Disk removal support for Raid10->Raid0 takeover
[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 #ifdef __BIG_ENDIAN
232 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
233 " between different architectured. Consider upgrading the Linux kernel.\n");
234 #endif
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 fprintf(stderr, Name ": failed to set internal bitmap.\n");
357 return 1;
358 }
359 } else {
360 int uuid[4];
361 int bitmap_fd;
362 int d;
363 int max_devs = st->max_devs;
364
365 /* try to load a superblock */
366 for (d=0; d<max_devs; d++) {
367 mdu_disk_info_t disk;
368 char *dv;
369 int fd2;
370 disk.number = d;
371 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
372 continue;
373 if ((disk.major==0 && disk.minor==0) ||
374 (disk.state & (1<<MD_DISK_REMOVED)))
375 continue;
376 dv = map_dev(disk.major, disk.minor, 1);
377 if (!dv) continue;
378 fd2 = dev_open(dv, O_RDONLY);
379 if (fd2 >= 0 &&
380 st->ss->load_super(st, fd2, NULL) == 0) {
381 close(fd2);
382 st->ss->uuid_from_super(st, uuid);
383 break;
384 }
385 close(fd2);
386 }
387 if (d == max_devs) {
388 fprintf(stderr, Name ": cannot find UUID for array!\n");
389 return 1;
390 }
391 if (CreateBitmap(file, force, (char*)uuid, chunk,
392 delay, write_behind, bitmapsize, major)) {
393 return 1;
394 }
395 bitmap_fd = open(file, O_RDWR);
396 if (bitmap_fd < 0) {
397 fprintf(stderr, Name ": weird: %s cannot be opened\n",
398 file);
399 return 1;
400 }
401 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
402 fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
403 devname, strerror(errno));
404 return 1;
405 }
406 }
407
408 return 0;
409 }
410
411
412 /*
413 * When reshaping an array we might need to backup some data.
414 * This is written to all spares with a 'super_block' describing it.
415 * The superblock goes 4K from the end of the used space on the
416 * device.
417 * It if written after the backup is complete.
418 * It has the following structure.
419 */
420
421 static struct mdp_backup_super {
422 char magic[16]; /* md_backup_data-1 or -2 */
423 __u8 set_uuid[16];
424 __u64 mtime;
425 /* start/sizes in 512byte sectors */
426 __u64 devstart; /* address on backup device/file of data */
427 __u64 arraystart;
428 __u64 length;
429 __u32 sb_csum; /* csum of preceeding bytes. */
430 __u32 pad1;
431 __u64 devstart2; /* offset in to data of second section */
432 __u64 arraystart2;
433 __u64 length2;
434 __u32 sb_csum2; /* csum of preceeding bytes. */
435 __u8 pad[512-68-32];
436 } __attribute__((aligned(512))) bsb, bsb2;
437
438 static __u32 bsb_csum(char *buf, int len)
439 {
440 int i;
441 int csum = 0;
442 for (i=0; i<len; i++)
443 csum = (csum<<3) + buf[0];
444 return __cpu_to_le32(csum);
445 }
446
447 static int child_grow(int afd, struct mdinfo *sra, unsigned long blocks,
448 int *fds, unsigned long long *offsets,
449 int disks, int chunk, int level, int layout, int data,
450 int dests, int *destfd, unsigned long long *destoffsets);
451 static int child_shrink(int afd, struct mdinfo *sra, unsigned long blocks,
452 int *fds, unsigned long long *offsets,
453 int disks, int chunk, int level, int layout, int data,
454 int dests, int *destfd, unsigned long long *destoffsets);
455 static int child_same_size(int afd, struct mdinfo *sra, unsigned long blocks,
456 int *fds, unsigned long long *offsets,
457 unsigned long long start,
458 int disks, int chunk, int level, int layout, int data,
459 int dests, int *destfd, unsigned long long *destoffsets);
460
461 static int freeze_container(struct supertype *st)
462 {
463 int container_dev = (st->container_dev != NoMdDev
464 ? st->container_dev : st->devnum);
465 char *container = devnum2devname(container_dev);
466
467 if (!container) {
468 fprintf(stderr, Name
469 ": could not determine container name, freeze aborted\n");
470 return -2;
471 }
472
473 if (block_monitor(container, 1)) {
474 fprintf(stderr, Name ": failed to freeze container\n");
475 return -2;
476 }
477
478 return 1;
479 }
480
481 static void unfreeze_container(struct supertype *st)
482 {
483 int container_dev = (st->container_dev != NoMdDev
484 ? st->container_dev : st->devnum);
485 char *container = devnum2devname(container_dev);
486
487 if (!container) {
488 fprintf(stderr, Name
489 ": could not determine container name, unfreeze aborted\n");
490 return;
491 }
492
493 unblock_monitor(container, 1);
494 }
495
496 static int freeze(struct supertype *st)
497 {
498 /* Try to freeze resync/rebuild on this array/container.
499 * Return -1 if the array is busy,
500 * return -2 container cannot be frozen,
501 * return 0 if this kernel doesn't support 'frozen'
502 * return 1 if it worked.
503 */
504 if (st->ss->external)
505 return freeze_container(st);
506 else {
507 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
508 int err;
509
510 if (!sra)
511 return -1;
512 err = sysfs_freeze_array(sra);
513 sysfs_free(sra);
514 return err;
515 }
516 }
517
518 static void unfreeze(struct supertype *st, int frozen)
519 {
520 /* If 'frozen' is 1, unfreeze the array */
521 if (frozen <= 0)
522 return;
523
524 if (st->ss->external)
525 return unfreeze_container(st);
526 else {
527 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
528
529 if (sra)
530 sysfs_set_str(sra, NULL, "sync_action", "idle");
531 else
532 fprintf(stderr, Name ": failed to unfreeze array\n");
533 sysfs_free(sra);
534 }
535 }
536
537 static void wait_reshape(struct mdinfo *sra)
538 {
539 int fd = sysfs_get_fd(sra, NULL, "sync_action");
540 char action[20];
541
542 do {
543 fd_set rfds;
544 FD_ZERO(&rfds);
545 FD_SET(fd, &rfds);
546 select(fd+1, NULL, NULL, &rfds, NULL);
547
548 if (sysfs_fd_get_str(fd, action, 20) < 0) {
549 close(fd);
550 return;
551 }
552 } while (strncmp(action, "reshape", 7) == 0);
553 }
554
555 static int reshape_super(struct supertype *st, long long size, int level,
556 int layout, int chunksize, int raid_disks,
557 char *backup_file, char *dev, int verbose)
558 {
559 /* nothing extra to check in the native case */
560 if (!st->ss->external)
561 return 0;
562 if (!st->ss->reshape_super ||
563 !st->ss->manage_reshape) {
564 fprintf(stderr, Name ": %s metadata does not support reshape\n",
565 st->ss->name);
566 return 1;
567 }
568
569 return st->ss->reshape_super(st, size, level, layout, chunksize,
570 raid_disks, backup_file, dev, verbose);
571 }
572
573 static void sync_metadata(struct supertype *st)
574 {
575 if (st->ss->external) {
576 if (st->update_tail)
577 flush_metadata_updates(st);
578 else
579 st->ss->sync_metadata(st);
580 }
581 }
582
583 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
584 {
585 /* when dealing with external metadata subarrays we need to be
586 * prepared to handle EAGAIN. The kernel may need to wait for
587 * mdmon to mark the array active so the kernel can handle
588 * allocations/writeback when preparing the reshape action
589 * (md_allow_write()). We temporarily disable safe_mode_delay
590 * to close a race with the array_state going clean before the
591 * next write to raid_disks / stripe_cache_size
592 */
593 char safe[50];
594 int rc;
595
596 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
597 if (strcmp(name, "raid_disks") != 0 &&
598 strcmp(name, "stripe_cache_size") != 0)
599 return sysfs_set_num(sra, NULL, name, n);
600
601 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
602 if (rc <= 0)
603 return -1;
604 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
605 rc = sysfs_set_num(sra, NULL, name, n);
606 if (rc < 0 && errno == EAGAIN) {
607 ping_monitor(container);
608 /* if we get EAGAIN here then the monitor is not active
609 * so stop trying
610 */
611 rc = sysfs_set_num(sra, NULL, name, n);
612 }
613 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
614 return rc;
615 }
616
617 static int reshape_container_raid_disks(char *container, int raid_disks)
618 {
619 /* for each subarray switch to a raid level that can
620 * support the reshape, and set raid disks
621 */
622 struct mdstat_ent *ent, *e;
623 int changed = 0, rv = 0, err = 0;
624
625 ent = mdstat_read(1, 0);
626 if (!ent) {
627 fprintf(stderr, Name ": unable to read /proc/mdstat\n");
628 return -1;
629 }
630
631 changed = 0;
632 for (e = ent; e; e = e->next) {
633 struct mdinfo *sub;
634 unsigned int cache;
635 int level, takeover_delta = 0;
636
637 if (!is_container_member(e, container))
638 continue;
639
640 level = map_name(pers, e->level);
641 if (level == 0) {
642 sub = sysfs_read(-1, e->devnum, GET_VERSION);
643 if (!sub)
644 break;
645 /* metadata records 'orig_level' */
646 rv = sysfs_set_num(sub, NULL, "level", 4);
647 if (rv < 0) {
648 err = errno;
649 break;
650 }
651 /* we want spares to be used for capacity
652 * expansion, not rebuild
653 */
654 takeover_delta = 1;
655
656 sysfs_free(sub);
657 level = 4;
658 }
659
660 sub = NULL;
661 switch (level) {
662 default:
663 rv = -1;
664 break;
665 case 4:
666 case 5:
667 case 6:
668 sub = sysfs_read(-1, e->devnum, GET_CHUNK|GET_CACHE);
669 if (!sub)
670 break;
671 cache = (sub->array.chunk_size / 4096) * 4;
672 if (cache > sub->cache_size)
673 rv = subarray_set_num(container, sub,
674 "stripe_cache_size", cache);
675 if (rv) {
676 err = errno;
677 break;
678 }
679 /* fall through */
680 case 1:
681 if (!sub)
682 sub = sysfs_read(-1, e->devnum, GET_VERSION);
683 if (!sub)
684 break;
685
686 rv = subarray_set_num(container, sub, "raid_disks",
687 raid_disks + takeover_delta);
688 if (rv)
689 err = errno;
690 else
691 changed++;
692 break;
693 }
694 sysfs_free(sub);
695 if (rv)
696 break;
697 }
698 free_mdstat(ent);
699 if (rv) {
700 fprintf(stderr, Name
701 ": failed to initiate container reshape%s%s\n",
702 err ? ": " : "", err ? strerror(err) : "");
703 return rv;
704 }
705
706 return changed;
707 }
708
709 static void revert_container_raid_disks(struct supertype *st, int fd, char *container)
710 {
711 /* we failed to prepare all subarrays in the container for
712 * reshape, so cancel the changes and restore the nominal raid
713 * level
714 */
715 struct mdstat_ent *ent, *e;
716
717 ent = mdstat_read(0, 0);
718 if (!ent) {
719 fprintf(stderr, Name
720 ": failed to read /proc/mdstat while aborting reshape\n");
721 return;
722 }
723
724 if (st->ss->load_container(st, fd, NULL)) {
725 fprintf(stderr, Name
726 ": failed read metadata while aborting reshape\n");
727 return ;
728 }
729
730
731 for (e = ent; e; e = e->next) {
732 int level_fixed = 0, disks_fixed = 0;
733 struct mdinfo *sub, *prev;
734 char *subarray;
735
736 if (!is_container_member(e, container))
737 continue;
738
739 subarray = to_subarray(e, container);
740 prev = st->ss->container_content(st, subarray);
741
742 /* changing level might change raid_disks so we do it
743 * first and then check if raid_disks still needs fixing
744 */
745 if (map_name(pers, e->level) != prev->array.level) {
746 sub = sysfs_read(-1, e->devnum, GET_VERSION);
747 if (sub &&
748 !sysfs_set_num(sub, NULL, "level", prev->array.level))
749 level_fixed = 1;
750 sysfs_free(sub);
751 } else
752 level_fixed = 1;
753
754 sub = sysfs_read(-1, e->devnum, GET_DISKS);
755 if (sub && sub->array.raid_disks != prev->array.raid_disks) {
756 if (!subarray_set_num(container, sub, "raid_disks",
757 prev->array.raid_disks))
758 disks_fixed = 1;
759 } else if (sub)
760 disks_fixed = 1;
761 sysfs_free(sub);
762
763 if (!disks_fixed || !level_fixed)
764 fprintf(stderr, Name
765 ": failed to restore %s to a %d-disk %s array\n",
766 e->dev, prev->array.raid_disks,
767 map_num(pers, prev->array.level));
768 free(prev);
769 }
770 st->ss->free_super(st);
771 free_mdstat(ent);
772 }
773
774 int remove_disks_on_raid10_to_raid0_takeover(struct supertype *st,
775 struct mdinfo *sra,
776 int layout)
777 {
778 int nr_of_copies;
779 struct mdinfo *remaining;
780 int slot;
781
782 nr_of_copies = layout & 0xff;
783
784 remaining = sra->devs;
785 sra->devs = NULL;
786 /* for each 'copy', select one device and remove from the list. */
787 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
788 struct mdinfo **diskp;
789 int found = 0;
790
791 /* Find a working device to keep */
792 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
793 struct mdinfo *disk = *diskp;
794
795 if (disk->disk.raid_disk < slot)
796 continue;
797 if (disk->disk.raid_disk >= slot + nr_of_copies)
798 continue;
799 if (disk->disk.state & (1<<MD_DISK_REMOVED))
800 continue;
801 if (disk->disk.state & (1<<MD_DISK_FAULTY))
802 continue;
803 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
804 continue;
805
806 /* We have found a good disk to use! */
807 *diskp = disk->next;
808 disk->next = sra->devs;
809 sra->devs = disk;
810 found = 1;
811 break;
812 }
813 if (!found)
814 break;
815 }
816
817 if (slot < sra->array.raid_disks) {
818 /* didn't find all slots */
819 struct mdinfo **e;
820 e = &remaining;
821 while (*e)
822 e = &(*e)->next;
823 *e = sra->devs;
824 sra->devs = remaining;
825 return 1;
826 }
827
828 /* Remove all 'remaining' devices from the array */
829 while (remaining) {
830 struct mdinfo *sd = remaining;
831 remaining = sd->next;
832
833 sysfs_set_str(sra, sd, "state", "faulty");
834 sysfs_set_str(sra, sd, "slot", "none");
835 sysfs_set_str(sra, sd, "state", "remove");
836 sd->disk.state |= (1<<MD_DISK_REMOVED);
837 sd->disk.state &= ~(1<<MD_DISK_SYNC);
838 sd->next = sra->devs;
839 sra->devs = sd;
840 }
841 return 0;
842 }
843
844 int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
845 long long size,
846 int level, char *layout_str, int chunksize, int raid_disks)
847 {
848 /* Make some changes in the shape of an array.
849 * The kernel must support the change.
850 *
851 * There are three different changes. Each can trigger
852 * a resync or recovery so we freeze that until we have
853 * requested everything (if kernel supports freezing - 2.6.30).
854 * The steps are:
855 * - change size (i.e. component_size)
856 * - change level
857 * - change layout/chunksize/ndisks
858 *
859 * The last can require a reshape. It is different on different
860 * levels so we need to check the level before actioning it.
861 * Some times the level change needs to be requested after the
862 * reshape (e.g. raid6->raid5, raid5->raid0)
863 *
864 */
865 struct mdu_array_info_s array, orig;
866 char *c;
867 int rv = 0;
868 struct supertype *st;
869 char *subarray = NULL;
870
871 int nchunk, ochunk;
872 int nlayout, olayout;
873 int ndisks, odisks;
874 unsigned int ndata, odata;
875 int orig_level = UnSet;
876 char alt_layout[40];
877 int *fdlist;
878 unsigned long long *offsets;
879 int d, i;
880 int nrdisks;
881 int err;
882 int frozen;
883 unsigned long a,b, blocks, stripes;
884 unsigned long cache;
885 unsigned long long array_size;
886 int changed = 0;
887 char *container = NULL;
888 int cfd = -1;
889 int done;
890
891 struct mdinfo *sra;
892 struct mdinfo *sd;
893
894 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
895 fprintf(stderr, Name ": %s is not an active md array - aborting\n",
896 devname);
897 return 1;
898 }
899
900 if (size >= 0 &&
901 (chunksize || level!= UnSet || layout_str || raid_disks)) {
902 fprintf(stderr, Name ": cannot change component size at the same time "
903 "as other changes.\n"
904 " Change size first, then check data is intact before "
905 "making other changes.\n");
906 return 1;
907 }
908
909 if (raid_disks && raid_disks < array.raid_disks && array.level > 1 &&
910 get_linux_version() < 2006032 &&
911 !check_env("MDADM_FORCE_FEWER")) {
912 fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n"
913 " Please use a newer kernel\n");
914 return 1;
915 }
916
917 st = super_by_fd(fd, &subarray);
918 if (!st) {
919 fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname);
920 return 1;
921 }
922
923 /* in the external case we need to check that the requested reshape is
924 * supported, and perform an initial check that the container holds the
925 * pre-requisite spare devices (mdmon owns final validation)
926 */
927 if (st->ss->external) {
928 int container_dev;
929 int rv;
930
931 if (subarray) {
932 container_dev = st->container_dev;
933 cfd = open_dev_excl(st->container_dev);
934 } else if (size >= 0 || layout_str != NULL || chunksize != 0 ||
935 level != UnSet) {
936 fprintf(stderr,
937 Name ": %s is a container, only 'raid-devices' can be changed\n",
938 devname);
939 return 1;
940 } else {
941 container_dev = st->devnum;
942 close(fd);
943 cfd = open_dev_excl(st->devnum);
944 fd = cfd;
945 }
946 if (cfd < 0) {
947 fprintf(stderr, Name ": Unable to open container for %s\n",
948 devname);
949 free(subarray);
950 return 1;
951 }
952
953 container = devnum2devname(st->devnum);
954 if (!container) {
955 fprintf(stderr, Name ": Could not determine container name\n");
956 free(subarray);
957 return 1;
958 }
959
960 if (subarray)
961 rv = st->ss->load_container(st, cfd, NULL);
962 else
963 rv = st->ss->load_super(st, cfd, NULL);
964 if (rv) {
965 fprintf(stderr, Name ": Cannot read superblock for %s\n",
966 devname);
967 free(subarray);
968 return 1;
969 }
970
971 if (mdmon_running(container_dev))
972 st->update_tail = &st->updates;
973 }
974
975 sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS | GET_STATE);
976 if (sra) {
977 if (st->ss->external && subarray == NULL) {
978 array.level = LEVEL_CONTAINER;
979 sra->array.level = LEVEL_CONTAINER;
980 }
981 } else {
982 fprintf(stderr, Name ": failed to read sysfs parameters for %s\n",
983 devname);
984 return 1;
985 }
986 frozen = freeze(st);
987 if (frozen < -1) {
988 /* freeze() already spewed the reason */
989 return 1;
990 } else if (frozen < 0) {
991 fprintf(stderr, Name ": %s is performing resync/recovery and cannot"
992 " be reshaped\n", devname);
993 return 1;
994 }
995
996 /* ========= set size =============== */
997 if (size >= 0 && (size == 0 || size != array.size)) {
998 long long orig_size = array.size;
999
1000 if (reshape_super(st, size, UnSet, UnSet, 0, 0, NULL, devname, !quiet)) {
1001 rv = 1;
1002 goto release;
1003 }
1004 sync_metadata(st);
1005 array.size = size;
1006 if (array.size != size) {
1007 /* got truncated to 32bit, write to
1008 * component_size instead
1009 */
1010 if (sra)
1011 rv = sysfs_set_num(sra, NULL,
1012 "component_size", size);
1013 else
1014 rv = -1;
1015 } else
1016 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1017 if (rv != 0) {
1018 int err = errno;
1019
1020 /* restore metadata */
1021 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1022 NULL, devname, !quiet) == 0)
1023 sync_metadata(st);
1024 fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
1025 devname, strerror(err));
1026 if (err == EBUSY &&
1027 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1028 fprintf(stderr, " Bitmap must be removed before size can be changed\n");
1029 rv = 1;
1030 goto release;
1031 }
1032 ioctl(fd, GET_ARRAY_INFO, &array);
1033 size = get_component_size(fd)/2;
1034 if (size == 0)
1035 size = array.size;
1036 if (!quiet)
1037 fprintf(stderr, Name ": component size of %s has been set to %lluK\n",
1038 devname, size);
1039 changed = 1;
1040 } else if (array.level != LEVEL_CONTAINER) {
1041 size = get_component_size(fd)/2;
1042 if (size == 0)
1043 size = array.size;
1044 }
1045
1046 /* ========= check for Raid10 -> Raid0 conversion ===============
1047 * current implemenation assumes that following conditions must be met:
1048 * - far_copies == 1
1049 * - near_copies == 2
1050 */
1051 if (level == 0 && array.level == 10 &&
1052 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) {
1053 int err;
1054 err = remove_disks_on_raid10_to_raid0_takeover(st, sra, array.layout);
1055 if (err) {
1056 dprintf(Name": Array cannot be reshaped\n");
1057 if (container)
1058 free(container);
1059 if (cfd > -1)
1060 close(cfd);
1061 return 1;
1062 }
1063 }
1064
1065 /* ======= set level =========== */
1066 if (level != UnSet && level != array.level) {
1067 /* Trying to change the level.
1068 * We might need to change layout first and schedule a
1069 * level change for later.
1070 * Level changes that can happen immediately are:
1071 * 0->4,5,6 1->5 4->5,6 5->1,6
1072 * Level changes that need a layout change first are:
1073 * 6->5,4,0 : need a -6 layout, or parity-last
1074 * 5->4,0 : need parity-last
1075 */
1076 if ((array.level == 6 || array.level == 5) &&
1077 (level == 5 || level == 4 || level == 0)) {
1078 /* Don't change level yet, but choose intermediate
1079 * layout
1080 */
1081 if (level == 5) {
1082 if (layout_str == NULL)
1083 switch (array.layout) {
1084 case ALGORITHM_LEFT_ASYMMETRIC:
1085 case ALGORITHM_LEFT_ASYMMETRIC_6:
1086 case ALGORITHM_ROTATING_N_RESTART:
1087 layout_str = "left-asymmetric-6";
1088 break;
1089 case ALGORITHM_LEFT_SYMMETRIC:
1090 case ALGORITHM_LEFT_SYMMETRIC_6:
1091 case ALGORITHM_ROTATING_N_CONTINUE:
1092 layout_str = "left-symmetric-6";
1093 break;
1094 case ALGORITHM_RIGHT_ASYMMETRIC:
1095 case ALGORITHM_RIGHT_ASYMMETRIC_6:
1096 case ALGORITHM_ROTATING_ZERO_RESTART:
1097 layout_str = "right-asymmetric-6";
1098 break;
1099 case ALGORITHM_RIGHT_SYMMETRIC:
1100 case ALGORITHM_RIGHT_SYMMETRIC_6:
1101 layout_str = "right-symmetric-6";
1102 break;
1103 case ALGORITHM_PARITY_0:
1104 case ALGORITHM_PARITY_0_6:
1105 layout_str = "parity-first-6";
1106 break;
1107 case ALGORITHM_PARITY_N:
1108 layout_str = "parity-last";
1109 break;
1110 default:
1111 fprintf(stderr, Name ": %s: cannot"
1112 "convert layout to RAID5 equivalent\n",
1113 devname);
1114 rv = 1;
1115 goto release;
1116 }
1117 else {
1118 int l = map_name(r5layout, layout_str);
1119 if (l == UnSet) {
1120 fprintf(stderr, Name ": %s: layout '%s' not recognised\n",
1121 devname, layout_str);
1122 rv = 1;
1123 goto release;
1124 }
1125 if (l != ALGORITHM_PARITY_N) {
1126 /* need the -6 version */
1127 char *ls = map_num(r5layout, l);
1128 strcat(strcpy(alt_layout, ls),
1129 "-6");
1130 layout_str = alt_layout;
1131 }
1132 }
1133 if (raid_disks)
1134 /* The final raid6->raid5 conversion
1135 * will reduce the number of disks,
1136 * so now we need to aim higher
1137 */
1138 raid_disks++;
1139 } else
1140 layout_str = "parity-last";
1141 } else {
1142 /* Level change is a simple takeover. In the external
1143 * case we don't check with the metadata handler until
1144 * we establish what the final layout will be. If the
1145 * level change is disallowed we will revert to
1146 * orig_level without disturbing the metadata, otherwise
1147 * we will send an update.
1148 */
1149 c = map_num(pers, level);
1150 if (c == NULL) {
1151 rv = 1;/* not possible */
1152 goto release;
1153 }
1154 err = sysfs_set_str(sra, NULL, "level", c);
1155 if (err) {
1156 err = errno;
1157 fprintf(stderr, Name ": %s: could not set level to %s\n",
1158 devname, c);
1159 if (err == EBUSY &&
1160 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1161 fprintf(stderr, " Bitmap must be removed before level can be changed\n");
1162 rv = 1;
1163 goto release;
1164 }
1165 orig = array;
1166 orig_level = orig.level;
1167 ioctl(fd, GET_ARRAY_INFO, &array);
1168 if (layout_str == NULL &&
1169 orig.level == 5 && level == 6 &&
1170 array.layout != orig.layout)
1171 layout_str = map_num(r5layout, orig.layout);
1172 if (!quiet)
1173 fprintf(stderr, Name " level of %s changed to %s\n",
1174 devname, c);
1175 changed = 1;
1176 }
1177 }
1178
1179 /* ========= set shape (chunk_size / layout / ndisks) ============== */
1180 /* Check if layout change is a no-op */
1181 switch (array.level) {
1182 case 5:
1183 if (layout_str && array.layout == map_name(r5layout, layout_str))
1184 layout_str = NULL;
1185 break;
1186 case 6:
1187 if (layout_str == NULL &&
1188 ((chunksize && chunksize * 1024 != array.chunk_size) ||
1189 (raid_disks && raid_disks != array.raid_disks)) &&
1190 array.layout >= 16) {
1191 fprintf(stderr, Name
1192 ": %s has a non-standard layout. If you wish to preserve this\n"
1193 " during the reshape, please specify --layout=preserve\n"
1194 " If you want to change it, specify a layout or use --layout=normalise\n",
1195 devname);
1196 rv = 1;
1197 goto release;
1198 }
1199 if (layout_str &&
1200 (strcmp(layout_str, "normalise") == 0 ||
1201 strcmp(layout_str, "normalize") == 0)) {
1202 char *hyphen;
1203 strcpy(alt_layout, map_num(r6layout, array.layout));
1204 hyphen = strrchr(alt_layout, '-');
1205 if (hyphen && strcmp(hyphen, "-6") == 0) {
1206 *hyphen = 0;
1207 layout_str = alt_layout;
1208 }
1209 }
1210
1211 if (layout_str && array.layout == map_name(r6layout, layout_str))
1212 layout_str = NULL;
1213 if (layout_str && strcmp(layout_str, "preserve") == 0)
1214 layout_str = NULL;
1215 break;
1216 }
1217 if (layout_str == NULL
1218 && (chunksize == 0 || chunksize*1024 == array.chunk_size)
1219 && (raid_disks == 0 || raid_disks == array.raid_disks)) {
1220 if (reshape_super(st, -1, level, UnSet, 0, 0, NULL, devname, !quiet)) {
1221 rv = 1;
1222 goto release;
1223 }
1224 sync_metadata(st);
1225 rv = 0;
1226 if (level != UnSet && level != array.level) {
1227 /* Looks like this level change doesn't need
1228 * a reshape after all.
1229 */
1230 c = map_num(pers, level);
1231 if (c) {
1232 rv = sysfs_set_str(sra, NULL, "level", c);
1233 if (rv) {
1234 int err = errno;
1235 fprintf(stderr, Name ": %s: could not set level to %s\n",
1236 devname, c);
1237 if (err == EBUSY &&
1238 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1239 fprintf(stderr, " Bitmap must be removed before level can be changed\n");
1240 rv = 1;
1241 }
1242 }
1243 } else if (!changed && !quiet)
1244 fprintf(stderr, Name ": %s: no change requested\n",
1245 devname);
1246
1247 if (st->ss->external && !mdmon_running(st->container_dev) &&
1248 level > 0) {
1249 start_mdmon(st->container_dev);
1250 ping_monitor(container);
1251 }
1252 goto release;
1253 }
1254
1255 c = map_num(pers, array.level);
1256 if (c == NULL) c = "-unknown-";
1257 switch (array.level) {
1258 default: /* raid0, linear, multipath cannot be reconfigured */
1259 fprintf(stderr, Name ": %s array %s cannot be reshaped.\n",
1260 c, devname);
1261 /* TODO raid0 raiddisks can be reshaped via raid4 */
1262 rv = 1;
1263 break;
1264 case LEVEL_CONTAINER: {
1265 int count;
1266
1267 /* double check that we are not changing anything but raid_disks */
1268 if (size >= 0 || layout_str != NULL || chunksize != 0 || level != UnSet) {
1269 fprintf(stderr,
1270 Name ": %s is a container, only 'raid-devices' can be changed\n",
1271 devname);
1272 rv = 1;
1273 goto release;
1274 }
1275
1276 st->update_tail = &st->updates;
1277 if (reshape_super(st, -1, UnSet, UnSet, 0, raid_disks,
1278 backup_file, devname, !quiet)) {
1279 rv = 1;
1280 goto release;
1281 }
1282
1283 count = reshape_container_raid_disks(container, raid_disks);
1284 if (count < 0) {
1285 revert_container_raid_disks(st, fd, container);
1286 rv = 1;
1287 goto release;
1288 } else if (count == 0) {
1289 if (!quiet)
1290 fprintf(stderr, Name
1291 ": no active subarrays to reshape\n");
1292 goto release;
1293 }
1294
1295 if (!mdmon_running(st->devnum)) {
1296 start_mdmon(st->devnum);
1297 ping_monitor(container);
1298 }
1299 sync_metadata(st);
1300
1301 /* give mdmon a chance to allocate spares */
1302 ping_manager(container);
1303
1304 /* manage_reshape takes care of releasing the array(s) */
1305 st->ss->manage_reshape(st, backup_file);
1306 frozen = 0;
1307 goto release;
1308 }
1309 case LEVEL_FAULTY: /* only 'layout' change is permitted */
1310
1311 if (chunksize || raid_disks) {
1312 fprintf(stderr, Name ": %s: Cannot change chunksize or disks of a 'faulty' array\n",
1313 devname);
1314 rv = 1;
1315 break;
1316 }
1317 if (layout_str == NULL)
1318 break; /* nothing to do.... */
1319
1320 array.layout = parse_layout_faulty(layout_str);
1321 if (array.layout < 0) {
1322 fprintf(stderr, Name ": %s: layout %s not understood for 'faulty' array\n",
1323 devname, layout_str);
1324 rv = 1;
1325 break;
1326 }
1327 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1328 fprintf(stderr, Name ": Cannot set layout for %s: %s\n",
1329 devname, strerror(errno));
1330 rv = 1;
1331 } else if (!quiet)
1332 printf("layout for %s set to %d\n", devname, array.layout);
1333 break;
1334
1335 case 1: /* only raid_disks can each be changed. */
1336
1337 if (chunksize || layout_str != NULL) {
1338 fprintf(stderr, Name ": %s: Cannot change chunk size or layout for a RAID1 array.\n",
1339 devname);
1340 rv = 1;
1341 break;
1342 }
1343 if (raid_disks > 0) {
1344 if (reshape_super(st, -1, UnSet, UnSet, 0, raid_disks,
1345 NULL, devname, !quiet)) {
1346 rv = 1;
1347 goto release;
1348 }
1349 sync_metadata(st);
1350 array.raid_disks = raid_disks;
1351 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1352 fprintf(stderr, Name ": Cannot set raid-devices for %s: %s\n",
1353 devname, strerror(errno));
1354 rv = 1;
1355 }
1356 }
1357 break;
1358
1359 case 4:
1360 case 5:
1361 case 6:
1362
1363 /*
1364 * layout/chunksize/raid_disks can be changed
1365 * though the kernel may not support it all.
1366 */
1367 if (subarray) {
1368 fprintf(stderr, Name ": Cannot reshape subarrays yet\n");
1369 break;
1370 }
1371
1372 /*
1373 * There are three possibilities.
1374 * 1/ The array will shrink.
1375 * We need to ensure the reshape will pause before reaching
1376 * the 'critical section'. We also need to fork and wait for
1377 * that to happen. When it does we
1378 * suspend/backup/complete/unfreeze
1379 *
1380 * 2/ The array will not change size.
1381 * This requires that we keep a backup of a sliding window
1382 * so that we can restore data after a crash. So we need
1383 * to fork and monitor progress.
1384 *
1385 * 3/ The array will grow. This is relatively easy.
1386 * However the kernel's restripe routines will cheerfully
1387 * overwrite some early data before it is safe. So we
1388 * need to make a backup of the early parts of the array
1389 * and be ready to restore it if rebuild aborts very early.
1390 *
1391 * We backup data by writing it to one spare, or to a
1392 * file which was given on command line.
1393 *
1394 * [FOLLOWING IS OLD AND PARTLY WRONG]
1395 * So: we enumerate the devices in the array and
1396 * make sure we can open all of them.
1397 * Then we freeze the early part of the array and
1398 * backup to the various spares.
1399 * Then we request changes and start the reshape.
1400 * Monitor progress until it has passed the danger zone.
1401 * and finally invalidate the copied data and unfreeze the
1402 * start of the array.
1403 *
1404 * In each case, we first make sure that storage is available
1405 * for the required backup.
1406 * Then we:
1407 * - request the shape change.
1408 * - for to handle backup etc.
1409 */
1410 nchunk = ochunk = array.chunk_size;
1411 nlayout = olayout = array.layout;
1412 ndisks = odisks = array.raid_disks;
1413
1414 if (chunksize) {
1415 nchunk = chunksize * 1024;
1416 if (size % chunksize) {
1417 fprintf(stderr, Name ": component size %lluK is not"
1418 " a multiple of chunksize %dK\n",
1419 size, chunksize);
1420 break;
1421 }
1422 }
1423 if (layout_str != NULL)
1424 switch(array.level) {
1425 case 4: /* ignore layout */
1426 break;
1427 case 5:
1428 nlayout = map_name(r5layout, layout_str);
1429 if (nlayout == UnSet) {
1430 fprintf(stderr, Name ": layout %s not understood for raid5.\n",
1431 layout_str);
1432 rv = 1;
1433 goto release;
1434 }
1435 break;
1436
1437 case 6:
1438 nlayout = map_name(r6layout, layout_str);
1439 if (nlayout == UnSet) {
1440 fprintf(stderr, Name ": layout %s not understood for raid6.\n",
1441 layout_str);
1442 rv = 1;
1443 goto release;
1444 }
1445 break;
1446 }
1447 if (raid_disks) ndisks = raid_disks;
1448
1449 odata = odisks-1;
1450 ndata = ndisks-1;
1451 if (array.level == 6) {
1452 odata--; /* number of data disks */
1453 ndata--;
1454 }
1455
1456 if (odata == ndata &&
1457 get_linux_version() < 2006032) {
1458 fprintf(stderr, Name ": in-place reshape is not safe before 2.6.32, sorry.\n");
1459 break;
1460 }
1461
1462 /* Check that we can hold all the data */
1463 get_dev_size(fd, NULL, &array_size);
1464 if (ndata * (unsigned long long)size < (array_size/1024)) {
1465 fprintf(stderr, Name ": this change will reduce the size of the array.\n"
1466 " use --grow --array-size first to truncate array.\n"
1467 " e.g. mdadm --grow %s --array-size %llu\n",
1468 devname, ndata * size);
1469 rv = 1;
1470 break;
1471 }
1472
1473 /* So how much do we need to backup.
1474 * We need an amount of data which is both a whole number of
1475 * old stripes and a whole number of new stripes.
1476 * So LCM for (chunksize*datadisks).
1477 */
1478 a = (ochunk/512) * odata;
1479 b = (nchunk/512) * ndata;
1480 /* Find GCD */
1481 while (a != b) {
1482 if (a < b)
1483 b -= a;
1484 if (b < a)
1485 a -= b;
1486 }
1487 /* LCM == product / GCD */
1488 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
1489
1490 sysfs_free(sra);
1491 sra = sysfs_read(fd, 0,
1492 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|
1493 GET_CACHE);
1494
1495 if (!sra) {
1496 fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
1497 devname);
1498 rv = 1;
1499 break;
1500 }
1501
1502 if (ndata == odata) {
1503 /* Make 'blocks' bigger for better throughput, but
1504 * not so big that we reject it below.
1505 * Try for 16 megabytes
1506 */
1507 while (blocks * 32 < sra->component_size &&
1508 blocks < 16*1024*2)
1509 blocks *= 2;
1510 } else
1511 fprintf(stderr, Name ": Need to backup %luK of critical "
1512 "section..\n", blocks/2);
1513
1514 if (blocks >= sra->component_size/2) {
1515 fprintf(stderr, Name ": %s: Something wrong - reshape aborted\n",
1516 devname);
1517 rv = 1;
1518 break;
1519 }
1520 nrdisks = array.raid_disks + sra->array.spare_disks;
1521 /* Now we need to open all these devices so we can read/write.
1522 */
1523 fdlist = malloc((1+nrdisks) * sizeof(int));
1524 offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
1525 if (!fdlist || !offsets) {
1526 fprintf(stderr, Name ": malloc failed: grow aborted\n");
1527 rv = 1;
1528 break;
1529 }
1530 for (d=0; d <= nrdisks; d++)
1531 fdlist[d] = -1;
1532 d = array.raid_disks;
1533 for (sd = sra->devs; sd; sd=sd->next) {
1534 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1535 continue;
1536 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
1537 char *dn = map_dev(sd->disk.major,
1538 sd->disk.minor, 1);
1539 fdlist[sd->disk.raid_disk]
1540 = dev_open(dn, O_RDONLY);
1541 offsets[sd->disk.raid_disk] = sd->data_offset*512;
1542 if (fdlist[sd->disk.raid_disk] < 0) {
1543 fprintf(stderr, Name ": %s: cannot open component %s\n",
1544 devname, dn?dn:"-unknown-");
1545 rv = 1;
1546 goto release;
1547 }
1548 } else if (backup_file == NULL) {
1549 /* spare */
1550 char *dn = map_dev(sd->disk.major,
1551 sd->disk.minor, 1);
1552 fdlist[d] = dev_open(dn, O_RDWR);
1553 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
1554 if (fdlist[d]<0) {
1555 fprintf(stderr, Name ": %s: cannot open component %s\n",
1556 devname, dn?dn:"-unknown");
1557 rv = 1;
1558 goto release;
1559 }
1560 d++;
1561 }
1562 }
1563 if (backup_file == NULL) {
1564 if (st->ss->external && !st->ss->manage_reshape) {
1565 fprintf(stderr, Name ": %s Grow operation not supported by %s metadata\n",
1566 devname, st->ss->name);
1567 rv = 1;
1568 break;
1569 }
1570 if (ndata <= odata) {
1571 fprintf(stderr, Name ": %s: Cannot grow - need backup-file\n",
1572 devname);
1573 rv = 1;
1574 break;
1575 } else if (sra->array.spare_disks == 0) {
1576 fprintf(stderr, Name ": %s: Cannot grow - need a spare or "
1577 "backup-file to backup critical section\n",
1578 devname);
1579 rv = 1;
1580 break;
1581 }
1582 if (d == array.raid_disks) {
1583 fprintf(stderr, Name ": %s: No spare device for backup\n",
1584 devname);
1585 rv = 1;
1586 break;
1587 }
1588 } else {
1589 /* need to check backup file is large enough */
1590 char buf[512];
1591 fdlist[d] = open(backup_file, O_RDWR|O_CREAT|O_EXCL,
1592 S_IRUSR | S_IWUSR);
1593 offsets[d] = 8 * 512;
1594 if (fdlist[d] < 0) {
1595 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
1596 devname, backup_file, strerror(errno));
1597 rv = 1;
1598 break;
1599 }
1600 memset(buf, 0, 512);
1601 for (i=0; i < (signed)blocks + 1 ; i++) {
1602 if (write(fdlist[d], buf, 512) != 512) {
1603 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
1604 devname, backup_file, strerror(errno));
1605 rv = 1;
1606 break;
1607 }
1608 }
1609 if (fsync(fdlist[d]) != 0) {
1610 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
1611 devname, backup_file, strerror(errno));
1612 rv = 1;
1613 break;
1614 }
1615 d++;
1616 }
1617
1618 /* check that the operation is supported by the metadata */
1619 if (reshape_super(st, -1, level, nlayout, nchunk, ndisks,
1620 backup_file, devname, !quiet)) {
1621 rv = 1;
1622 break;
1623 }
1624
1625 /* lastly, check that the internal stripe cache is
1626 * large enough, or it won't work.
1627 */
1628
1629 cache = (nchunk < ochunk) ? ochunk : nchunk;
1630 cache = cache * 4 / 4096;
1631 if (cache < blocks / 8 / odisks + 16)
1632 /* Make it big enough to hold 'blocks' */
1633 cache = blocks / 8 / odisks + 16;
1634 if (sra->cache_size < cache)
1635 sysfs_set_num(sra, NULL, "stripe_cache_size",
1636 cache+1);
1637 /* Right, everything seems fine. Let's kick things off.
1638 * If only changing raid_disks, use ioctl, else use
1639 * sysfs.
1640 */
1641 sync_metadata(st);
1642 if (ochunk == nchunk && olayout == nlayout) {
1643 array.raid_disks = ndisks;
1644 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1645 int err = errno;
1646 rv = 1;
1647 fprintf(stderr, Name ": Cannot set device shape for %s: %s\n",
1648 devname, strerror(errno));
1649 if (ndisks < odisks &&
1650 get_linux_version() < 2006030)
1651 fprintf(stderr, Name ": linux 2.6.30 or later required\n");
1652 if (err == EBUSY &&
1653 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1654 fprintf(stderr, " Bitmap must be removed before shape can be changed\n");
1655
1656 break;
1657 }
1658 } else {
1659 /* set them all just in case some old 'new_*' value
1660 * persists from some earlier problem
1661 */
1662 int err = err; /* only used if rv==1, and always set if
1663 * rv==1, so initialisation not needed,
1664 * despite gcc warning
1665 */
1666 if (sysfs_set_num(sra, NULL, "chunk_size", nchunk) < 0)
1667 rv = 1, err = errno;
1668 if (!rv && sysfs_set_num(sra, NULL, "layout", nlayout) < 0)
1669 rv = 1, err = errno;
1670 if (!rv && sysfs_set_num(sra, NULL, "raid_disks", ndisks) < 0)
1671 rv = 1, err = errno;
1672 if (rv) {
1673 fprintf(stderr, Name ": Cannot set device shape for %s\n",
1674 devname);
1675 if (get_linux_version() < 2006030)
1676 fprintf(stderr, Name ": linux 2.6.30 or later required\n");
1677 if (err == EBUSY &&
1678 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1679 fprintf(stderr, " Bitmap must be removed before shape can be changed\n");
1680 break;
1681 }
1682 }
1683
1684 if (ndisks == 2 && odisks == 2) {
1685 /* No reshape is needed in this trivial case */
1686 rv = 0;
1687 break;
1688 }
1689
1690 if (st->ss->external) {
1691 /* metadata handler takes it from here */
1692 ping_manager(container);
1693 st->ss->manage_reshape(st, backup_file);
1694 frozen = 0;
1695 break;
1696 }
1697
1698 /* set up the backup-super-block. This requires the
1699 * uuid from the array.
1700 */
1701 /* Find a superblock */
1702 for (sd = sra->devs; sd; sd = sd->next) {
1703 char *dn;
1704 int devfd;
1705 int ok;
1706 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1707 continue;
1708 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
1709 devfd = dev_open(dn, O_RDONLY);
1710 if (devfd < 0)
1711 continue;
1712 ok = st->ss->load_super(st, devfd, NULL);
1713 close(devfd);
1714 if (ok >= 0)
1715 break;
1716 }
1717 if (!sd) {
1718 fprintf(stderr, Name ": %s: Cannot find a superblock\n",
1719 devname);
1720 rv = 1;
1721 break;
1722 }
1723
1724 memset(&bsb, 0, 512);
1725 memcpy(bsb.magic, "md_backup_data-1", 16);
1726 st->ss->uuid_from_super(st, (int*)&bsb.set_uuid);
1727 bsb.mtime = __cpu_to_le64(time(0));
1728 bsb.devstart2 = blocks;
1729 stripes = blocks / (ochunk/512) / odata;
1730 /* Now we just need to kick off the reshape and watch, while
1731 * handling backups of the data...
1732 * This is all done by a forked background process.
1733 */
1734 switch(fork()) {
1735 case 0:
1736 close(fd);
1737 if (check_env("MDADM_GROW_VERIFY"))
1738 fd = open(devname, O_RDONLY | O_DIRECT);
1739 else
1740 fd = -1;
1741 mlockall(MCL_FUTURE);
1742
1743 if (odata < ndata)
1744 done = child_grow(fd, sra, stripes,
1745 fdlist, offsets,
1746 odisks, ochunk, array.level, olayout, odata,
1747 d - odisks, fdlist+odisks, offsets+odisks);
1748 else if (odata > ndata)
1749 done = child_shrink(fd, sra, stripes,
1750 fdlist, offsets,
1751 odisks, ochunk, array.level, olayout, odata,
1752 d - odisks, fdlist+odisks, offsets+odisks);
1753 else
1754 done = child_same_size(fd, sra, stripes,
1755 fdlist, offsets,
1756 0,
1757 odisks, ochunk, array.level, olayout, odata,
1758 d - odisks, fdlist+odisks, offsets+odisks);
1759 if (backup_file && done)
1760 unlink(backup_file);
1761 if (level != UnSet && level != array.level) {
1762 /* We need to wait for the reshape to finish
1763 * (which will have happened unless odata < ndata)
1764 * and then set the level
1765 */
1766
1767 c = map_num(pers, level);
1768 if (c == NULL)
1769 exit(0);/* not possible */
1770
1771 if (odata < ndata)
1772 wait_reshape(sra);
1773 err = sysfs_set_str(sra, NULL, "level", c);
1774 if (err)
1775 fprintf(stderr, Name ": %s: could not set level to %s\n",
1776 devname, c);
1777 }
1778 exit(0);
1779 case -1:
1780 fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n",
1781 strerror(errno));
1782 rv = 1;
1783 break;
1784 default:
1785 /* The child will take care of unfreezing the array */
1786 frozen = 0;
1787 break;
1788 }
1789 break;
1790
1791 }
1792
1793 release:
1794 if (rv && orig_level != UnSet && sra) {
1795 c = map_num(pers, orig_level);
1796 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
1797 fprintf(stderr, Name ": aborting level change\n");
1798 }
1799 unfreeze(st, frozen);
1800 return rv;
1801 }
1802
1803 /*
1804 * We run a child process in the background which performs the following
1805 * steps:
1806 * - wait for resync to reach a certain point
1807 * - suspend io to the following section
1808 * - backup that section
1809 * - allow resync to proceed further
1810 * - resume io
1811 * - discard the backup.
1812 *
1813 * When are combined in slightly different ways in the three cases.
1814 * Grow:
1815 * - suspend/backup/allow/wait/resume/discard
1816 * Shrink:
1817 * - allow/wait/suspend/backup/allow/wait/resume/discard
1818 * same-size:
1819 * - wait/resume/discard/suspend/backup/allow
1820 *
1821 * suspend/backup/allow always come together
1822 * wait/resume/discard do too.
1823 * For the same-size case we have two backups to improve flow.
1824 *
1825 */
1826
1827 /* FIXME return status is never checked */
1828 static int grow_backup(struct mdinfo *sra,
1829 unsigned long long offset, /* per device */
1830 unsigned long stripes, /* per device */
1831 int *sources, unsigned long long *offsets,
1832 int disks, int chunk, int level, int layout,
1833 int dests, int *destfd, unsigned long long *destoffsets,
1834 int part, int *degraded,
1835 char *buf)
1836 {
1837 /* Backup 'blocks' sectors at 'offset' on each device of the array,
1838 * to storage 'destfd' (offset 'destoffsets'), after first
1839 * suspending IO. Then allow resync to continue
1840 * over the suspended section.
1841 * Use part 'part' of the backup-super-block.
1842 */
1843 int odata = disks;
1844 int rv = 0;
1845 int i;
1846 unsigned long long ll;
1847 int new_degraded;
1848 //printf("offset %llu\n", offset);
1849 if (level >= 4)
1850 odata--;
1851 if (level == 6)
1852 odata--;
1853 sysfs_set_num(sra, NULL, "suspend_hi", (offset + stripes * (chunk/512)) * odata);
1854 /* Check that array hasn't become degraded, else we might backup the wrong data */
1855 sysfs_get_ll(sra, NULL, "degraded", &ll);
1856 new_degraded = (int)ll;
1857 if (new_degraded != *degraded) {
1858 /* check each device to ensure it is still working */
1859 struct mdinfo *sd;
1860 for (sd = sra->devs ; sd ; sd = sd->next) {
1861 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1862 continue;
1863 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
1864 char sbuf[20];
1865 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
1866 strstr(sbuf, "faulty") ||
1867 strstr(sbuf, "in_sync") == NULL) {
1868 /* this device is dead */
1869 sd->disk.state = (1<<MD_DISK_FAULTY);
1870 if (sd->disk.raid_disk >= 0 &&
1871 sources[sd->disk.raid_disk] >= 0) {
1872 close(sources[sd->disk.raid_disk]);
1873 sources[sd->disk.raid_disk] = -1;
1874 }
1875 }
1876 }
1877 }
1878 *degraded = new_degraded;
1879 }
1880 if (part) {
1881 bsb.arraystart2 = __cpu_to_le64(offset * odata);
1882 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
1883 } else {
1884 bsb.arraystart = __cpu_to_le64(offset * odata);
1885 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
1886 }
1887 if (part)
1888 bsb.magic[15] = '2';
1889 for (i = 0; i < dests; i++)
1890 if (part)
1891 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
1892 else
1893 lseek64(destfd[i], destoffsets[i], 0);
1894
1895 rv = save_stripes(sources, offsets,
1896 disks, chunk, level, layout,
1897 dests, destfd,
1898 offset*512*odata, stripes * chunk * odata,
1899 buf);
1900
1901 if (rv)
1902 return rv;
1903 bsb.mtime = __cpu_to_le64(time(0));
1904 for (i = 0; i < dests; i++) {
1905 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
1906
1907 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
1908 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
1909 bsb.sb_csum2 = bsb_csum((char*)&bsb,
1910 ((char*)&bsb.sb_csum2)-((char*)&bsb));
1911
1912 rv = -1;
1913 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
1914 != destoffsets[i] - 4096)
1915 break;
1916 if (write(destfd[i], &bsb, 512) != 512)
1917 break;
1918 if (destoffsets[i] > 4096) {
1919 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
1920 destoffsets[i]+stripes*chunk*odata)
1921 break;
1922 if (write(destfd[i], &bsb, 512) != 512)
1923 break;
1924 }
1925 fsync(destfd[i]);
1926 rv = 0;
1927 }
1928
1929 return rv;
1930 }
1931
1932 /* in 2.6.30, the value reported by sync_completed can be
1933 * less that it should be by one stripe.
1934 * This only happens when reshape hits sync_max and pauses.
1935 * So allow wait_backup to either extent sync_max further
1936 * than strictly necessary, or return before the
1937 * sync has got quite as far as we would really like.
1938 * This is what 'blocks2' is for.
1939 * The various caller give appropriate values so that
1940 * every works.
1941 */
1942 /* FIXME return value is often ignored */
1943 static int wait_backup(struct mdinfo *sra,
1944 unsigned long long offset, /* per device */
1945 unsigned long long blocks, /* per device */
1946 unsigned long long blocks2, /* per device - hack */
1947 int dests, int *destfd, unsigned long long *destoffsets,
1948 int part)
1949 {
1950 /* Wait for resync to pass the section that was backed up
1951 * then erase the backup and allow IO
1952 */
1953 int fd = sysfs_get_fd(sra, NULL, "sync_completed");
1954 unsigned long long completed;
1955 int i;
1956 int rv;
1957
1958 if (fd < 0)
1959 return -1;
1960 sysfs_set_num(sra, NULL, "sync_max", offset + blocks + blocks2);
1961 if (offset == 0)
1962 sysfs_set_str(sra, NULL, "sync_action", "reshape");
1963 do {
1964 char action[20];
1965 fd_set rfds;
1966 FD_ZERO(&rfds);
1967 FD_SET(fd, &rfds);
1968 select(fd+1, NULL, NULL, &rfds, NULL);
1969 if (sysfs_fd_get_ll(fd, &completed) < 0) {
1970 close(fd);
1971 return -1;
1972 }
1973 if (sysfs_get_str(sra, NULL, "sync_action",
1974 action, 20) > 0 &&
1975 strncmp(action, "reshape", 7) != 0)
1976 break;
1977 } while (completed < offset + blocks);
1978 close(fd);
1979
1980 if (part) {
1981 bsb.arraystart2 = __cpu_to_le64(0);
1982 bsb.length2 = __cpu_to_le64(0);
1983 } else {
1984 bsb.arraystart = __cpu_to_le64(0);
1985 bsb.length = __cpu_to_le64(0);
1986 }
1987 bsb.mtime = __cpu_to_le64(time(0));
1988 rv = 0;
1989 for (i = 0; i < dests; i++) {
1990 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
1991 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
1992 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
1993 bsb.sb_csum2 = bsb_csum((char*)&bsb,
1994 ((char*)&bsb.sb_csum2)-((char*)&bsb));
1995 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
1996 destoffsets[i]-4096)
1997 rv = -1;
1998 if (rv == 0 &&
1999 write(destfd[i], &bsb, 512) != 512)
2000 rv = -1;
2001 fsync(destfd[i]);
2002 }
2003 return rv;
2004 }
2005
2006 static void fail(char *msg)
2007 {
2008 int rv;
2009 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
2010 rv |= (write(2, "\n", 1) != 1);
2011 exit(rv ? 1 : 2);
2012 }
2013
2014 static char *abuf, *bbuf;
2015 static unsigned long long abuflen;
2016 static void validate(int afd, int bfd, unsigned long long offset)
2017 {
2018 /* check that the data in the backup against the array.
2019 * This is only used for regression testing and should not
2020 * be used while the array is active
2021 */
2022 if (afd < 0)
2023 return;
2024 lseek64(bfd, offset - 4096, 0);
2025 if (read(bfd, &bsb2, 512) != 512)
2026 fail("cannot read bsb");
2027 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
2028 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
2029 fail("first csum bad");
2030 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
2031 fail("magic is bad");
2032 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
2033 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
2034 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
2035 fail("second csum bad");
2036
2037 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
2038 fail("devstart is wrong");
2039
2040 if (bsb2.length) {
2041 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
2042
2043 if (abuflen < len) {
2044 free(abuf);
2045 free(bbuf);
2046 abuflen = len;
2047 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
2048 posix_memalign((void**)&bbuf, 4096, abuflen)) {
2049 abuflen = 0;
2050 /* just stop validating on mem-alloc failure */
2051 return;
2052 }
2053 }
2054
2055 lseek64(bfd, offset, 0);
2056 if ((unsigned long long)read(bfd, bbuf, len) != len) {
2057 //printf("len %llu\n", len);
2058 fail("read first backup failed");
2059 }
2060 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
2061 if ((unsigned long long)read(afd, abuf, len) != len)
2062 fail("read first from array failed");
2063 if (memcmp(bbuf, abuf, len) != 0) {
2064 #if 0
2065 int i;
2066 printf("offset=%llu len=%llu\n",
2067 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
2068 for (i=0; i<len; i++)
2069 if (bbuf[i] != abuf[i]) {
2070 printf("first diff byte %d\n", i);
2071 break;
2072 }
2073 #endif
2074 fail("data1 compare failed");
2075 }
2076 }
2077 if (bsb2.length2) {
2078 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
2079
2080 if (abuflen < len) {
2081 free(abuf);
2082 free(bbuf);
2083 abuflen = len;
2084 abuf = malloc(abuflen);
2085 bbuf = malloc(abuflen);
2086 }
2087
2088 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
2089 if ((unsigned long long)read(bfd, bbuf, len) != len)
2090 fail("read second backup failed");
2091 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
2092 if ((unsigned long long)read(afd, abuf, len) != len)
2093 fail("read second from array failed");
2094 if (memcmp(bbuf, abuf, len) != 0)
2095 fail("data2 compare failed");
2096 }
2097 }
2098
2099 static int child_grow(int afd, struct mdinfo *sra, unsigned long stripes,
2100 int *fds, unsigned long long *offsets,
2101 int disks, int chunk, int level, int layout, int data,
2102 int dests, int *destfd, unsigned long long *destoffsets)
2103 {
2104 char *buf;
2105 int degraded = 0;
2106
2107 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2108 /* Don't start the 'reshape' */
2109 return 0;
2110 sysfs_set_num(sra, NULL, "suspend_hi", 0);
2111 sysfs_set_num(sra, NULL, "suspend_lo", 0);
2112 grow_backup(sra, 0, stripes,
2113 fds, offsets, disks, chunk, level, layout,
2114 dests, destfd, destoffsets,
2115 0, &degraded, buf);
2116 validate(afd, destfd[0], destoffsets[0]);
2117 wait_backup(sra, 0, stripes * (chunk / 512), stripes * (chunk / 512),
2118 dests, destfd, destoffsets,
2119 0);
2120 sysfs_set_num(sra, NULL, "suspend_lo", (stripes * (chunk/512)) * data);
2121 free(buf);
2122 /* FIXME this should probably be numeric */
2123 sysfs_set_str(sra, NULL, "sync_max", "max");
2124 return 1;
2125 }
2126
2127 static int child_shrink(int afd, struct mdinfo *sra, unsigned long stripes,
2128 int *fds, unsigned long long *offsets,
2129 int disks, int chunk, int level, int layout, int data,
2130 int dests, int *destfd, unsigned long long *destoffsets)
2131 {
2132 char *buf;
2133 unsigned long long start;
2134 int rv;
2135 int degraded = 0;
2136
2137 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2138 return 0;
2139 start = sra->component_size - stripes * (chunk/512);
2140 sysfs_set_num(sra, NULL, "sync_max", start);
2141 sysfs_set_str(sra, NULL, "sync_action", "reshape");
2142 sysfs_set_num(sra, NULL, "suspend_lo", 0);
2143 sysfs_set_num(sra, NULL, "suspend_hi", 0);
2144 rv = wait_backup(sra, 0, start - stripes * (chunk/512), stripes * (chunk/512),
2145 dests, destfd, destoffsets, 0);
2146 if (rv < 0)
2147 return 0;
2148 grow_backup(sra, 0, stripes,
2149 fds, offsets,
2150 disks, chunk, level, layout,
2151 dests, destfd, destoffsets,
2152 0, &degraded, buf);
2153 validate(afd, destfd[0], destoffsets[0]);
2154 wait_backup(sra, start, stripes*(chunk/512), 0,
2155 dests, destfd, destoffsets, 0);
2156 sysfs_set_num(sra, NULL, "suspend_lo", (stripes * (chunk/512)) * data);
2157 free(buf);
2158 /* FIXME this should probably be numeric */
2159 sysfs_set_str(sra, NULL, "sync_max", "max");
2160 return 1;
2161 }
2162
2163 static int child_same_size(int afd, struct mdinfo *sra, unsigned long stripes,
2164 int *fds, unsigned long long *offsets,
2165 unsigned long long start,
2166 int disks, int chunk, int level, int layout, int data,
2167 int dests, int *destfd, unsigned long long *destoffsets)
2168 {
2169 unsigned long long size;
2170 unsigned long tailstripes = stripes;
2171 int part;
2172 char *buf;
2173 unsigned long long speed;
2174 int degraded = 0;
2175
2176
2177 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2178 return 0;
2179
2180 sysfs_set_num(sra, NULL, "suspend_lo", 0);
2181 sysfs_set_num(sra, NULL, "suspend_hi", 0);
2182
2183 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
2184 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
2185
2186 grow_backup(sra, start, stripes,
2187 fds, offsets,
2188 disks, chunk, level, layout,
2189 dests, destfd, destoffsets,
2190 0, &degraded, buf);
2191 grow_backup(sra, (start + stripes) * (chunk/512), stripes,
2192 fds, offsets,
2193 disks, chunk, level, layout,
2194 dests, destfd, destoffsets,
2195 1, &degraded, buf);
2196 validate(afd, destfd[0], destoffsets[0]);
2197 part = 0;
2198 start += stripes * 2; /* where to read next */
2199 size = sra->component_size / (chunk/512);
2200 while (start < size) {
2201 if (wait_backup(sra, (start-stripes*2)*(chunk/512),
2202 stripes*(chunk/512), 0,
2203 dests, destfd, destoffsets,
2204 part) < 0)
2205 return 0;
2206 sysfs_set_num(sra, NULL, "suspend_lo", start*(chunk/512) * data);
2207 if (start + stripes > size)
2208 tailstripes = (size - start);
2209
2210 grow_backup(sra, start*(chunk/512), tailstripes,
2211 fds, offsets,
2212 disks, chunk, level, layout,
2213 dests, destfd, destoffsets,
2214 part, &degraded, buf);
2215 start += stripes;
2216 part = 1 - part;
2217 validate(afd, destfd[0], destoffsets[0]);
2218 }
2219 if (wait_backup(sra, (start-stripes*2) * (chunk/512), stripes * (chunk/512), 0,
2220 dests, destfd, destoffsets,
2221 part) < 0)
2222 return 0;
2223 sysfs_set_num(sra, NULL, "suspend_lo", ((start-stripes)*(chunk/512)) * data);
2224 wait_backup(sra, (start-stripes) * (chunk/512), tailstripes * (chunk/512), 0,
2225 dests, destfd, destoffsets,
2226 1-part);
2227 sysfs_set_num(sra, NULL, "suspend_lo", (size*(chunk/512)) * data);
2228 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
2229 free(buf);
2230 return 1;
2231 }
2232
2233 /*
2234 * If any spare contains md_back_data-1 which is recent wrt mtime,
2235 * write that data into the array and update the super blocks with
2236 * the new reshape_progress
2237 */
2238 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
2239 char *backup_file, int verbose)
2240 {
2241 int i, j;
2242 int old_disks;
2243 unsigned long long *offsets;
2244 unsigned long long nstripe, ostripe;
2245 int ndata, odata;
2246
2247 if (info->new_level != info->array.level)
2248 return 1; /* Cannot handle level changes (they are instantaneous) */
2249
2250 odata = info->array.raid_disks - info->delta_disks - 1;
2251 if (info->array.level == 6) odata--; /* number of data disks */
2252 ndata = info->array.raid_disks - 1;
2253 if (info->new_level == 6) ndata--;
2254
2255 old_disks = info->array.raid_disks - info->delta_disks;
2256
2257 if (info->delta_disks <= 0)
2258 /* Didn't grow, so the backup file must have
2259 * been used
2260 */
2261 old_disks = cnt;
2262 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
2263 struct mdinfo dinfo;
2264 int fd;
2265 int bsbsize;
2266 char *devname, namebuf[20];
2267
2268 /* This was a spare and may have some saved data on it.
2269 * Load the superblock, find and load the
2270 * backup_super_block.
2271 * If either fail, go on to next device.
2272 * If the backup contains no new info, just return
2273 * else restore data and update all superblocks
2274 */
2275 if (i == old_disks-1) {
2276 fd = open(backup_file, O_RDONLY);
2277 if (fd<0) {
2278 fprintf(stderr, Name ": backup file %s inaccessible: %s\n",
2279 backup_file, strerror(errno));
2280 continue;
2281 }
2282 devname = backup_file;
2283 } else {
2284 fd = fdlist[i];
2285 if (fd < 0)
2286 continue;
2287 if (st->ss->load_super(st, fd, NULL))
2288 continue;
2289
2290 st->ss->getinfo_super(st, &dinfo, NULL);
2291 st->ss->free_super(st);
2292
2293 if (lseek64(fd,
2294 (dinfo.data_offset + dinfo.component_size - 8) <<9,
2295 0) < 0) {
2296 fprintf(stderr, Name ": Cannot seek on device %d\n", i);
2297 continue; /* Cannot seek */
2298 }
2299 sprintf(namebuf, "device-%d", i);
2300 devname = namebuf;
2301 }
2302 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
2303 if (verbose)
2304 fprintf(stderr, Name ": Cannot read from %s\n", devname);
2305 continue; /* Cannot read */
2306 }
2307 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
2308 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
2309 if (verbose)
2310 fprintf(stderr, Name ": No backup metadata on %s\n", devname);
2311 continue;
2312 }
2313 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
2314 if (verbose)
2315 fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname);
2316 continue; /* bad checksum */
2317 }
2318 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
2319 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
2320 if (verbose)
2321 fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname);
2322 continue; /* Bad second checksum */
2323 }
2324 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
2325 if (verbose)
2326 fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname);
2327 continue; /* Wrong uuid */
2328 }
2329
2330 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2331 * sometimes they aren't... So allow considerable flexability in matching, and allow
2332 * this test to be overridden by an environment variable.
2333 */
2334 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
2335 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
2336 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2337 fprintf(stderr, Name ": accepting backup with timestamp %lu "
2338 "for array with timestamp %lu\n",
2339 (unsigned long)__le64_to_cpu(bsb.mtime),
2340 (unsigned long)info->array.utime);
2341 } else {
2342 if (verbose)
2343 fprintf(stderr, Name ": too-old timestamp on "
2344 "backup-metadata on %s\n", devname);
2345 continue; /* time stamp is too bad */
2346 }
2347 }
2348
2349 if (bsb.magic[15] == '1') {
2350 if (info->delta_disks >= 0) {
2351 /* reshape_progress is increasing */
2352 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2353 info->reshape_progress) {
2354 nonew:
2355 if (verbose)
2356 fprintf(stderr, Name ": backup-metadata found on %s but is not needed\n", devname);
2357 continue; /* No new data here */
2358 }
2359 } else {
2360 /* reshape_progress is decreasing */
2361 if (__le64_to_cpu(bsb.arraystart) >=
2362 info->reshape_progress)
2363 goto nonew; /* No new data here */
2364 }
2365 } else {
2366 if (info->delta_disks >= 0) {
2367 /* reshape_progress is increasing */
2368 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2369 info->reshape_progress &&
2370 __le64_to_cpu(bsb.arraystart2) + __le64_to_cpu(bsb.length2) <
2371 info->reshape_progress)
2372 goto nonew; /* No new data here */
2373 } else {
2374 /* reshape_progress is decreasing */
2375 if (__le64_to_cpu(bsb.arraystart) >=
2376 info->reshape_progress &&
2377 __le64_to_cpu(bsb.arraystart2) >=
2378 info->reshape_progress)
2379 goto nonew; /* No new data here */
2380 }
2381 }
2382 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
2383 second_fail:
2384 if (verbose)
2385 fprintf(stderr, Name ": Failed to verify secondary backup-metadata block on %s\n",
2386 devname);
2387 continue; /* Cannot seek */
2388 }
2389 /* There should be a duplicate backup superblock 4k before here */
2390 if (lseek64(fd, -4096, 1) < 0 ||
2391 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
2392 goto second_fail; /* Cannot find leading superblock */
2393 if (bsb.magic[15] == '1')
2394 bsbsize = offsetof(struct mdp_backup_super, pad1);
2395 else
2396 bsbsize = offsetof(struct mdp_backup_super, pad);
2397 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
2398 goto second_fail; /* Cannot find leading superblock */
2399
2400 /* Now need the data offsets for all devices. */
2401 offsets = malloc(sizeof(*offsets)*info->array.raid_disks);
2402 for(j=0; j<info->array.raid_disks; j++) {
2403 if (fdlist[j] < 0)
2404 continue;
2405 if (st->ss->load_super(st, fdlist[j], NULL))
2406 /* FIXME should be this be an error */
2407 continue;
2408 st->ss->getinfo_super(st, &dinfo, NULL);
2409 st->ss->free_super(st);
2410 offsets[j] = dinfo.data_offset * 512;
2411 }
2412 printf(Name ": restoring critical section\n");
2413
2414 if (restore_stripes(fdlist, offsets,
2415 info->array.raid_disks,
2416 info->new_chunk,
2417 info->new_level,
2418 info->new_layout,
2419 fd, __le64_to_cpu(bsb.devstart)*512,
2420 __le64_to_cpu(bsb.arraystart)*512,
2421 __le64_to_cpu(bsb.length)*512)) {
2422 /* didn't succeed, so giveup */
2423 if (verbose)
2424 fprintf(stderr, Name ": Error restoring backup from %s\n",
2425 devname);
2426 return 1;
2427 }
2428
2429 if (bsb.magic[15] == '2' &&
2430 restore_stripes(fdlist, offsets,
2431 info->array.raid_disks,
2432 info->new_chunk,
2433 info->new_level,
2434 info->new_layout,
2435 fd, __le64_to_cpu(bsb.devstart)*512 +
2436 __le64_to_cpu(bsb.devstart2)*512,
2437 __le64_to_cpu(bsb.arraystart2)*512,
2438 __le64_to_cpu(bsb.length2)*512)) {
2439 /* didn't succeed, so giveup */
2440 if (verbose)
2441 fprintf(stderr, Name ": Error restoring second backup from %s\n",
2442 devname);
2443 return 1;
2444 }
2445
2446
2447 /* Ok, so the data is restored. Let's update those superblocks. */
2448
2449 if (info->delta_disks >= 0) {
2450 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
2451 __le64_to_cpu(bsb.length);
2452 if (bsb.magic[15] == '2') {
2453 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
2454 __le64_to_cpu(bsb.length2);
2455 if (p2 > info->reshape_progress)
2456 info->reshape_progress = p2;
2457 }
2458 } else {
2459 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
2460 if (bsb.magic[15] == '2') {
2461 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
2462 if (p2 < info->reshape_progress)
2463 info->reshape_progress = p2;
2464 }
2465 }
2466 for (j=0; j<info->array.raid_disks; j++) {
2467 if (fdlist[j] < 0) continue;
2468 if (st->ss->load_super(st, fdlist[j], NULL))
2469 continue;
2470 st->ss->getinfo_super(st, &dinfo, NULL);
2471 dinfo.reshape_progress = info->reshape_progress;
2472 st->ss->update_super(st, &dinfo,
2473 "_reshape_progress",
2474 NULL,0, 0, NULL);
2475 st->ss->store_super(st, fdlist[j]);
2476 st->ss->free_super(st);
2477 }
2478 return 0;
2479 }
2480 /* Didn't find any backup data, try to see if any
2481 * was needed.
2482 */
2483 if (info->delta_disks < 0) {
2484 /* When shrinking, the critical section is at the end.
2485 * So see if we are before the critical section.
2486 */
2487 unsigned long long first_block;
2488 nstripe = ostripe = 0;
2489 first_block = 0;
2490 while (ostripe >= nstripe) {
2491 ostripe += info->array.chunk_size / 512;
2492 first_block = ostripe * odata;
2493 nstripe = first_block / ndata / (info->new_chunk/512) *
2494 (info->new_chunk/512);
2495 }
2496
2497 if (info->reshape_progress >= first_block)
2498 return 0;
2499 }
2500 if (info->delta_disks > 0) {
2501 /* See if we are beyond the critical section. */
2502 unsigned long long last_block;
2503 nstripe = ostripe = 0;
2504 last_block = 0;
2505 while (nstripe >= ostripe) {
2506 nstripe += info->new_chunk / 512;
2507 last_block = nstripe * ndata;
2508 ostripe = last_block / odata / (info->array.chunk_size/512) *
2509 (info->array.chunk_size/512);
2510 }
2511
2512 if (info->reshape_progress >= last_block)
2513 return 0;
2514 }
2515 /* needed to recover critical section! */
2516 if (verbose)
2517 fprintf(stderr, Name ": Failed to find backup of critical section\n");
2518 return 1;
2519 }
2520
2521 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
2522 char *backup_file)
2523 {
2524 /* Array is assembled and ready to be started, but
2525 * monitoring is probably required.
2526 * So:
2527 * - start read-only
2528 * - set upper bound for resync
2529 * - initialise the 'suspend' boundaries
2530 * - switch to read-write
2531 * - fork and continue monitoring
2532 */
2533 int err;
2534 int backup_list[1];
2535 unsigned long long backup_offsets[1];
2536 int odisks, ndisks, ochunk, nchunk,odata,ndata;
2537 unsigned long a,b,blocks,stripes;
2538 int backup_fd;
2539 int *fds;
2540 unsigned long long *offsets;
2541 int d;
2542 struct mdinfo *sra, *sd;
2543 int rv;
2544 unsigned long cache;
2545 int done = 0;
2546
2547 err = sysfs_set_str(info, NULL, "array_state", "readonly");
2548 if (err)
2549 return err;
2550
2551 /* make sure reshape doesn't progress until we are ready */
2552 sysfs_set_str(info, NULL, "sync_max", "0");
2553 sysfs_set_str(info, NULL, "array_state", "active"); /* FIXME or clean */
2554
2555 sra = sysfs_read(-1, devname2devnum(info->sys_name),
2556 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|
2557 GET_CACHE);
2558 if (!sra)
2559 return 1;
2560
2561 /* ndisks is not growing, so raid_disks is old and +delta is new */
2562 odisks = info->array.raid_disks;
2563 ndisks = odisks + info->delta_disks;
2564 odata = odisks - 1;
2565 ndata = ndisks - 1;
2566 if (info->array.level == 6) {
2567 odata--;
2568 ndata--;
2569 }
2570 ochunk = info->array.chunk_size;
2571 nchunk = info->new_chunk;
2572
2573 a = (ochunk/512) * odata;
2574 b = (nchunk/512) * ndata;
2575 /* Find GCD */
2576 while (a != b) {
2577 if (a < b)
2578 b -= a;
2579 if (b < a)
2580 a -= b;
2581 }
2582 /* LCM == product / GCD */
2583 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
2584
2585 if (ndata == odata)
2586 while (blocks * 32 < sra->component_size &&
2587 blocks < 16*1024*2)
2588 blocks *= 2;
2589 stripes = blocks / (info->array.chunk_size/512) / odata;
2590
2591 /* check that the internal stripe cache is
2592 * large enough, or it won't work.
2593 */
2594 cache = (nchunk < ochunk) ? ochunk : nchunk;
2595 cache = cache * 4 / 4096;
2596 if (cache < blocks / 8 / odisks + 16)
2597 /* Make it big enough to hold 'blocks' */
2598 cache = blocks / 8 / odisks + 16;
2599 if (sra->cache_size < cache)
2600 sysfs_set_num(sra, NULL, "stripe_cache_size",
2601 cache+1);
2602
2603 memset(&bsb, 0, 512);
2604 memcpy(bsb.magic, "md_backup_data-1", 16);
2605 memcpy(&bsb.set_uuid, info->uuid, 16);
2606 bsb.mtime = __cpu_to_le64(time(0));
2607 bsb.devstart2 = blocks;
2608
2609 backup_fd = open(backup_file, O_RDWR|O_CREAT, S_IRUSR | S_IWUSR);
2610 backup_list[0] = backup_fd;
2611 backup_offsets[0] = 8 * 512;
2612 fds = malloc(odisks * sizeof(fds[0]));
2613 offsets = malloc(odisks * sizeof(offsets[0]));
2614 for (d=0; d<odisks; d++)
2615 fds[d] = -1;
2616
2617 for (sd = sra->devs; sd; sd = sd->next) {
2618 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2619 continue;
2620 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
2621 char *dn = map_dev(sd->disk.major,
2622 sd->disk.minor, 1);
2623 fds[sd->disk.raid_disk]
2624 = dev_open(dn, O_RDONLY);
2625 offsets[sd->disk.raid_disk] = sd->data_offset*512;
2626 if (fds[sd->disk.raid_disk] < 0) {
2627 fprintf(stderr, Name ": %s: cannot open component %s\n",
2628 info->sys_name, dn?dn:"-unknown-");
2629 rv = 1;
2630 goto release;
2631 }
2632 free(dn);
2633 }
2634 }
2635
2636 switch(fork()) {
2637 case 0:
2638 close(mdfd);
2639 mlockall(MCL_FUTURE);
2640 if (info->delta_disks < 0)
2641 done = child_shrink(-1, info, stripes,
2642 fds, offsets,
2643 info->array.raid_disks,
2644 info->array.chunk_size,
2645 info->array.level, info->array.layout,
2646 odata,
2647 1, backup_list, backup_offsets);
2648 else if (info->delta_disks == 0) {
2649 /* The 'start' is a per-device stripe number.
2650 * reshape_progress is a per-array sector number.
2651 * So divide by ndata * chunk_size
2652 */
2653 unsigned long long start = info->reshape_progress / ndata;
2654 start /= (info->array.chunk_size/512);
2655 done = child_same_size(-1, info, stripes,
2656 fds, offsets,
2657 start,
2658 info->array.raid_disks,
2659 info->array.chunk_size,
2660 info->array.level, info->array.layout,
2661 odata,
2662 1, backup_list, backup_offsets);
2663 }
2664 if (backup_file && done)
2665 unlink(backup_file);
2666 /* FIXME should I intuit a level change */
2667 exit(0);
2668 case -1:
2669 fprintf(stderr, Name ": Cannot run child to continue monitoring reshape: %s\n",
2670 strerror(errno));
2671 return 1;
2672 default:
2673 break;
2674 }
2675 release:
2676 return 0;
2677 }
2678
2679