]>
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, | |
41784c88 AK |
568 | int delta_disks, char *backup_file, char *dev, |
569 | int verbose) | |
7bc71196 DW |
570 | { |
571 | /* nothing extra to check in the native case */ | |
572 | if (!st->ss->external) | |
573 | return 0; | |
574 | if (!st->ss->reshape_super || | |
575 | !st->ss->manage_reshape) { | |
576 | fprintf(stderr, Name ": %s metadata does not support reshape\n", | |
577 | st->ss->name); | |
578 | return 1; | |
579 | } | |
580 | ||
581 | return st->ss->reshape_super(st, size, level, layout, chunksize, | |
41784c88 AK |
582 | raid_disks, delta_disks, backup_file, dev, |
583 | verbose); | |
7bc71196 DW |
584 | } |
585 | ||
586 | static void sync_metadata(struct supertype *st) | |
587 | { | |
588 | if (st->ss->external) { | |
76266030 | 589 | if (st->update_tail) { |
7bc71196 | 590 | flush_metadata_updates(st); |
76266030 N |
591 | st->update_tail = &st->updates; |
592 | } else | |
7bc71196 DW |
593 | st->ss->sync_metadata(st); |
594 | } | |
595 | } | |
596 | ||
597 | static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n) | |
598 | { | |
599 | /* when dealing with external metadata subarrays we need to be | |
600 | * prepared to handle EAGAIN. The kernel may need to wait for | |
601 | * mdmon to mark the array active so the kernel can handle | |
602 | * allocations/writeback when preparing the reshape action | |
603 | * (md_allow_write()). We temporarily disable safe_mode_delay | |
604 | * to close a race with the array_state going clean before the | |
605 | * next write to raid_disks / stripe_cache_size | |
606 | */ | |
607 | char safe[50]; | |
608 | int rc; | |
609 | ||
610 | /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */ | |
5da9ab98 N |
611 | if (!container || |
612 | (strcmp(name, "raid_disks") != 0 && | |
613 | strcmp(name, "stripe_cache_size") != 0)) | |
7bc71196 DW |
614 | return sysfs_set_num(sra, NULL, name, n); |
615 | ||
616 | rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe)); | |
617 | if (rc <= 0) | |
618 | return -1; | |
619 | sysfs_set_num(sra, NULL, "safe_mode_delay", 0); | |
620 | rc = sysfs_set_num(sra, NULL, name, n); | |
621 | if (rc < 0 && errno == EAGAIN) { | |
622 | ping_monitor(container); | |
623 | /* if we get EAGAIN here then the monitor is not active | |
624 | * so stop trying | |
625 | */ | |
626 | rc = sysfs_set_num(sra, NULL, name, n); | |
627 | } | |
628 | sysfs_set_str(sra, NULL, "safe_mode_delay", safe); | |
629 | return rc; | |
630 | } | |
631 | ||
1e971e71 | 632 | int start_reshape(struct mdinfo *sra, int already_running) |
47eb4d5a N |
633 | { |
634 | int err; | |
0f28668b | 635 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); |
47eb4d5a N |
636 | err = sysfs_set_num(sra, NULL, "suspend_hi", 0); |
637 | err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
1e971e71 N |
638 | if (!already_running) |
639 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
47eb4d5a | 640 | err = err ?: sysfs_set_num(sra, NULL, "sync_max", 0); |
1e971e71 N |
641 | if (!already_running) |
642 | err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape"); | |
47eb4d5a N |
643 | |
644 | return err; | |
645 | } | |
646 | ||
647 | void abort_reshape(struct mdinfo *sra) | |
648 | { | |
649 | sysfs_set_str(sra, NULL, "sync_action", "idle"); | |
650 | sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL); | |
651 | sysfs_set_num(sra, NULL, "suspend_hi", 0); | |
652 | sysfs_set_num(sra, NULL, "suspend_lo", 0); | |
653 | sysfs_set_num(sra, NULL, "sync_min", 0); | |
654 | sysfs_set_str(sra, NULL, "sync_max", "max"); | |
655 | } | |
656 | ||
dfe77a9e KW |
657 | int remove_disks_for_takeover(struct supertype *st, |
658 | struct mdinfo *sra, | |
659 | int layout) | |
62a48395 AK |
660 | { |
661 | int nr_of_copies; | |
662 | struct mdinfo *remaining; | |
663 | int slot; | |
664 | ||
dfe77a9e KW |
665 | if (sra->array.level == 10) |
666 | nr_of_copies = layout & 0xff; | |
667 | else if (sra->array.level == 1) | |
668 | nr_of_copies = sra->array.raid_disks; | |
669 | else | |
670 | return 1; | |
62a48395 AK |
671 | |
672 | remaining = sra->devs; | |
673 | sra->devs = NULL; | |
674 | /* for each 'copy', select one device and remove from the list. */ | |
675 | for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) { | |
676 | struct mdinfo **diskp; | |
677 | int found = 0; | |
678 | ||
679 | /* Find a working device to keep */ | |
680 | for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) { | |
681 | struct mdinfo *disk = *diskp; | |
682 | ||
683 | if (disk->disk.raid_disk < slot) | |
684 | continue; | |
685 | if (disk->disk.raid_disk >= slot + nr_of_copies) | |
686 | continue; | |
687 | if (disk->disk.state & (1<<MD_DISK_REMOVED)) | |
688 | continue; | |
689 | if (disk->disk.state & (1<<MD_DISK_FAULTY)) | |
690 | continue; | |
691 | if (!(disk->disk.state & (1<<MD_DISK_SYNC))) | |
692 | continue; | |
693 | ||
694 | /* We have found a good disk to use! */ | |
695 | *diskp = disk->next; | |
696 | disk->next = sra->devs; | |
697 | sra->devs = disk; | |
698 | found = 1; | |
699 | break; | |
700 | } | |
701 | if (!found) | |
702 | break; | |
703 | } | |
704 | ||
705 | if (slot < sra->array.raid_disks) { | |
706 | /* didn't find all slots */ | |
707 | struct mdinfo **e; | |
708 | e = &remaining; | |
709 | while (*e) | |
710 | e = &(*e)->next; | |
711 | *e = sra->devs; | |
712 | sra->devs = remaining; | |
713 | return 1; | |
714 | } | |
715 | ||
716 | /* Remove all 'remaining' devices from the array */ | |
717 | while (remaining) { | |
718 | struct mdinfo *sd = remaining; | |
719 | remaining = sd->next; | |
720 | ||
721 | sysfs_set_str(sra, sd, "state", "faulty"); | |
722 | sysfs_set_str(sra, sd, "slot", "none"); | |
d5ca4a23 KW |
723 | /* for external metadata disks should be removed in mdmon */ |
724 | if (!st->ss->external) | |
725 | sysfs_set_str(sra, sd, "state", "remove"); | |
62a48395 AK |
726 | sd->disk.state |= (1<<MD_DISK_REMOVED); |
727 | sd->disk.state &= ~(1<<MD_DISK_SYNC); | |
728 | sd->next = sra->devs; | |
729 | sra->devs = sd; | |
730 | } | |
731 | return 0; | |
732 | } | |
733 | ||
130994cb AK |
734 | void reshape_free_fdlist(int *fdlist, |
735 | unsigned long long *offsets, | |
736 | int size) | |
737 | { | |
738 | int i; | |
739 | ||
740 | for (i = 0; i < size; i++) | |
741 | if (fdlist[i] >= 0) | |
742 | close(fdlist[i]); | |
743 | ||
744 | free(fdlist); | |
745 | free(offsets); | |
746 | } | |
747 | ||
748 | int reshape_prepare_fdlist(char *devname, | |
749 | struct mdinfo *sra, | |
750 | int raid_disks, | |
751 | int nrdisks, | |
752 | unsigned long blocks, | |
753 | char *backup_file, | |
754 | int *fdlist, | |
755 | unsigned long long *offsets) | |
756 | { | |
757 | int d = 0; | |
758 | struct mdinfo *sd; | |
759 | ||
760 | for (d = 0; d <= nrdisks; d++) | |
761 | fdlist[d] = -1; | |
762 | d = raid_disks; | |
763 | for (sd = sra->devs; sd; sd = sd->next) { | |
764 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
765 | continue; | |
766 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
767 | char *dn = map_dev(sd->disk.major, | |
768 | sd->disk.minor, 1); | |
769 | fdlist[sd->disk.raid_disk] | |
770 | = dev_open(dn, O_RDONLY); | |
771 | offsets[sd->disk.raid_disk] = sd->data_offset*512; | |
772 | if (fdlist[sd->disk.raid_disk] < 0) { | |
773 | fprintf(stderr, | |
774 | Name ": %s: cannot open component %s\n", | |
775 | devname, dn ? dn : "-unknown-"); | |
776 | d = -1; | |
777 | goto release; | |
778 | } | |
779 | } else if (backup_file == NULL) { | |
780 | /* spare */ | |
781 | char *dn = map_dev(sd->disk.major, | |
782 | sd->disk.minor, 1); | |
783 | fdlist[d] = dev_open(dn, O_RDWR); | |
784 | offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512; | |
785 | if (fdlist[d] < 0) { | |
786 | fprintf(stderr, Name ": %s: cannot open component %s\n", | |
787 | devname, dn ? dn : "-unknown-"); | |
788 | d = -1; | |
789 | goto release; | |
790 | } | |
791 | d++; | |
792 | } | |
793 | } | |
794 | release: | |
795 | return d; | |
796 | } | |
797 | ||
e6e9d47b AK |
798 | int reshape_open_backup_file(char *backup_file, |
799 | int fd, | |
800 | char *devname, | |
801 | long blocks, | |
802 | int *fdlist, | |
a93f87ee N |
803 | unsigned long long *offsets, |
804 | int restart) | |
e6e9d47b AK |
805 | { |
806 | /* Return 1 on success, 0 on any form of failure */ | |
807 | /* need to check backup file is large enough */ | |
808 | char buf[512]; | |
809 | struct stat stb; | |
810 | unsigned int dev; | |
811 | int i; | |
812 | ||
a93f87ee | 813 | *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL), |
e6e9d47b AK |
814 | S_IRUSR | S_IWUSR); |
815 | *offsets = 8 * 512; | |
816 | if (*fdlist < 0) { | |
817 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
818 | devname, backup_file, strerror(errno)); | |
819 | return 0; | |
820 | } | |
821 | /* Guard against backup file being on array device. | |
822 | * If array is partitioned or if LVM etc is in the | |
823 | * way this will not notice, but it is better than | |
824 | * nothing. | |
825 | */ | |
826 | fstat(*fdlist, &stb); | |
827 | dev = stb.st_dev; | |
828 | fstat(fd, &stb); | |
829 | if (stb.st_rdev == dev) { | |
830 | fprintf(stderr, Name ": backup file must NOT be" | |
831 | " on the array being reshaped.\n"); | |
832 | close(*fdlist); | |
833 | return 0; | |
834 | } | |
835 | ||
836 | memset(buf, 0, 512); | |
ca4fe0bf | 837 | for (i=0; i < blocks + 8 ; i++) { |
e6e9d47b AK |
838 | if (write(*fdlist, buf, 512) != 512) { |
839 | fprintf(stderr, Name ": %s: cannot create" | |
840 | " backup file %s: %s\n", | |
841 | devname, backup_file, strerror(errno)); | |
842 | return 0; | |
843 | } | |
844 | } | |
845 | if (fsync(*fdlist) != 0) { | |
846 | fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n", | |
847 | devname, backup_file, strerror(errno)); | |
848 | return 0; | |
849 | } | |
850 | ||
851 | return 1; | |
852 | } | |
853 | ||
1c009fc2 AK |
854 | unsigned long compute_backup_blocks(int nchunk, int ochunk, |
855 | unsigned int ndata, unsigned int odata) | |
856 | { | |
857 | unsigned long a, b, blocks; | |
858 | /* So how much do we need to backup. | |
859 | * We need an amount of data which is both a whole number of | |
860 | * old stripes and a whole number of new stripes. | |
861 | * So LCM for (chunksize*datadisks). | |
862 | */ | |
863 | a = (ochunk/512) * odata; | |
864 | b = (nchunk/512) * ndata; | |
865 | /* Find GCD */ | |
866 | while (a != b) { | |
867 | if (a < b) | |
868 | b -= a; | |
869 | if (b < a) | |
870 | a -= b; | |
871 | } | |
872 | /* LCM == product / GCD */ | |
873 | blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a; | |
874 | ||
875 | return blocks; | |
876 | } | |
877 | ||
5da9ab98 N |
878 | char *analyse_change(struct mdinfo *info, struct reshape *re) |
879 | { | |
880 | /* Based on the current array state in info->array and | |
881 | * the changes in info->new_* etc, determine: | |
882 | * - whether the change is possible | |
883 | * - Intermediate level/raid_disks/layout | |
884 | * - whether a restriping reshape is needed | |
885 | * - number of sectors in minimum change unit. This | |
886 | * will cover a whole number of stripes in 'before' and | |
887 | * 'after'. | |
888 | * | |
889 | * Return message if the change should be rejected | |
890 | * NULL if the change can be achieved | |
891 | * | |
892 | * This can be called as part of starting a reshape, or | |
893 | * when assembling an array that is undergoing reshape. | |
894 | */ | |
895 | int new_disks; | |
b545e14a N |
896 | /* delta_parity records change in number of devices |
897 | * caused by level change | |
898 | */ | |
899 | int delta_parity = 0; | |
5da9ab98 N |
900 | |
901 | /* If a new level not explicitly given, we assume no-change */ | |
902 | if (info->new_level == UnSet) | |
903 | info->new_level = info->array.level; | |
904 | ||
905 | if (info->new_chunk) | |
906 | switch (info->new_level) { | |
907 | case 0: | |
908 | case 4: | |
909 | case 5: | |
910 | case 6: | |
911 | case 10: | |
912 | /* chunk size is meaningful, must divide component_size | |
913 | * evenly | |
914 | */ | |
915 | if (info->component_size % (info->new_chunk/512)) | |
916 | return "New chunk size does not" | |
917 | " divide component size"; | |
918 | break; | |
919 | default: | |
920 | return "chunk size not meaningful for this level"; | |
921 | } | |
922 | else | |
923 | info->new_chunk = info->array.chunk_size; | |
924 | ||
925 | switch (info->array.level) { | |
926 | case 1: | |
927 | /* RAID1 can convert to RAID1 with different disks, or | |
dfe77a9e KW |
928 | * raid5 with 2 disks, or |
929 | * raid0 with 1 disk | |
5da9ab98 | 930 | */ |
dfe77a9e | 931 | if (info->new_level == 0) { |
b545e14a N |
932 | if (info->delta_disks != UnSet && |
933 | info->delta_disks != 0) | |
934 | return "Cannot change number of disks " | |
935 | "with RAID1->RAID0 conversion"; | |
dfe77a9e KW |
936 | re->level = 0; |
937 | re->before.data_disks = 1; | |
938 | re->after.data_disks = 1; | |
939 | re->before.layout = 0; | |
940 | re->backup_blocks = 0; | |
941 | re->parity = 0; | |
942 | return NULL; | |
943 | } | |
5da9ab98 N |
944 | if (info->new_level == 1) { |
945 | if (info->delta_disks == UnSet) | |
946 | /* Don't know what to do */ | |
947 | return "no change requested for Growing RAID1"; | |
948 | re->level = 1; | |
b4f8e38b | 949 | re->backup_blocks = 0; |
5da9ab98 N |
950 | re->parity = 0; |
951 | return NULL; | |
952 | } | |
953 | if (info->array.raid_disks == 2 && | |
446d2a5a | 954 | info->new_level == 5) { |
5652f2d9 | 955 | |
5da9ab98 | 956 | re->level = 5; |
5da9ab98 | 957 | re->before.data_disks = 1; |
5652f2d9 N |
958 | if (info->delta_disks != UnSet && |
959 | info->delta_disks != 0) | |
960 | re->after.data_disks = 1 + info->delta_disks; | |
961 | else | |
962 | re->after.data_disks = 1; | |
963 | if (re->after.data_disks < 1) | |
964 | return "Number of disks too small for RAID5"; | |
965 | ||
5da9ab98 | 966 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC; |
446d2a5a N |
967 | info->array.chunk_size = 65536; |
968 | break; | |
5da9ab98 N |
969 | } |
970 | /* Could do some multi-stage conversions, but leave that to | |
971 | * later. | |
972 | */ | |
973 | return "Impossibly level change request for RAID1"; | |
974 | ||
975 | case 10: | |
976 | /* RAID10 can only be converted from near mode to | |
977 | * RAID0 by removing some devices | |
978 | */ | |
979 | if ((info->array.layout & ~0xff) != 0x100) | |
980 | return "Cannot Grow RAID10 with far/offset layout"; | |
981 | /* number of devices must be multiple of number of copies */ | |
982 | if (info->array.raid_disks % (info->array.layout & 0xff)) | |
983 | return "RAID10 layout too complex for Grow operation"; | |
984 | ||
985 | if (info->new_level != 0) | |
986 | return "RAID10 can only be changed to RAID0"; | |
987 | new_disks = (info->array.raid_disks | |
988 | / (info->array.layout & 0xff)); | |
b545e14a | 989 | if (info->delta_disks == UnSet) |
5da9ab98 N |
990 | info->delta_disks = (new_disks |
991 | - info->array.raid_disks); | |
b545e14a | 992 | |
5da9ab98 N |
993 | if (info->delta_disks != new_disks - info->array.raid_disks) |
994 | return "New number of raid-devices impossible for RAID10"; | |
995 | if (info->new_chunk && | |
996 | info->new_chunk != info->array.chunk_size) | |
997 | return "Cannot change chunk-size with RAID10 Grow"; | |
998 | ||
999 | /* looks good */ | |
1000 | re->level = 0; | |
1001 | re->parity = 0; | |
1002 | re->before.data_disks = new_disks; | |
031d445c | 1003 | re->after.data_disks = re->before.data_disks; |
5da9ab98 | 1004 | re->before.layout = 0; |
b4f8e38b | 1005 | re->backup_blocks = 0; |
5da9ab98 N |
1006 | return NULL; |
1007 | ||
1008 | case 0: | |
1009 | /* RAID0 can be converted to RAID10, or to RAID456 */ | |
1010 | if (info->new_level == 10) { | |
1011 | if (info->new_layout == UnSet && info->delta_disks == UnSet) { | |
1012 | /* Assume near=2 layout */ | |
1013 | info->new_layout = 0x102; | |
1014 | info->delta_disks = info->array.raid_disks; | |
1015 | } | |
1016 | if (info->new_layout == UnSet) { | |
1017 | int copies = 1 + (info->delta_disks | |
1018 | / info->array.raid_disks); | |
1019 | if (info->array.raid_disks * (copies-1) | |
1020 | != info->delta_disks) | |
1021 | return "Impossible number of devices" | |
1022 | " for RAID0->RAID10"; | |
1023 | info->new_layout = 0x100 + copies; | |
1024 | } | |
1025 | if (info->delta_disks == UnSet) { | |
1026 | int copies = info->new_layout & 0xff; | |
1027 | if (info->new_layout != 0x100 + copies) | |
1028 | return "New layout impossible" | |
1029 | " for RAID0->RAID10";; | |
1030 | info->delta_disks = (copies - 1) * | |
1031 | info->array.raid_disks; | |
1032 | } | |
1033 | if (info->new_chunk && | |
1034 | info->new_chunk != info->array.chunk_size) | |
1035 | return "Cannot change chunk-size with RAID0->RAID10"; | |
1036 | /* looks good */ | |
1037 | re->level = 10; | |
1038 | re->parity = 0; | |
1039 | re->before.data_disks = (info->array.raid_disks + | |
1040 | info->delta_disks); | |
031d445c | 1041 | re->after.data_disks = re->before.data_disks; |
5da9ab98 | 1042 | re->before.layout = info->new_layout; |
b4f8e38b | 1043 | re->backup_blocks = 0; |
5da9ab98 N |
1044 | return NULL; |
1045 | } | |
1046 | ||
1047 | /* RAID0 can also covert to RAID0/4/5/6 by first converting to | |
1048 | * a raid4 style layout of the final level. | |
1049 | */ | |
1050 | switch (info->new_level) { | |
5da9ab98 | 1051 | case 4: |
b545e14a N |
1052 | delta_parity = 1; |
1053 | case 0: | |
5da9ab98 N |
1054 | re->level = 4; |
1055 | re->before.layout = 0; | |
1056 | break; | |
1057 | case 5: | |
b545e14a | 1058 | delta_parity = 1; |
5da9ab98 N |
1059 | re->level = 5; |
1060 | re->before.layout = ALGORITHM_PARITY_N; | |
1061 | break; | |
1062 | case 6: | |
b545e14a | 1063 | delta_parity = 2; |
5da9ab98 N |
1064 | re->level = 6; |
1065 | re->before.layout = ALGORITHM_PARITY_N; | |
1066 | break; | |
1067 | default: | |
1068 | return "Impossible level change requested"; | |
1069 | } | |
1070 | re->before.data_disks = info->array.raid_disks; | |
1071 | /* determining 'after' layout happens outside this 'switch' */ | |
1072 | break; | |
1073 | ||
1074 | case 4: | |
1075 | info->array.layout = ALGORITHM_PARITY_N; | |
1076 | case 5: | |
1077 | switch (info->new_level) { | |
f0cce442 | 1078 | case 0: |
b545e14a | 1079 | delta_parity = -1; |
5da9ab98 N |
1080 | case 4: |
1081 | re->level = info->array.level; | |
1082 | re->before.data_disks = info->array.raid_disks - 1; | |
1083 | re->before.layout = info->array.layout; | |
1084 | break; | |
1085 | case 5: | |
1086 | re->level = 5; | |
1087 | re->before.data_disks = info->array.raid_disks - 1; | |
1088 | re->before.layout = info->array.layout; | |
1089 | break; | |
1090 | case 6: | |
b545e14a | 1091 | delta_parity = 1; |
5da9ab98 N |
1092 | re->level = 6; |
1093 | re->before.data_disks = info->array.raid_disks - 1; | |
1094 | switch (info->array.layout) { | |
1095 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1096 | re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
1097 | break; | |
1098 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1099 | re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
1100 | break; | |
1101 | case ALGORITHM_LEFT_SYMMETRIC: | |
1102 | re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
1103 | break; | |
1104 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1105 | re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
1106 | break; | |
1107 | case ALGORITHM_PARITY_0: | |
1108 | re->before.layout = ALGORITHM_PARITY_0_6; | |
1109 | break; | |
1110 | case ALGORITHM_PARITY_N: | |
1111 | re->before.layout = ALGORITHM_PARITY_N_6; | |
1112 | break; | |
1113 | default: | |
1114 | return "Cannot convert an array with this layout"; | |
1115 | } | |
1116 | break; | |
1117 | case 1: | |
1118 | if (info->array.raid_disks != 2) | |
1119 | return "Can only convert a 2-device array to RAID1"; | |
b545e14a N |
1120 | if (info->delta_disks != UnSet && |
1121 | info->delta_disks != 0) | |
1122 | return "Cannot set raid_disk when " | |
1123 | "converting RAID5->RAID1"; | |
5da9ab98 | 1124 | re->level = 1; |
5da9ab98 N |
1125 | break; |
1126 | default: | |
1127 | return "Impossible level change requested"; | |
1128 | } | |
1129 | break; | |
1130 | case 6: | |
1131 | switch (info->new_level) { | |
1132 | case 4: | |
1133 | case 5: | |
b545e14a | 1134 | delta_parity = -1; |
5da9ab98 N |
1135 | case 6: |
1136 | re->level = 6; | |
1137 | re->before.data_disks = info->array.raid_disks - 2; | |
1138 | re->before.layout = info->array.layout; | |
1139 | break; | |
1140 | default: | |
1141 | return "Impossible level change requested"; | |
1142 | } | |
1143 | break; | |
1144 | } | |
1145 | ||
1146 | /* If we reached here then it looks like a re-stripe is | |
1147 | * happening. We have determined the intermediate level | |
1148 | * and initial raid_disks/layout and stored these in 're'. | |
1149 | * | |
1150 | * We need to deduce the final layout that can be atomically | |
1151 | * converted to the end state. | |
1152 | */ | |
1153 | switch (info->new_level) { | |
1154 | case 0: | |
1155 | /* We can only get to RAID0 from RAID4 or RAID5 | |
1156 | * with appropriate layout and one extra device | |
1157 | */ | |
1158 | if (re->level != 4 && re->level != 5) | |
1159 | return "Cannot covert to RAID0 from this level"; | |
b545e14a | 1160 | |
5da9ab98 N |
1161 | switch (re->level) { |
1162 | case 4: | |
1163 | re->after.layout = 0 ; break; | |
1164 | case 5: | |
1165 | re->after.layout = ALGORITHM_PARITY_N; break; | |
1166 | } | |
1167 | break; | |
1168 | ||
1169 | case 4: | |
1170 | /* We can only get to RAID4 from RAID5 */ | |
1171 | if (re->level != 4 && re->level != 5) | |
1172 | return "Cannot convert to RAID4 from this level"; | |
b545e14a | 1173 | |
5da9ab98 N |
1174 | switch (re->level) { |
1175 | case 4: | |
1176 | re->after.layout = 0 ; break; | |
1177 | case 5: | |
1178 | re->after.layout = ALGORITHM_PARITY_N; break; | |
1179 | } | |
1180 | break; | |
1181 | ||
1182 | case 5: | |
1183 | /* We get to RAID5 for RAID5 or RAID6 */ | |
1184 | if (re->level != 5 && re->level != 6) | |
1185 | return "Cannot convert to RAID5 from this level"; | |
b545e14a | 1186 | |
5da9ab98 N |
1187 | switch (re->level) { |
1188 | case 5: | |
1189 | if (info->new_layout == UnSet) | |
1190 | re->after.layout = re->before.layout; | |
1191 | else | |
1192 | re->after.layout = info->new_layout; | |
1193 | break; | |
1194 | case 6: | |
4a870364 N |
1195 | if (info->new_layout == UnSet) |
1196 | info->new_layout = re->before.layout; | |
1197 | ||
5da9ab98 N |
1198 | /* after.layout needs to be raid6 version of new_layout */ |
1199 | if (info->new_layout == ALGORITHM_PARITY_N) | |
1200 | re->after.layout = ALGORITHM_PARITY_N; | |
1201 | else { | |
1202 | char layout[40]; | |
1203 | char *ls = map_num(r5layout, info->new_layout); | |
1204 | int l; | |
1205 | strcat(strcpy(layout, ls), "-6"); | |
1206 | l = map_name(r6layout, layout); | |
1207 | if (l == UnSet) | |
1208 | return "Cannot find RAID6 layout" | |
1209 | " to convert to"; | |
1210 | re->after.layout = l; | |
1211 | } | |
1212 | } | |
1213 | break; | |
1214 | ||
1215 | case 6: | |
1216 | /* We must already be at level 6 */ | |
1217 | if (re->level != 6) | |
1218 | return "Impossible level change"; | |
5da9ab98 | 1219 | if (info->new_layout == UnSet) |
4a870364 | 1220 | re->after.layout = info->array.layout; |
5da9ab98 N |
1221 | else |
1222 | re->after.layout = info->new_layout; | |
1223 | break; | |
1224 | default: | |
1225 | return "Impossible level change requested"; | |
1226 | } | |
b545e14a N |
1227 | if (info->delta_disks == UnSet) |
1228 | info->delta_disks = delta_parity; | |
1229 | ||
1230 | re->after.data_disks = (re->before.data_disks | |
1231 | + info->delta_disks | |
1232 | - delta_parity); | |
5da9ab98 N |
1233 | switch (re->level) { |
1234 | case 6: re->parity = 2; break; | |
1235 | case 4: | |
1236 | case 5: re->parity = 1; break; | |
1237 | default: re->parity = 0; break; | |
1238 | } | |
1239 | /* So we have a restripe operation, we need to calculate the number | |
1240 | * of blocks per reshape operation. | |
1241 | */ | |
1242 | if (info->new_chunk == 0) | |
1243 | info->new_chunk = info->array.chunk_size; | |
1244 | if (re->after.data_disks == re->before.data_disks && | |
1245 | re->after.layout == re->before.layout && | |
1246 | info->new_chunk == info->array.chunk_size) { | |
1247 | /* Nothing to change */ | |
b4f8e38b | 1248 | re->backup_blocks = 0; |
5da9ab98 N |
1249 | return NULL; |
1250 | } | |
1251 | if (re->after.data_disks == 1 && re->before.data_disks == 1) { | |
446d2a5a | 1252 | /* chunk and layout changes make no difference */ |
b4f8e38b | 1253 | re->backup_blocks = 0; |
5da9ab98 N |
1254 | return NULL; |
1255 | } | |
1256 | ||
1257 | if (re->after.data_disks == re->before.data_disks && | |
1258 | get_linux_version() < 2006032) | |
1259 | return "in-place reshape is not safe before 2.6.32 - sorry."; | |
1260 | ||
1261 | if (re->after.data_disks < re->before.data_disks && | |
1262 | get_linux_version() < 2006030) | |
1263 | return "reshape to fewer devices is not supported before 2.6.32 - sorry."; | |
1264 | ||
b4f8e38b | 1265 | re->backup_blocks = compute_backup_blocks( |
5da9ab98 N |
1266 | info->new_chunk, info->array.chunk_size, |
1267 | re->after.data_disks, | |
1268 | re->before.data_disks); | |
1269 | ||
1270 | re->new_size = info->component_size * re->after.data_disks; | |
1271 | return NULL; | |
1272 | } | |
1273 | ||
1274 | static int reshape_array(char *container, int fd, char *devname, | |
1275 | struct supertype *st, struct mdinfo *info, | |
a93f87ee N |
1276 | int force, char *backup_file, int quiet, int forked, |
1277 | int restart); | |
493f5dd6 | 1278 | static int reshape_container(char *container, char *devname, |
5da9ab98 N |
1279 | struct supertype *st, |
1280 | struct mdinfo *info, | |
1281 | int force, | |
1282 | char *backup_file, | |
493f5dd6 | 1283 | int quiet, int restart); |
1c009fc2 | 1284 | |
06b0d786 | 1285 | int Grow_reshape(char *devname, int fd, int quiet, char *backup_file, |
e86c9dd6 | 1286 | long long size, |
691a36b7 N |
1287 | int level, char *layout_str, int chunksize, int raid_disks, |
1288 | int force) | |
e86c9dd6 NB |
1289 | { |
1290 | /* Make some changes in the shape of an array. | |
1291 | * The kernel must support the change. | |
7236ee7a N |
1292 | * |
1293 | * There are three different changes. Each can trigger | |
1294 | * a resync or recovery so we freeze that until we have | |
1295 | * requested everything (if kernel supports freezing - 2.6.30). | |
1296 | * The steps are: | |
1297 | * - change size (i.e. component_size) | |
1298 | * - change level | |
1299 | * - change layout/chunksize/ndisks | |
1300 | * | |
1301 | * The last can require a reshape. It is different on different | |
1302 | * levels so we need to check the level before actioning it. | |
1303 | * Some times the level change needs to be requested after the | |
1304 | * reshape (e.g. raid6->raid5, raid5->raid0) | |
1305 | * | |
e86c9dd6 | 1306 | */ |
5da9ab98 | 1307 | struct mdu_array_info_s array; |
7236ee7a | 1308 | int rv = 0; |
e86c9dd6 | 1309 | struct supertype *st; |
4725bc31 | 1310 | char *subarray = NULL; |
e86c9dd6 | 1311 | |
7236ee7a | 1312 | int frozen; |
7236ee7a | 1313 | int changed = 0; |
7bc71196 | 1314 | char *container = NULL; |
5da9ab98 | 1315 | char container_buf[20]; |
7bc71196 | 1316 | int cfd = -1; |
e86c9dd6 | 1317 | |
5da9ab98 | 1318 | struct mdinfo info; |
7e0f6979 | 1319 | struct mdinfo *sra; |
e86c9dd6 NB |
1320 | |
1321 | if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) { | |
1322 | fprintf(stderr, Name ": %s is not an active md array - aborting\n", | |
1323 | devname); | |
1324 | return 1; | |
1325 | } | |
24d40069 | 1326 | |
9ce510be N |
1327 | if (size >= 0 && |
1328 | (chunksize || level!= UnSet || layout_str || raid_disks)) { | |
1329 | fprintf(stderr, Name ": cannot change component size at the same time " | |
1330 | "as other changes.\n" | |
1331 | " Change size first, then check data is intact before " | |
1332 | "making other changes.\n"); | |
1333 | return 1; | |
1334 | } | |
1335 | ||
24d40069 N |
1336 | if (raid_disks && raid_disks < array.raid_disks && array.level > 1 && |
1337 | get_linux_version() < 2006032 && | |
1338 | !check_env("MDADM_FORCE_FEWER")) { | |
1339 | fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n" | |
1340 | " Please use a newer kernel\n"); | |
1341 | return 1; | |
1342 | } | |
7bc71196 DW |
1343 | |
1344 | st = super_by_fd(fd, &subarray); | |
1345 | if (!st) { | |
1346 | fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname); | |
1347 | return 1; | |
1348 | } | |
acab7bb1 N |
1349 | if (raid_disks > st->max_devs) { |
1350 | fprintf(stderr, Name ": Cannot increase raid-disks on this array" | |
1351 | " beyond %d\n", st->max_devs); | |
1352 | return 1; | |
1353 | } | |
7bc71196 DW |
1354 | |
1355 | /* in the external case we need to check that the requested reshape is | |
1356 | * supported, and perform an initial check that the container holds the | |
1357 | * pre-requisite spare devices (mdmon owns final validation) | |
1358 | */ | |
1359 | if (st->ss->external) { | |
1360 | int container_dev; | |
7f2ba464 | 1361 | int rv; |
7bc71196 DW |
1362 | |
1363 | if (subarray) { | |
1364 | container_dev = st->container_dev; | |
1365 | cfd = open_dev_excl(st->container_dev); | |
7bc71196 DW |
1366 | } else { |
1367 | container_dev = st->devnum; | |
1368 | close(fd); | |
1369 | cfd = open_dev_excl(st->devnum); | |
1370 | fd = cfd; | |
1371 | } | |
1372 | if (cfd < 0) { | |
1373 | fprintf(stderr, Name ": Unable to open container for %s\n", | |
1374 | devname); | |
7f2ba464 | 1375 | free(subarray); |
7bc71196 DW |
1376 | return 1; |
1377 | } | |
1378 | ||
5da9ab98 N |
1379 | fmt_devname(container_buf, container_dev); |
1380 | container = container_buf; | |
7bc71196 | 1381 | |
eb744788 AK |
1382 | rv = st->ss->load_container(st, cfd, NULL); |
1383 | ||
7f2ba464 | 1384 | if (rv) { |
7bc71196 DW |
1385 | fprintf(stderr, Name ": Cannot read superblock for %s\n", |
1386 | devname); | |
7f2ba464 | 1387 | free(subarray); |
7bc71196 DW |
1388 | return 1; |
1389 | } | |
1390 | ||
1391 | if (mdmon_running(container_dev)) | |
1392 | st->update_tail = &st->updates; | |
691a36b7 N |
1393 | } |
1394 | ||
1395 | if (raid_disks > array.raid_disks && | |
1396 | array.spare_disks < (raid_disks - array.raid_disks) && | |
1397 | !force) { | |
1398 | fprintf(stderr, | |
1399 | Name ": Need %d spare%s to avoid degraded array," | |
1400 | " and only have %d.\n" | |
1401 | " Use --force to over-ride this check.\n", | |
1402 | raid_disks - array.raid_disks, | |
1403 | raid_disks - array.raid_disks == 1 ? "" : "s", | |
1404 | array.spare_disks); | |
1405 | return 1; | |
7bc71196 DW |
1406 | } |
1407 | ||
3fa436ac ML |
1408 | sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS |
1409 | | GET_STATE | GET_VERSION); | |
1410 | if (sra) { | |
7f2ba464 | 1411 | if (st->ss->external && subarray == NULL) { |
7bc71196 DW |
1412 | array.level = LEVEL_CONTAINER; |
1413 | sra->array.level = LEVEL_CONTAINER; | |
1414 | } | |
7bc71196 | 1415 | } else { |
b526e52d DW |
1416 | fprintf(stderr, Name ": failed to read sysfs parameters for %s\n", |
1417 | devname); | |
1418 | return 1; | |
1419 | } | |
7f2ba464 DW |
1420 | frozen = freeze(st); |
1421 | if (frozen < -1) { | |
1422 | /* freeze() already spewed the reason */ | |
1423 | return 1; | |
1424 | } else if (frozen < 0) { | |
7236ee7a N |
1425 | fprintf(stderr, Name ": %s is performing resync/recovery and cannot" |
1426 | " be reshaped\n", devname); | |
1427 | return 1; | |
1428 | } | |
1429 | ||
1430 | /* ========= set size =============== */ | |
1431 | if (size >= 0 && (size == 0 || size != array.size)) { | |
7bc71196 DW |
1432 | long long orig_size = array.size; |
1433 | ||
41784c88 AK |
1434 | if (reshape_super(st, size, UnSet, UnSet, 0, 0, UnSet, NULL, |
1435 | devname, !quiet)) { | |
7bc71196 DW |
1436 | rv = 1; |
1437 | goto release; | |
1438 | } | |
1439 | sync_metadata(st); | |
7236ee7a N |
1440 | array.size = size; |
1441 | if (array.size != size) { | |
1442 | /* got truncated to 32bit, write to | |
1443 | * component_size instead | |
1444 | */ | |
1445 | if (sra) | |
1446 | rv = sysfs_set_num(sra, NULL, | |
1447 | "component_size", size); | |
1448 | else | |
1449 | rv = -1; | |
1450 | } else | |
1451 | rv = ioctl(fd, SET_ARRAY_INFO, &array); | |
1452 | if (rv != 0) { | |
39bbb392 | 1453 | int err = errno; |
7bc71196 DW |
1454 | |
1455 | /* restore metadata */ | |
1456 | if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0, | |
41784c88 | 1457 | UnSet, NULL, devname, !quiet) == 0) |
7bc71196 | 1458 | sync_metadata(st); |
7236ee7a | 1459 | fprintf(stderr, Name ": Cannot set device size for %s: %s\n", |
39bbb392 N |
1460 | devname, strerror(err)); |
1461 | if (err == EBUSY && | |
1462 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1463 | fprintf(stderr, " Bitmap must be removed before size can be changed\n"); | |
7236ee7a N |
1464 | rv = 1; |
1465 | goto release; | |
1466 | } | |
1467 | ioctl(fd, GET_ARRAY_INFO, &array); | |
be1cabbd | 1468 | size = get_component_size(fd)/2; |
f98841b3 N |
1469 | if (size == 0) |
1470 | size = array.size; | |
7236ee7a | 1471 | if (!quiet) |
f98841b3 N |
1472 | fprintf(stderr, Name ": component size of %s has been set to %lluK\n", |
1473 | devname, size); | |
7236ee7a | 1474 | changed = 1; |
7bc71196 | 1475 | } else if (array.level != LEVEL_CONTAINER) { |
be1cabbd | 1476 | size = get_component_size(fd)/2; |
f98841b3 N |
1477 | if (size == 0) |
1478 | size = array.size; | |
7236ee7a N |
1479 | } |
1480 | ||
dfe77a9e | 1481 | /* ========= check for Raid10/Raid1 -> Raid0 conversion =============== |
5da9ab98 | 1482 | * current implementation assumes that following conditions must be met: |
dfe77a9e KW |
1483 | * - RAID10: |
1484 | * - far_copies == 1 | |
1485 | * - near_copies == 2 | |
62a48395 | 1486 | */ |
dfe77a9e KW |
1487 | if ((level == 0 && array.level == 10 && sra && |
1488 | array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) || | |
1489 | (level == 0 && array.level == 1 && sra)) { | |
62a48395 | 1490 | int err; |
dfe77a9e | 1491 | err = remove_disks_for_takeover(st, sra, array.layout); |
62a48395 AK |
1492 | if (err) { |
1493 | dprintf(Name": Array cannot be reshaped\n"); | |
62a48395 AK |
1494 | if (cfd > -1) |
1495 | close(cfd); | |
5da9ab98 N |
1496 | rv = 1; |
1497 | goto release; | |
7236ee7a | 1498 | } |
bb025c2f KW |
1499 | /* FIXME this is added with no justification - why is it here */ |
1500 | ping_monitor(container); | |
7236ee7a N |
1501 | } |
1502 | ||
5da9ab98 | 1503 | info.array = array; |
7443ee81 | 1504 | sysfs_init(&info, fd, NoMdDev); |
3fa436ac | 1505 | strcpy(info.text_version, sra->text_version); |
7443ee81 | 1506 | info.component_size = size*2; |
5da9ab98 N |
1507 | info.new_level = level; |
1508 | info.new_chunk = chunksize * 1024; | |
1509 | if (raid_disks) | |
1510 | info.delta_disks = raid_disks - info.array.raid_disks; | |
1511 | else | |
1512 | info.delta_disks = UnSet; | |
1513 | if (layout_str == NULL) { | |
1514 | info.new_layout = UnSet; | |
1515 | if (info.array.level == 6 && | |
1516 | (info.new_level == 6 || info.new_level == UnSet) && | |
1517 | info.array.layout >= 16) { | |
7236ee7a | 1518 | fprintf(stderr, Name |
5da9ab98 N |
1519 | ": %s has a non-standard layout. If you" |
1520 | " wish to preserve this\n" | |
1521 | " during the reshape, please specify" | |
1522 | " --layout=preserve\n" | |
1523 | " If you want to change it, specify a" | |
1524 | " layout or use --layout=normalise\n", | |
7236ee7a N |
1525 | devname); |
1526 | rv = 1; | |
1527 | goto release; | |
1528 | } | |
5da9ab98 N |
1529 | } else if (strcmp(layout_str, "normalise") == 0 || |
1530 | strcmp(layout_str, "normalize") == 0) { | |
1531 | /* If we have a -6 RAID6 layout, remove the '-6'. */ | |
1532 | info.new_layout = UnSet; | |
1533 | if (info.array.level == 6 && info.new_level == UnSet) { | |
1534 | char l[40], *h; | |
1535 | strcpy(l, map_num(r6layout, info.array.layout)); | |
1536 | h = strrchr(l, '-'); | |
1537 | if (h && strcmp(h, "-6") == 0) { | |
1538 | *h = 0; | |
1539 | info.new_layout = map_name(r6layout, l); | |
7236ee7a N |
1540 | } |
1541 | } | |
5da9ab98 N |
1542 | } else if (strcmp(layout_str, "preserve") == 0) { |
1543 | info.new_layout = UnSet; | |
1544 | } else { | |
1545 | int l = info.new_level; | |
1546 | if (l == UnSet) | |
1547 | l = info.array.level; | |
1548 | switch (l) { | |
1549 | case 5: | |
1550 | info.new_layout = map_name(r5layout, layout_str); | |
1551 | break; | |
1552 | case 6: | |
1553 | info.new_layout = map_name(r6layout, layout_str); | |
1554 | break; | |
1555 | case 10: | |
1556 | info.new_layout = parse_layout_10(layout_str); | |
1557 | break; | |
1558 | case LEVEL_FAULTY: | |
1559 | info.new_layout = parse_layout_faulty(layout_str); | |
1560 | break; | |
1561 | default: | |
1562 | fprintf(stderr, Name ": layout not meaningful" | |
1563 | " with this level\n"); | |
7bc71196 DW |
1564 | rv = 1; |
1565 | goto release; | |
1566 | } | |
5da9ab98 N |
1567 | if (info.new_layout == UnSet) { |
1568 | fprintf(stderr, Name ": layout %s not understood" | |
1569 | " for this level\n", | |
1570 | layout_str); | |
1571 | rv = 1; | |
1572 | goto release; | |
7bc71196 | 1573 | } |
7236ee7a N |
1574 | } |
1575 | ||
5da9ab98 N |
1576 | if (array.level == LEVEL_CONTAINER) { |
1577 | /* This change is to be applied to every array in the | |
1578 | * container. This is only needed when the metadata imposes | |
1579 | * restraints of the various arrays in the container. | |
1580 | * Currently we only know that IMSM requires all arrays | |
1581 | * to have the same number of devices so changing the | |
1582 | * number of devices (On-Line Capacity Expansion) must be | |
1583 | * performed at the level of the container | |
1584 | */ | |
493f5dd6 N |
1585 | rv = reshape_container(container, devname, st, &info, |
1586 | force, backup_file, quiet, 0); | |
9202b8eb | 1587 | frozen = 0; |
5da9ab98 | 1588 | } else { |
08f9e34b AK |
1589 | /* get spare devices from external metadata |
1590 | */ | |
1591 | if (st->ss->external) { | |
1592 | struct mdinfo *info2; | |
1593 | ||
1594 | info2 = st->ss->container_content(st, subarray); | |
1595 | if (info2) { | |
1596 | info.array.spare_disks = | |
1597 | info2->array.spare_disks; | |
1598 | sysfs_free(info2); | |
1599 | } | |
1600 | } | |
1601 | ||
5da9ab98 N |
1602 | /* Impose these changes on a single array. First |
1603 | * check that the metadata is OK with the change. */ | |
7bc71196 | 1604 | |
5da9ab98 N |
1605 | if (reshape_super(st, info.component_size, info.new_level, |
1606 | info.new_layout, info.new_chunk, | |
41784c88 | 1607 | info.array.raid_disks, info.delta_disks, |
5da9ab98 | 1608 | backup_file, devname, quiet)) { |
7bc71196 DW |
1609 | rv = 1; |
1610 | goto release; | |
1611 | } | |
5da9ab98 N |
1612 | sync_metadata(st); |
1613 | rv = reshape_array(container, fd, devname, st, &info, force, | |
a93f87ee | 1614 | backup_file, quiet, 0, 0); |
9202b8eb | 1615 | frozen = 0; |
5da9ab98 N |
1616 | } |
1617 | release: | |
9202b8eb N |
1618 | if (frozen > 0) |
1619 | unfreeze(st); | |
5da9ab98 N |
1620 | return rv; |
1621 | } | |
7bc71196 | 1622 | |
5da9ab98 N |
1623 | static int reshape_array(char *container, int fd, char *devname, |
1624 | struct supertype *st, struct mdinfo *info, | |
1625 | int force, | |
a93f87ee N |
1626 | char *backup_file, int quiet, int forked, |
1627 | int restart) | |
5da9ab98 N |
1628 | { |
1629 | struct reshape reshape; | |
1630 | int spares_needed; | |
1631 | char *msg; | |
1632 | int orig_level = UnSet; | |
1633 | int disks, odisks; | |
7bc71196 | 1634 | |
5da9ab98 N |
1635 | struct mdu_array_info_s array; |
1636 | char *c; | |
e86c9dd6 | 1637 | |
5da9ab98 N |
1638 | int *fdlist; |
1639 | unsigned long long *offsets; | |
1640 | int d; | |
1641 | int nrdisks; | |
1642 | int err; | |
da8100ca | 1643 | unsigned long blocks; |
5da9ab98 N |
1644 | unsigned long cache; |
1645 | unsigned long long array_size; | |
1646 | int done; | |
cf6ac177 | 1647 | struct mdinfo *sra = NULL; |
7bc71196 | 1648 | |
5da9ab98 N |
1649 | msg = analyse_change(info, &reshape); |
1650 | if (msg) { | |
1651 | fprintf(stderr, Name ": %s\n", msg); | |
631d7405 | 1652 | goto release; |
5da9ab98 | 1653 | } |
20a40eca N |
1654 | if (reshape.level != info->array.level || |
1655 | reshape.before.layout != info->array.layout || | |
1656 | reshape.before.data_disks + reshape.parity != info->array.raid_disks) { | |
1657 | fprintf(stderr, Name ": reshape info is not in native format -" | |
1658 | " cannot continue.\n"); | |
1659 | goto release; | |
1660 | } | |
9ff87c16 | 1661 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { |
20c8698d | 1662 | dprintf("Cannot get array information.\n"); |
631d7405 | 1663 | goto release; |
9ff87c16 | 1664 | } |
a93f87ee N |
1665 | |
1666 | if (restart) { | |
1667 | /* reshape already started. just skip to monitoring the reshape */ | |
1668 | if (reshape.backup_blocks == 0) | |
1669 | return 0; | |
1670 | goto started; | |
1671 | } | |
a9c3e78f AK |
1672 | /* The container is frozen but the array may not be. |
1673 | * So freeze the array so spares don't get put to the wrong use | |
1674 | * FIXME there should probably be a cleaner separation between | |
1675 | * freeze_array and freeze_container. | |
1676 | */ | |
1677 | sysfs_freeze_array(info); | |
5da9ab98 N |
1678 | spares_needed = max(reshape.before.data_disks, |
1679 | reshape.after.data_disks) | |
1680 | + reshape.parity - array.raid_disks; | |
7bc71196 | 1681 | |
88c1a083 | 1682 | if (!force && |
d6bd632c | 1683 | info->new_level > 1 && |
88c1a083 | 1684 | spares_needed > info->array.spare_disks) { |
5da9ab98 N |
1685 | fprintf(stderr, |
1686 | Name ": Need %d spare%s to avoid degraded array," | |
1687 | " and only have %d.\n" | |
1688 | " Use --force to over-ride this check.\n", | |
1689 | spares_needed, | |
1690 | spares_needed == 1 ? "" : "s", | |
1691 | info->array.spare_disks); | |
631d7405 | 1692 | goto release; |
7bc71196 | 1693 | } |
e86c9dd6 | 1694 | |
5da9ab98 N |
1695 | if (reshape.level != info->array.level) { |
1696 | char *c = map_num(pers, reshape.level); | |
1697 | int err; | |
1698 | if (c == NULL) | |
631d7405 | 1699 | goto release; |
e86c9dd6 | 1700 | |
5da9ab98 N |
1701 | err = sysfs_set_str(info, NULL, "level", c); |
1702 | if (err) { | |
1703 | err = errno; | |
1704 | fprintf(stderr, Name ": %s: could not set level to %s\n", | |
1705 | devname, c); | |
1706 | if (err == EBUSY && | |
1707 | (info->array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1708 | fprintf(stderr, " Bitmap must be removed" | |
1709 | " before level can be changed\n"); | |
631d7405 | 1710 | goto release; |
19678e53 | 1711 | } |
5da9ab98 | 1712 | if (!quiet) |
b6b95155 | 1713 | fprintf(stderr, Name ": level of %s changed to %s\n", |
5da9ab98 N |
1714 | devname, c); |
1715 | orig_level = info->array.level; | |
3cd4e7c4 | 1716 | sysfs_freeze_array(info); |
e86c9dd6 | 1717 | |
16d4d84e AK |
1718 | if (reshape.level > 0 && st->ss->external) { |
1719 | /* make sure mdmon is aware of the new level */ | |
1720 | if (!mdmon_running(st->container_dev)) | |
1721 | start_mdmon(st->container_dev); | |
1722 | ping_monitor(container); | |
1723 | } | |
5da9ab98 | 1724 | } |
5da9ab98 N |
1725 | /* ->reshape_super might have chosen some spares from the |
1726 | * container that it wants to be part of the new array. | |
1727 | * We can collect them with ->container_content and give | |
1728 | * them to the kernel. | |
1729 | */ | |
1730 | if (st->ss->reshape_super && st->ss->container_content) { | |
1731 | char *subarray = strchr(info->text_version+1, '/')+1; | |
1732 | struct mdinfo *info2 = | |
1733 | st->ss->container_content(st, subarray); | |
1734 | struct mdinfo *d; | |
1735 | ||
2dd0d697 AK |
1736 | if (info2) { |
1737 | sysfs_init(info2, fd, st->devnum); | |
5da9ab98 N |
1738 | for (d = info2->devs; d; d = d->next) { |
1739 | if (d->disk.state == 0 && | |
1740 | d->disk.raid_disk >= 0) { | |
1741 | /* This is a spare that wants to | |
1742 | * be part of the array. | |
1743 | */ | |
1744 | add_disk(fd, st, info2, d); | |
1745 | } | |
9860f271 | 1746 | } |
2dd0d697 AK |
1747 | sysfs_free(info2); |
1748 | } | |
5da9ab98 | 1749 | } |
1686dc25 | 1750 | |
b4f8e38b | 1751 | if (reshape.backup_blocks == 0) { |
5da9ab98 N |
1752 | /* No restriping needed, but we might need to impose |
1753 | * some more changes: layout, raid_disks, chunk_size | |
e86c9dd6 | 1754 | */ |
24aebf3a KW |
1755 | /* read current array info */ |
1756 | if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) { | |
83732c28 | 1757 | dprintf("Cannot get array information.\n"); |
24aebf3a KW |
1758 | goto release; |
1759 | } | |
1760 | /* compare current array info with new values and if | |
1761 | * it is different update them to new */ | |
5da9ab98 | 1762 | if (info->new_layout != UnSet && |
24aebf3a KW |
1763 | info->new_layout != array.layout) { |
1764 | array.layout = info->new_layout; | |
1765 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
5da9ab98 | 1766 | fprintf(stderr, Name ": failed to set new layout\n"); |
631d7405 | 1767 | goto release; |
5da9ab98 N |
1768 | } else if (!quiet) |
1769 | printf("layout for %s set to %d\n", | |
24aebf3a | 1770 | devname, array.layout); |
5da9ab98 N |
1771 | } |
1772 | if (info->delta_disks != UnSet && | |
24aebf3a KW |
1773 | info->delta_disks != 0 && |
1774 | array.raid_disks != (info->array.raid_disks + info->delta_disks)) { | |
1775 | array.raid_disks += info->delta_disks; | |
1776 | if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) { | |
5da9ab98 | 1777 | fprintf(stderr, Name ": failed to set raid disks\n"); |
631d7405 | 1778 | goto release; |
24aebf3a | 1779 | } else if (!quiet) { |
5da9ab98 | 1780 | printf("raid_disks for %s set to %d\n", |
24aebf3a KW |
1781 | devname, array.raid_disks); |
1782 | } | |
5da9ab98 N |
1783 | } |
1784 | if (info->new_chunk != 0 && | |
24aebf3a | 1785 | info->new_chunk != array.chunk_size) { |
5da9ab98 N |
1786 | if (sysfs_set_num(info, NULL, |
1787 | "chunk_size", info->new_chunk) != 0) { | |
1788 | fprintf(stderr, Name ": failed to set chunk size\n"); | |
631d7405 | 1789 | goto release; |
5da9ab98 N |
1790 | } else if (!quiet) |
1791 | printf("chunk size for %s set to %d\n", | |
24aebf3a | 1792 | devname, array.chunk_size); |
4725bc31 | 1793 | } |
e35b189b | 1794 | unfreeze(st); |
631d7405 | 1795 | return 0; |
5da9ab98 | 1796 | } |
19678e53 | 1797 | |
5da9ab98 N |
1798 | /* |
1799 | * There are three possibilities. | |
1800 | * 1/ The array will shrink. | |
1801 | * We need to ensure the reshape will pause before reaching | |
1802 | * the 'critical section'. We also need to fork and wait for | |
1803 | * that to happen. When it does we | |
1804 | * suspend/backup/complete/unfreeze | |
1805 | * | |
1806 | * 2/ The array will not change size. | |
1807 | * This requires that we keep a backup of a sliding window | |
1808 | * so that we can restore data after a crash. So we need | |
1809 | * to fork and monitor progress. | |
1810 | * In future we will allow the data_offset to change, so | |
1811 | * a sliding backup becomes unnecessary. | |
1812 | * | |
1813 | * 3/ The array will grow. This is relatively easy. | |
1814 | * However the kernel's restripe routines will cheerfully | |
1815 | * overwrite some early data before it is safe. So we | |
1816 | * need to make a backup of the early parts of the array | |
1817 | * and be ready to restore it if rebuild aborts very early. | |
1818 | * For externally managed metadata, we still need a forked | |
1819 | * child to monitor the reshape and suspend IO over the region | |
1820 | * that is being reshaped. | |
1821 | * | |
1822 | * We backup data by writing it to one spare, or to a | |
1823 | * file which was given on command line. | |
1824 | * | |
1825 | * In each case, we first make sure that storage is available | |
1826 | * for the required backup. | |
1827 | * Then we: | |
1828 | * - request the shape change. | |
1829 | * - fork to handle backup etc. | |
1830 | */ | |
a93f87ee | 1831 | started: |
5da9ab98 N |
1832 | /* Check that we can hold all the data */ |
1833 | get_dev_size(fd, NULL, &array_size); | |
1834 | if (reshape.new_size < (array_size/512)) { | |
1835 | fprintf(stderr, | |
1836 | Name ": this change will reduce the size of the array.\n" | |
1837 | " use --grow --array-size first to truncate array.\n" | |
1838 | " e.g. mdadm --grow %s --array-size %llu\n", | |
1839 | devname, reshape.new_size/2); | |
5da9ab98 N |
1840 | goto release; |
1841 | } | |
7236ee7a | 1842 | |
5da9ab98 | 1843 | sra = sysfs_read(fd, 0, |
3656edcd | 1844 | GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK| |
5da9ab98 | 1845 | GET_CACHE); |
5da9ab98 N |
1846 | if (!sra) { |
1847 | fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n", | |
1848 | devname); | |
5da9ab98 N |
1849 | goto release; |
1850 | } | |
7236ee7a | 1851 | |
5da9ab98 N |
1852 | /* Decide how many blocks (sectors) for a reshape |
1853 | * unit. The number we have so far is just a minimum | |
1854 | */ | |
b4f8e38b | 1855 | blocks = reshape.backup_blocks; |
5da9ab98 N |
1856 | if (reshape.before.data_disks == |
1857 | reshape.after.data_disks) { | |
1858 | /* Make 'blocks' bigger for better throughput, but | |
1859 | * not so big that we reject it below. | |
1860 | * Try for 16 megabytes | |
1861 | */ | |
1862 | while (blocks * 32 < sra->component_size && | |
1863 | blocks < 16*1024*2) | |
1864 | blocks *= 2; | |
1865 | } else | |
1866 | fprintf(stderr, Name ": Need to backup %luK of critical " | |
1867 | "section..\n", blocks/2); | |
1868 | ||
1869 | if (blocks >= sra->component_size/2) { | |
1870 | fprintf(stderr, Name ": %s: Something wrong" | |
1871 | " - reshape aborted\n", | |
1872 | devname); | |
5da9ab98 N |
1873 | goto release; |
1874 | } | |
7236ee7a | 1875 | |
5da9ab98 N |
1876 | /* Now we need to open all these devices so we can read/write. |
1877 | */ | |
b8b286a6 N |
1878 | nrdisks = max(reshape.before.data_disks, |
1879 | reshape.after.data_disks) + reshape.parity | |
1880 | + sra->array.spare_disks; | |
5da9ab98 N |
1881 | fdlist = malloc((1+nrdisks) * sizeof(int)); |
1882 | offsets = malloc((1+nrdisks) * sizeof(offsets[0])); | |
1883 | if (!fdlist || !offsets) { | |
1884 | fprintf(stderr, Name ": malloc failed: grow aborted\n"); | |
5da9ab98 N |
1885 | goto release; |
1886 | } | |
e380d3be | 1887 | |
b8b286a6 N |
1888 | odisks = reshape.before.data_disks + reshape.parity; |
1889 | d = reshape_prepare_fdlist(devname, sra, odisks, | |
5da9ab98 N |
1890 | nrdisks, blocks, backup_file, |
1891 | fdlist, offsets); | |
1892 | if (d < 0) { | |
5da9ab98 N |
1893 | goto release; |
1894 | } | |
1895 | if (backup_file == NULL) { | |
374ba335 | 1896 | if (reshape.after.data_disks <= reshape.before.data_disks) { |
5da9ab98 N |
1897 | fprintf(stderr, |
1898 | Name ": %s: Cannot grow - need backup-file\n", | |
c03ef02d | 1899 | devname); |
5da9ab98 N |
1900 | goto release; |
1901 | } else if (sra->array.spare_disks == 0) { | |
1902 | fprintf(stderr, Name ": %s: Cannot grow - need a spare or " | |
1903 | "backup-file to backup critical section\n", | |
e86c9dd6 | 1904 | devname); |
5da9ab98 | 1905 | goto release; |
e86c9dd6 | 1906 | } |
5da9ab98 N |
1907 | } else { |
1908 | if (!reshape_open_backup_file(backup_file, fd, devname, | |
1909 | (signed)blocks, | |
a93f87ee | 1910 | fdlist+d, offsets+d, restart)) { |
130994cb | 1911 | goto release; |
e86c9dd6 | 1912 | } |
5da9ab98 N |
1913 | d++; |
1914 | } | |
7236ee7a | 1915 | |
5da9ab98 N |
1916 | /* lastly, check that the internal stripe cache is |
1917 | * large enough, or it won't work. | |
1918 | * It must hold at least 4 stripes of the larger | |
1919 | * chunk size | |
1920 | */ | |
1921 | cache = max(info->array.chunk_size, info->new_chunk); | |
1922 | cache *= 4; /* 4 stripes minimum */ | |
1923 | cache /= 512; /* convert to sectors */ | |
1924 | disks = min(reshape.before.data_disks, reshape.after.data_disks); | |
1925 | /* make sure there is room for 'blocks' with a bit to spare */ | |
1926 | if (cache < 16 + blocks / disks) | |
1927 | cache = 16 + blocks / disks; | |
1928 | cache /= (4096/512); /* Covert from sectors to pages */ | |
1929 | ||
1930 | if (sra->cache_size < cache) | |
1931 | subarray_set_num(container, sra, "stripe_cache_size", | |
1932 | cache+1); | |
1933 | ||
1934 | /* Right, everything seems fine. Let's kick things off. | |
1935 | * If only changing raid_disks, use ioctl, else use | |
1936 | * sysfs. | |
1937 | */ | |
1938 | sync_metadata(st); | |
1939 | ||
b4f8e38b | 1940 | sra->new_chunk = info->new_chunk; |
ddee071d | 1941 | |
20a40eca | 1942 | if (restart) |
ddee071d | 1943 | sra->reshape_progress = info->reshape_progress; |
f897078e N |
1944 | else { |
1945 | sra->reshape_progress = 0; | |
1946 | if (reshape.after.data_disks < reshape.before.data_disks) | |
1947 | /* start from the end of the new array */ | |
1948 | sra->reshape_progress = (sra->component_size | |
1949 | * reshape.after.data_disks); | |
1950 | } | |
1951 | ||
1952 | if (info->array.chunk_size == info->new_chunk && | |
7477d513 AK |
1953 | reshape.before.layout == reshape.after.layout && |
1954 | st->ss->external == 0) { | |
f897078e | 1955 | /* use SET_ARRAY_INFO but only if reshape hasn't started */ |
6ef421be | 1956 | ioctl(fd, GET_ARRAY_INFO, &array); |
5da9ab98 | 1957 | array.raid_disks = reshape.after.data_disks + reshape.parity; |
20a40eca | 1958 | if (!restart && |
f897078e | 1959 | ioctl(fd, SET_ARRAY_INFO, &array) != 0) { |
5da9ab98 | 1960 | int err = errno; |
631d7405 | 1961 | |
5da9ab98 N |
1962 | fprintf(stderr, |
1963 | Name ": Cannot set device shape for %s: %s\n", | |
1964 | devname, strerror(errno)); | |
7bc71196 | 1965 | |
5da9ab98 N |
1966 | if (err == EBUSY && |
1967 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1968 | fprintf(stderr, | |
1969 | " Bitmap must be removed before" | |
1970 | " shape can be changed\n"); | |
b5420ef3 | 1971 | |
5da9ab98 N |
1972 | goto release; |
1973 | } | |
20a40eca | 1974 | } else if (!restart) { |
5da9ab98 | 1975 | /* set them all just in case some old 'new_*' value |
f897078e | 1976 | * persists from some earlier problem. |
7236ee7a | 1977 | */ |
631d7405 | 1978 | int err = 0; |
5da9ab98 | 1979 | if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0) |
631d7405 N |
1980 | err = errno; |
1981 | if (!err && sysfs_set_num(sra, NULL, "layout", | |
5da9ab98 | 1982 | reshape.after.layout) < 0) |
631d7405 N |
1983 | err = errno; |
1984 | if (!err && subarray_set_num(container, sra, "raid_disks", | |
5da9ab98 N |
1985 | reshape.after.data_disks + |
1986 | reshape.parity) < 0) | |
631d7405 N |
1987 | err = errno; |
1988 | if (err) { | |
5da9ab98 N |
1989 | fprintf(stderr, Name ": Cannot set device shape for %s\n", |
1990 | devname); | |
7236ee7a | 1991 | |
5da9ab98 N |
1992 | if (err == EBUSY && |
1993 | (array.state & (1<<MD_SB_BITMAP_PRESENT))) | |
1994 | fprintf(stderr, | |
1995 | " Bitmap must be removed before" | |
1996 | " shape can be changed\n"); | |
1997 | goto release; | |
e86c9dd6 | 1998 | } |
5da9ab98 | 1999 | } |
e86c9dd6 | 2000 | |
20a40eca | 2001 | err = start_reshape(sra, restart); |
fab32c97 | 2002 | if (err) { |
20a40eca N |
2003 | fprintf(stderr, |
2004 | Name ": Cannot %s reshape for %s\n", | |
2005 | restart ? "continue" : "start", | |
fab32c97 AK |
2006 | devname); |
2007 | goto release; | |
2008 | } | |
a93f87ee N |
2009 | if (restart) |
2010 | sysfs_set_str(sra, NULL, "array_state", "active"); | |
5da9ab98 N |
2011 | |
2012 | /* Now we just need to kick off the reshape and watch, while | |
2013 | * handling backups of the data... | |
2014 | * This is all done by a forked background process. | |
2015 | */ | |
2016 | switch(forked ? 0 : fork()) { | |
6b2d630c N |
2017 | case -1: |
2018 | fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n", | |
2019 | strerror(errno)); | |
6b2d630c | 2020 | abort_reshape(sra); |
631d7405 | 2021 | goto release; |
6b2d630c N |
2022 | default: |
2023 | return 0; | |
5da9ab98 | 2024 | case 0: |
6b2d630c N |
2025 | break; |
2026 | } | |
5da9ab98 | 2027 | |
6b2d630c N |
2028 | close(fd); |
2029 | if (check_env("MDADM_GROW_VERIFY")) | |
2030 | fd = open(devname, O_RDONLY | O_DIRECT); | |
2031 | else | |
2032 | fd = -1; | |
2033 | mlockall(MCL_FUTURE); | |
5da9ab98 | 2034 | |
6b2d630c N |
2035 | if (st->ss->external) { |
2036 | /* metadata handler takes it from here */ | |
2037 | done = st->ss->manage_reshape( | |
2038 | fd, sra, &reshape, st, blocks, | |
2039 | fdlist, offsets, | |
2040 | d - odisks, fdlist+odisks, | |
2041 | offsets+odisks); | |
2042 | } else | |
2043 | done = child_monitor( | |
2044 | fd, sra, &reshape, st, blocks, | |
2045 | fdlist, offsets, | |
2046 | d - odisks, fdlist+odisks, | |
2047 | offsets+odisks); | |
2048 | ||
2049 | if (backup_file && done) | |
2050 | unlink(backup_file); | |
2051 | if (!done) { | |
2052 | abort_reshape(sra); | |
2053 | goto out; | |
2054 | } | |
1cbc5680 N |
2055 | |
2056 | if (!st->ss->external && | |
2057 | !(reshape.before.data_disks != reshape.after.data_disks | |
2058 | && info->custom_array_size) && | |
2059 | info->new_level == reshape.level && | |
2060 | !forked) { | |
2061 | /* no need to wait for the reshape to finish as | |
2062 | * there is nothing more to do. | |
2063 | */ | |
2064 | exit(0); | |
2065 | } | |
2066 | wait_reshape(sra); | |
2067 | ||
2068 | if (st->ss->external) { | |
2069 | /* Re-load the metadata as much could have changed */ | |
2070 | int cfd = open_dev(st->container_dev); | |
2071 | if (cfd >= 0) { | |
2072 | ping_monitor(container); | |
2073 | st->ss->free_super(st); | |
2074 | st->ss->load_container(st, cfd, container); | |
2075 | close(cfd); | |
2076 | } | |
2077 | } | |
2078 | ||
6b2d630c N |
2079 | /* set new array size if required customer_array_size is used |
2080 | * by this metadata. | |
2081 | */ | |
2082 | if (reshape.before.data_disks != | |
2083 | reshape.after.data_disks && | |
2084 | info->custom_array_size) { | |
2085 | struct mdinfo *info2; | |
2086 | char *subarray = strchr(info->text_version+1, '/')+1; | |
582496b2 | 2087 | |
6b2d630c N |
2088 | info2 = st->ss->container_content(st, subarray); |
2089 | if (info2) { | |
2090 | unsigned long long current_size = 0; | |
2091 | unsigned long long new_size = | |
2092 | info2->custom_array_size/2; | |
2093 | ||
2094 | if (sysfs_get_ll(sra, | |
2095 | NULL, | |
2096 | "array_size", | |
2097 | ¤t_size) == 0 && | |
2098 | new_size > current_size) { | |
2099 | if (sysfs_set_num(sra, NULL, | |
2100 | "array_size", new_size) | |
2101 | < 0) | |
2102 | dprintf("Error: Cannot" | |
2103 | " set array size"); | |
2104 | else | |
2105 | dprintf("Array size " | |
2106 | "changed"); | |
2107 | dprintf(" from %llu to %llu.\n", | |
2108 | current_size, new_size); | |
582496b2 | 2109 | } |
6b2d630c | 2110 | sysfs_free(info2); |
582496b2 | 2111 | } |
6b2d630c | 2112 | } |
582496b2 | 2113 | |
6b2d630c | 2114 | if (info->new_level != reshape.level) { |
582496b2 | 2115 | |
6b2d630c | 2116 | c = map_num(pers, info->new_level); |
1cbc5680 N |
2117 | if (c) { |
2118 | err = sysfs_set_str(sra, NULL, "level", c); | |
2119 | if (err) | |
2120 | fprintf(stderr, Name\ | |
2121 | ": %s: could not set level " | |
2122 | "to %s\n", devname, c); | |
2123 | } | |
7236ee7a | 2124 | } |
6b2d630c N |
2125 | out: |
2126 | if (forked) | |
2127 | return 0; | |
e84f2c00 | 2128 | unfreeze(st); |
6b2d630c | 2129 | exit(0); |
e86c9dd6 | 2130 | |
6b2d630c | 2131 | release: |
1cbc5680 | 2132 | if (orig_level != UnSet && sra) { |
7236ee7a N |
2133 | c = map_num(pers, orig_level); |
2134 | if (c && sysfs_set_str(sra, NULL, "level", c) == 0) | |
2135 | fprintf(stderr, Name ": aborting level change\n"); | |
2136 | } | |
9202b8eb N |
2137 | if (!forked) |
2138 | unfreeze(st); | |
631d7405 | 2139 | return 1; |
7236ee7a | 2140 | } |
e86c9dd6 | 2141 | |
493f5dd6 | 2142 | int reshape_container(char *container, char *devname, |
5da9ab98 N |
2143 | struct supertype *st, |
2144 | struct mdinfo *info, | |
2145 | int force, | |
2146 | char *backup_file, | |
493f5dd6 | 2147 | int quiet, int restart) |
5da9ab98 | 2148 | { |
1bb174ba AK |
2149 | struct mdinfo *cc = NULL; |
2150 | ||
d8eb27f7 KA |
2151 | /* component_size is not meaningful for a container, |
2152 | * so pass '-1' meaning 'no change' | |
2153 | */ | |
493f5dd6 N |
2154 | if (!restart && |
2155 | reshape_super(st, -1, info->new_level, | |
5da9ab98 | 2156 | info->new_layout, info->new_chunk, |
41784c88 | 2157 | info->array.raid_disks, info->delta_disks, |
9e325442 AK |
2158 | backup_file, devname, quiet)) { |
2159 | unfreeze(st); | |
5da9ab98 | 2160 | return 1; |
9e325442 | 2161 | } |
5da9ab98 N |
2162 | |
2163 | sync_metadata(st); | |
2164 | ||
1bb174ba AK |
2165 | /* ping monitor to be sure that update is on disk |
2166 | */ | |
2167 | ping_monitor(container); | |
5da9ab98 N |
2168 | |
2169 | switch (fork()) { | |
2170 | case -1: /* error */ | |
2171 | perror("Cannot fork to complete reshape\n"); | |
9e325442 | 2172 | unfreeze(st); |
5da9ab98 N |
2173 | return 1; |
2174 | default: /* parent */ | |
2175 | printf(Name ": multi-array reshape continues in background\n"); | |
2176 | return 0; | |
2177 | case 0: /* child */ | |
2178 | break; | |
2179 | } | |
2180 | ||
1bb174ba AK |
2181 | while(1) { |
2182 | /* For each member array with reshape_active, | |
2183 | * we need to perform the reshape. | |
2184 | * We pick the first array that needs reshaping and | |
2185 | * reshape it. reshape_array() will re-read the metadata | |
2186 | * so the next time through a different array should be | |
2187 | * ready for reshape. | |
493f5dd6 N |
2188 | * It is possible that the 'different' array will not |
2189 | * be assembled yet. In that case we simple exit. | |
2190 | * When it is assembled, the mdadm which assembles it | |
2191 | * will take over the reshape. | |
1bb174ba AK |
2192 | */ |
2193 | struct mdinfo *content; | |
5da9ab98 N |
2194 | int rv; |
2195 | int fd; | |
2196 | struct mdstat_ent *mdstat; | |
5da9ab98 N |
2197 | char *adev; |
2198 | ||
1bb174ba | 2199 | sysfs_free(cc); |
5da9ab98 | 2200 | |
1bb174ba AK |
2201 | cc = st->ss->container_content(st, NULL); |
2202 | ||
2203 | for (content = cc; content ; content = content->next) { | |
2204 | char *subarray; | |
2205 | if (!content->reshape_active) | |
2206 | continue; | |
2207 | ||
2208 | subarray = strchr(content->text_version+1, '/')+1; | |
2209 | mdstat = mdstat_by_subdev(subarray, | |
2210 | devname2devnum(container)); | |
2211 | if (!mdstat) | |
2212 | continue; | |
2213 | break; | |
2214 | } | |
2215 | if (!content) | |
2216 | break; | |
5da9ab98 | 2217 | |
55f14721 | 2218 | fd = open_dev(mdstat->devnum); |
5da9ab98 N |
2219 | if (fd < 0) |
2220 | break; | |
2221 | adev = map_dev(dev2major(mdstat->devnum), | |
2222 | dev2minor(mdstat->devnum), | |
2223 | 0); | |
2224 | if (!adev) | |
1bb174ba AK |
2225 | adev = content->text_version; |
2226 | ||
2227 | sysfs_init(content, fd, mdstat->devnum); | |
5da9ab98 | 2228 | |
1bb174ba AK |
2229 | rv = reshape_array(container, fd, adev, st, |
2230 | content, force, | |
493f5dd6 | 2231 | backup_file, quiet, 1, restart); |
5da9ab98 | 2232 | close(fd); |
493f5dd6 | 2233 | restart = 0; |
5da9ab98 N |
2234 | if (rv) |
2235 | break; | |
2236 | } | |
9202b8eb | 2237 | unfreeze(st); |
1bb174ba | 2238 | sysfs_free(cc); |
5da9ab98 N |
2239 | exit(0); |
2240 | } | |
2241 | ||
7236ee7a N |
2242 | /* |
2243 | * We run a child process in the background which performs the following | |
2244 | * steps: | |
2245 | * - wait for resync to reach a certain point | |
2246 | * - suspend io to the following section | |
2247 | * - backup that section | |
2248 | * - allow resync to proceed further | |
2249 | * - resume io | |
2250 | * - discard the backup. | |
2251 | * | |
2252 | * When are combined in slightly different ways in the three cases. | |
2253 | * Grow: | |
2254 | * - suspend/backup/allow/wait/resume/discard | |
2255 | * Shrink: | |
2256 | * - allow/wait/suspend/backup/allow/wait/resume/discard | |
2257 | * same-size: | |
2258 | * - wait/resume/discard/suspend/backup/allow | |
2259 | * | |
2260 | * suspend/backup/allow always come together | |
2261 | * wait/resume/discard do too. | |
2262 | * For the same-size case we have two backups to improve flow. | |
2263 | * | |
2264 | */ | |
e86c9dd6 | 2265 | |
7443ee81 N |
2266 | int progress_reshape(struct mdinfo *info, struct reshape *reshape, |
2267 | unsigned long long backup_point, | |
2268 | unsigned long long wait_point, | |
2269 | unsigned long long *suspend_point, | |
2270 | unsigned long long *reshape_completed) | |
2271 | { | |
2272 | /* This function is called repeatedly by the reshape manager. | |
2273 | * It determines how much progress can safely be made and allows | |
2274 | * that progress. | |
2275 | * - 'info' identifies the array and particularly records in | |
2276 | * ->reshape_progress the metadata's knowledge of progress | |
2277 | * This is a sector offset from the start of the array | |
2278 | * of the next array block to be relocated. This number | |
2279 | * may increase from 0 or decrease from array_size, depending | |
2280 | * on the type of reshape that is happening. | |
2281 | * Note that in contrast, 'sync_completed' is a block count of the | |
2282 | * reshape so far. It gives the distance between the start point | |
2283 | * (head or tail of device) and the next place that data will be | |
2284 | * written. It always increases. | |
2285 | * - 'reshape' is the structure created by analyse_change | |
2286 | * - 'backup_point' shows how much the metadata manager has backed-up | |
2287 | * data. For reshapes with increasing progress, it is the next address | |
2288 | * to be backed up, previous addresses have been backed-up. For | |
2289 | * decreasing progress, it is the earliest address that has been | |
2290 | * backed up - later address are also backed up. | |
2291 | * So addresses between reshape_progress and backup_point are | |
2292 | * backed up providing those are in the 'correct' order. | |
2293 | * - 'wait_point' is an array address. When reshape_completed | |
2294 | * passes this point, progress_reshape should return. It might | |
2295 | * return earlier if it determines that ->reshape_progress needs | |
2296 | * to be updated or further backup is needed. | |
2297 | * - suspend_point is maintained by progress_reshape and the caller | |
2298 | * should not touch it except to initialise to zero. | |
2299 | * It is an array address and it only increases in 2.6.37 and earlier. | |
b6b95155 | 2300 | * This makes it difficult to handle reducing reshapes with |
7443ee81 N |
2301 | * external metadata. |
2302 | * However: it is similar to backup_point in that it records the | |
2303 | * other end of a suspended region from reshape_progress. | |
2304 | * it is moved to extend the region that is safe to backup and/or | |
2305 | * reshape | |
2306 | * - reshape_completed is read from sysfs and returned. The caller | |
2307 | * should copy this into ->reshape_progress when it has reason to | |
2308 | * believe that the metadata knows this, and any backup outside this | |
2309 | * has been erased. | |
2310 | * | |
2311 | * Return value is: | |
2312 | * 1 if more data from backup_point - but only as far as suspend_point, | |
2313 | * should be backed up | |
2314 | * 0 if things are progressing smoothly | |
9e034f70 N |
2315 | * -1 if the reshape is finished because it is all done, |
2316 | * -2 if the reshape is finished due to an error. | |
7443ee81 N |
2317 | */ |
2318 | ||
2319 | int advancing = (reshape->after.data_disks | |
2320 | >= reshape->before.data_disks); | |
38dad34a N |
2321 | unsigned long long need_backup; /* All data between start of array and |
2322 | * here will at some point need to | |
2323 | * be backed up. | |
d0ab945e | 2324 | */ |
7443ee81 | 2325 | unsigned long long read_offset, write_offset; |
b4f8e38b | 2326 | unsigned long long write_range; |
7443ee81 | 2327 | unsigned long long max_progress, target, completed; |
d0ab945e N |
2328 | unsigned long long array_size = (info->component_size |
2329 | * reshape->before.data_disks); | |
7443ee81 | 2330 | int fd; |
77a73a17 | 2331 | char buf[20]; |
7443ee81 N |
2332 | |
2333 | /* First, we unsuspend any region that is now known to be safe. | |
2334 | * If suspend_point is on the 'wrong' side of reshape_progress, then | |
2335 | * we don't have or need suspension at the moment. This is true for | |
2336 | * native metadata when we don't need to back-up. | |
2337 | */ | |
2338 | if (advancing) { | |
49cd738b | 2339 | if (info->reshape_progress <= *suspend_point) |
7443ee81 N |
2340 | sysfs_set_num(info, NULL, "suspend_lo", |
2341 | info->reshape_progress); | |
2342 | } else { | |
2343 | /* Note: this won't work in 2.6.37 and before. | |
2344 | * Something somewhere should make sure we don't need it! | |
2345 | */ | |
49cd738b | 2346 | if (info->reshape_progress >= *suspend_point) |
7443ee81 N |
2347 | sysfs_set_num(info, NULL, "suspend_hi", |
2348 | info->reshape_progress); | |
2349 | } | |
2350 | ||
2351 | /* Now work out how far it is safe to progress. | |
2352 | * If the read_offset for ->reshape_progress is less than | |
2353 | * 'blocks' beyond the write_offset, we can only progress as far | |
2354 | * as a backup. | |
2355 | * Otherwise we can progress until the write_offset for the new location | |
2356 | * reaches (within 'blocks' of) the read_offset at the current location. | |
2357 | * However that region must be suspended unless we are using native | |
2358 | * metadata. | |
2359 | * If we need to suspend more, we limit it to 128M per device, which is | |
2360 | * rather arbitrary and should be some time-based calculation. | |
2361 | */ | |
ec757320 N |
2362 | read_offset = info->reshape_progress / reshape->before.data_disks; |
2363 | write_offset = info->reshape_progress / reshape->after.data_disks; | |
b4f8e38b | 2364 | write_range = info->new_chunk/512; |
38dad34a N |
2365 | if (reshape->before.data_disks == reshape->after.data_disks) |
2366 | need_backup = array_size; | |
2367 | else | |
2368 | need_backup = reshape->backup_blocks; | |
7443ee81 | 2369 | if (advancing) { |
38dad34a | 2370 | if (read_offset < write_offset + write_range) |
7443ee81 | 2371 | max_progress = backup_point; |
ddee071d | 2372 | else |
7443ee81 | 2373 | max_progress = |
2f3dd5e4 N |
2374 | read_offset * |
2375 | reshape->after.data_disks; | |
7443ee81 | 2376 | } else { |
38dad34a N |
2377 | if (read_offset > write_offset - write_range) |
2378 | /* Can only progress as far as has been backed up, | |
2379 | * which must be suspended */ | |
7443ee81 | 2380 | max_progress = backup_point; |
38dad34a N |
2381 | else if (info->reshape_progress <= need_backup) |
2382 | max_progress = backup_point; | |
2383 | else { | |
2384 | if (info->array.major_version >= 0) | |
2385 | /* Can progress until backup is needed */ | |
2386 | max_progress = need_backup; | |
2387 | else { | |
2388 | /* Can progress until metadata update is required */ | |
2389 | max_progress = | |
2390 | read_offset * | |
2391 | reshape->after.data_disks; | |
2392 | /* but data must be suspended */ | |
2393 | if (max_progress < *suspend_point) | |
2394 | max_progress = *suspend_point; | |
2395 | } | |
7443ee81 N |
2396 | } |
2397 | } | |
2398 | ||
2399 | /* We know it is safe to progress to 'max_progress' providing | |
2400 | * it is suspended or we are using native metadata. | |
2401 | * Consider extending suspend_point 128M per device if it | |
2402 | * is less than 64M per device beyond reshape_progress. | |
2403 | * But always do a multiple of 'blocks' | |
832b2b9a N |
2404 | * FIXME this is too big - it takes to long to complete |
2405 | * this much. | |
7443ee81 N |
2406 | */ |
2407 | target = 64*1024*2 * min(reshape->before.data_disks, | |
2408 | reshape->after.data_disks); | |
b4f8e38b | 2409 | target /= reshape->backup_blocks; |
7443ee81 N |
2410 | if (target < 2) |
2411 | target = 2; | |
b4f8e38b | 2412 | target *= reshape->backup_blocks; |
7443ee81 N |
2413 | |
2414 | /* For externally managed metadata we always need to suspend IO to | |
2415 | * the area being reshaped so we regularly push suspend_point forward. | |
2416 | * For native metadata we only need the suspend if we are going to do | |
2417 | * a backup. | |
2418 | */ | |
2419 | if (advancing) { | |
d0ab945e N |
2420 | if ((need_backup > info->reshape_progress |
2421 | || info->array.major_version < 0) && | |
7443ee81 | 2422 | *suspend_point < info->reshape_progress + target) { |
d0ab945e N |
2423 | if (need_backup < *suspend_point + 2 * target) |
2424 | *suspend_point = need_backup; | |
2425 | else if (*suspend_point + 2 * target < array_size) | |
7443ee81 | 2426 | *suspend_point += 2 * target; |
d0ab945e N |
2427 | else |
2428 | *suspend_point = array_size; | |
7443ee81 | 2429 | sysfs_set_num(info, NULL, "suspend_hi", *suspend_point); |
d0ab945e N |
2430 | if (max_progress > *suspend_point) |
2431 | max_progress = *suspend_point; | |
7443ee81 N |
2432 | } |
2433 | } else { | |
38dad34a N |
2434 | if (info->array.major_version >= 0) { |
2435 | /* Only need to suspend when about to backup */ | |
2436 | if (info->reshape_progress < need_backup * 2 && | |
2437 | *suspend_point > 0) { | |
d0ab945e | 2438 | *suspend_point = 0; |
38dad34a N |
2439 | sysfs_set_num(info, NULL, "suspend_lo", 0); |
2440 | sysfs_set_num(info, NULL, "suspend_hi", need_backup); | |
2441 | } | |
2442 | } else { | |
2443 | /* Need to suspend continually */ | |
2444 | if (info->reshape_progress < *suspend_point) | |
2445 | *suspend_point = info->reshape_progress; | |
2446 | if (*suspend_point + target < info->reshape_progress) | |
2447 | /* No need to move suspend region yet */; | |
2448 | else { | |
2449 | if (*suspend_point >= 2 * target) | |
2450 | *suspend_point -= 2 * target; | |
2451 | else | |
2452 | *suspend_point = 0; | |
2453 | sysfs_set_num(info, NULL, "suspend_lo", | |
2454 | *suspend_point); | |
2455 | } | |
d0ab945e N |
2456 | if (max_progress < *suspend_point) |
2457 | max_progress = *suspend_point; | |
7443ee81 N |
2458 | } |
2459 | } | |
2460 | ||
2461 | /* now set sync_max to allow that progress. sync_max, like | |
2462 | * sync_completed is a count of sectors written per device, so | |
2463 | * we find the difference between max_progress and the start point, | |
2464 | * and divide that by after.data_disks to get a sync_max | |
2465 | * number. | |
2466 | * At the same time we convert wait_point to a similar number | |
2467 | * for comparing against sync_completed. | |
2468 | */ | |
92d1991f | 2469 | /* scale down max_progress to per_disk */ |
7443ee81 | 2470 | max_progress /= reshape->after.data_disks; |
92d1991f N |
2471 | /* Round to chunk size as some kernels give an erroneously high number */ |
2472 | max_progress /= info->new_chunk/512; | |
2473 | max_progress *= info->new_chunk/512; | |
7f913e9b N |
2474 | /* And round to old chunk size as the kernel wants that */ |
2475 | max_progress /= info->array.chunk_size/512; | |
2476 | max_progress *= info->array.chunk_size/512; | |
92d1991f N |
2477 | /* Limit progress to the whole device */ |
2478 | if (max_progress > info->component_size) | |
2479 | max_progress = info->component_size; | |
7443ee81 | 2480 | wait_point /= reshape->after.data_disks; |
92d1991f N |
2481 | if (!advancing) { |
2482 | /* switch from 'device offset' to 'processed block count' */ | |
2483 | max_progress = info->component_size - max_progress; | |
2484 | wait_point = info->component_size - wait_point; | |
2485 | } | |
7443ee81 N |
2486 | |
2487 | sysfs_set_num(info, NULL, "sync_max", max_progress); | |
2488 | ||
2489 | /* Now wait. If we have already reached the point that we were | |
2490 | * asked to wait to, don't wait at all, else wait for any change. | |
2491 | * We need to select on 'sync_completed' as that is the place that | |
2492 | * notifications happen, but we are really interested in | |
2493 | * 'reshape_position' | |
2494 | */ | |
2495 | fd = sysfs_get_fd(info, NULL, "sync_completed"); | |
2496 | if (fd < 0) | |
77a73a17 | 2497 | goto check_progress; |
7443ee81 N |
2498 | |
2499 | if (sysfs_fd_get_ll(fd, &completed) < 0) { | |
2500 | close(fd); | |
77a73a17 | 2501 | goto check_progress; |
7443ee81 N |
2502 | } |
2503 | while (completed < max_progress && completed < wait_point) { | |
2504 | /* Check that sync_action is still 'reshape' to avoid | |
2505 | * waiting forever on a dead array | |
2506 | */ | |
2507 | char action[20]; | |
2508 | fd_set rfds; | |
2509 | if (sysfs_get_str(info, NULL, "sync_action", | |
2510 | action, 20) <= 0 || | |
2511 | strncmp(action, "reshape", 7) != 0) | |
2512 | break; | |
26d6e157 AK |
2513 | /* Some kernels reset 'sync_completed' to zero |
2514 | * before setting 'sync_action' to 'idle'. | |
2515 | * So we need these extra tests. | |
2516 | */ | |
2517 | if (completed == 0 && advancing | |
2518 | && info->reshape_progress > 0) | |
2519 | break; | |
2520 | if (completed == 0 && !advancing | |
2521 | && info->reshape_progress < (info->component_size | |
2522 | * reshape->after.data_disks)) | |
2523 | break; | |
7443ee81 N |
2524 | FD_ZERO(&rfds); |
2525 | FD_SET(fd, &rfds); | |
2526 | select(fd+1, NULL, NULL, &rfds, NULL); | |
2527 | if (sysfs_fd_get_ll(fd, &completed) < 0) { | |
2528 | close(fd); | |
77a73a17 | 2529 | goto check_progress; |
7443ee81 N |
2530 | } |
2531 | } | |
10d0d365 AK |
2532 | /* Some kernels reset 'sync_completed' to zero, |
2533 | * we need to have real point we are in md | |
2534 | */ | |
2535 | if (completed == 0) | |
2536 | completed = max_progress; | |
2537 | ||
92d1991f N |
2538 | /* some kernels can give an incorrectly high 'completed' number */ |
2539 | completed /= (info->new_chunk/512); | |
2540 | completed *= (info->new_chunk/512); | |
7443ee81 N |
2541 | /* Convert 'completed' back in to a 'progress' number */ |
2542 | completed *= reshape->after.data_disks; | |
2543 | if (!advancing) { | |
2544 | completed = info->component_size * reshape->after.data_disks | |
2545 | - completed; | |
2546 | } | |
2547 | *reshape_completed = completed; | |
2548 | ||
2549 | close(fd); | |
2550 | ||
2551 | /* We return the need_backup flag. Caller will decide | |
d0ab945e | 2552 | * how much - a multiple of ->backup_blocks up to *suspend_point |
7443ee81 | 2553 | */ |
38dad34a N |
2554 | if (advancing) |
2555 | return need_backup > info->reshape_progress; | |
2556 | else | |
2557 | return need_backup >= info->reshape_progress; | |
77a73a17 N |
2558 | |
2559 | check_progress: | |
2560 | /* if we couldn't read a number from sync_completed, then | |
2561 | * either the reshape did complete, or it aborted. | |
2562 | * We can tell which by checking for 'none' in reshape_position. | |
2563 | */ | |
2564 | strcpy(buf, "hi"); | |
2565 | if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0 | |
2566 | || strncmp(buf, "none", 4) != 0) | |
2567 | return -2; /* abort */ | |
6d5316f6 N |
2568 | else { |
2569 | /* Maybe racing with array shutdown - check state */ | |
2570 | if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0 | |
2571 | || strncmp(buf, "inactive", 8) == 0 | |
2572 | || strncmp(buf, "clear",5) == 0) | |
2573 | return -2; /* abort */ | |
77a73a17 | 2574 | return -1; /* complete */ |
6d5316f6 | 2575 | } |
7443ee81 N |
2576 | } |
2577 | ||
2578 | ||
fcf57625 | 2579 | /* FIXME return status is never checked */ |
4411fb17 | 2580 | static int grow_backup(struct mdinfo *sra, |
7236ee7a | 2581 | unsigned long long offset, /* per device */ |
7443ee81 | 2582 | unsigned long stripes, /* per device, in old chunks */ |
7236ee7a N |
2583 | int *sources, unsigned long long *offsets, |
2584 | int disks, int chunk, int level, int layout, | |
2585 | int dests, int *destfd, unsigned long long *destoffsets, | |
d4445387 | 2586 | int part, int *degraded, |
7236ee7a N |
2587 | char *buf) |
2588 | { | |
2589 | /* Backup 'blocks' sectors at 'offset' on each device of the array, | |
2590 | * to storage 'destfd' (offset 'destoffsets'), after first | |
2591 | * suspending IO. Then allow resync to continue | |
2592 | * over the suspended section. | |
2593 | * Use part 'part' of the backup-super-block. | |
2594 | */ | |
2595 | int odata = disks; | |
2596 | int rv = 0; | |
2597 | int i; | |
f21e18ca N |
2598 | unsigned long long ll; |
2599 | int new_degraded; | |
7236ee7a N |
2600 | //printf("offset %llu\n", offset); |
2601 | if (level >= 4) | |
2602 | odata--; | |
2603 | if (level == 6) | |
2604 | odata--; | |
7443ee81 | 2605 | |
d4445387 | 2606 | /* Check that array hasn't become degraded, else we might backup the wrong data */ |
2c6ac128 N |
2607 | if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0) |
2608 | return -1; /* FIXME this error is ignored */ | |
f21e18ca | 2609 | new_degraded = (int)ll; |
d4445387 N |
2610 | if (new_degraded != *degraded) { |
2611 | /* check each device to ensure it is still working */ | |
2612 | struct mdinfo *sd; | |
2613 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
2614 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2615 | continue; | |
2616 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
2617 | char sbuf[20]; | |
2618 | if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 || | |
2619 | strstr(sbuf, "faulty") || | |
2620 | strstr(sbuf, "in_sync") == NULL) { | |
2621 | /* this device is dead */ | |
2622 | sd->disk.state = (1<<MD_DISK_FAULTY); | |
2623 | if (sd->disk.raid_disk >= 0 && | |
2624 | sources[sd->disk.raid_disk] >= 0) { | |
2625 | close(sources[sd->disk.raid_disk]); | |
2626 | sources[sd->disk.raid_disk] = -1; | |
2627 | } | |
2628 | } | |
2629 | } | |
2630 | } | |
2631 | *degraded = new_degraded; | |
2632 | } | |
7236ee7a N |
2633 | if (part) { |
2634 | bsb.arraystart2 = __cpu_to_le64(offset * odata); | |
200871ad | 2635 | bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
2636 | } else { |
2637 | bsb.arraystart = __cpu_to_le64(offset * odata); | |
200871ad | 2638 | bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata); |
7236ee7a N |
2639 | } |
2640 | if (part) | |
2641 | bsb.magic[15] = '2'; | |
2642 | for (i = 0; i < dests; i++) | |
2643 | if (part) | |
2644 | lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0); | |
2645 | else | |
2646 | lseek64(destfd[i], destoffsets[i], 0); | |
2647 | ||
2648 | rv = save_stripes(sources, offsets, | |
2649 | disks, chunk, level, layout, | |
2650 | dests, destfd, | |
2651 | offset*512*odata, stripes * chunk * odata, | |
2652 | buf); | |
2653 | ||
2654 | if (rv) | |
2655 | return rv; | |
97396422 | 2656 | bsb.mtime = __cpu_to_le64(time(0)); |
7236ee7a N |
2657 | for (i = 0; i < dests; i++) { |
2658 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
2659 | ||
2660 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
2661 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
2662 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
2663 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
2664 | ||
72044953 | 2665 | rv = -1; |
f21e18ca N |
2666 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0) |
2667 | != destoffsets[i] - 4096) | |
72044953 N |
2668 | break; |
2669 | if (write(destfd[i], &bsb, 512) != 512) | |
2670 | break; | |
ff94fb86 | 2671 | if (destoffsets[i] > 4096) { |
f21e18ca | 2672 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) != |
a847575a | 2673 | destoffsets[i]+stripes*chunk*odata) |
72044953 N |
2674 | break; |
2675 | if (write(destfd[i], &bsb, 512) != 512) | |
2676 | break; | |
ff94fb86 | 2677 | } |
7236ee7a | 2678 | fsync(destfd[i]); |
72044953 | 2679 | rv = 0; |
7236ee7a | 2680 | } |
2efedc7b | 2681 | |
fcf57625 | 2682 | return rv; |
7236ee7a N |
2683 | } |
2684 | ||
2685 | /* in 2.6.30, the value reported by sync_completed can be | |
2686 | * less that it should be by one stripe. | |
2687 | * This only happens when reshape hits sync_max and pauses. | |
2688 | * So allow wait_backup to either extent sync_max further | |
2689 | * than strictly necessary, or return before the | |
2690 | * sync has got quite as far as we would really like. | |
2691 | * This is what 'blocks2' is for. | |
2692 | * The various caller give appropriate values so that | |
2693 | * every works. | |
2694 | */ | |
fcf57625 | 2695 | /* FIXME return value is often ignored */ |
7443ee81 | 2696 | static int forget_backup( |
7236ee7a N |
2697 | int dests, int *destfd, unsigned long long *destoffsets, |
2698 | int part) | |
2699 | { | |
7443ee81 N |
2700 | /* |
2701 | * Erase backup 'part' (which is 0 or 1) | |
7236ee7a | 2702 | */ |
7236ee7a | 2703 | int i; |
fcf57625 | 2704 | int rv; |
7236ee7a | 2705 | |
7236ee7a N |
2706 | if (part) { |
2707 | bsb.arraystart2 = __cpu_to_le64(0); | |
2708 | bsb.length2 = __cpu_to_le64(0); | |
2709 | } else { | |
2710 | bsb.arraystart = __cpu_to_le64(0); | |
2711 | bsb.length = __cpu_to_le64(0); | |
2712 | } | |
97396422 | 2713 | bsb.mtime = __cpu_to_le64(time(0)); |
fcf57625 | 2714 | rv = 0; |
7236ee7a N |
2715 | for (i = 0; i < dests; i++) { |
2716 | bsb.devstart = __cpu_to_le64(destoffsets[i]/512); | |
2717 | bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb)); | |
2718 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0) | |
2719 | bsb.sb_csum2 = bsb_csum((char*)&bsb, | |
2720 | ((char*)&bsb.sb_csum2)-((char*)&bsb)); | |
f21e18ca | 2721 | if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) != |
a847575a | 2722 | destoffsets[i]-4096) |
72044953 N |
2723 | rv = -1; |
2724 | if (rv == 0 && | |
2725 | write(destfd[i], &bsb, 512) != 512) | |
2726 | rv = -1; | |
7236ee7a N |
2727 | fsync(destfd[i]); |
2728 | } | |
fcf57625 | 2729 | return rv; |
7236ee7a | 2730 | } |
e86c9dd6 | 2731 | |
7236ee7a N |
2732 | static void fail(char *msg) |
2733 | { | |
fcf57625 | 2734 | int rv; |
72044953 N |
2735 | rv = (write(2, msg, strlen(msg)) != (int)strlen(msg)); |
2736 | rv |= (write(2, "\n", 1) != 1); | |
fcf57625 | 2737 | exit(rv ? 1 : 2); |
7236ee7a N |
2738 | } |
2739 | ||
2740 | static char *abuf, *bbuf; | |
f21e18ca | 2741 | static unsigned long long abuflen; |
7236ee7a N |
2742 | static void validate(int afd, int bfd, unsigned long long offset) |
2743 | { | |
2744 | /* check that the data in the backup against the array. | |
2745 | * This is only used for regression testing and should not | |
2746 | * be used while the array is active | |
2747 | */ | |
7236ee7a N |
2748 | if (afd < 0) |
2749 | return; | |
2750 | lseek64(bfd, offset - 4096, 0); | |
2751 | if (read(bfd, &bsb2, 512) != 512) | |
2752 | fail("cannot read bsb"); | |
2753 | if (bsb2.sb_csum != bsb_csum((char*)&bsb2, | |
2754 | ((char*)&bsb2.sb_csum)-((char*)&bsb2))) | |
2755 | fail("first csum bad"); | |
2756 | if (memcmp(bsb2.magic, "md_backup_data", 14) != 0) | |
2757 | fail("magic is bad"); | |
2758 | if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 && | |
2759 | bsb2.sb_csum2 != bsb_csum((char*)&bsb2, | |
2760 | ((char*)&bsb2.sb_csum2)-((char*)&bsb2))) | |
2761 | fail("second csum bad"); | |
2762 | ||
2763 | if (__le64_to_cpu(bsb2.devstart)*512 != offset) | |
2764 | fail("devstart is wrong"); | |
2765 | ||
2766 | if (bsb2.length) { | |
2767 | unsigned long long len = __le64_to_cpu(bsb2.length)*512; | |
2768 | ||
2769 | if (abuflen < len) { | |
2770 | free(abuf); | |
2771 | free(bbuf); | |
2772 | abuflen = len; | |
fcf57625 N |
2773 | if (posix_memalign((void**)&abuf, 4096, abuflen) || |
2774 | posix_memalign((void**)&bbuf, 4096, abuflen)) { | |
2775 | abuflen = 0; | |
2776 | /* just stop validating on mem-alloc failure */ | |
2777 | return; | |
2778 | } | |
e86c9dd6 | 2779 | } |
48924014 | 2780 | |
7236ee7a | 2781 | lseek64(bfd, offset, 0); |
f21e18ca | 2782 | if ((unsigned long long)read(bfd, bbuf, len) != len) { |
080fd005 | 2783 | //printf("len %llu\n", len); |
7236ee7a N |
2784 | fail("read first backup failed"); |
2785 | } | |
2786 | lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0); | |
f21e18ca | 2787 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
2788 | fail("read first from array failed"); |
2789 | if (memcmp(bbuf, abuf, len) != 0) { | |
080fd005 | 2790 | #if 0 |
7236ee7a N |
2791 | int i; |
2792 | printf("offset=%llu len=%llu\n", | |
080fd005 | 2793 | (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len); |
7236ee7a N |
2794 | for (i=0; i<len; i++) |
2795 | if (bbuf[i] != abuf[i]) { | |
2796 | printf("first diff byte %d\n", i); | |
93ecfa01 | 2797 | break; |
7236ee7a | 2798 | } |
080fd005 | 2799 | #endif |
7236ee7a | 2800 | fail("data1 compare failed"); |
e86c9dd6 | 2801 | } |
7236ee7a N |
2802 | } |
2803 | if (bsb2.length2) { | |
2804 | unsigned long long len = __le64_to_cpu(bsb2.length2)*512; | |
2805 | ||
2806 | if (abuflen < len) { | |
2807 | free(abuf); | |
2808 | free(bbuf); | |
2809 | abuflen = len; | |
2810 | abuf = malloc(abuflen); | |
2811 | bbuf = malloc(abuflen); | |
e86c9dd6 NB |
2812 | } |
2813 | ||
7236ee7a | 2814 | lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0); |
f21e18ca | 2815 | if ((unsigned long long)read(bfd, bbuf, len) != len) |
7236ee7a N |
2816 | fail("read second backup failed"); |
2817 | lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0); | |
f21e18ca | 2818 | if ((unsigned long long)read(afd, abuf, len) != len) |
7236ee7a N |
2819 | fail("read second from array failed"); |
2820 | if (memcmp(bbuf, abuf, len) != 0) | |
2821 | fail("data2 compare failed"); | |
e86c9dd6 | 2822 | } |
7236ee7a | 2823 | } |
e86c9dd6 | 2824 | |
999b4972 N |
2825 | int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape, |
2826 | struct supertype *st, unsigned long blocks, | |
2827 | int *fds, unsigned long long *offsets, | |
2828 | int dests, int *destfd, unsigned long long *destoffsets) | |
7236ee7a | 2829 | { |
7443ee81 N |
2830 | /* Monitor a reshape where backup is being performed using |
2831 | * 'native' mechanism - either to a backup file, or | |
2832 | * to some space in a spare. | |
2833 | */ | |
7236ee7a | 2834 | char *buf; |
7443ee81 N |
2835 | int degraded = -1; |
2836 | unsigned long long speed; | |
2837 | unsigned long long suspend_point, array_size; | |
2838 | unsigned long long backup_point, wait_point; | |
2839 | unsigned long long reshape_completed; | |
2840 | int done = 0; | |
2841 | int increasing = reshape->after.data_disks >= reshape->before.data_disks; | |
2842 | int part = 0; /* The next part of the backup area to fill. It may already | |
2843 | * be full, so we need to check */ | |
2844 | int level = reshape->level; | |
2845 | int layout = reshape->before.layout; | |
2846 | int data = reshape->before.data_disks; | |
2847 | int disks = reshape->before.data_disks + reshape->parity; | |
2848 | int chunk = sra->array.chunk_size; | |
da8100ca N |
2849 | struct mdinfo *sd; |
2850 | unsigned long stripes; | |
2851 | ||
2852 | /* set up the backup-super-block. This requires the | |
2853 | * uuid from the array. | |
2854 | */ | |
2855 | /* Find a superblock */ | |
2856 | for (sd = sra->devs; sd; sd = sd->next) { | |
2857 | char *dn; | |
2858 | int devfd; | |
2859 | int ok; | |
2860 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
2861 | continue; | |
2862 | dn = map_dev(sd->disk.major, sd->disk.minor, 1); | |
2863 | devfd = dev_open(dn, O_RDONLY); | |
2864 | if (devfd < 0) | |
2865 | continue; | |
2866 | ok = st->ss->load_super(st, devfd, NULL); | |
2867 | close(devfd); | |
2868 | if (ok >= 0) | |
2869 | break; | |
2870 | } | |
2871 | if (!sd) { | |
2872 | fprintf(stderr, Name ": Cannot find a superblock\n"); | |
2873 | return 0; | |
2874 | } | |
2875 | ||
2876 | memset(&bsb, 0, 512); | |
2877 | memcpy(bsb.magic, "md_backup_data-1", 16); | |
2878 | st->ss->uuid_from_super(st, (int*)&bsb.set_uuid); | |
2879 | bsb.mtime = __cpu_to_le64(time(0)); | |
2880 | bsb.devstart2 = blocks; | |
2881 | ||
2882 | stripes = blocks / (sra->array.chunk_size/512) / | |
2883 | reshape->before.data_disks; | |
e86c9dd6 | 2884 | |
fcf57625 N |
2885 | if (posix_memalign((void**)&buf, 4096, disks * chunk)) |
2886 | /* Don't start the 'reshape' */ | |
2887 | return 0; | |
7443ee81 N |
2888 | if (reshape->before.data_disks == reshape->after.data_disks) { |
2889 | sysfs_get_ll(sra, NULL, "sync_speed_min", &speed); | |
2890 | sysfs_set_num(sra, NULL, "sync_speed_min", 200000); | |
2891 | } | |
e86c9dd6 | 2892 | |
7443ee81 | 2893 | if (increasing) { |
96533072 | 2894 | array_size = sra->component_size * reshape->after.data_disks; |
7443ee81 N |
2895 | backup_point = sra->reshape_progress; |
2896 | suspend_point = 0; | |
2897 | } else { | |
96533072 | 2898 | array_size = sra->component_size * reshape->before.data_disks; |
25da62d9 | 2899 | backup_point = reshape->backup_blocks; |
7443ee81 N |
2900 | suspend_point = array_size; |
2901 | } | |
7236ee7a | 2902 | |
7443ee81 N |
2903 | while (!done) { |
2904 | int rv; | |
7236ee7a | 2905 | |
7443ee81 N |
2906 | /* Want to return as soon the oldest backup slot can |
2907 | * be released as that allows us to start backing up | |
2908 | * some more, providing suspend_point has been | |
b6b95155 | 2909 | * advanced, which it should have. |
7443ee81 N |
2910 | */ |
2911 | if (increasing) { | |
2912 | wait_point = array_size; | |
2913 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
2914 | wait_point = (__le64_to_cpu(bsb.arraystart) + | |
2915 | __le64_to_cpu(bsb.length)); | |
2916 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
2917 | wait_point = (__le64_to_cpu(bsb.arraystart2) + | |
2918 | __le64_to_cpu(bsb.length2)); | |
2919 | } else { | |
2920 | wait_point = 0; | |
2921 | if (part == 0 && __le64_to_cpu(bsb.length) > 0) | |
2922 | wait_point = __le64_to_cpu(bsb.arraystart); | |
2923 | if (part == 1 && __le64_to_cpu(bsb.length2) > 0) | |
2924 | wait_point = __le64_to_cpu(bsb.arraystart2); | |
2925 | } | |
2926 | ||
2927 | rv = progress_reshape(sra, reshape, | |
2928 | backup_point, wait_point, | |
2929 | &suspend_point, &reshape_completed); | |
7443ee81 N |
2930 | /* external metadata would need to ping_monitor here */ |
2931 | sra->reshape_progress = reshape_completed; | |
7236ee7a | 2932 | |
7443ee81 N |
2933 | /* Clear any backup region that is before 'here' */ |
2934 | if (increasing) { | |
2935 | if (reshape_completed >= (__le64_to_cpu(bsb.arraystart) + | |
2936 | __le64_to_cpu(bsb.length))) | |
2937 | forget_backup(dests, destfd, | |
2938 | destoffsets, 0); | |
2939 | if (reshape_completed >= (__le64_to_cpu(bsb.arraystart2) + | |
2940 | __le64_to_cpu(bsb.length2))) | |
2941 | forget_backup(dests, destfd, | |
2942 | destoffsets, 1); | |
2943 | } else { | |
2944 | if (reshape_completed <= (__le64_to_cpu(bsb.arraystart))) | |
2945 | forget_backup(dests, destfd, | |
2946 | destoffsets, 0); | |
2947 | if (reshape_completed <= (__le64_to_cpu(bsb.arraystart2))) | |
2948 | forget_backup(dests, destfd, | |
2949 | destoffsets, 1); | |
2950 | } | |
7236ee7a | 2951 | |
223c5c99 | 2952 | if (rv < 0) { |
77a73a17 N |
2953 | if (rv == -1) |
2954 | done = 1; | |
223c5c99 N |
2955 | break; |
2956 | } | |
2957 | ||
60204e4e | 2958 | while (rv) { |
7443ee81 | 2959 | unsigned long long offset; |
e97ca002 | 2960 | unsigned long actual_stripes; |
60204e4e N |
2961 | /* Need to backup some data. |
2962 | * If 'part' is not used and the desired | |
2963 | * backup size is suspended, do a backup, | |
2964 | * then consider the next part. | |
2965 | */ | |
7443ee81 N |
2966 | /* Check that 'part' is unused */ |
2967 | if (part == 0 && __le64_to_cpu(bsb.length) != 0) | |
60204e4e | 2968 | break; |
7443ee81 | 2969 | if (part == 1 && __le64_to_cpu(bsb.length2) != 0) |
60204e4e | 2970 | break; |
7443ee81 N |
2971 | |
2972 | offset = backup_point / data; | |
e97ca002 | 2973 | actual_stripes = stripes; |
60204e4e | 2974 | if (increasing) { |
e97ca002 N |
2975 | if (offset + actual_stripes * (chunk/512) > |
2976 | sra->component_size) | |
2977 | actual_stripes = ((sra->component_size - offset) | |
2978 | / (chunk/512)); | |
2979 | if (offset + actual_stripes * (chunk/512) > | |
60204e4e N |
2980 | suspend_point/data) |
2981 | break; | |
2982 | } else { | |
e97ca002 N |
2983 | if (offset < actual_stripes * (chunk/512)) |
2984 | actual_stripes = offset / (chunk/512); | |
2985 | offset -= actual_stripes * (chunk/512); | |
2986 | if (offset < suspend_point/data) | |
60204e4e N |
2987 | break; |
2988 | } | |
e4b11073 N |
2989 | if (actual_stripes == 0) |
2990 | break; | |
e97ca002 | 2991 | grow_backup(sra, offset, actual_stripes, |
7443ee81 N |
2992 | fds, offsets, |
2993 | disks, chunk, level, layout, | |
2994 | dests, destfd, destoffsets, | |
2995 | part, °raded, buf); | |
2996 | validate(afd, destfd[0], destoffsets[0]); | |
2997 | /* record where 'part' is up to */ | |
2998 | part = !part; | |
2999 | if (increasing) | |
e97ca002 | 3000 | backup_point += actual_stripes * (chunk/512) * data; |
7443ee81 | 3001 | else |
e97ca002 | 3002 | backup_point -= actual_stripes * (chunk/512) * data; |
7443ee81 N |
3003 | } |
3004 | } | |
3005 | ||
223c5c99 N |
3006 | /* FIXME maybe call progress_reshape one more time instead */ |
3007 | abort_reshape(sra); /* remove any remaining suspension */ | |
7443ee81 N |
3008 | if (reshape->before.data_disks == reshape->after.data_disks) |
3009 | sysfs_set_num(sra, NULL, "sync_speed_min", speed); | |
7236ee7a | 3010 | free(buf); |
77a73a17 | 3011 | return done; |
e86c9dd6 | 3012 | } |
353632d9 NB |
3013 | |
3014 | /* | |
3015 | * If any spare contains md_back_data-1 which is recent wrt mtime, | |
3016 | * write that data into the array and update the super blocks with | |
3017 | * the new reshape_progress | |
3018 | */ | |
ea0ebe96 N |
3019 | int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt, |
3020 | char *backup_file, int verbose) | |
353632d9 NB |
3021 | { |
3022 | int i, j; | |
3023 | int old_disks; | |
353632d9 | 3024 | unsigned long long *offsets; |
82f2d6ab | 3025 | unsigned long long nstripe, ostripe; |
6e9eac4f | 3026 | int ndata, odata; |
353632d9 | 3027 | |
e9e43ec3 N |
3028 | if (info->new_level != info->array.level) |
3029 | return 1; /* Cannot handle level changes (they are instantaneous) */ | |
3030 | ||
3031 | odata = info->array.raid_disks - info->delta_disks - 1; | |
3032 | if (info->array.level == 6) odata--; /* number of data disks */ | |
3033 | ndata = info->array.raid_disks - 1; | |
3034 | if (info->new_level == 6) ndata--; | |
353632d9 NB |
3035 | |
3036 | old_disks = info->array.raid_disks - info->delta_disks; | |
3037 | ||
e9e43ec3 N |
3038 | if (info->delta_disks <= 0) |
3039 | /* Didn't grow, so the backup file must have | |
3040 | * been used | |
3041 | */ | |
3042 | old_disks = cnt; | |
06b0d786 | 3043 | for (i=old_disks-(backup_file?1:0); i<cnt; i++) { |
353632d9 | 3044 | struct mdinfo dinfo; |
06b0d786 | 3045 | int fd; |
e9e43ec3 | 3046 | int bsbsize; |
ea0ebe96 | 3047 | char *devname, namebuf[20]; |
10af14c4 | 3048 | unsigned long long lo, hi; |
353632d9 NB |
3049 | |
3050 | /* This was a spare and may have some saved data on it. | |
3051 | * Load the superblock, find and load the | |
3052 | * backup_super_block. | |
3053 | * If either fail, go on to next device. | |
3054 | * If the backup contains no new info, just return | |
206c5eae | 3055 | * else restore data and update all superblocks |
353632d9 | 3056 | */ |
06b0d786 NB |
3057 | if (i == old_disks-1) { |
3058 | fd = open(backup_file, O_RDONLY); | |
e9e43ec3 N |
3059 | if (fd<0) { |
3060 | fprintf(stderr, Name ": backup file %s inaccessible: %s\n", | |
3061 | backup_file, strerror(errno)); | |
06b0d786 | 3062 | continue; |
e9e43ec3 | 3063 | } |
ea0ebe96 | 3064 | devname = backup_file; |
06b0d786 NB |
3065 | } else { |
3066 | fd = fdlist[i]; | |
3067 | if (fd < 0) | |
3068 | continue; | |
3da92f27 | 3069 | if (st->ss->load_super(st, fd, NULL)) |
06b0d786 | 3070 | continue; |
353632d9 | 3071 | |
a5d85af7 | 3072 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 NB |
3073 | st->ss->free_super(st); |
3074 | ||
06b0d786 NB |
3075 | if (lseek64(fd, |
3076 | (dinfo.data_offset + dinfo.component_size - 8) <<9, | |
ea0ebe96 N |
3077 | 0) < 0) { |
3078 | fprintf(stderr, Name ": Cannot seek on device %d\n", i); | |
06b0d786 | 3079 | continue; /* Cannot seek */ |
ea0ebe96 N |
3080 | } |
3081 | sprintf(namebuf, "device-%d", i); | |
3082 | devname = namebuf; | |
06b0d786 | 3083 | } |
ea0ebe96 N |
3084 | if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) { |
3085 | if (verbose) | |
3086 | fprintf(stderr, Name ": Cannot read from %s\n", devname); | |
353632d9 | 3087 | continue; /* Cannot read */ |
ea0ebe96 | 3088 | } |
e9e43ec3 | 3089 | if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 && |
ea0ebe96 N |
3090 | memcmp(bsb.magic, "md_backup_data-2", 16) != 0) { |
3091 | if (verbose) | |
3092 | fprintf(stderr, Name ": No backup metadata on %s\n", devname); | |
353632d9 | 3093 | continue; |
ea0ebe96 N |
3094 | } |
3095 | if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) { | |
3096 | if (verbose) | |
3097 | fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname); | |
353632d9 | 3098 | continue; /* bad checksum */ |
ea0ebe96 | 3099 | } |
e9e43ec3 | 3100 | if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 && |
ea0ebe96 N |
3101 | bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) { |
3102 | if (verbose) | |
3103 | fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname); | |
22e30516 | 3104 | continue; /* Bad second checksum */ |
ea0ebe96 N |
3105 | } |
3106 | if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) { | |
3107 | if (verbose) | |
3108 | fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname); | |
353632d9 | 3109 | continue; /* Wrong uuid */ |
ea0ebe96 | 3110 | } |
353632d9 | 3111 | |
097075b6 N |
3112 | /* array utime and backup-mtime should be updated at much the same time, but it seems that |
3113 | * sometimes they aren't... So allow considerable flexability in matching, and allow | |
3114 | * this test to be overridden by an environment variable. | |
3115 | */ | |
f21e18ca N |
3116 | if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 || |
3117 | info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) { | |
097075b6 N |
3118 | if (check_env("MDADM_GROW_ALLOW_OLD")) { |
3119 | fprintf(stderr, Name ": accepting backup with timestamp %lu " | |
3120 | "for array with timestamp %lu\n", | |
3121 | (unsigned long)__le64_to_cpu(bsb.mtime), | |
3122 | (unsigned long)info->array.utime); | |
3123 | } else { | |
3124 | if (verbose) | |
3125 | fprintf(stderr, Name ": too-old timestamp on " | |
3126 | "backup-metadata on %s\n", devname); | |
3127 | continue; /* time stamp is too bad */ | |
3128 | } | |
ea0ebe96 | 3129 | } |
353632d9 | 3130 | |
e9e43ec3 | 3131 | if (bsb.magic[15] == '1') { |
10af14c4 N |
3132 | if (bsb.length == 0) |
3133 | continue; | |
e7a71c6b N |
3134 | if (info->delta_disks >= 0) { |
3135 | /* reshape_progress is increasing */ | |
3136 | if (__le64_to_cpu(bsb.arraystart) | |
3137 | + __le64_to_cpu(bsb.length) | |
3138 | < info->reshape_progress) { | |
3139 | nonew: | |
3140 | if (verbose) | |
3141 | fprintf(stderr, Name | |
3142 | ": backup-metadata found on %s but is not needed\n", devname); | |
3143 | continue; /* No new data here */ | |
3144 | } | |
3145 | } else { | |
3146 | /* reshape_progress is decreasing */ | |
3147 | if (__le64_to_cpu(bsb.arraystart) >= | |
3148 | info->reshape_progress) | |
3149 | goto nonew; /* No new data here */ | |
ea0ebe96 | 3150 | } |
e9e43ec3 | 3151 | } else { |
10af14c4 N |
3152 | if (bsb.length == 0 && bsb.length2 == 0) |
3153 | continue; | |
e7a71c6b N |
3154 | if (info->delta_disks >= 0) { |
3155 | /* reshape_progress is increasing */ | |
3156 | if ((__le64_to_cpu(bsb.arraystart) | |
3157 | + __le64_to_cpu(bsb.length) | |
3158 | < info->reshape_progress) | |
3159 | && | |
3160 | (__le64_to_cpu(bsb.arraystart2) | |
3161 | + __le64_to_cpu(bsb.length2) | |
3162 | < info->reshape_progress)) | |
3163 | goto nonew; /* No new data here */ | |
3164 | } else { | |
3165 | /* reshape_progress is decreasing */ | |
3166 | if (__le64_to_cpu(bsb.arraystart) >= | |
3167 | info->reshape_progress && | |
3168 | __le64_to_cpu(bsb.arraystart2) >= | |
3169 | info->reshape_progress) | |
3170 | goto nonew; /* No new data here */ | |
3171 | } | |
e9e43ec3 | 3172 | } |
ea0ebe96 N |
3173 | if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) { |
3174 | second_fail: | |
3175 | if (verbose) | |
e7a71c6b N |
3176 | fprintf(stderr, Name |
3177 | ": Failed to verify secondary backup-metadata block on %s\n", | |
ea0ebe96 | 3178 | devname); |
353632d9 | 3179 | continue; /* Cannot seek */ |
ea0ebe96 | 3180 | } |
2efedc7b | 3181 | /* There should be a duplicate backup superblock 4k before here */ |
06b0d786 | 3182 | if (lseek64(fd, -4096, 1) < 0 || |
0155af90 | 3183 | read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2)) |
ea0ebe96 | 3184 | goto second_fail; /* Cannot find leading superblock */ |
e9e43ec3 N |
3185 | if (bsb.magic[15] == '1') |
3186 | bsbsize = offsetof(struct mdp_backup_super, pad1); | |
3187 | else | |
3188 | bsbsize = offsetof(struct mdp_backup_super, pad); | |
ff94fb86 | 3189 | if (memcmp(&bsb2, &bsb, bsbsize) != 0) |
ea0ebe96 | 3190 | goto second_fail; /* Cannot find leading superblock */ |
2efedc7b | 3191 | |
353632d9 NB |
3192 | /* Now need the data offsets for all devices. */ |
3193 | offsets = malloc(sizeof(*offsets)*info->array.raid_disks); | |
3194 | for(j=0; j<info->array.raid_disks; j++) { | |
3195 | if (fdlist[j] < 0) | |
3196 | continue; | |
3da92f27 | 3197 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 NB |
3198 | /* FIXME should be this be an error */ |
3199 | continue; | |
a5d85af7 | 3200 | st->ss->getinfo_super(st, &dinfo, NULL); |
3da92f27 | 3201 | st->ss->free_super(st); |
14e5b4d7 | 3202 | offsets[j] = dinfo.data_offset * 512; |
353632d9 NB |
3203 | } |
3204 | printf(Name ": restoring critical section\n"); | |
3205 | ||
3206 | if (restore_stripes(fdlist, offsets, | |
3207 | info->array.raid_disks, | |
3208 | info->new_chunk, | |
3209 | info->new_level, | |
3210 | info->new_layout, | |
06b0d786 | 3211 | fd, __le64_to_cpu(bsb.devstart)*512, |
92dcdf7c | 3212 | __le64_to_cpu(bsb.arraystart)*512, |
e9e43ec3 N |
3213 | __le64_to_cpu(bsb.length)*512)) { |
3214 | /* didn't succeed, so giveup */ | |
ea0ebe96 N |
3215 | if (verbose) |
3216 | fprintf(stderr, Name ": Error restoring backup from %s\n", | |
3217 | devname); | |
e9e43ec3 N |
3218 | return 1; |
3219 | } | |
3220 | ||
3221 | if (bsb.magic[15] == '2' && | |
3222 | restore_stripes(fdlist, offsets, | |
3223 | info->array.raid_disks, | |
3224 | info->new_chunk, | |
3225 | info->new_level, | |
3226 | info->new_layout, | |
3227 | fd, __le64_to_cpu(bsb.devstart)*512 + | |
3228 | __le64_to_cpu(bsb.devstart2)*512, | |
92dcdf7c | 3229 | __le64_to_cpu(bsb.arraystart2)*512, |
e9e43ec3 | 3230 | __le64_to_cpu(bsb.length2)*512)) { |
353632d9 | 3231 | /* didn't succeed, so giveup */ |
ea0ebe96 N |
3232 | if (verbose) |
3233 | fprintf(stderr, Name ": Error restoring second backup from %s\n", | |
3234 | devname); | |
2295250a | 3235 | return 1; |
353632d9 NB |
3236 | } |
3237 | ||
e9e43ec3 | 3238 | |
353632d9 NB |
3239 | /* Ok, so the data is restored. Let's update those superblocks. */ |
3240 | ||
10af14c4 N |
3241 | lo = hi = 0; |
3242 | if (bsb.length) { | |
3243 | lo = __le64_to_cpu(bsb.arraystart); | |
3244 | hi = lo + __le64_to_cpu(bsb.length); | |
3245 | } | |
3246 | if (bsb.magic[15] == '2' && bsb.length2) { | |
3247 | unsigned long long lo1, hi1; | |
3248 | lo1 = __le64_to_cpu(bsb.arraystart2); | |
3249 | hi1 = lo1 + __le64_to_cpu(bsb.length2); | |
3250 | if (lo == hi) { | |
3251 | lo = lo1; | |
3252 | hi = hi1; | |
3253 | } else if (lo < lo1) | |
3254 | hi = hi1; | |
3255 | else | |
3256 | lo = lo1; | |
3257 | } | |
3258 | if (lo < hi && | |
3259 | (info->reshape_progress < lo || | |
3260 | info->reshape_progress > hi)) | |
3261 | /* backup does not affect reshape_progress*/ ; | |
3262 | else if (info->delta_disks >= 0) { | |
e9e43ec3 N |
3263 | info->reshape_progress = __le64_to_cpu(bsb.arraystart) + |
3264 | __le64_to_cpu(bsb.length); | |
3265 | if (bsb.magic[15] == '2') { | |
3266 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) + | |
3267 | __le64_to_cpu(bsb.length2); | |
3268 | if (p2 > info->reshape_progress) | |
3269 | info->reshape_progress = p2; | |
3270 | } | |
3271 | } else { | |
3272 | info->reshape_progress = __le64_to_cpu(bsb.arraystart); | |
3273 | if (bsb.magic[15] == '2') { | |
3274 | unsigned long long p2 = __le64_to_cpu(bsb.arraystart2); | |
3275 | if (p2 < info->reshape_progress) | |
3276 | info->reshape_progress = p2; | |
3277 | } | |
3278 | } | |
353632d9 NB |
3279 | for (j=0; j<info->array.raid_disks; j++) { |
3280 | if (fdlist[j] < 0) continue; | |
3da92f27 | 3281 | if (st->ss->load_super(st, fdlist[j], NULL)) |
353632d9 | 3282 | continue; |
a5d85af7 | 3283 | st->ss->getinfo_super(st, &dinfo, NULL); |
e9e43ec3 | 3284 | dinfo.reshape_progress = info->reshape_progress; |
3da92f27 | 3285 | st->ss->update_super(st, &dinfo, |
68c7d6d7 NB |
3286 | "_reshape_progress", |
3287 | NULL,0, 0, NULL); | |
3da92f27 NB |
3288 | st->ss->store_super(st, fdlist[j]); |
3289 | st->ss->free_super(st); | |
353632d9 | 3290 | } |
353632d9 NB |
3291 | return 0; |
3292 | } | |
6e9eac4f NB |
3293 | /* Didn't find any backup data, try to see if any |
3294 | * was needed. | |
3295 | */ | |
82f2d6ab N |
3296 | if (info->delta_disks < 0) { |
3297 | /* When shrinking, the critical section is at the end. | |
3298 | * So see if we are before the critical section. | |
3299 | */ | |
3300 | unsigned long long first_block; | |
3301 | nstripe = ostripe = 0; | |
3302 | first_block = 0; | |
3303 | while (ostripe >= nstripe) { | |
3304 | ostripe += info->array.chunk_size / 512; | |
3305 | first_block = ostripe * odata; | |
3306 | nstripe = first_block / ndata / (info->new_chunk/512) * | |
3307 | (info->new_chunk/512); | |
3308 | } | |
3309 | ||
3310 | if (info->reshape_progress >= first_block) | |
3311 | return 0; | |
6e9eac4f | 3312 | } |
82f2d6ab N |
3313 | if (info->delta_disks > 0) { |
3314 | /* See if we are beyond the critical section. */ | |
3315 | unsigned long long last_block; | |
3316 | nstripe = ostripe = 0; | |
3317 | last_block = 0; | |
3318 | while (nstripe >= ostripe) { | |
3319 | nstripe += info->new_chunk / 512; | |
3320 | last_block = nstripe * ndata; | |
3321 | ostripe = last_block / odata / (info->array.chunk_size/512) * | |
3322 | (info->array.chunk_size/512); | |
3323 | } | |
6e9eac4f | 3324 | |
82f2d6ab N |
3325 | if (info->reshape_progress >= last_block) |
3326 | return 0; | |
3327 | } | |
6e9eac4f | 3328 | /* needed to recover critical section! */ |
ea0ebe96 N |
3329 | if (verbose) |
3330 | fprintf(stderr, Name ": Failed to find backup of critical section\n"); | |
2295250a | 3331 | return 1; |
353632d9 | 3332 | } |
e9e43ec3 N |
3333 | |
3334 | int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info, | |
3335 | char *backup_file) | |
3336 | { | |
864a004f AK |
3337 | char buf[40]; |
3338 | char *container = NULL; | |
ef5414b2 | 3339 | int err; |
864a004f | 3340 | |
20a40eca N |
3341 | err = sysfs_set_str(info, NULL, "array_state", "readonly"); |
3342 | if (err) | |
3343 | return err; | |
3344 | if (st->ss->external) { | |
864a004f AK |
3345 | fmt_devname(buf, st->container_dev); |
3346 | container = buf; | |
493f5dd6 N |
3347 | freeze(st); |
3348 | if (info->reshape_active == 2) | |
3349 | return reshape_container(container, NULL, | |
3350 | st, info, 0, backup_file, | |
3351 | 0, 1); | |
864a004f AK |
3352 | } |
3353 | return reshape_array(container, mdfd, "array", st, info, 1, | |
3354 | backup_file, 0, 0, 1); | |
e9e43ec3 | 3355 | } |