically | |
1159 | * converted to the end state. | |
1160 | */ | |
1161 | switch (info->new_level) { | |
1162 | case 0: | |
1163 | /* We can only get to RAID0 from RAID4 or RAID5 | |
1164 | * with appropriate layout and one extra device | |
1165 | */ | |
1166 | if (re->level != 4 && re->level != 5) | |
1167 | return "Cannot covert to RAID0 from this level"; | |
b545e14a | 1168 | |
5da9ab98 N |
1169 | switch (re->level) { |
1170 | case 4: | |
1171 | re->after.layout = 0 ; break; | |
1172 | case 5: | |
1173 | re->after.layout = ALGORITHM_PARITY_N; break; | |
1174 | } | |
1175 | break; | |
1176 | ||
1177 | case 4: | |
1178 | /* We can only get to RAID4 from RAID5 */ | |
1179 | if (re->level != 4 && re->level != 5) | |
1180 | return "Cannot convert to RAID4 from this level"; | |
b545e14a | 1181 | |
5da9ab98 N |
1182 | switch (re->level) { |
1183 | case 4: | |
1184 | re->after.layout = 0 ; break; | |
1185 | case 5: | |
1186 | re->after.layout = ALGORITHM_PARITY_N; break; | |
1187 | } | |
1188 | break; | |
1189 | ||
1190 | case 5: | |
1191 | /* We get to RAID5 for RAID5 or RAID6 */ | |
1192 | if (re->level != 5 && re->level != 6) | |
1193 | return "Cannot convert to RAID5 from this level"; | |
b545e14a | 1194 | |
5da9ab98 N |
1195 | switch (re->level) { |
1196 | case 5: | |
1197 | if (info->new_layout == UnSet) | |
1198 | re->after.layout = re->before.layout; | |
1199 | else | |
1200 | re->after.layout = info->new_layout; | |
1201 | break; | |
1202 | case 6: | |
4a870364 N |
1203 | if (info->new_layout == UnSet) |
1204 | info->new_layout = re->before.layout; | |
1205 | ||
5da9ab98 N |
1206 | /* after.layout needs to be raid6 version of new_layout */ |
1207 | if (info->new_layout == ALGORITHM_PARITY_N) | |
1208 | re->after.layout = ALGORITHM_PARITY_N; | |
1209 | else { | |
1210 | char layout[40]; | |
1211 | char *ls = map_num(r5layout, info->new_layout); | |
1212 | int l; | |
1213 | strcat(strcpy(layout, ls), "-6"); | |
1214 | l = map_name(r6layout, layout); | |
1215 | if (l == UnSet) | |
1216 | return "Cannot find RAID6 layout" | |
1217 | " to convert to"; | |
1218 | re->after.layout = l; | |
1219 | } | |
1220 | } | |
1221 | break; | |
1222 | ||
1223 | case 6: | |
1224 | /* We must already be at level 6 */ | |
1225 | if (re->level != 6) | |
1226 | return "Impossible level change"; | |
5da9ab98 | 1227 | if (info->new_layout == UnSet) |
4a870364 | 1228 | re->after.layout = info->array.layout; |
5da9ab98 N |
1229 | else |
1230 | re->after.layout = info->new_layout; | |
1231 | break; | |
1232 | default: | |
1233 | return "Impossible level change requested"; | |
1234 | } | |
b545e14a N |
1235 | if (info->delta_disks == UnSet) |
1236 | info->delta_disks = delta_parity; | |
1237 | ||
1238 | re->after.data_disks = (re->before.data_disks | |
1239 | + info->delta_disks | |
1240 | - delta_parity); | |
5da9ab98 N |
1241 | switch (re->level) { |
1242 | case 6: re->parity = 2; break; | |
1243 | case 4: | |
1244 | case 5: re->parity = 1; break; | |
1245 | default: re->parity = 0; break; | |
1246 | } | |
1247 | /* So we have a restripe operation, we need to calculate the number | |
1248 | * of blocks per reshape operation. | |
1249 | */ | |
1250 | if (info->new_chunk == 0) | |
1251 | info->new_chunk = info->array.chunk_size; | |
1252 | if (re->after.data_disks == re->before.data_disks && | |
1253 | re->after.layout == re->before.layout && | |
1254 | info->new_chunk == info->array.chunk_size) { | |
1255 | /* Nothing to change */ | |
b4f8e38b | 1256 | re->backup_blocks = 0; |
5da9ab98 N |
1257 | return NULL; |
1258 | } | |
1259 | if (re->after.data_disks == 1 && re->before.data_disks == 1) { | |
446d2a5a | 1260 | /* chunk and layout changes make no difference */ |
b4f8e38b | 1261 | re->backup_blocks = 0; |
5da9ab98 N |
1262 | return NULL; |
1263 | } | |
1264 | ||
1265 | if (re->after.data_disks == re->before.data_disks && | |
1266 | get_linux_version() < 2006032) | |
1267 | return "in-place reshape is not safe before 2.6.32 - sorry."; | |
1268 | ||
1269 | if (re->after.data_disks < re->before.data_disks && | |
1270 | get_linux_version() < 2006030) | |
508ede86 | 1271 | return "reshape to fewer devices is not supported before 2.6.30 - sorry."; |
5da9ab98 | 1272 | |
b4f8e38b | 1273 | re->backup_blocks = compute_backup_blocks( |
5da9ab98 N |
1274 | info->new_chunk, info->array.chunk_size, |
1275 | re->after.data_disks, | |
1276 | re->before.data_disks); | |
1277 | ||
1278 | re->new_size = info->component_size * re->after.data_disks; | |
1279 | return NULL; | |
1280 | } | |
1281 | ||
1282 | static int reshape_array(char *container, int fd, char *devname, | |
1283 | struct supertype *st, struct mdinfo *info, | |
e2e53a2d N |
1284 | int force, struct mddev_dev *devlist, |
1285 | char *backup_file, int quiet, int forked, | |
a93f87ee | 1286 | int restart); |
493f5dd6 | 1287 | static int reshape_container(char *container, char *devname, |
5da9ab98 N |
1288 | struct supertype *st, |
1289 | struct mdinfo *info, | |
1290 | int force, | |
1291 | char *backup_file, | |
493f5dd6 | 1292 | int quiet, int restart); |
1c009fc2 | 1293 | |
06b0d786 | 1294 | int Grow_reshape(char *devname, int fd, int quiet, char *backup_file, |
e86c9dd6 | 1295 | long long size, |
691a36b7 | 1296 | int level, char *layout_str, int chunksize, int raid_disks, |
e2e53a2d | 1297 | struct mddev_dev *devlist, |
ce52f92f | 1298 | int assume_clean, int force) |
e86c9dd6 NB |
1299 | { |
1300 | /* Make some changes in the shape of an array. | |
1301 | * The kernel must support the change. | |
7236ee7a N |
1302 | * |
1303 | * There are three different changes. Each can trigger | |
1304 | * a resync or recovery so we freeze that until we have | |
1305 | * requested everything (if kernel supports freezing - 2.6.30). | |
1306 | * The steps are: | |
1307 | * - change size (i.e. component_size) | |
1308 | * - change level | |
1309 | * - change layout/chunksize/ndisks | |
1310 | * | |
1311 | * The last can require a reshape. It is different on different | |
1312 | * levels so we need to check the level before actioning it. | |
1313 | * Some times the level change needs to be requested after the | |
1314 | * reshape (e.g. raid6->raid5, raid5->raid0) | |
1315 | * | |
e86c9dd6 | 1316 | */ |
5da9ab98 | 1317 | struct mdu_array_info_s array; |
7236ee7a | 1318 | int rv = 0; |
e86c9dd6 | 1319 | struct supertype *st; |
4725bc31 | 1320 | char *subarray = NULL; |
e86c9dd6 | 1321 | |
7236ee7a | 1322 | int frozen; |
7236ee7a | 1323 | int changed = 0; |
7bc71196 | 1324 | char *container = NULL; |
5da9ab98 | 1325 | char container_buf[20]; |
7bc71196 | 1326 | int cfd = -1; |
e86c9dd6 | 1327 | |
e2e53a2d N |
1328 | struct mddev_dev *dv; |
1329 | int added_disks; | |
1330 | ||
5da9ab98 | 1331 | struct mdinfo info; |
7e0f6979 | 1332 | struct mdinfo *sra; |
e86c9dd6 NB |
1333 | |
1334 | if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) { | |
1335 | fprintf(stderr, Name ": %s is not an active md array - aborting\n", | |
1336 | devname); | |
1337 | return 1; | |
1338 | } | |
24d40069 | 1339 | |
9ce510be N |
1340 | if (size >= 0 && |
1341 | (chunksize || level!= UnSet || layout_str || raid_disks)) { | |
1342 | fprintf(stderr, Name ": cannot change component size at the same time " | |
1343 | "as other changes.\n" | |
1344 | " Change size first, then check data is intact before " | |
1345 | "making other changes.\n"); | |
1346 | return 1; | |
1347 | } | |
1348 | ||
24d40069 N |
1349 | if (raid_disks && raid_disks < array.raid_disks && array.level > 1 && |
1350 | get_linux_version() < 2006032 && | |
1351 | !check_env("MDADM_FORCE_FEWER")) { | |
1352 | fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n" | |
1353 | " Please use a newer kernel\n"); | |
1354 | return 1; | |
1355 | } | |
7bc71196 DW |
1356 | |
1357 | st = super_by_fd(fd, &subarray); | |
1358 | if (!st) { | |
1359 | fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname); | |
1360 | return 1; | |
1361 | } | |
acab7bb1 N |
1362 | if (raid_disks > st->max_devs) { |
1363 | fprintf(stderr, Name ": Cannot increase raid-disks on this array" | |
1364 | " beyond %d\n", st->max_devs); | |
1365 | return 1; | |
1366 | } | |
7bc71196 DW |
1367 | |
1368 | /* in the external case we need to check that the requested reshape is | |
1369 | * supported, and perform an initial check that the container holds the | |
1370 | * pre-requisite spare devices (mdmon owns final validation) | |
1371 | */ | |
1372 | if (st->ss->external) { | |
1373 | int container_dev; | |
7f2ba464 | 1374 | int rv; |
7bc71196 DW |
1375 | |
1376 | if (subarray) { | |
1377 | container_dev = st->container_dev; | |
1378 | cfd = open_dev_excl(st->container_dev); | |
7bc71196 DW |
1379 | } else { |
1380 | container_dev = st->devnum; | |
1381 | close(fd); | |
1382 | cfd = open_dev_excl(st->devnum); | |
1383 | fd = cfd; | |
1384 | } | |
1385 | if (cfd < 0) { | |
1386 | fprintf(stderr, Name ": Unable to open container for %s\n", | |
1387 | devname); | |
7f2ba464 | 1388 | free(subarray); |
7bc71196 DW |
1389 | return 1; |
1390 | } | |
1391 | ||
5da9ab98 N |
1392 | fmt_devname(container_buf, container_dev); |
1393 | container = container_buf; | |
7bc71196 | 1394 | |
eb744788 AK |
1395 | rv = st->ss->load_container(st, cfd, NULL); |
1396 | ||
7f2ba464 | 1397 | if (rv) { |
7bc71196 DW |
1398 | fprintf(stderr, Name ": Cannot read superblock for %s\n", |
1399 | devname); | |
7f2ba464 | 1400 | free(subarray); |
7bc71196 DW |
1401 | return 1; |
1402 | } | |
1403 | ||
1404 | if (mdmon_running(container_dev)) | |
1405 | st->update_tail = &st->updates; | |
e2e53a2d | 1406 | } |
691a36b7 | 1407 | |
e2e53a2d N |
1408 | added_disks = 0; |
1409 | for (dv = devlist; dv; dv = dv->next) | |
1410 | added_disks++; | |
691a36b7 | 1411 | if (raid_disks > array.raid_disks && |
e2e53a2d | 1412 | array.spare_disks +added_disks < (raid_disks - array.raid_disks) && |
691a36b7 N |
1413 | !force) { |
1414 | fprintf(stderr, | |
1415 | Name ": Need %d spare%s to avoid degraded array," | |
1416 | " and only have %d.\n" | |
1417 | " Use --force to over-ride this check.\n", | |
1418 | raid_disks - array.raid_disks, | |
1419 | raid_disks - array.raid_disks == 1 ? "" : "s", | |
e2e53a2d | 1420 | array.spare_disks + added_disks); |
691a36b7 | 1421 | return 1; |
7bc71196 DW |
1422 | } |
1423 | ||
3fa436ac ML |
1424 | sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS |
1425 | | GET_STATE | GET_VERSION); | |
1426 | if (sra) { | |
7f2ba464 | 1427 | if (st->ss->external && subarray == NULL) { |
7bc71196 DW |
1428 | array.level = LEVEL_CONTAINER; |
1429 | sra->array.level = LEVEL_CONTAINER; | |
1430 | } | |
7bc71196 | 1431 | } else { |
b526e52d DW |
1432 | fprintf(stderr, Name ": failed to read sysfs parameters for %s\n", |
1433 | devname); | |
1434 | return 1; | |
1435 | } | |
7f2ba464 DW |
1436 | frozen = freeze(st); |
1437 | if (frozen < -1) { | |
1438 | /* freeze() already spewed the reason */ | |
1439 | return 1; | |
1440 | } else if (frozen < 0) { | |
7236ee7a N |
1441 | fprintf(stderr, Name ": %s is performing resync/recovery and cannot" |
1442 | " be reshaped\n", devname); | |
1443 | return 1; | |
1444 | } | |
1445 | ||
1446 | /* ========= set size =============== */ | |
1447 | if (size >= 0 && (size == 0 || size != array.size)) { | |
85f10287 | 1448 | long long orig_size = get_component_size(fd)/2; |
d1537ed1 | 1449 | struct mdinfo *mdi; |
7bc71196 | 1450 | |
85f10287 N |
1451 | if (orig_size == 0) |
1452 | orig_size = array.size; | |
1453 | ||
41784c88 AK |
1454 | if (reshape_super(st, size, UnSet, UnSet, 0, 0, UnSet, NULL, |
1455 | devname, !quiet)) { | |
7bc71196 DW |
1456 | rv = 1; |
1457 | goto release; | |
1458 | } | |
1459 | sync_metadata(st); | |
d1537ed1 N |
1460 | |
1461 | /* Update the size of each member device in case | |
1462 | * they have been resized. This will never reduce | |
1463 | * below the current used-size. The "size" attribute | |
1464 | * understand '0' to mean 'max'. | |
1465 | */ | |
1466 | for (mdi = sra->devs; mdi; mdi = mdi->next) | |
1467 | sysfs_set_num(sra, mdi, "size", size); | |
1468 | ||
7236ee7a N |
1469 | array.size = size; |
1470 | if (array.size != size) { | |
1471 | /* got truncated to 32bit, write to | |
1472 | * component_size instead | |
1473 | */ | |
1474 | if (sra) | |
1475 | rv = sysfs_set_num(sra, NULL, | |
1476 | "component_size", size); | |
1477 | else | |
1478 | rv = -1; | |
1479 | } else | |
1480 | rv = ioctl(fd, SET_ARRAY_INFO, &array); | |
1481 | if (rv != 0) { | |
39bbb392 | 1482 | int err = errno; |
7bc71196 DW |
1483 | |
1484 | /* restore metadata */ | |
1485 | if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0, | |
41784c88 | 1486 | UnSet, NULL, devname, !quiet) == 0) |
7bc71196 | 1487 | sync_metadata(st); |
7236ee7a | 1488 | fprintf(stderr, Name ": Cannot set device size for %s: %s\n", |
39bbb392 N |
1489 | devname, strerror(err)); |
1490 | if (err == EBUSY && | |
1491 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1492 | fprintf(stderr, " Bitmap must be removed before size can be changed\n"); | |
7236ee7a N |
1493 | rv = 1; |
1494 | goto release; | |
1495 | } | |
ce52f92f | 1496 | if (assume_clean) { |
6cbf8fb8 | 1497 | /* This will fail on kernels newer than 3.0 unless |
ce52f92f N |
1498 | * a backport has been arranged. |
1499 | */ | |
1500 | if (sra == NULL || | |
1501 | sysfs_set_str(sra, NULL, "resync_start", "none") < 0) | |
1502 | fprintf(stderr, Name ": --assume-clean not support with --grow on this kernel\n"); | |
1503 | } | |
7236ee7a | 1504 | ioctl(fd, GET_ARRAY_INFO, &array); |
be1cabbd | 1505 | size = get_component_size(fd)/2; |
f98841b3 N |
1506 | if (size == 0) |
1507 | size = array.size; | |
85f10287 N |
1508 | if (!quiet) { |
1509 | if (size == orig_size) | |
1510 | fprintf(stderr, Name ": component size of %s " | |
1511 | "unchanged at %lluK\n", | |
1512 | devname, size); | |
1513 | else | |
1514 | fprintf(stderr, Name ": component size of %s " | |
1515 | "has been set to %lluK\n", | |
1516 | devname, size); | |
1517 | } | |
7236ee7a | 1518 | changed = 1; |
7bc71196 | 1519 | } else if (array.level != LEVEL_CONTAINER) { |
be1cabbd | 1520 | size = get_component_size(fd)/2; |
f98841b3 N |
1521 | if (size == 0) |
1522 | size = array.size; | |
7236ee7a N |
1523 | } |
1524 | ||
d515d27f N |
1525 | /* See if there is anything else to do */ |
1526 | if ((level == UnSet || level == array.level) && | |
1527 | (layout_str == NULL) && | |
1528 | (chunksize == 0 || chunksize == array.chunk_size) && | |
1529 | (raid_disks == 0 || raid_disks == array.raid_disks)) { | |
1530 | /* Nothing more to do */ | |
1531 | if (!changed && !quiet) | |
1532 | fprintf(stderr, Name ": %s: no change requested\n", | |
1533 | devname); | |
1534 | goto release; | |
1535 | } | |
1536 | ||
dfe77a9e | 1537 | /* ========= check for Raid10/Raid1 -> Raid0 conversion =============== |
5da9ab98 | 1538 | * current implementation assumes that following conditions must be met: |
dfe77a9e KW |
1539 | * - RAID10: |
1540 | * - far_copies == 1 | |
1541 | * - near_copies == 2 | |
62a48395 | 1542 | */ |
dfe77a9e KW |
1543 | if ((level == 0 && array.level == 10 && sra && |
1544 | array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) || | |
1545 | (level == 0 && array.level == 1 && sra)) { | |
62a48395 | 1546 | int err; |
dfe77a9e | 1547 | err = remove_disks_for_takeover(st, sra, array.layout); |
62a48395 AK |
1548 | if (err) { |
1549 | dprintf(Name": Array cannot be reshaped\n"); | |
62a48395 AK |
1550 | if (cfd > -1) |
1551 | close(cfd); | |
5da9ab98 N |
1552 | rv = 1; |
1553 | goto release; | |
7236ee7a | 1554 | } |
bb025c2f KW |
1555 | /* FIXME this is added with no justification - why is it here */ |
1556 | ping_monitor(container); | |
7236ee7a N |
1557 | } |
1558 | ||
8ff6d094 | 1559 | memset(&info, 0, sizeof(info)); |
5da9ab98 | 1560 | info.array = array; |
7443ee81 | 1561 | sysfs_init(&info, fd, NoMdDev); |
3fa436ac | 1562 | strcpy(info.text_version, sra->text_version); |
7443ee81 | 1563 | info.component_size = size*2; |
5da9ab98 N |
1564 | info.new_level = level; |
1565 | info.new_chunk = chunksize * 1024; | |
eff4954d N |
1566 | if (info.array.level == LEVEL_CONTAINER) { |
1567 | info.delta_disks = UnSet; | |
1568 | info.array.raid_disks = raid_disks; | |
1569 | } else if (raid_disks) | |
5da9ab98 N |
1570 | info.delta_disks = raid_disks - info.array.raid_disks; |
1571 | else | |
1572 | info.delta_disks = UnSet; | |
1573 | if (layout_str == NULL) { | |
1574 | info.new_layout = UnSet; | |
1575 | if (info.array.level == 6 && | |
1576 | (info.new_level == 6 || info.new_level == UnSet) && | |
1577 | info.array.layout >= 16) { | |
7236ee7a | 1578 | fprintf(stderr, Name |
5da9ab98 N |
1579 | ": %s has a non-standard layout. If you" |
1580 | " wish to preserve this\n" | |
1581 | " during the reshape, please specify" | |
1582 | " --layout=preserve\n" | |
1583 | " If you want to change it, specify a" | |
1584 | " layout or use --layout=normalise\n", | |
7236ee7a N |
1585 | devname); |
1586 | rv = 1; | |
1587 | goto release; | |
1588 | } | |
5da9ab98 N |
1589 | } else if (strcmp(layout_str, "normalise") == 0 || |
1590 | strcmp(layout_str, "normalize") == 0) { | |
1591 | /* If we have a -6 RAID6 layout, remove the '-6'. */ | |
1592 | info.new_layout = UnSet; | |
1593 | if (info.array.level == 6 && info.new_level == UnSet) { | |
1594 | char l[40], *h; | |
1595 | strcpy(l, map_num(r6layout, info.array.layout)); | |
1596 | h = strrchr(l, '-'); | |
1597 | if (h && strcmp(h, "-6") == 0) { | |
1598 | *h = 0; | |
1599 | info.new_layout = map_name(r6layout, l); | |
7236ee7a N |
1600 | } |
1601 | } | |
5da9ab98 N |
1602 | } else if (strcmp(layout_str, "preserve") == 0) { |
1603 | info.new_layout = UnSet; | |
1604 | } else { | |
1605 | int l = info.new_level; | |
1606 | if (l == UnSet) | |
1607 | l = info.array.level; | |
1608 | switch (l) { | |
1609 | case 5: | |
1610 | info.new_layout = map_name(r5layout, layout_str); | |
1611 | break; | |
1612 | case 6: | |
1613 | info.new_layout = map_name(r6layout, layout_str); | |
1614 | break; | |
1615 | case 10: | |
1616 | info.new_layout = parse_layout_10(layout_str); | |
1617 | break; | |
1618 | case LEVEL_FAULTY: | |
1619 | info.new_layout = parse_layout_faulty(layout_str); | |
1620 | break; | |
1621 | default: | |
1622 | fprintf(stderr, Name ": layout not meaningful" | |
1623 | " with this level\n"); | |
7bc71196 DW |
1624 | rv = 1; |
1625 | goto release; | |
1626 | } | |
5da9ab98 N |
1627 | if (info.new_layout == UnSet) { |
1628 | fprintf(stderr, Name ": layout %s not understood" | |
1629 | " for this level\n", | |
1630 | layout_str); | |
1631 | rv = 1; | |
1632 | goto release; | |
7bc71196 | 1633 | } |
7236ee7a N |
1634 | } |
1635 | ||
907ea753 N |
1636 | if (array.level == LEVEL_FAULTY) { |
1637 | if (level != UnSet && level != array.level) { | |
1638 | fprintf(stderr, Name ": cannot change level of Faulty device\n"); | |
1639 | rv =1 ; | |
1640 | } | |
1641 | if (chunksize) { | |
1642 | fprintf(stderr, Name ": cannot set chunksize of Faulty device\n"); | |
1643 | rv =1 ; | |
1644 | } | |
1645 | if (raid_disks && raid_disks != 1) { | |
1646 | fprintf(stderr, Name ": cannot set raid_disks of Faulty device\n"); | |
1647 | rv =1 ; | |
1648 | } | |
1649 | if (layout_str) { | |
1650 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
1651 | dprintf("Cannot get array information.\n"); | |
1652 | goto release; | |
1653 | } | |
1654 | array.layout = info.new_layout; | |
1655 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
1656 | fprintf(stderr, Name ": failed to set new layout\n"); | |
1657 | rv = 1; | |
1658 | } else if (!quiet) | |
1659 | printf("layout for %s set to %d\n", | |
1660 | devname, array.layout); | |
1661 | } | |
1662 | } else if (array.level == LEVEL_CONTAINER) { | |
5da9ab98 N |
1663 | /* This change is to be applied to every array in the |
1664 | * container. This is only needed when the metadata imposes | |
1665 | * restraints of the various arrays in the container. | |
1666 | * Currently we only know that IMSM requires all arrays | |
1667 | * to have the same number of devices so changing the | |
1668 | * number of devices (On-Line Capacity Expansion) must be | |
1669 | * performed at the level of the container | |
1670 | */ | |
493f5dd6 N |
1671 | rv = reshape_container(container, devname, st, &info, |
1672 | force, backup_file, quiet, 0); | |
9202b8eb | 1673 | frozen = 0; |
5da9ab98 | 1674 | } else { |
08f9e34b AK |
1675 | /* get spare devices from external metadata |
1676 | */ | |
1677 | if (st->ss->external) { | |
1678 | struct mdinfo *info2; | |
1679 | ||
1680 | info2 = st->ss->container_content(st, subarray); | |
1681 | if (info2) { | |
1682 | info.array.spare_disks = | |
1683 | info2->array.spare_disks; | |
1684 | sysfs_free(info2); | |
1685 | } | |
1686 | } | |
1687 | ||
5da9ab98 N |
1688 | /* Impose these changes on a single array. First |
1689 | * check that the metadata is OK with the change. */ | |
7bc71196 | 1690 | |
5da9ab98 N |
1691 | if (reshape_super(st, info.component_size, info.new_level, |
1692 | info.new_layout, info.new_chunk, | |
41784c88 | 1693 | info.array.raid_disks, info.delta_disks, |
5da9ab98 | 1694 | backup_file, devname, quiet)) { |
7bc71196 DW |
1695 | rv = 1; |
1696 | goto release; | |
1697 | } | |
5da9ab98 N |
1698 | sync_metadata(st); |
1699 | rv = reshape_array(container, fd, devname, st, &info, force, | |
e2e53a2d | 1700 | devlist, backup_file, quiet, 0, 0); |
9202b8eb | 1701 | frozen = 0; |
5da9ab98 N |
1702 | } |
1703 | release: | |
9202b8eb N |
1704 | if (frozen > 0) |
1705 | unfreeze(st); | |
5da9ab98 N |
1706 | return rv; |
1707 | } | |
7bc71196 | 1708 | |
5da9ab98 N |
1709 | static int reshape_array(char *container, int fd, char *devname, |
1710 | struct supertype *st, struct mdinfo *info, | |
e2e53a2d | 1711 | int force, struct mddev_dev *devlist, |
a93f87ee N |
1712 | char *backup_file, int quiet, int forked, |
1713 | int restart) | |
5da9ab98 N |
1714 | { |
1715 | struct reshape reshape; | |
1716 | int spares_needed; | |
1717 | char *msg; | |
1718 | int orig_level = UnSet; | |
1719 | int disks, odisks; | |
7bc71196 | 1720 | |
5da9ab98 N |
1721 | struct mdu_array_info_s array; |
1722 | char *c; | |
e86c9dd6 | 1723 | |
e2e53a2d N |
1724 | struct mddev_dev *dv; |
1725 | int added_disks; | |
1726 | ||
5da9ab98 N |
1727 | int *fdlist; |
1728 | unsigned long long *offsets; | |
1729 | int d; | |
1730 | int nrdisks; | |
1731 | int err; | |
da8100ca | 1732 | unsigned long blocks; |
5da9ab98 N |
1733 | unsigned long cache; |
1734 | unsigned long long array_size; | |
1735 | int done; | |
cf6ac177 | 1736 | struct mdinfo *sra = NULL; |
7bc71196 | 1737 | |
9468aeac N |
1738 | /* when reshaping a RAID0, the component_size might be zero. |
1739 | * So try to fix that up. | |
1740 | */ | |
1741 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
1742 | dprintf("Cannot get array information.\n"); | |
1743 | goto release; | |
1744 | } | |
1745 | if (array.level == 0 && info->component_size == 0) { | |
1746 | get_dev_size(fd, NULL, &array_size); | |
1747 | info->component_size = array_size / array.raid_disks; | |
1748 | } | |
1749 | ||
3cb2aed2 N |
1750 | if (info->reshape_active) { |
1751 | int new_level = info->new_level; | |
1752 | info->new_level = UnSet; | |
178b8f35 | 1753 | info->array.raid_disks -= info->delta_disks; |
3cb2aed2 N |
1754 | msg = analyse_change(info, &reshape); |
1755 | info->new_level = new_level; | |
178b8f35 | 1756 | info->array.raid_disks += info->delta_disks; |
fcdfb814 AK |
1757 | if (!restart) |
1758 | /* Make sure the array isn't read-only */ | |
1759 | ioctl(fd, RESTART_ARRAY_RW, 0); | |
3cb2aed2 N |
1760 | } else |
1761 | msg = analyse_change(info, &reshape); | |
5da9ab98 N |
1762 | if (msg) { |
1763 | fprintf(stderr, Name ": %s\n", msg); | |
631d7405 | 1764 | goto release; |
5da9ab98 | 1765 | } |
817ed7d6 N |
1766 | if (restart && |
1767 | (reshape.level != info->array.level || | |
1768 | reshape.before.layout != info->array.layout || | |
384e9be1 AK |
1769 | reshape.before.data_disks + reshape.parity |
1770 | != info->array.raid_disks - info->delta_disks)) { | |
20a40eca N |
1771 | fprintf(stderr, Name ": reshape info is not in native format -" |
1772 | " cannot continue.\n"); | |
1773 | goto release; | |
1774 | } | |
a93f87ee N |
1775 | |
1776 | if (restart) { | |
1777 | /* reshape already started. just skip to monitoring the reshape */ | |
1778 | if (reshape.backup_blocks == 0) | |
1779 | return 0; | |
1780 | goto started; | |
1781 | } | |
a9c3e78f AK |
1782 | /* The container is frozen but the array may not be. |
1783 | * So freeze the array so spares don't get put to the wrong use | |
1784 | * FIXME there should probably be a cleaner separation between | |
1785 | * freeze_array and freeze_container. | |
1786 | */ | |
1787 | sysfs_freeze_array(info); | |
e06c4e59 | 1788 | /* Check we have enough spares to not be degraded */ |
e2e53a2d N |
1789 | added_disks = 0; |
1790 | for (dv = devlist; dv ; dv=dv->next) | |
1791 | added_disks++; | |
5da9ab98 N |
1792 | spares_needed = max(reshape.before.data_disks, |
1793 | reshape.after.data_disks) | |
1794 | + reshape.parity - array.raid_disks; | |
7bc71196 | 1795 | |
88c1a083 | 1796 | if (!force && |
e2e53a2d N |
1797 | info->new_level > 1 && info->array.level > 1 && |
1798 | spares_needed > info->array.spare_disks + added_disks) { | |
5da9ab98 N |
1799 | fprintf(stderr, |
1800 | Name ": Need %d spare%s to avoid degraded array," | |
1801 | " and only have %d.\n" | |
1802 | " Use --force to over-ride this check.\n", | |
1803 | spares_needed, | |
1804 | spares_needed == 1 ? "" : "s", | |
e2e53a2d | 1805 | info->array.spare_disks + added_disks); |
631d7405 | 1806 | goto release; |
7bc71196 | 1807 | } |
e06c4e59 N |
1808 | /* Check we have enough spares to not fail */ |
1809 | spares_needed = max(reshape.before.data_disks, | |
1810 | reshape.after.data_disks) | |
1811 | - array.raid_disks; | |
1812 | if ((info->new_level > 1 || info->new_level == 0) && | |
e2e53a2d | 1813 | spares_needed > info->array.spare_disks +added_disks) { |
e06c4e59 N |
1814 | fprintf(stderr, |
1815 | Name ": Need %d spare%s to create working array," | |
1816 | " and only have %d.\n", | |
1817 | spares_needed, | |
1818 | spares_needed == 1 ? "" : "s", | |
e2e53a2d | 1819 | info->array.spare_disks + added_disks); |
e06c4e59 N |
1820 | goto release; |
1821 | } | |
e86c9dd6 | 1822 | |
eae35f5c | 1823 | if (reshape.level != array.level) { |
5da9ab98 N |
1824 | char *c = map_num(pers, reshape.level); |
1825 | int err; | |
1826 | if (c == NULL) | |
631d7405 | 1827 | goto release; |
e86c9dd6 | 1828 | |
5da9ab98 N |
1829 | err = sysfs_set_str(info, NULL, "level", c); |
1830 | if (err) { | |
1831 | err = errno; | |
1832 | fprintf(stderr, Name ": %s: could not set level to %s\n", | |
1833 | devname, c); | |
1834 | if (err == EBUSY && | |
1835 | (info->array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1836 | fprintf(stderr, " Bitmap must be removed" | |
1837 | " before level can be changed\n"); | |
631d7405 | 1838 | goto release; |
19678e53 | 1839 | } |
5da9ab98 | 1840 | if (!quiet) |
b6b95155 | 1841 | fprintf(stderr, Name ": level of %s changed to %s\n", |
5da9ab98 | 1842 | devname, c); |
eae35f5c | 1843 | orig_level = array.level; |
3cd4e7c4 | 1844 | sysfs_freeze_array(info); |
e86c9dd6 | 1845 | |
16d4d84e AK |
1846 | if (reshape.level > 0 && st->ss->external) { |
1847 | /* make sure mdmon is aware of the new level */ | |
1848 | if (!mdmon_running(st->container_dev)) | |
1849 | start_mdmon(st->container_dev); | |
1850 | ping_monitor(container); | |
e919fb0a AK |
1851 | if (mdmon_running(st->container_dev) && |
1852 | st->update_tail == NULL) | |
1853 | st->update_tail = &st->updates; | |
16d4d84e | 1854 | } |
5da9ab98 | 1855 | } |
5da9ab98 N |
1856 | /* ->reshape_super might have chosen some spares from the |
1857 | * container that it wants to be part of the new array. | |
1858 | * We can collect them with ->container_content and give | |
1859 | * them to the kernel. | |
1860 | */ | |
1861 | if (st->ss->reshape_super && st->ss->container_content) { | |
1862 | char *subarray = strchr(info->text_version+1, '/')+1; | |
1863 | struct mdinfo *info2 = | |
1864 | st->ss->container_content(st, subarray); | |
1865 | struct mdinfo *d; | |
1866 | ||
2dd0d697 AK |
1867 | if (info2) { |
1868 | sysfs_init(info2, fd, st->devnum); | |
0d5ac3c6 N |
1869 | /* When increasing number of devices, we need to set |
1870 | * new raid_disks before adding these, or they might | |
1871 | * be rejected. | |
1872 | */ | |
1873 | if (reshape.backup_blocks && | |
1874 | reshape.after.data_disks > reshape.before.data_disks) | |
1875 | subarray_set_num(container, info2, "raid_disks", | |
1876 | reshape.after.data_disks + | |
1877 | reshape.parity); | |
5da9ab98 N |
1878 | for (d = info2->devs; d; d = d->next) { |
1879 | if (d->disk.state == 0 && | |
1880 | d->disk.raid_disk >= 0) { | |
1881 | /* This is a spare that wants to | |
1882 | * be part of the array. | |
1883 | */ | |
1884 | add_disk(fd, st, info2, d); | |
1885 | } | |
9860f271 | 1886 | } |
2dd0d697 AK |
1887 | sysfs_free(info2); |
1888 | } | |
5da9ab98 | 1889 | } |
e2e53a2d N |
1890 | /* We might have been given some devices to add to the |
1891 | * array. Now that the array has been changed to the right | |
1892 | * level and frozen, we can safely add them. | |
1893 | */ | |
1894 | if (devlist) | |
1895 | Manage_subdevs(devname, fd, devlist, !quiet, | |
1896 | 0,NULL); | |
1686dc25 | 1897 | |
b4f8e38b | 1898 | if (reshape.backup_blocks == 0) { |
5da9ab98 N |
1899 | /* No restriping needed, but we might need to impose |
1900 | * some more changes: layout, raid_disks, chunk_size | |
e86c9dd6 | 1901 | */ |
24aebf3a KW |
1902 | /* read current array info */ |
1903 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
83732c28 | 1904 | dprintf("Cannot get array information.\n"); |
24aebf3a KW |
1905 | goto release; |
1906 | } | |
1907 | /* compare current array info with new values and if | |
1908 | * it is different update them to new */ | |
5da9ab98 | 1909 | if (info->new_layout != UnSet && |
24aebf3a KW |
1910 | info->new_layout != array.layout) { |
1911 | array.layout = info->new_layout; | |
1912 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
5da9ab98 | 1913 | fprintf(stderr, Name ": failed to set new layout\n"); |
631d7405 | 1914 | goto release; |
5da9ab98 N |
1915 | } else if (!quiet) |
1916 | printf("layout for %s set to %d\n", | |
24aebf3a | 1917 | devname, array.layout); |
5da9ab98 N |
1918 | } |
1919 | if (info->delta_disks != UnSet && | |
24aebf3a KW |
1920 | info->delta_disks != 0 && |
1921 | array.raid_disks != (info->array.raid_disks + info->delta_disks)) { | |
1922 | array.raid_disks += info->delta_disks; | |
1923 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
5da9ab98 | 1924 | fprintf(stderr, Name ": failed to set raid disks\n"); |
631d7405 | 1925 | goto release; |
24aebf3a | 1926 | } else if (!quiet) { |
5da9ab98 | 1927 | printf("raid_disks for %s set to %d\n", |
24aebf3a KW |
1928 | devname, array.raid_disks); |
1929 | } | |
5da9ab98 N |
1930 | } |
1931 | if (info->new_chunk != 0 && | |
24aebf3a | 1932 | info->new_chunk != array.chunk_size) { |
5da9ab98 N |
1933 | if (sysfs_set_num(info, NULL, |
1934 | "chunk_size", info->new_chunk) != 0) { | |
1935 | fprintf(stderr, Name ": failed to set chunk size\n"); | |
631d7405 | 1936 | goto release; |
5da9ab98 N |
1937 | } else if (!quiet) |
1938 | printf("chunk size for %s set to %d\n", | |
24aebf3a | 1939 | devname, array.chunk_size); |
4725bc31 | 1940 | } |
e35b189b | 1941 | unfreeze(st); |
631d7405 | 1942 | return 0; |
5da9ab98 | 1943 | } |
19678e53 | 1944 | |
5da9ab98 N |
1945 | /* |
1946 | * There are three possibilities. | |
1947 | * 1/ The array will shrink. | |
1948 | * We need to ensure the reshape will pause before reaching | |
1949 | * the 'critical section'. We also need to fork and wait for | |
1950 | * that to happen. When it does we | |
1951 | * suspend/backup/complete/unfreeze | |
1952 | * | |
1953 | * 2/ The array will not change size. | |
1954 | * This requires that we keep a backup of a sliding window | |
1955 | * so that we can restore data after a crash. So we need | |
1956 | * to fork and monitor progress. | |
1957 | * In future we will allow the data_offset to change, so | |
1958 | * a sliding backup becomes unnecessary. | |
1959 | * | |
1960 | * 3/ The array will grow. This is relatively easy. | |
1961 | * However the kernel's restripe routines will cheerfully | |
1962 | * overwrite some early data before it is safe. So we | |
1963 | * need to make a backup of the early parts of the array | |
1964 | * and be ready to restore it if rebuild aborts very early. | |
1965 | * For externally managed metadata, we still need a forked | |
1966 | * child to monitor the reshape and suspend IO over the region | |
1967 | * that is being reshaped. | |
1968 | * | |
1969 | * We backup data by writing it to one spare, or to a | |
1970 | * file which was given on command line. | |
1971 | * | |
1972 | * In each case, we first make sure that storage is available | |
1973 | * for the required backup. | |
1974 | * Then we: | |
1975 | * - request the shape change. | |
1976 | * - fork to handle backup etc. | |
1977 | */ | |
a93f87ee | 1978 | started: |
5da9ab98 N |
1979 | /* Check that we can hold all the data */ |
1980 | get_dev_size(fd, NULL, &array_size); | |
1981 | if (reshape.new_size < (array_size/512)) { | |
1982 | fprintf(stderr, | |
1983 | Name ": this change will reduce the size of the array.\n" | |
1984 | " use --grow --array-size first to truncate array.\n" | |
1985 | " e.g. mdadm --grow %s --array-size %llu\n", | |
1986 | devname, reshape.new_size/2); | |
5da9ab98 N |
1987 | goto release; |
1988 | } | |
7236ee7a | 1989 | |
5da9ab98 | 1990 | sra = sysfs_read(fd, 0, |
3656edcd | 1991 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK| |
5da9ab98 | 1992 | GET_CACHE); |
5da9ab98 N |
1993 | if (!sra) { |
1994 | fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n", | |
1995 | devname); | |
5da9ab98 N |
1996 | goto release; |
1997 | } | |
7236ee7a | 1998 | |
5da9ab98 N |
1999 | /* Decide how many blocks (sectors) for a reshape |
2000 | * unit. The number we have so far is just a minimum | |
2001 | */ | |
b4f8e38b | 2002 | blocks = reshape.backup_blocks; |
5da9ab98 N |
2003 | if (reshape.before.data_disks == |
2004 | reshape.after.data_disks) { | |
2005 | /* Make 'blocks' bigger for better throughput, but | |
2006 | * not so big that we reject it below. | |
2007 | * Try for 16 megabytes | |
2008 | */ | |
2009 | while (blocks * 32 < sra->component_size && | |
2010 | blocks < 16*1024*2) | |
2011 | blocks *= 2; | |
2012 | } else | |
2013 | fprintf(stderr, Name ": Need to backup %luK of critical " | |
2014 | "section..\n", blocks/2); | |
2015 | ||
2016 | if (blocks >= sra->component_size/2) { | |
2017 | fprintf(stderr, Name ": %s: Something wrong" | |
2018 | " - reshape aborted\n", | |
2019 | devname); | |
5da9ab98 N |
2020 | goto release; |
2021 | } | |
7236ee7a | 2022 | |
5da9ab98 N |
2023 | /* Now we need to open all these devices so we can read/write. |
2024 | */ | |
b8b286a6 N |
2025 | nrdisks = max(reshape.before.data_disks, |
2026 | reshape.after.data_disks) + reshape.parity | |
2027 | + sra->array.spare_disks; | |
5da9ab98 N |
2028 | fdlist = malloc((1+nrdisks) * sizeof(int)); |
2029 | offsets = malloc((1+nrdisks) * sizeof(offsets[0])); | |
2030 | if (!fdlist || !offsets) { | |
2031 | fprintf(stderr, Name ": malloc failed: grow aborted\n"); | |
5da9ab98 N |
2032 | goto release; |
2033 | } | |
e380d3be | 2034 | |
b8b286a6 N |
2035 | odisks = reshape.before.data_disks + reshape.parity; |
2036 | d = reshape_prepare_fdlist(devname, sra, odisks, | |
5da9ab98 N |
2037 | nrdisks, blocks, backup_file, |
2038 | fdlist, offsets); | |
2039 | if (d < 0) { | |
5da9ab98 N |
2040 | goto release; |
2041 | } | |
13c37ad3 AK |
2042 | if ((st->ss->manage_reshape == NULL) || |
2043 | (st->ss->recover_backup == NULL)) { | |
2044 | if (backup_file == NULL) { | |
2045 | if (reshape.after.data_disks <= | |
2046 | reshape.before.data_disks) { | |
2047 | fprintf(stderr, Name ": %s: Cannot grow - " | |
2048 | "need backup-file\n", devname); | |
2049 | goto release; | |
2050 | } else if (sra->array.spare_disks == 0) { | |
2051 | fprintf(stderr, Name ": %s: Cannot grow - " | |
2052 | "need a spare or backup-file to backup " | |
2053 | "critical section\n", devname); | |
2054 | goto release; | |
2055 | } | |
2056 | } else { | |
2057 | if (!reshape_open_backup_file(backup_file, fd, devname, | |
2058 | (signed)blocks, | |
2059 | fdlist+d, offsets+d, | |
2060 | restart)) { | |
2061 | goto release; | |
2062 | } | |
2063 | d++; | |
e86c9dd6 | 2064 | } |
5da9ab98 | 2065 | } |
7236ee7a | 2066 | |
5da9ab98 N |
2067 | /* lastly, check that the internal stripe cache is |
2068 | * large enough, or it won't work. | |
2069 | * It must hold at least 4 stripes of the larger | |
2070 | * chunk size | |
2071 | */ | |
2072 | cache = max(info->array.chunk_size, info->new_chunk); | |
2073 | cache *= 4; /* 4 stripes minimum */ | |
2074 | cache /= 512; /* convert to sectors */ | |
2075 | disks = min(reshape.before.data_disks, reshape.after.data_disks); | |
2076 | /* make sure there is room for 'blocks' with a bit to spare */ | |
2077 | if (cache < 16 + blocks / disks) | |
2078 | cache = 16 + blocks / disks; | |
2079 | cache /= (4096/512); /* Covert from sectors to pages */ | |
2080 | ||
2081 | if (sra->cache_size < cache) | |
2082 | subarray_set_num(container, sra, "stripe_cache_size", | |
2083 | cache+1); | |
2084 | ||
2085 | /* Right, everything seems fine. Let's kick things off. | |
2086 | * If only changing raid_disks, use ioctl, else use | |
2087 | * sysfs. | |
2088 | */ | |
2089 | sync_metadata(st); | |
2090 | ||
b4f8e38b | 2091 | sra->new_chunk = info->new_chunk; |
ddee071d | 2092 | |
20a40eca | 2093 | if (restart) |
ddee071d | 2094 | sra->reshape_progress = info->reshape_progress; |
f897078e N |
2095 | else { |
2096 | sra->reshape_progress = 0; | |
2097 | if (reshape.after.data_disks < reshape.before.data_disks) | |
2098 | /* start from the end of the new array */ | |
2099 | sra->reshape_progress = (sra->component_size | |
2100 | * reshape.after.data_disks); | |
2101 | } | |
2102 | ||
2103 | if (info->array.chunk_size == info->new_chunk && | |
7477d513 AK |
2104 | reshape.before.layout == reshape.after.layout && |
2105 | st->ss->external == 0) { | |
f897078e | 2106 | /* use SET_ARRAY_INFO but only if reshape hasn't started */ |
6ef421be | 2107 | ioctl(fd, GET_ARRAY_INFO, &array); |
5da9ab98 | 2108 | array.raid_disks = reshape.after.data_disks + reshape.parity; |
20a40eca | 2109 | if (!restart && |
f897078e | 2110 | ioctl(fd, SET_ARRAY_INFO, &array) != 0) { |
5da9ab98 | 2111 | int err = errno; |
631d7405 | 2112 | |
5da9ab98 N |
2113 | fprintf(stderr, |
2114 | Name ": Cannot set device shape for %s: %s\n", | |
2115 | devname, strerror(errno)); | |
7bc71196 | 2116 | |
5da9ab98 N |
2117 | if (err == EBUSY && |
2118 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
2119 | fprintf(stderr, | |
2120 | " Bitmap must be removed before" | |
2121 | " shape can be changed\n"); | |
b5420ef3 | 2122 | |
5da9ab98 N |
2123 | goto release; |
2124 | } | |
20a40eca | 2125 | } else if (!restart) { |
5da9ab98 | 2126 | /* set them all just in case some old 'new_*' value |
f897078e | 2127 | * persists from some earlier problem. |
7236ee7a | 2128 | */ |
631d7405 | 2129 | int err = 0; |
5da9ab98 | 2130 | if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0) |
631d7405 N |
2131 | err = errno; |
2132 | if (!err && sysfs_set_num(sra, NULL, "layout", | |
5da9ab98 | 2133 | reshape.after.layout) < 0) |
631d7405 N |
2134 | err = errno; |
2135 | if (!err && subarray_set_num(container, sra, "raid_disks", | |
5da9ab98 N |
2136 | reshape.after.data_disks + |
2137 | reshape.parity) < 0) | |
631d7405 N |
2138 | err = errno; |
2139 | if (err) { | |
5da9ab98 N |
2140 | fprintf(stderr, Name ": Cannot set device shape for %s\n", |
2141 | devname); | |
7236ee7a | 2142 | |
5da9ab98 N |
2143 | if (err == EBUSY && |
2144 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
2145 | fprintf(stderr, | |
2146 | " Bitmap must be removed before" | |
2147 | " shape can be changed\n"); | |
2148 | goto release; | |
e86c9dd6 | 2149 | } |
5da9ab98 | 2150 | } |
e86c9dd6 | 2151 | |
20a40eca | 2152 | err = start_reshape(sra, restart); |
fab32c97 | 2153 | if (err) { |
20a40eca N |
2154 | fprintf(stderr, |
2155 | Name ": Cannot %s reshape for %s\n", | |
2156 | restart ? "continue" : "start", | |
fab32c97 AK |
2157 | devname); |
2158 | goto release; | |
2159 | } | |
a93f87ee N |
2160 | if (restart) |
2161 | sysfs_set_str(sra, NULL, "array_state", "active"); | |
5da9ab98 N |
2162 | |
2163 | /* Now we just need to kick off the reshape and watch, while | |
2164 | * handling backups of the data... | |
2165 | * This is all done by a forked background process. | |
2166 | */ | |
2167 | switch(forked ? 0 : fork()) { | |
6b2d630c N |
2168 | case -1: |
2169 | fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n", | |
2170 | strerror(errno)); | |
6b2d630c | 2171 | abort_reshape(sra); |
631d7405 | 2172 | goto release; |
6b2d630c N |
2173 | default: |
2174 | return 0; | |
5da9ab98 | 2175 | case 0: |
6b2d630c N |
2176 | break; |
2177 | } | |
5da9ab98 | 2178 | |
6b2d630c N |
2179 | close(fd); |
2180 | if (check_env("MDADM_GROW_VERIFY")) | |
2181 | fd = open(devname, O_RDONLY | O_DIRECT); | |
2182 | else | |
2183 | fd = -1; | |
2184 | mlockall(MCL_FUTURE); | |
5da9ab98 | 2185 | |
6b2d630c N |
2186 | if (st->ss->external) { |
2187 | /* metadata handler takes it from here */ | |
2188 | done = st->ss->manage_reshape( | |
2189 | fd, sra, &reshape, st, blocks, | |
2190 | fdlist, offsets, | |
2191 | d - odisks, fdlist+odisks, | |
2192 | offsets+odisks); | |
2193 | } else | |
2194 | done = child_monitor( | |
2195 | fd, sra, &reshape, st, blocks, | |
2196 | fdlist, offsets, | |
2197 | d - odisks, fdlist+odisks, | |
2198 | offsets+odisks); | |
2199 | ||
2200 | if (backup_file && done) | |
2201 | unlink(backup_file); | |
2202 | if (!done) { | |
2203 | abort_reshape(sra); | |
2204 | goto out; | |
2205 | } | |
1cbc5680 N |
2206 | |
2207 | if (!st->ss->external && | |
2208 | !(reshape.before.data_disks != reshape.after.data_disks | |
2209 | && info->custom_array_size) && | |
2210 | info->new_level == reshape.level && | |
2211 | !forked) { | |
2212 | /* no need to wait for the reshape to finish as | |
2213 | * there is nothing more to do. | |
2214 | */ | |
2215 | exit(0); | |
2216 | } | |
2217 | wait_reshape(sra); | |
2218 | ||
2219 | if (st->ss->external) { | |
2220 | /* Re-load the metadata as much could have changed */ | |
2221 | int cfd = open_dev(st->container_dev); | |
2222 | if (cfd >= 0) { | |
2223 | ping_monitor(container); | |
2224 | st->ss->free_super(st); | |
2225 | st->ss->load_container(st, cfd, container); | |
2226 | close(cfd); | |
2227 | } | |
2228 | } | |
2229 | ||
6b2d630c N |
2230 | /* set new array size if required customer_array_size is used |
2231 | * by this metadata. | |
2232 | */ | |
2233 | if (reshape.before.data_disks != | |
2234 | reshape.after.data_disks && | |
2235 | info->custom_array_size) { | |
2236 | struct mdinfo *info2; | |
2237 | char *subarray = strchr(info->text_version+1, '/')+1; | |
582496b2 | 2238 | |
6b2d630c N |
2239 | info2 = st->ss->container_content(st, subarray); |
2240 | if (info2) { | |
2241 | unsigned long long current_size = 0; | |
2242 | unsigned long long new_size = | |
2243 | info2->custom_array_size/2; | |
2244 | ||
2245 | if (sysfs_get_ll(sra, | |
2246 | NULL, | |
2247 | "array_size", | |
2248 | ¤t_size) == 0 && | |
2249 | new_size > current_size) { | |
2250 | if (sysfs_set_num(sra, NULL, | |
2251 | "array_size", new_size) | |
2252 | < 0) | |
2253 | dprintf("Error: Cannot" | |
2254 | " set array size"); | |
2255 | else | |
2256 | dprintf("Array size " | |
2257 | "changed"); | |
2258 | dprintf(" from %llu to %llu.\n", | |
2259 | current_size, new_size); | |
582496b2 | 2260 | } |
6b2d630c | 2261 | sysfs_free(info2); |
582496b2 | 2262 | } |
6b2d630c | 2263 | } |
582496b2 | 2264 | |
6b2d630c | 2265 | if (info->new_level != reshape.level) { |
582496b2 | 2266 | |
6b2d630c | 2267 | c = map_num(pers, info->new_level); |
1cbc5680 N |
2268 | if (c) { |
2269 | err = sysfs_set_str(sra, NULL, "level", c); | |
2270 | if (err) | |
2271 | fprintf(stderr, Name\ | |
2272 | ": %s: could not set level " | |
2273 | "to %s\n", devname, c); | |
2274 | } | |
e919fb0a AK |
2275 | if (info->new_level == 0) |
2276 | st->update_tail = NULL; | |
7236ee7a | 2277 | } |
6b2d630c N |
2278 | out: |
2279 | if (forked) | |
2280 | return 0; | |
e84f2c00 | 2281 | unfreeze(st); |
6b2d630c | 2282 | exit(0); |
e86c9dd6 | 2283 | |
6b2d630c | 2284 | release: |
1cbc5680 | 2285 | if (orig_level != UnSet && sra) { |
7236ee7a N |
2286 | c = map_num(pers, orig_level); |
2287 | if (c && sysfs_set_str(sra, NULL, "level", c) == 0) | |
2288 | fprintf(stderr, Name ": aborting level change\n"); | |
2289 | } | |
9202b8eb N |
2290 | if (!forked) |
2291 | unfreeze(st); | |
631d7405 | 2292 | return 1; |
7236ee7a | 2293 | } |
e86c9dd6 | 2294 | |
493f5dd6 | 2295 | int reshape_container(char *container, char *devname, |
5da9ab98 N |
2296 | struct supertype *st, |
2297 | struct mdinfo *info, | |
2298 | int force, | |
2299 | char *backup_file, | |
493f5dd6 | 2300 | int quiet, int restart) |
5da9ab98 | 2301 | { |
1bb174ba | 2302 | struct mdinfo *cc = NULL; |
bcc9e9ed | 2303 | int rv = restart; |
1bb174ba | 2304 | |
d8eb27f7 KA |
2305 | /* component_size is not meaningful for a container, |
2306 | * so pass '-1' meaning 'no change' | |
2307 | */ | |
493f5dd6 N |
2308 | if (!restart && |
2309 | reshape_super(st, -1, info->new_level, | |
5da9ab98 | 2310 | info->new_layout, info->new_chunk, |
41784c88 | 2311 | info->array.raid_disks, info->delta_disks, |
9e325442 AK |
2312 | backup_file, devname, quiet)) { |
2313 | unfreeze(st); | |
5da9ab98 | 2314 | return 1; |
9e325442 | 2315 | } |
5da9ab98 N |
2316 | |
2317 | sync_metadata(st); | |
2318 | ||
1bb174ba AK |
2319 | /* ping monitor to be sure that update is on disk |
2320 | */ | |
2321 | ping_monitor(container); | |
5da9ab98 N |
2322 | |
2323 | switch (fork()) { | |
2324 | case -1: /* error */ | |
2325 | perror("Cannot fork to complete reshape\n"); | |
9e325442 | 2326 | unfreeze(st); |
5da9ab98 N |
2327 | return 1; |
2328 | default: /* parent */ | |
2329 | printf(Name ": multi-array reshape continues in background\n"); | |
2330 | return 0; | |
2331 | case 0: /* child */ | |
2332 | break; | |
2333 | } | |
2334 | ||
1bb174ba AK |
2335 | while(1) { |
2336 | /* For each member array with reshape_active, | |
2337 | * we need to perform the reshape. | |
2338 | * We pick the first array that needs reshaping and | |
2339 | * reshape it. reshape_array() will re-read the metadata | |
2340 | * so the next time through a different array should be | |
2341 | * ready for reshape. | |
493f5dd6 N |
2342 | * It is possible that the 'different' array will not |
2343 | * be assembled yet. In that case we simple exit. | |
2344 | * When it is assembled, the mdadm which assembles it | |
2345 | * will take over the reshape. | |
1bb174ba AK |
2346 | */ |
2347 | struct mdinfo *content; | |
5da9ab98 N |
2348 | int fd; |
2349 | struct mdstat_ent *mdstat; | |
5da9ab98 N |
2350 | char *adev; |
2351 | ||
1bb174ba | 2352 | sysfs_free(cc); |
5da9ab98 | 2353 | |
1bb174ba AK |
2354 | cc = st->ss->container_content(st, NULL); |
2355 | ||
2356 | for (content = cc; content ; content = content->next) { | |
2357 | char *subarray; | |
2358 | if (!content->reshape_active) | |
2359 | continue; | |
2360 | ||
2361 | subarray = strchr(content->text_version+1, '/')+1; | |
2362 | mdstat = mdstat_by_subdev(subarray, | |
2363 | devname2devnum(container)); | |
2364 | if (!mdstat) | |
2365 | continue; | |
2366 | break; | |
2367 | } | |
2368 | if (!content) | |
2369 | break; | |
5da9ab98 | 2370 | |
55f14721 | 2371 | fd = open_dev(mdstat->devnum); |
5da9ab98 N |
2372 | if (fd < 0) |
2373 | break; | |
2374 | adev = map_dev(dev2major(mdstat->devnum), | |
2375 | dev2minor(mdstat->devnum), | |
2376 | 0); | |
2377 | if (!adev) | |
1bb174ba AK |
2378 | adev = content->text_version; |
2379 | ||
2380 | sysfs_init(content, fd, mdstat->devnum); | |
5da9ab98 | 2381 | |
1bb174ba | 2382 | rv = reshape_array(container, fd, adev, st, |
e2e53a2d | 2383 | content, force, NULL, |
493f5dd6 | 2384 | backup_file, quiet, 1, restart); |
5da9ab98 | 2385 | close(fd); |
493f5dd6 | 2386 | restart = 0; |
5da9ab98 N |
2387 | if (rv) |
2388 | break; | |
2389 | } | |
bcc9e9ed AK |
2390 | if (!rv) |
2391 | unfreeze(st); | |
1bb174ba | 2392 | sysfs_free(cc); |
5da9ab98 N |
2393 | exit(0); |
2394 | } | |
2395 | ||
7236ee7a N |
2396 | /* |
2397 | * We run a child process in the background which performs the following | |
2398 | * steps: | |
2399 | * - wait for resync to reach a certain point | |
2400 | * - suspend io to the following section | |
2401 | * - backup that section | |
2402 | * - allow resync to proceed further | |
2403 | * - resume io | |
2404 | * - discard the backup. | |
2405 | * | |
2406 | * When are combined in slightly different ways in the three cases. | |
2407 | * Grow: | |
2408 | * - suspend/backup/allow/wait/resume/discard | |
2409 | * Shrink: | |
2410 | * - allow/wait/suspend/backup/allow/wait/resume/discard | |
2411 | * same-size: | |
2412 | * - wait/resume/discard/suspend/backup/allow | |
2413 | * | |
2414 | * suspend/backup/allow always come together | |
2415 | * wait/resume/discard do too. | |
2416 | * For the same-size case we have two backups to improve flow. | |
2417 | * | |
2418 | */ | |
e86c9dd6 | 2419 | |
7443ee81 N |
2420 | int progress_reshape(struct mdinfo *info, struct reshape *reshape, |
2421 | unsigned long long backup_point, | |
2422 | unsigned long long wait_point, | |
2423 | unsigned long long *suspend_point, | |
2424 | unsigned long long *reshape_completed) | |
2425 | { | |
2426 | /* This function is called repeatedly by the reshape manager. | |
2427 | * It determines how much progress can safely be made and allows | |
2428 | * that progress. | |
2429 | * - 'info' identifies the array and particularly records in | |
2430 | * ->reshape_progress the metadata's knowledge of progress | |
2431 | * This is a sector offset from the start of the array | |
2432 | * of the next array block to be relocated. This number | |
2433 | * may increase from 0 or decrease from array_size, depending | |
2434 | * on the type of reshape that is happening. | |
2435 | * Note that in contrast, 'sync_completed' is a block count of the | |
2436 | * reshape so far. It gives the distance between the start point | |
2437 | * (head or tail of device) and the next place that data will be | |
2438 | * written. It always increases. | |
2439 | * - 'reshape' is the structure created by analyse_change | |
2440 | * - 'backup_point' shows how much the metadata manager has backed-up | |
2441 | * data. For reshapes with increasing progress, it is the next address | |
2442 | * to be backed up, previous addresses have been backed-up. For | |
2443 | * decreasing progress, it is the earliest address that has been | |
2444 | * backed up - later address are also backed up. | |
2445 | * So addresses between reshape_progress and backup_point are | |
2446 | * backed up providing those are in the 'correct' order. | |
2447 | * - 'wait_point' is an array address. When reshape_completed | |
2448 | * passes this point, progress_reshape should return. It might | |
2449 | * return earlier if it determines that ->reshape_progress needs | |
2450 | * to be updated or further backup is needed. | |
2451 | * - suspend_point is maintained by progress_reshape and the caller | |
2452 | * should not touch it except to initialise to zero. | |
2453 | * It is an array address and it only increases in 2.6.37 and earlier. | |
b6b95155 | 2454 | * This makes it difficult to handle reducing reshapes with |
7443ee81 N |
2455 | * external metadata. |
2456 | * However: it is similar to backup_point in that it records the | |
2457 | * other end of a suspended region from reshape_progress. | |
2458 | * it is moved to extend the region that is safe to backup and/or | |
2459 | * reshape | |
2460 | * - reshape_completed is read from sysfs and returned. The caller | |
2461 | * should copy this into ->reshape_progress when it has reason to | |
2462 | * believe that the metadata knows this, and any backup outside this | |
2463 | * has been erased. | |
2464 | * | |
2465 | * Return value is: | |
2466 | * 1 if more data from backup_point - but only as far as suspend_point, | |
2467 | * should be backed up | |
2468 | * 0 if things are progressing smoothly | |
9e034f70 N |
2469 | * -1 if the reshape is finished because it is all done, |
2470 | * -2 if the reshape is finished due to an error. | |
7443ee81 N |
2471 | */ |
2472 | ||
2473 | int advancing = (reshape->after.data_disks | |
2474 | >= reshape->before.data_disks); | |
38dad34a N |
2475 | unsigned long long need_backup; /* All data between start of array and |
2476 | * here will at some point need to | |
2477 | * be backed up. | |
d0ab945e | 2478 | */ |
7443ee81 | 2479 | unsigned long long read_offset, write_offset; |
b4f8e38b | 2480 | unsigned long long write_range; |
7443ee81 | 2481 | unsigned long long max_progress, target, completed; |
d0ab945e N |
2482 | unsigned long long array_size = (info->component_size |
2483 | * reshape->before.data_disks); | |
7443ee81 | 2484 | int fd; |
77a73a17 | 2485 | char buf[20]; |
7443ee81 N |
2486 | |
2487 | /* First, we unsuspend any region that is now known to be safe. | |
2488 | * If suspend_point is on the 'wrong' side of reshape_progress, then | |
2489 | * we don't have or need suspension at the moment. This is true for | |
2490 | * native metadata when we don't need to back-up. | |
2491 | */ | |
2492 | if (advancing) { | |
49cd738b | 2493 | if (info->reshape_progress <= *suspend_point) |
7443ee81 N |
2494 | sysfs_set_num(info, NULL, "suspend_lo", |
2495 | info->reshape_progress); | |
2496 | } else { | |
2497 | /* Note: this won't work in 2.6.37 and before. | |
2498 | * Something somewhere should make sure we don't need it! | |
2499 | */ | |
49cd738b | 2500 | if (info->reshape_progress >= *suspend_point) |
7443ee81 N |
2501 | sysfs_set_num(info, NULL, "suspend_hi", |
2502 | info->reshape_progress); | |
2503 | } | |
2504 | ||
2505 | /* Now work out how far it is safe to progress. | |
2506 | * If the read_offset for ->reshape_progress is less than | |
2507 | * 'blocks' beyond the write_offset, we can only progress as far | |
2508 | * as a backup. | |
2509 | * Otherwise we can progress until the write_offset for the new location | |
2510 | * reaches (within 'blocks' of) the read_offset at the current location. | |
2511 | * However that region must be suspended unless we are using native | |
2512 | * metadata. | |
2513 | * If we need to suspend more, we limit it to 128M per device, which is | |
2514 | * rather arbitrary and should be some time-based calculation. | |
2515 | */ | |
ec757320 N |
2516 | read_offset = info->reshape_progress / reshape->before.data_disks; |
2517 | write_offset = info->reshape_progress / reshape->after.data_disks; | |
b4f8e38b | 2518 | write_range = info->new_chunk/512; |
38dad34a N |
2519 | if (reshape->before.data_disks == reshape->after.data_disks) |
2520 | need_backup = array_size; | |
2521 | else | |
2522 | need_backup = reshape->backup_blocks; | |
7443ee81 | 2523 | if (advancing) { |
38dad34a | 2524 | if (read_offset < write_offset + write_range) |
7443ee81 | 2525 | max_progress = backup_point; |
ddee071d | 2526 | else |
7443ee81 | 2527 | max_progress = |
2f3dd5e4 N |
2528 | read_offset * |
2529 | reshape->after.data_disks; | |
7443ee81 | 2530 | } else { |
38dad34a N |
2531 | if (read_offset > write_offset - write_range) |
2532 | /* Can only progress as far as has been backed up, | |
2533 | * which must be suspended */ | |
7443ee81 | 2534 | max_progress = backup_point; |
38dad34a N |
2535 | else if (info->reshape_progress <= need_backup) |
2536 | max_progress = backup_point; | |
2537 | else { | |
2538 | if (info->array.major_version >= 0) | |
2539 | /* Can progress until backup is needed */ | |
2540 | max_progress = need_backup; | |
2541 | else { | |
2542 | /* Can progress until metadata update is required */ | |
2543 | max_progress = | |
2544 | read_offset * | |
2545 | reshape->after.data_disks; | |
2546 | /* but data must be suspended */ | |
2547 | if (max_progress < *suspend_point) | |
2548 | max_progress = *suspend_point; | |
2549 | } | |
7443ee81 N |
2550 | } |
2551 | } | |
2552 | ||
2553 | /* We know it is safe to progress to 'max_progress' providing | |
2554 | * it is suspended or we are using native metadata. | |
2555 | * Consider extending suspend_point 128M per device if it | |
2556 | * is less than 64M per device beyond reshape_progress. | |
2557 | * But always do a multiple of 'blocks' | |
832b2b9a N |
2558 | * FIXME this is too big - it takes to long to complete |
2559 | * this much. | |
7443ee81 N |
2560 | */ |
2561 | target = 64*1024*2 * min(reshape->before.data_disks, | |
2562 | reshape->after.data_disks); | |
b4f8e38b | 2563 | target /= reshape->backup_blocks; |
7443ee81 N |
2564 | if (target < 2) |
2565 | target = 2; | |
b4f8e38b | 2566 | target *= reshape->backup_blocks; |
7443ee81 N |
2567 | |
2568 | /* For externally managed metadata we always need to suspend IO to | |
2569 | * the area being reshaped so we regularly push suspend_point forward. | |
2570 | * For native metadata we only need the suspend if we are going to do | |
2571 | * a backup. | |
2572 | */ | |
2573 | if (advancing) { | |
d0ab945e N |
2574 | if ((need_backup > info->reshape_progress |
2575 | || info->array.major_version < 0) && | |
7443ee81 | 2576 | *suspend_point < info->reshape_progress + target) { |
d0ab945e N |
2577 | if (need_backup < *suspend_point + 2 * target) |
2578 | *suspend_point = need_backup; | |
2579 | else if (*suspend_point + 2 * target < array_size) | |
7443ee81 | 2580 | *suspend_point += 2 * target; |
d0ab945e N |
2581 | else |
2582 | *suspend_point = array_size; | |
7443ee81 | 2583 | sysfs_set_num(info, NULL, "suspend_hi", *suspend_point); |
d0ab945e N |
2584 | if (max_progress > *suspend_point) |
2585 | max_progress = *suspend_point; | |
7443ee81 N |
2586 | } |
2587 | } else { | |
38dad34a N |
2588 | if (info->array.major_version >= 0) { |
2589 | /* Only need to suspend when about to backup */ | |
2590 | if (info->reshape_progress < need_backup * 2 && | |
2591 | *suspend_point > 0) { | |
d0ab945e | 2592 | *suspend_point = 0; |
38dad34a N |
2593 | sysfs_set_num(info, NULL, "suspend_lo", 0); |
2594 | sysfs_set_num(info, NULL, "suspend_hi", need_backup); | |
2595 | } | |
2596 | } else { | |
2597 | /* Need to suspend continually */ | |
2598 | if (info->reshape_progress < *suspend_point) | |
2599 | *suspend_point = info->reshape_progress; | |
2600 | if (*suspend_point + target < info->reshape_progress) | |
2601 | /* No need to move suspend region yet */; | |
2602 | else { | |
2603 | if (*suspend_point >= 2 * target) | |
2604 | *suspend_point -= 2 * target; | |
2605 | else | |
2606 | *suspend_point = 0; | |
2607 | sysfs_set_num(info, NULL, "suspend_lo", | |
2608 | *suspend_point); | |
2609 | } | |
d0ab945e N |
2610 | if (max_progress < *suspend_point) |
2611 | max_progress = *suspend_point; | |
7443ee81 N |
2612 | } |
2613 | } | |
2614 | ||
2615 | /* now set sync_max to allow that progress. sync_max, like | |
2616 | * sync_completed is a count of sectors written per device, so | |
2617 | * we find the difference between max_progress and the start point, | |
2618 | * and divide that by after.data_disks to get a sync_max | |
2619 | * number. | |
2620 | * At the same time we convert wait_point to a similar number | |
2621 | * for comparing against sync_completed. | |
2622 | */ | |
92d1991f | 2623 | /* scale down max_progress to per_disk */ |
7443ee81 | 2624 | max_progress /= reshape->after.data_disks; |
92d1991f N |
2625 | /* Round to chunk size as some kernels give an erroneously high number */ |
2626 | max_progress /= info->new_chunk/512; | |
2627 | max_progress *= info->new_chunk/512; | |
7f913e9b N |
2628 | /* And round to old chunk size as the kernel wants that */ |
2629 | max_progress /= info->array.chunk_size/512; | |
2630 | max_progress *= info->array.chunk_size/512; | |
92d1991f N |
2631 | /* Limit progress to the whole device */ |
2632 | if (max_progress > info->component_size) | |
2633 | max_progress = info->component_size; | |
7443ee81 | 2634 | wait_point /= reshape->after.data_disks; |
92d1991f N |
2635 | if (!advancing) { |
2636 | /* switch from 'device offset' to 'processed block count' */ | |
2637 | max_progress = info->component_size - max_progress; | |
2638 | wait_point = info->component_size - wait_point; | |
2639 | } | |
7443ee81 N |
2640 | |
2641 | sysfs_set_num(info, NULL, "sync_max", max_progress); | |
2642 | ||
2643 | /* Now wait. If we have already reached the point that we were | |
2644 | * asked to wait to, don't wait at all, else wait for any change. | |
2645 | * We need to select on 'sync_completed' as that is the place that | |
2646 | * notifications happen, but we are really interested in | |
2647 | * 'reshape_position' | |
2648 | */ | |
2649 | fd = sysfs_get_fd(info, NULL, "sync_completed"); | |
2650 | if (fd < 0) | |
77a73a17 | 2651 | goto check_progress; |
7443ee81 | 2652 | |
621ea11b | 2653 | if (sysfs_fd_get_ll(fd, &completed) < 0) |
77a73a17 | 2654 | goto check_progress; |
621ea11b | 2655 | |
7443ee81 N |
2656 | while (completed < max_progress && completed < wait_point) { |
2657 | /* Check that sync_action is still 'reshape' to avoid | |
2658 | * waiting forever on a dead array | |
2659 | */ | |
2660 | char action[20]; | |
2661 | fd_set rfds; | |
2662 | if (sysfs_get_str(info, NULL, "sync_action", | |
2663 | action, 20) <= 0 || | |
2664 | strncmp(action, "reshape", 7) != 0) | |
2665 | break; | |
26d6e157 AK |
2666 | /* Some kernels reset 'sync_completed' to zero |
2667 | * before setting 'sync_action' to 'idle'. | |
2668 | * So we need these extra tests. | |
2669 | */ | |
2670 | if (completed == 0 && advancing | |
2671 | && info->reshape_progress > 0) | |
2672 | break; | |
2673 | if (completed == 0 && !advancing | |
2674 | && info->reshape_progress < (info->component_size | |
2675 | * reshape->after.data_disks)) | |
2676 | break; | |
7443ee81 N |
2677 | FD_ZERO(&rfds); |
2678 | FD_SET(fd, &rfds); | |
2679 | select(fd+1, NULL, NULL, &rfds, NULL); | |
621ea11b | 2680 | if (sysfs_fd_get_ll(fd, &completed) < 0) |
77a73a17 | 2681 | goto check_progress; |
7443ee81 | 2682 | } |
10d0d365 AK |
2683 | /* Some kernels reset 'sync_completed' to zero, |
2684 | * we need to have real point we are in md | |
2685 | */ | |
2686 | if (completed == 0) | |
2687 | completed = max_progress; | |
2688 | ||
92d1991f N |
2689 | /* some kernels can give an incorrectly high 'completed' number */ |
2690 | completed /= (info->new_chunk/512); | |
2691 | completed *= (info->new_chunk/512); | |
7443ee81 N |
2692 | /* Convert 'completed' back in to a 'progress' number */ |
2693 | completed *= reshape->after.data_disks; | |
2694 | if (!advancing) { | |
2695 | completed = info->component_size * reshape->after.data_disks | |
2696 | - completed; | |
2697 | } | |
2698 | *reshape_completed = completed; | |
2699 | ||
2700 | close(fd); | |
2701 | ||
2702 | /* We return the need_backup flag. Caller will decide | |
d0ab945e | 2703 | * how much - a multiple of ->backup_blocks up to *suspend_point |
7443ee81 | 2704 | */ |
38dad34a N |
2705 | if (advancing) |
2706 | return need_backup > info->reshape_progress; | |
2707 | else | |
2708 | return need_backup >= info->reshape_progress; | |
77a73a17 N |
2709 | |
2710 | check_progress: | |
2711 | /* if we couldn't read a number from sync_completed, then | |
2712 | * either the reshape did complete, or it aborted. | |
2713 | * We can tell which by checking for 'none' in reshape_position. | |
621ea11b N |
2714 | * If it did abort, then it might immediately restart if it |
2715 | * it was just a device failure that leaves us degraded but | |
2716 | * functioning. | |
77a73a17 N |
2717 | */ |
2718 | strcpy(buf, "hi"); | |
2719 | if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0 | |
621ea11b N |
2720 | || strncmp(buf, "none", 4) != 0) { |
2721 | /* The abort might only be temporary. Wait up to 10 | |
2722 | * seconds for fd to contain a valid number again. | |
2723 | */ | |
2724 | struct timeval tv; | |
2725 | int rv = -2; | |
2726 | tv.tv_sec = 10; | |
2727 | tv.tv_usec = 0; | |
6560987b | 2728 | while (fd >= 0 && rv < 0 && tv.tv_sec > 0) { |
621ea11b N |
2729 | fd_set rfds; |
2730 | FD_ZERO(&rfds); | |
2731 | FD_SET(fd, &rfds); | |
2732 | if (select(fd+1, NULL, NULL, &rfds, &tv) != 1) | |
2733 | break; | |
6560987b N |
2734 | switch (sysfs_fd_get_ll(fd, &completed)) { |
2735 | case 0: | |
621ea11b N |
2736 | /* all good again */ |
2737 | rv = 1; | |
6560987b N |
2738 | break; |
2739 | case -2: /* read error - abort */ | |
2740 | tv.tv_sec = 0; | |
2741 | break; | |
2742 | } | |
621ea11b N |
2743 | } |
2744 | if (fd >= 0) | |
2745 | close(fd); | |
2746 | return rv; /* abort */ | |
2747 | } else { | |
6d5316f6 | 2748 | /* Maybe racing with array shutdown - check state */ |
621ea11b N |
2749 | if (fd >= 0) |
2750 | close(fd); | |
6d5316f6 N |
2751 | if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0 |
2752 | || strncmp(buf, "inactive", 8) == 0 | |
2753 | || strncmp(buf, "clear",5) == 0) | |
2754 | return -2; /* abort */ | |
77a73a17 | 2755 | return -1; /* complete */ |
6d5316f6 | 2756 | } |
7443ee81 N |
2757 | } |
2758 | ||
2759 | ||
fcf57625 | 2760 | /* FIXME return status is never checked */ |
4411fb17 | 2761 | static int grow_backup(struct mdinfo *sra, |
7236ee7a | 2762 | unsigned long long offset, /* per device */ |
7443ee81 | 2763 | unsigned long stripes, /* per device, in old chunks */ |
7236ee7a N |
2764 | int *sources, unsigned long long *offsets, |
2765 | int disks, int chunk, int level, int layout, | |
2766 | int dests, int *destfd, unsigned long long *destoffsets, | |
d4445387 | 2767 | int part, int *degraded, |
7236ee7a N |
2768 | char *buf) |
2769 | { | |
2770 | /* Backup 'blocks' sectors at 'offset' on each device of the array, | |
2771 | * to storage 'destfd' (offset 'destoffsets'), after first | |
2772 | * suspending IO. Then allow resync to continue | |
2773 | * over the suspended section. | |
2774 | * Use part 'part' of the backup-super-block. | |
2775 | */ | |
2776 | int odata = disks; | |
2777 | int rv = 0; | |
2778 | int i; | |
f21e18ca N |
2779 | unsigned long long ll; |
2780 | int new_degraded; | |
7236ee7a N |
2781 | //printf("offset %llu\n", offset); |
2782 | if (level >= 4) | |
2783 | odata--; | |
2784 | if (level == 6) | |
2785 | odata--; | |
7443ee81 | 2786 | |
d4445387 | 2787 | /* Check that array hasn't become degraded, else we might backup the wrong data */ |
2c6ac128 N |
2788 | if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0) |
2789 | return -1; /* FIXME this error is ignored */ | |
f21e18ca | 2790 | new_degraded = (int)ll; |
d4445387 N |
2791 | if (new_degraded != *degraded) { |
2792 | /* check each device to ensure it is still working */ | |
2793 | struct mdinfo *sd; | |
2794 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
2795 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2796 | continue; | |
2797 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
2798 | char sbuf[20]; | |
2799 | if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 || | |
2800 | strstr(sbuf, "faulty") || | |
2801 | strstr(sbuf, "in_sync") == NULL) { | |
2802 | /* this device is dead */ | |
2803 | sd->disk.state = (1<<MD_DISK_FAULTY); | |
2804 | if (sd->disk.raid_disk >= 0 && | |
2805 | sources[sd->disk.raid_disk] >= 0) { | |
2806 | close(sources[sd->disk.raid_disk]); | |
2807 | sources[sd->disk.raid_disk] = -1; | |
2808 | } | |
2809 | } | |
2810 | } | |
2811 | } | |
2812 | *degraded = new_degraded; | |
2813 | } | |
7236ee7a N |
2814 | if (part) { |
2815 | bsb.arraystart2 = __cpu_to_le64(offset * odata); | |
200871ad | 2816 | bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
2817 | } else { |
2818 | bsb.arraystart = __cpu_to_le64(offset * odata); | |
200871ad | 2819 | bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
2820 | } |
2821 | if (part) | |
2822 | bsb.magic[15] = '2'; | |
2823 | for (i = 0; i < dests; i++) | |
2824 | if (part) | |
2825 | lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0); | |
2826 | else | |
2827 | lseek64(destfd[i], destoffsets[i], 0); | |
2828 | ||
2829 | rv = save_stripes(sources, offsets, | |
2830 | disks, chunk, level, layout, | |
2831 | dests, destfd, | |
2832 | offset*512*odata, stripes * chunk * odata, | |
2833 | buf); | |
2834 | ||
2835 | if (rv) | |
2836 | return rv; | |
97396422 | 2837 | bsb.mtime = __cpu_to_le64(time(0)); |
7236ee7a N |
2838 | for (i = 0; i < dests; i++) { |
2839 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
2840 | ||
2841 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
2842 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
2843 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
2844 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
2845 | ||
72044953 | 2846 | rv = -1; |
f21e18ca N |
2847 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0) |
2848 | != destoffsets[i] - 4096) | |
72044953 N |
2849 | break; |
2850 | if (write(destfd[i], &bsb, 512) != 512) | |
2851 | break; | |
ff94fb86 | 2852 | if (destoffsets[i] > 4096) { |
f21e18ca | 2853 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) != |
a847575a | 2854 | destoffsets[i]+stripes*chunk*odata) |
72044953 N |
2855 | break; |
2856 | if (write(destfd[i], &bsb, 512) != 512) | |
2857 | break; | |
ff94fb86 | 2858 | } |
7236ee7a | 2859 | fsync(destfd[i]); |
72044953 | 2860 | rv = 0; |
7236ee7a | 2861 | } |
2efedc7b | 2862 | |
fcf57625 | 2863 | return rv; |
7236ee7a N |
2864 | } |
2865 | ||
2866 | /* in 2.6.30, the value reported by sync_completed can be | |
2867 | * less that it should be by one stripe. | |
2868 | * This only happens when reshape hits sync_max and pauses. | |
2869 | * So allow wait_backup to either extent sync_max further | |
2870 | * than strictly necessary, or return before the | |
2871 | * sync has got quite as far as we would really like. | |
2872 | * This is what 'blocks2' is for. | |
2873 | * The various caller give appropriate values so that | |
2874 | * every works. | |
2875 | */ | |
fcf57625 | 2876 | /* FIXME return value is often ignored */ |
7443ee81 | 2877 | static int forget_backup( |
7236ee7a N |
2878 | int dests, int *destfd, unsigned long long *destoffsets, |
2879 | int part) | |
2880 | { | |
7443ee81 N |
2881 | /* |
2882 | * Erase backup 'part' (which is 0 or 1) | |
7236ee7a | 2883 | */ |
7236ee7a | 2884 | int i; |
fcf57625 | 2885 | int rv; |
7236ee7a | 2886 | |
7236ee7a N |
2887 | if (part) { |
2888 | bsb.arraystart2 = __cpu_to_le64(0); | |
2889 | bsb.length2 = __cpu_to_le64(0); | |
2890 | } else { | |
2891 | bsb.arraystart = __cpu_to_le64(0); | |
2892 | bsb.length = __cpu_to_le64(0); | |
2893 | } | |
97396422 | 2894 | bsb.mtime = __cpu_to_le64(time(0)); |
fcf57625 | 2895 | rv = 0; |
7236ee7a N |
2896 | for (i = 0; i < dests; i++) { |
2897 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
2898 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
2899 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
2900 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
2901 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
f21e18ca | 2902 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) != |
a847575a | 2903 | destoffsets[i]-4096) |
72044953 N |
2904 | rv = -1; |
2905 | if (rv == 0 && | |
2906 | write(destfd[i], &bsb, 512) != 512) | |
2907 | rv = -1; | |
7236ee7a N |
2908 | fsync(destfd[i]); |
2909 | } | |
fcf57625 | 2910 | return rv; |
7236ee7a | 2911 | } |
e86c9dd6 | 2912 | |
7236ee7a N |
2913 | static void fail(char *msg) |
2914 | { | |
fcf57625 | 2915 | int rv; |
72044953 N |
2916 | rv = (write(2, msg, strlen(msg)) != (int)strlen(msg)); |
2917 | rv |= (write(2, "\n", 1) != 1); | |
fcf57625 | 2918 | exit(rv ? 1 : 2); |
7236ee7a N |
2919 | } |
2920 | ||
2921 | static char *abuf, *bbuf; | |
f21e18ca | 2922 | static unsigned long long abuflen; |
7236ee7a N |
2923 | static void validate(int afd, int bfd, unsigned long long offset) |
2924 | { | |
2925 | /* check that the data in the backup against the array. | |
2926 | * This is only used for regression testing and should not | |
2927 | * be used while the array is active | |
2928 | */ | |
7236ee7a N |
2929 | if (afd < 0) |
2930 | return; | |
2931 | lseek64(bfd, offset - 4096, 0); | |
2932 | if (read(bfd, &bsb2, 512) != 512) | |
2933 | fail("cannot read bsb"); | |
2934 | if (bsb2.sb_csum != bsb_csum((char*)&bsb2, | |
2935 | ((char*)&bsb2.sb_csum)-((char*)&bsb2))) | |
2936 | fail("first csum bad"); | |
2937 | if (memcmp(bsb2.magic, "md_backup_data", 14) != 0) | |
2938 | fail("magic is bad"); | |
2939 | if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 && | |
2940 | bsb2.sb_csum2 != bsb_csum((char*)&bsb2, | |
2941 | ((char*)&bsb2.sb_csum2)-((char*)&bsb2))) | |
2942 | fail("second csum bad"); | |
2943 | ||
2944 | if (__le64_to_cpu(bsb2.devstart)*512 != offset) | |
2945 | fail("devstart is wrong"); | |
2946 | ||
2947 | if (bsb2.length) { | |
2948 | unsigned long long len = __le64_to_cpu(bsb2.length)*512; | |
2949 | ||
2950 | if (abuflen < len) { | |
2951 | free(abuf); | |
2952 | free(bbuf); | |
2953 | abuflen = len; | |
fcf57625 N |
2954 | if (posix_memalign((void**)&abuf, 4096, abuflen) || |
2955 | posix_memalign((void**)&bbuf, 4096, abuflen)) { | |
2956 | abuflen = 0; | |
2957 | /* just stop validating on mem-alloc failure */ | |
2958 | return; | |
2959 | } | |
e86c9dd6 | 2960 | } |
48924014 | 2961 | |
7236ee7a | 2962 | lseek64(bfd, offset, 0); |
f21e18ca | 2963 | if ((unsigned long long)read(bfd, bbuf, len) != len) { |
080fd005 | 2964 | //printf("len %llu\n", len); |
7236ee7a N |
2965 | fail("read first backup failed"); |
2966 | } | |
2967 | lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0); | |
f21e18ca | 2968 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
2969 | fail("read first from array failed"); |
2970 | if (memcmp(bbuf, abuf, len) != 0) { | |
080fd005 | 2971 | #if 0 |
7236ee7a N |
2972 | int i; |
2973 | printf("offset=%llu len=%llu\n", | |
080fd005 | 2974 | (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len); |
7236ee7a N |
2975 | for (i=0; i<len; i++) |
2976 | if (bbuf[i] != abuf[i]) { | |
2977 | printf("first diff byte %d\n", i); | |
93ecfa01 | 2978 | break; |
7236ee7a | 2979 | } |
080fd005 | 2980 | #endif |
7236ee7a | 2981 | fail("data1 compare failed"); |
e86c9dd6 | 2982 | } |
7236ee7a N |
2983 | } |
2984 | if (bsb2.length2) { | |
2985 | unsigned long long len = __le64_to_cpu(bsb2.length2)*512; | |
2986 | ||
2987 | if (abuflen < len) { | |
2988 | free(abuf); | |
2989 | free(bbuf); | |
2990 | abuflen = len; | |
2991 | abuf = malloc(abuflen); | |
2992 | bbuf = malloc(abuflen); | |
e86c9dd6 NB |
2993 | } |
2994 | ||
7236ee7a | 2995 | lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0); |
f21e18ca | 2996 | if ((unsigned long long)read(bfd, bbuf, len) != len) |
7236ee7a N |
2997 | fail("read second backup failed"); |
2998 | lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0); | |
f21e18ca | 2999 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
3000 | fail("read second from array failed"); |
3001 | if (memcmp(bbuf, abuf, len) != 0) | |
3002 | fail("data2 compare failed"); | |
e86c9dd6 | 3003 | } |
7236ee7a | 3004 | } |
e86c9dd6 | 3005 | |
999b4972 N |
3006 | int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape, |
3007 | struct supertype *st, unsigned long blocks, | |
3008 | int *fds, unsigned long long *offsets, | |
3009 | int dests, int *destfd, unsigned long long *destoffsets) | |
7236ee7a | 3010 | { |
7443ee81 N |
3011 | /* Monitor a reshape where backup is being performed using |
3012 | * 'native' mechanism - either to a backup file, or | |
3013 | * to some space in a spare. | |
3014 | */ | |
7236ee7a | 3015 | char *buf; |
7443ee81 N |
3016 | int degraded = -1; |
3017 | unsigned long long speed; | |
3018 | unsigned long long suspend_point, array_size; | |
3019 | unsigned long long backup_point, wait_point; | |
3020 | unsigned long long reshape_completed; | |
3021 | int done = 0; | |
3022 | int increasing = reshape->after.data_disks >= reshape->before.data_disks; | |
3023 | int part = 0; /* The next part of the backup area to fill. It may already | |
3024 | * be full, so we need to check */ | |
3025 | int level = reshape->level; | |
3026 | int layout = reshape->before.layout; | |
3027 | int data = reshape->before.data_disks; | |
3028 | int disks = reshape->before.data_disks + reshape->parity; | |
3029 | int chunk = sra->array.chunk_size; | |
da8100ca N |
3030 | struct mdinfo *sd; |
3031 | unsigned long stripes; | |
3b7e9d0c | 3032 | int uuid[4]; |
da8100ca N |
3033 | |
3034 | /* set up the backup-super-block. This requires the | |
3035 | * uuid from the array. | |
3036 | */ | |
3037 | /* Find a superblock */ | |
3038 | for (sd = sra->devs; sd; sd = sd->next) { | |
3039 | char *dn; | |
3040 | int devfd; | |
3041 | int ok; | |
3042 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
3043 | continue; | |
3044 | dn = map_dev(sd->disk.major, sd->disk.minor, 1); | |
3045 | devfd = dev_open(dn, O_RDONLY); | |
3046 | if (devfd < 0) | |
3047 | continue; | |
3048 | ok = st->ss->load_super(st, devfd, NULL); | |
3049 | close(devfd); | |
77f8d358 | 3050 | if (ok == 0) |
da8100ca N |
3051 | break; |
3052 | } | |
3053 | if (!sd) { | |
3054 | fprintf(stderr, Name ": Cannot find a superblock\n"); | |
3055 | return 0; | |
3056 | } | |
3057 | ||
3058 | memset(&bsb, 0, 512); | |
3059 | memcpy(bsb.magic, "md_backup_data-1", 16); | |
3b7e9d0c LB |
3060 | st->ss->uuid_from_super(st, uuid); |
3061 | memcpy(bsb.set_uuid, uuid, 16); | |
da8100ca N |
3062 | bsb.mtime = __cpu_to_le64(time(0)); |
3063 | bsb.devstart2 = blocks; | |
3064 | ||
3065 | stripes = blocks / (sra->array.chunk_size/512) / | |
3066 | reshape->before.data_disks; | |
e86c9dd6 | 3067 | |
fcf57625 N |
3068 | if (posix_memalign((void**)&buf, 4096, disks * chunk)) |
3069 | /* Don't start the 'reshape' */ | |
3070 | return 0; | |
7443ee81 N |
3071 | if (reshape->before.data_disks == reshape->after.data_disks) { |
3072 | sysfs_get_ll(sra, NULL, "sync_speed_min", &speed); | |
3073 | sysfs_set_num(sra, NULL, "sync_speed_min", 200000); | |
3074 | } | |
e86c9dd6 | 3075 | |
7443ee81 | 3076 | if (increasing) { |
96533072 | 3077 | array_size = sra->component_size * reshape->after.data_disks; |
7443ee81 N |
3078 | backup_point = sra->reshape_progress; |
3079 | suspend_point = 0; | |
3080 | } else { | |
96533072 | 3081 | array_size = sra->component_size * reshape->before.data_disks; |
25da62d9 | 3082 | backup_point = reshape->backup_blocks; |
7443ee81 N |
3083 | suspend_point = array_size; |
3084 | } | |
7236ee7a | 3085 | |
7443ee81 N |
3086 | while (!done) { |
3087 | int rv; | |
7236ee7a | 3088 | |
7443ee81 N |
3089 | /* Want to return as soon the oldest backup slot can |
3090 | * be released as that allows us to start backing up | |
3091 | * some more, providing suspend_point has been | |
b6b95155 | 3092 | * advanced, which it should have. |
7443ee81 N |
3093 | */ |
3094 | if (increasing) { | |
3095 | wait_point = array_size; | |
3096 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
3097 | wait_point = (__le64_to_cpu(bsb.arraystart) + | |
3098 | __le64_to_cpu(bsb.length)); | |
3099 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
3100 | wait_point = (__le64_to_cpu(bsb.arraystart2) + | |
3101 | __le64_to_cpu(bsb.length2)); | |
3102 | } else { | |
3103 | wait_point = 0; | |
3104 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
3105 | wait_point = __le64_to_cpu(bsb.arraystart); | |
3106 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
3107 | wait_point = __le64_to_cpu(bsb.arraystart2); | |
3108 | } | |
3109 | ||
3110 | rv = progress_reshape(sra, reshape, | |
3111 | backup_point, wait_point, | |
3112 | &suspend_point, &reshape_completed); | |
7443ee81 N |
3113 | /* external metadata would need to ping_monitor here */ |
3114 | sra->reshape_progress = reshape_completed; | |
7236ee7a | 3115 | |
7443ee81 N |
3116 | /* Clear any backup region that is before 'here' */ |
3117 | if (increasing) { | |
b3cf095a N |
3118 | if (__le64_to_cpu(bsb.length) > 0 && |
3119 | reshape_completed >= (__le64_to_cpu(bsb.arraystart) + | |
7443ee81 N |
3120 | __le64_to_cpu(bsb.length))) |
3121 | forget_backup(dests, destfd, | |
3122 | destoffsets, 0); | |
b3cf095a N |
3123 | if (__le64_to_cpu(bsb.length2) > 0 && |
3124 | reshape_completed >= (__le64_to_cpu(bsb.arraystart2) + | |
7443ee81 N |
3125 | __le64_to_cpu(bsb.length2))) |
3126 | forget_backup(dests, destfd, | |
3127 | destoffsets, 1); | |
3128 | } else { | |
b3cf095a N |
3129 | if (__le64_to_cpu(bsb.length) > 0 && |
3130 | reshape_completed <= (__le64_to_cpu(bsb.arraystart))) | |
7443ee81 N |
3131 | forget_backup(dests, destfd, |
3132 | destoffsets, 0); | |
b3cf095a N |
3133 | if (__le64_to_cpu(bsb.length2) > 0 && |
3134 | reshape_completed <= (__le64_to_cpu(bsb.arraystart2))) | |
7443ee81 N |
3135 | forget_backup(dests, destfd, |
3136 | destoffsets, 1); | |
3137 | } | |
7236ee7a | 3138 | |
223c5c99 | 3139 | if (rv < 0) { |
77a73a17 N |
3140 | if (rv == -1) |
3141 | done = 1; | |
223c5c99 N |
3142 | break; |
3143 | } | |
2d4de5f9 N |
3144 | if (rv == 0 && increasing && !st->ss->external) { |
3145 | /* No longer need to monitor this reshape */ | |
3146 | done = 1; | |
3147 | break; | |
3148 | } | |
223c5c99 | 3149 | |
60204e4e | 3150 | while (rv) { |
7443ee81 | 3151 | unsigned long long offset; |
e97ca002 | 3152 | unsigned long actual_stripes; |
60204e4e N |
3153 | /* Need to backup some data. |
3154 | * If 'part' is not used and the desired | |
3155 | * backup size is suspended, do a backup, | |
3156 | * then consider the next part. | |
3157 | */ | |
7443ee81 N |
3158 | /* Check that 'part' is unused */ |
3159 | if (part == 0 && __le64_to_cpu(bsb.length) != 0) | |
60204e4e | 3160 | break; |
7443ee81 | 3161 | if (part == 1 && __le64_to_cpu(bsb.length2) != 0) |
60204e4e | 3162 | break; |
7443ee81 N |
3163 | |
3164 | offset = backup_point / data; | |
e97ca002 | 3165 | actual_stripes = stripes; |
60204e4e | 3166 | if (increasing) { |
e97ca002 N |
3167 | if (offset + actual_stripes * (chunk/512) > |
3168 | sra->component_size) | |
3169 | actual_stripes = ((sra->component_size - offset) | |
3170 | / (chunk/512)); | |
3171 | if (offset + actual_stripes * (chunk/512) > | |
60204e4e N |
3172 | suspend_point/data) |
3173 | break; | |
3174 | } else { | |
e97ca002 N |
3175 | if (offset < actual_stripes * (chunk/512)) |
3176 | actual_stripes = offset / (chunk/512); | |
3177 | offset -= actual_stripes * (chunk/512); | |
3178 | if (offset < suspend_point/data) | |
60204e4e N |
3179 | break; |
3180 | } | |
e4b11073 N |
3181 | if (actual_stripes == 0) |
3182 | break; | |
e97ca002 | 3183 | grow_backup(sra, offset, actual_stripes, |
7443ee81 N |
3184 | fds, offsets, |
3185 | disks, chunk, level, layout, | |
3186 | dests, destfd, destoffsets, | |
3187 | part, °raded, buf); | |
3188 | validate(afd, destfd[0], destoffsets[0]); | |
3189 | /* record where 'part' is up to */ | |
3190 | part = !part; | |
3191 | if (increasing) | |
e97ca002 | 3192 | backup_point += actual_stripes * (chunk/512) * data; |
7443ee81 | 3193 | else |
e97ca002 | 3194 | backup_point -= actual_stripes * (chunk/512) * data; |
7443ee81 N |
3195 | } |
3196 | } | |
3197 | ||
223c5c99 N |
3198 | /* FIXME maybe call progress_reshape one more time instead */ |
3199 | abort_reshape(sra); /* remove any remaining suspension */ | |
7443ee81 N |
3200 | if (reshape->before.data_disks == reshape->after.data_disks) |
3201 | sysfs_set_num(sra, NULL, "sync_speed_min", speed); | |
7236ee7a | 3202 | free(buf); |
77a73a17 | 3203 | return done; |
e86c9dd6 | 3204 | } |
353632d9 NB |
3205 | |
3206 | /* | |
3207 | * If any spare contains md_back_data-1 which is recent wrt mtime, | |
3208 | * write that data into the array and update the super blocks with | |
3209 | * the new reshape_progress | |
3210 | */ | |
ea0ebe96 N |
3211 | int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt, |
3212 | char *backup_file, int verbose) | |
353632d9 NB |
3213 | { |
3214 | int i, j; | |
3215 | int old_disks; | |
353632d9 | 3216 | unsigned long long *offsets; |
82f2d6ab | 3217 | unsigned long long nstripe, ostripe; |
6e9eac4f | 3218 | int ndata, odata; |
353632d9 | 3219 | |
e9e43ec3 N |
3220 | odata = info->array.raid_disks - info->delta_disks - 1; |
3221 | if (info->array.level == 6) odata--; /* number of data disks */ | |
3222 | ndata = info->array.raid_disks - 1; | |
3223 | if (info->new_level == 6) ndata--; | |
353632d9 NB |
3224 | |
3225 | old_disks = info->array.raid_disks - info->delta_disks; | |
3226 | ||
e9e43ec3 N |
3227 | if (info->delta_disks <= 0) |
3228 | /* Didn't grow, so the backup file must have | |
3229 | * been used | |
3230 | */ | |
3231 | old_disks = cnt; | |
06b0d786 | 3232 | for (i=old_disks-(backup_file?1:0); i<cnt; i++) { |
353632d9 | 3233 | struct mdinfo dinfo; |
06b0d786 | 3234 | int fd; |
e9e43ec3 | 3235 | int bsbsize; |
ea0ebe96 | 3236 | char *devname, namebuf[20]; |
10af14c4 | 3237 | unsigned long long lo, hi; |
353632d9 NB |
3238 | |
3239 | /* This was a spare and may have some saved data on it. | |
3240 | * Load the superblock, find and load the | |
3241 | * backup_super_block. | |
3242 | * If either fail, go on to next device. | |
3243 | * If the backup contains no new info, just return | |
206c5eae | 3244 | * else restore data and update all superblocks |
353632d9 | 3245 | */ |
06b0d786 NB |
3246 | if (i == old_disks-1) { |
3247 | fd = open(backup_file, O_RDONLY); | |
e9e43ec3 N |
3248 | if (fd<0) { |
3249 | fprintf(stderr, Name ": backup file %s inaccessible: %s\n", | |
3250 | backup_file, strerror(errno)); | |
06b0d786 | 3251 | continue; |
e9e43ec3 | 3252 | } |
ea0ebe96 | 3253 | devname = backup_file; |
06b0d786 NB |
3254 | } else { |
3255 | fd = fdlist[i]; | |
3256 | if (fd < 0) | |
3257 | continue; | |
3da92f27 | 3258 | if (st->ss->load_super(st, fd, NULL)) |
06b0d786 | 3259 | continue; |
353632d9 | 3260 | |
a5d85af7 | 3261 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 NB |
3262 | st->ss->free_super(st); |
3263 | ||
06b0d786 NB |
3264 | if (lseek64(fd, |
3265 | (dinfo.data_offset + dinfo.component_size - 8) <<9, | |
ea0ebe96 N |
3266 | 0) < 0) { |
3267 | fprintf(stderr, Name ": Cannot seek on device %d\n", i); | |
06b0d786 | 3268 | continue; /* Cannot seek */ |
ea0ebe96 N |
3269 | } |
3270 | sprintf(namebuf, "device-%d", i); | |
3271 | devname = namebuf; | |
06b0d786 | 3272 | } |
ea0ebe96 N |
3273 | if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) { |
3274 | if (verbose) | |
3275 | fprintf(stderr, Name ": Cannot read from %s\n", devname); | |
353632d9 | 3276 | continue; /* Cannot read */ |
ea0ebe96 | 3277 | } |
e9e43ec3 | 3278 | if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 && |
ea0ebe96 N |
3279 | memcmp(bsb.magic, "md_backup_data-2", 16) != 0) { |
3280 | if (verbose) | |
3281 | fprintf(stderr, Name ": No backup metadata on %s\n", devname); | |
353632d9 | 3282 | continue; |
ea0ebe96 N |
3283 | } |
3284 | if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) { | |
3285 | if (verbose) | |
3286 | fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname); | |
353632d9 | 3287 | continue; /* bad checksum */ |
ea0ebe96 | 3288 | } |
e9e43ec3 | 3289 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 && |
ea0ebe96 N |
3290 | bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) { |
3291 | if (verbose) | |
3292 | fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname); | |
22e30516 | 3293 | continue; /* Bad second checksum */ |
ea0ebe96 N |
3294 | } |
3295 | if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) { | |
3296 | if (verbose) | |
3297 | fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname); | |
353632d9 | 3298 | continue; /* Wrong uuid */ |
ea0ebe96 | 3299 | } |
353632d9 | 3300 | |
097075b6 N |
3301 | /* array utime and backup-mtime should be updated at much the same time, but it seems that |
3302 | * sometimes they aren't... So allow considerable flexability in matching, and allow | |
3303 | * this test to be overridden by an environment variable. | |
3304 | */ | |
f21e18ca N |
3305 | if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 || |
3306 | info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) { | |
097075b6 N |
3307 | if (check_env("MDADM_GROW_ALLOW_OLD")) { |
3308 | fprintf(stderr, Name ": accepting backup with timestamp %lu " | |
3309 | "for array with timestamp %lu\n", | |
3310 | (unsigned long)__le64_to_cpu(bsb.mtime), | |
3311 | (unsigned long)info->array.utime); | |
3312 | } else { | |
3313 | if (verbose) | |
3314 | fprintf(stderr, Name ": too-old timestamp on " | |
3315 | "backup-metadata on %s\n", devname); | |
3316 | continue; /* time stamp is too bad */ | |
3317 | } | |
ea0ebe96 | 3318 | } |
353632d9 | 3319 | |
e9e43ec3 | 3320 | if (bsb.magic[15] == '1') { |
10af14c4 N |
3321 | if (bsb.length == 0) |
3322 | continue; | |
e7a71c6b N |
3323 | if (info->delta_disks >= 0) { |
3324 | /* reshape_progress is increasing */ | |
3325 | if (__le64_to_cpu(bsb.arraystart) | |
3326 | + __le64_to_cpu(bsb.length) | |
3327 | < info->reshape_progress) { | |
3328 | nonew: | |
3329 | if (verbose) | |
3330 | fprintf(stderr, Name | |
3331 | ": backup-metadata found on %s but is not needed\n", devname); | |
3332 | continue; /* No new data here */ | |
3333 | } | |
3334 | } else { | |
3335 | /* reshape_progress is decreasing */ | |
3336 | if (__le64_to_cpu(bsb.arraystart) >= | |
3337 | info->reshape_progress) | |
3338 | goto nonew; /* No new data here */ | |
ea0ebe96 | 3339 | } |
e9e43ec3 | 3340 | } else { |
10af14c4 N |
3341 | if (bsb.length == 0 && bsb.length2 == 0) |
3342 | continue; | |
e7a71c6b N |
3343 | if (info->delta_disks >= 0) { |
3344 | /* reshape_progress is increasing */ | |
3345 | if ((__le64_to_cpu(bsb.arraystart) | |
3346 | + __le64_to_cpu(bsb.length) | |
3347 | < info->reshape_progress) | |
3348 | && | |
3349 | (__le64_to_cpu(bsb.arraystart2) | |
3350 | + __le64_to_cpu(bsb.length2) | |
3351 | < info->reshape_progress)) | |
3352 | goto nonew; /* No new data here */ | |
3353 | } else { | |
3354 | /* reshape_progress is decreasing */ | |
3355 | if (__le64_to_cpu(bsb.arraystart) >= | |
3356 | info->reshape_progress && | |
3357 | __le64_to_cpu(bsb.arraystart2) >= | |
3358 | info->reshape_progress) | |
3359 | goto nonew; /* No new data here */ | |
3360 | } | |
e9e43ec3 | 3361 | } |
ea0ebe96 N |
3362 | if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) { |
3363 | second_fail: | |
3364 | if (verbose) | |
e7a71c6b N |
3365 | fprintf(stderr, Name |
3366 | ": Failed to verify secondary backup-metadata block on %s\n", | |
ea0ebe96 | 3367 | devname); |
353632d9 | 3368 | continue; /* Cannot seek */ |
ea0ebe96 | 3369 | } |
2efedc7b | 3370 | /* There should be a duplicate backup superblock 4k before here */ |
06b0d786 | 3371 | if (lseek64(fd, -4096, 1) < 0 || |
0155af90 | 3372 | read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2)) |
ea0ebe96 | 3373 | goto second_fail; /* Cannot find leading superblock */ |
e9e43ec3 N |
3374 | if (bsb.magic[15] == '1') |
3375 | bsbsize = offsetof(struct mdp_backup_super, pad1); | |
3376 | else | |
3377 | bsbsize = offsetof(struct mdp_backup_super, pad); | |
ff94fb86 | 3378 | if (memcmp(&bsb2, &bsb, bsbsize) != 0) |
ea0ebe96 | 3379 | goto second_fail; /* Cannot find leading superblock */ |
2efedc7b | 3380 | |
353632d9 NB |
3381 | /* Now need the data offsets for all devices. */ |
3382 | offsets = malloc(sizeof(*offsets)*info->array.raid_disks); | |
3383 | for(j=0; j<info->array.raid_disks; j++) { | |
3384 | if (fdlist[j] < 0) | |
3385 | continue; | |
3da92f27 | 3386 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 NB |
3387 | /* FIXME should be this be an error */ |
3388 | continue; | |
a5d85af7 | 3389 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 | 3390 | st->ss->free_super(st); |
14e5b4d7 | 3391 | offsets[j] = dinfo.data_offset * 512; |
353632d9 NB |
3392 | } |
3393 | printf(Name ": restoring critical section\n"); | |
3394 | ||
3395 | if (restore_stripes(fdlist, offsets, | |
3396 | info->array.raid_disks, | |
3397 | info->new_chunk, | |
3398 | info->new_level, | |
3399 | info->new_layout, | |
06b0d786 | 3400 | fd, __le64_to_cpu(bsb.devstart)*512, |
92dcdf7c | 3401 | __le64_to_cpu(bsb.arraystart)*512, |
2fcb75ae | 3402 | __le64_to_cpu(bsb.length)*512, NULL)) { |
e9e43ec3 | 3403 | /* didn't succeed, so giveup */ |
ea0ebe96 N |
3404 | if (verbose) |
3405 | fprintf(stderr, Name ": Error restoring backup from %s\n", | |
3406 | devname); | |
e9e43ec3 N |
3407 | return 1; |
3408 | } | |
3409 | ||
3410 | if (bsb.magic[15] == '2' && | |
3411 | restore_stripes(fdlist, offsets, | |
3412 | info->array.raid_disks, | |
3413 | info->new_chunk, | |
3414 | info->new_level, | |
3415 | info->new_layout, | |
3416 | fd, __le64_to_cpu(bsb.devstart)*512 + | |
3417 | __le64_to_cpu(bsb.devstart2)*512, | |
92dcdf7c | 3418 | __le64_to_cpu(bsb.arraystart2)*512, |
2fcb75ae | 3419 | __le64_to_cpu(bsb.length2)*512, NULL)) { |
353632d9 | 3420 | /* didn't succeed, so giveup */ |
ea0ebe96 N |
3421 | if (verbose) |
3422 | fprintf(stderr, Name ": Error restoring second backup from %s\n", | |
3423 | devname); | |
2295250a | 3424 | return 1; |
353632d9 NB |
3425 | } |
3426 | ||
e9e43ec3 | 3427 | |
353632d9 NB |
3428 | /* Ok, so the data is restored. Let's update those superblocks. */ |
3429 | ||
10af14c4 N |
3430 | lo = hi = 0; |
3431 | if (bsb.length) { | |
3432 | lo = __le64_to_cpu(bsb.arraystart); | |
3433 | hi = lo + __le64_to_cpu(bsb.length); | |
3434 | } | |
3435 | if (bsb.magic[15] == '2' && bsb.length2) { | |
3436 | unsigned long long lo1, hi1; | |
3437 | lo1 = __le64_to_cpu(bsb.arraystart2); | |
3438 | hi1 = lo1 + __le64_to_cpu(bsb.length2); | |
3439 | if (lo == hi) { | |
3440 | lo = lo1; | |
3441 | hi = hi1; | |
3442 | } else if (lo < lo1) | |
3443 | hi = hi1; | |
3444 | else | |
3445 | lo = lo1; | |
3446 | } | |
3447 | if (lo < hi && | |
3448 | (info->reshape_progress < lo || | |
3449 | info->reshape_progress > hi)) | |
3450 | /* backup does not affect reshape_progress*/ ; | |
3451 | else if (info->delta_disks >= 0) { | |
e9e43ec3 N |
3452 | info->reshape_progress = __le64_to_cpu(bsb.arraystart) + |
3453 | __le64_to_cpu(bsb.length); | |
3454 | if (bsb.magic[15] == '2') { | |
3455 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) + | |
3456 | __le64_to_cpu(bsb.length2); | |
3457 | if (p2 > info->reshape_progress) | |
3458 | info->reshape_progress = p2; | |
3459 | } | |
3460 | } else { | |
3461 | info->reshape_progress = __le64_to_cpu(bsb.arraystart); | |
3462 | if (bsb.magic[15] == '2') { | |
3463 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2); | |
3464 | if (p2 < info->reshape_progress) | |
3465 | info->reshape_progress = p2; | |
3466 | } | |
3467 | } | |
353632d9 NB |
3468 | for (j=0; j<info->array.raid_disks; j++) { |
3469 | if (fdlist[j] < 0) continue; | |
3da92f27 | 3470 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 | 3471 | continue; |
a5d85af7 | 3472 | st->ss->getinfo_super(st, &dinfo, NULL); |
e9e43ec3 | 3473 | dinfo.reshape_progress = info->reshape_progress; |
3da92f27 | 3474 | st->ss->update_super(st, &dinfo, |
68c7d6d7 NB |
3475 | "_reshape_progress", |
3476 | NULL,0, 0, NULL); | |
3da92f27 NB |
3477 | st->ss->store_super(st, fdlist[j]); |
3478 | st->ss->free_super(st); | |
353632d9 | 3479 | } |
353632d9 NB |
3480 | return 0; |
3481 | } | |
6e9eac4f NB |
3482 | /* Didn't find any backup data, try to see if any |
3483 | * was needed. | |
3484 | */ | |
82f2d6ab N |
3485 | if (info->delta_disks < 0) { |
3486 | /* When shrinking, the critical section is at the end. | |
3487 | * So see if we are before the critical section. | |
3488 | */ | |
3489 | unsigned long long first_block; | |
3490 | nstripe = ostripe = 0; | |
3491 | first_block = 0; | |
3492 | while (ostripe >= nstripe) { | |
3493 | ostripe += info->array.chunk_size / 512; | |
3494 | first_block = ostripe * odata; | |
3495 | nstripe = first_block / ndata / (info->new_chunk/512) * | |
3496 | (info->new_chunk/512); | |
3497 | } | |
3498 | ||
3499 | if (info->reshape_progress >= first_block) | |
3500 | return 0; | |
6e9eac4f | 3501 | } |
82f2d6ab N |
3502 | if (info->delta_disks > 0) { |
3503 | /* See if we are beyond the critical section. */ | |
3504 | unsigned long long last_block; | |
3505 | nstripe = ostripe = 0; | |
3506 | last_block = 0; | |
3507 | while (nstripe >= ostripe) { | |
3508 | nstripe += info->new_chunk / 512; | |
3509 | last_block = nstripe * ndata; | |
3510 | ostripe = last_block / odata / (info->array.chunk_size/512) * | |
3511 | (info->array.chunk_size/512); | |
3512 | } | |
6e9eac4f | 3513 | |
82f2d6ab N |
3514 | if (info->reshape_progress >= last_block) |
3515 | return 0; | |
3516 | } | |
6e9eac4f | 3517 | /* needed to recover critical section! */ |
ea0ebe96 N |
3518 | if (verbose) |
3519 | fprintf(stderr, Name ": Failed to find backup of critical section\n"); | |
2295250a | 3520 | return 1; |
353632d9 | 3521 | } |
e9e43ec3 N |
3522 | |
3523 | int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info, | |
3524 | char *backup_file) | |
3525 | { | |
864a004f AK |
3526 | char buf[40]; |
3527 | char *container = NULL; | |
ef5414b2 | 3528 | int err; |
864a004f | 3529 | |
20a40eca N |
3530 | err = sysfs_set_str(info, NULL, "array_state", "readonly"); |
3531 | if (err) | |
3532 | return err; | |
3533 | if (st->ss->external) { | |
864a004f AK |
3534 | fmt_devname(buf, st->container_dev); |
3535 | container = buf; | |
493f5dd6 | 3536 | freeze(st); |
3db2fdd8 | 3537 | |
f362d22b N |
3538 | if (!mdmon_running(st->container_dev)) |
3539 | start_mdmon(st->container_dev); | |
983fff45 | 3540 | ping_monitor_by_id(st->container_dev); |
f362d22b N |
3541 | |
3542 | ||
3db2fdd8 AK |
3543 | if (info->reshape_active == 2) { |
3544 | int cfd = open_dev(st->container_dev); | |
3545 | if (cfd < 0) | |
3546 | return 1; | |
3547 | st->ss->load_container(st, cfd, container); | |
3548 | close(cfd); | |
493f5dd6 N |
3549 | return reshape_container(container, NULL, |
3550 | st, info, 0, backup_file, | |
3551 | 0, 1); | |
3db2fdd8 | 3552 | } |
864a004f AK |
3553 | } |
3554 | return reshape_array(container, mdfd, "array", st, info, 1, | |
e2e53a2d | 3555 | NULL, backup_file, 0, 0, 1); |
e9e43ec3 | 3556 | } |