]>
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; | |
524 | ||
525 | if (!sra) | |
526 | return -1; | |
527 | err = sysfs_freeze_array(sra); | |
528 | sysfs_free(sra); | |
529 | return err; | |
530 | } | |
7236ee7a N |
531 | } |
532 | ||
9202b8eb | 533 | static void unfreeze(struct supertype *st) |
7236ee7a | 534 | { |
7f2ba464 DW |
535 | if (st->ss->external) |
536 | return unfreeze_container(st); | |
537 | else { | |
538 | struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION); | |
539 | ||
540 | if (sra) | |
541 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
542 | else | |
543 | fprintf(stderr, Name ": failed to unfreeze array\n"); | |
544 | sysfs_free(sra); | |
545 | } | |
7236ee7a N |
546 | } |
547 | ||
4411fb17 | 548 | static void wait_reshape(struct mdinfo *sra) |
7236ee7a N |
549 | { |
550 | int fd = sysfs_get_fd(sra, NULL, "sync_action"); | |
551 | char action[20]; | |
552 | ||
92a19f1a AK |
553 | if (fd < 0) |
554 | return; | |
555 | ||
556 | while (sysfs_fd_get_str(fd, action, 20) > 0 && | |
557 | strncmp(action, "reshape", 7) == 0) { | |
7236ee7a N |
558 | fd_set rfds; |
559 | FD_ZERO(&rfds); | |
560 | FD_SET(fd, &rfds); | |
561 | select(fd+1, NULL, NULL, &rfds, NULL); | |
92a19f1a AK |
562 | } |
563 | close(fd); | |
7236ee7a | 564 | } |
7bc71196 DW |
565 | |
566 | static int reshape_super(struct supertype *st, long long size, int level, | |
567 | int layout, int chunksize, int raid_disks, | |
568 | char *backup_file, char *dev, int verbose) | |
569 | { | |
570 | /* nothing extra to check in the native case */ | |
571 | if (!st->ss->external) | |
572 | return 0; | |
573 | if (!st->ss->reshape_super || | |
574 | !st->ss->manage_reshape) { | |
575 | fprintf(stderr, Name ": %s metadata does not support reshape\n", | |
576 | st->ss->name); | |
577 | return 1; | |
578 | } | |
579 | ||
580 | return st->ss->reshape_super(st, size, level, layout, chunksize, | |
581 | raid_disks, backup_file, dev, verbose); | |
582 | } | |
583 | ||
584 | static void sync_metadata(struct supertype *st) | |
585 | { | |
586 | if (st->ss->external) { | |
76266030 | 587 | if (st->update_tail) { |
7bc71196 | 588 | flush_metadata_updates(st); |
76266030 N |
589 | st->update_tail = &st->updates; |
590 | } else | |
7bc71196 DW |
591 | st->ss->sync_metadata(st); |
592 | } | |
593 | } | |
594 | ||
595 | static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n) | |
596 | { | |
597 | /* when dealing with external metadata subarrays we need to be | |
598 | * prepared to handle EAGAIN. The kernel may need to wait for | |
599 | * mdmon to mark the array active so the kernel can handle | |
600 | * allocations/writeback when preparing the reshape action | |
601 | * (md_allow_write()). We temporarily disable safe_mode_delay | |
602 | * to close a race with the array_state going clean before the | |
603 | * next write to raid_disks / stripe_cache_size | |
604 | */ | |
605 | char safe[50]; | |
606 | int rc; | |
607 | ||
608 | /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */ | |
5da9ab98 N |
609 | if (!container || |
610 | (strcmp(name, "raid_disks") != 0 && | |
611 | strcmp(name, "stripe_cache_size") != 0)) | |
7bc71196 DW |
612 | return sysfs_set_num(sra, NULL, name, n); |
613 | ||
614 | rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe)); | |
615 | if (rc <= 0) | |
616 | return -1; | |
617 | sysfs_set_num(sra, NULL, "safe_mode_delay", 0); | |
618 | rc = sysfs_set_num(sra, NULL, name, n); | |
619 | if (rc < 0 && errno == EAGAIN) { | |
620 | ping_monitor(container); | |
621 | /* if we get EAGAIN here then the monitor is not active | |
622 | * so stop trying | |
623 | */ | |
624 | rc = sysfs_set_num(sra, NULL, name, n); | |
625 | } | |
626 | sysfs_set_str(sra, NULL, "safe_mode_delay", safe); | |
627 | return rc; | |
628 | } | |
629 | ||
47eb4d5a N |
630 | int start_reshape(struct mdinfo *sra) |
631 | { | |
632 | int err; | |
0f28668b | 633 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); |
47eb4d5a N |
634 | err = sysfs_set_num(sra, NULL, "suspend_hi", 0); |
635 | err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
18eaf9e5 N |
636 | /* Setting sync_min can fail if the recovery is already 'running', |
637 | * which can happen when restarting an array which is reshaping. | |
638 | * So don't worry about errors here */ | |
639 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
47eb4d5a N |
640 | err = err ?: sysfs_set_num(sra, NULL, "sync_max", 0); |
641 | err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape"); | |
642 | ||
643 | return err; | |
644 | } | |
645 | ||
646 | void abort_reshape(struct mdinfo *sra) | |
647 | { | |
648 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
649 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); | |
650 | sysfs_set_num(sra, NULL, "suspend_hi", 0); | |
651 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
652 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
653 | sysfs_set_str(sra, NULL, "sync_max", "max"); | |
654 | } | |
655 | ||
dfe77a9e KW |
656 | int remove_disks_for_takeover(struct supertype *st, |
657 | struct mdinfo *sra, | |
658 | int layout) | |
62a48395 AK |
659 | { |
660 | int nr_of_copies; | |
661 | struct mdinfo *remaining; | |
662 | int slot; | |
663 | ||
dfe77a9e KW |
664 | if (sra->array.level == 10) |
665 | nr_of_copies = layout & 0xff; | |
666 | else if (sra->array.level == 1) | |
667 | nr_of_copies = sra->array.raid_disks; | |
668 | else | |
669 | return 1; | |
62a48395 AK |
670 | |
671 | remaining = sra->devs; | |
672 | sra->devs = NULL; | |
673 | /* for each 'copy', select one device and remove from the list. */ | |
674 | for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) { | |
675 | struct mdinfo **diskp; | |
676 | int found = 0; | |
677 | ||
678 | /* Find a working device to keep */ | |
679 | for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) { | |
680 | struct mdinfo *disk = *diskp; | |
681 | ||
682 | if (disk->disk.raid_disk < slot) | |
683 | continue; | |
684 | if (disk->disk.raid_disk >= slot + nr_of_copies) | |
685 | continue; | |
686 | if (disk->disk.state & (1<<MD_DISK_REMOVED)) | |
687 | continue; | |
688 | if (disk->disk.state & (1<<MD_DISK_FAULTY)) | |
689 | continue; | |
690 | if (!(disk->disk.state & (1<<MD_DISK_SYNC))) | |
691 | continue; | |
692 | ||
693 | /* We have found a good disk to use! */ | |
694 | *diskp = disk->next; | |
695 | disk->next = sra->devs; | |
696 | sra->devs = disk; | |
697 | found = 1; | |
698 | break; | |
699 | } | |
700 | if (!found) | |
701 | break; | |
702 | } | |
703 | ||
704 | if (slot < sra->array.raid_disks) { | |
705 | /* didn't find all slots */ | |
706 | struct mdinfo **e; | |
707 | e = &remaining; | |
708 | while (*e) | |
709 | e = &(*e)->next; | |
710 | *e = sra->devs; | |
711 | sra->devs = remaining; | |
712 | return 1; | |
713 | } | |
714 | ||
715 | /* Remove all 'remaining' devices from the array */ | |
716 | while (remaining) { | |
717 | struct mdinfo *sd = remaining; | |
718 | remaining = sd->next; | |
719 | ||
720 | sysfs_set_str(sra, sd, "state", "faulty"); | |
721 | sysfs_set_str(sra, sd, "slot", "none"); | |
722 | sysfs_set_str(sra, sd, "state", "remove"); | |
723 | sd->disk.state |= (1<<MD_DISK_REMOVED); | |
724 | sd->disk.state &= ~(1<<MD_DISK_SYNC); | |
725 | sd->next = sra->devs; | |
726 | sra->devs = sd; | |
727 | } | |
728 | return 0; | |
729 | } | |
730 | ||
130994cb AK |
731 | void reshape_free_fdlist(int *fdlist, |
732 | unsigned long long *offsets, | |
733 | int size) | |
734 | { | |
735 | int i; | |
736 | ||
737 | for (i = 0; i < size; i++) | |
738 | if (fdlist[i] >= 0) | |
739 | close(fdlist[i]); | |
740 | ||
741 | free(fdlist); | |
742 | free(offsets); | |
743 | } | |
744 | ||
745 | int reshape_prepare_fdlist(char *devname, | |
746 | struct mdinfo *sra, | |
747 | int raid_disks, | |
748 | int nrdisks, | |
749 | unsigned long blocks, | |
750 | char *backup_file, | |
751 | int *fdlist, | |
752 | unsigned long long *offsets) | |
753 | { | |
754 | int d = 0; | |
755 | struct mdinfo *sd; | |
756 | ||
757 | for (d = 0; d <= nrdisks; d++) | |
758 | fdlist[d] = -1; | |
759 | d = raid_disks; | |
760 | for (sd = sra->devs; sd; sd = sd->next) { | |
761 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
762 | continue; | |
763 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
764 | char *dn = map_dev(sd->disk.major, | |
765 | sd->disk.minor, 1); | |
766 | fdlist[sd->disk.raid_disk] | |
767 | = dev_open(dn, O_RDONLY); | |
768 | offsets[sd->disk.raid_disk] = sd->data_offset*512; | |
769 | if (fdlist[sd->disk.raid_disk] < 0) { | |
770 | fprintf(stderr, | |
771 | Name ": %s: cannot open component %s\n", | |
772 | devname, dn ? dn : "-unknown-"); | |
773 | d = -1; | |
774 | goto release; | |
775 | } | |
776 | } else if (backup_file == NULL) { | |
777 | /* spare */ | |
778 | char *dn = map_dev(sd->disk.major, | |
779 | sd->disk.minor, 1); | |
780 | fdlist[d] = dev_open(dn, O_RDWR); | |
781 | offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512; | |
782 | if (fdlist[d] < 0) { | |
783 | fprintf(stderr, Name ": %s: cannot open component %s\n", | |
784 | devname, dn ? dn : "-unknown-"); | |
785 | d = -1; | |
786 | goto release; | |
787 | } | |
788 | d++; | |
789 | } | |
790 | } | |
791 | release: | |
792 | return d; | |
793 | } | |
794 | ||
e6e9d47b AK |
795 | int reshape_open_backup_file(char *backup_file, |
796 | int fd, | |
797 | char *devname, | |
798 | long blocks, | |
799 | int *fdlist, | |
a93f87ee N |
800 | unsigned long long *offsets, |
801 | int restart) | |
e6e9d47b AK |
802 | { |
803 | /* Return 1 on success, 0 on any form of failure */ | |
804 | /* need to check backup file is large enough */ | |
805 | char buf[512]; | |
806 | struct stat stb; | |
807 | unsigned int dev; | |
808 | int i; | |
809 | ||
a93f87ee | 810 | *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL), |
e6e9d47b AK |
811 | S_IRUSR | S_IWUSR); |
812 | *offsets = 8 * 512; | |
813 | if (*fdlist < 0) { | |
814 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
815 | devname, backup_file, strerror(errno)); | |
816 | return 0; | |
817 | } | |
818 | /* Guard against backup file being on array device. | |
819 | * If array is partitioned or if LVM etc is in the | |
820 | * way this will not notice, but it is better than | |
821 | * nothing. | |
822 | */ | |
823 | fstat(*fdlist, &stb); | |
824 | dev = stb.st_dev; | |
825 | fstat(fd, &stb); | |
826 | if (stb.st_rdev == dev) { | |
827 | fprintf(stderr, Name ": backup file must NOT be" | |
828 | " on the array being reshaped.\n"); | |
829 | close(*fdlist); | |
830 | return 0; | |
831 | } | |
832 | ||
833 | memset(buf, 0, 512); | |
834 | for (i=0; i < blocks + 1 ; i++) { | |
835 | if (write(*fdlist, buf, 512) != 512) { | |
836 | fprintf(stderr, Name ": %s: cannot create" | |
837 | " backup file %s: %s\n", | |
838 | devname, backup_file, strerror(errno)); | |
839 | return 0; | |
840 | } | |
841 | } | |
842 | if (fsync(*fdlist) != 0) { | |
843 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
844 | devname, backup_file, strerror(errno)); | |
845 | return 0; | |
846 | } | |
847 | ||
848 | return 1; | |
849 | } | |
850 | ||
1c009fc2 AK |
851 | unsigned long compute_backup_blocks(int nchunk, int ochunk, |
852 | unsigned int ndata, unsigned int odata) | |
853 | { | |
854 | unsigned long a, b, blocks; | |
855 | /* So how much do we need to backup. | |
856 | * We need an amount of data which is both a whole number of | |
857 | * old stripes and a whole number of new stripes. | |
858 | * So LCM for (chunksize*datadisks). | |
859 | */ | |
860 | a = (ochunk/512) * odata; | |
861 | b = (nchunk/512) * ndata; | |
862 | /* Find GCD */ | |
863 | while (a != b) { | |
864 | if (a < b) | |
865 | b -= a; | |
866 | if (b < a) | |
867 | a -= b; | |
868 | } | |
869 | /* LCM == product / GCD */ | |
870 | blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a; | |
871 | ||
872 | return blocks; | |
873 | } | |
874 | ||
5da9ab98 N |
875 | char *analyse_change(struct mdinfo *info, struct reshape *re) |
876 | { | |
877 | /* Based on the current array state in info->array and | |
878 | * the changes in info->new_* etc, determine: | |
879 | * - whether the change is possible | |
880 | * - Intermediate level/raid_disks/layout | |
881 | * - whether a restriping reshape is needed | |
882 | * - number of sectors in minimum change unit. This | |
883 | * will cover a whole number of stripes in 'before' and | |
884 | * 'after'. | |
885 | * | |
886 | * Return message if the change should be rejected | |
887 | * NULL if the change can be achieved | |
888 | * | |
889 | * This can be called as part of starting a reshape, or | |
890 | * when assembling an array that is undergoing reshape. | |
891 | */ | |
892 | int new_disks; | |
893 | ||
894 | /* If a new level not explicitly given, we assume no-change */ | |
895 | if (info->new_level == UnSet) | |
896 | info->new_level = info->array.level; | |
897 | ||
898 | if (info->new_chunk) | |
899 | switch (info->new_level) { | |
900 | case 0: | |
901 | case 4: | |
902 | case 5: | |
903 | case 6: | |
904 | case 10: | |
905 | /* chunk size is meaningful, must divide component_size | |
906 | * evenly | |
907 | */ | |
908 | if (info->component_size % (info->new_chunk/512)) | |
909 | return "New chunk size does not" | |
910 | " divide component size"; | |
911 | break; | |
912 | default: | |
913 | return "chunk size not meaningful for this level"; | |
914 | } | |
915 | else | |
916 | info->new_chunk = info->array.chunk_size; | |
917 | ||
918 | switch (info->array.level) { | |
919 | case 1: | |
920 | /* RAID1 can convert to RAID1 with different disks, or | |
dfe77a9e KW |
921 | * raid5 with 2 disks, or |
922 | * raid0 with 1 disk | |
5da9ab98 | 923 | */ |
dfe77a9e KW |
924 | if (info->new_level == 0) { |
925 | re->level = 0; | |
926 | re->before.data_disks = 1; | |
927 | re->after.data_disks = 1; | |
928 | re->before.layout = 0; | |
929 | re->backup_blocks = 0; | |
930 | re->parity = 0; | |
931 | return NULL; | |
932 | } | |
5da9ab98 N |
933 | if (info->new_level == 1) { |
934 | if (info->delta_disks == UnSet) | |
935 | /* Don't know what to do */ | |
936 | return "no change requested for Growing RAID1"; | |
937 | re->level = 1; | |
b4f8e38b | 938 | re->backup_blocks = 0; |
5da9ab98 N |
939 | re->parity = 0; |
940 | return NULL; | |
941 | } | |
942 | if (info->array.raid_disks == 2 && | |
446d2a5a | 943 | info->new_level == 5) { |
5da9ab98 | 944 | re->level = 5; |
5da9ab98 | 945 | re->before.data_disks = 1; |
c8b06d82 | 946 | re->after.data_disks = 1; |
5da9ab98 | 947 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC; |
446d2a5a N |
948 | info->array.chunk_size = 65536; |
949 | break; | |
5da9ab98 N |
950 | } |
951 | /* Could do some multi-stage conversions, but leave that to | |
952 | * later. | |
953 | */ | |
954 | return "Impossibly level change request for RAID1"; | |
955 | ||
956 | case 10: | |
957 | /* RAID10 can only be converted from near mode to | |
958 | * RAID0 by removing some devices | |
959 | */ | |
960 | if ((info->array.layout & ~0xff) != 0x100) | |
961 | return "Cannot Grow RAID10 with far/offset layout"; | |
962 | /* number of devices must be multiple of number of copies */ | |
963 | if (info->array.raid_disks % (info->array.layout & 0xff)) | |
964 | return "RAID10 layout too complex for Grow operation"; | |
965 | ||
966 | if (info->new_level != 0) | |
967 | return "RAID10 can only be changed to RAID0"; | |
968 | new_disks = (info->array.raid_disks | |
969 | / (info->array.layout & 0xff)); | |
5dad078d | 970 | if (info->delta_disks == UnSet) { |
5da9ab98 N |
971 | info->delta_disks = (new_disks |
972 | - info->array.raid_disks); | |
973 | } | |
974 | if (info->delta_disks != new_disks - info->array.raid_disks) | |
975 | return "New number of raid-devices impossible for RAID10"; | |
976 | if (info->new_chunk && | |
977 | info->new_chunk != info->array.chunk_size) | |
978 | return "Cannot change chunk-size with RAID10 Grow"; | |
979 | ||
980 | /* looks good */ | |
981 | re->level = 0; | |
982 | re->parity = 0; | |
983 | re->before.data_disks = new_disks; | |
031d445c | 984 | re->after.data_disks = re->before.data_disks; |
5da9ab98 | 985 | re->before.layout = 0; |
b4f8e38b | 986 | re->backup_blocks = 0; |
5da9ab98 N |
987 | return NULL; |
988 | ||
989 | case 0: | |
990 | /* RAID0 can be converted to RAID10, or to RAID456 */ | |
991 | if (info->new_level == 10) { | |
992 | if (info->new_layout == UnSet && info->delta_disks == UnSet) { | |
993 | /* Assume near=2 layout */ | |
994 | info->new_layout = 0x102; | |
995 | info->delta_disks = info->array.raid_disks; | |
996 | } | |
997 | if (info->new_layout == UnSet) { | |
998 | int copies = 1 + (info->delta_disks | |
999 | / info->array.raid_disks); | |
1000 | if (info->array.raid_disks * (copies-1) | |
1001 | != info->delta_disks) | |
1002 | return "Impossible number of devices" | |
1003 | " for RAID0->RAID10"; | |
1004 | info->new_layout = 0x100 + copies; | |
1005 | } | |
1006 | if (info->delta_disks == UnSet) { | |
1007 | int copies = info->new_layout & 0xff; | |
1008 | if (info->new_layout != 0x100 + copies) | |
1009 | return "New layout impossible" | |
1010 | " for RAID0->RAID10";; | |
1011 | info->delta_disks = (copies - 1) * | |
1012 | info->array.raid_disks; | |
1013 | } | |
1014 | if (info->new_chunk && | |
1015 | info->new_chunk != info->array.chunk_size) | |
1016 | return "Cannot change chunk-size with RAID0->RAID10"; | |
1017 | /* looks good */ | |
1018 | re->level = 10; | |
1019 | re->parity = 0; | |
1020 | re->before.data_disks = (info->array.raid_disks + | |
1021 | info->delta_disks); | |
031d445c | 1022 | re->after.data_disks = re->before.data_disks; |
5da9ab98 | 1023 | re->before.layout = info->new_layout; |
b4f8e38b | 1024 | re->backup_blocks = 0; |
5da9ab98 N |
1025 | return NULL; |
1026 | } | |
1027 | ||
1028 | /* RAID0 can also covert to RAID0/4/5/6 by first converting to | |
1029 | * a raid4 style layout of the final level. | |
1030 | */ | |
1031 | switch (info->new_level) { | |
1032 | case 0: | |
1033 | case 4: | |
1034 | re->level = 4; | |
1035 | re->before.layout = 0; | |
1036 | break; | |
1037 | case 5: | |
1038 | re->level = 5; | |
1039 | re->before.layout = ALGORITHM_PARITY_N; | |
1040 | break; | |
1041 | case 6: | |
1042 | re->level = 6; | |
1043 | re->before.layout = ALGORITHM_PARITY_N; | |
1044 | break; | |
1045 | default: | |
1046 | return "Impossible level change requested"; | |
1047 | } | |
1048 | re->before.data_disks = info->array.raid_disks; | |
1049 | /* determining 'after' layout happens outside this 'switch' */ | |
1050 | break; | |
1051 | ||
1052 | case 4: | |
1053 | info->array.layout = ALGORITHM_PARITY_N; | |
1054 | case 5: | |
1055 | switch (info->new_level) { | |
1056 | case 4: | |
1057 | re->level = info->array.level; | |
1058 | re->before.data_disks = info->array.raid_disks - 1; | |
1059 | re->before.layout = info->array.layout; | |
1060 | break; | |
1061 | case 5: | |
1062 | re->level = 5; | |
1063 | re->before.data_disks = info->array.raid_disks - 1; | |
1064 | re->before.layout = info->array.layout; | |
1065 | break; | |
1066 | case 6: | |
1067 | re->level = 6; | |
1068 | re->before.data_disks = info->array.raid_disks - 1; | |
1069 | switch (info->array.layout) { | |
1070 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1071 | re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
1072 | break; | |
1073 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1074 | re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
1075 | break; | |
1076 | case ALGORITHM_LEFT_SYMMETRIC: | |
1077 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
1078 | break; | |
1079 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1080 | re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
1081 | break; | |
1082 | case ALGORITHM_PARITY_0: | |
1083 | re->before.layout = ALGORITHM_PARITY_0_6; | |
1084 | break; | |
1085 | case ALGORITHM_PARITY_N: | |
1086 | re->before.layout = ALGORITHM_PARITY_N_6; | |
1087 | break; | |
1088 | default: | |
1089 | return "Cannot convert an array with this layout"; | |
1090 | } | |
1091 | break; | |
1092 | case 1: | |
1093 | if (info->array.raid_disks != 2) | |
1094 | return "Can only convert a 2-device array to RAID1"; | |
1095 | re->level = 1; | |
5da9ab98 N |
1096 | break; |
1097 | default: | |
1098 | return "Impossible level change requested"; | |
1099 | } | |
1100 | break; | |
1101 | case 6: | |
1102 | switch (info->new_level) { | |
1103 | case 4: | |
1104 | case 5: | |
1105 | case 6: | |
1106 | re->level = 6; | |
1107 | re->before.data_disks = info->array.raid_disks - 2; | |
1108 | re->before.layout = info->array.layout; | |
1109 | break; | |
1110 | default: | |
1111 | return "Impossible level change requested"; | |
1112 | } | |
1113 | break; | |
1114 | } | |
1115 | ||
1116 | /* If we reached here then it looks like a re-stripe is | |
1117 | * happening. We have determined the intermediate level | |
1118 | * and initial raid_disks/layout and stored these in 're'. | |
1119 | * | |
1120 | * We need to deduce the final layout that can be atomically | |
1121 | * converted to the end state. | |
1122 | */ | |
1123 | switch (info->new_level) { | |
1124 | case 0: | |
1125 | /* We can only get to RAID0 from RAID4 or RAID5 | |
1126 | * with appropriate layout and one extra device | |
1127 | */ | |
1128 | if (re->level != 4 && re->level != 5) | |
1129 | return "Cannot covert to RAID0 from this level"; | |
1130 | if (info->delta_disks == UnSet) | |
1131 | re->after.data_disks = re->before.data_disks; | |
1132 | else | |
1133 | re->after.data_disks = | |
1134 | info->array.raid_disks + info->delta_disks; | |
1135 | switch (re->level) { | |
1136 | case 4: | |
1137 | re->after.layout = 0 ; break; | |
1138 | case 5: | |
1139 | re->after.layout = ALGORITHM_PARITY_N; break; | |
1140 | } | |
1141 | break; | |
1142 | ||
1143 | case 4: | |
1144 | /* We can only get to RAID4 from RAID5 */ | |
1145 | if (re->level != 4 && re->level != 5) | |
1146 | return "Cannot convert to RAID4 from this level"; | |
1147 | if (info->delta_disks == UnSet) | |
1148 | re->after.data_disks = re->before.data_disks; | |
1149 | else | |
1150 | re->after.data_disks = | |
1151 | re->before.data_disks + info->delta_disks; | |
1152 | switch (re->level) { | |
1153 | case 4: | |
1154 | re->after.layout = 0 ; break; | |
1155 | case 5: | |
1156 | re->after.layout = ALGORITHM_PARITY_N; break; | |
1157 | } | |
1158 | break; | |
1159 | ||
1160 | case 5: | |
1161 | /* We get to RAID5 for RAID5 or RAID6 */ | |
1162 | if (re->level != 5 && re->level != 6) | |
1163 | return "Cannot convert to RAID5 from this level"; | |
1164 | if (info->delta_disks == UnSet) | |
1165 | re->after.data_disks = re->before.data_disks; | |
1166 | else if (re->level == 5) | |
1167 | re->after.data_disks = | |
1168 | re->before.data_disks + info->delta_disks; | |
1169 | else | |
1170 | re->after.data_disks = | |
1171 | info->array.raid_disks + info->delta_disks - 1; | |
1172 | switch (re->level) { | |
1173 | case 5: | |
1174 | if (info->new_layout == UnSet) | |
1175 | re->after.layout = re->before.layout; | |
1176 | else | |
1177 | re->after.layout = info->new_layout; | |
1178 | break; | |
1179 | case 6: | |
4a870364 N |
1180 | if (info->new_layout == UnSet) |
1181 | info->new_layout = re->before.layout; | |
1182 | ||
5da9ab98 N |
1183 | /* after.layout needs to be raid6 version of new_layout */ |
1184 | if (info->new_layout == ALGORITHM_PARITY_N) | |
1185 | re->after.layout = ALGORITHM_PARITY_N; | |
1186 | else { | |
1187 | char layout[40]; | |
1188 | char *ls = map_num(r5layout, info->new_layout); | |
1189 | int l; | |
1190 | strcat(strcpy(layout, ls), "-6"); | |
1191 | l = map_name(r6layout, layout); | |
1192 | if (l == UnSet) | |
1193 | return "Cannot find RAID6 layout" | |
1194 | " to convert to"; | |
1195 | re->after.layout = l; | |
1196 | } | |
1197 | } | |
1198 | break; | |
1199 | ||
1200 | case 6: | |
1201 | /* We must already be at level 6 */ | |
1202 | if (re->level != 6) | |
1203 | return "Impossible level change"; | |
1204 | if (info->delta_disks == UnSet) | |
1205 | re->after.data_disks = re->before.data_disks; | |
1206 | else | |
1207 | re->after.data_disks = (info->array.raid_disks + | |
1208 | info->delta_disks) - 2; | |
1209 | if (info->new_layout == UnSet) | |
4a870364 | 1210 | re->after.layout = info->array.layout; |
5da9ab98 N |
1211 | else |
1212 | re->after.layout = info->new_layout; | |
1213 | break; | |
1214 | default: | |
1215 | return "Impossible level change requested"; | |
1216 | } | |
1217 | switch (re->level) { | |
1218 | case 6: re->parity = 2; break; | |
1219 | case 4: | |
1220 | case 5: re->parity = 1; break; | |
1221 | default: re->parity = 0; break; | |
1222 | } | |
1223 | /* So we have a restripe operation, we need to calculate the number | |
1224 | * of blocks per reshape operation. | |
1225 | */ | |
1226 | if (info->new_chunk == 0) | |
1227 | info->new_chunk = info->array.chunk_size; | |
1228 | if (re->after.data_disks == re->before.data_disks && | |
1229 | re->after.layout == re->before.layout && | |
1230 | info->new_chunk == info->array.chunk_size) { | |
1231 | /* Nothing to change */ | |
b4f8e38b | 1232 | re->backup_blocks = 0; |
5da9ab98 N |
1233 | return NULL; |
1234 | } | |
1235 | if (re->after.data_disks == 1 && re->before.data_disks == 1) { | |
446d2a5a | 1236 | /* chunk and layout changes make no difference */ |
b4f8e38b | 1237 | re->backup_blocks = 0; |
5da9ab98 N |
1238 | return NULL; |
1239 | } | |
1240 | ||
1241 | if (re->after.data_disks == re->before.data_disks && | |
1242 | get_linux_version() < 2006032) | |
1243 | return "in-place reshape is not safe before 2.6.32 - sorry."; | |
1244 | ||
1245 | if (re->after.data_disks < re->before.data_disks && | |
1246 | get_linux_version() < 2006030) | |
1247 | return "reshape to fewer devices is not supported before 2.6.32 - sorry."; | |
1248 | ||
b4f8e38b | 1249 | re->backup_blocks = compute_backup_blocks( |
5da9ab98 N |
1250 | info->new_chunk, info->array.chunk_size, |
1251 | re->after.data_disks, | |
1252 | re->before.data_disks); | |
1253 | ||
1254 | re->new_size = info->component_size * re->after.data_disks; | |
1255 | return NULL; | |
1256 | } | |
1257 | ||
1258 | static int reshape_array(char *container, int fd, char *devname, | |
1259 | struct supertype *st, struct mdinfo *info, | |
a93f87ee N |
1260 | int force, char *backup_file, int quiet, int forked, |
1261 | int restart); | |
5da9ab98 N |
1262 | static int reshape_container(char *container, int cfd, char *devname, |
1263 | struct supertype *st, | |
1264 | struct mdinfo *info, | |
1265 | int force, | |
1266 | char *backup_file, | |
1267 | int quiet); | |
1c009fc2 | 1268 | |
06b0d786 | 1269 | int Grow_reshape(char *devname, int fd, int quiet, char *backup_file, |
e86c9dd6 | 1270 | long long size, |
691a36b7 N |
1271 | int level, char *layout_str, int chunksize, int raid_disks, |
1272 | int force) | |
e86c9dd6 NB |
1273 | { |
1274 | /* Make some changes in the shape of an array. | |
1275 | * The kernel must support the change. | |
7236ee7a N |
1276 | * |
1277 | * There are three different changes. Each can trigger | |
1278 | * a resync or recovery so we freeze that until we have | |
1279 | * requested everything (if kernel supports freezing - 2.6.30). | |
1280 | * The steps are: | |
1281 | * - change size (i.e. component_size) | |
1282 | * - change level | |
1283 | * - change layout/chunksize/ndisks | |
1284 | * | |
1285 | * The last can require a reshape. It is different on different | |
1286 | * levels so we need to check the level before actioning it. | |
1287 | * Some times the level change needs to be requested after the | |
1288 | * reshape (e.g. raid6->raid5, raid5->raid0) | |
1289 | * | |
e86c9dd6 | 1290 | */ |
5da9ab98 | 1291 | struct mdu_array_info_s array; |
7236ee7a | 1292 | int rv = 0; |
e86c9dd6 | 1293 | struct supertype *st; |
4725bc31 | 1294 | char *subarray = NULL; |
e86c9dd6 | 1295 | |
7236ee7a | 1296 | int frozen; |
7236ee7a | 1297 | int changed = 0; |
7bc71196 | 1298 | char *container = NULL; |
5da9ab98 | 1299 | char container_buf[20]; |
7bc71196 | 1300 | int cfd = -1; |
e86c9dd6 | 1301 | |
5da9ab98 | 1302 | struct mdinfo info; |
7e0f6979 | 1303 | struct mdinfo *sra; |
e86c9dd6 NB |
1304 | |
1305 | if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) { | |
1306 | fprintf(stderr, Name ": %s is not an active md array - aborting\n", | |
1307 | devname); | |
1308 | return 1; | |
1309 | } | |
24d40069 | 1310 | |
9ce510be N |
1311 | if (size >= 0 && |
1312 | (chunksize || level!= UnSet || layout_str || raid_disks)) { | |
1313 | fprintf(stderr, Name ": cannot change component size at the same time " | |
1314 | "as other changes.\n" | |
1315 | " Change size first, then check data is intact before " | |
1316 | "making other changes.\n"); | |
1317 | return 1; | |
1318 | } | |
1319 | ||
24d40069 N |
1320 | if (raid_disks && raid_disks < array.raid_disks && array.level > 1 && |
1321 | get_linux_version() < 2006032 && | |
1322 | !check_env("MDADM_FORCE_FEWER")) { | |
1323 | fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n" | |
1324 | " Please use a newer kernel\n"); | |
1325 | return 1; | |
1326 | } | |
7bc71196 DW |
1327 | |
1328 | st = super_by_fd(fd, &subarray); | |
1329 | if (!st) { | |
1330 | fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname); | |
1331 | return 1; | |
1332 | } | |
acab7bb1 N |
1333 | if (raid_disks > st->max_devs) { |
1334 | fprintf(stderr, Name ": Cannot increase raid-disks on this array" | |
1335 | " beyond %d\n", st->max_devs); | |
1336 | return 1; | |
1337 | } | |
7bc71196 DW |
1338 | |
1339 | /* in the external case we need to check that the requested reshape is | |
1340 | * supported, and perform an initial check that the container holds the | |
1341 | * pre-requisite spare devices (mdmon owns final validation) | |
1342 | */ | |
1343 | if (st->ss->external) { | |
1344 | int container_dev; | |
7f2ba464 | 1345 | int rv; |
7bc71196 DW |
1346 | |
1347 | if (subarray) { | |
1348 | container_dev = st->container_dev; | |
1349 | cfd = open_dev_excl(st->container_dev); | |
7bc71196 DW |
1350 | } else { |
1351 | container_dev = st->devnum; | |
1352 | close(fd); | |
1353 | cfd = open_dev_excl(st->devnum); | |
1354 | fd = cfd; | |
1355 | } | |
1356 | if (cfd < 0) { | |
1357 | fprintf(stderr, Name ": Unable to open container for %s\n", | |
1358 | devname); | |
7f2ba464 | 1359 | free(subarray); |
7bc71196 DW |
1360 | return 1; |
1361 | } | |
1362 | ||
5da9ab98 N |
1363 | fmt_devname(container_buf, container_dev); |
1364 | container = container_buf; | |
7bc71196 | 1365 | |
eb744788 AK |
1366 | rv = st->ss->load_container(st, cfd, NULL); |
1367 | ||
7f2ba464 | 1368 | if (rv) { |
7bc71196 DW |
1369 | fprintf(stderr, Name ": Cannot read superblock for %s\n", |
1370 | devname); | |
7f2ba464 | 1371 | free(subarray); |
7bc71196 DW |
1372 | return 1; |
1373 | } | |
1374 | ||
1375 | if (mdmon_running(container_dev)) | |
1376 | st->update_tail = &st->updates; | |
691a36b7 N |
1377 | } |
1378 | ||
1379 | if (raid_disks > array.raid_disks && | |
1380 | array.spare_disks < (raid_disks - array.raid_disks) && | |
1381 | !force) { | |
1382 | fprintf(stderr, | |
1383 | Name ": Need %d spare%s to avoid degraded array," | |
1384 | " and only have %d.\n" | |
1385 | " Use --force to over-ride this check.\n", | |
1386 | raid_disks - array.raid_disks, | |
1387 | raid_disks - array.raid_disks == 1 ? "" : "s", | |
1388 | array.spare_disks); | |
1389 | return 1; | |
7bc71196 DW |
1390 | } |
1391 | ||
3fa436ac ML |
1392 | sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS |
1393 | | GET_STATE | GET_VERSION); | |
1394 | if (sra) { | |
7f2ba464 | 1395 | if (st->ss->external && subarray == NULL) { |
7bc71196 DW |
1396 | array.level = LEVEL_CONTAINER; |
1397 | sra->array.level = LEVEL_CONTAINER; | |
1398 | } | |
7bc71196 | 1399 | } else { |
b526e52d DW |
1400 | fprintf(stderr, Name ": failed to read sysfs parameters for %s\n", |
1401 | devname); | |
1402 | return 1; | |
1403 | } | |
7f2ba464 DW |
1404 | frozen = freeze(st); |
1405 | if (frozen < -1) { | |
1406 | /* freeze() already spewed the reason */ | |
1407 | return 1; | |
1408 | } else if (frozen < 0) { | |
7236ee7a N |
1409 | fprintf(stderr, Name ": %s is performing resync/recovery and cannot" |
1410 | " be reshaped\n", devname); | |
1411 | return 1; | |
1412 | } | |
1413 | ||
1414 | /* ========= set size =============== */ | |
1415 | if (size >= 0 && (size == 0 || size != array.size)) { | |
7bc71196 DW |
1416 | long long orig_size = array.size; |
1417 | ||
1418 | if (reshape_super(st, size, UnSet, UnSet, 0, 0, NULL, devname, !quiet)) { | |
1419 | rv = 1; | |
1420 | goto release; | |
1421 | } | |
1422 | sync_metadata(st); | |
7236ee7a N |
1423 | array.size = size; |
1424 | if (array.size != size) { | |
1425 | /* got truncated to 32bit, write to | |
1426 | * component_size instead | |
1427 | */ | |
1428 | if (sra) | |
1429 | rv = sysfs_set_num(sra, NULL, | |
1430 | "component_size", size); | |
1431 | else | |
1432 | rv = -1; | |
1433 | } else | |
1434 | rv = ioctl(fd, SET_ARRAY_INFO, &array); | |
1435 | if (rv != 0) { | |
39bbb392 | 1436 | int err = errno; |
7bc71196 DW |
1437 | |
1438 | /* restore metadata */ | |
1439 | if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0, | |
1440 | NULL, devname, !quiet) == 0) | |
1441 | sync_metadata(st); | |
7236ee7a | 1442 | fprintf(stderr, Name ": Cannot set device size for %s: %s\n", |
39bbb392 N |
1443 | devname, strerror(err)); |
1444 | if (err == EBUSY && | |
1445 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1446 | fprintf(stderr, " Bitmap must be removed before size can be changed\n"); | |
7236ee7a N |
1447 | rv = 1; |
1448 | goto release; | |
1449 | } | |
1450 | ioctl(fd, GET_ARRAY_INFO, &array); | |
be1cabbd | 1451 | size = get_component_size(fd)/2; |
f98841b3 N |
1452 | if (size == 0) |
1453 | size = array.size; | |
7236ee7a | 1454 | if (!quiet) |
f98841b3 N |
1455 | fprintf(stderr, Name ": component size of %s has been set to %lluK\n", |
1456 | devname, size); | |
7236ee7a | 1457 | changed = 1; |
7bc71196 | 1458 | } else if (array.level != LEVEL_CONTAINER) { |
be1cabbd | 1459 | size = get_component_size(fd)/2; |
f98841b3 N |
1460 | if (size == 0) |
1461 | size = array.size; | |
7236ee7a N |
1462 | } |
1463 | ||
dfe77a9e | 1464 | /* ========= check for Raid10/Raid1 -> Raid0 conversion =============== |
5da9ab98 | 1465 | * current implementation assumes that following conditions must be met: |
dfe77a9e KW |
1466 | * - RAID10: |
1467 | * - far_copies == 1 | |
1468 | * - near_copies == 2 | |
62a48395 | 1469 | */ |
dfe77a9e KW |
1470 | if ((level == 0 && array.level == 10 && sra && |
1471 | array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) || | |
1472 | (level == 0 && array.level == 1 && sra)) { | |
62a48395 | 1473 | int err; |
dfe77a9e | 1474 | err = remove_disks_for_takeover(st, sra, array.layout); |
62a48395 AK |
1475 | if (err) { |
1476 | dprintf(Name": Array cannot be reshaped\n"); | |
62a48395 AK |
1477 | if (cfd > -1) |
1478 | close(cfd); | |
5da9ab98 N |
1479 | rv = 1; |
1480 | goto release; | |
7236ee7a | 1481 | } |
bb025c2f KW |
1482 | /* FIXME this is added with no justification - why is it here */ |
1483 | ping_monitor(container); | |
7236ee7a N |
1484 | } |
1485 | ||
5da9ab98 | 1486 | info.array = array; |
7443ee81 | 1487 | sysfs_init(&info, fd, NoMdDev); |
3fa436ac | 1488 | strcpy(info.text_version, sra->text_version); |
7443ee81 | 1489 | info.component_size = size*2; |
5da9ab98 N |
1490 | info.new_level = level; |
1491 | info.new_chunk = chunksize * 1024; | |
1492 | if (raid_disks) | |
1493 | info.delta_disks = raid_disks - info.array.raid_disks; | |
1494 | else | |
1495 | info.delta_disks = UnSet; | |
1496 | if (layout_str == NULL) { | |
1497 | info.new_layout = UnSet; | |
1498 | if (info.array.level == 6 && | |
1499 | (info.new_level == 6 || info.new_level == UnSet) && | |
1500 | info.array.layout >= 16) { | |
7236ee7a | 1501 | fprintf(stderr, Name |
5da9ab98 N |
1502 | ": %s has a non-standard layout. If you" |
1503 | " wish to preserve this\n" | |
1504 | " during the reshape, please specify" | |
1505 | " --layout=preserve\n" | |
1506 | " If you want to change it, specify a" | |
1507 | " layout or use --layout=normalise\n", | |
7236ee7a N |
1508 | devname); |
1509 | rv = 1; | |
1510 | goto release; | |
1511 | } | |
5da9ab98 N |
1512 | } else if (strcmp(layout_str, "normalise") == 0 || |
1513 | strcmp(layout_str, "normalize") == 0) { | |
1514 | /* If we have a -6 RAID6 layout, remove the '-6'. */ | |
1515 | info.new_layout = UnSet; | |
1516 | if (info.array.level == 6 && info.new_level == UnSet) { | |
1517 | char l[40], *h; | |
1518 | strcpy(l, map_num(r6layout, info.array.layout)); | |
1519 | h = strrchr(l, '-'); | |
1520 | if (h && strcmp(h, "-6") == 0) { | |
1521 | *h = 0; | |
1522 | info.new_layout = map_name(r6layout, l); | |
7236ee7a N |
1523 | } |
1524 | } | |
5da9ab98 N |
1525 | } else if (strcmp(layout_str, "preserve") == 0) { |
1526 | info.new_layout = UnSet; | |
1527 | } else { | |
1528 | int l = info.new_level; | |
1529 | if (l == UnSet) | |
1530 | l = info.array.level; | |
1531 | switch (l) { | |
1532 | case 5: | |
1533 | info.new_layout = map_name(r5layout, layout_str); | |
1534 | break; | |
1535 | case 6: | |
1536 | info.new_layout = map_name(r6layout, layout_str); | |
1537 | break; | |
1538 | case 10: | |
1539 | info.new_layout = parse_layout_10(layout_str); | |
1540 | break; | |
1541 | case LEVEL_FAULTY: | |
1542 | info.new_layout = parse_layout_faulty(layout_str); | |
1543 | break; | |
1544 | default: | |
1545 | fprintf(stderr, Name ": layout not meaningful" | |
1546 | " with this level\n"); | |
7bc71196 DW |
1547 | rv = 1; |
1548 | goto release; | |
1549 | } | |
5da9ab98 N |
1550 | if (info.new_layout == UnSet) { |
1551 | fprintf(stderr, Name ": layout %s not understood" | |
1552 | " for this level\n", | |
1553 | layout_str); | |
1554 | rv = 1; | |
1555 | goto release; | |
7bc71196 | 1556 | } |
7236ee7a N |
1557 | } |
1558 | ||
5da9ab98 N |
1559 | if (array.level == LEVEL_CONTAINER) { |
1560 | /* This change is to be applied to every array in the | |
1561 | * container. This is only needed when the metadata imposes | |
1562 | * restraints of the various arrays in the container. | |
1563 | * Currently we only know that IMSM requires all arrays | |
1564 | * to have the same number of devices so changing the | |
1565 | * number of devices (On-Line Capacity Expansion) must be | |
1566 | * performed at the level of the container | |
1567 | */ | |
1568 | rv = reshape_container(container, fd, devname, st, &info, | |
1569 | force, backup_file, quiet); | |
9202b8eb | 1570 | frozen = 0; |
5da9ab98 N |
1571 | } else { |
1572 | /* Impose these changes on a single array. First | |
1573 | * check that the metadata is OK with the change. */ | |
7bc71196 | 1574 | |
5da9ab98 N |
1575 | if (reshape_super(st, info.component_size, info.new_level, |
1576 | info.new_layout, info.new_chunk, | |
1577 | info.array.raid_disks + info.delta_disks, | |
1578 | backup_file, devname, quiet)) { | |
7bc71196 DW |
1579 | rv = 1; |
1580 | goto release; | |
1581 | } | |
5da9ab98 N |
1582 | sync_metadata(st); |
1583 | rv = reshape_array(container, fd, devname, st, &info, force, | |
a93f87ee | 1584 | backup_file, quiet, 0, 0); |
9202b8eb | 1585 | frozen = 0; |
5da9ab98 N |
1586 | } |
1587 | release: | |
9202b8eb N |
1588 | if (frozen > 0) |
1589 | unfreeze(st); | |
5da9ab98 N |
1590 | return rv; |
1591 | } | |
7bc71196 | 1592 | |
5da9ab98 N |
1593 | static int reshape_array(char *container, int fd, char *devname, |
1594 | struct supertype *st, struct mdinfo *info, | |
1595 | int force, | |
a93f87ee N |
1596 | char *backup_file, int quiet, int forked, |
1597 | int restart) | |
5da9ab98 N |
1598 | { |
1599 | struct reshape reshape; | |
1600 | int spares_needed; | |
1601 | char *msg; | |
1602 | int orig_level = UnSet; | |
1603 | int disks, odisks; | |
7bc71196 | 1604 | |
5da9ab98 N |
1605 | struct mdu_array_info_s array; |
1606 | char *c; | |
e86c9dd6 | 1607 | |
5da9ab98 N |
1608 | int *fdlist; |
1609 | unsigned long long *offsets; | |
1610 | int d; | |
1611 | int nrdisks; | |
1612 | int err; | |
da8100ca | 1613 | unsigned long blocks; |
5da9ab98 N |
1614 | unsigned long cache; |
1615 | unsigned long long array_size; | |
1616 | int done; | |
cf6ac177 | 1617 | struct mdinfo *sra = NULL; |
7bc71196 | 1618 | |
5da9ab98 N |
1619 | msg = analyse_change(info, &reshape); |
1620 | if (msg) { | |
1621 | fprintf(stderr, Name ": %s\n", msg); | |
631d7405 | 1622 | goto release; |
5da9ab98 | 1623 | } |
9ff87c16 AK |
1624 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { |
1625 | dprintf("Canot get array information.\n"); | |
631d7405 | 1626 | goto release; |
9ff87c16 | 1627 | } |
a93f87ee N |
1628 | |
1629 | if (restart) { | |
1630 | /* reshape already started. just skip to monitoring the reshape */ | |
1631 | if (reshape.backup_blocks == 0) | |
1632 | return 0; | |
1633 | goto started; | |
1634 | } | |
a9c3e78f AK |
1635 | /* The container is frozen but the array may not be. |
1636 | * So freeze the array so spares don't get put to the wrong use | |
1637 | * FIXME there should probably be a cleaner separation between | |
1638 | * freeze_array and freeze_container. | |
1639 | */ | |
1640 | sysfs_freeze_array(info); | |
5da9ab98 N |
1641 | spares_needed = max(reshape.before.data_disks, |
1642 | reshape.after.data_disks) | |
1643 | + reshape.parity - array.raid_disks; | |
7bc71196 | 1644 | |
88c1a083 | 1645 | if (!force && |
d6bd632c | 1646 | info->new_level > 1 && |
88c1a083 | 1647 | spares_needed > info->array.spare_disks) { |
5da9ab98 N |
1648 | fprintf(stderr, |
1649 | Name ": Need %d spare%s to avoid degraded array," | |
1650 | " and only have %d.\n" | |
1651 | " Use --force to over-ride this check.\n", | |
1652 | spares_needed, | |
1653 | spares_needed == 1 ? "" : "s", | |
1654 | info->array.spare_disks); | |
631d7405 | 1655 | goto release; |
7bc71196 | 1656 | } |
e86c9dd6 | 1657 | |
5da9ab98 N |
1658 | if (reshape.level != info->array.level) { |
1659 | char *c = map_num(pers, reshape.level); | |
1660 | int err; | |
1661 | if (c == NULL) | |
631d7405 | 1662 | goto release; |
e86c9dd6 | 1663 | |
5da9ab98 N |
1664 | err = sysfs_set_str(info, NULL, "level", c); |
1665 | if (err) { | |
1666 | err = errno; | |
1667 | fprintf(stderr, Name ": %s: could not set level to %s\n", | |
1668 | devname, c); | |
1669 | if (err == EBUSY && | |
1670 | (info->array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1671 | fprintf(stderr, " Bitmap must be removed" | |
1672 | " before level can be changed\n"); | |
631d7405 | 1673 | goto release; |
19678e53 | 1674 | } |
5da9ab98 | 1675 | if (!quiet) |
b6b95155 | 1676 | fprintf(stderr, Name ": level of %s changed to %s\n", |
5da9ab98 N |
1677 | devname, c); |
1678 | orig_level = info->array.level; | |
e86c9dd6 | 1679 | |
16d4d84e AK |
1680 | if (reshape.level > 0 && st->ss->external) { |
1681 | /* make sure mdmon is aware of the new level */ | |
1682 | if (!mdmon_running(st->container_dev)) | |
1683 | start_mdmon(st->container_dev); | |
1684 | ping_monitor(container); | |
1685 | } | |
5da9ab98 | 1686 | } |
5da9ab98 N |
1687 | /* ->reshape_super might have chosen some spares from the |
1688 | * container that it wants to be part of the new array. | |
1689 | * We can collect them with ->container_content and give | |
1690 | * them to the kernel. | |
1691 | */ | |
1692 | if (st->ss->reshape_super && st->ss->container_content) { | |
1693 | char *subarray = strchr(info->text_version+1, '/')+1; | |
1694 | struct mdinfo *info2 = | |
1695 | st->ss->container_content(st, subarray); | |
1696 | struct mdinfo *d; | |
1697 | ||
2dd0d697 AK |
1698 | if (info2) { |
1699 | sysfs_init(info2, fd, st->devnum); | |
5da9ab98 N |
1700 | for (d = info2->devs; d; d = d->next) { |
1701 | if (d->disk.state == 0 && | |
1702 | d->disk.raid_disk >= 0) { | |
1703 | /* This is a spare that wants to | |
1704 | * be part of the array. | |
1705 | */ | |
1706 | add_disk(fd, st, info2, d); | |
1707 | } | |
9860f271 | 1708 | } |
2dd0d697 AK |
1709 | sysfs_free(info2); |
1710 | } | |
5da9ab98 | 1711 | } |
1686dc25 | 1712 | |
b4f8e38b | 1713 | if (reshape.backup_blocks == 0) { |
5da9ab98 N |
1714 | /* No restriping needed, but we might need to impose |
1715 | * some more changes: layout, raid_disks, chunk_size | |
e86c9dd6 | 1716 | */ |
5da9ab98 N |
1717 | if (info->new_layout != UnSet && |
1718 | info->new_layout != info->array.layout) { | |
1719 | info->array.layout = info->new_layout; | |
1720 | if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) { | |
1721 | fprintf(stderr, Name ": failed to set new layout\n"); | |
631d7405 | 1722 | goto release; |
5da9ab98 N |
1723 | } else if (!quiet) |
1724 | printf("layout for %s set to %d\n", | |
1725 | devname, info->array.layout); | |
1726 | } | |
1727 | if (info->delta_disks != UnSet && | |
1728 | info->delta_disks != 0) { | |
1729 | info->array.raid_disks += info->delta_disks; | |
1730 | if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) { | |
1731 | fprintf(stderr, Name ": failed to set raid disks\n"); | |
631d7405 | 1732 | goto release; |
5da9ab98 N |
1733 | } else if (!quiet) |
1734 | printf("raid_disks for %s set to %d\n", | |
1735 | devname, info->array.raid_disks); | |
1736 | } | |
1737 | if (info->new_chunk != 0 && | |
1738 | info->new_chunk != info->array.chunk_size) { | |
1739 | if (sysfs_set_num(info, NULL, | |
1740 | "chunk_size", info->new_chunk) != 0) { | |
1741 | fprintf(stderr, Name ": failed to set chunk size\n"); | |
631d7405 | 1742 | goto release; |
5da9ab98 N |
1743 | } else if (!quiet) |
1744 | printf("chunk size for %s set to %d\n", | |
1745 | devname, info->array.chunk_size); | |
4725bc31 | 1746 | } |
e35b189b | 1747 | unfreeze(st); |
631d7405 | 1748 | return 0; |
5da9ab98 | 1749 | } |
19678e53 | 1750 | |
5da9ab98 N |
1751 | /* |
1752 | * There are three possibilities. | |
1753 | * 1/ The array will shrink. | |
1754 | * We need to ensure the reshape will pause before reaching | |
1755 | * the 'critical section'. We also need to fork and wait for | |
1756 | * that to happen. When it does we | |
1757 | * suspend/backup/complete/unfreeze | |
1758 | * | |
1759 | * 2/ The array will not change size. | |
1760 | * This requires that we keep a backup of a sliding window | |
1761 | * so that we can restore data after a crash. So we need | |
1762 | * to fork and monitor progress. | |
1763 | * In future we will allow the data_offset to change, so | |
1764 | * a sliding backup becomes unnecessary. | |
1765 | * | |
1766 | * 3/ The array will grow. This is relatively easy. | |
1767 | * However the kernel's restripe routines will cheerfully | |
1768 | * overwrite some early data before it is safe. So we | |
1769 | * need to make a backup of the early parts of the array | |
1770 | * and be ready to restore it if rebuild aborts very early. | |
1771 | * For externally managed metadata, we still need a forked | |
1772 | * child to monitor the reshape and suspend IO over the region | |
1773 | * that is being reshaped. | |
1774 | * | |
1775 | * We backup data by writing it to one spare, or to a | |
1776 | * file which was given on command line. | |
1777 | * | |
1778 | * In each case, we first make sure that storage is available | |
1779 | * for the required backup. | |
1780 | * Then we: | |
1781 | * - request the shape change. | |
1782 | * - fork to handle backup etc. | |
1783 | */ | |
a93f87ee | 1784 | started: |
5da9ab98 N |
1785 | /* Check that we can hold all the data */ |
1786 | get_dev_size(fd, NULL, &array_size); | |
1787 | if (reshape.new_size < (array_size/512)) { | |
1788 | fprintf(stderr, | |
1789 | Name ": this change will reduce the size of the array.\n" | |
1790 | " use --grow --array-size first to truncate array.\n" | |
1791 | " e.g. mdadm --grow %s --array-size %llu\n", | |
1792 | devname, reshape.new_size/2); | |
5da9ab98 N |
1793 | goto release; |
1794 | } | |
7236ee7a | 1795 | |
5da9ab98 | 1796 | sra = sysfs_read(fd, 0, |
3656edcd | 1797 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK| |
5da9ab98 | 1798 | GET_CACHE); |
5da9ab98 N |
1799 | if (!sra) { |
1800 | fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n", | |
1801 | devname); | |
5da9ab98 N |
1802 | goto release; |
1803 | } | |
7236ee7a | 1804 | |
5da9ab98 N |
1805 | /* Decide how many blocks (sectors) for a reshape |
1806 | * unit. The number we have so far is just a minimum | |
1807 | */ | |
b4f8e38b | 1808 | blocks = reshape.backup_blocks; |
5da9ab98 N |
1809 | if (reshape.before.data_disks == |
1810 | reshape.after.data_disks) { | |
1811 | /* Make 'blocks' bigger for better throughput, but | |
1812 | * not so big that we reject it below. | |
1813 | * Try for 16 megabytes | |
1814 | */ | |
1815 | while (blocks * 32 < sra->component_size && | |
1816 | blocks < 16*1024*2) | |
1817 | blocks *= 2; | |
1818 | } else | |
1819 | fprintf(stderr, Name ": Need to backup %luK of critical " | |
1820 | "section..\n", blocks/2); | |
1821 | ||
1822 | if (blocks >= sra->component_size/2) { | |
1823 | fprintf(stderr, Name ": %s: Something wrong" | |
1824 | " - reshape aborted\n", | |
1825 | devname); | |
5da9ab98 N |
1826 | goto release; |
1827 | } | |
7236ee7a | 1828 | |
5da9ab98 N |
1829 | /* Now we need to open all these devices so we can read/write. |
1830 | */ | |
1831 | nrdisks = array.raid_disks + sra->array.spare_disks; | |
1832 | fdlist = malloc((1+nrdisks) * sizeof(int)); | |
1833 | offsets = malloc((1+nrdisks) * sizeof(offsets[0])); | |
1834 | if (!fdlist || !offsets) { | |
1835 | fprintf(stderr, Name ": malloc failed: grow aborted\n"); | |
5da9ab98 N |
1836 | goto release; |
1837 | } | |
e380d3be | 1838 | |
5da9ab98 N |
1839 | d = reshape_prepare_fdlist(devname, sra, array.raid_disks, |
1840 | nrdisks, blocks, backup_file, | |
1841 | fdlist, offsets); | |
1842 | if (d < 0) { | |
5da9ab98 N |
1843 | goto release; |
1844 | } | |
1845 | if (backup_file == NULL) { | |
374ba335 | 1846 | if (reshape.after.data_disks <= reshape.before.data_disks) { |
5da9ab98 N |
1847 | fprintf(stderr, |
1848 | Name ": %s: Cannot grow - need backup-file\n", | |
c03ef02d | 1849 | devname); |
5da9ab98 N |
1850 | goto release; |
1851 | } else if (sra->array.spare_disks == 0) { | |
1852 | fprintf(stderr, Name ": %s: Cannot grow - need a spare or " | |
1853 | "backup-file to backup critical section\n", | |
e86c9dd6 | 1854 | devname); |
5da9ab98 | 1855 | goto release; |
e86c9dd6 | 1856 | } |
5da9ab98 N |
1857 | } else { |
1858 | if (!reshape_open_backup_file(backup_file, fd, devname, | |
1859 | (signed)blocks, | |
a93f87ee | 1860 | fdlist+d, offsets+d, restart)) { |
130994cb | 1861 | goto release; |
e86c9dd6 | 1862 | } |
5da9ab98 N |
1863 | d++; |
1864 | } | |
7236ee7a | 1865 | |
5da9ab98 N |
1866 | /* lastly, check that the internal stripe cache is |
1867 | * large enough, or it won't work. | |
1868 | * It must hold at least 4 stripes of the larger | |
1869 | * chunk size | |
1870 | */ | |
1871 | cache = max(info->array.chunk_size, info->new_chunk); | |
1872 | cache *= 4; /* 4 stripes minimum */ | |
1873 | cache /= 512; /* convert to sectors */ | |
1874 | disks = min(reshape.before.data_disks, reshape.after.data_disks); | |
1875 | /* make sure there is room for 'blocks' with a bit to spare */ | |
1876 | if (cache < 16 + blocks / disks) | |
1877 | cache = 16 + blocks / disks; | |
1878 | cache /= (4096/512); /* Covert from sectors to pages */ | |
1879 | ||
1880 | if (sra->cache_size < cache) | |
1881 | subarray_set_num(container, sra, "stripe_cache_size", | |
1882 | cache+1); | |
1883 | ||
1884 | /* Right, everything seems fine. Let's kick things off. | |
1885 | * If only changing raid_disks, use ioctl, else use | |
1886 | * sysfs. | |
1887 | */ | |
1888 | sync_metadata(st); | |
1889 | ||
b4f8e38b | 1890 | sra->new_chunk = info->new_chunk; |
ddee071d | 1891 | |
a93f87ee | 1892 | if (info->reshape_active) |
ddee071d | 1893 | sra->reshape_progress = info->reshape_progress; |
f897078e N |
1894 | else { |
1895 | sra->reshape_progress = 0; | |
1896 | if (reshape.after.data_disks < reshape.before.data_disks) | |
1897 | /* start from the end of the new array */ | |
1898 | sra->reshape_progress = (sra->component_size | |
1899 | * reshape.after.data_disks); | |
1900 | } | |
1901 | ||
1902 | if (info->array.chunk_size == info->new_chunk && | |
7477d513 AK |
1903 | reshape.before.layout == reshape.after.layout && |
1904 | st->ss->external == 0) { | |
f897078e | 1905 | /* use SET_ARRAY_INFO but only if reshape hasn't started */ |
5da9ab98 | 1906 | array.raid_disks = reshape.after.data_disks + reshape.parity; |
f897078e N |
1907 | if (!info->reshape_active && |
1908 | ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
5da9ab98 | 1909 | int err = errno; |
631d7405 | 1910 | |
5da9ab98 N |
1911 | fprintf(stderr, |
1912 | Name ": Cannot set device shape for %s: %s\n", | |
1913 | devname, strerror(errno)); | |
7bc71196 | 1914 | |
5da9ab98 N |
1915 | if (err == EBUSY && |
1916 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1917 | fprintf(stderr, | |
1918 | " Bitmap must be removed before" | |
1919 | " shape can be changed\n"); | |
b5420ef3 | 1920 | |
5da9ab98 N |
1921 | goto release; |
1922 | } | |
1923 | } else { | |
1924 | /* set them all just in case some old 'new_*' value | |
f897078e N |
1925 | * persists from some earlier problem. |
1926 | * We even set them when restarting in the middle. They will | |
1927 | * already be set in that case so this will be a no-op, | |
1928 | * but it is hard to tell the difference. | |
7236ee7a | 1929 | */ |
631d7405 | 1930 | int err = 0; |
5da9ab98 | 1931 | if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0) |
631d7405 N |
1932 | err = errno; |
1933 | if (!err && sysfs_set_num(sra, NULL, "layout", | |
5da9ab98 | 1934 | reshape.after.layout) < 0) |
631d7405 N |
1935 | err = errno; |
1936 | if (!err && subarray_set_num(container, sra, "raid_disks", | |
5da9ab98 N |
1937 | reshape.after.data_disks + |
1938 | reshape.parity) < 0) | |
631d7405 N |
1939 | err = errno; |
1940 | if (err) { | |
5da9ab98 N |
1941 | fprintf(stderr, Name ": Cannot set device shape for %s\n", |
1942 | devname); | |
7236ee7a | 1943 | |
5da9ab98 N |
1944 | if (err == EBUSY && |
1945 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1946 | fprintf(stderr, | |
1947 | " Bitmap must be removed before" | |
1948 | " shape can be changed\n"); | |
1949 | goto release; | |
e86c9dd6 | 1950 | } |
5da9ab98 | 1951 | } |
e86c9dd6 | 1952 | |
fab32c97 AK |
1953 | err = start_reshape(sra); |
1954 | if (err) { | |
1955 | fprintf(stderr, Name ": Cannot start reshape for %s\n", | |
1956 | devname); | |
1957 | goto release; | |
1958 | } | |
a93f87ee N |
1959 | if (restart) |
1960 | sysfs_set_str(sra, NULL, "array_state", "active"); | |
5da9ab98 N |
1961 | |
1962 | /* Now we just need to kick off the reshape and watch, while | |
1963 | * handling backups of the data... | |
1964 | * This is all done by a forked background process. | |
1965 | */ | |
1966 | switch(forked ? 0 : fork()) { | |
6b2d630c N |
1967 | case -1: |
1968 | fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n", | |
1969 | strerror(errno)); | |
6b2d630c | 1970 | abort_reshape(sra); |
631d7405 | 1971 | goto release; |
6b2d630c N |
1972 | default: |
1973 | return 0; | |
5da9ab98 | 1974 | case 0: |
6b2d630c N |
1975 | break; |
1976 | } | |
5da9ab98 | 1977 | |
6b2d630c N |
1978 | close(fd); |
1979 | if (check_env("MDADM_GROW_VERIFY")) | |
1980 | fd = open(devname, O_RDONLY | O_DIRECT); | |
1981 | else | |
1982 | fd = -1; | |
1983 | mlockall(MCL_FUTURE); | |
5da9ab98 | 1984 | |
6b2d630c | 1985 | odisks = reshape.before.data_disks + reshape.parity; |
582496b2 | 1986 | |
6b2d630c N |
1987 | if (st->ss->external) { |
1988 | /* metadata handler takes it from here */ | |
1989 | done = st->ss->manage_reshape( | |
1990 | fd, sra, &reshape, st, blocks, | |
1991 | fdlist, offsets, | |
1992 | d - odisks, fdlist+odisks, | |
1993 | offsets+odisks); | |
1994 | } else | |
1995 | done = child_monitor( | |
1996 | fd, sra, &reshape, st, blocks, | |
1997 | fdlist, offsets, | |
1998 | d - odisks, fdlist+odisks, | |
1999 | offsets+odisks); | |
2000 | ||
2001 | if (backup_file && done) | |
2002 | unlink(backup_file); | |
2003 | if (!done) { | |
2004 | abort_reshape(sra); | |
2005 | goto out; | |
2006 | } | |
1cbc5680 N |
2007 | |
2008 | if (!st->ss->external && | |
2009 | !(reshape.before.data_disks != reshape.after.data_disks | |
2010 | && info->custom_array_size) && | |
2011 | info->new_level == reshape.level && | |
2012 | !forked) { | |
2013 | /* no need to wait for the reshape to finish as | |
2014 | * there is nothing more to do. | |
2015 | */ | |
2016 | exit(0); | |
2017 | } | |
2018 | wait_reshape(sra); | |
2019 | ||
2020 | if (st->ss->external) { | |
2021 | /* Re-load the metadata as much could have changed */ | |
2022 | int cfd = open_dev(st->container_dev); | |
2023 | if (cfd >= 0) { | |
2024 | ping_monitor(container); | |
2025 | st->ss->free_super(st); | |
2026 | st->ss->load_container(st, cfd, container); | |
2027 | close(cfd); | |
2028 | } | |
2029 | } | |
2030 | ||
6b2d630c N |
2031 | /* set new array size if required customer_array_size is used |
2032 | * by this metadata. | |
2033 | */ | |
2034 | if (reshape.before.data_disks != | |
2035 | reshape.after.data_disks && | |
2036 | info->custom_array_size) { | |
2037 | struct mdinfo *info2; | |
2038 | char *subarray = strchr(info->text_version+1, '/')+1; | |
582496b2 | 2039 | |
6b2d630c N |
2040 | info2 = st->ss->container_content(st, subarray); |
2041 | if (info2) { | |
2042 | unsigned long long current_size = 0; | |
2043 | unsigned long long new_size = | |
2044 | info2->custom_array_size/2; | |
2045 | ||
2046 | if (sysfs_get_ll(sra, | |
2047 | NULL, | |
2048 | "array_size", | |
2049 | ¤t_size) == 0 && | |
2050 | new_size > current_size) { | |
2051 | if (sysfs_set_num(sra, NULL, | |
2052 | "array_size", new_size) | |
2053 | < 0) | |
2054 | dprintf("Error: Cannot" | |
2055 | " set array size"); | |
2056 | else | |
2057 | dprintf("Array size " | |
2058 | "changed"); | |
2059 | dprintf(" from %llu to %llu.\n", | |
2060 | current_size, new_size); | |
582496b2 | 2061 | } |
6b2d630c | 2062 | sysfs_free(info2); |
582496b2 | 2063 | } |
6b2d630c | 2064 | } |
582496b2 | 2065 | |
6b2d630c | 2066 | if (info->new_level != reshape.level) { |
582496b2 | 2067 | |
6b2d630c | 2068 | c = map_num(pers, info->new_level); |
1cbc5680 N |
2069 | if (c) { |
2070 | err = sysfs_set_str(sra, NULL, "level", c); | |
2071 | if (err) | |
2072 | fprintf(stderr, Name\ | |
2073 | ": %s: could not set level " | |
2074 | "to %s\n", devname, c); | |
2075 | } | |
7236ee7a | 2076 | } |
6b2d630c N |
2077 | out: |
2078 | if (forked) | |
2079 | return 0; | |
2080 | exit(0); | |
e86c9dd6 | 2081 | |
6b2d630c | 2082 | release: |
1cbc5680 | 2083 | if (orig_level != UnSet && sra) { |
7236ee7a N |
2084 | c = map_num(pers, orig_level); |
2085 | if (c && sysfs_set_str(sra, NULL, "level", c) == 0) | |
2086 | fprintf(stderr, Name ": aborting level change\n"); | |
2087 | } | |
9202b8eb N |
2088 | if (!forked) |
2089 | unfreeze(st); | |
631d7405 | 2090 | return 1; |
7236ee7a | 2091 | } |
e86c9dd6 | 2092 | |
5da9ab98 N |
2093 | int reshape_container(char *container, int cfd, char *devname, |
2094 | struct supertype *st, | |
2095 | struct mdinfo *info, | |
2096 | int force, | |
2097 | char *backup_file, | |
2098 | int quiet) | |
2099 | { | |
1bb174ba AK |
2100 | struct mdinfo *cc = NULL; |
2101 | ||
d8eb27f7 KA |
2102 | /* component_size is not meaningful for a container, |
2103 | * so pass '-1' meaning 'no change' | |
2104 | */ | |
2105 | if (reshape_super(st, -1, info->new_level, | |
5da9ab98 N |
2106 | info->new_layout, info->new_chunk, |
2107 | info->array.raid_disks + info->delta_disks, | |
2108 | backup_file, devname, quiet)) | |
2109 | return 1; | |
2110 | ||
2111 | sync_metadata(st); | |
2112 | ||
1bb174ba AK |
2113 | /* ping monitor to be sure that update is on disk |
2114 | */ | |
2115 | ping_monitor(container); | |
5da9ab98 N |
2116 | |
2117 | switch (fork()) { | |
2118 | case -1: /* error */ | |
2119 | perror("Cannot fork to complete reshape\n"); | |
2120 | return 1; | |
2121 | default: /* parent */ | |
2122 | printf(Name ": multi-array reshape continues in background\n"); | |
2123 | return 0; | |
2124 | case 0: /* child */ | |
2125 | break; | |
2126 | } | |
2127 | ||
1bb174ba AK |
2128 | while(1) { |
2129 | /* For each member array with reshape_active, | |
2130 | * we need to perform the reshape. | |
2131 | * We pick the first array that needs reshaping and | |
2132 | * reshape it. reshape_array() will re-read the metadata | |
2133 | * so the next time through a different array should be | |
2134 | * ready for reshape. | |
2135 | */ | |
2136 | struct mdinfo *content; | |
5da9ab98 N |
2137 | int rv; |
2138 | int fd; | |
2139 | struct mdstat_ent *mdstat; | |
5da9ab98 N |
2140 | char *adev; |
2141 | ||
1bb174ba | 2142 | sysfs_free(cc); |
5da9ab98 | 2143 | |
1bb174ba AK |
2144 | cc = st->ss->container_content(st, NULL); |
2145 | ||
2146 | for (content = cc; content ; content = content->next) { | |
2147 | char *subarray; | |
2148 | if (!content->reshape_active) | |
2149 | continue; | |
2150 | ||
2151 | subarray = strchr(content->text_version+1, '/')+1; | |
2152 | mdstat = mdstat_by_subdev(subarray, | |
2153 | devname2devnum(container)); | |
2154 | if (!mdstat) | |
2155 | continue; | |
2156 | break; | |
2157 | } | |
2158 | if (!content) | |
2159 | break; | |
5da9ab98 | 2160 | |
55f14721 | 2161 | fd = open_dev(mdstat->devnum); |
5da9ab98 N |
2162 | if (fd < 0) |
2163 | break; | |
2164 | adev = map_dev(dev2major(mdstat->devnum), | |
2165 | dev2minor(mdstat->devnum), | |
2166 | 0); | |
2167 | if (!adev) | |
1bb174ba AK |
2168 | adev = content->text_version; |
2169 | ||
2170 | sysfs_init(content, fd, mdstat->devnum); | |
5da9ab98 | 2171 | |
1bb174ba AK |
2172 | rv = reshape_array(container, fd, adev, st, |
2173 | content, force, | |
a93f87ee | 2174 | backup_file, quiet, 1, 0); |
5da9ab98 N |
2175 | close(fd); |
2176 | if (rv) | |
2177 | break; | |
2178 | } | |
9202b8eb | 2179 | unfreeze(st); |
1bb174ba | 2180 | sysfs_free(cc); |
5da9ab98 N |
2181 | exit(0); |
2182 | } | |
2183 | ||
7236ee7a N |
2184 | /* |
2185 | * We run a child process in the background which performs the following | |
2186 | * steps: | |
2187 | * - wait for resync to reach a certain point | |
2188 | * - suspend io to the following section | |
2189 | * - backup that section | |
2190 | * - allow resync to proceed further | |
2191 | * - resume io | |
2192 | * - discard the backup. | |
2193 | * | |
2194 | * When are combined in slightly different ways in the three cases. | |
2195 | * Grow: | |
2196 | * - suspend/backup/allow/wait/resume/discard | |
2197 | * Shrink: | |
2198 | * - allow/wait/suspend/backup/allow/wait/resume/discard | |
2199 | * same-size: | |
2200 | * - wait/resume/discard/suspend/backup/allow | |
2201 | * | |
2202 | * suspend/backup/allow always come together | |
2203 | * wait/resume/discard do too. | |
2204 | * For the same-size case we have two backups to improve flow. | |
2205 | * | |
2206 | */ | |
e86c9dd6 | 2207 | |
7443ee81 N |
2208 | int progress_reshape(struct mdinfo *info, struct reshape *reshape, |
2209 | unsigned long long backup_point, | |
2210 | unsigned long long wait_point, | |
2211 | unsigned long long *suspend_point, | |
2212 | unsigned long long *reshape_completed) | |
2213 | { | |
2214 | /* This function is called repeatedly by the reshape manager. | |
2215 | * It determines how much progress can safely be made and allows | |
2216 | * that progress. | |
2217 | * - 'info' identifies the array and particularly records in | |
2218 | * ->reshape_progress the metadata's knowledge of progress | |
2219 | * This is a sector offset from the start of the array | |
2220 | * of the next array block to be relocated. This number | |
2221 | * may increase from 0 or decrease from array_size, depending | |
2222 | * on the type of reshape that is happening. | |
2223 | * Note that in contrast, 'sync_completed' is a block count of the | |
2224 | * reshape so far. It gives the distance between the start point | |
2225 | * (head or tail of device) and the next place that data will be | |
2226 | * written. It always increases. | |
2227 | * - 'reshape' is the structure created by analyse_change | |
2228 | * - 'backup_point' shows how much the metadata manager has backed-up | |
2229 | * data. For reshapes with increasing progress, it is the next address | |
2230 | * to be backed up, previous addresses have been backed-up. For | |
2231 | * decreasing progress, it is the earliest address that has been | |
2232 | * backed up - later address are also backed up. | |
2233 | * So addresses between reshape_progress and backup_point are | |
2234 | * backed up providing those are in the 'correct' order. | |
2235 | * - 'wait_point' is an array address. When reshape_completed | |
2236 | * passes this point, progress_reshape should return. It might | |
2237 | * return earlier if it determines that ->reshape_progress needs | |
2238 | * to be updated or further backup is needed. | |
2239 | * - suspend_point is maintained by progress_reshape and the caller | |
2240 | * should not touch it except to initialise to zero. | |
2241 | * It is an array address and it only increases in 2.6.37 and earlier. | |
b6b95155 | 2242 | * This makes it difficult to handle reducing reshapes with |
7443ee81 N |
2243 | * external metadata. |
2244 | * However: it is similar to backup_point in that it records the | |
2245 | * other end of a suspended region from reshape_progress. | |
2246 | * it is moved to extend the region that is safe to backup and/or | |
2247 | * reshape | |
2248 | * - reshape_completed is read from sysfs and returned. The caller | |
2249 | * should copy this into ->reshape_progress when it has reason to | |
2250 | * believe that the metadata knows this, and any backup outside this | |
2251 | * has been erased. | |
2252 | * | |
2253 | * Return value is: | |
2254 | * 1 if more data from backup_point - but only as far as suspend_point, | |
2255 | * should be backed up | |
2256 | * 0 if things are progressing smoothly | |
2257 | * -1 if the reshape is finished, either because it is all done, | |
2258 | * or due to an error. | |
2259 | */ | |
2260 | ||
2261 | int advancing = (reshape->after.data_disks | |
2262 | >= reshape->before.data_disks); | |
38dad34a N |
2263 | unsigned long long need_backup; /* All data between start of array and |
2264 | * here will at some point need to | |
2265 | * be backed up. | |
d0ab945e | 2266 | */ |
7443ee81 | 2267 | unsigned long long read_offset, write_offset; |
b4f8e38b | 2268 | unsigned long long write_range; |
7443ee81 | 2269 | unsigned long long max_progress, target, completed; |
d0ab945e N |
2270 | unsigned long long array_size = (info->component_size |
2271 | * reshape->before.data_disks); | |
7443ee81 | 2272 | int fd; |
77a73a17 | 2273 | char buf[20]; |
7443ee81 N |
2274 | |
2275 | /* First, we unsuspend any region that is now known to be safe. | |
2276 | * If suspend_point is on the 'wrong' side of reshape_progress, then | |
2277 | * we don't have or need suspension at the moment. This is true for | |
2278 | * native metadata when we don't need to back-up. | |
2279 | */ | |
2280 | if (advancing) { | |
49cd738b | 2281 | if (info->reshape_progress <= *suspend_point) |
7443ee81 N |
2282 | sysfs_set_num(info, NULL, "suspend_lo", |
2283 | info->reshape_progress); | |
2284 | } else { | |
2285 | /* Note: this won't work in 2.6.37 and before. | |
2286 | * Something somewhere should make sure we don't need it! | |
2287 | */ | |
49cd738b | 2288 | if (info->reshape_progress >= *suspend_point) |
7443ee81 N |
2289 | sysfs_set_num(info, NULL, "suspend_hi", |
2290 | info->reshape_progress); | |
2291 | } | |
2292 | ||
2293 | /* Now work out how far it is safe to progress. | |
2294 | * If the read_offset for ->reshape_progress is less than | |
2295 | * 'blocks' beyond the write_offset, we can only progress as far | |
2296 | * as a backup. | |
2297 | * Otherwise we can progress until the write_offset for the new location | |
2298 | * reaches (within 'blocks' of) the read_offset at the current location. | |
2299 | * However that region must be suspended unless we are using native | |
2300 | * metadata. | |
2301 | * If we need to suspend more, we limit it to 128M per device, which is | |
2302 | * rather arbitrary and should be some time-based calculation. | |
2303 | */ | |
ec757320 N |
2304 | read_offset = info->reshape_progress / reshape->before.data_disks; |
2305 | write_offset = info->reshape_progress / reshape->after.data_disks; | |
b4f8e38b | 2306 | write_range = info->new_chunk/512; |
38dad34a N |
2307 | if (reshape->before.data_disks == reshape->after.data_disks) |
2308 | need_backup = array_size; | |
2309 | else | |
2310 | need_backup = reshape->backup_blocks; | |
7443ee81 | 2311 | if (advancing) { |
38dad34a | 2312 | if (read_offset < write_offset + write_range) |
7443ee81 | 2313 | max_progress = backup_point; |
ddee071d | 2314 | else |
7443ee81 | 2315 | max_progress = |
2f3dd5e4 N |
2316 | read_offset * |
2317 | reshape->after.data_disks; | |
7443ee81 | 2318 | } else { |
38dad34a N |
2319 | if (read_offset > write_offset - write_range) |
2320 | /* Can only progress as far as has been backed up, | |
2321 | * which must be suspended */ | |
7443ee81 | 2322 | max_progress = backup_point; |
38dad34a N |
2323 | else if (info->reshape_progress <= need_backup) |
2324 | max_progress = backup_point; | |
2325 | else { | |
2326 | if (info->array.major_version >= 0) | |
2327 | /* Can progress until backup is needed */ | |
2328 | max_progress = need_backup; | |
2329 | else { | |
2330 | /* Can progress until metadata update is required */ | |
2331 | max_progress = | |
2332 | read_offset * | |
2333 | reshape->after.data_disks; | |
2334 | /* but data must be suspended */ | |
2335 | if (max_progress < *suspend_point) | |
2336 | max_progress = *suspend_point; | |
2337 | } | |
7443ee81 N |
2338 | } |
2339 | } | |
2340 | ||
2341 | /* We know it is safe to progress to 'max_progress' providing | |
2342 | * it is suspended or we are using native metadata. | |
2343 | * Consider extending suspend_point 128M per device if it | |
2344 | * is less than 64M per device beyond reshape_progress. | |
2345 | * But always do a multiple of 'blocks' | |
832b2b9a N |
2346 | * FIXME this is too big - it takes to long to complete |
2347 | * this much. | |
7443ee81 N |
2348 | */ |
2349 | target = 64*1024*2 * min(reshape->before.data_disks, | |
2350 | reshape->after.data_disks); | |
b4f8e38b | 2351 | target /= reshape->backup_blocks; |
7443ee81 N |
2352 | if (target < 2) |
2353 | target = 2; | |
b4f8e38b | 2354 | target *= reshape->backup_blocks; |
7443ee81 N |
2355 | |
2356 | /* For externally managed metadata we always need to suspend IO to | |
2357 | * the area being reshaped so we regularly push suspend_point forward. | |
2358 | * For native metadata we only need the suspend if we are going to do | |
2359 | * a backup. | |
2360 | */ | |
2361 | if (advancing) { | |
d0ab945e N |
2362 | if ((need_backup > info->reshape_progress |
2363 | || info->array.major_version < 0) && | |
7443ee81 | 2364 | *suspend_point < info->reshape_progress + target) { |
d0ab945e N |
2365 | if (need_backup < *suspend_point + 2 * target) |
2366 | *suspend_point = need_backup; | |
2367 | else if (*suspend_point + 2 * target < array_size) | |
7443ee81 | 2368 | *suspend_point += 2 * target; |
d0ab945e N |
2369 | else |
2370 | *suspend_point = array_size; | |
7443ee81 | 2371 | sysfs_set_num(info, NULL, "suspend_hi", *suspend_point); |
d0ab945e N |
2372 | if (max_progress > *suspend_point) |
2373 | max_progress = *suspend_point; | |
7443ee81 N |
2374 | } |
2375 | } else { | |
38dad34a N |
2376 | if (info->array.major_version >= 0) { |
2377 | /* Only need to suspend when about to backup */ | |
2378 | if (info->reshape_progress < need_backup * 2 && | |
2379 | *suspend_point > 0) { | |
d0ab945e | 2380 | *suspend_point = 0; |
38dad34a N |
2381 | sysfs_set_num(info, NULL, "suspend_lo", 0); |
2382 | sysfs_set_num(info, NULL, "suspend_hi", need_backup); | |
2383 | } | |
2384 | } else { | |
2385 | /* Need to suspend continually */ | |
2386 | if (info->reshape_progress < *suspend_point) | |
2387 | *suspend_point = info->reshape_progress; | |
2388 | if (*suspend_point + target < info->reshape_progress) | |
2389 | /* No need to move suspend region yet */; | |
2390 | else { | |
2391 | if (*suspend_point >= 2 * target) | |
2392 | *suspend_point -= 2 * target; | |
2393 | else | |
2394 | *suspend_point = 0; | |
2395 | sysfs_set_num(info, NULL, "suspend_lo", | |
2396 | *suspend_point); | |
2397 | } | |
d0ab945e N |
2398 | if (max_progress < *suspend_point) |
2399 | max_progress = *suspend_point; | |
7443ee81 N |
2400 | } |
2401 | } | |
2402 | ||
2403 | /* now set sync_max to allow that progress. sync_max, like | |
2404 | * sync_completed is a count of sectors written per device, so | |
2405 | * we find the difference between max_progress and the start point, | |
2406 | * and divide that by after.data_disks to get a sync_max | |
2407 | * number. | |
2408 | * At the same time we convert wait_point to a similar number | |
2409 | * for comparing against sync_completed. | |
2410 | */ | |
92d1991f | 2411 | /* scale down max_progress to per_disk */ |
7443ee81 | 2412 | max_progress /= reshape->after.data_disks; |
92d1991f N |
2413 | /* Round to chunk size as some kernels give an erroneously high number */ |
2414 | max_progress /= info->new_chunk/512; | |
2415 | max_progress *= info->new_chunk/512; | |
2416 | /* Limit progress to the whole device */ | |
2417 | if (max_progress > info->component_size) | |
2418 | max_progress = info->component_size; | |
7443ee81 | 2419 | wait_point /= reshape->after.data_disks; |
92d1991f N |
2420 | if (!advancing) { |
2421 | /* switch from 'device offset' to 'processed block count' */ | |
2422 | max_progress = info->component_size - max_progress; | |
2423 | wait_point = info->component_size - wait_point; | |
2424 | } | |
7443ee81 N |
2425 | |
2426 | sysfs_set_num(info, NULL, "sync_max", max_progress); | |
2427 | ||
2428 | /* Now wait. If we have already reached the point that we were | |
2429 | * asked to wait to, don't wait at all, else wait for any change. | |
2430 | * We need to select on 'sync_completed' as that is the place that | |
2431 | * notifications happen, but we are really interested in | |
2432 | * 'reshape_position' | |
2433 | */ | |
2434 | fd = sysfs_get_fd(info, NULL, "sync_completed"); | |
2435 | if (fd < 0) | |
77a73a17 | 2436 | goto check_progress; |
7443ee81 N |
2437 | |
2438 | if (sysfs_fd_get_ll(fd, &completed) < 0) { | |
2439 | close(fd); | |
77a73a17 | 2440 | goto check_progress; |
7443ee81 N |
2441 | } |
2442 | while (completed < max_progress && completed < wait_point) { | |
2443 | /* Check that sync_action is still 'reshape' to avoid | |
2444 | * waiting forever on a dead array | |
2445 | */ | |
2446 | char action[20]; | |
2447 | fd_set rfds; | |
2448 | if (sysfs_get_str(info, NULL, "sync_action", | |
2449 | action, 20) <= 0 || | |
2450 | strncmp(action, "reshape", 7) != 0) | |
2451 | break; | |
26d6e157 AK |
2452 | /* Some kernels reset 'sync_completed' to zero |
2453 | * before setting 'sync_action' to 'idle'. | |
2454 | * So we need these extra tests. | |
2455 | */ | |
2456 | if (completed == 0 && advancing | |
2457 | && info->reshape_progress > 0) | |
2458 | break; | |
2459 | if (completed == 0 && !advancing | |
2460 | && info->reshape_progress < (info->component_size | |
2461 | * reshape->after.data_disks)) | |
2462 | break; | |
7443ee81 N |
2463 | FD_ZERO(&rfds); |
2464 | FD_SET(fd, &rfds); | |
2465 | select(fd+1, NULL, NULL, &rfds, NULL); | |
2466 | if (sysfs_fd_get_ll(fd, &completed) < 0) { | |
2467 | close(fd); | |
77a73a17 | 2468 | goto check_progress; |
7443ee81 N |
2469 | } |
2470 | } | |
10d0d365 AK |
2471 | /* Some kernels reset 'sync_completed' to zero, |
2472 | * we need to have real point we are in md | |
2473 | */ | |
2474 | if (completed == 0) | |
2475 | completed = max_progress; | |
2476 | ||
92d1991f N |
2477 | /* some kernels can give an incorrectly high 'completed' number */ |
2478 | completed /= (info->new_chunk/512); | |
2479 | completed *= (info->new_chunk/512); | |
7443ee81 N |
2480 | /* Convert 'completed' back in to a 'progress' number */ |
2481 | completed *= reshape->after.data_disks; | |
2482 | if (!advancing) { | |
2483 | completed = info->component_size * reshape->after.data_disks | |
2484 | - completed; | |
2485 | } | |
2486 | *reshape_completed = completed; | |
2487 | ||
2488 | close(fd); | |
2489 | ||
2490 | /* We return the need_backup flag. Caller will decide | |
d0ab945e | 2491 | * how much - a multiple of ->backup_blocks up to *suspend_point |
7443ee81 | 2492 | */ |
38dad34a N |
2493 | if (advancing) |
2494 | return need_backup > info->reshape_progress; | |
2495 | else | |
2496 | return need_backup >= info->reshape_progress; | |
77a73a17 N |
2497 | |
2498 | check_progress: | |
2499 | /* if we couldn't read a number from sync_completed, then | |
2500 | * either the reshape did complete, or it aborted. | |
2501 | * We can tell which by checking for 'none' in reshape_position. | |
2502 | */ | |
2503 | strcpy(buf, "hi"); | |
2504 | if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0 | |
2505 | || strncmp(buf, "none", 4) != 0) | |
2506 | return -2; /* abort */ | |
6d5316f6 N |
2507 | else { |
2508 | /* Maybe racing with array shutdown - check state */ | |
2509 | if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0 | |
2510 | || strncmp(buf, "inactive", 8) == 0 | |
2511 | || strncmp(buf, "clear",5) == 0) | |
2512 | return -2; /* abort */ | |
77a73a17 | 2513 | return -1; /* complete */ |
6d5316f6 | 2514 | } |
7443ee81 N |
2515 | } |
2516 | ||
2517 | ||
fcf57625 | 2518 | /* FIXME return status is never checked */ |
4411fb17 | 2519 | static int grow_backup(struct mdinfo *sra, |
7236ee7a | 2520 | unsigned long long offset, /* per device */ |
7443ee81 | 2521 | unsigned long stripes, /* per device, in old chunks */ |
7236ee7a N |
2522 | int *sources, unsigned long long *offsets, |
2523 | int disks, int chunk, int level, int layout, | |
2524 | int dests, int *destfd, unsigned long long *destoffsets, | |
d4445387 | 2525 | int part, int *degraded, |
7236ee7a N |
2526 | char *buf) |
2527 | { | |
2528 | /* Backup 'blocks' sectors at 'offset' on each device of the array, | |
2529 | * to storage 'destfd' (offset 'destoffsets'), after first | |
2530 | * suspending IO. Then allow resync to continue | |
2531 | * over the suspended section. | |
2532 | * Use part 'part' of the backup-super-block. | |
2533 | */ | |
2534 | int odata = disks; | |
2535 | int rv = 0; | |
2536 | int i; | |
f21e18ca N |
2537 | unsigned long long ll; |
2538 | int new_degraded; | |
7236ee7a N |
2539 | //printf("offset %llu\n", offset); |
2540 | if (level >= 4) | |
2541 | odata--; | |
2542 | if (level == 6) | |
2543 | odata--; | |
7443ee81 | 2544 | |
d4445387 | 2545 | /* Check that array hasn't become degraded, else we might backup the wrong data */ |
2c6ac128 N |
2546 | if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0) |
2547 | return -1; /* FIXME this error is ignored */ | |
f21e18ca | 2548 | new_degraded = (int)ll; |
d4445387 N |
2549 | if (new_degraded != *degraded) { |
2550 | /* check each device to ensure it is still working */ | |
2551 | struct mdinfo *sd; | |
2552 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
2553 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2554 | continue; | |
2555 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
2556 | char sbuf[20]; | |
2557 | if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 || | |
2558 | strstr(sbuf, "faulty") || | |
2559 | strstr(sbuf, "in_sync") == NULL) { | |
2560 | /* this device is dead */ | |
2561 | sd->disk.state = (1<<MD_DISK_FAULTY); | |
2562 | if (sd->disk.raid_disk >= 0 && | |
2563 | sources[sd->disk.raid_disk] >= 0) { | |
2564 | close(sources[sd->disk.raid_disk]); | |
2565 | sources[sd->disk.raid_disk] = -1; | |
2566 | } | |
2567 | } | |
2568 | } | |
2569 | } | |
2570 | *degraded = new_degraded; | |
2571 | } | |
7236ee7a N |
2572 | if (part) { |
2573 | bsb.arraystart2 = __cpu_to_le64(offset * odata); | |
200871ad | 2574 | bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
2575 | } else { |
2576 | bsb.arraystart = __cpu_to_le64(offset * odata); | |
200871ad | 2577 | bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
2578 | } |
2579 | if (part) | |
2580 | bsb.magic[15] = '2'; | |
2581 | for (i = 0; i < dests; i++) | |
2582 | if (part) | |
2583 | lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0); | |
2584 | else | |
2585 | lseek64(destfd[i], destoffsets[i], 0); | |
2586 | ||
2587 | rv = save_stripes(sources, offsets, | |
2588 | disks, chunk, level, layout, | |
2589 | dests, destfd, | |
2590 | offset*512*odata, stripes * chunk * odata, | |
2591 | buf); | |
2592 | ||
2593 | if (rv) | |
2594 | return rv; | |
97396422 | 2595 | bsb.mtime = __cpu_to_le64(time(0)); |
7236ee7a N |
2596 | for (i = 0; i < dests; i++) { |
2597 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
2598 | ||
2599 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
2600 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
2601 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
2602 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
2603 | ||
72044953 | 2604 | rv = -1; |
f21e18ca N |
2605 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0) |
2606 | != destoffsets[i] - 4096) | |
72044953 N |
2607 | break; |
2608 | if (write(destfd[i], &bsb, 512) != 512) | |
2609 | break; | |
ff94fb86 | 2610 | if (destoffsets[i] > 4096) { |
f21e18ca | 2611 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) != |
a847575a | 2612 | destoffsets[i]+stripes*chunk*odata) |
72044953 N |
2613 | break; |
2614 | if (write(destfd[i], &bsb, 512) != 512) | |
2615 | break; | |
ff94fb86 | 2616 | } |
7236ee7a | 2617 | fsync(destfd[i]); |
72044953 | 2618 | rv = 0; |
7236ee7a | 2619 | } |
2efedc7b | 2620 | |
fcf57625 | 2621 | return rv; |
7236ee7a N |
2622 | } |
2623 | ||
2624 | /* in 2.6.30, the value reported by sync_completed can be | |
2625 | * less that it should be by one stripe. | |
2626 | * This only happens when reshape hits sync_max and pauses. | |
2627 | * So allow wait_backup to either extent sync_max further | |
2628 | * than strictly necessary, or return before the | |
2629 | * sync has got quite as far as we would really like. | |
2630 | * This is what 'blocks2' is for. | |
2631 | * The various caller give appropriate values so that | |
2632 | * every works. | |
2633 | */ | |
fcf57625 | 2634 | /* FIXME return value is often ignored */ |
7443ee81 | 2635 | static int forget_backup( |
7236ee7a N |
2636 | int dests, int *destfd, unsigned long long *destoffsets, |
2637 | int part) | |
2638 | { | |
7443ee81 N |
2639 | /* |
2640 | * Erase backup 'part' (which is 0 or 1) | |
7236ee7a | 2641 | */ |
7236ee7a | 2642 | int i; |
fcf57625 | 2643 | int rv; |
7236ee7a | 2644 | |
7236ee7a N |
2645 | if (part) { |
2646 | bsb.arraystart2 = __cpu_to_le64(0); | |
2647 | bsb.length2 = __cpu_to_le64(0); | |
2648 | } else { | |
2649 | bsb.arraystart = __cpu_to_le64(0); | |
2650 | bsb.length = __cpu_to_le64(0); | |
2651 | } | |
97396422 | 2652 | bsb.mtime = __cpu_to_le64(time(0)); |
fcf57625 | 2653 | rv = 0; |
7236ee7a N |
2654 | for (i = 0; i < dests; i++) { |
2655 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
2656 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
2657 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
2658 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
2659 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
f21e18ca | 2660 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) != |
a847575a | 2661 | destoffsets[i]-4096) |
72044953 N |
2662 | rv = -1; |
2663 | if (rv == 0 && | |
2664 | write(destfd[i], &bsb, 512) != 512) | |
2665 | rv = -1; | |
7236ee7a N |
2666 | fsync(destfd[i]); |
2667 | } | |
fcf57625 | 2668 | return rv; |
7236ee7a | 2669 | } |
e86c9dd6 | 2670 | |
7236ee7a N |
2671 | static void fail(char *msg) |
2672 | { | |
fcf57625 | 2673 | int rv; |
72044953 N |
2674 | rv = (write(2, msg, strlen(msg)) != (int)strlen(msg)); |
2675 | rv |= (write(2, "\n", 1) != 1); | |
fcf57625 | 2676 | exit(rv ? 1 : 2); |
7236ee7a N |
2677 | } |
2678 | ||
2679 | static char *abuf, *bbuf; | |
f21e18ca | 2680 | static unsigned long long abuflen; |
7236ee7a N |
2681 | static void validate(int afd, int bfd, unsigned long long offset) |
2682 | { | |
2683 | /* check that the data in the backup against the array. | |
2684 | * This is only used for regression testing and should not | |
2685 | * be used while the array is active | |
2686 | */ | |
7236ee7a N |
2687 | if (afd < 0) |
2688 | return; | |
2689 | lseek64(bfd, offset - 4096, 0); | |
2690 | if (read(bfd, &bsb2, 512) != 512) | |
2691 | fail("cannot read bsb"); | |
2692 | if (bsb2.sb_csum != bsb_csum((char*)&bsb2, | |
2693 | ((char*)&bsb2.sb_csum)-((char*)&bsb2))) | |
2694 | fail("first csum bad"); | |
2695 | if (memcmp(bsb2.magic, "md_backup_data", 14) != 0) | |
2696 | fail("magic is bad"); | |
2697 | if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 && | |
2698 | bsb2.sb_csum2 != bsb_csum((char*)&bsb2, | |
2699 | ((char*)&bsb2.sb_csum2)-((char*)&bsb2))) | |
2700 | fail("second csum bad"); | |
2701 | ||
2702 | if (__le64_to_cpu(bsb2.devstart)*512 != offset) | |
2703 | fail("devstart is wrong"); | |
2704 | ||
2705 | if (bsb2.length) { | |
2706 | unsigned long long len = __le64_to_cpu(bsb2.length)*512; | |
2707 | ||
2708 | if (abuflen < len) { | |
2709 | free(abuf); | |
2710 | free(bbuf); | |
2711 | abuflen = len; | |
fcf57625 N |
2712 | if (posix_memalign((void**)&abuf, 4096, abuflen) || |
2713 | posix_memalign((void**)&bbuf, 4096, abuflen)) { | |
2714 | abuflen = 0; | |
2715 | /* just stop validating on mem-alloc failure */ | |
2716 | return; | |
2717 | } | |
e86c9dd6 | 2718 | } |
48924014 | 2719 | |
7236ee7a | 2720 | lseek64(bfd, offset, 0); |
f21e18ca | 2721 | if ((unsigned long long)read(bfd, bbuf, len) != len) { |
080fd005 | 2722 | //printf("len %llu\n", len); |
7236ee7a N |
2723 | fail("read first backup failed"); |
2724 | } | |
2725 | lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0); | |
f21e18ca | 2726 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
2727 | fail("read first from array failed"); |
2728 | if (memcmp(bbuf, abuf, len) != 0) { | |
080fd005 | 2729 | #if 0 |
7236ee7a N |
2730 | int i; |
2731 | printf("offset=%llu len=%llu\n", | |
080fd005 | 2732 | (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len); |
7236ee7a N |
2733 | for (i=0; i<len; i++) |
2734 | if (bbuf[i] != abuf[i]) { | |
2735 | printf("first diff byte %d\n", i); | |
93ecfa01 | 2736 | break; |
7236ee7a | 2737 | } |
080fd005 | 2738 | #endif |
7236ee7a | 2739 | fail("data1 compare failed"); |
e86c9dd6 | 2740 | } |
7236ee7a N |
2741 | } |
2742 | if (bsb2.length2) { | |
2743 | unsigned long long len = __le64_to_cpu(bsb2.length2)*512; | |
2744 | ||
2745 | if (abuflen < len) { | |
2746 | free(abuf); | |
2747 | free(bbuf); | |
2748 | abuflen = len; | |
2749 | abuf = malloc(abuflen); | |
2750 | bbuf = malloc(abuflen); | |
e86c9dd6 NB |
2751 | } |
2752 | ||
7236ee7a | 2753 | lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0); |
f21e18ca | 2754 | if ((unsigned long long)read(bfd, bbuf, len) != len) |
7236ee7a N |
2755 | fail("read second backup failed"); |
2756 | lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0); | |
f21e18ca | 2757 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
2758 | fail("read second from array failed"); |
2759 | if (memcmp(bbuf, abuf, len) != 0) | |
2760 | fail("data2 compare failed"); | |
e86c9dd6 | 2761 | } |
7236ee7a | 2762 | } |
e86c9dd6 | 2763 | |
999b4972 N |
2764 | int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape, |
2765 | struct supertype *st, unsigned long blocks, | |
2766 | int *fds, unsigned long long *offsets, | |
2767 | int dests, int *destfd, unsigned long long *destoffsets) | |
7236ee7a | 2768 | { |
7443ee81 N |
2769 | /* Monitor a reshape where backup is being performed using |
2770 | * 'native' mechanism - either to a backup file, or | |
2771 | * to some space in a spare. | |
2772 | */ | |
7236ee7a | 2773 | char *buf; |
7443ee81 N |
2774 | int degraded = -1; |
2775 | unsigned long long speed; | |
2776 | unsigned long long suspend_point, array_size; | |
2777 | unsigned long long backup_point, wait_point; | |
2778 | unsigned long long reshape_completed; | |
2779 | int done = 0; | |
2780 | int increasing = reshape->after.data_disks >= reshape->before.data_disks; | |
2781 | int part = 0; /* The next part of the backup area to fill. It may already | |
2782 | * be full, so we need to check */ | |
2783 | int level = reshape->level; | |
2784 | int layout = reshape->before.layout; | |
2785 | int data = reshape->before.data_disks; | |
2786 | int disks = reshape->before.data_disks + reshape->parity; | |
2787 | int chunk = sra->array.chunk_size; | |
da8100ca N |
2788 | struct mdinfo *sd; |
2789 | unsigned long stripes; | |
2790 | ||
2791 | /* set up the backup-super-block. This requires the | |
2792 | * uuid from the array. | |
2793 | */ | |
2794 | /* Find a superblock */ | |
2795 | for (sd = sra->devs; sd; sd = sd->next) { | |
2796 | char *dn; | |
2797 | int devfd; | |
2798 | int ok; | |
2799 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2800 | continue; | |
2801 | dn = map_dev(sd->disk.major, sd->disk.minor, 1); | |
2802 | devfd = dev_open(dn, O_RDONLY); | |
2803 | if (devfd < 0) | |
2804 | continue; | |
2805 | ok = st->ss->load_super(st, devfd, NULL); | |
2806 | close(devfd); | |
2807 | if (ok >= 0) | |
2808 | break; | |
2809 | } | |
2810 | if (!sd) { | |
2811 | fprintf(stderr, Name ": Cannot find a superblock\n"); | |
2812 | return 0; | |
2813 | } | |
2814 | ||
2815 | memset(&bsb, 0, 512); | |
2816 | memcpy(bsb.magic, "md_backup_data-1", 16); | |
2817 | st->ss->uuid_from_super(st, (int*)&bsb.set_uuid); | |
2818 | bsb.mtime = __cpu_to_le64(time(0)); | |
2819 | bsb.devstart2 = blocks; | |
2820 | ||
2821 | stripes = blocks / (sra->array.chunk_size/512) / | |
2822 | reshape->before.data_disks; | |
e86c9dd6 | 2823 | |
fcf57625 N |
2824 | if (posix_memalign((void**)&buf, 4096, disks * chunk)) |
2825 | /* Don't start the 'reshape' */ | |
2826 | return 0; | |
7443ee81 N |
2827 | if (reshape->before.data_disks == reshape->after.data_disks) { |
2828 | sysfs_get_ll(sra, NULL, "sync_speed_min", &speed); | |
2829 | sysfs_set_num(sra, NULL, "sync_speed_min", 200000); | |
2830 | } | |
e86c9dd6 | 2831 | |
7443ee81 | 2832 | if (increasing) { |
96533072 | 2833 | array_size = sra->component_size * reshape->after.data_disks; |
7443ee81 N |
2834 | backup_point = sra->reshape_progress; |
2835 | suspend_point = 0; | |
2836 | } else { | |
96533072 | 2837 | array_size = sra->component_size * reshape->before.data_disks; |
25da62d9 | 2838 | backup_point = reshape->backup_blocks; |
7443ee81 N |
2839 | suspend_point = array_size; |
2840 | } | |
7236ee7a | 2841 | |
7443ee81 N |
2842 | while (!done) { |
2843 | int rv; | |
7236ee7a | 2844 | |
7443ee81 N |
2845 | /* Want to return as soon the oldest backup slot can |
2846 | * be released as that allows us to start backing up | |
2847 | * some more, providing suspend_point has been | |
b6b95155 | 2848 | * advanced, which it should have. |
7443ee81 N |
2849 | */ |
2850 | if (increasing) { | |
2851 | wait_point = array_size; | |
2852 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
2853 | wait_point = (__le64_to_cpu(bsb.arraystart) + | |
2854 | __le64_to_cpu(bsb.length)); | |
2855 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
2856 | wait_point = (__le64_to_cpu(bsb.arraystart2) + | |
2857 | __le64_to_cpu(bsb.length2)); | |
2858 | } else { | |
2859 | wait_point = 0; | |
2860 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
2861 | wait_point = __le64_to_cpu(bsb.arraystart); | |
2862 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
2863 | wait_point = __le64_to_cpu(bsb.arraystart2); | |
2864 | } | |
2865 | ||
2866 | rv = progress_reshape(sra, reshape, | |
2867 | backup_point, wait_point, | |
2868 | &suspend_point, &reshape_completed); | |
7443ee81 N |
2869 | /* external metadata would need to ping_monitor here */ |
2870 | sra->reshape_progress = reshape_completed; | |
7236ee7a | 2871 | |
7443ee81 N |
2872 | /* Clear any backup region that is before 'here' */ |
2873 | if (increasing) { | |
2874 | if (reshape_completed >= (__le64_to_cpu(bsb.arraystart) + | |
2875 | __le64_to_cpu(bsb.length))) | |
2876 | forget_backup(dests, destfd, | |
2877 | destoffsets, 0); | |
2878 | if (reshape_completed >= (__le64_to_cpu(bsb.arraystart2) + | |
2879 | __le64_to_cpu(bsb.length2))) | |
2880 | forget_backup(dests, destfd, | |
2881 | destoffsets, 1); | |
2882 | } else { | |
2883 | if (reshape_completed <= (__le64_to_cpu(bsb.arraystart))) | |
2884 | forget_backup(dests, destfd, | |
2885 | destoffsets, 0); | |
2886 | if (reshape_completed <= (__le64_to_cpu(bsb.arraystart2))) | |
2887 | forget_backup(dests, destfd, | |
2888 | destoffsets, 1); | |
2889 | } | |
7236ee7a | 2890 | |
223c5c99 | 2891 | if (rv < 0) { |
77a73a17 N |
2892 | if (rv == -1) |
2893 | done = 1; | |
223c5c99 N |
2894 | break; |
2895 | } | |
2896 | ||
60204e4e | 2897 | while (rv) { |
7443ee81 | 2898 | unsigned long long offset; |
e97ca002 | 2899 | unsigned long actual_stripes; |
60204e4e N |
2900 | /* Need to backup some data. |
2901 | * If 'part' is not used and the desired | |
2902 | * backup size is suspended, do a backup, | |
2903 | * then consider the next part. | |
2904 | */ | |
7443ee81 N |
2905 | /* Check that 'part' is unused */ |
2906 | if (part == 0 && __le64_to_cpu(bsb.length) != 0) | |
60204e4e | 2907 | break; |
7443ee81 | 2908 | if (part == 1 && __le64_to_cpu(bsb.length2) != 0) |
60204e4e | 2909 | break; |
7443ee81 N |
2910 | |
2911 | offset = backup_point / data; | |
e97ca002 | 2912 | actual_stripes = stripes; |
60204e4e | 2913 | if (increasing) { |
e97ca002 N |
2914 | if (offset + actual_stripes * (chunk/512) > |
2915 | sra->component_size) | |
2916 | actual_stripes = ((sra->component_size - offset) | |
2917 | / (chunk/512)); | |
2918 | if (offset + actual_stripes * (chunk/512) > | |
60204e4e N |
2919 | suspend_point/data) |
2920 | break; | |
2921 | } else { | |
e97ca002 N |
2922 | if (offset < actual_stripes * (chunk/512)) |
2923 | actual_stripes = offset / (chunk/512); | |
2924 | offset -= actual_stripes * (chunk/512); | |
2925 | if (offset < suspend_point/data) | |
60204e4e N |
2926 | break; |
2927 | } | |
e97ca002 | 2928 | grow_backup(sra, offset, actual_stripes, |
7443ee81 N |
2929 | fds, offsets, |
2930 | disks, chunk, level, layout, | |
2931 | dests, destfd, destoffsets, | |
2932 | part, °raded, buf); | |
2933 | validate(afd, destfd[0], destoffsets[0]); | |
2934 | /* record where 'part' is up to */ | |
2935 | part = !part; | |
2936 | if (increasing) | |
e97ca002 | 2937 | backup_point += actual_stripes * (chunk/512) * data; |
7443ee81 | 2938 | else |
e97ca002 | 2939 | backup_point -= actual_stripes * (chunk/512) * data; |
7443ee81 N |
2940 | } |
2941 | } | |
2942 | ||
223c5c99 N |
2943 | /* FIXME maybe call progress_reshape one more time instead */ |
2944 | abort_reshape(sra); /* remove any remaining suspension */ | |
7443ee81 N |
2945 | if (reshape->before.data_disks == reshape->after.data_disks) |
2946 | sysfs_set_num(sra, NULL, "sync_speed_min", speed); | |
7236ee7a | 2947 | free(buf); |
77a73a17 | 2948 | return done; |
e86c9dd6 | 2949 | } |
353632d9 NB |
2950 | |
2951 | /* | |
2952 | * If any spare contains md_back_data-1 which is recent wrt mtime, | |
2953 | * write that data into the array and update the super blocks with | |
2954 | * the new reshape_progress | |
2955 | */ | |
ea0ebe96 N |
2956 | int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt, |
2957 | char *backup_file, int verbose) | |
353632d9 NB |
2958 | { |
2959 | int i, j; | |
2960 | int old_disks; | |
353632d9 | 2961 | unsigned long long *offsets; |
82f2d6ab | 2962 | unsigned long long nstripe, ostripe; |
6e9eac4f | 2963 | int ndata, odata; |
353632d9 | 2964 | |
e9e43ec3 N |
2965 | if (info->new_level != info->array.level) |
2966 | return 1; /* Cannot handle level changes (they are instantaneous) */ | |
2967 | ||
2968 | odata = info->array.raid_disks - info->delta_disks - 1; | |
2969 | if (info->array.level == 6) odata--; /* number of data disks */ | |
2970 | ndata = info->array.raid_disks - 1; | |
2971 | if (info->new_level == 6) ndata--; | |
353632d9 NB |
2972 | |
2973 | old_disks = info->array.raid_disks - info->delta_disks; | |
2974 | ||
e9e43ec3 N |
2975 | if (info->delta_disks <= 0) |
2976 | /* Didn't grow, so the backup file must have | |
2977 | * been used | |
2978 | */ | |
2979 | old_disks = cnt; | |
06b0d786 | 2980 | for (i=old_disks-(backup_file?1:0); i<cnt; i++) { |
353632d9 | 2981 | struct mdinfo dinfo; |
06b0d786 | 2982 | int fd; |
e9e43ec3 | 2983 | int bsbsize; |
ea0ebe96 | 2984 | char *devname, namebuf[20]; |
10af14c4 | 2985 | unsigned long long lo, hi; |
353632d9 NB |
2986 | |
2987 | /* This was a spare and may have some saved data on it. | |
2988 | * Load the superblock, find and load the | |
2989 | * backup_super_block. | |
2990 | * If either fail, go on to next device. | |
2991 | * If the backup contains no new info, just return | |
206c5eae | 2992 | * else restore data and update all superblocks |
353632d9 | 2993 | */ |
06b0d786 NB |
2994 | if (i == old_disks-1) { |
2995 | fd = open(backup_file, O_RDONLY); | |
e9e43ec3 N |
2996 | if (fd<0) { |
2997 | fprintf(stderr, Name ": backup file %s inaccessible: %s\n", | |
2998 | backup_file, strerror(errno)); | |
06b0d786 | 2999 | continue; |
e9e43ec3 | 3000 | } |
ea0ebe96 | 3001 | devname = backup_file; |
06b0d786 NB |
3002 | } else { |
3003 | fd = fdlist[i]; | |
3004 | if (fd < 0) | |
3005 | continue; | |
3da92f27 | 3006 | if (st->ss->load_super(st, fd, NULL)) |
06b0d786 | 3007 | continue; |
353632d9 | 3008 | |
a5d85af7 | 3009 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 NB |
3010 | st->ss->free_super(st); |
3011 | ||
06b0d786 NB |
3012 | if (lseek64(fd, |
3013 | (dinfo.data_offset + dinfo.component_size - 8) <<9, | |
ea0ebe96 N |
3014 | 0) < 0) { |
3015 | fprintf(stderr, Name ": Cannot seek on device %d\n", i); | |
06b0d786 | 3016 | continue; /* Cannot seek */ |
ea0ebe96 N |
3017 | } |
3018 | sprintf(namebuf, "device-%d", i); | |
3019 | devname = namebuf; | |
06b0d786 | 3020 | } |
ea0ebe96 N |
3021 | if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) { |
3022 | if (verbose) | |
3023 | fprintf(stderr, Name ": Cannot read from %s\n", devname); | |
353632d9 | 3024 | continue; /* Cannot read */ |
ea0ebe96 | 3025 | } |
e9e43ec3 | 3026 | if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 && |
ea0ebe96 N |
3027 | memcmp(bsb.magic, "md_backup_data-2", 16) != 0) { |
3028 | if (verbose) | |
3029 | fprintf(stderr, Name ": No backup metadata on %s\n", devname); | |
353632d9 | 3030 | continue; |
ea0ebe96 N |
3031 | } |
3032 | if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) { | |
3033 | if (verbose) | |
3034 | fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname); | |
353632d9 | 3035 | continue; /* bad checksum */ |
ea0ebe96 | 3036 | } |
e9e43ec3 | 3037 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 && |
ea0ebe96 N |
3038 | bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) { |
3039 | if (verbose) | |
3040 | fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname); | |
22e30516 | 3041 | continue; /* Bad second checksum */ |
ea0ebe96 N |
3042 | } |
3043 | if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) { | |
3044 | if (verbose) | |
3045 | fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname); | |
353632d9 | 3046 | continue; /* Wrong uuid */ |
ea0ebe96 | 3047 | } |
353632d9 | 3048 | |
097075b6 N |
3049 | /* array utime and backup-mtime should be updated at much the same time, but it seems that |
3050 | * sometimes they aren't... So allow considerable flexability in matching, and allow | |
3051 | * this test to be overridden by an environment variable. | |
3052 | */ | |
f21e18ca N |
3053 | if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 || |
3054 | info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) { | |
097075b6 N |
3055 | if (check_env("MDADM_GROW_ALLOW_OLD")) { |
3056 | fprintf(stderr, Name ": accepting backup with timestamp %lu " | |
3057 | "for array with timestamp %lu\n", | |
3058 | (unsigned long)__le64_to_cpu(bsb.mtime), | |
3059 | (unsigned long)info->array.utime); | |
3060 | } else { | |
3061 | if (verbose) | |
3062 | fprintf(stderr, Name ": too-old timestamp on " | |
3063 | "backup-metadata on %s\n", devname); | |
3064 | continue; /* time stamp is too bad */ | |
3065 | } | |
ea0ebe96 | 3066 | } |
353632d9 | 3067 | |
e9e43ec3 | 3068 | if (bsb.magic[15] == '1') { |
10af14c4 N |
3069 | if (bsb.length == 0) |
3070 | continue; | |
e7a71c6b N |
3071 | if (info->delta_disks >= 0) { |
3072 | /* reshape_progress is increasing */ | |
3073 | if (__le64_to_cpu(bsb.arraystart) | |
3074 | + __le64_to_cpu(bsb.length) | |
3075 | < info->reshape_progress) { | |
3076 | nonew: | |
3077 | if (verbose) | |
3078 | fprintf(stderr, Name | |
3079 | ": backup-metadata found on %s but is not needed\n", devname); | |
3080 | continue; /* No new data here */ | |
3081 | } | |
3082 | } else { | |
3083 | /* reshape_progress is decreasing */ | |
3084 | if (__le64_to_cpu(bsb.arraystart) >= | |
3085 | info->reshape_progress) | |
3086 | goto nonew; /* No new data here */ | |
ea0ebe96 | 3087 | } |
e9e43ec3 | 3088 | } else { |
10af14c4 N |
3089 | if (bsb.length == 0 && bsb.length2 == 0) |
3090 | continue; | |
e7a71c6b N |
3091 | if (info->delta_disks >= 0) { |
3092 | /* reshape_progress is increasing */ | |
3093 | if ((__le64_to_cpu(bsb.arraystart) | |
3094 | + __le64_to_cpu(bsb.length) | |
3095 | < info->reshape_progress) | |
3096 | && | |
3097 | (__le64_to_cpu(bsb.arraystart2) | |
3098 | + __le64_to_cpu(bsb.length2) | |
3099 | < info->reshape_progress)) | |
3100 | goto nonew; /* No new data here */ | |
3101 | } else { | |
3102 | /* reshape_progress is decreasing */ | |
3103 | if (__le64_to_cpu(bsb.arraystart) >= | |
3104 | info->reshape_progress && | |
3105 | __le64_to_cpu(bsb.arraystart2) >= | |
3106 | info->reshape_progress) | |
3107 | goto nonew; /* No new data here */ | |
3108 | } | |
e9e43ec3 | 3109 | } |
ea0ebe96 N |
3110 | if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) { |
3111 | second_fail: | |
3112 | if (verbose) | |
e7a71c6b N |
3113 | fprintf(stderr, Name |
3114 | ": Failed to verify secondary backup-metadata block on %s\n", | |
ea0ebe96 | 3115 | devname); |
353632d9 | 3116 | continue; /* Cannot seek */ |
ea0ebe96 | 3117 | } |
2efedc7b | 3118 | /* There should be a duplicate backup superblock 4k before here */ |
06b0d786 | 3119 | if (lseek64(fd, -4096, 1) < 0 || |
0155af90 | 3120 | read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2)) |
ea0ebe96 | 3121 | goto second_fail; /* Cannot find leading superblock */ |
e9e43ec3 N |
3122 | if (bsb.magic[15] == '1') |
3123 | bsbsize = offsetof(struct mdp_backup_super, pad1); | |
3124 | else | |
3125 | bsbsize = offsetof(struct mdp_backup_super, pad); | |
ff94fb86 | 3126 | if (memcmp(&bsb2, &bsb, bsbsize) != 0) |
ea0ebe96 | 3127 | goto second_fail; /* Cannot find leading superblock */ |
2efedc7b | 3128 | |
353632d9 NB |
3129 | /* Now need the data offsets for all devices. */ |
3130 | offsets = malloc(sizeof(*offsets)*info->array.raid_disks); | |
3131 | for(j=0; j<info->array.raid_disks; j++) { | |
3132 | if (fdlist[j] < 0) | |
3133 | continue; | |
3da92f27 | 3134 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 NB |
3135 | /* FIXME should be this be an error */ |
3136 | continue; | |
a5d85af7 | 3137 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 | 3138 | st->ss->free_super(st); |
14e5b4d7 | 3139 | offsets[j] = dinfo.data_offset * 512; |
353632d9 NB |
3140 | } |
3141 | printf(Name ": restoring critical section\n"); | |
3142 | ||
3143 | if (restore_stripes(fdlist, offsets, | |
3144 | info->array.raid_disks, | |
3145 | info->new_chunk, | |
3146 | info->new_level, | |
3147 | info->new_layout, | |
06b0d786 | 3148 | fd, __le64_to_cpu(bsb.devstart)*512, |
92dcdf7c | 3149 | __le64_to_cpu(bsb.arraystart)*512, |
e9e43ec3 N |
3150 | __le64_to_cpu(bsb.length)*512)) { |
3151 | /* didn't succeed, so giveup */ | |
ea0ebe96 N |
3152 | if (verbose) |
3153 | fprintf(stderr, Name ": Error restoring backup from %s\n", | |
3154 | devname); | |
e9e43ec3 N |
3155 | return 1; |
3156 | } | |
3157 | ||
3158 | if (bsb.magic[15] == '2' && | |
3159 | restore_stripes(fdlist, offsets, | |
3160 | info->array.raid_disks, | |
3161 | info->new_chunk, | |
3162 | info->new_level, | |
3163 | info->new_layout, | |
3164 | fd, __le64_to_cpu(bsb.devstart)*512 + | |
3165 | __le64_to_cpu(bsb.devstart2)*512, | |
92dcdf7c | 3166 | __le64_to_cpu(bsb.arraystart2)*512, |
e9e43ec3 | 3167 | __le64_to_cpu(bsb.length2)*512)) { |
353632d9 | 3168 | /* didn't succeed, so giveup */ |
ea0ebe96 N |
3169 | if (verbose) |
3170 | fprintf(stderr, Name ": Error restoring second backup from %s\n", | |
3171 | devname); | |
2295250a | 3172 | return 1; |
353632d9 NB |
3173 | } |
3174 | ||
e9e43ec3 | 3175 | |
353632d9 NB |
3176 | /* Ok, so the data is restored. Let's update those superblocks. */ |
3177 | ||
10af14c4 N |
3178 | lo = hi = 0; |
3179 | if (bsb.length) { | |
3180 | lo = __le64_to_cpu(bsb.arraystart); | |
3181 | hi = lo + __le64_to_cpu(bsb.length); | |
3182 | } | |
3183 | if (bsb.magic[15] == '2' && bsb.length2) { | |
3184 | unsigned long long lo1, hi1; | |
3185 | lo1 = __le64_to_cpu(bsb.arraystart2); | |
3186 | hi1 = lo1 + __le64_to_cpu(bsb.length2); | |
3187 | if (lo == hi) { | |
3188 | lo = lo1; | |
3189 | hi = hi1; | |
3190 | } else if (lo < lo1) | |
3191 | hi = hi1; | |
3192 | else | |
3193 | lo = lo1; | |
3194 | } | |
3195 | if (lo < hi && | |
3196 | (info->reshape_progress < lo || | |
3197 | info->reshape_progress > hi)) | |
3198 | /* backup does not affect reshape_progress*/ ; | |
3199 | else if (info->delta_disks >= 0) { | |
e9e43ec3 N |
3200 | info->reshape_progress = __le64_to_cpu(bsb.arraystart) + |
3201 | __le64_to_cpu(bsb.length); | |
3202 | if (bsb.magic[15] == '2') { | |
3203 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) + | |
3204 | __le64_to_cpu(bsb.length2); | |
3205 | if (p2 > info->reshape_progress) | |
3206 | info->reshape_progress = p2; | |
3207 | } | |
3208 | } else { | |
3209 | info->reshape_progress = __le64_to_cpu(bsb.arraystart); | |
3210 | if (bsb.magic[15] == '2') { | |
3211 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2); | |
3212 | if (p2 < info->reshape_progress) | |
3213 | info->reshape_progress = p2; | |
3214 | } | |
3215 | } | |
353632d9 NB |
3216 | for (j=0; j<info->array.raid_disks; j++) { |
3217 | if (fdlist[j] < 0) continue; | |
3da92f27 | 3218 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 | 3219 | continue; |
a5d85af7 | 3220 | st->ss->getinfo_super(st, &dinfo, NULL); |
e9e43ec3 | 3221 | dinfo.reshape_progress = info->reshape_progress; |
3da92f27 | 3222 | st->ss->update_super(st, &dinfo, |
68c7d6d7 NB |
3223 | "_reshape_progress", |
3224 | NULL,0, 0, NULL); | |
3da92f27 NB |
3225 | st->ss->store_super(st, fdlist[j]); |
3226 | st->ss->free_super(st); | |
353632d9 | 3227 | } |
353632d9 NB |
3228 | return 0; |
3229 | } | |
6e9eac4f NB |
3230 | /* Didn't find any backup data, try to see if any |
3231 | * was needed. | |
3232 | */ | |
82f2d6ab N |
3233 | if (info->delta_disks < 0) { |
3234 | /* When shrinking, the critical section is at the end. | |
3235 | * So see if we are before the critical section. | |
3236 | */ | |
3237 | unsigned long long first_block; | |
3238 | nstripe = ostripe = 0; | |
3239 | first_block = 0; | |
3240 | while (ostripe >= nstripe) { | |
3241 | ostripe += info->array.chunk_size / 512; | |
3242 | first_block = ostripe * odata; | |
3243 | nstripe = first_block / ndata / (info->new_chunk/512) * | |
3244 | (info->new_chunk/512); | |
3245 | } | |
3246 | ||
3247 | if (info->reshape_progress >= first_block) | |
3248 | return 0; | |
6e9eac4f | 3249 | } |
82f2d6ab N |
3250 | if (info->delta_disks > 0) { |
3251 | /* See if we are beyond the critical section. */ | |
3252 | unsigned long long last_block; | |
3253 | nstripe = ostripe = 0; | |
3254 | last_block = 0; | |
3255 | while (nstripe >= ostripe) { | |
3256 | nstripe += info->new_chunk / 512; | |
3257 | last_block = nstripe * ndata; | |
3258 | ostripe = last_block / odata / (info->array.chunk_size/512) * | |
3259 | (info->array.chunk_size/512); | |
3260 | } | |
6e9eac4f | 3261 | |
82f2d6ab N |
3262 | if (info->reshape_progress >= last_block) |
3263 | return 0; | |
3264 | } | |
6e9eac4f | 3265 | /* needed to recover critical section! */ |
ea0ebe96 N |
3266 | if (verbose) |
3267 | fprintf(stderr, Name ": Failed to find backup of critical section\n"); | |
2295250a | 3268 | return 1; |
353632d9 | 3269 | } |
e9e43ec3 N |
3270 | |
3271 | int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info, | |
3272 | char *backup_file) | |
3273 | { | |
7443ee81 | 3274 | int err = sysfs_set_str(info, NULL, "array_state", "readonly"); |
e9e43ec3 N |
3275 | if (err) |
3276 | return err; | |
a93f87ee | 3277 | return reshape_array(NULL, mdfd, "array", st, info, 1, backup_file, 0, 0, 1); |
e9e43ec3 N |
3278 | } |
3279 | ||
3280 |