]> git.ipfire.org Git - thirdparty/mdadm.git/blame - Grow.c
Make text_version available to reshape_array for subarray processing.
[thirdparty/mdadm.git] / Grow.c
CommitLineData
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
38int 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 210int 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 430static 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 447static __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
456static 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 481static 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
500static 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
511static 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
7f2ba464 533static void unfreeze(struct supertype *st, int frozen)
7236ee7a
N
534{
535 /* If 'frozen' is 1, unfreeze the array */
7f2ba464
DW
536 if (frozen <= 0)
537 return;
538
539 if (st->ss->external)
540 return unfreeze_container(st);
541 else {
542 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
543
544 if (sra)
545 sysfs_set_str(sra, NULL, "sync_action", "idle");
546 else
547 fprintf(stderr, Name ": failed to unfreeze array\n");
548 sysfs_free(sra);
549 }
7236ee7a
N
550}
551
4411fb17 552static void wait_reshape(struct mdinfo *sra)
7236ee7a
N
553{
554 int fd = sysfs_get_fd(sra, NULL, "sync_action");
555 char action[20];
556
92a19f1a
AK
557 if (fd < 0)
558 return;
559
560 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
561 strncmp(action, "reshape", 7) == 0) {
7236ee7a
N
562 fd_set rfds;
563 FD_ZERO(&rfds);
564 FD_SET(fd, &rfds);
565 select(fd+1, NULL, NULL, &rfds, NULL);
92a19f1a
AK
566 }
567 close(fd);
7236ee7a 568}
7bc71196
DW
569
570static int reshape_super(struct supertype *st, long long size, int level,
571 int layout, int chunksize, int raid_disks,
572 char *backup_file, char *dev, int verbose)
573{
574 /* nothing extra to check in the native case */
575 if (!st->ss->external)
576 return 0;
577 if (!st->ss->reshape_super ||
578 !st->ss->manage_reshape) {
579 fprintf(stderr, Name ": %s metadata does not support reshape\n",
580 st->ss->name);
581 return 1;
582 }
583
584 return st->ss->reshape_super(st, size, level, layout, chunksize,
585 raid_disks, backup_file, dev, verbose);
586}
587
588static void sync_metadata(struct supertype *st)
589{
590 if (st->ss->external) {
76266030 591 if (st->update_tail) {
7bc71196 592 flush_metadata_updates(st);
76266030
N
593 st->update_tail = &st->updates;
594 } else
7bc71196
DW
595 st->ss->sync_metadata(st);
596 }
597}
598
599static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
600{
601 /* when dealing with external metadata subarrays we need to be
602 * prepared to handle EAGAIN. The kernel may need to wait for
603 * mdmon to mark the array active so the kernel can handle
604 * allocations/writeback when preparing the reshape action
605 * (md_allow_write()). We temporarily disable safe_mode_delay
606 * to close a race with the array_state going clean before the
607 * next write to raid_disks / stripe_cache_size
608 */
609 char safe[50];
610 int rc;
611
612 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
5da9ab98
N
613 if (!container ||
614 (strcmp(name, "raid_disks") != 0 &&
615 strcmp(name, "stripe_cache_size") != 0))
7bc71196
DW
616 return sysfs_set_num(sra, NULL, name, n);
617
618 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
619 if (rc <= 0)
620 return -1;
621 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
622 rc = sysfs_set_num(sra, NULL, name, n);
623 if (rc < 0 && errno == EAGAIN) {
624 ping_monitor(container);
625 /* if we get EAGAIN here then the monitor is not active
626 * so stop trying
627 */
628 rc = sysfs_set_num(sra, NULL, name, n);
629 }
630 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
631 return rc;
632}
633
47eb4d5a
N
634int start_reshape(struct mdinfo *sra)
635{
636 int err;
0f28668b 637 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
47eb4d5a
N
638 err = sysfs_set_num(sra, NULL, "suspend_hi", 0);
639 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", 0);
640 err = err ?: sysfs_set_num(sra, NULL, "sync_min", 0);
641 err = err ?: sysfs_set_num(sra, NULL, "sync_max", 0);
642 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
643
644 return err;
645}
646
647void 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
62a48395
AK
657int remove_disks_on_raid10_to_raid0_takeover(struct supertype *st,
658 struct mdinfo *sra,
659 int layout)
660{
661 int nr_of_copies;
662 struct mdinfo *remaining;
663 int slot;
664
665 nr_of_copies = layout & 0xff;
666
667 remaining = sra->devs;
668 sra->devs = NULL;
669 /* for each 'copy', select one device and remove from the list. */
670 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
671 struct mdinfo **diskp;
672 int found = 0;
673
674 /* Find a working device to keep */
675 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
676 struct mdinfo *disk = *diskp;
677
678 if (disk->disk.raid_disk < slot)
679 continue;
680 if (disk->disk.raid_disk >= slot + nr_of_copies)
681 continue;
682 if (disk->disk.state & (1<<MD_DISK_REMOVED))
683 continue;
684 if (disk->disk.state & (1<<MD_DISK_FAULTY))
685 continue;
686 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
687 continue;
688
689 /* We have found a good disk to use! */
690 *diskp = disk->next;
691 disk->next = sra->devs;
692 sra->devs = disk;
693 found = 1;
694 break;
695 }
696 if (!found)
697 break;
698 }
699
700 if (slot < sra->array.raid_disks) {
701 /* didn't find all slots */
702 struct mdinfo **e;
703 e = &remaining;
704 while (*e)
705 e = &(*e)->next;
706 *e = sra->devs;
707 sra->devs = remaining;
708 return 1;
709 }
710
711 /* Remove all 'remaining' devices from the array */
712 while (remaining) {
713 struct mdinfo *sd = remaining;
714 remaining = sd->next;
715
716 sysfs_set_str(sra, sd, "state", "faulty");
717 sysfs_set_str(sra, sd, "slot", "none");
718 sysfs_set_str(sra, sd, "state", "remove");
719 sd->disk.state |= (1<<MD_DISK_REMOVED);
720 sd->disk.state &= ~(1<<MD_DISK_SYNC);
721 sd->next = sra->devs;
722 sra->devs = sd;
723 }
724 return 0;
725}
726
130994cb
AK
727void reshape_free_fdlist(int *fdlist,
728 unsigned long long *offsets,
729 int size)
730{
731 int i;
732
733 for (i = 0; i < size; i++)
734 if (fdlist[i] >= 0)
735 close(fdlist[i]);
736
737 free(fdlist);
738 free(offsets);
739}
740
741int reshape_prepare_fdlist(char *devname,
742 struct mdinfo *sra,
743 int raid_disks,
744 int nrdisks,
745 unsigned long blocks,
746 char *backup_file,
747 int *fdlist,
748 unsigned long long *offsets)
749{
750 int d = 0;
751 struct mdinfo *sd;
752
753 for (d = 0; d <= nrdisks; d++)
754 fdlist[d] = -1;
755 d = raid_disks;
756 for (sd = sra->devs; sd; sd = sd->next) {
757 if (sd->disk.state & (1<<MD_DISK_FAULTY))
758 continue;
759 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
760 char *dn = map_dev(sd->disk.major,
761 sd->disk.minor, 1);
762 fdlist[sd->disk.raid_disk]
763 = dev_open(dn, O_RDONLY);
764 offsets[sd->disk.raid_disk] = sd->data_offset*512;
765 if (fdlist[sd->disk.raid_disk] < 0) {
766 fprintf(stderr,
767 Name ": %s: cannot open component %s\n",
768 devname, dn ? dn : "-unknown-");
769 d = -1;
770 goto release;
771 }
772 } else if (backup_file == NULL) {
773 /* spare */
774 char *dn = map_dev(sd->disk.major,
775 sd->disk.minor, 1);
776 fdlist[d] = dev_open(dn, O_RDWR);
777 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
778 if (fdlist[d] < 0) {
779 fprintf(stderr, Name ": %s: cannot open component %s\n",
780 devname, dn ? dn : "-unknown-");
781 d = -1;
782 goto release;
783 }
784 d++;
785 }
786 }
787release:
788 return d;
789}
790
e6e9d47b
AK
791int reshape_open_backup_file(char *backup_file,
792 int fd,
793 char *devname,
794 long blocks,
795 int *fdlist,
796 unsigned long long *offsets)
797{
798 /* Return 1 on success, 0 on any form of failure */
799 /* need to check backup file is large enough */
800 char buf[512];
801 struct stat stb;
802 unsigned int dev;
803 int i;
804
805 *fdlist = open(backup_file, O_RDWR|O_CREAT|O_EXCL,
806 S_IRUSR | S_IWUSR);
807 *offsets = 8 * 512;
808 if (*fdlist < 0) {
809 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
810 devname, backup_file, strerror(errno));
811 return 0;
812 }
813 /* Guard against backup file being on array device.
814 * If array is partitioned or if LVM etc is in the
815 * way this will not notice, but it is better than
816 * nothing.
817 */
818 fstat(*fdlist, &stb);
819 dev = stb.st_dev;
820 fstat(fd, &stb);
821 if (stb.st_rdev == dev) {
822 fprintf(stderr, Name ": backup file must NOT be"
823 " on the array being reshaped.\n");
824 close(*fdlist);
825 return 0;
826 }
827
828 memset(buf, 0, 512);
829 for (i=0; i < blocks + 1 ; i++) {
830 if (write(*fdlist, buf, 512) != 512) {
831 fprintf(stderr, Name ": %s: cannot create"
832 " backup file %s: %s\n",
833 devname, backup_file, strerror(errno));
834 return 0;
835 }
836 }
837 if (fsync(*fdlist) != 0) {
838 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
839 devname, backup_file, strerror(errno));
840 return 0;
841 }
842
843 return 1;
844}
845
1c009fc2
AK
846unsigned long compute_backup_blocks(int nchunk, int ochunk,
847 unsigned int ndata, unsigned int odata)
848{
849 unsigned long a, b, blocks;
850 /* So how much do we need to backup.
851 * We need an amount of data which is both a whole number of
852 * old stripes and a whole number of new stripes.
853 * So LCM for (chunksize*datadisks).
854 */
855 a = (ochunk/512) * odata;
856 b = (nchunk/512) * ndata;
857 /* Find GCD */
858 while (a != b) {
859 if (a < b)
860 b -= a;
861 if (b < a)
862 a -= b;
863 }
864 /* LCM == product / GCD */
865 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
866
867 return blocks;
868}
869
5da9ab98
N
870char *analyse_change(struct mdinfo *info, struct reshape *re)
871{
872 /* Based on the current array state in info->array and
873 * the changes in info->new_* etc, determine:
874 * - whether the change is possible
875 * - Intermediate level/raid_disks/layout
876 * - whether a restriping reshape is needed
877 * - number of sectors in minimum change unit. This
878 * will cover a whole number of stripes in 'before' and
879 * 'after'.
880 *
881 * Return message if the change should be rejected
882 * NULL if the change can be achieved
883 *
884 * This can be called as part of starting a reshape, or
885 * when assembling an array that is undergoing reshape.
886 */
887 int new_disks;
888
889 /* If a new level not explicitly given, we assume no-change */
890 if (info->new_level == UnSet)
891 info->new_level = info->array.level;
892
893 if (info->new_chunk)
894 switch (info->new_level) {
895 case 0:
896 case 4:
897 case 5:
898 case 6:
899 case 10:
900 /* chunk size is meaningful, must divide component_size
901 * evenly
902 */
903 if (info->component_size % (info->new_chunk/512))
904 return "New chunk size does not"
905 " divide component size";
906 break;
907 default:
908 return "chunk size not meaningful for this level";
909 }
910 else
911 info->new_chunk = info->array.chunk_size;
912
913 switch (info->array.level) {
914 case 1:
915 /* RAID1 can convert to RAID1 with different disks, or
916 * raid5 with 2 disks
917 */
918 if (info->new_level == 1) {
919 if (info->delta_disks == UnSet)
920 /* Don't know what to do */
921 return "no change requested for Growing RAID1";
922 re->level = 1;
923 re->before.data_disks = (info->array.raid_disks +
924 info->delta_disks);
925 re->before.layout = 0;
b4f8e38b 926 re->backup_blocks = 0;
5da9ab98
N
927 re->parity = 0;
928 return NULL;
929 }
930 if (info->array.raid_disks == 2 &&
931 info->array.raid_disks == 5) {
932 /* simple in-place conversion */
933 re->level = 5;
934 re->parity = 1;
935 re->before.data_disks = 1;
936 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
b4f8e38b 937 re->backup_blocks = 0;
5da9ab98
N
938 return NULL;
939 }
940 /* Could do some multi-stage conversions, but leave that to
941 * later.
942 */
943 return "Impossibly level change request for RAID1";
944
945 case 10:
946 /* RAID10 can only be converted from near mode to
947 * RAID0 by removing some devices
948 */
949 if ((info->array.layout & ~0xff) != 0x100)
950 return "Cannot Grow RAID10 with far/offset layout";
951 /* number of devices must be multiple of number of copies */
952 if (info->array.raid_disks % (info->array.layout & 0xff))
953 return "RAID10 layout too complex for Grow operation";
954
955 if (info->new_level != 0)
956 return "RAID10 can only be changed to RAID0";
957 new_disks = (info->array.raid_disks
958 / (info->array.layout & 0xff));
959 if (info->delta_disks != UnSet) {
960 info->delta_disks = (new_disks
961 - info->array.raid_disks);
962 }
963 if (info->delta_disks != new_disks - info->array.raid_disks)
964 return "New number of raid-devices impossible for RAID10";
965 if (info->new_chunk &&
966 info->new_chunk != info->array.chunk_size)
967 return "Cannot change chunk-size with RAID10 Grow";
968
969 /* looks good */
970 re->level = 0;
971 re->parity = 0;
972 re->before.data_disks = new_disks;
973 re->before.layout = 0;
b4f8e38b 974 re->backup_blocks = 0;
5da9ab98
N
975 return NULL;
976
977 case 0:
978 /* RAID0 can be converted to RAID10, or to RAID456 */
979 if (info->new_level == 10) {
980 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
981 /* Assume near=2 layout */
982 info->new_layout = 0x102;
983 info->delta_disks = info->array.raid_disks;
984 }
985 if (info->new_layout == UnSet) {
986 int copies = 1 + (info->delta_disks
987 / info->array.raid_disks);
988 if (info->array.raid_disks * (copies-1)
989 != info->delta_disks)
990 return "Impossible number of devices"
991 " for RAID0->RAID10";
992 info->new_layout = 0x100 + copies;
993 }
994 if (info->delta_disks == UnSet) {
995 int copies = info->new_layout & 0xff;
996 if (info->new_layout != 0x100 + copies)
997 return "New layout impossible"
998 " for RAID0->RAID10";;
999 info->delta_disks = (copies - 1) *
1000 info->array.raid_disks;
1001 }
1002 if (info->new_chunk &&
1003 info->new_chunk != info->array.chunk_size)
1004 return "Cannot change chunk-size with RAID0->RAID10";
1005 /* looks good */
1006 re->level = 10;
1007 re->parity = 0;
1008 re->before.data_disks = (info->array.raid_disks +
1009 info->delta_disks);
1010 re->before.layout = info->new_layout;
b4f8e38b 1011 re->backup_blocks = 0;
5da9ab98
N
1012 return NULL;
1013 }
1014
1015 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1016 * a raid4 style layout of the final level.
1017 */
1018 switch (info->new_level) {
1019 case 0:
1020 case 4:
1021 re->level = 4;
1022 re->before.layout = 0;
1023 break;
1024 case 5:
1025 re->level = 5;
1026 re->before.layout = ALGORITHM_PARITY_N;
1027 break;
1028 case 6:
1029 re->level = 6;
1030 re->before.layout = ALGORITHM_PARITY_N;
1031 break;
1032 default:
1033 return "Impossible level change requested";
1034 }
1035 re->before.data_disks = info->array.raid_disks;
1036 /* determining 'after' layout happens outside this 'switch' */
1037 break;
1038
1039 case 4:
1040 info->array.layout = ALGORITHM_PARITY_N;
1041 case 5:
1042 switch (info->new_level) {
1043 case 4:
1044 re->level = info->array.level;
1045 re->before.data_disks = info->array.raid_disks - 1;
1046 re->before.layout = info->array.layout;
1047 break;
1048 case 5:
1049 re->level = 5;
1050 re->before.data_disks = info->array.raid_disks - 1;
1051 re->before.layout = info->array.layout;
1052 break;
1053 case 6:
1054 re->level = 6;
1055 re->before.data_disks = info->array.raid_disks - 1;
1056 switch (info->array.layout) {
1057 case ALGORITHM_LEFT_ASYMMETRIC:
1058 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1059 break;
1060 case ALGORITHM_RIGHT_ASYMMETRIC:
1061 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1062 break;
1063 case ALGORITHM_LEFT_SYMMETRIC:
1064 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1065 break;
1066 case ALGORITHM_RIGHT_SYMMETRIC:
1067 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1068 break;
1069 case ALGORITHM_PARITY_0:
1070 re->before.layout = ALGORITHM_PARITY_0_6;
1071 break;
1072 case ALGORITHM_PARITY_N:
1073 re->before.layout = ALGORITHM_PARITY_N_6;
1074 break;
1075 default:
1076 return "Cannot convert an array with this layout";
1077 }
1078 break;
1079 case 1:
1080 if (info->array.raid_disks != 2)
1081 return "Can only convert a 2-device array to RAID1";
1082 re->level = 1;
1083 re->before.data_disks = 2;
1084 re->before.layout = 0;
1085 break;
1086 default:
1087 return "Impossible level change requested";
1088 }
1089 break;
1090 case 6:
1091 switch (info->new_level) {
1092 case 4:
1093 case 5:
1094 case 6:
1095 re->level = 6;
1096 re->before.data_disks = info->array.raid_disks - 2;
1097 re->before.layout = info->array.layout;
1098 break;
1099 default:
1100 return "Impossible level change requested";
1101 }
1102 break;
1103 }
1104
1105 /* If we reached here then it looks like a re-stripe is
1106 * happening. We have determined the intermediate level
1107 * and initial raid_disks/layout and stored these in 're'.
1108 *
1109 * We need to deduce the final layout that can be atomically
1110 * converted to the end state.
1111 */
1112 switch (info->new_level) {
1113 case 0:
1114 /* We can only get to RAID0 from RAID4 or RAID5
1115 * with appropriate layout and one extra device
1116 */
1117 if (re->level != 4 && re->level != 5)
1118 return "Cannot covert to RAID0 from this level";
1119 if (info->delta_disks == UnSet)
1120 re->after.data_disks = re->before.data_disks;
1121 else
1122 re->after.data_disks =
1123 info->array.raid_disks + info->delta_disks;
1124 switch (re->level) {
1125 case 4:
1126 re->after.layout = 0 ; break;
1127 case 5:
1128 re->after.layout = ALGORITHM_PARITY_N; break;
1129 }
1130 break;
1131
1132 case 4:
1133 /* We can only get to RAID4 from RAID5 */
1134 if (re->level != 4 && re->level != 5)
1135 return "Cannot convert to RAID4 from this level";
1136 if (info->delta_disks == UnSet)
1137 re->after.data_disks = re->before.data_disks;
1138 else
1139 re->after.data_disks =
1140 re->before.data_disks + info->delta_disks;
1141 switch (re->level) {
1142 case 4:
1143 re->after.layout = 0 ; break;
1144 case 5:
1145 re->after.layout = ALGORITHM_PARITY_N; break;
1146 }
1147 break;
1148
1149 case 5:
1150 /* We get to RAID5 for RAID5 or RAID6 */
1151 if (re->level != 5 && re->level != 6)
1152 return "Cannot convert to RAID5 from this level";
1153 if (info->delta_disks == UnSet)
1154 re->after.data_disks = re->before.data_disks;
1155 else if (re->level == 5)
1156 re->after.data_disks =
1157 re->before.data_disks + info->delta_disks;
1158 else
1159 re->after.data_disks =
1160 info->array.raid_disks + info->delta_disks - 1;
1161 switch (re->level) {
1162 case 5:
1163 if (info->new_layout == UnSet)
1164 re->after.layout = re->before.layout;
1165 else
1166 re->after.layout = info->new_layout;
1167 break;
1168 case 6:
4a870364
N
1169 if (info->new_layout == UnSet)
1170 info->new_layout = re->before.layout;
1171
5da9ab98
N
1172 /* after.layout needs to be raid6 version of new_layout */
1173 if (info->new_layout == ALGORITHM_PARITY_N)
1174 re->after.layout = ALGORITHM_PARITY_N;
1175 else {
1176 char layout[40];
1177 char *ls = map_num(r5layout, info->new_layout);
1178 int l;
1179 strcat(strcpy(layout, ls), "-6");
1180 l = map_name(r6layout, layout);
1181 if (l == UnSet)
1182 return "Cannot find RAID6 layout"
1183 " to convert to";
1184 re->after.layout = l;
1185 }
1186 }
1187 break;
1188
1189 case 6:
1190 /* We must already be at level 6 */
1191 if (re->level != 6)
1192 return "Impossible level change";
1193 if (info->delta_disks == UnSet)
1194 re->after.data_disks = re->before.data_disks;
1195 else
1196 re->after.data_disks = (info->array.raid_disks +
1197 info->delta_disks) - 2;
1198 if (info->new_layout == UnSet)
4a870364 1199 re->after.layout = info->array.layout;
5da9ab98
N
1200 else
1201 re->after.layout = info->new_layout;
1202 break;
1203 default:
1204 return "Impossible level change requested";
1205 }
1206 switch (re->level) {
1207 case 6: re->parity = 2; break;
1208 case 4:
1209 case 5: re->parity = 1; break;
1210 default: re->parity = 0; break;
1211 }
1212 /* So we have a restripe operation, we need to calculate the number
1213 * of blocks per reshape operation.
1214 */
1215 if (info->new_chunk == 0)
1216 info->new_chunk = info->array.chunk_size;
1217 if (re->after.data_disks == re->before.data_disks &&
1218 re->after.layout == re->before.layout &&
1219 info->new_chunk == info->array.chunk_size) {
1220 /* Nothing to change */
b4f8e38b 1221 re->backup_blocks = 0;
5da9ab98
N
1222 return NULL;
1223 }
1224 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1225 /* chunks can layout changes make no difference */
b4f8e38b 1226 re->backup_blocks = 0;
5da9ab98
N
1227 return NULL;
1228 }
1229
1230 if (re->after.data_disks == re->before.data_disks &&
1231 get_linux_version() < 2006032)
1232 return "in-place reshape is not safe before 2.6.32 - sorry.";
1233
1234 if (re->after.data_disks < re->before.data_disks &&
1235 get_linux_version() < 2006030)
1236 return "reshape to fewer devices is not supported before 2.6.32 - sorry.";
1237
b4f8e38b 1238 re->backup_blocks = compute_backup_blocks(
5da9ab98
N
1239 info->new_chunk, info->array.chunk_size,
1240 re->after.data_disks,
1241 re->before.data_disks);
1242
1243 re->new_size = info->component_size * re->after.data_disks;
1244 return NULL;
1245}
1246
1247static int reshape_array(char *container, int fd, char *devname,
1248 struct supertype *st, struct mdinfo *info,
1249 int force, char *backup_file, int quiet, int forked);
1250static int reshape_container(char *container, int cfd, char *devname,
1251 struct supertype *st,
1252 struct mdinfo *info,
1253 int force,
1254 char *backup_file,
1255 int quiet);
1c009fc2 1256
06b0d786 1257int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
e86c9dd6 1258 long long size,
691a36b7
N
1259 int level, char *layout_str, int chunksize, int raid_disks,
1260 int force)
e86c9dd6
NB
1261{
1262 /* Make some changes in the shape of an array.
1263 * The kernel must support the change.
7236ee7a
N
1264 *
1265 * There are three different changes. Each can trigger
1266 * a resync or recovery so we freeze that until we have
1267 * requested everything (if kernel supports freezing - 2.6.30).
1268 * The steps are:
1269 * - change size (i.e. component_size)
1270 * - change level
1271 * - change layout/chunksize/ndisks
1272 *
1273 * The last can require a reshape. It is different on different
1274 * levels so we need to check the level before actioning it.
1275 * Some times the level change needs to be requested after the
1276 * reshape (e.g. raid6->raid5, raid5->raid0)
1277 *
e86c9dd6 1278 */
5da9ab98 1279 struct mdu_array_info_s array;
7236ee7a 1280 int rv = 0;
e86c9dd6 1281 struct supertype *st;
4725bc31 1282 char *subarray = NULL;
e86c9dd6 1283
7236ee7a 1284 int frozen;
7236ee7a 1285 int changed = 0;
7bc71196 1286 char *container = NULL;
5da9ab98 1287 char container_buf[20];
7bc71196 1288 int cfd = -1;
e86c9dd6 1289
5da9ab98 1290 struct mdinfo info;
7e0f6979 1291 struct mdinfo *sra;
e86c9dd6
NB
1292
1293 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1294 fprintf(stderr, Name ": %s is not an active md array - aborting\n",
1295 devname);
1296 return 1;
1297 }
24d40069 1298
9ce510be
N
1299 if (size >= 0 &&
1300 (chunksize || level!= UnSet || layout_str || raid_disks)) {
1301 fprintf(stderr, Name ": cannot change component size at the same time "
1302 "as other changes.\n"
1303 " Change size first, then check data is intact before "
1304 "making other changes.\n");
1305 return 1;
1306 }
1307
24d40069
N
1308 if (raid_disks && raid_disks < array.raid_disks && array.level > 1 &&
1309 get_linux_version() < 2006032 &&
1310 !check_env("MDADM_FORCE_FEWER")) {
1311 fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n"
1312 " Please use a newer kernel\n");
1313 return 1;
1314 }
7bc71196
DW
1315
1316 st = super_by_fd(fd, &subarray);
1317 if (!st) {
1318 fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname);
1319 return 1;
1320 }
acab7bb1
N
1321 if (raid_disks > st->max_devs) {
1322 fprintf(stderr, Name ": Cannot increase raid-disks on this array"
1323 " beyond %d\n", st->max_devs);
1324 return 1;
1325 }
7bc71196
DW
1326
1327 /* in the external case we need to check that the requested reshape is
1328 * supported, and perform an initial check that the container holds the
1329 * pre-requisite spare devices (mdmon owns final validation)
1330 */
1331 if (st->ss->external) {
1332 int container_dev;
7f2ba464 1333 int rv;
7bc71196
DW
1334
1335 if (subarray) {
1336 container_dev = st->container_dev;
1337 cfd = open_dev_excl(st->container_dev);
7bc71196
DW
1338 } else {
1339 container_dev = st->devnum;
1340 close(fd);
1341 cfd = open_dev_excl(st->devnum);
1342 fd = cfd;
1343 }
1344 if (cfd < 0) {
1345 fprintf(stderr, Name ": Unable to open container for %s\n",
1346 devname);
7f2ba464 1347 free(subarray);
7bc71196
DW
1348 return 1;
1349 }
1350
5da9ab98
N
1351 fmt_devname(container_buf, container_dev);
1352 container = container_buf;
7bc71196 1353
7f2ba464
DW
1354 if (subarray)
1355 rv = st->ss->load_container(st, cfd, NULL);
1356 else
1357 rv = st->ss->load_super(st, cfd, NULL);
1358 if (rv) {
7bc71196
DW
1359 fprintf(stderr, Name ": Cannot read superblock for %s\n",
1360 devname);
7f2ba464 1361 free(subarray);
7bc71196
DW
1362 return 1;
1363 }
1364
1365 if (mdmon_running(container_dev))
1366 st->update_tail = &st->updates;
691a36b7
N
1367 }
1368
1369 if (raid_disks > array.raid_disks &&
1370 array.spare_disks < (raid_disks - array.raid_disks) &&
1371 !force) {
1372 fprintf(stderr,
1373 Name ": Need %d spare%s to avoid degraded array,"
1374 " and only have %d.\n"
1375 " Use --force to over-ride this check.\n",
1376 raid_disks - array.raid_disks,
1377 raid_disks - array.raid_disks == 1 ? "" : "s",
1378 array.spare_disks);
1379 return 1;
7bc71196
DW
1380 }
1381
3fa436ac
ML
1382 sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS
1383 | GET_STATE | GET_VERSION);
1384 if (sra) {
7f2ba464 1385 if (st->ss->external && subarray == NULL) {
7bc71196
DW
1386 array.level = LEVEL_CONTAINER;
1387 sra->array.level = LEVEL_CONTAINER;
1388 }
7bc71196 1389 } else {
b526e52d
DW
1390 fprintf(stderr, Name ": failed to read sysfs parameters for %s\n",
1391 devname);
1392 return 1;
1393 }
7f2ba464
DW
1394 frozen = freeze(st);
1395 if (frozen < -1) {
1396 /* freeze() already spewed the reason */
1397 return 1;
1398 } else if (frozen < 0) {
7236ee7a
N
1399 fprintf(stderr, Name ": %s is performing resync/recovery and cannot"
1400 " be reshaped\n", devname);
1401 return 1;
1402 }
1403
1404 /* ========= set size =============== */
1405 if (size >= 0 && (size == 0 || size != array.size)) {
7bc71196
DW
1406 long long orig_size = array.size;
1407
1408 if (reshape_super(st, size, UnSet, UnSet, 0, 0, NULL, devname, !quiet)) {
1409 rv = 1;
1410 goto release;
1411 }
1412 sync_metadata(st);
7236ee7a
N
1413 array.size = size;
1414 if (array.size != size) {
1415 /* got truncated to 32bit, write to
1416 * component_size instead
1417 */
1418 if (sra)
1419 rv = sysfs_set_num(sra, NULL,
1420 "component_size", size);
1421 else
1422 rv = -1;
1423 } else
1424 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1425 if (rv != 0) {
39bbb392 1426 int err = errno;
7bc71196
DW
1427
1428 /* restore metadata */
1429 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1430 NULL, devname, !quiet) == 0)
1431 sync_metadata(st);
7236ee7a 1432 fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
39bbb392
N
1433 devname, strerror(err));
1434 if (err == EBUSY &&
1435 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1436 fprintf(stderr, " Bitmap must be removed before size can be changed\n");
7236ee7a
N
1437 rv = 1;
1438 goto release;
1439 }
1440 ioctl(fd, GET_ARRAY_INFO, &array);
be1cabbd 1441 size = get_component_size(fd)/2;
f98841b3
N
1442 if (size == 0)
1443 size = array.size;
7236ee7a 1444 if (!quiet)
f98841b3
N
1445 fprintf(stderr, Name ": component size of %s has been set to %lluK\n",
1446 devname, size);
7236ee7a 1447 changed = 1;
7bc71196 1448 } else if (array.level != LEVEL_CONTAINER) {
be1cabbd 1449 size = get_component_size(fd)/2;
f98841b3
N
1450 if (size == 0)
1451 size = array.size;
7236ee7a
N
1452 }
1453
62a48395 1454 /* ========= check for Raid10 -> Raid0 conversion ===============
5da9ab98 1455 * current implementation assumes that following conditions must be met:
62a48395
AK
1456 * - far_copies == 1
1457 * - near_copies == 2
1458 */
5da9ab98 1459 if (level == 0 && array.level == 10 && sra &&
62a48395
AK
1460 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) {
1461 int err;
1462 err = remove_disks_on_raid10_to_raid0_takeover(st, sra, array.layout);
1463 if (err) {
1464 dprintf(Name": Array cannot be reshaped\n");
62a48395
AK
1465 if (cfd > -1)
1466 close(cfd);
5da9ab98
N
1467 rv = 1;
1468 goto release;
7236ee7a
N
1469 }
1470 }
1471
5da9ab98 1472 info.array = array;
7443ee81 1473 sysfs_init(&info, fd, NoMdDev);
3fa436ac 1474 strcpy(info.text_version, sra->text_version);
7443ee81 1475 info.component_size = size*2;
5da9ab98
N
1476 info.new_level = level;
1477 info.new_chunk = chunksize * 1024;
1478 if (raid_disks)
1479 info.delta_disks = raid_disks - info.array.raid_disks;
1480 else
1481 info.delta_disks = UnSet;
1482 if (layout_str == NULL) {
1483 info.new_layout = UnSet;
1484 if (info.array.level == 6 &&
1485 (info.new_level == 6 || info.new_level == UnSet) &&
1486 info.array.layout >= 16) {
7236ee7a 1487 fprintf(stderr, Name
5da9ab98
N
1488 ": %s has a non-standard layout. If you"
1489 " wish to preserve this\n"
1490 " during the reshape, please specify"
1491 " --layout=preserve\n"
1492 " If you want to change it, specify a"
1493 " layout or use --layout=normalise\n",
7236ee7a
N
1494 devname);
1495 rv = 1;
1496 goto release;
1497 }
5da9ab98
N
1498 } else if (strcmp(layout_str, "normalise") == 0 ||
1499 strcmp(layout_str, "normalize") == 0) {
1500 /* If we have a -6 RAID6 layout, remove the '-6'. */
1501 info.new_layout = UnSet;
1502 if (info.array.level == 6 && info.new_level == UnSet) {
1503 char l[40], *h;
1504 strcpy(l, map_num(r6layout, info.array.layout));
1505 h = strrchr(l, '-');
1506 if (h && strcmp(h, "-6") == 0) {
1507 *h = 0;
1508 info.new_layout = map_name(r6layout, l);
7236ee7a
N
1509 }
1510 }
5da9ab98
N
1511 } else if (strcmp(layout_str, "preserve") == 0) {
1512 info.new_layout = UnSet;
1513 } else {
1514 int l = info.new_level;
1515 if (l == UnSet)
1516 l = info.array.level;
1517 switch (l) {
1518 case 5:
1519 info.new_layout = map_name(r5layout, layout_str);
1520 break;
1521 case 6:
1522 info.new_layout = map_name(r6layout, layout_str);
1523 break;
1524 case 10:
1525 info.new_layout = parse_layout_10(layout_str);
1526 break;
1527 case LEVEL_FAULTY:
1528 info.new_layout = parse_layout_faulty(layout_str);
1529 break;
1530 default:
1531 fprintf(stderr, Name ": layout not meaningful"
1532 " with this level\n");
7bc71196
DW
1533 rv = 1;
1534 goto release;
1535 }
5da9ab98
N
1536 if (info.new_layout == UnSet) {
1537 fprintf(stderr, Name ": layout %s not understood"
1538 " for this level\n",
1539 layout_str);
1540 rv = 1;
1541 goto release;
7bc71196 1542 }
7236ee7a
N
1543 }
1544
5da9ab98
N
1545 if (array.level == LEVEL_CONTAINER) {
1546 /* This change is to be applied to every array in the
1547 * container. This is only needed when the metadata imposes
1548 * restraints of the various arrays in the container.
1549 * Currently we only know that IMSM requires all arrays
1550 * to have the same number of devices so changing the
1551 * number of devices (On-Line Capacity Expansion) must be
1552 * performed at the level of the container
1553 */
1554 rv = reshape_container(container, fd, devname, st, &info,
1555 force, backup_file, quiet);
1556 } else {
1557 /* Impose these changes on a single array. First
1558 * check that the metadata is OK with the change. */
7bc71196 1559
5da9ab98
N
1560 if (reshape_super(st, info.component_size, info.new_level,
1561 info.new_layout, info.new_chunk,
1562 info.array.raid_disks + info.delta_disks,
1563 backup_file, devname, quiet)) {
7bc71196
DW
1564 rv = 1;
1565 goto release;
1566 }
5da9ab98
N
1567 sync_metadata(st);
1568 rv = reshape_array(container, fd, devname, st, &info, force,
1569 backup_file, quiet, 0);
1570 }
f94eedaf
N
1571 /* reshape_* released the array */
1572 return rv;
5da9ab98
N
1573release:
1574 unfreeze(st, frozen);
1575 return rv;
1576}
7bc71196 1577
5da9ab98
N
1578static int reshape_array(char *container, int fd, char *devname,
1579 struct supertype *st, struct mdinfo *info,
1580 int force,
1581 char *backup_file, int quiet, int forked)
1582{
1583 struct reshape reshape;
1584 int spares_needed;
1585 char *msg;
1586 int orig_level = UnSet;
1587 int disks, odisks;
7bc71196 1588
5da9ab98
N
1589 struct mdu_array_info_s array;
1590 char *c;
1591 int rv = 0;
e86c9dd6 1592
5da9ab98
N
1593 int *fdlist;
1594 unsigned long long *offsets;
1595 int d;
1596 int nrdisks;
1597 int err;
1598 int frozen;
da8100ca 1599 unsigned long blocks;
5da9ab98
N
1600 unsigned long cache;
1601 unsigned long long array_size;
1602 int done;
da8100ca 1603 struct mdinfo *sra;
7bc71196 1604
5da9ab98
N
1605 msg = analyse_change(info, &reshape);
1606 if (msg) {
1607 fprintf(stderr, Name ": %s\n", msg);
1608 return 1;
1609 }
9ff87c16
AK
1610 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
1611 dprintf("Canot get array information.\n");
1612 return 1;
1613 }
5da9ab98
N
1614 spares_needed = max(reshape.before.data_disks,
1615 reshape.after.data_disks)
1616 + reshape.parity - array.raid_disks;
7bc71196 1617
88c1a083
N
1618 if (!force &&
1619 info->new_level > 0 &&
1620 spares_needed > info->array.spare_disks) {
5da9ab98
N
1621 fprintf(stderr,
1622 Name ": Need %d spare%s to avoid degraded array,"
1623 " and only have %d.\n"
1624 " Use --force to over-ride this check.\n",
1625 spares_needed,
1626 spares_needed == 1 ? "" : "s",
1627 info->array.spare_disks);
1628 return 1;
7bc71196 1629 }
e86c9dd6 1630
5da9ab98
N
1631 if (reshape.level != info->array.level) {
1632 char *c = map_num(pers, reshape.level);
1633 int err;
1634 if (c == NULL)
1635 return 1; /* This should not be possible */
e86c9dd6 1636
5da9ab98
N
1637 err = sysfs_set_str(info, NULL, "level", c);
1638 if (err) {
1639 err = errno;
1640 fprintf(stderr, Name ": %s: could not set level to %s\n",
1641 devname, c);
1642 if (err == EBUSY &&
1643 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
1644 fprintf(stderr, " Bitmap must be removed"
1645 " before level can be changed\n");
1646 return 1;
19678e53 1647 }
5da9ab98 1648 if (!quiet)
b6b95155 1649 fprintf(stderr, Name ": level of %s changed to %s\n",
5da9ab98
N
1650 devname, c);
1651 orig_level = info->array.level;
1652 }
e86c9dd6 1653
5da9ab98
N
1654 if (reshape.level > 0 && st->ss->external &&
1655 !mdmon_running(st->container_dev)) {
1656 start_mdmon(st->container_dev);
1657 ping_monitor(container);
1658 }
e86c9dd6 1659
5da9ab98
N
1660 /* ->reshape_super might have chosen some spares from the
1661 * container that it wants to be part of the new array.
1662 * We can collect them with ->container_content and give
1663 * them to the kernel.
1664 */
1665 if (st->ss->reshape_super && st->ss->container_content) {
1666 char *subarray = strchr(info->text_version+1, '/')+1;
1667 struct mdinfo *info2 =
1668 st->ss->container_content(st, subarray);
1669 struct mdinfo *d;
1670
2dd0d697
AK
1671 if (info2) {
1672 sysfs_init(info2, fd, st->devnum);
5da9ab98
N
1673 for (d = info2->devs; d; d = d->next) {
1674 if (d->disk.state == 0 &&
1675 d->disk.raid_disk >= 0) {
1676 /* This is a spare that wants to
1677 * be part of the array.
1678 */
1679 add_disk(fd, st, info2, d);
1680 }
9860f271 1681 }
2dd0d697
AK
1682 sysfs_free(info2);
1683 }
5da9ab98 1684 }
1686dc25 1685
b4f8e38b 1686 if (reshape.backup_blocks == 0) {
5da9ab98
N
1687 /* No restriping needed, but we might need to impose
1688 * some more changes: layout, raid_disks, chunk_size
e86c9dd6 1689 */
5da9ab98
N
1690 if (info->new_layout != UnSet &&
1691 info->new_layout != info->array.layout) {
1692 info->array.layout = info->new_layout;
1693 if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) {
1694 fprintf(stderr, Name ": failed to set new layout\n");
1695 rv = 1;
1696 } else if (!quiet)
1697 printf("layout for %s set to %d\n",
1698 devname, info->array.layout);
1699 }
1700 if (info->delta_disks != UnSet &&
1701 info->delta_disks != 0) {
1702 info->array.raid_disks += info->delta_disks;
1703 if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) {
1704 fprintf(stderr, Name ": failed to set raid disks\n");
1705 rv = 1;
1706 } else if (!quiet)
1707 printf("raid_disks for %s set to %d\n",
1708 devname, info->array.raid_disks);
1709 }
1710 if (info->new_chunk != 0 &&
1711 info->new_chunk != info->array.chunk_size) {
1712 if (sysfs_set_num(info, NULL,
1713 "chunk_size", info->new_chunk) != 0) {
1714 fprintf(stderr, Name ": failed to set chunk size\n");
1715 rv = 1;
1716 } else if (!quiet)
1717 printf("chunk size for %s set to %d\n",
1718 devname, info->array.chunk_size);
4725bc31 1719 }
7236ee7a 1720
5da9ab98
N
1721 return rv;
1722 }
19678e53 1723
5da9ab98
N
1724 /*
1725 * There are three possibilities.
1726 * 1/ The array will shrink.
1727 * We need to ensure the reshape will pause before reaching
1728 * the 'critical section'. We also need to fork and wait for
1729 * that to happen. When it does we
1730 * suspend/backup/complete/unfreeze
1731 *
1732 * 2/ The array will not change size.
1733 * This requires that we keep a backup of a sliding window
1734 * so that we can restore data after a crash. So we need
1735 * to fork and monitor progress.
1736 * In future we will allow the data_offset to change, so
1737 * a sliding backup becomes unnecessary.
1738 *
1739 * 3/ The array will grow. This is relatively easy.
1740 * However the kernel's restripe routines will cheerfully
1741 * overwrite some early data before it is safe. So we
1742 * need to make a backup of the early parts of the array
1743 * and be ready to restore it if rebuild aborts very early.
1744 * For externally managed metadata, we still need a forked
1745 * child to monitor the reshape and suspend IO over the region
1746 * that is being reshaped.
1747 *
1748 * We backup data by writing it to one spare, or to a
1749 * file which was given on command line.
1750 *
1751 * In each case, we first make sure that storage is available
1752 * for the required backup.
1753 * Then we:
1754 * - request the shape change.
1755 * - fork to handle backup etc.
1756 */
e86c9dd6 1757
5da9ab98
N
1758 /* Check that we can hold all the data */
1759 get_dev_size(fd, NULL, &array_size);
1760 if (reshape.new_size < (array_size/512)) {
1761 fprintf(stderr,
1762 Name ": this change will reduce the size of the array.\n"
1763 " use --grow --array-size first to truncate array.\n"
1764 " e.g. mdadm --grow %s --array-size %llu\n",
1765 devname, reshape.new_size/2);
1766 rv = 1;
1767 goto release;
1768 }
7236ee7a 1769
5da9ab98 1770 sra = sysfs_read(fd, 0,
3656edcd 1771 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
5da9ab98 1772 GET_CACHE);
d2505cff 1773
5da9ab98
N
1774 if (!sra) {
1775 fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
1776 devname);
1777 rv = 1;
1778 goto release;
1779 }
7236ee7a 1780
5da9ab98
N
1781 /* Decide how many blocks (sectors) for a reshape
1782 * unit. The number we have so far is just a minimum
1783 */
b4f8e38b 1784 blocks = reshape.backup_blocks;
5da9ab98
N
1785 if (reshape.before.data_disks ==
1786 reshape.after.data_disks) {
1787 /* Make 'blocks' bigger for better throughput, but
1788 * not so big that we reject it below.
1789 * Try for 16 megabytes
1790 */
1791 while (blocks * 32 < sra->component_size &&
1792 blocks < 16*1024*2)
1793 blocks *= 2;
1794 } else
1795 fprintf(stderr, Name ": Need to backup %luK of critical "
1796 "section..\n", blocks/2);
1797
1798 if (blocks >= sra->component_size/2) {
1799 fprintf(stderr, Name ": %s: Something wrong"
1800 " - reshape aborted\n",
1801 devname);
1802 rv = 1;
1803 goto release;
1804 }
7236ee7a 1805
5da9ab98
N
1806 /* Now we need to open all these devices so we can read/write.
1807 */
1808 nrdisks = array.raid_disks + sra->array.spare_disks;
1809 fdlist = malloc((1+nrdisks) * sizeof(int));
1810 offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
1811 if (!fdlist || !offsets) {
1812 fprintf(stderr, Name ": malloc failed: grow aborted\n");
1813 rv = 1;
1814 goto release;
1815 }
e380d3be 1816
5da9ab98
N
1817 d = reshape_prepare_fdlist(devname, sra, array.raid_disks,
1818 nrdisks, blocks, backup_file,
1819 fdlist, offsets);
1820 if (d < 0) {
1821 rv = 1;
1822 goto release;
1823 }
1824 if (backup_file == NULL) {
374ba335 1825 if (reshape.after.data_disks <= reshape.before.data_disks) {
5da9ab98
N
1826 fprintf(stderr,
1827 Name ": %s: Cannot grow - need backup-file\n",
c03ef02d
N
1828 devname);
1829 rv = 1;
5da9ab98
N
1830 goto release;
1831 } else if (sra->array.spare_disks == 0) {
1832 fprintf(stderr, Name ": %s: Cannot grow - need a spare or "
1833 "backup-file to backup critical section\n",
e86c9dd6 1834 devname);
7236ee7a 1835 rv = 1;
5da9ab98 1836 goto release;
e86c9dd6 1837 }
5da9ab98
N
1838 } else {
1839 if (!reshape_open_backup_file(backup_file, fd, devname,
1840 (signed)blocks,
1841 fdlist+d, offsets+d)) {
130994cb
AK
1842 rv = 1;
1843 goto release;
e86c9dd6 1844 }
5da9ab98
N
1845 d++;
1846 }
7236ee7a 1847
5da9ab98
N
1848 /* lastly, check that the internal stripe cache is
1849 * large enough, or it won't work.
1850 * It must hold at least 4 stripes of the larger
1851 * chunk size
1852 */
1853 cache = max(info->array.chunk_size, info->new_chunk);
1854 cache *= 4; /* 4 stripes minimum */
1855 cache /= 512; /* convert to sectors */
1856 disks = min(reshape.before.data_disks, reshape.after.data_disks);
1857 /* make sure there is room for 'blocks' with a bit to spare */
1858 if (cache < 16 + blocks / disks)
1859 cache = 16 + blocks / disks;
1860 cache /= (4096/512); /* Covert from sectors to pages */
1861
1862 if (sra->cache_size < cache)
1863 subarray_set_num(container, sra, "stripe_cache_size",
1864 cache+1);
1865
1866 /* Right, everything seems fine. Let's kick things off.
1867 * If only changing raid_disks, use ioctl, else use
1868 * sysfs.
1869 */
1870 sync_metadata(st);
1871
b4f8e38b
N
1872 sra->new_chunk = info->new_chunk;
1873
5da9ab98 1874 if (info->array.chunk_size == info->new_chunk &&
7477d513
AK
1875 reshape.before.layout == reshape.after.layout &&
1876 st->ss->external == 0) {
5da9ab98
N
1877 array.raid_disks = reshape.after.data_disks + reshape.parity;
1878 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1879 int err = errno;
7bc71196 1880 rv = 1;
5da9ab98
N
1881 fprintf(stderr,
1882 Name ": Cannot set device shape for %s: %s\n",
1883 devname, strerror(errno));
7bc71196 1884
5da9ab98
N
1885 if (err == EBUSY &&
1886 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1887 fprintf(stderr,
1888 " Bitmap must be removed before"
1889 " shape can be changed\n");
b5420ef3 1890
5da9ab98
N
1891 goto release;
1892 }
1893 } else {
1894 /* set them all just in case some old 'new_*' value
1895 * persists from some earlier problem
7236ee7a 1896 */
5da9ab98
N
1897 int err = err; /* only used if rv==1, and always set if
1898 * rv==1, so initialisation not needed,
1899 * despite gcc warning
1900 */
1901 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
1902 rv = 1, err = errno;
1903 if (!rv && sysfs_set_num(sra, NULL, "layout",
1904 reshape.after.layout) < 0)
1905 rv = 1, err = errno;
1906 if (!rv && subarray_set_num(container, sra, "raid_disks",
1907 reshape.after.data_disks +
1908 reshape.parity) < 0)
1909 rv = 1, err = errno;
1910 if (rv) {
1911 fprintf(stderr, Name ": Cannot set device shape for %s\n",
1912 devname);
7236ee7a 1913
5da9ab98
N
1914 if (err == EBUSY &&
1915 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1916 fprintf(stderr,
1917 " Bitmap must be removed before"
1918 " shape can be changed\n");
1919 goto release;
e86c9dd6 1920 }
5da9ab98 1921 }
e86c9dd6 1922
5da9ab98 1923 start_reshape(sra);
5da9ab98
N
1924
1925 /* Now we just need to kick off the reshape and watch, while
1926 * handling backups of the data...
1927 * This is all done by a forked background process.
1928 */
1929 switch(forked ? 0 : fork()) {
1930 case 0:
1931 close(fd);
1932 if (check_env("MDADM_GROW_VERIFY"))
1933 fd = open(devname, O_RDONLY | O_DIRECT);
1934 else
1935 fd = -1;
1936 mlockall(MCL_FUTURE);
1937
1938 odisks = reshape.before.data_disks + reshape.parity;
1939
999b4972
N
1940 if (st->ss->external) {
1941 /* metadata handler takes it from here */
1942 done = st->ss->manage_reshape(
1943 fd, sra, &reshape, st, blocks,
1944 fdlist, offsets,
1945 d - odisks, fdlist+odisks,
1946 offsets+odisks);
1947 } else
1948 done = child_monitor(
1949 fd, sra, &reshape, st, blocks,
1950 fdlist, offsets,
1951 d - odisks, fdlist+odisks,
1952 offsets+odisks);
7443ee81 1953
5da9ab98
N
1954 if (backup_file && done)
1955 unlink(backup_file);
582496b2 1956 if (!done) {
43475447 1957 abort_reshape(sra);
582496b2
AK
1958 goto out;
1959 }
1960 /* set new array size if required customer_array_size is used
1961 * by this metadata.
1962 */
1963 if (reshape.before.data_disks !=
1964 reshape.after.data_disks &&
1965 info->custom_array_size) {
1966 struct mdinfo *info2;
1967 char *subarray = strchr(info->text_version+1, '/')+1;
1968
1969 wait_reshape(sra);
1970 ping_monitor(container);
1971
1972 info2 = st->ss->container_content(st, subarray);
1973 if (info2) {
1974 unsigned long long current_size = 0;
1975 unsigned long long new_size =
1976 info2->custom_array_size/2;
1977
1978 if (sysfs_get_ll(sra,
1979 NULL,
1980 "array_size",
1981 &current_size) == 0 &&
1982 new_size > current_size) {
1983 if (sysfs_set_num(sra, NULL,
1984 "array_size", new_size)
1985 < 0)
1986 dprintf("Error: Cannot"
1987 " set array size");
1988 else
1989 dprintf("Array size "
1990 "changed");
1991 dprintf(" from %llu to %llu.\n",
1992 current_size, new_size);
1993 }
1994 sysfs_free(info2);
1995 }
1996 }
1997
1b3cbac5 1998 if (info->new_level != reshape.level) {
5da9ab98 1999 /* We need to wait for the reshape to finish
582496b2
AK
2000 * (which will have happened unless
2001 * odata < ndata) and then set the level
5da9ab98 2002 */
e86c9dd6 2003
582496b2
AK
2004 if (reshape.before.data_disks <
2005 reshape.after.data_disks)
2006 wait_reshape(sra);
2007
5da9ab98 2008 c = map_num(pers, info->new_level);
374ba335
N
2009 if (c == NULL)
2010 goto out;/* not possible */
5da9ab98 2011
5da9ab98
N
2012 err = sysfs_set_str(sra, NULL, "level", c);
2013 if (err)
582496b2 2014 fprintf(stderr, Name\
b6b95155 2015 ": %s: could not set level "
582496b2 2016 "to %s\n", devname, c);
e86c9dd6 2017 }
582496b2 2018 out:
5da9ab98
N
2019 if (forked)
2020 return 0;
2021 exit(0);
2022 case -1:
2023 fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n",
2024 strerror(errno));
2025 rv = 1;
2026 abort_reshape(sra);
2027 break;
2028 default:
2029 /* The child will take care of unfreezing the array */
2030 frozen = 0;
7236ee7a 2031 break;
7236ee7a 2032 }
e86c9dd6 2033
5da9ab98 2034
7236ee7a 2035 release:
6eb48844
N
2036 if (!rv) {
2037 if (container)
2038 ping_monitor(container);
2039 if (st->ss->external) {
2040 /* Re-load the metadata as much could have changed */
2041 int cfd = open_dev(st->container_dev);
2042 if (cfd >= 0) {
2043 st->ss->free_super(st);
2044 st->ss->load_container(st, cfd, container);
2045 close(cfd);
2046 }
1bb174ba
AK
2047 }
2048 }
7236ee7a
N
2049 if (rv && orig_level != UnSet && sra) {
2050 c = map_num(pers, orig_level);
2051 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
2052 fprintf(stderr, Name ": aborting level change\n");
2053 }
7f2ba464 2054 unfreeze(st, frozen);
7236ee7a
N
2055 return rv;
2056}
e86c9dd6 2057
5da9ab98
N
2058int reshape_container(char *container, int cfd, char *devname,
2059 struct supertype *st,
2060 struct mdinfo *info,
2061 int force,
2062 char *backup_file,
2063 int quiet)
2064{
1bb174ba
AK
2065 struct mdinfo *cc = NULL;
2066
5da9ab98
N
2067 if (reshape_super(st, info->component_size, info->new_level,
2068 info->new_layout, info->new_chunk,
2069 info->array.raid_disks + info->delta_disks,
2070 backup_file, devname, quiet))
2071 return 1;
2072
2073 sync_metadata(st);
2074
1bb174ba
AK
2075 /* ping monitor to be sure that update is on disk
2076 */
2077 ping_monitor(container);
5da9ab98
N
2078
2079 switch (fork()) {
2080 case -1: /* error */
2081 perror("Cannot fork to complete reshape\n");
2082 return 1;
2083 default: /* parent */
2084 printf(Name ": multi-array reshape continues in background\n");
2085 return 0;
2086 case 0: /* child */
2087 break;
2088 }
2089
1bb174ba
AK
2090 while(1) {
2091 /* For each member array with reshape_active,
2092 * we need to perform the reshape.
2093 * We pick the first array that needs reshaping and
2094 * reshape it. reshape_array() will re-read the metadata
2095 * so the next time through a different array should be
2096 * ready for reshape.
2097 */
2098 struct mdinfo *content;
5da9ab98
N
2099 int rv;
2100 int fd;
2101 struct mdstat_ent *mdstat;
5da9ab98
N
2102 char *adev;
2103
1bb174ba 2104 sysfs_free(cc);
5da9ab98 2105
1bb174ba
AK
2106 cc = st->ss->container_content(st, NULL);
2107
2108 for (content = cc; content ; content = content->next) {
2109 char *subarray;
2110 if (!content->reshape_active)
2111 continue;
2112
2113 subarray = strchr(content->text_version+1, '/')+1;
2114 mdstat = mdstat_by_subdev(subarray,
2115 devname2devnum(container));
2116 if (!mdstat)
2117 continue;
2118 break;
2119 }
2120 if (!content)
2121 break;
5da9ab98 2122
5da9ab98
N
2123 fd = open_dev_excl(mdstat->devnum);
2124 if (fd < 0)
2125 break;
2126 adev = map_dev(dev2major(mdstat->devnum),
2127 dev2minor(mdstat->devnum),
2128 0);
2129 if (!adev)
1bb174ba
AK
2130 adev = content->text_version;
2131
2132 sysfs_init(content, fd, mdstat->devnum);
5da9ab98 2133
1bb174ba
AK
2134 rv = reshape_array(container, fd, adev, st,
2135 content, force,
5da9ab98
N
2136 backup_file, quiet, 1);
2137 close(fd);
2138 if (rv)
2139 break;
2140 }
1bb174ba 2141 sysfs_free(cc);
5da9ab98
N
2142 exit(0);
2143}
2144
7236ee7a
N
2145/*
2146 * We run a child process in the background which performs the following
2147 * steps:
2148 * - wait for resync to reach a certain point
2149 * - suspend io to the following section
2150 * - backup that section
2151 * - allow resync to proceed further
2152 * - resume io
2153 * - discard the backup.
2154 *
2155 * When are combined in slightly different ways in the three cases.
2156 * Grow:
2157 * - suspend/backup/allow/wait/resume/discard
2158 * Shrink:
2159 * - allow/wait/suspend/backup/allow/wait/resume/discard
2160 * same-size:
2161 * - wait/resume/discard/suspend/backup/allow
2162 *
2163 * suspend/backup/allow always come together
2164 * wait/resume/discard do too.
2165 * For the same-size case we have two backups to improve flow.
2166 *
2167 */
e86c9dd6 2168
7443ee81
N
2169int progress_reshape(struct mdinfo *info, struct reshape *reshape,
2170 unsigned long long backup_point,
2171 unsigned long long wait_point,
2172 unsigned long long *suspend_point,
2173 unsigned long long *reshape_completed)
2174{
2175 /* This function is called repeatedly by the reshape manager.
2176 * It determines how much progress can safely be made and allows
2177 * that progress.
2178 * - 'info' identifies the array and particularly records in
2179 * ->reshape_progress the metadata's knowledge of progress
2180 * This is a sector offset from the start of the array
2181 * of the next array block to be relocated. This number
2182 * may increase from 0 or decrease from array_size, depending
2183 * on the type of reshape that is happening.
2184 * Note that in contrast, 'sync_completed' is a block count of the
2185 * reshape so far. It gives the distance between the start point
2186 * (head or tail of device) and the next place that data will be
2187 * written. It always increases.
2188 * - 'reshape' is the structure created by analyse_change
2189 * - 'backup_point' shows how much the metadata manager has backed-up
2190 * data. For reshapes with increasing progress, it is the next address
2191 * to be backed up, previous addresses have been backed-up. For
2192 * decreasing progress, it is the earliest address that has been
2193 * backed up - later address are also backed up.
2194 * So addresses between reshape_progress and backup_point are
2195 * backed up providing those are in the 'correct' order.
2196 * - 'wait_point' is an array address. When reshape_completed
2197 * passes this point, progress_reshape should return. It might
2198 * return earlier if it determines that ->reshape_progress needs
2199 * to be updated or further backup is needed.
2200 * - suspend_point is maintained by progress_reshape and the caller
2201 * should not touch it except to initialise to zero.
2202 * It is an array address and it only increases in 2.6.37 and earlier.
b6b95155 2203 * This makes it difficult to handle reducing reshapes with
7443ee81
N
2204 * external metadata.
2205 * However: it is similar to backup_point in that it records the
2206 * other end of a suspended region from reshape_progress.
2207 * it is moved to extend the region that is safe to backup and/or
2208 * reshape
2209 * - reshape_completed is read from sysfs and returned. The caller
2210 * should copy this into ->reshape_progress when it has reason to
2211 * believe that the metadata knows this, and any backup outside this
2212 * has been erased.
2213 *
2214 * Return value is:
2215 * 1 if more data from backup_point - but only as far as suspend_point,
2216 * should be backed up
2217 * 0 if things are progressing smoothly
2218 * -1 if the reshape is finished, either because it is all done,
2219 * or due to an error.
2220 */
2221
2222 int advancing = (reshape->after.data_disks
2223 >= reshape->before.data_disks);
d0ab945e
N
2224 unsigned long long need_backup; /* need to eventually backup all the way
2225 * to here
2226 */
7443ee81 2227 unsigned long long read_offset, write_offset;
b4f8e38b 2228 unsigned long long write_range;
7443ee81 2229 unsigned long long max_progress, target, completed;
d0ab945e
N
2230 unsigned long long array_size = (info->component_size
2231 * reshape->before.data_disks);
7443ee81
N
2232 int fd;
2233
2234 /* First, we unsuspend any region that is now known to be safe.
2235 * If suspend_point is on the 'wrong' side of reshape_progress, then
2236 * we don't have or need suspension at the moment. This is true for
2237 * native metadata when we don't need to back-up.
2238 */
2239 if (advancing) {
49cd738b 2240 if (info->reshape_progress <= *suspend_point)
7443ee81
N
2241 sysfs_set_num(info, NULL, "suspend_lo",
2242 info->reshape_progress);
2243 } else {
2244 /* Note: this won't work in 2.6.37 and before.
2245 * Something somewhere should make sure we don't need it!
2246 */
49cd738b 2247 if (info->reshape_progress >= *suspend_point)
7443ee81
N
2248 sysfs_set_num(info, NULL, "suspend_hi",
2249 info->reshape_progress);
2250 }
2251
2252 /* Now work out how far it is safe to progress.
2253 * If the read_offset for ->reshape_progress is less than
2254 * 'blocks' beyond the write_offset, we can only progress as far
2255 * as a backup.
2256 * Otherwise we can progress until the write_offset for the new location
2257 * reaches (within 'blocks' of) the read_offset at the current location.
2258 * However that region must be suspended unless we are using native
2259 * metadata.
2260 * If we need to suspend more, we limit it to 128M per device, which is
2261 * rather arbitrary and should be some time-based calculation.
2262 */
ec757320
N
2263 read_offset = info->reshape_progress / reshape->before.data_disks;
2264 write_offset = info->reshape_progress / reshape->after.data_disks;
b4f8e38b 2265 write_range = info->new_chunk/512;
7443ee81 2266 if (advancing) {
d0ab945e 2267 need_backup = 0;
7443ee81
N
2268 if (read_offset < write_offset + write_range) {
2269 max_progress = backup_point;
d0ab945e
N
2270 if (reshape->before.data_disks == reshape->after.data_disks)
2271 need_backup = array_size;
2272 else
2273 need_backup = reshape->backup_blocks;
7443ee81
N
2274 } else {
2275 max_progress =
2f3dd5e4
N
2276 read_offset *
2277 reshape->after.data_disks;
7443ee81
N
2278 }
2279 } else {
d0ab945e 2280 need_backup = array_size;
7443ee81
N
2281 if (read_offset > write_offset - write_range) {
2282 max_progress = backup_point;
2283 if (max_progress >= info->reshape_progress)
d0ab945e 2284 need_backup = 0;
7443ee81
N
2285 } else {
2286 max_progress =
2f3dd5e4
N
2287 read_offset *
2288 reshape->after.data_disks;
7443ee81
N
2289 /* If we are using internal metadata, then we can
2290 * progress all the way to the suspend_point without
2291 * worrying about backing-up/suspending along the
2292 * way.
2293 */
2294 if (max_progress < *suspend_point &&
2295 info->array.major_version >= 0)
2296 max_progress = *suspend_point;
2297 }
2298 }
2299
2300 /* We know it is safe to progress to 'max_progress' providing
2301 * it is suspended or we are using native metadata.
2302 * Consider extending suspend_point 128M per device if it
2303 * is less than 64M per device beyond reshape_progress.
2304 * But always do a multiple of 'blocks'
832b2b9a
N
2305 * FIXME this is too big - it takes to long to complete
2306 * this much.
7443ee81
N
2307 */
2308 target = 64*1024*2 * min(reshape->before.data_disks,
2309 reshape->after.data_disks);
b4f8e38b 2310 target /= reshape->backup_blocks;
7443ee81
N
2311 if (target < 2)
2312 target = 2;
b4f8e38b 2313 target *= reshape->backup_blocks;
7443ee81
N
2314
2315 /* For externally managed metadata we always need to suspend IO to
2316 * the area being reshaped so we regularly push suspend_point forward.
2317 * For native metadata we only need the suspend if we are going to do
2318 * a backup.
2319 */
2320 if (advancing) {
d0ab945e
N
2321 if ((need_backup > info->reshape_progress
2322 || info->array.major_version < 0) &&
7443ee81 2323 *suspend_point < info->reshape_progress + target) {
d0ab945e
N
2324 if (need_backup < *suspend_point + 2 * target)
2325 *suspend_point = need_backup;
2326 else if (*suspend_point + 2 * target < array_size)
7443ee81 2327 *suspend_point += 2 * target;
d0ab945e
N
2328 else
2329 *suspend_point = array_size;
7443ee81 2330 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
d0ab945e
N
2331 if (max_progress > *suspend_point)
2332 max_progress = *suspend_point;
7443ee81
N
2333 }
2334 } else {
d0ab945e
N
2335 if ((need_backup < info->reshape_progress
2336 || info->array.major_version < 0) &&
7443ee81 2337 *suspend_point > info->reshape_progress - target) {
d0ab945e
N
2338 if (need_backup > *suspend_point - 2 * target)
2339 *suspend_point = need_backup;
2340 else if (*suspend_point >= 2 * target)
7443ee81 2341 *suspend_point -= 2 * target;
d0ab945e
N
2342 else
2343 *suspend_point = 0;
7443ee81 2344 sysfs_set_num(info, NULL, "suspend_lo", *suspend_point);
d0ab945e
N
2345 if (max_progress < *suspend_point)
2346 max_progress = *suspend_point;
7443ee81
N
2347 }
2348 }
2349
2350 /* now set sync_max to allow that progress. sync_max, like
2351 * sync_completed is a count of sectors written per device, so
2352 * we find the difference between max_progress and the start point,
2353 * and divide that by after.data_disks to get a sync_max
2354 * number.
2355 * At the same time we convert wait_point to a similar number
2356 * for comparing against sync_completed.
2357 */
92d1991f 2358 /* scale down max_progress to per_disk */
7443ee81 2359 max_progress /= reshape->after.data_disks;
92d1991f
N
2360 /* Round to chunk size as some kernels give an erroneously high number */
2361 max_progress /= info->new_chunk/512;
2362 max_progress *= info->new_chunk/512;
2363 /* Limit progress to the whole device */
2364 if (max_progress > info->component_size)
2365 max_progress = info->component_size;
7443ee81 2366 wait_point /= reshape->after.data_disks;
92d1991f
N
2367 if (!advancing) {
2368 /* switch from 'device offset' to 'processed block count' */
2369 max_progress = info->component_size - max_progress;
2370 wait_point = info->component_size - wait_point;
2371 }
7443ee81
N
2372
2373 sysfs_set_num(info, NULL, "sync_max", max_progress);
2374
2375 /* Now wait. If we have already reached the point that we were
2376 * asked to wait to, don't wait at all, else wait for any change.
2377 * We need to select on 'sync_completed' as that is the place that
2378 * notifications happen, but we are really interested in
2379 * 'reshape_position'
2380 */
2381 fd = sysfs_get_fd(info, NULL, "sync_completed");
2382 if (fd < 0)
2383 return -1;
2384
2385 if (sysfs_fd_get_ll(fd, &completed) < 0) {
2386 close(fd);
2387 return -1;
2388 }
2389 while (completed < max_progress && completed < wait_point) {
2390 /* Check that sync_action is still 'reshape' to avoid
2391 * waiting forever on a dead array
2392 */
2393 char action[20];
2394 fd_set rfds;
2395 if (sysfs_get_str(info, NULL, "sync_action",
2396 action, 20) <= 0 ||
2397 strncmp(action, "reshape", 7) != 0)
2398 break;
2399 FD_ZERO(&rfds);
2400 FD_SET(fd, &rfds);
2401 select(fd+1, NULL, NULL, &rfds, NULL);
2402 if (sysfs_fd_get_ll(fd, &completed) < 0) {
2403 close(fd);
2404 return -1;
2405 }
2406 }
92d1991f
N
2407 /* some kernels can give an incorrectly high 'completed' number */
2408 completed /= (info->new_chunk/512);
2409 completed *= (info->new_chunk/512);
7443ee81
N
2410 /* Convert 'completed' back in to a 'progress' number */
2411 completed *= reshape->after.data_disks;
2412 if (!advancing) {
2413 completed = info->component_size * reshape->after.data_disks
2414 - completed;
2415 }
2416 *reshape_completed = completed;
2417
2418 close(fd);
2419
2420 /* We return the need_backup flag. Caller will decide
d0ab945e 2421 * how much - a multiple of ->backup_blocks up to *suspend_point
7443ee81 2422 */
d0ab945e
N
2423 return advancing
2424 ? (need_backup > info->reshape_progress)
2425 : (need_backup < info->reshape_progress);
7443ee81
N
2426}
2427
2428
fcf57625 2429/* FIXME return status is never checked */
4411fb17 2430static int grow_backup(struct mdinfo *sra,
7236ee7a 2431 unsigned long long offset, /* per device */
7443ee81 2432 unsigned long stripes, /* per device, in old chunks */
7236ee7a
N
2433 int *sources, unsigned long long *offsets,
2434 int disks, int chunk, int level, int layout,
2435 int dests, int *destfd, unsigned long long *destoffsets,
d4445387 2436 int part, int *degraded,
7236ee7a
N
2437 char *buf)
2438{
2439 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2440 * to storage 'destfd' (offset 'destoffsets'), after first
2441 * suspending IO. Then allow resync to continue
2442 * over the suspended section.
2443 * Use part 'part' of the backup-super-block.
2444 */
2445 int odata = disks;
2446 int rv = 0;
2447 int i;
f21e18ca
N
2448 unsigned long long ll;
2449 int new_degraded;
7236ee7a
N
2450 //printf("offset %llu\n", offset);
2451 if (level >= 4)
2452 odata--;
2453 if (level == 6)
2454 odata--;
7443ee81 2455
d4445387 2456 /* Check that array hasn't become degraded, else we might backup the wrong data */
f21e18ca
N
2457 sysfs_get_ll(sra, NULL, "degraded", &ll);
2458 new_degraded = (int)ll;
d4445387
N
2459 if (new_degraded != *degraded) {
2460 /* check each device to ensure it is still working */
2461 struct mdinfo *sd;
2462 for (sd = sra->devs ; sd ; sd = sd->next) {
2463 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2464 continue;
2465 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
2466 char sbuf[20];
2467 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
2468 strstr(sbuf, "faulty") ||
2469 strstr(sbuf, "in_sync") == NULL) {
2470 /* this device is dead */
2471 sd->disk.state = (1<<MD_DISK_FAULTY);
2472 if (sd->disk.raid_disk >= 0 &&
2473 sources[sd->disk.raid_disk] >= 0) {
2474 close(sources[sd->disk.raid_disk]);
2475 sources[sd->disk.raid_disk] = -1;
2476 }
2477 }
2478 }
2479 }
2480 *degraded = new_degraded;
2481 }
7236ee7a
N
2482 if (part) {
2483 bsb.arraystart2 = __cpu_to_le64(offset * odata);
200871ad 2484 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
7236ee7a
N
2485 } else {
2486 bsb.arraystart = __cpu_to_le64(offset * odata);
200871ad 2487 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
7236ee7a
N
2488 }
2489 if (part)
2490 bsb.magic[15] = '2';
2491 for (i = 0; i < dests; i++)
2492 if (part)
2493 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
2494 else
2495 lseek64(destfd[i], destoffsets[i], 0);
2496
2497 rv = save_stripes(sources, offsets,
2498 disks, chunk, level, layout,
2499 dests, destfd,
2500 offset*512*odata, stripes * chunk * odata,
2501 buf);
2502
2503 if (rv)
2504 return rv;
97396422 2505 bsb.mtime = __cpu_to_le64(time(0));
7236ee7a
N
2506 for (i = 0; i < dests; i++) {
2507 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
2508
2509 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
2510 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
2511 bsb.sb_csum2 = bsb_csum((char*)&bsb,
2512 ((char*)&bsb.sb_csum2)-((char*)&bsb));
2513
72044953 2514 rv = -1;
f21e18ca
N
2515 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
2516 != destoffsets[i] - 4096)
72044953
N
2517 break;
2518 if (write(destfd[i], &bsb, 512) != 512)
2519 break;
ff94fb86 2520 if (destoffsets[i] > 4096) {
f21e18ca 2521 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
a847575a 2522 destoffsets[i]+stripes*chunk*odata)
72044953
N
2523 break;
2524 if (write(destfd[i], &bsb, 512) != 512)
2525 break;
ff94fb86 2526 }
7236ee7a 2527 fsync(destfd[i]);
72044953 2528 rv = 0;
7236ee7a 2529 }
2efedc7b 2530
fcf57625 2531 return rv;
7236ee7a
N
2532}
2533
2534/* in 2.6.30, the value reported by sync_completed can be
2535 * less that it should be by one stripe.
2536 * This only happens when reshape hits sync_max and pauses.
2537 * So allow wait_backup to either extent sync_max further
2538 * than strictly necessary, or return before the
2539 * sync has got quite as far as we would really like.
2540 * This is what 'blocks2' is for.
2541 * The various caller give appropriate values so that
2542 * every works.
2543 */
fcf57625 2544/* FIXME return value is often ignored */
7443ee81 2545static int forget_backup(
7236ee7a
N
2546 int dests, int *destfd, unsigned long long *destoffsets,
2547 int part)
2548{
7443ee81
N
2549 /*
2550 * Erase backup 'part' (which is 0 or 1)
7236ee7a 2551 */
7236ee7a 2552 int i;
fcf57625 2553 int rv;
7236ee7a 2554
7236ee7a
N
2555 if (part) {
2556 bsb.arraystart2 = __cpu_to_le64(0);
2557 bsb.length2 = __cpu_to_le64(0);
2558 } else {
2559 bsb.arraystart = __cpu_to_le64(0);
2560 bsb.length = __cpu_to_le64(0);
2561 }
97396422 2562 bsb.mtime = __cpu_to_le64(time(0));
fcf57625 2563 rv = 0;
7236ee7a
N
2564 for (i = 0; i < dests; i++) {
2565 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
2566 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
2567 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
2568 bsb.sb_csum2 = bsb_csum((char*)&bsb,
2569 ((char*)&bsb.sb_csum2)-((char*)&bsb));
f21e18ca 2570 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
a847575a 2571 destoffsets[i]-4096)
72044953
N
2572 rv = -1;
2573 if (rv == 0 &&
2574 write(destfd[i], &bsb, 512) != 512)
2575 rv = -1;
7236ee7a
N
2576 fsync(destfd[i]);
2577 }
fcf57625 2578 return rv;
7236ee7a 2579}
e86c9dd6 2580
7236ee7a
N
2581static void fail(char *msg)
2582{
fcf57625 2583 int rv;
72044953
N
2584 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
2585 rv |= (write(2, "\n", 1) != 1);
fcf57625 2586 exit(rv ? 1 : 2);
7236ee7a
N
2587}
2588
2589static char *abuf, *bbuf;
f21e18ca 2590static unsigned long long abuflen;
7236ee7a
N
2591static void validate(int afd, int bfd, unsigned long long offset)
2592{
2593 /* check that the data in the backup against the array.
2594 * This is only used for regression testing and should not
2595 * be used while the array is active
2596 */
7236ee7a
N
2597 if (afd < 0)
2598 return;
2599 lseek64(bfd, offset - 4096, 0);
2600 if (read(bfd, &bsb2, 512) != 512)
2601 fail("cannot read bsb");
2602 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
2603 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
2604 fail("first csum bad");
2605 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
2606 fail("magic is bad");
2607 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
2608 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
2609 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
2610 fail("second csum bad");
2611
2612 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
2613 fail("devstart is wrong");
2614
2615 if (bsb2.length) {
2616 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
2617
2618 if (abuflen < len) {
2619 free(abuf);
2620 free(bbuf);
2621 abuflen = len;
fcf57625
N
2622 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
2623 posix_memalign((void**)&bbuf, 4096, abuflen)) {
2624 abuflen = 0;
2625 /* just stop validating on mem-alloc failure */
2626 return;
2627 }
e86c9dd6 2628 }
48924014 2629
7236ee7a 2630 lseek64(bfd, offset, 0);
f21e18ca 2631 if ((unsigned long long)read(bfd, bbuf, len) != len) {
080fd005 2632 //printf("len %llu\n", len);
7236ee7a
N
2633 fail("read first backup failed");
2634 }
2635 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
f21e18ca 2636 if ((unsigned long long)read(afd, abuf, len) != len)
7236ee7a
N
2637 fail("read first from array failed");
2638 if (memcmp(bbuf, abuf, len) != 0) {
080fd005 2639 #if 0
7236ee7a
N
2640 int i;
2641 printf("offset=%llu len=%llu\n",
080fd005 2642 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
7236ee7a
N
2643 for (i=0; i<len; i++)
2644 if (bbuf[i] != abuf[i]) {
2645 printf("first diff byte %d\n", i);
93ecfa01 2646 break;
7236ee7a 2647 }
080fd005 2648 #endif
7236ee7a 2649 fail("data1 compare failed");
e86c9dd6 2650 }
7236ee7a
N
2651 }
2652 if (bsb2.length2) {
2653 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
2654
2655 if (abuflen < len) {
2656 free(abuf);
2657 free(bbuf);
2658 abuflen = len;
2659 abuf = malloc(abuflen);
2660 bbuf = malloc(abuflen);
e86c9dd6
NB
2661 }
2662
7236ee7a 2663 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
f21e18ca 2664 if ((unsigned long long)read(bfd, bbuf, len) != len)
7236ee7a
N
2665 fail("read second backup failed");
2666 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
f21e18ca 2667 if ((unsigned long long)read(afd, abuf, len) != len)
7236ee7a
N
2668 fail("read second from array failed");
2669 if (memcmp(bbuf, abuf, len) != 0)
2670 fail("data2 compare failed");
e86c9dd6 2671 }
7236ee7a 2672}
e86c9dd6 2673
999b4972
N
2674int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
2675 struct supertype *st, unsigned long blocks,
2676 int *fds, unsigned long long *offsets,
2677 int dests, int *destfd, unsigned long long *destoffsets)
7236ee7a 2678{
7443ee81
N
2679 /* Monitor a reshape where backup is being performed using
2680 * 'native' mechanism - either to a backup file, or
2681 * to some space in a spare.
2682 */
7236ee7a 2683 char *buf;
7443ee81
N
2684 int degraded = -1;
2685 unsigned long long speed;
2686 unsigned long long suspend_point, array_size;
2687 unsigned long long backup_point, wait_point;
2688 unsigned long long reshape_completed;
2689 int done = 0;
2690 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
2691 int part = 0; /* The next part of the backup area to fill. It may already
2692 * be full, so we need to check */
2693 int level = reshape->level;
2694 int layout = reshape->before.layout;
2695 int data = reshape->before.data_disks;
2696 int disks = reshape->before.data_disks + reshape->parity;
2697 int chunk = sra->array.chunk_size;
da8100ca
N
2698 struct mdinfo *sd;
2699 unsigned long stripes;
2700
2701 /* set up the backup-super-block. This requires the
2702 * uuid from the array.
2703 */
2704 /* Find a superblock */
2705 for (sd = sra->devs; sd; sd = sd->next) {
2706 char *dn;
2707 int devfd;
2708 int ok;
2709 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2710 continue;
2711 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
2712 devfd = dev_open(dn, O_RDONLY);
2713 if (devfd < 0)
2714 continue;
2715 ok = st->ss->load_super(st, devfd, NULL);
2716 close(devfd);
2717 if (ok >= 0)
2718 break;
2719 }
2720 if (!sd) {
2721 fprintf(stderr, Name ": Cannot find a superblock\n");
2722 return 0;
2723 }
2724
2725 memset(&bsb, 0, 512);
2726 memcpy(bsb.magic, "md_backup_data-1", 16);
2727 st->ss->uuid_from_super(st, (int*)&bsb.set_uuid);
2728 bsb.mtime = __cpu_to_le64(time(0));
2729 bsb.devstart2 = blocks;
2730
2731 stripes = blocks / (sra->array.chunk_size/512) /
2732 reshape->before.data_disks;
e86c9dd6 2733
fcf57625
N
2734 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2735 /* Don't start the 'reshape' */
2736 return 0;
7443ee81
N
2737 if (reshape->before.data_disks == reshape->after.data_disks) {
2738 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
2739 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
2740 }
e86c9dd6 2741
7443ee81 2742 if (increasing) {
96533072 2743 array_size = sra->component_size * reshape->after.data_disks;
7443ee81
N
2744 backup_point = sra->reshape_progress;
2745 suspend_point = 0;
2746 } else {
96533072 2747 array_size = sra->component_size * reshape->before.data_disks;
7443ee81
N
2748 backup_point = array_size;
2749 suspend_point = array_size;
2750 }
7236ee7a 2751
7443ee81
N
2752 while (!done) {
2753 int rv;
7236ee7a 2754
7443ee81
N
2755 /* Want to return as soon the oldest backup slot can
2756 * be released as that allows us to start backing up
2757 * some more, providing suspend_point has been
b6b95155 2758 * advanced, which it should have.
7443ee81
N
2759 */
2760 if (increasing) {
2761 wait_point = array_size;
2762 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
2763 wait_point = (__le64_to_cpu(bsb.arraystart) +
2764 __le64_to_cpu(bsb.length));
2765 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
2766 wait_point = (__le64_to_cpu(bsb.arraystart2) +
2767 __le64_to_cpu(bsb.length2));
2768 } else {
2769 wait_point = 0;
2770 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
2771 wait_point = __le64_to_cpu(bsb.arraystart);
2772 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
2773 wait_point = __le64_to_cpu(bsb.arraystart2);
2774 }
2775
2776 rv = progress_reshape(sra, reshape,
2777 backup_point, wait_point,
2778 &suspend_point, &reshape_completed);
7443ee81
N
2779 /* external metadata would need to ping_monitor here */
2780 sra->reshape_progress = reshape_completed;
7236ee7a 2781
7443ee81
N
2782 /* Clear any backup region that is before 'here' */
2783 if (increasing) {
2784 if (reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
2785 __le64_to_cpu(bsb.length)))
2786 forget_backup(dests, destfd,
2787 destoffsets, 0);
2788 if (reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
2789 __le64_to_cpu(bsb.length2)))
2790 forget_backup(dests, destfd,
2791 destoffsets, 1);
2792 } else {
2793 if (reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
2794 forget_backup(dests, destfd,
2795 destoffsets, 0);
2796 if (reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
2797 forget_backup(dests, destfd,
2798 destoffsets, 1);
2799 }
7236ee7a 2800
223c5c99
N
2801 if (rv < 0) {
2802 done = 1;
2803 break;
2804 }
2805
60204e4e 2806 while (rv) {
7443ee81 2807 unsigned long long offset;
e97ca002 2808 unsigned long actual_stripes;
60204e4e
N
2809 /* Need to backup some data.
2810 * If 'part' is not used and the desired
2811 * backup size is suspended, do a backup,
2812 * then consider the next part.
2813 */
7443ee81
N
2814 /* Check that 'part' is unused */
2815 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
60204e4e 2816 break;
7443ee81 2817 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
60204e4e 2818 break;
7443ee81
N
2819
2820 offset = backup_point / data;
e97ca002 2821 actual_stripes = stripes;
60204e4e 2822 if (increasing) {
e97ca002
N
2823 if (offset + actual_stripes * (chunk/512) >
2824 sra->component_size)
2825 actual_stripes = ((sra->component_size - offset)
2826 / (chunk/512));
2827 if (offset + actual_stripes * (chunk/512) >
60204e4e
N
2828 suspend_point/data)
2829 break;
2830 } else {
e97ca002
N
2831 if (offset < actual_stripes * (chunk/512))
2832 actual_stripes = offset / (chunk/512);
2833 offset -= actual_stripes * (chunk/512);
2834 if (offset < suspend_point/data)
60204e4e
N
2835 break;
2836 }
e97ca002 2837 grow_backup(sra, offset, actual_stripes,
7443ee81
N
2838 fds, offsets,
2839 disks, chunk, level, layout,
2840 dests, destfd, destoffsets,
2841 part, &degraded, buf);
2842 validate(afd, destfd[0], destoffsets[0]);
2843 /* record where 'part' is up to */
2844 part = !part;
2845 if (increasing)
e97ca002 2846 backup_point += actual_stripes * (chunk/512) * data;
7443ee81 2847 else
e97ca002 2848 backup_point -= actual_stripes * (chunk/512) * data;
7443ee81
N
2849 }
2850 }
2851
223c5c99
N
2852 /* FIXME maybe call progress_reshape one more time instead */
2853 abort_reshape(sra); /* remove any remaining suspension */
7443ee81
N
2854 if (reshape->before.data_disks == reshape->after.data_disks)
2855 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
7236ee7a 2856 free(buf);
7443ee81 2857 return 1; /* FIXME what does this mean? */
e86c9dd6 2858}
353632d9
NB
2859
2860/*
2861 * If any spare contains md_back_data-1 which is recent wrt mtime,
2862 * write that data into the array and update the super blocks with
2863 * the new reshape_progress
2864 */
ea0ebe96
N
2865int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
2866 char *backup_file, int verbose)
353632d9
NB
2867{
2868 int i, j;
2869 int old_disks;
353632d9 2870 unsigned long long *offsets;
82f2d6ab 2871 unsigned long long nstripe, ostripe;
6e9eac4f 2872 int ndata, odata;
353632d9 2873
e9e43ec3
N
2874 if (info->new_level != info->array.level)
2875 return 1; /* Cannot handle level changes (they are instantaneous) */
2876
2877 odata = info->array.raid_disks - info->delta_disks - 1;
2878 if (info->array.level == 6) odata--; /* number of data disks */
2879 ndata = info->array.raid_disks - 1;
2880 if (info->new_level == 6) ndata--;
353632d9
NB
2881
2882 old_disks = info->array.raid_disks - info->delta_disks;
2883
e9e43ec3
N
2884 if (info->delta_disks <= 0)
2885 /* Didn't grow, so the backup file must have
2886 * been used
2887 */
2888 old_disks = cnt;
06b0d786 2889 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
353632d9 2890 struct mdinfo dinfo;
06b0d786 2891 int fd;
e9e43ec3 2892 int bsbsize;
ea0ebe96 2893 char *devname, namebuf[20];
353632d9
NB
2894
2895 /* This was a spare and may have some saved data on it.
2896 * Load the superblock, find and load the
2897 * backup_super_block.
2898 * If either fail, go on to next device.
2899 * If the backup contains no new info, just return
206c5eae 2900 * else restore data and update all superblocks
353632d9 2901 */
06b0d786
NB
2902 if (i == old_disks-1) {
2903 fd = open(backup_file, O_RDONLY);
e9e43ec3
N
2904 if (fd<0) {
2905 fprintf(stderr, Name ": backup file %s inaccessible: %s\n",
2906 backup_file, strerror(errno));
06b0d786 2907 continue;
e9e43ec3 2908 }
ea0ebe96 2909 devname = backup_file;
06b0d786
NB
2910 } else {
2911 fd = fdlist[i];
2912 if (fd < 0)
2913 continue;
3da92f27 2914 if (st->ss->load_super(st, fd, NULL))
06b0d786 2915 continue;
353632d9 2916
a5d85af7 2917 st->ss->getinfo_super(st, &dinfo, NULL);
3da92f27
NB
2918 st->ss->free_super(st);
2919
06b0d786
NB
2920 if (lseek64(fd,
2921 (dinfo.data_offset + dinfo.component_size - 8) <<9,
ea0ebe96
N
2922 0) < 0) {
2923 fprintf(stderr, Name ": Cannot seek on device %d\n", i);
06b0d786 2924 continue; /* Cannot seek */
ea0ebe96
N
2925 }
2926 sprintf(namebuf, "device-%d", i);
2927 devname = namebuf;
06b0d786 2928 }
ea0ebe96
N
2929 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
2930 if (verbose)
2931 fprintf(stderr, Name ": Cannot read from %s\n", devname);
353632d9 2932 continue; /* Cannot read */
ea0ebe96 2933 }
e9e43ec3 2934 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
ea0ebe96
N
2935 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
2936 if (verbose)
2937 fprintf(stderr, Name ": No backup metadata on %s\n", devname);
353632d9 2938 continue;
ea0ebe96
N
2939 }
2940 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
2941 if (verbose)
2942 fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname);
353632d9 2943 continue; /* bad checksum */
ea0ebe96 2944 }
e9e43ec3 2945 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
ea0ebe96
N
2946 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
2947 if (verbose)
2948 fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname);
22e30516 2949 continue; /* Bad second checksum */
ea0ebe96
N
2950 }
2951 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
2952 if (verbose)
2953 fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname);
353632d9 2954 continue; /* Wrong uuid */
ea0ebe96 2955 }
353632d9 2956
097075b6
N
2957 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2958 * sometimes they aren't... So allow considerable flexability in matching, and allow
2959 * this test to be overridden by an environment variable.
2960 */
f21e18ca
N
2961 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
2962 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
097075b6
N
2963 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2964 fprintf(stderr, Name ": accepting backup with timestamp %lu "
2965 "for array with timestamp %lu\n",
2966 (unsigned long)__le64_to_cpu(bsb.mtime),
2967 (unsigned long)info->array.utime);
2968 } else {
2969 if (verbose)
2970 fprintf(stderr, Name ": too-old timestamp on "
2971 "backup-metadata on %s\n", devname);
2972 continue; /* time stamp is too bad */
2973 }
ea0ebe96 2974 }
353632d9 2975
e9e43ec3
N
2976 if (bsb.magic[15] == '1') {
2977 if (info->delta_disks >= 0) {
2978 /* reshape_progress is increasing */
2979 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
ea0ebe96
N
2980 info->reshape_progress) {
2981 nonew:
2982 if (verbose)
2983 fprintf(stderr, Name ": backup-metadata found on %s but is not needed\n", devname);
e9e43ec3 2984 continue; /* No new data here */
ea0ebe96 2985 }
e9e43ec3
N
2986 } else {
2987 /* reshape_progress is decreasing */
2988 if (__le64_to_cpu(bsb.arraystart) >=
2989 info->reshape_progress)
ea0ebe96 2990 goto nonew; /* No new data here */
e9e43ec3
N
2991 }
2992 } else {
2993 if (info->delta_disks >= 0) {
2994 /* reshape_progress is increasing */
2995 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2996 info->reshape_progress &&
2997 __le64_to_cpu(bsb.arraystart2) + __le64_to_cpu(bsb.length2) <
2998 info->reshape_progress)
ea0ebe96 2999 goto nonew; /* No new data here */
e9e43ec3
N
3000 } else {
3001 /* reshape_progress is decreasing */
3002 if (__le64_to_cpu(bsb.arraystart) >=
3003 info->reshape_progress &&
3004 __le64_to_cpu(bsb.arraystart2) >=
3005 info->reshape_progress)
ea0ebe96 3006 goto nonew; /* No new data here */
e9e43ec3
N
3007 }
3008 }
ea0ebe96
N
3009 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
3010 second_fail:
3011 if (verbose)
3012 fprintf(stderr, Name ": Failed to verify secondary backup-metadata block on %s\n",
3013 devname);
353632d9 3014 continue; /* Cannot seek */
ea0ebe96 3015 }
2efedc7b 3016 /* There should be a duplicate backup superblock 4k before here */
06b0d786 3017 if (lseek64(fd, -4096, 1) < 0 ||
0155af90 3018 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
ea0ebe96 3019 goto second_fail; /* Cannot find leading superblock */
e9e43ec3
N
3020 if (bsb.magic[15] == '1')
3021 bsbsize = offsetof(struct mdp_backup_super, pad1);
3022 else
3023 bsbsize = offsetof(struct mdp_backup_super, pad);
ff94fb86 3024 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
ea0ebe96 3025 goto second_fail; /* Cannot find leading superblock */
2efedc7b 3026
353632d9
NB
3027 /* Now need the data offsets for all devices. */
3028 offsets = malloc(sizeof(*offsets)*info->array.raid_disks);
3029 for(j=0; j<info->array.raid_disks; j++) {
3030 if (fdlist[j] < 0)
3031 continue;
3da92f27 3032 if (st->ss->load_super(st, fdlist[j], NULL))
353632d9
NB
3033 /* FIXME should be this be an error */
3034 continue;
a5d85af7 3035 st->ss->getinfo_super(st, &dinfo, NULL);
3da92f27 3036 st->ss->free_super(st);
14e5b4d7 3037 offsets[j] = dinfo.data_offset * 512;
353632d9
NB
3038 }
3039 printf(Name ": restoring critical section\n");
3040
3041 if (restore_stripes(fdlist, offsets,
3042 info->array.raid_disks,
3043 info->new_chunk,
3044 info->new_level,
3045 info->new_layout,
06b0d786 3046 fd, __le64_to_cpu(bsb.devstart)*512,
92dcdf7c 3047 __le64_to_cpu(bsb.arraystart)*512,
e9e43ec3
N
3048 __le64_to_cpu(bsb.length)*512)) {
3049 /* didn't succeed, so giveup */
ea0ebe96
N
3050 if (verbose)
3051 fprintf(stderr, Name ": Error restoring backup from %s\n",
3052 devname);
e9e43ec3
N
3053 return 1;
3054 }
3055
3056 if (bsb.magic[15] == '2' &&
3057 restore_stripes(fdlist, offsets,
3058 info->array.raid_disks,
3059 info->new_chunk,
3060 info->new_level,
3061 info->new_layout,
3062 fd, __le64_to_cpu(bsb.devstart)*512 +
3063 __le64_to_cpu(bsb.devstart2)*512,
92dcdf7c 3064 __le64_to_cpu(bsb.arraystart2)*512,
e9e43ec3 3065 __le64_to_cpu(bsb.length2)*512)) {
353632d9 3066 /* didn't succeed, so giveup */
ea0ebe96
N
3067 if (verbose)
3068 fprintf(stderr, Name ": Error restoring second backup from %s\n",
3069 devname);
2295250a 3070 return 1;
353632d9
NB
3071 }
3072
e9e43ec3 3073
353632d9
NB
3074 /* Ok, so the data is restored. Let's update those superblocks. */
3075
e9e43ec3
N
3076 if (info->delta_disks >= 0) {
3077 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
3078 __le64_to_cpu(bsb.length);
3079 if (bsb.magic[15] == '2') {
3080 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
3081 __le64_to_cpu(bsb.length2);
3082 if (p2 > info->reshape_progress)
3083 info->reshape_progress = p2;
3084 }
3085 } else {
3086 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
3087 if (bsb.magic[15] == '2') {
3088 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
3089 if (p2 < info->reshape_progress)
3090 info->reshape_progress = p2;
3091 }
3092 }
353632d9
NB
3093 for (j=0; j<info->array.raid_disks; j++) {
3094 if (fdlist[j] < 0) continue;
3da92f27 3095 if (st->ss->load_super(st, fdlist[j], NULL))
353632d9 3096 continue;
a5d85af7 3097 st->ss->getinfo_super(st, &dinfo, NULL);
e9e43ec3 3098 dinfo.reshape_progress = info->reshape_progress;
3da92f27 3099 st->ss->update_super(st, &dinfo,
68c7d6d7
NB
3100 "_reshape_progress",
3101 NULL,0, 0, NULL);
3da92f27
NB
3102 st->ss->store_super(st, fdlist[j]);
3103 st->ss->free_super(st);
353632d9 3104 }
353632d9
NB
3105 return 0;
3106 }
6e9eac4f
NB
3107 /* Didn't find any backup data, try to see if any
3108 * was needed.
3109 */
82f2d6ab
N
3110 if (info->delta_disks < 0) {
3111 /* When shrinking, the critical section is at the end.
3112 * So see if we are before the critical section.
3113 */
3114 unsigned long long first_block;
3115 nstripe = ostripe = 0;
3116 first_block = 0;
3117 while (ostripe >= nstripe) {
3118 ostripe += info->array.chunk_size / 512;
3119 first_block = ostripe * odata;
3120 nstripe = first_block / ndata / (info->new_chunk/512) *
3121 (info->new_chunk/512);
3122 }
3123
3124 if (info->reshape_progress >= first_block)
3125 return 0;
6e9eac4f 3126 }
82f2d6ab
N
3127 if (info->delta_disks > 0) {
3128 /* See if we are beyond the critical section. */
3129 unsigned long long last_block;
3130 nstripe = ostripe = 0;
3131 last_block = 0;
3132 while (nstripe >= ostripe) {
3133 nstripe += info->new_chunk / 512;
3134 last_block = nstripe * ndata;
3135 ostripe = last_block / odata / (info->array.chunk_size/512) *
3136 (info->array.chunk_size/512);
3137 }
6e9eac4f 3138
82f2d6ab
N
3139 if (info->reshape_progress >= last_block)
3140 return 0;
3141 }
6e9eac4f 3142 /* needed to recover critical section! */
ea0ebe96
N
3143 if (verbose)
3144 fprintf(stderr, Name ": Failed to find backup of critical section\n");
2295250a 3145 return 1;
353632d9 3146}
e9e43ec3
N
3147
3148int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
3149 char *backup_file)
3150{
7443ee81 3151 int err = sysfs_set_str(info, NULL, "array_state", "readonly");
e9e43ec3
N
3152 if (err)
3153 return err;
7443ee81 3154 return reshape_array(NULL, mdfd, "array", st, info, 1, backup_file, 0, 0);
e9e43ec3
N
3155}
3156
3157