]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Assemble: allow an array undergoing reshape to be started without backup file
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
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
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
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
34 #ifndef offsetof
35 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
36 #endif
37
38 int Grow_Add_device(char *devname, int fd, char *newdev)
39 {
40 /* Add a device to an active array.
41 * Currently, just extend a linear array.
42 * This requires writing a new superblock on the
43 * new device, calling the kernel to add the device,
44 * and if that succeeds, update the superblock on
45 * all other devices.
46 * This means that we need to *find* all other devices.
47 */
48 struct mdinfo info;
49
50 struct stat stb;
51 int nfd, fd2;
52 int d, nd;
53 struct supertype *st = NULL;
54 char *subarray = NULL;
55
56 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
57 fprintf(stderr, Name ": cannot get array info for %s\n", devname);
58 return 1;
59 }
60
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);
67 if (!st) {
68 fprintf(stderr, Name ": cannot handle arrays with superblock version %d\n", info.array.major_version);
69 return 1;
70 }
71
72 if (subarray) {
73 fprintf(stderr, Name ": Cannot grow linear sub-arrays yet\n");
74 free(subarray);
75 free(st);
76 }
77
78 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
79 if (nfd < 0) {
80 fprintf(stderr, Name ": cannot open %s\n", newdev);
81 free(st);
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);
88 free(st);
89 return 1;
90 }
91 /* now check out all the devices and make sure we can read the superblock */
92 for (d=0 ; d < info.array.raid_disks ; d++) {
93 mdu_disk_info_t disk;
94 char *dv;
95
96 st->ss->free_super(st);
97
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);
102 close(nfd);
103 free(st);
104 return 1;
105 }
106 dv = map_dev(disk.major, disk.minor, 1);
107 if (!dv) {
108 fprintf(stderr, Name ": cannot find device file for device %d\n",
109 d);
110 close(nfd);
111 free(st);
112 return 1;
113 }
114 fd2 = dev_open(dv, O_RDWR);
115 if (!fd2) {
116 fprintf(stderr, Name ": cannot open device file %s\n", dv);
117 close(nfd);
118 free(st);
119 return 1;
120 }
121
122 if (st->ss->load_super(st, fd2, NULL)) {
123 fprintf(stderr, Name ": cannot find super block on %s\n", dv);
124 close(nfd);
125 close(fd2);
126 free(st);
127 return 1;
128 }
129 close(fd2);
130 }
131 /* Ok, looks good. Lets update the superblock and write it out to
132 * newdev.
133 */
134
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);
140 st->ss->update_super(st, &info, "linear-grow-new", newdev,
141 0, 0, NULL);
142
143 if (st->ss->store_super(st, nfd)) {
144 fprintf(stderr, Name ": Cannot store new superblock on %s\n",
145 newdev);
146 close(nfd);
147 return 1;
148 }
149 close(nfd);
150
151 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
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
159 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
160 fprintf(stderr, Name ": cannot get array info for %s\n", devname);
161 return 1;
162 }
163
164 nd = d;
165 for (d=0 ; d < info.array.raid_disks ; d++) {
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 }
175 dv = map_dev(disk.major, disk.minor, 1);
176 if (!dv) {
177 fprintf(stderr, Name ": cannot find device file for device %d\n",
178 d);
179 return 1;
180 }
181 fd2 = dev_open(dv, O_RDWR);
182 if (fd2 < 0) {
183 fprintf(stderr, Name ": cannot open device file %s\n", dv);
184 return 1;
185 }
186 if (st->ss->load_super(st, fd2, NULL)) {
187 fprintf(stderr, Name ": cannot find super block on %s\n", dv);
188 close(fd);
189 return 1;
190 }
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;
195
196 st->ss->update_super(st, &info, "linear-grow-update", dv,
197 0, 0, NULL);
198
199 if (st->ss->store_super(st, fd2)) {
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 }
209
210 int Grow_addbitmap(char *devname, int fd, char *file, int chunk, int delay, int write_behind, int force)
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;
224 char *subarray = NULL;
225 int major = BITMAP_MAJOR_HI;
226 int vers = md_get_version(fd);
227 unsigned long long bitmapsize, array_size;
228
229 if (vers < 9003) {
230 major = BITMAP_MAJOR_HOSTENDIAN;
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");
235 }
236
237 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
238 if (errno == ENOMEM)
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]) {
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 }
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)) {
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 }
270 fprintf(stderr, Name ": Internal bitmap already present on %s\n",
271 devname);
272 return 1;
273 }
274
275 if (strcmp(file, "none") == 0) {
276 fprintf(stderr, Name ": no bitmap found on %s\n", devname);
277 return 1;
278 }
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 }
284 bitmapsize = array.size;
285 bitmapsize <<= 1;
286 if (get_dev_size(fd, NULL, &array_size) &&
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 }
293 if (bitmapsize == 0) {
294 fprintf(stderr, Name ": Cannot reliably determine size of array to create bitmap - sorry.\n");
295 return 1;
296 }
297
298 if (array.level == 10) {
299 int ncopies = (array.layout&255)*((array.layout>>8)&255);
300 bitmapsize = bitmapsize * array.raid_disks / ncopies;
301 }
302
303 st = super_by_fd(fd, &subarray);
304 if (!st) {
305 fprintf(stderr, Name ": Cannot understand version %d.%d\n",
306 array.major_version, array.minor_version);
307 return 1;
308 }
309 if (subarray) {
310 fprintf(stderr, Name ": Cannot add bitmaps to sub-arrays yet\n");
311 free(subarray);
312 free(st);
313 return 1;
314 }
315 if (strcmp(file, "internal") == 0) {
316 int d;
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 }
322 for (d=0; d< st->max_devs; d++) {
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;
333 dv = map_dev(disk.major, disk.minor, 1);
334 if (dv) {
335 int fd2 = dev_open(dv, O_RDWR);
336 if (fd2 < 0)
337 continue;
338 if (st->ss->load_super(st, fd2, NULL)==0) {
339 if (st->ss->add_internal_bitmap(
340 st,
341 &chunk, delay, write_behind,
342 bitmapsize, 0, major)
343 )
344 st->ss->write_bitmap(st, fd2);
345 else {
346 fprintf(stderr, Name ": failed to create internal bitmap - chunksize problem.\n");
347 close(fd2);
348 return 1;
349 }
350 }
351 close(fd2);
352 }
353 }
354 array.state |= (1<<MD_SB_BITMAP_PRESENT);
355 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
356 if (errno == EBUSY)
357 fprintf(stderr, Name
358 ": Cannot add bitmap while array is"
359 " resyncing or reshaping etc.\n");
360 fprintf(stderr, Name ": failed to set internal bitmap.\n");
361 return 1;
362 }
363 } else {
364 int uuid[4];
365 int bitmap_fd;
366 int d;
367 int max_devs = st->max_devs;
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;
380 dv = map_dev(disk.major, disk.minor, 1);
381 if (!dv) continue;
382 fd2 = dev_open(dv, O_RDONLY);
383 if (fd2 >= 0 &&
384 st->ss->load_super(st, fd2, NULL) == 0) {
385 close(fd2);
386 st->ss->uuid_from_super(st, uuid);
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 }
395 if (CreateBitmap(file, force, (char*)uuid, chunk,
396 delay, write_behind, bitmapsize, major)) {
397 return 1;
398 }
399 bitmap_fd = open(file, O_RDWR);
400 if (bitmap_fd < 0) {
401 fprintf(stderr, Name ": weird: %s cannot be opened\n",
402 file);
403 return 1;
404 }
405 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
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");
411 fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
412 devname, strerror(err));
413 return 1;
414 }
415 }
416
417 return 0;
418 }
419
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.
424 * The superblock goes 4K from the end of the used space on the
425 * device.
426 * It if written after the backup is complete.
427 * It has the following structure.
428 */
429
430 static struct mdp_backup_super {
431 char magic[16]; /* md_backup_data-1 or -2 */
432 __u8 set_uuid[16];
433 __u64 mtime;
434 /* start/sizes in 512byte sectors */
435 __u64 devstart; /* address on backup device/file of data */
436 __u64 arraystart;
437 __u64 length;
438 __u32 sb_csum; /* csum of preceeding bytes. */
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];
445 } __attribute__((aligned(512))) bsb, bsb2;
446
447 static __u32 bsb_csum(char *buf, int len)
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
456 static int child_grow(int afd, struct mdinfo *sra, unsigned long blocks,
457 int *fds, unsigned long long *offsets,
458 int disks, int chunk, int level, int layout, int data,
459 int dests, int *destfd, unsigned long long *destoffsets);
460 static int child_shrink(int afd, struct mdinfo *sra, unsigned long blocks,
461 int *fds, unsigned long long *offsets,
462 int disks, int chunk, int level, int layout, int data,
463 int dests, int *destfd, unsigned long long *destoffsets);
464 static int child_same_size(int afd, struct mdinfo *sra, unsigned long blocks,
465 int *fds, unsigned long long *offsets,
466 unsigned long long start,
467 int disks, int chunk, int level, int layout, int data,
468 int dests, int *destfd, unsigned long long *destoffsets);
469
470 static int freeze_container(struct supertype *st)
471 {
472 int container_dev = (st->container_dev != NoMdDev
473 ? st->container_dev : st->devnum);
474 char *container = devnum2devname(container_dev);
475
476 if (!container) {
477 fprintf(stderr, Name
478 ": could not determine container name, freeze aborted\n");
479 return -2;
480 }
481
482 if (block_monitor(container, 1)) {
483 fprintf(stderr, Name ": failed to freeze container\n");
484 return -2;
485 }
486
487 return 1;
488 }
489
490 static void unfreeze_container(struct supertype *st)
491 {
492 int container_dev = (st->container_dev != NoMdDev
493 ? st->container_dev : st->devnum);
494 char *container = devnum2devname(container_dev);
495
496 if (!container) {
497 fprintf(stderr, Name
498 ": could not determine container name, unfreeze aborted\n");
499 return;
500 }
501
502 unblock_monitor(container, 1);
503 }
504
505 static int freeze(struct supertype *st)
506 {
507 /* Try to freeze resync/rebuild on this array/container.
508 * Return -1 if the array is busy,
509 * return -2 container cannot be frozen,
510 * return 0 if this kernel doesn't support 'frozen'
511 * return 1 if it worked.
512 */
513 if (st->ss->external)
514 return freeze_container(st);
515 else {
516 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
517 int err;
518
519 if (!sra)
520 return -1;
521 err = sysfs_freeze_array(sra);
522 sysfs_free(sra);
523 return err;
524 }
525 }
526
527 static void unfreeze(struct supertype *st, int frozen)
528 {
529 /* If 'frozen' is 1, unfreeze the array */
530 if (frozen <= 0)
531 return;
532
533 if (st->ss->external)
534 return unfreeze_container(st);
535 else {
536 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
537
538 if (sra)
539 sysfs_set_str(sra, NULL, "sync_action", "idle");
540 else
541 fprintf(stderr, Name ": failed to unfreeze array\n");
542 sysfs_free(sra);
543 }
544 }
545
546 static void wait_reshape(struct mdinfo *sra)
547 {
548 int fd = sysfs_get_fd(sra, NULL, "sync_action");
549 char action[20];
550
551 do {
552 fd_set rfds;
553 FD_ZERO(&rfds);
554 FD_SET(fd, &rfds);
555 select(fd+1, NULL, NULL, &rfds, NULL);
556
557 if (sysfs_fd_get_str(fd, action, 20) < 0) {
558 close(fd);
559 return;
560 }
561 } while (strncmp(action, "reshape", 7) == 0);
562 }
563
564 static int reshape_super(struct supertype *st, long long size, int level,
565 int layout, int chunksize, int raid_disks,
566 char *backup_file, char *dev, int verbose)
567 {
568 /* nothing extra to check in the native case */
569 if (!st->ss->external)
570 return 0;
571 if (!st->ss->reshape_super ||
572 !st->ss->manage_reshape) {
573 fprintf(stderr, Name ": %s metadata does not support reshape\n",
574 st->ss->name);
575 return 1;
576 }
577
578 return st->ss->reshape_super(st, size, level, layout, chunksize,
579 raid_disks, backup_file, dev, verbose);
580 }
581
582 static void sync_metadata(struct supertype *st)
583 {
584 if (st->ss->external) {
585 if (st->update_tail)
586 flush_metadata_updates(st);
587 else
588 st->ss->sync_metadata(st);
589 }
590 }
591
592 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
593 {
594 /* when dealing with external metadata subarrays we need to be
595 * prepared to handle EAGAIN. The kernel may need to wait for
596 * mdmon to mark the array active so the kernel can handle
597 * allocations/writeback when preparing the reshape action
598 * (md_allow_write()). We temporarily disable safe_mode_delay
599 * to close a race with the array_state going clean before the
600 * next write to raid_disks / stripe_cache_size
601 */
602 char safe[50];
603 int rc;
604
605 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
606 if (strcmp(name, "raid_disks") != 0 &&
607 strcmp(name, "stripe_cache_size") != 0)
608 return sysfs_set_num(sra, NULL, name, n);
609
610 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
611 if (rc <= 0)
612 return -1;
613 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
614 rc = sysfs_set_num(sra, NULL, name, n);
615 if (rc < 0 && errno == EAGAIN) {
616 ping_monitor(container);
617 /* if we get EAGAIN here then the monitor is not active
618 * so stop trying
619 */
620 rc = sysfs_set_num(sra, NULL, name, n);
621 }
622 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
623 return rc;
624 }
625
626 static int reshape_container_raid_disks(char *container, int raid_disks)
627 {
628 /* for each subarray switch to a raid level that can
629 * support the reshape, and set raid disks
630 */
631 struct mdstat_ent *ent, *e;
632 int changed = 0, rv = 0, err = 0;
633
634 ent = mdstat_read(1, 0);
635 if (!ent) {
636 fprintf(stderr, Name ": unable to read /proc/mdstat\n");
637 return -1;
638 }
639
640 changed = 0;
641 for (e = ent; e; e = e->next) {
642 struct mdinfo *sub;
643 unsigned int cache;
644 int level, takeover_delta = 0;
645
646 if (!is_container_member(e, container))
647 continue;
648
649 level = map_name(pers, e->level);
650 if (level == 0) {
651 sub = sysfs_read(-1, e->devnum, GET_VERSION);
652 if (!sub)
653 break;
654 /* metadata records 'orig_level' */
655 rv = sysfs_set_num(sub, NULL, "level", 4);
656 if (rv < 0) {
657 err = errno;
658 break;
659 }
660 /* we want spares to be used for capacity
661 * expansion, not rebuild
662 */
663 takeover_delta = 1;
664
665 sysfs_free(sub);
666 level = 4;
667 }
668
669 sub = NULL;
670 switch (level) {
671 default:
672 rv = -1;
673 break;
674 case 4:
675 case 5:
676 case 6:
677 sub = sysfs_read(-1, e->devnum, GET_CHUNK|GET_CACHE);
678 if (!sub)
679 break;
680 cache = (sub->array.chunk_size / 4096) * 4;
681 if (cache > sub->cache_size)
682 rv = subarray_set_num(container, sub,
683 "stripe_cache_size", cache);
684 if (rv) {
685 err = errno;
686 break;
687 }
688 /* fall through */
689 case 1:
690 if (!sub)
691 sub = sysfs_read(-1, e->devnum, GET_VERSION);
692 if (!sub)
693 break;
694
695 rv = subarray_set_num(container, sub, "raid_disks",
696 raid_disks + takeover_delta);
697 if (rv)
698 err = errno;
699 else
700 changed++;
701 break;
702 }
703 sysfs_free(sub);
704 if (rv)
705 break;
706 }
707 free_mdstat(ent);
708 if (rv) {
709 fprintf(stderr, Name
710 ": failed to initiate container reshape%s%s\n",
711 err ? ": " : "", err ? strerror(err) : "");
712 return rv;
713 }
714
715 return changed;
716 }
717
718 static void revert_container_raid_disks(struct supertype *st, int fd, char *container)
719 {
720 /* we failed to prepare all subarrays in the container for
721 * reshape, so cancel the changes and restore the nominal raid
722 * level
723 */
724 struct mdstat_ent *ent, *e;
725
726 ent = mdstat_read(0, 0);
727 if (!ent) {
728 fprintf(stderr, Name
729 ": failed to read /proc/mdstat while aborting reshape\n");
730 return;
731 }
732
733 if (st->ss->load_container(st, fd, NULL)) {
734 fprintf(stderr, Name
735 ": failed read metadata while aborting reshape\n");
736 return ;
737 }
738
739
740 for (e = ent; e; e = e->next) {
741 int level_fixed = 0, disks_fixed = 0;
742 struct mdinfo *sub, *prev;
743 char *subarray;
744
745 if (!is_container_member(e, container))
746 continue;
747
748 subarray = to_subarray(e, container);
749 prev = st->ss->container_content(st, subarray);
750
751 /* changing level might change raid_disks so we do it
752 * first and then check if raid_disks still needs fixing
753 */
754 if (map_name(pers, e->level) != prev->array.level) {
755 sub = sysfs_read(-1, e->devnum, GET_VERSION);
756 if (sub &&
757 !sysfs_set_num(sub, NULL, "level", prev->array.level))
758 level_fixed = 1;
759 sysfs_free(sub);
760 } else
761 level_fixed = 1;
762
763 sub = sysfs_read(-1, e->devnum, GET_DISKS);
764 if (sub && sub->array.raid_disks != prev->array.raid_disks) {
765 if (!subarray_set_num(container, sub, "raid_disks",
766 prev->array.raid_disks))
767 disks_fixed = 1;
768 } else if (sub)
769 disks_fixed = 1;
770 sysfs_free(sub);
771
772 if (!disks_fixed || !level_fixed)
773 fprintf(stderr, Name
774 ": failed to restore %s to a %d-disk %s array\n",
775 e->dev, prev->array.raid_disks,
776 map_num(pers, prev->array.level));
777 free(prev);
778 }
779 st->ss->free_super(st);
780 free_mdstat(ent);
781 }
782
783 int remove_disks_on_raid10_to_raid0_takeover(struct supertype *st,
784 struct mdinfo *sra,
785 int layout)
786 {
787 int nr_of_copies;
788 struct mdinfo *remaining;
789 int slot;
790
791 nr_of_copies = layout & 0xff;
792
793 remaining = sra->devs;
794 sra->devs = NULL;
795 /* for each 'copy', select one device and remove from the list. */
796 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
797 struct mdinfo **diskp;
798 int found = 0;
799
800 /* Find a working device to keep */
801 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
802 struct mdinfo *disk = *diskp;
803
804 if (disk->disk.raid_disk < slot)
805 continue;
806 if (disk->disk.raid_disk >= slot + nr_of_copies)
807 continue;
808 if (disk->disk.state & (1<<MD_DISK_REMOVED))
809 continue;
810 if (disk->disk.state & (1<<MD_DISK_FAULTY))
811 continue;
812 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
813 continue;
814
815 /* We have found a good disk to use! */
816 *diskp = disk->next;
817 disk->next = sra->devs;
818 sra->devs = disk;
819 found = 1;
820 break;
821 }
822 if (!found)
823 break;
824 }
825
826 if (slot < sra->array.raid_disks) {
827 /* didn't find all slots */
828 struct mdinfo **e;
829 e = &remaining;
830 while (*e)
831 e = &(*e)->next;
832 *e = sra->devs;
833 sra->devs = remaining;
834 return 1;
835 }
836
837 /* Remove all 'remaining' devices from the array */
838 while (remaining) {
839 struct mdinfo *sd = remaining;
840 remaining = sd->next;
841
842 sysfs_set_str(sra, sd, "state", "faulty");
843 sysfs_set_str(sra, sd, "slot", "none");
844 sysfs_set_str(sra, sd, "state", "remove");
845 sd->disk.state |= (1<<MD_DISK_REMOVED);
846 sd->disk.state &= ~(1<<MD_DISK_SYNC);
847 sd->next = sra->devs;
848 sra->devs = sd;
849 }
850 return 0;
851 }
852
853 void reshape_free_fdlist(int *fdlist,
854 unsigned long long *offsets,
855 int size)
856 {
857 int i;
858
859 for (i = 0; i < size; i++)
860 if (fdlist[i] >= 0)
861 close(fdlist[i]);
862
863 free(fdlist);
864 free(offsets);
865 }
866
867 int reshape_prepare_fdlist(char *devname,
868 struct mdinfo *sra,
869 int raid_disks,
870 int nrdisks,
871 unsigned long blocks,
872 char *backup_file,
873 int *fdlist,
874 unsigned long long *offsets)
875 {
876 int d = 0;
877 struct mdinfo *sd;
878
879 for (d = 0; d <= nrdisks; d++)
880 fdlist[d] = -1;
881 d = raid_disks;
882 for (sd = sra->devs; sd; sd = sd->next) {
883 if (sd->disk.state & (1<<MD_DISK_FAULTY))
884 continue;
885 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
886 char *dn = map_dev(sd->disk.major,
887 sd->disk.minor, 1);
888 fdlist[sd->disk.raid_disk]
889 = dev_open(dn, O_RDONLY);
890 offsets[sd->disk.raid_disk] = sd->data_offset*512;
891 if (fdlist[sd->disk.raid_disk] < 0) {
892 fprintf(stderr,
893 Name ": %s: cannot open component %s\n",
894 devname, dn ? dn : "-unknown-");
895 d = -1;
896 goto release;
897 }
898 } else if (backup_file == NULL) {
899 /* spare */
900 char *dn = map_dev(sd->disk.major,
901 sd->disk.minor, 1);
902 fdlist[d] = dev_open(dn, O_RDWR);
903 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
904 if (fdlist[d] < 0) {
905 fprintf(stderr, Name ": %s: cannot open component %s\n",
906 devname, dn ? dn : "-unknown-");
907 d = -1;
908 goto release;
909 }
910 d++;
911 }
912 }
913 release:
914 return d;
915 }
916
917 unsigned long compute_backup_blocks(int nchunk, int ochunk,
918 unsigned int ndata, unsigned int odata)
919 {
920 unsigned long a, b, blocks;
921 /* So how much do we need to backup.
922 * We need an amount of data which is both a whole number of
923 * old stripes and a whole number of new stripes.
924 * So LCM for (chunksize*datadisks).
925 */
926 a = (ochunk/512) * odata;
927 b = (nchunk/512) * ndata;
928 /* Find GCD */
929 while (a != b) {
930 if (a < b)
931 b -= a;
932 if (b < a)
933 a -= b;
934 }
935 /* LCM == product / GCD */
936 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
937
938 return blocks;
939 }
940
941
942 int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
943 long long size,
944 int level, char *layout_str, int chunksize, int raid_disks)
945 {
946 /* Make some changes in the shape of an array.
947 * The kernel must support the change.
948 *
949 * There are three different changes. Each can trigger
950 * a resync or recovery so we freeze that until we have
951 * requested everything (if kernel supports freezing - 2.6.30).
952 * The steps are:
953 * - change size (i.e. component_size)
954 * - change level
955 * - change layout/chunksize/ndisks
956 *
957 * The last can require a reshape. It is different on different
958 * levels so we need to check the level before actioning it.
959 * Some times the level change needs to be requested after the
960 * reshape (e.g. raid6->raid5, raid5->raid0)
961 *
962 */
963 struct mdu_array_info_s array, orig;
964 char *c;
965 int rv = 0;
966 struct supertype *st;
967 char *subarray = NULL;
968
969 int nchunk, ochunk;
970 int nlayout, olayout;
971 int ndisks, odisks;
972 unsigned int ndata, odata;
973 int orig_level = UnSet;
974 char alt_layout[40];
975 int *fdlist;
976 unsigned long long *offsets;
977 int d, i;
978 int nrdisks;
979 int err;
980 int frozen;
981 unsigned long blocks, stripes;
982 unsigned long cache;
983 unsigned long long array_size;
984 int changed = 0;
985 char *container = NULL;
986 int cfd = -1;
987 int done;
988
989 struct mdinfo *sra;
990 struct mdinfo *sd;
991
992 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
993 fprintf(stderr, Name ": %s is not an active md array - aborting\n",
994 devname);
995 return 1;
996 }
997
998 if (size >= 0 &&
999 (chunksize || level!= UnSet || layout_str || raid_disks)) {
1000 fprintf(stderr, Name ": cannot change component size at the same time "
1001 "as other changes.\n"
1002 " Change size first, then check data is intact before "
1003 "making other changes.\n");
1004 return 1;
1005 }
1006
1007 if (raid_disks && raid_disks < array.raid_disks && array.level > 1 &&
1008 get_linux_version() < 2006032 &&
1009 !check_env("MDADM_FORCE_FEWER")) {
1010 fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n"
1011 " Please use a newer kernel\n");
1012 return 1;
1013 }
1014
1015 st = super_by_fd(fd, &subarray);
1016 if (!st) {
1017 fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname);
1018 return 1;
1019 }
1020
1021 /* in the external case we need to check that the requested reshape is
1022 * supported, and perform an initial check that the container holds the
1023 * pre-requisite spare devices (mdmon owns final validation)
1024 */
1025 if (st->ss->external) {
1026 int container_dev;
1027 int rv;
1028
1029 if (subarray) {
1030 container_dev = st->container_dev;
1031 cfd = open_dev_excl(st->container_dev);
1032 } else if (size >= 0 || layout_str != NULL || chunksize != 0 ||
1033 level != UnSet) {
1034 fprintf(stderr,
1035 Name ": %s is a container, only 'raid-devices' can be changed\n",
1036 devname);
1037 return 1;
1038 } else {
1039 container_dev = st->devnum;
1040 close(fd);
1041 cfd = open_dev_excl(st->devnum);
1042 fd = cfd;
1043 }
1044 if (cfd < 0) {
1045 fprintf(stderr, Name ": Unable to open container for %s\n",
1046 devname);
1047 free(subarray);
1048 return 1;
1049 }
1050
1051 container = devnum2devname(st->devnum);
1052 if (!container) {
1053 fprintf(stderr, Name ": Could not determine container name\n");
1054 free(subarray);
1055 return 1;
1056 }
1057
1058 if (subarray)
1059 rv = st->ss->load_container(st, cfd, NULL);
1060 else
1061 rv = st->ss->load_super(st, cfd, NULL);
1062 if (rv) {
1063 fprintf(stderr, Name ": Cannot read superblock for %s\n",
1064 devname);
1065 free(subarray);
1066 return 1;
1067 }
1068
1069 if (mdmon_running(container_dev))
1070 st->update_tail = &st->updates;
1071 }
1072
1073 sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS | GET_STATE);
1074 if (sra) {
1075 if (st->ss->external && subarray == NULL) {
1076 array.level = LEVEL_CONTAINER;
1077 sra->array.level = LEVEL_CONTAINER;
1078 }
1079 } else {
1080 fprintf(stderr, Name ": failed to read sysfs parameters for %s\n",
1081 devname);
1082 return 1;
1083 }
1084 frozen = freeze(st);
1085 if (frozen < -1) {
1086 /* freeze() already spewed the reason */
1087 return 1;
1088 } else if (frozen < 0) {
1089 fprintf(stderr, Name ": %s is performing resync/recovery and cannot"
1090 " be reshaped\n", devname);
1091 return 1;
1092 }
1093
1094 /* ========= set size =============== */
1095 if (size >= 0 && (size == 0 || size != array.size)) {
1096 long long orig_size = array.size;
1097
1098 if (reshape_super(st, size, UnSet, UnSet, 0, 0, NULL, devname, !quiet)) {
1099 rv = 1;
1100 goto release;
1101 }
1102 sync_metadata(st);
1103 array.size = size;
1104 if (array.size != size) {
1105 /* got truncated to 32bit, write to
1106 * component_size instead
1107 */
1108 if (sra)
1109 rv = sysfs_set_num(sra, NULL,
1110 "component_size", size);
1111 else
1112 rv = -1;
1113 } else
1114 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1115 if (rv != 0) {
1116 int err = errno;
1117
1118 /* restore metadata */
1119 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1120 NULL, devname, !quiet) == 0)
1121 sync_metadata(st);
1122 fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
1123 devname, strerror(err));
1124 if (err == EBUSY &&
1125 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1126 fprintf(stderr, " Bitmap must be removed before size can be changed\n");
1127 rv = 1;
1128 goto release;
1129 }
1130 ioctl(fd, GET_ARRAY_INFO, &array);
1131 size = get_component_size(fd)/2;
1132 if (size == 0)
1133 size = array.size;
1134 if (!quiet)
1135 fprintf(stderr, Name ": component size of %s has been set to %lluK\n",
1136 devname, size);
1137 changed = 1;
1138 } else if (array.level != LEVEL_CONTAINER) {
1139 size = get_component_size(fd)/2;
1140 if (size == 0)
1141 size = array.size;
1142 }
1143
1144 /* ========= check for Raid10 -> Raid0 conversion ===============
1145 * current implemenation assumes that following conditions must be met:
1146 * - far_copies == 1
1147 * - near_copies == 2
1148 */
1149 if (level == 0 && array.level == 10 &&
1150 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) {
1151 int err;
1152 err = remove_disks_on_raid10_to_raid0_takeover(st, sra, array.layout);
1153 if (err) {
1154 dprintf(Name": Array cannot be reshaped\n");
1155 if (container)
1156 free(container);
1157 if (cfd > -1)
1158 close(cfd);
1159 return 1;
1160 }
1161 }
1162
1163 /* ======= set level =========== */
1164 if (level != UnSet && level != array.level) {
1165 /* Trying to change the level.
1166 * We might need to change layout first and schedule a
1167 * level change for later.
1168 * Level changes that can happen immediately are:
1169 * 0->4,5,6 1->5 4->5,6 5->1,6
1170 * Level changes that need a layout change first are:
1171 * 6->5,4,0 : need a -6 layout, or parity-last
1172 * 5->4,0 : need parity-last
1173 */
1174 if ((array.level == 6 || array.level == 5) &&
1175 (level == 5 || level == 4 || level == 0)) {
1176 /* Don't change level yet, but choose intermediate
1177 * layout
1178 */
1179 if (level == 5) {
1180 if (layout_str == NULL)
1181 switch (array.layout) {
1182 case ALGORITHM_LEFT_ASYMMETRIC:
1183 case ALGORITHM_LEFT_ASYMMETRIC_6:
1184 case ALGORITHM_ROTATING_N_RESTART:
1185 layout_str = "left-asymmetric-6";
1186 break;
1187 case ALGORITHM_LEFT_SYMMETRIC:
1188 case ALGORITHM_LEFT_SYMMETRIC_6:
1189 case ALGORITHM_ROTATING_N_CONTINUE:
1190 layout_str = "left-symmetric-6";
1191 break;
1192 case ALGORITHM_RIGHT_ASYMMETRIC:
1193 case ALGORITHM_RIGHT_ASYMMETRIC_6:
1194 case ALGORITHM_ROTATING_ZERO_RESTART:
1195 layout_str = "right-asymmetric-6";
1196 break;
1197 case ALGORITHM_RIGHT_SYMMETRIC:
1198 case ALGORITHM_RIGHT_SYMMETRIC_6:
1199 layout_str = "right-symmetric-6";
1200 break;
1201 case ALGORITHM_PARITY_0:
1202 case ALGORITHM_PARITY_0_6:
1203 layout_str = "parity-first-6";
1204 break;
1205 case ALGORITHM_PARITY_N:
1206 layout_str = "parity-last";
1207 break;
1208 default:
1209 fprintf(stderr, Name ": %s: cannot"
1210 "convert layout to RAID5 equivalent\n",
1211 devname);
1212 rv = 1;
1213 goto release;
1214 }
1215 else {
1216 int l = map_name(r5layout, layout_str);
1217 if (l == UnSet) {
1218 fprintf(stderr, Name ": %s: layout '%s' not recognised\n",
1219 devname, layout_str);
1220 rv = 1;
1221 goto release;
1222 }
1223 if (l != ALGORITHM_PARITY_N) {
1224 /* need the -6 version */
1225 char *ls = map_num(r5layout, l);
1226 strcat(strcpy(alt_layout, ls),
1227 "-6");
1228 layout_str = alt_layout;
1229 }
1230 }
1231 if (raid_disks)
1232 /* The final raid6->raid5 conversion
1233 * will reduce the number of disks,
1234 * so now we need to aim higher
1235 */
1236 raid_disks++;
1237 } else
1238 layout_str = "parity-last";
1239 } else {
1240 /* Level change is a simple takeover. In the external
1241 * case we don't check with the metadata handler until
1242 * we establish what the final layout will be. If the
1243 * level change is disallowed we will revert to
1244 * orig_level without disturbing the metadata, otherwise
1245 * we will send an update.
1246 */
1247 c = map_num(pers, level);
1248 if (c == NULL) {
1249 rv = 1;/* not possible */
1250 goto release;
1251 }
1252 err = sysfs_set_str(sra, NULL, "level", c);
1253 if (err) {
1254 err = errno;
1255 fprintf(stderr, Name ": %s: could not set level to %s\n",
1256 devname, c);
1257 if (err == EBUSY &&
1258 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1259 fprintf(stderr, " Bitmap must be removed before level can be changed\n");
1260 rv = 1;
1261 goto release;
1262 }
1263 orig = array;
1264 orig_level = orig.level;
1265 ioctl(fd, GET_ARRAY_INFO, &array);
1266 if (layout_str == NULL &&
1267 orig.level == 5 && level == 6 &&
1268 array.layout != orig.layout)
1269 layout_str = map_num(r5layout, orig.layout);
1270 if (!quiet)
1271 fprintf(stderr, Name " level of %s changed to %s\n",
1272 devname, c);
1273 changed = 1;
1274 }
1275 }
1276
1277 /* ========= set shape (chunk_size / layout / ndisks) ============== */
1278 /* Check if layout change is a no-op */
1279 switch (array.level) {
1280 case 5:
1281 if (layout_str && array.layout == map_name(r5layout, layout_str))
1282 layout_str = NULL;
1283 break;
1284 case 6:
1285 if (layout_str == NULL &&
1286 ((chunksize && chunksize * 1024 != array.chunk_size) ||
1287 (raid_disks && raid_disks != array.raid_disks)) &&
1288 array.layout >= 16) {
1289 fprintf(stderr, Name
1290 ": %s has a non-standard layout. If you wish to preserve this\n"
1291 " during the reshape, please specify --layout=preserve\n"
1292 " If you want to change it, specify a layout or use --layout=normalise\n",
1293 devname);
1294 rv = 1;
1295 goto release;
1296 }
1297 if (layout_str &&
1298 (strcmp(layout_str, "normalise") == 0 ||
1299 strcmp(layout_str, "normalize") == 0)) {
1300 char *hyphen;
1301 strcpy(alt_layout, map_num(r6layout, array.layout));
1302 hyphen = strrchr(alt_layout, '-');
1303 if (hyphen && strcmp(hyphen, "-6") == 0) {
1304 *hyphen = 0;
1305 layout_str = alt_layout;
1306 }
1307 }
1308
1309 if (layout_str && array.layout == map_name(r6layout, layout_str))
1310 layout_str = NULL;
1311 if (layout_str && strcmp(layout_str, "preserve") == 0)
1312 layout_str = NULL;
1313 break;
1314 }
1315 if (layout_str == NULL
1316 && (chunksize == 0 || chunksize*1024 == array.chunk_size)
1317 && (raid_disks == 0 || raid_disks == array.raid_disks)) {
1318 if (reshape_super(st, -1, level, UnSet, 0, 0, NULL, devname, !quiet)) {
1319 rv = 1;
1320 goto release;
1321 }
1322 sync_metadata(st);
1323 rv = 0;
1324 if (level != UnSet && level != array.level) {
1325 /* Looks like this level change doesn't need
1326 * a reshape after all.
1327 */
1328 c = map_num(pers, level);
1329 if (c) {
1330 rv = sysfs_set_str(sra, NULL, "level", c);
1331 if (rv) {
1332 int err = errno;
1333 fprintf(stderr, Name ": %s: could not set level to %s\n",
1334 devname, c);
1335 if (err == EBUSY &&
1336 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1337 fprintf(stderr, " Bitmap must be removed before level can be changed\n");
1338 rv = 1;
1339 }
1340 }
1341 } else if (!changed && !quiet)
1342 fprintf(stderr, Name ": %s: no change requested\n",
1343 devname);
1344
1345 if (st->ss->external && !mdmon_running(st->container_dev) &&
1346 level > 0) {
1347 start_mdmon(st->container_dev);
1348 ping_monitor(container);
1349 }
1350 goto release;
1351 }
1352
1353 c = map_num(pers, array.level);
1354 if (c == NULL) c = "-unknown-";
1355 switch (array.level) {
1356 default: /* raid0, linear, multipath cannot be reconfigured */
1357 fprintf(stderr, Name ": %s array %s cannot be reshaped.\n",
1358 c, devname);
1359 /* TODO raid0 raiddisks can be reshaped via raid4 */
1360 rv = 1;
1361 break;
1362 case LEVEL_CONTAINER: {
1363 int count;
1364
1365 /* double check that we are not changing anything but raid_disks */
1366 if (size >= 0 || layout_str != NULL || chunksize != 0 || level != UnSet) {
1367 fprintf(stderr,
1368 Name ": %s is a container, only 'raid-devices' can be changed\n",
1369 devname);
1370 rv = 1;
1371 goto release;
1372 }
1373
1374 st->update_tail = &st->updates;
1375 if (reshape_super(st, -1, UnSet, UnSet, 0, raid_disks,
1376 backup_file, devname, !quiet)) {
1377 rv = 1;
1378 goto release;
1379 }
1380
1381 count = reshape_container_raid_disks(container, raid_disks);
1382 if (count < 0) {
1383 revert_container_raid_disks(st, fd, container);
1384 rv = 1;
1385 goto release;
1386 } else if (count == 0) {
1387 if (!quiet)
1388 fprintf(stderr, Name
1389 ": no active subarrays to reshape\n");
1390 goto release;
1391 }
1392
1393 if (!mdmon_running(st->devnum)) {
1394 start_mdmon(st->devnum);
1395 ping_monitor(container);
1396 }
1397 sync_metadata(st);
1398
1399 /* give mdmon a chance to allocate spares */
1400 ping_manager(container);
1401
1402 /* manage_reshape takes care of releasing the array(s) */
1403 st->ss->manage_reshape(st, backup_file);
1404 frozen = 0;
1405 goto release;
1406 }
1407 case LEVEL_FAULTY: /* only 'layout' change is permitted */
1408
1409 if (chunksize || raid_disks) {
1410 fprintf(stderr, Name ": %s: Cannot change chunksize or disks of a 'faulty' array\n",
1411 devname);
1412 rv = 1;
1413 break;
1414 }
1415 if (layout_str == NULL)
1416 break; /* nothing to do.... */
1417
1418 array.layout = parse_layout_faulty(layout_str);
1419 if (array.layout < 0) {
1420 fprintf(stderr, Name ": %s: layout %s not understood for 'faulty' array\n",
1421 devname, layout_str);
1422 rv = 1;
1423 break;
1424 }
1425 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1426 fprintf(stderr, Name ": Cannot set layout for %s: %s\n",
1427 devname, strerror(errno));
1428 rv = 1;
1429 } else if (!quiet)
1430 printf("layout for %s set to %d\n", devname, array.layout);
1431 break;
1432
1433 case 1: /* only raid_disks can each be changed. */
1434
1435 if (chunksize || layout_str != NULL) {
1436 fprintf(stderr, Name ": %s: Cannot change chunk size or layout for a RAID1 array.\n",
1437 devname);
1438 rv = 1;
1439 break;
1440 }
1441 if (raid_disks > 0) {
1442 if (reshape_super(st, -1, UnSet, UnSet, 0, raid_disks,
1443 NULL, devname, !quiet)) {
1444 rv = 1;
1445 goto release;
1446 }
1447 sync_metadata(st);
1448 array.raid_disks = raid_disks;
1449 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1450 fprintf(stderr, Name ": Cannot set raid-devices for %s: %s\n",
1451 devname, strerror(errno));
1452 rv = 1;
1453 }
1454 }
1455 break;
1456
1457 case 4:
1458 case 5:
1459 case 6:
1460
1461 /*
1462 * layout/chunksize/raid_disks can be changed
1463 * though the kernel may not support it all.
1464 */
1465 if (subarray) {
1466 fprintf(stderr, Name ": Cannot reshape subarrays yet\n");
1467 break;
1468 }
1469
1470 /*
1471 * There are three possibilities.
1472 * 1/ The array will shrink.
1473 * We need to ensure the reshape will pause before reaching
1474 * the 'critical section'. We also need to fork and wait for
1475 * that to happen. When it does we
1476 * suspend/backup/complete/unfreeze
1477 *
1478 * 2/ The array will not change size.
1479 * This requires that we keep a backup of a sliding window
1480 * so that we can restore data after a crash. So we need
1481 * to fork and monitor progress.
1482 *
1483 * 3/ The array will grow. This is relatively easy.
1484 * However the kernel's restripe routines will cheerfully
1485 * overwrite some early data before it is safe. So we
1486 * need to make a backup of the early parts of the array
1487 * and be ready to restore it if rebuild aborts very early.
1488 *
1489 * We backup data by writing it to one spare, or to a
1490 * file which was given on command line.
1491 *
1492 * [FOLLOWING IS OLD AND PARTLY WRONG]
1493 * So: we enumerate the devices in the array and
1494 * make sure we can open all of them.
1495 * Then we freeze the early part of the array and
1496 * backup to the various spares.
1497 * Then we request changes and start the reshape.
1498 * Monitor progress until it has passed the danger zone.
1499 * and finally invalidate the copied data and unfreeze the
1500 * start of the array.
1501 *
1502 * In each case, we first make sure that storage is available
1503 * for the required backup.
1504 * Then we:
1505 * - request the shape change.
1506 * - for to handle backup etc.
1507 */
1508 nchunk = ochunk = array.chunk_size;
1509 nlayout = olayout = array.layout;
1510 ndisks = odisks = array.raid_disks;
1511
1512 if (chunksize) {
1513 nchunk = chunksize * 1024;
1514 if (size % chunksize) {
1515 fprintf(stderr, Name ": component size %lluK is not"
1516 " a multiple of chunksize %dK\n",
1517 size, chunksize);
1518 break;
1519 }
1520 }
1521 if (layout_str != NULL)
1522 switch(array.level) {
1523 case 4: /* ignore layout */
1524 break;
1525 case 5:
1526 nlayout = map_name(r5layout, layout_str);
1527 if (nlayout == UnSet) {
1528 fprintf(stderr, Name ": layout %s not understood for raid5.\n",
1529 layout_str);
1530 rv = 1;
1531 goto release;
1532 }
1533 break;
1534
1535 case 6:
1536 nlayout = map_name(r6layout, layout_str);
1537 if (nlayout == UnSet) {
1538 fprintf(stderr, Name ": layout %s not understood for raid6.\n",
1539 layout_str);
1540 rv = 1;
1541 goto release;
1542 }
1543 break;
1544 }
1545 if (raid_disks) ndisks = raid_disks;
1546
1547 odata = odisks-1;
1548 ndata = ndisks-1;
1549 if (array.level == 6) {
1550 odata--; /* number of data disks */
1551 ndata--;
1552 }
1553
1554 if (odata == ndata &&
1555 get_linux_version() < 2006032) {
1556 fprintf(stderr, Name ": in-place reshape is not safe before 2.6.32, sorry.\n");
1557 break;
1558 }
1559
1560 /* Check that we can hold all the data */
1561 get_dev_size(fd, NULL, &array_size);
1562 if (ndata * (unsigned long long)size < (array_size/1024)) {
1563 fprintf(stderr, Name ": this change will reduce the size of the array.\n"
1564 " use --grow --array-size first to truncate array.\n"
1565 " e.g. mdadm --grow %s --array-size %llu\n",
1566 devname, ndata * size);
1567 rv = 1;
1568 break;
1569 }
1570
1571 blocks = compute_backup_blocks(nchunk, ochunk, ndata, odata);
1572
1573 sysfs_free(sra);
1574 sra = sysfs_read(fd, 0,
1575 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|
1576 GET_CACHE);
1577
1578 if (!sra) {
1579 fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
1580 devname);
1581 rv = 1;
1582 break;
1583 }
1584
1585 if (ndata == odata) {
1586 /* Make 'blocks' bigger for better throughput, but
1587 * not so big that we reject it below.
1588 * Try for 16 megabytes
1589 */
1590 while (blocks * 32 < sra->component_size &&
1591 blocks < 16*1024*2)
1592 blocks *= 2;
1593 } else
1594 fprintf(stderr, Name ": Need to backup %luK of critical "
1595 "section..\n", blocks/2);
1596
1597 if (blocks >= sra->component_size/2) {
1598 fprintf(stderr, Name ": %s: Something wrong - reshape aborted\n",
1599 devname);
1600 rv = 1;
1601 break;
1602 }
1603 nrdisks = array.raid_disks + sra->array.spare_disks;
1604 /* Now we need to open all these devices so we can read/write.
1605 */
1606 fdlist = malloc((1+nrdisks) * sizeof(int));
1607 offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
1608 if (!fdlist || !offsets) {
1609 fprintf(stderr, Name ": malloc failed: grow aborted\n");
1610 rv = 1;
1611 break;
1612 }
1613
1614 d = reshape_prepare_fdlist(devname, sra, array.raid_disks,
1615 nrdisks, blocks, backup_file,
1616 fdlist, offsets);
1617 if (d < 0) {
1618 rv = 1;
1619 goto release;
1620 }
1621 if (backup_file == NULL) {
1622 if (st->ss->external && !st->ss->manage_reshape) {
1623 fprintf(stderr, Name ": %s Grow operation not supported by %s metadata\n",
1624 devname, st->ss->name);
1625 rv = 1;
1626 break;
1627 }
1628 if (ndata <= odata) {
1629 fprintf(stderr, Name ": %s: Cannot grow - need backup-file\n",
1630 devname);
1631 rv = 1;
1632 break;
1633 } else if (sra->array.spare_disks == 0) {
1634 fprintf(stderr, Name ": %s: Cannot grow - need a spare or "
1635 "backup-file to backup critical section\n",
1636 devname);
1637 rv = 1;
1638 break;
1639 }
1640 if (d == array.raid_disks) {
1641 fprintf(stderr, Name ": %s: No spare device for backup\n",
1642 devname);
1643 rv = 1;
1644 break;
1645 }
1646 } else {
1647 /* need to check backup file is large enough */
1648 char buf[512];
1649 fdlist[d] = open(backup_file, O_RDWR|O_CREAT|O_EXCL,
1650 S_IRUSR | S_IWUSR);
1651 offsets[d] = 8 * 512;
1652 if (fdlist[d] < 0) {
1653 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
1654 devname, backup_file, strerror(errno));
1655 rv = 1;
1656 break;
1657 }
1658 memset(buf, 0, 512);
1659 for (i=0; i < (signed)blocks + 1 ; i++) {
1660 if (write(fdlist[d], buf, 512) != 512) {
1661 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
1662 devname, backup_file, strerror(errno));
1663 rv = 1;
1664 break;
1665 }
1666 }
1667 if (fsync(fdlist[d]) != 0) {
1668 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
1669 devname, backup_file, strerror(errno));
1670 rv = 1;
1671 break;
1672 }
1673 d++;
1674 }
1675
1676 /* check that the operation is supported by the metadata */
1677 if (reshape_super(st, -1, level, nlayout, nchunk, ndisks,
1678 backup_file, devname, !quiet)) {
1679 rv = 1;
1680 break;
1681 }
1682
1683 /* lastly, check that the internal stripe cache is
1684 * large enough, or it won't work.
1685 */
1686
1687 cache = (nchunk < ochunk) ? ochunk : nchunk;
1688 cache = cache * 4 / 4096;
1689 if (cache < blocks / 8 / odisks + 16)
1690 /* Make it big enough to hold 'blocks' */
1691 cache = blocks / 8 / odisks + 16;
1692 if (sra->cache_size < cache)
1693 sysfs_set_num(sra, NULL, "stripe_cache_size",
1694 cache+1);
1695 /* Right, everything seems fine. Let's kick things off.
1696 * If only changing raid_disks, use ioctl, else use
1697 * sysfs.
1698 */
1699 sync_metadata(st);
1700 if (ochunk == nchunk && olayout == nlayout) {
1701 array.raid_disks = ndisks;
1702 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1703 int err = errno;
1704 rv = 1;
1705 fprintf(stderr, Name ": Cannot set device shape for %s: %s\n",
1706 devname, strerror(errno));
1707 if (ndisks < odisks &&
1708 get_linux_version() < 2006030)
1709 fprintf(stderr, Name ": linux 2.6.30 or later required\n");
1710 if (err == EBUSY &&
1711 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1712 fprintf(stderr, " Bitmap must be removed before shape can be changed\n");
1713
1714 break;
1715 }
1716 } else {
1717 /* set them all just in case some old 'new_*' value
1718 * persists from some earlier problem
1719 */
1720 int err = err; /* only used if rv==1, and always set if
1721 * rv==1, so initialisation not needed,
1722 * despite gcc warning
1723 */
1724 if (sysfs_set_num(sra, NULL, "chunk_size", nchunk) < 0)
1725 rv = 1, err = errno;
1726 if (!rv && sysfs_set_num(sra, NULL, "layout", nlayout) < 0)
1727 rv = 1, err = errno;
1728 if (!rv && sysfs_set_num(sra, NULL, "raid_disks", ndisks) < 0)
1729 rv = 1, err = errno;
1730 if (rv) {
1731 fprintf(stderr, Name ": Cannot set device shape for %s\n",
1732 devname);
1733 if (get_linux_version() < 2006030)
1734 fprintf(stderr, Name ": linux 2.6.30 or later required\n");
1735 if (err == EBUSY &&
1736 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1737 fprintf(stderr, " Bitmap must be removed before shape can be changed\n");
1738 break;
1739 }
1740 }
1741
1742 if (ndisks == 2 && odisks == 2) {
1743 /* No reshape is needed in this trivial case */
1744 rv = 0;
1745 break;
1746 }
1747
1748 if (st->ss->external) {
1749 /* metadata handler takes it from here */
1750 ping_manager(container);
1751 st->ss->manage_reshape(st, backup_file);
1752 frozen = 0;
1753 break;
1754 }
1755
1756 /* set up the backup-super-block. This requires the
1757 * uuid from the array.
1758 */
1759 /* Find a superblock */
1760 for (sd = sra->devs; sd; sd = sd->next) {
1761 char *dn;
1762 int devfd;
1763 int ok;
1764 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1765 continue;
1766 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
1767 devfd = dev_open(dn, O_RDONLY);
1768 if (devfd < 0)
1769 continue;
1770 ok = st->ss->load_super(st, devfd, NULL);
1771 close(devfd);
1772 if (ok >= 0)
1773 break;
1774 }
1775 if (!sd) {
1776 fprintf(stderr, Name ": %s: Cannot find a superblock\n",
1777 devname);
1778 rv = 1;
1779 break;
1780 }
1781
1782 memset(&bsb, 0, 512);
1783 memcpy(bsb.magic, "md_backup_data-1", 16);
1784 st->ss->uuid_from_super(st, (int*)&bsb.set_uuid);
1785 bsb.mtime = __cpu_to_le64(time(0));
1786 bsb.devstart2 = blocks;
1787 stripes = blocks / (ochunk/512) / odata;
1788 /* Now we just need to kick off the reshape and watch, while
1789 * handling backups of the data...
1790 * This is all done by a forked background process.
1791 */
1792 switch(fork()) {
1793 case 0:
1794 close(fd);
1795 if (check_env("MDADM_GROW_VERIFY"))
1796 fd = open(devname, O_RDONLY | O_DIRECT);
1797 else
1798 fd = -1;
1799 mlockall(MCL_FUTURE);
1800
1801 if (odata < ndata)
1802 done = child_grow(fd, sra, stripes,
1803 fdlist, offsets,
1804 odisks, ochunk, array.level, olayout, odata,
1805 d - odisks, fdlist+odisks, offsets+odisks);
1806 else if (odata > ndata)
1807 done = child_shrink(fd, sra, stripes,
1808 fdlist, offsets,
1809 odisks, ochunk, array.level, olayout, odata,
1810 d - odisks, fdlist+odisks, offsets+odisks);
1811 else
1812 done = child_same_size(fd, sra, stripes,
1813 fdlist, offsets,
1814 0,
1815 odisks, ochunk, array.level, olayout, odata,
1816 d - odisks, fdlist+odisks, offsets+odisks);
1817 if (backup_file && done)
1818 unlink(backup_file);
1819 if (level != UnSet && level != array.level) {
1820 /* We need to wait for the reshape to finish
1821 * (which will have happened unless odata < ndata)
1822 * and then set the level
1823 */
1824
1825 c = map_num(pers, level);
1826 if (c == NULL)
1827 exit(0);/* not possible */
1828
1829 if (odata < ndata)
1830 wait_reshape(sra);
1831 err = sysfs_set_str(sra, NULL, "level", c);
1832 if (err)
1833 fprintf(stderr, Name ": %s: could not set level to %s\n",
1834 devname, c);
1835 }
1836 exit(0);
1837 case -1:
1838 fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n",
1839 strerror(errno));
1840 rv = 1;
1841 break;
1842 default:
1843 /* The child will take care of unfreezing the array */
1844 frozen = 0;
1845 break;
1846 }
1847 break;
1848
1849 }
1850
1851 release:
1852 if (rv && orig_level != UnSet && sra) {
1853 c = map_num(pers, orig_level);
1854 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
1855 fprintf(stderr, Name ": aborting level change\n");
1856 }
1857 unfreeze(st, frozen);
1858 return rv;
1859 }
1860
1861 /*
1862 * We run a child process in the background which performs the following
1863 * steps:
1864 * - wait for resync to reach a certain point
1865 * - suspend io to the following section
1866 * - backup that section
1867 * - allow resync to proceed further
1868 * - resume io
1869 * - discard the backup.
1870 *
1871 * When are combined in slightly different ways in the three cases.
1872 * Grow:
1873 * - suspend/backup/allow/wait/resume/discard
1874 * Shrink:
1875 * - allow/wait/suspend/backup/allow/wait/resume/discard
1876 * same-size:
1877 * - wait/resume/discard/suspend/backup/allow
1878 *
1879 * suspend/backup/allow always come together
1880 * wait/resume/discard do too.
1881 * For the same-size case we have two backups to improve flow.
1882 *
1883 */
1884
1885 /* FIXME return status is never checked */
1886 static int grow_backup(struct mdinfo *sra,
1887 unsigned long long offset, /* per device */
1888 unsigned long stripes, /* per device */
1889 int *sources, unsigned long long *offsets,
1890 int disks, int chunk, int level, int layout,
1891 int dests, int *destfd, unsigned long long *destoffsets,
1892 int part, int *degraded,
1893 char *buf)
1894 {
1895 /* Backup 'blocks' sectors at 'offset' on each device of the array,
1896 * to storage 'destfd' (offset 'destoffsets'), after first
1897 * suspending IO. Then allow resync to continue
1898 * over the suspended section.
1899 * Use part 'part' of the backup-super-block.
1900 */
1901 int odata = disks;
1902 int rv = 0;
1903 int i;
1904 unsigned long long ll;
1905 int new_degraded;
1906 //printf("offset %llu\n", offset);
1907 if (level >= 4)
1908 odata--;
1909 if (level == 6)
1910 odata--;
1911 sysfs_set_num(sra, NULL, "suspend_hi", (offset + stripes * (chunk/512)) * odata);
1912 /* Check that array hasn't become degraded, else we might backup the wrong data */
1913 sysfs_get_ll(sra, NULL, "degraded", &ll);
1914 new_degraded = (int)ll;
1915 if (new_degraded != *degraded) {
1916 /* check each device to ensure it is still working */
1917 struct mdinfo *sd;
1918 for (sd = sra->devs ; sd ; sd = sd->next) {
1919 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1920 continue;
1921 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
1922 char sbuf[20];
1923 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
1924 strstr(sbuf, "faulty") ||
1925 strstr(sbuf, "in_sync") == NULL) {
1926 /* this device is dead */
1927 sd->disk.state = (1<<MD_DISK_FAULTY);
1928 if (sd->disk.raid_disk >= 0 &&
1929 sources[sd->disk.raid_disk] >= 0) {
1930 close(sources[sd->disk.raid_disk]);
1931 sources[sd->disk.raid_disk] = -1;
1932 }
1933 }
1934 }
1935 }
1936 *degraded = new_degraded;
1937 }
1938 if (part) {
1939 bsb.arraystart2 = __cpu_to_le64(offset * odata);
1940 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
1941 } else {
1942 bsb.arraystart = __cpu_to_le64(offset * odata);
1943 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
1944 }
1945 if (part)
1946 bsb.magic[15] = '2';
1947 for (i = 0; i < dests; i++)
1948 if (part)
1949 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
1950 else
1951 lseek64(destfd[i], destoffsets[i], 0);
1952
1953 rv = save_stripes(sources, offsets,
1954 disks, chunk, level, layout,
1955 dests, destfd,
1956 offset*512*odata, stripes * chunk * odata,
1957 buf);
1958
1959 if (rv)
1960 return rv;
1961 bsb.mtime = __cpu_to_le64(time(0));
1962 for (i = 0; i < dests; i++) {
1963 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
1964
1965 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
1966 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
1967 bsb.sb_csum2 = bsb_csum((char*)&bsb,
1968 ((char*)&bsb.sb_csum2)-((char*)&bsb));
1969
1970 rv = -1;
1971 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
1972 != destoffsets[i] - 4096)
1973 break;
1974 if (write(destfd[i], &bsb, 512) != 512)
1975 break;
1976 if (destoffsets[i] > 4096) {
1977 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
1978 destoffsets[i]+stripes*chunk*odata)
1979 break;
1980 if (write(destfd[i], &bsb, 512) != 512)
1981 break;
1982 }
1983 fsync(destfd[i]);
1984 rv = 0;
1985 }
1986
1987 return rv;
1988 }
1989
1990 /* in 2.6.30, the value reported by sync_completed can be
1991 * less that it should be by one stripe.
1992 * This only happens when reshape hits sync_max and pauses.
1993 * So allow wait_backup to either extent sync_max further
1994 * than strictly necessary, or return before the
1995 * sync has got quite as far as we would really like.
1996 * This is what 'blocks2' is for.
1997 * The various caller give appropriate values so that
1998 * every works.
1999 */
2000 /* FIXME return value is often ignored */
2001 static int wait_backup(struct mdinfo *sra,
2002 unsigned long long offset, /* per device */
2003 unsigned long long blocks, /* per device */
2004 unsigned long long blocks2, /* per device - hack */
2005 int dests, int *destfd, unsigned long long *destoffsets,
2006 int part)
2007 {
2008 /* Wait for resync to pass the section that was backed up
2009 * then erase the backup and allow IO
2010 */
2011 int fd = sysfs_get_fd(sra, NULL, "sync_completed");
2012 unsigned long long completed;
2013 int i;
2014 int rv;
2015
2016 if (fd < 0)
2017 return -1;
2018 sysfs_set_num(sra, NULL, "sync_max", offset + blocks + blocks2);
2019 if (offset == 0)
2020 sysfs_set_str(sra, NULL, "sync_action", "reshape");
2021 do {
2022 char action[20];
2023 fd_set rfds;
2024 FD_ZERO(&rfds);
2025 FD_SET(fd, &rfds);
2026 select(fd+1, NULL, NULL, &rfds, NULL);
2027 if (sysfs_fd_get_ll(fd, &completed) < 0) {
2028 close(fd);
2029 return -1;
2030 }
2031 if (sysfs_get_str(sra, NULL, "sync_action",
2032 action, 20) > 0 &&
2033 strncmp(action, "reshape", 7) != 0)
2034 break;
2035 } while (completed < offset + blocks);
2036 close(fd);
2037
2038 if (part) {
2039 bsb.arraystart2 = __cpu_to_le64(0);
2040 bsb.length2 = __cpu_to_le64(0);
2041 } else {
2042 bsb.arraystart = __cpu_to_le64(0);
2043 bsb.length = __cpu_to_le64(0);
2044 }
2045 bsb.mtime = __cpu_to_le64(time(0));
2046 rv = 0;
2047 for (i = 0; i < dests; i++) {
2048 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
2049 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
2050 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
2051 bsb.sb_csum2 = bsb_csum((char*)&bsb,
2052 ((char*)&bsb.sb_csum2)-((char*)&bsb));
2053 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
2054 destoffsets[i]-4096)
2055 rv = -1;
2056 if (rv == 0 &&
2057 write(destfd[i], &bsb, 512) != 512)
2058 rv = -1;
2059 fsync(destfd[i]);
2060 }
2061 return rv;
2062 }
2063
2064 static void fail(char *msg)
2065 {
2066 int rv;
2067 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
2068 rv |= (write(2, "\n", 1) != 1);
2069 exit(rv ? 1 : 2);
2070 }
2071
2072 static char *abuf, *bbuf;
2073 static unsigned long long abuflen;
2074 static void validate(int afd, int bfd, unsigned long long offset)
2075 {
2076 /* check that the data in the backup against the array.
2077 * This is only used for regression testing and should not
2078 * be used while the array is active
2079 */
2080 if (afd < 0)
2081 return;
2082 lseek64(bfd, offset - 4096, 0);
2083 if (read(bfd, &bsb2, 512) != 512)
2084 fail("cannot read bsb");
2085 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
2086 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
2087 fail("first csum bad");
2088 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
2089 fail("magic is bad");
2090 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
2091 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
2092 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
2093 fail("second csum bad");
2094
2095 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
2096 fail("devstart is wrong");
2097
2098 if (bsb2.length) {
2099 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
2100
2101 if (abuflen < len) {
2102 free(abuf);
2103 free(bbuf);
2104 abuflen = len;
2105 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
2106 posix_memalign((void**)&bbuf, 4096, abuflen)) {
2107 abuflen = 0;
2108 /* just stop validating on mem-alloc failure */
2109 return;
2110 }
2111 }
2112
2113 lseek64(bfd, offset, 0);
2114 if ((unsigned long long)read(bfd, bbuf, len) != len) {
2115 //printf("len %llu\n", len);
2116 fail("read first backup failed");
2117 }
2118 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
2119 if ((unsigned long long)read(afd, abuf, len) != len)
2120 fail("read first from array failed");
2121 if (memcmp(bbuf, abuf, len) != 0) {
2122 #if 0
2123 int i;
2124 printf("offset=%llu len=%llu\n",
2125 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
2126 for (i=0; i<len; i++)
2127 if (bbuf[i] != abuf[i]) {
2128 printf("first diff byte %d\n", i);
2129 break;
2130 }
2131 #endif
2132 fail("data1 compare failed");
2133 }
2134 }
2135 if (bsb2.length2) {
2136 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
2137
2138 if (abuflen < len) {
2139 free(abuf);
2140 free(bbuf);
2141 abuflen = len;
2142 abuf = malloc(abuflen);
2143 bbuf = malloc(abuflen);
2144 }
2145
2146 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
2147 if ((unsigned long long)read(bfd, bbuf, len) != len)
2148 fail("read second backup failed");
2149 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
2150 if ((unsigned long long)read(afd, abuf, len) != len)
2151 fail("read second from array failed");
2152 if (memcmp(bbuf, abuf, len) != 0)
2153 fail("data2 compare failed");
2154 }
2155 }
2156
2157 static int child_grow(int afd, struct mdinfo *sra, unsigned long stripes,
2158 int *fds, unsigned long long *offsets,
2159 int disks, int chunk, int level, int layout, int data,
2160 int dests, int *destfd, unsigned long long *destoffsets)
2161 {
2162 char *buf;
2163 int degraded = 0;
2164
2165 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2166 /* Don't start the 'reshape' */
2167 return 0;
2168 sysfs_set_num(sra, NULL, "suspend_hi", 0);
2169 sysfs_set_num(sra, NULL, "suspend_lo", 0);
2170 grow_backup(sra, 0, stripes,
2171 fds, offsets, disks, chunk, level, layout,
2172 dests, destfd, destoffsets,
2173 0, &degraded, buf);
2174 validate(afd, destfd[0], destoffsets[0]);
2175 wait_backup(sra, 0, stripes * (chunk / 512), stripes * (chunk / 512),
2176 dests, destfd, destoffsets,
2177 0);
2178 sysfs_set_num(sra, NULL, "suspend_lo", (stripes * (chunk/512)) * data);
2179 free(buf);
2180 /* FIXME this should probably be numeric */
2181 sysfs_set_str(sra, NULL, "sync_max", "max");
2182 return 1;
2183 }
2184
2185 static int child_shrink(int afd, struct mdinfo *sra, unsigned long stripes,
2186 int *fds, unsigned long long *offsets,
2187 int disks, int chunk, int level, int layout, int data,
2188 int dests, int *destfd, unsigned long long *destoffsets)
2189 {
2190 char *buf;
2191 unsigned long long start;
2192 int rv;
2193 int degraded = 0;
2194
2195 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2196 return 0;
2197 start = sra->component_size - stripes * (chunk/512);
2198 sysfs_set_num(sra, NULL, "sync_max", start);
2199 sysfs_set_str(sra, NULL, "sync_action", "reshape");
2200 sysfs_set_num(sra, NULL, "suspend_lo", 0);
2201 sysfs_set_num(sra, NULL, "suspend_hi", 0);
2202 rv = wait_backup(sra, 0, start - stripes * (chunk/512), stripes * (chunk/512),
2203 dests, destfd, destoffsets, 0);
2204 if (rv < 0)
2205 return 0;
2206 grow_backup(sra, 0, stripes,
2207 fds, offsets,
2208 disks, chunk, level, layout,
2209 dests, destfd, destoffsets,
2210 0, &degraded, buf);
2211 validate(afd, destfd[0], destoffsets[0]);
2212 wait_backup(sra, start, stripes*(chunk/512), 0,
2213 dests, destfd, destoffsets, 0);
2214 sysfs_set_num(sra, NULL, "suspend_lo", (stripes * (chunk/512)) * data);
2215 free(buf);
2216 /* FIXME this should probably be numeric */
2217 sysfs_set_str(sra, NULL, "sync_max", "max");
2218 return 1;
2219 }
2220
2221 static int child_same_size(int afd, struct mdinfo *sra, unsigned long stripes,
2222 int *fds, unsigned long long *offsets,
2223 unsigned long long start,
2224 int disks, int chunk, int level, int layout, int data,
2225 int dests, int *destfd, unsigned long long *destoffsets)
2226 {
2227 unsigned long long size;
2228 unsigned long tailstripes = stripes;
2229 int part;
2230 char *buf;
2231 unsigned long long speed;
2232 int degraded = 0;
2233
2234
2235 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2236 return 0;
2237
2238 sysfs_set_num(sra, NULL, "suspend_lo", 0);
2239 sysfs_set_num(sra, NULL, "suspend_hi", 0);
2240
2241 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
2242 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
2243
2244 grow_backup(sra, start, stripes,
2245 fds, offsets,
2246 disks, chunk, level, layout,
2247 dests, destfd, destoffsets,
2248 0, &degraded, buf);
2249 grow_backup(sra, (start + stripes) * (chunk/512), stripes,
2250 fds, offsets,
2251 disks, chunk, level, layout,
2252 dests, destfd, destoffsets,
2253 1, &degraded, buf);
2254 validate(afd, destfd[0], destoffsets[0]);
2255 part = 0;
2256 start += stripes * 2; /* where to read next */
2257 size = sra->component_size / (chunk/512);
2258 while (start < size) {
2259 if (wait_backup(sra, (start-stripes*2)*(chunk/512),
2260 stripes*(chunk/512), 0,
2261 dests, destfd, destoffsets,
2262 part) < 0)
2263 return 0;
2264 sysfs_set_num(sra, NULL, "suspend_lo", start*(chunk/512) * data);
2265 if (start + stripes > size)
2266 tailstripes = (size - start);
2267
2268 grow_backup(sra, start*(chunk/512), tailstripes,
2269 fds, offsets,
2270 disks, chunk, level, layout,
2271 dests, destfd, destoffsets,
2272 part, &degraded, buf);
2273 start += stripes;
2274 part = 1 - part;
2275 validate(afd, destfd[0], destoffsets[0]);
2276 }
2277 if (wait_backup(sra, (start-stripes*2) * (chunk/512), stripes * (chunk/512), 0,
2278 dests, destfd, destoffsets,
2279 part) < 0)
2280 return 0;
2281 sysfs_set_num(sra, NULL, "suspend_lo", ((start-stripes)*(chunk/512)) * data);
2282 wait_backup(sra, (start-stripes) * (chunk/512), tailstripes * (chunk/512), 0,
2283 dests, destfd, destoffsets,
2284 1-part);
2285 sysfs_set_num(sra, NULL, "suspend_lo", (size*(chunk/512)) * data);
2286 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
2287 free(buf);
2288 return 1;
2289 }
2290
2291 /*
2292 * If any spare contains md_back_data-1 which is recent wrt mtime,
2293 * write that data into the array and update the super blocks with
2294 * the new reshape_progress
2295 */
2296 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
2297 char *backup_file, int verbose)
2298 {
2299 int i, j;
2300 int old_disks;
2301 unsigned long long *offsets;
2302 unsigned long long nstripe, ostripe;
2303 int ndata, odata;
2304
2305 if (info->new_level != info->array.level)
2306 return 1; /* Cannot handle level changes (they are instantaneous) */
2307
2308 odata = info->array.raid_disks - info->delta_disks - 1;
2309 if (info->array.level == 6) odata--; /* number of data disks */
2310 ndata = info->array.raid_disks - 1;
2311 if (info->new_level == 6) ndata--;
2312
2313 old_disks = info->array.raid_disks - info->delta_disks;
2314
2315 if (info->delta_disks <= 0)
2316 /* Didn't grow, so the backup file must have
2317 * been used
2318 */
2319 old_disks = cnt;
2320 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
2321 struct mdinfo dinfo;
2322 int fd;
2323 int bsbsize;
2324 char *devname, namebuf[20];
2325
2326 /* This was a spare and may have some saved data on it.
2327 * Load the superblock, find and load the
2328 * backup_super_block.
2329 * If either fail, go on to next device.
2330 * If the backup contains no new info, just return
2331 * else restore data and update all superblocks
2332 */
2333 if (i == old_disks-1) {
2334 fd = open(backup_file, O_RDONLY);
2335 if (fd<0) {
2336 fprintf(stderr, Name ": backup file %s inaccessible: %s\n",
2337 backup_file, strerror(errno));
2338 continue;
2339 }
2340 devname = backup_file;
2341 } else {
2342 fd = fdlist[i];
2343 if (fd < 0)
2344 continue;
2345 if (st->ss->load_super(st, fd, NULL))
2346 continue;
2347
2348 st->ss->getinfo_super(st, &dinfo, NULL);
2349 st->ss->free_super(st);
2350
2351 if (lseek64(fd,
2352 (dinfo.data_offset + dinfo.component_size - 8) <<9,
2353 0) < 0) {
2354 fprintf(stderr, Name ": Cannot seek on device %d\n", i);
2355 continue; /* Cannot seek */
2356 }
2357 sprintf(namebuf, "device-%d", i);
2358 devname = namebuf;
2359 }
2360 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
2361 if (verbose)
2362 fprintf(stderr, Name ": Cannot read from %s\n", devname);
2363 continue; /* Cannot read */
2364 }
2365 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
2366 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
2367 if (verbose)
2368 fprintf(stderr, Name ": No backup metadata on %s\n", devname);
2369 continue;
2370 }
2371 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
2372 if (verbose)
2373 fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname);
2374 continue; /* bad checksum */
2375 }
2376 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
2377 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
2378 if (verbose)
2379 fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname);
2380 continue; /* Bad second checksum */
2381 }
2382 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
2383 if (verbose)
2384 fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname);
2385 continue; /* Wrong uuid */
2386 }
2387
2388 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2389 * sometimes they aren't... So allow considerable flexability in matching, and allow
2390 * this test to be overridden by an environment variable.
2391 */
2392 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
2393 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
2394 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2395 fprintf(stderr, Name ": accepting backup with timestamp %lu "
2396 "for array with timestamp %lu\n",
2397 (unsigned long)__le64_to_cpu(bsb.mtime),
2398 (unsigned long)info->array.utime);
2399 } else {
2400 if (verbose)
2401 fprintf(stderr, Name ": too-old timestamp on "
2402 "backup-metadata on %s\n", devname);
2403 continue; /* time stamp is too bad */
2404 }
2405 }
2406
2407 if (bsb.magic[15] == '1') {
2408 if (info->delta_disks >= 0) {
2409 /* reshape_progress is increasing */
2410 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2411 info->reshape_progress) {
2412 nonew:
2413 if (verbose)
2414 fprintf(stderr, Name ": backup-metadata found on %s but is not needed\n", devname);
2415 continue; /* No new data here */
2416 }
2417 } else {
2418 /* reshape_progress is decreasing */
2419 if (__le64_to_cpu(bsb.arraystart) >=
2420 info->reshape_progress)
2421 goto nonew; /* No new data here */
2422 }
2423 } else {
2424 if (info->delta_disks >= 0) {
2425 /* reshape_progress is increasing */
2426 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2427 info->reshape_progress &&
2428 __le64_to_cpu(bsb.arraystart2) + __le64_to_cpu(bsb.length2) <
2429 info->reshape_progress)
2430 goto nonew; /* No new data here */
2431 } else {
2432 /* reshape_progress is decreasing */
2433 if (__le64_to_cpu(bsb.arraystart) >=
2434 info->reshape_progress &&
2435 __le64_to_cpu(bsb.arraystart2) >=
2436 info->reshape_progress)
2437 goto nonew; /* No new data here */
2438 }
2439 }
2440 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
2441 second_fail:
2442 if (verbose)
2443 fprintf(stderr, Name ": Failed to verify secondary backup-metadata block on %s\n",
2444 devname);
2445 continue; /* Cannot seek */
2446 }
2447 /* There should be a duplicate backup superblock 4k before here */
2448 if (lseek64(fd, -4096, 1) < 0 ||
2449 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
2450 goto second_fail; /* Cannot find leading superblock */
2451 if (bsb.magic[15] == '1')
2452 bsbsize = offsetof(struct mdp_backup_super, pad1);
2453 else
2454 bsbsize = offsetof(struct mdp_backup_super, pad);
2455 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
2456 goto second_fail; /* Cannot find leading superblock */
2457
2458 /* Now need the data offsets for all devices. */
2459 offsets = malloc(sizeof(*offsets)*info->array.raid_disks);
2460 for(j=0; j<info->array.raid_disks; j++) {
2461 if (fdlist[j] < 0)
2462 continue;
2463 if (st->ss->load_super(st, fdlist[j], NULL))
2464 /* FIXME should be this be an error */
2465 continue;
2466 st->ss->getinfo_super(st, &dinfo, NULL);
2467 st->ss->free_super(st);
2468 offsets[j] = dinfo.data_offset * 512;
2469 }
2470 printf(Name ": restoring critical section\n");
2471
2472 if (restore_stripes(fdlist, offsets,
2473 info->array.raid_disks,
2474 info->new_chunk,
2475 info->new_level,
2476 info->new_layout,
2477 fd, __le64_to_cpu(bsb.devstart)*512,
2478 __le64_to_cpu(bsb.arraystart)*512,
2479 __le64_to_cpu(bsb.length)*512)) {
2480 /* didn't succeed, so giveup */
2481 if (verbose)
2482 fprintf(stderr, Name ": Error restoring backup from %s\n",
2483 devname);
2484 return 1;
2485 }
2486
2487 if (bsb.magic[15] == '2' &&
2488 restore_stripes(fdlist, offsets,
2489 info->array.raid_disks,
2490 info->new_chunk,
2491 info->new_level,
2492 info->new_layout,
2493 fd, __le64_to_cpu(bsb.devstart)*512 +
2494 __le64_to_cpu(bsb.devstart2)*512,
2495 __le64_to_cpu(bsb.arraystart2)*512,
2496 __le64_to_cpu(bsb.length2)*512)) {
2497 /* didn't succeed, so giveup */
2498 if (verbose)
2499 fprintf(stderr, Name ": Error restoring second backup from %s\n",
2500 devname);
2501 return 1;
2502 }
2503
2504
2505 /* Ok, so the data is restored. Let's update those superblocks. */
2506
2507 if (info->delta_disks >= 0) {
2508 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
2509 __le64_to_cpu(bsb.length);
2510 if (bsb.magic[15] == '2') {
2511 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
2512 __le64_to_cpu(bsb.length2);
2513 if (p2 > info->reshape_progress)
2514 info->reshape_progress = p2;
2515 }
2516 } else {
2517 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
2518 if (bsb.magic[15] == '2') {
2519 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
2520 if (p2 < info->reshape_progress)
2521 info->reshape_progress = p2;
2522 }
2523 }
2524 for (j=0; j<info->array.raid_disks; j++) {
2525 if (fdlist[j] < 0) continue;
2526 if (st->ss->load_super(st, fdlist[j], NULL))
2527 continue;
2528 st->ss->getinfo_super(st, &dinfo, NULL);
2529 dinfo.reshape_progress = info->reshape_progress;
2530 st->ss->update_super(st, &dinfo,
2531 "_reshape_progress",
2532 NULL,0, 0, NULL);
2533 st->ss->store_super(st, fdlist[j]);
2534 st->ss->free_super(st);
2535 }
2536 return 0;
2537 }
2538 /* Didn't find any backup data, try to see if any
2539 * was needed.
2540 */
2541 if (info->delta_disks < 0) {
2542 /* When shrinking, the critical section is at the end.
2543 * So see if we are before the critical section.
2544 */
2545 unsigned long long first_block;
2546 nstripe = ostripe = 0;
2547 first_block = 0;
2548 while (ostripe >= nstripe) {
2549 ostripe += info->array.chunk_size / 512;
2550 first_block = ostripe * odata;
2551 nstripe = first_block / ndata / (info->new_chunk/512) *
2552 (info->new_chunk/512);
2553 }
2554
2555 if (info->reshape_progress >= first_block)
2556 return 0;
2557 }
2558 if (info->delta_disks > 0) {
2559 /* See if we are beyond the critical section. */
2560 unsigned long long last_block;
2561 nstripe = ostripe = 0;
2562 last_block = 0;
2563 while (nstripe >= ostripe) {
2564 nstripe += info->new_chunk / 512;
2565 last_block = nstripe * ndata;
2566 ostripe = last_block / odata / (info->array.chunk_size/512) *
2567 (info->array.chunk_size/512);
2568 }
2569
2570 if (info->reshape_progress >= last_block)
2571 return 0;
2572 }
2573 /* needed to recover critical section! */
2574 if (verbose)
2575 fprintf(stderr, Name ": Failed to find backup of critical section\n");
2576 return 1;
2577 }
2578
2579 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
2580 char *backup_file)
2581 {
2582 /* Array is assembled and ready to be started, but
2583 * monitoring is probably required.
2584 * So:
2585 * - start read-only
2586 * - set upper bound for resync
2587 * - initialise the 'suspend' boundaries
2588 * - switch to read-write
2589 * - fork and continue monitoring
2590 */
2591 int err;
2592 int backup_list[1];
2593 unsigned long long backup_offsets[1];
2594 int odisks, ndisks, ochunk, nchunk,odata,ndata;
2595 unsigned long a,b,blocks,stripes;
2596 int backup_fd;
2597 int *fds;
2598 unsigned long long *offsets;
2599 int d;
2600 struct mdinfo *sra, *sd;
2601 int rv;
2602 unsigned long cache;
2603 int done = 0;
2604
2605 err = sysfs_set_str(info, NULL, "array_state", "readonly");
2606 if (err)
2607 return err;
2608
2609 /* make sure reshape doesn't progress until we are ready */
2610 sysfs_set_str(info, NULL, "sync_max", "0");
2611 sysfs_set_str(info, NULL, "array_state", "active"); /* FIXME or clean */
2612
2613 sra = sysfs_read(-1, devname2devnum(info->sys_name),
2614 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|
2615 GET_CACHE);
2616 if (!sra)
2617 return 1;
2618
2619 /* ndisks is not growing, so raid_disks is old and +delta is new */
2620 odisks = info->array.raid_disks;
2621 ndisks = odisks + info->delta_disks;
2622 odata = odisks - 1;
2623 ndata = ndisks - 1;
2624 if (info->array.level == 6) {
2625 odata--;
2626 ndata--;
2627 }
2628 ochunk = info->array.chunk_size;
2629 nchunk = info->new_chunk;
2630
2631 a = (ochunk/512) * odata;
2632 b = (nchunk/512) * ndata;
2633 /* Find GCD */
2634 while (a != b) {
2635 if (a < b)
2636 b -= a;
2637 if (b < a)
2638 a -= b;
2639 }
2640 /* LCM == product / GCD */
2641 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
2642
2643 if (ndata == odata)
2644 while (blocks * 32 < sra->component_size &&
2645 blocks < 16*1024*2)
2646 blocks *= 2;
2647 stripes = blocks / (info->array.chunk_size/512) / odata;
2648
2649 /* check that the internal stripe cache is
2650 * large enough, or it won't work.
2651 */
2652 cache = (nchunk < ochunk) ? ochunk : nchunk;
2653 cache = cache * 4 / 4096;
2654 if (cache < blocks / 8 / odisks + 16)
2655 /* Make it big enough to hold 'blocks' */
2656 cache = blocks / 8 / odisks + 16;
2657 if (sra->cache_size < cache)
2658 sysfs_set_num(sra, NULL, "stripe_cache_size",
2659 cache+1);
2660
2661 memset(&bsb, 0, 512);
2662 memcpy(bsb.magic, "md_backup_data-1", 16);
2663 memcpy(&bsb.set_uuid, info->uuid, 16);
2664 bsb.mtime = __cpu_to_le64(time(0));
2665 bsb.devstart2 = blocks;
2666
2667 backup_fd = open(backup_file, O_RDWR|O_CREAT, S_IRUSR | S_IWUSR);
2668 if (backup_fd < 0) {
2669 fprintf(stderr, Name ": Cannot open backup file %s\n",
2670 backup_file ?: "- no backup-file given");
2671 return 1;
2672 }
2673 backup_list[0] = backup_fd;
2674 backup_offsets[0] = 8 * 512;
2675 fds = malloc(odisks * sizeof(fds[0]));
2676 offsets = malloc(odisks * sizeof(offsets[0]));
2677 for (d=0; d<odisks; d++)
2678 fds[d] = -1;
2679
2680 for (sd = sra->devs; sd; sd = sd->next) {
2681 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2682 continue;
2683 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
2684 char *dn = map_dev(sd->disk.major,
2685 sd->disk.minor, 1);
2686 fds[sd->disk.raid_disk]
2687 = dev_open(dn, O_RDONLY);
2688 offsets[sd->disk.raid_disk] = sd->data_offset*512;
2689 if (fds[sd->disk.raid_disk] < 0) {
2690 fprintf(stderr, Name ": %s: cannot open component %s\n",
2691 info->sys_name, dn?dn:"-unknown-");
2692 rv = 1;
2693 goto release;
2694 }
2695 free(dn);
2696 }
2697 }
2698
2699 switch(fork()) {
2700 case 0:
2701 close(mdfd);
2702 mlockall(MCL_FUTURE);
2703 if (info->delta_disks < 0)
2704 done = child_shrink(-1, info, stripes,
2705 fds, offsets,
2706 info->array.raid_disks,
2707 info->array.chunk_size,
2708 info->array.level, info->array.layout,
2709 odata,
2710 1, backup_list, backup_offsets);
2711 else if (info->delta_disks == 0) {
2712 /* The 'start' is a per-device stripe number.
2713 * reshape_progress is a per-array sector number.
2714 * So divide by ndata * chunk_size
2715 */
2716 unsigned long long start = info->reshape_progress / ndata;
2717 start /= (info->array.chunk_size/512);
2718 done = child_same_size(-1, info, stripes,
2719 fds, offsets,
2720 start,
2721 info->array.raid_disks,
2722 info->array.chunk_size,
2723 info->array.level, info->array.layout,
2724 odata,
2725 1, backup_list, backup_offsets);
2726 }
2727 if (backup_file && done)
2728 unlink(backup_file);
2729 /* FIXME should I intuit a level change */
2730 exit(0);
2731 case -1:
2732 fprintf(stderr, Name ": Cannot run child to continue monitoring reshape: %s\n",
2733 strerror(errno));
2734 return 1;
2735 default:
2736 break;
2737 }
2738 release:
2739 return 0;
2740 }
2741
2742
Unnamed repository; edit this file 'description' to name the repository.