]> 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
mdadm/clustermd_tests: add test case to test grow_resize cluster-raid10
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 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 #include <stddef.h>
28 #include <stdint.h>
29 #include <signal.h>
30 #include <sys/wait.h>
31
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
34 #endif
35 #include "md_u.h"
36 #include "md_p.h"
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char **backup_filep,
43 int verbose)
44 {
45 int i;
46 int *fdlist;
47 struct mdinfo *dev;
48 int err;
49 int disk_count = next_spare + working_disks;
50 char *backup_file = *backup_filep;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61
62 sprintf(buf, "%d:%d", dev->disk.major, dev->disk.minor);
63 fd = dev_open(buf, O_RDWR);
64
65 if (dev->disk.raid_disk >= 0)
66 fdlist[dev->disk.raid_disk] = fd;
67 else
68 fdlist[next_spare++] = fd;
69 }
70
71 if (!backup_file) {
72 backup_file = locate_backup(content->sys_name);
73 *backup_filep = backup_file;
74 }
75
76 if (st->ss->external && st->ss->recover_backup)
77 err = st->ss->recover_backup(st, content);
78 else
79 err = Grow_restart(st, content, fdlist, next_spare,
80 backup_file, verbose > 0);
81
82 while (next_spare > 0) {
83 next_spare--;
84 if (fdlist[next_spare] >= 0)
85 close(fdlist[next_spare]);
86 }
87 free(fdlist);
88 if (err) {
89 pr_err("Failed to restore critical section for reshape - sorry.\n");
90 if (!backup_file)
91 pr_err("Possibly you need to specify a --backup-file\n");
92 return 1;
93 }
94
95 dprintf("restore_backup() returns status OK.\n");
96 return 0;
97 }
98
99 int Grow_Add_device(char *devname, int fd, char *newdev)
100 {
101 /* Add a device to an active array.
102 * Currently, just extend a linear array.
103 * This requires writing a new superblock on the
104 * new device, calling the kernel to add the device,
105 * and if that succeeds, update the superblock on
106 * all other devices.
107 * This means that we need to *find* all other devices.
108 */
109 struct mdinfo info;
110
111 dev_t rdev;
112 int nfd, fd2;
113 int d, nd;
114 struct supertype *st = NULL;
115 char *subarray = NULL;
116
117 if (md_get_array_info(fd, &info.array) < 0) {
118 pr_err("cannot get array info for %s\n", devname);
119 return 1;
120 }
121
122 if (info.array.level != -1) {
123 pr_err("can only add devices to linear arrays\n");
124 return 1;
125 }
126
127 st = super_by_fd(fd, &subarray);
128 if (!st) {
129 pr_err("cannot handle arrays with superblock version %d\n",
130 info.array.major_version);
131 return 1;
132 }
133
134 if (subarray) {
135 pr_err("Cannot grow linear sub-arrays yet\n");
136 free(subarray);
137 free(st);
138 return 1;
139 }
140
141 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
142 if (nfd < 0) {
143 pr_err("cannot open %s\n", newdev);
144 free(st);
145 return 1;
146 }
147 if (!fstat_is_blkdev(nfd, newdev, &rdev)) {
148 close(nfd);
149 free(st);
150 return 1;
151 }
152 /* now check out all the devices and make sure we can read the
153 * superblock */
154 for (d=0 ; d < info.array.raid_disks ; d++) {
155 mdu_disk_info_t disk;
156 char *dv;
157
158 st->ss->free_super(st);
159
160 disk.number = d;
161 if (md_get_disk_info(fd, &disk) < 0) {
162 pr_err("cannot get device detail for device %d\n", d);
163 close(nfd);
164 free(st);
165 return 1;
166 }
167 dv = map_dev(disk.major, disk.minor, 1);
168 if (!dv) {
169 pr_err("cannot find device file for device %d\n", d);
170 close(nfd);
171 free(st);
172 return 1;
173 }
174 fd2 = dev_open(dv, O_RDWR);
175 if (fd2 < 0) {
176 pr_err("cannot open device file %s\n", dv);
177 close(nfd);
178 free(st);
179 return 1;
180 }
181
182 if (st->ss->load_super(st, fd2, NULL)) {
183 pr_err("cannot find super block on %s\n", dv);
184 close(nfd);
185 close(fd2);
186 free(st);
187 return 1;
188 }
189 close(fd2);
190 }
191 /* Ok, looks good. Lets update the superblock and write it out to
192 * newdev.
193 */
194
195 info.disk.number = d;
196 info.disk.major = major(rdev);
197 info.disk.minor = minor(rdev);
198 info.disk.raid_disk = d;
199 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
200 st->ss->update_super(st, &info, "linear-grow-new", newdev, 0, 0, NULL);
201
202 if (st->ss->store_super(st, nfd)) {
203 pr_err("Cannot store new superblock on %s\n", newdev);
204 close(nfd);
205 return 1;
206 }
207 close(nfd);
208
209 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
210 pr_err("Cannot add new disk to this array\n");
211 return 1;
212 }
213 /* Well, that seems to have worked.
214 * Now go through and update all superblocks
215 */
216
217 if (md_get_array_info(fd, &info.array) < 0) {
218 pr_err("cannot get array info for %s\n", devname);
219 return 1;
220 }
221
222 nd = d;
223 for (d=0 ; d < info.array.raid_disks ; d++) {
224 mdu_disk_info_t disk;
225 char *dv;
226
227 disk.number = d;
228 if (md_get_disk_info(fd, &disk) < 0) {
229 pr_err("cannot get device detail for device %d\n", d);
230 return 1;
231 }
232 dv = map_dev(disk.major, disk.minor, 1);
233 if (!dv) {
234 pr_err("cannot find device file for device %d\n", d);
235 return 1;
236 }
237 fd2 = dev_open(dv, O_RDWR);
238 if (fd2 < 0) {
239 pr_err("cannot open device file %s\n", dv);
240 return 1;
241 }
242 if (st->ss->load_super(st, fd2, NULL)) {
243 pr_err("cannot find super block on %s\n", dv);
244 close(fd);
245 return 1;
246 }
247 info.array.raid_disks = nd+1;
248 info.array.nr_disks = nd+1;
249 info.array.active_disks = nd+1;
250 info.array.working_disks = nd+1;
251
252 st->ss->update_super(st, &info, "linear-grow-update", dv,
253 0, 0, NULL);
254
255 if (st->ss->store_super(st, fd2)) {
256 pr_err("Cannot store new superblock on %s\n", dv);
257 close(fd2);
258 return 1;
259 }
260 close(fd2);
261 }
262
263 return 0;
264 }
265
266 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
267 {
268 /*
269 * First check that array doesn't have a bitmap
270 * Then create the bitmap
271 * Then add it
272 *
273 * For internal bitmaps, we need to check the version,
274 * find all the active devices, and write the bitmap block
275 * to all devices
276 */
277 mdu_bitmap_file_t bmf;
278 mdu_array_info_t array;
279 struct supertype *st;
280 char *subarray = NULL;
281 int major = BITMAP_MAJOR_HI;
282 unsigned long long bitmapsize, array_size;
283 struct mdinfo *mdi;
284
285 /*
286 * We only ever get called if s->bitmap_file is != NULL, so this check
287 * is just here to quiet down static code checkers.
288 */
289 if (!s->bitmap_file)
290 return 1;
291
292 if (strcmp(s->bitmap_file, "clustered") == 0)
293 major = BITMAP_MAJOR_CLUSTERED;
294
295 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
296 if (errno == ENOMEM)
297 pr_err("Memory allocation failure.\n");
298 else
299 pr_err("bitmaps not supported by this kernel.\n");
300 return 1;
301 }
302 if (bmf.pathname[0]) {
303 if (strcmp(s->bitmap_file,"none") == 0) {
304 if (ioctl(fd, SET_BITMAP_FILE, -1) != 0) {
305 pr_err("failed to remove bitmap %s\n",
306 bmf.pathname);
307 return 1;
308 }
309 return 0;
310 }
311 pr_err("%s already has a bitmap (%s)\n", devname, bmf.pathname);
312 return 1;
313 }
314 if (md_get_array_info(fd, &array) != 0) {
315 pr_err("cannot get array status for %s\n", devname);
316 return 1;
317 }
318 if (array.state & (1 << MD_SB_BITMAP_PRESENT)) {
319 if (strcmp(s->bitmap_file, "none")==0) {
320 array.state &= ~(1 << MD_SB_BITMAP_PRESENT);
321 if (md_set_array_info(fd, &array) != 0) {
322 if (array.state & (1 << MD_SB_CLUSTERED))
323 pr_err("failed to remove clustered bitmap.\n");
324 else
325 pr_err("failed to remove internal bitmap.\n");
326 return 1;
327 }
328 return 0;
329 }
330 pr_err("bitmap already present on %s\n", devname);
331 return 1;
332 }
333
334 if (strcmp(s->bitmap_file, "none") == 0) {
335 pr_err("no bitmap found on %s\n", devname);
336 return 1;
337 }
338 if (array.level <= 0) {
339 pr_err("Bitmaps not meaningful with level %s\n",
340 map_num(pers, array.level)?:"of this array");
341 return 1;
342 }
343 bitmapsize = array.size;
344 bitmapsize <<= 1;
345 if (get_dev_size(fd, NULL, &array_size) &&
346 array_size > (0x7fffffffULL << 9)) {
347 /* Array is big enough that we cannot trust array.size
348 * try other approaches
349 */
350 bitmapsize = get_component_size(fd);
351 }
352 if (bitmapsize == 0) {
353 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
354 return 1;
355 }
356
357 if (array.level == 10) {
358 int ncopies;
359
360 ncopies = (array.layout & 255) * ((array.layout >> 8) & 255);
361 bitmapsize = bitmapsize * array.raid_disks / ncopies;
362
363 if (strcmp(s->bitmap_file, "clustered") == 0 &&
364 !is_near_layout_10(array.layout)) {
365 pr_err("only near layout is supported with clustered raid10\n");
366 return 1;
367 }
368 }
369
370 st = super_by_fd(fd, &subarray);
371 if (!st) {
372 pr_err("Cannot understand version %d.%d\n",
373 array.major_version, array.minor_version);
374 return 1;
375 }
376 if (subarray) {
377 pr_err("Cannot add bitmaps to sub-arrays yet\n");
378 free(subarray);
379 free(st);
380 return 1;
381 }
382
383 mdi = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY);
384 if (mdi) {
385 if (mdi->consistency_policy == CONSISTENCY_POLICY_PPL) {
386 pr_err("Cannot add bitmap to array with PPL\n");
387 free(mdi);
388 free(st);
389 return 1;
390 }
391 free(mdi);
392 }
393
394 if (strcmp(s->bitmap_file, "internal") == 0 ||
395 strcmp(s->bitmap_file, "clustered") == 0) {
396 int rv;
397 int d;
398 int offset_setable = 0;
399 if (st->ss->add_internal_bitmap == NULL) {
400 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
401 return 1;
402 }
403 st->nodes = c->nodes;
404 st->cluster_name = c->homecluster;
405 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
406 if (mdi)
407 offset_setable = 1;
408 for (d = 0; d < st->max_devs; d++) {
409 mdu_disk_info_t disk;
410 char *dv;
411 int fd2;
412
413 disk.number = d;
414 if (md_get_disk_info(fd, &disk) < 0)
415 continue;
416 if (disk.major == 0 && disk.minor == 0)
417 continue;
418 if ((disk.state & (1 << MD_DISK_SYNC)) == 0)
419 continue;
420 dv = map_dev(disk.major, disk.minor, 1);
421 if (!dv)
422 continue;
423 fd2 = dev_open(dv, O_RDWR);
424 if (fd2 < 0)
425 continue;
426 rv = st->ss->load_super(st, fd2, NULL);
427 if (!rv) {
428 rv = st->ss->add_internal_bitmap(
429 st, &s->bitmap_chunk, c->delay,
430 s->write_behind, bitmapsize,
431 offset_setable, major);
432 if (!rv) {
433 st->ss->write_bitmap(st, fd2,
434 NodeNumUpdate);
435 } else {
436 pr_err("failed to create internal bitmap - chunksize problem.\n");
437 }
438 } else {
439 pr_err("failed to load super-block.\n");
440 }
441 close(fd2);
442 if (rv)
443 return 1;
444 }
445 if (offset_setable) {
446 st->ss->getinfo_super(st, mdi, NULL);
447 if (sysfs_init(mdi, fd, NULL)) {
448 pr_err("failed to intialize sysfs.\n");
449 free(mdi);
450 }
451 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
452 mdi->bitmap_offset);
453 free(mdi);
454 } else {
455 if (strcmp(s->bitmap_file, "clustered") == 0)
456 array.state |= (1 << MD_SB_CLUSTERED);
457 array.state |= (1 << MD_SB_BITMAP_PRESENT);
458 rv = md_set_array_info(fd, &array);
459 }
460 if (rv < 0) {
461 if (errno == EBUSY)
462 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
463 pr_err("failed to set internal bitmap.\n");
464 return 1;
465 }
466 } else {
467 int uuid[4];
468 int bitmap_fd;
469 int d;
470 int max_devs = st->max_devs;
471
472 /* try to load a superblock */
473 for (d = 0; d < max_devs; d++) {
474 mdu_disk_info_t disk;
475 char *dv;
476 int fd2;
477 disk.number = d;
478 if (md_get_disk_info(fd, &disk) < 0)
479 continue;
480 if ((disk.major==0 && disk.minor == 0) ||
481 (disk.state & (1 << MD_DISK_REMOVED)))
482 continue;
483 dv = map_dev(disk.major, disk.minor, 1);
484 if (!dv)
485 continue;
486 fd2 = dev_open(dv, O_RDONLY);
487 if (fd2 >= 0) {
488 if (st->ss->load_super(st, fd2, NULL) == 0) {
489 close(fd2);
490 st->ss->uuid_from_super(st, uuid);
491 break;
492 }
493 close(fd2);
494 }
495 }
496 if (d == max_devs) {
497 pr_err("cannot find UUID for array!\n");
498 return 1;
499 }
500 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid,
501 s->bitmap_chunk, c->delay, s->write_behind,
502 bitmapsize, major)) {
503 return 1;
504 }
505 bitmap_fd = open(s->bitmap_file, O_RDWR);
506 if (bitmap_fd < 0) {
507 pr_err("weird: %s cannot be opened\n", s->bitmap_file);
508 return 1;
509 }
510 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
511 int err = errno;
512 if (errno == EBUSY)
513 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
514 pr_err("Cannot set bitmap file for %s: %s\n",
515 devname, strerror(err));
516 return 1;
517 }
518 }
519
520 return 0;
521 }
522
523 int Grow_consistency_policy(char *devname, int fd, struct context *c, struct shape *s)
524 {
525 struct supertype *st;
526 struct mdinfo *sra;
527 struct mdinfo *sd;
528 char *subarray = NULL;
529 int ret = 0;
530 char container_dev[PATH_MAX];
531 char buf[20];
532
533 if (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
534 s->consistency_policy != CONSISTENCY_POLICY_PPL) {
535 pr_err("Operation not supported for consistency policy %s\n",
536 map_num(consistency_policies, s->consistency_policy));
537 return 1;
538 }
539
540 st = super_by_fd(fd, &subarray);
541 if (!st)
542 return 1;
543
544 sra = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY|GET_LEVEL|
545 GET_DEVS|GET_STATE);
546 if (!sra) {
547 ret = 1;
548 goto free_st;
549 }
550
551 if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
552 !st->ss->write_init_ppl) {
553 pr_err("%s metadata does not support PPL\n", st->ss->name);
554 ret = 1;
555 goto free_info;
556 }
557
558 if (sra->array.level != 5) {
559 pr_err("Operation not supported for array level %d\n",
560 sra->array.level);
561 ret = 1;
562 goto free_info;
563 }
564
565 if (sra->consistency_policy == (unsigned)s->consistency_policy) {
566 pr_err("Consistency policy is already %s\n",
567 map_num(consistency_policies, s->consistency_policy));
568 ret = 1;
569 goto free_info;
570 } else if (sra->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
571 sra->consistency_policy != CONSISTENCY_POLICY_PPL) {
572 pr_err("Current consistency policy is %s, cannot change to %s\n",
573 map_num(consistency_policies, sra->consistency_policy),
574 map_num(consistency_policies, s->consistency_policy));
575 ret = 1;
576 goto free_info;
577 }
578
579 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
580 if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0) {
581 ret = 1;
582 goto free_info;
583 } else if (strcmp(buf, "reshape\n") == 0) {
584 pr_err("PPL cannot be enabled when reshape is in progress\n");
585 ret = 1;
586 goto free_info;
587 }
588 }
589
590 if (subarray) {
591 char *update;
592
593 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
594 update = "ppl";
595 else
596 update = "no-ppl";
597
598 sprintf(container_dev, "/dev/%s", st->container_devnm);
599
600 ret = Update_subarray(container_dev, subarray, update, NULL,
601 c->verbose);
602 if (ret)
603 goto free_info;
604 }
605
606 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
607 struct mdinfo info;
608
609 if (subarray) {
610 struct mdinfo *mdi;
611 int cfd;
612
613 cfd = open(container_dev, O_RDWR|O_EXCL);
614 if (cfd < 0) {
615 pr_err("Failed to open %s\n", container_dev);
616 ret = 1;
617 goto free_info;
618 }
619
620 ret = st->ss->load_container(st, cfd, st->container_devnm);
621 close(cfd);
622
623 if (ret) {
624 pr_err("Cannot read superblock for %s\n",
625 container_dev);
626 goto free_info;
627 }
628
629 mdi = st->ss->container_content(st, subarray);
630 info = *mdi;
631 free(mdi);
632 }
633
634 for (sd = sra->devs; sd; sd = sd->next) {
635 int dfd;
636 char *devpath;
637
638 devpath = map_dev(sd->disk.major, sd->disk.minor, 0);
639 dfd = dev_open(devpath, O_RDWR);
640 if (dfd < 0) {
641 pr_err("Failed to open %s\n", devpath);
642 ret = 1;
643 goto free_info;
644 }
645
646 if (!subarray) {
647 ret = st->ss->load_super(st, dfd, NULL);
648 if (ret) {
649 pr_err("Failed to load super-block.\n");
650 close(dfd);
651 goto free_info;
652 }
653
654 ret = st->ss->update_super(st, sra, "ppl",
655 devname,
656 c->verbose, 0, NULL);
657 if (ret) {
658 close(dfd);
659 st->ss->free_super(st);
660 goto free_info;
661 }
662 st->ss->getinfo_super(st, &info, NULL);
663 }
664
665 ret |= sysfs_set_num(sra, sd, "ppl_sector",
666 info.ppl_sector);
667 ret |= sysfs_set_num(sra, sd, "ppl_size",
668 info.ppl_size);
669
670 if (ret) {
671 pr_err("Failed to set PPL attributes for %s\n",
672 sd->sys_name);
673 close(dfd);
674 st->ss->free_super(st);
675 goto free_info;
676 }
677
678 ret = st->ss->write_init_ppl(st, &info, dfd);
679 if (ret)
680 pr_err("Failed to write PPL\n");
681
682 close(dfd);
683
684 if (!subarray)
685 st->ss->free_super(st);
686
687 if (ret)
688 goto free_info;
689 }
690 }
691
692 ret = sysfs_set_str(sra, NULL, "consistency_policy",
693 map_num(consistency_policies,
694 s->consistency_policy));
695 if (ret)
696 pr_err("Failed to change array consistency policy\n");
697
698 free_info:
699 sysfs_free(sra);
700 free_st:
701 free(st);
702 free(subarray);
703
704 return ret;
705 }
706
707 /*
708 * When reshaping an array we might need to backup some data.
709 * This is written to all spares with a 'super_block' describing it.
710 * The superblock goes 4K from the end of the used space on the
711 * device.
712 * It if written after the backup is complete.
713 * It has the following structure.
714 */
715
716 static struct mdp_backup_super {
717 char magic[16]; /* md_backup_data-1 or -2 */
718 __u8 set_uuid[16];
719 __u64 mtime;
720 /* start/sizes in 512byte sectors */
721 __u64 devstart; /* address on backup device/file of data */
722 __u64 arraystart;
723 __u64 length;
724 __u32 sb_csum; /* csum of preceeding bytes. */
725 __u32 pad1;
726 __u64 devstart2; /* offset in to data of second section */
727 __u64 arraystart2;
728 __u64 length2;
729 __u32 sb_csum2; /* csum of preceeding bytes. */
730 __u8 pad[512-68-32];
731 } __attribute__((aligned(512))) bsb, bsb2;
732
733 static __u32 bsb_csum(char *buf, int len)
734 {
735 int i;
736 int csum = 0;
737 for (i = 0; i < len; i++)
738 csum = (csum<<3) + buf[0];
739 return __cpu_to_le32(csum);
740 }
741
742 static int check_idle(struct supertype *st)
743 {
744 /* Check that all member arrays for this container, or the
745 * container of this array, are idle
746 */
747 char *container = (st->container_devnm[0]
748 ? st->container_devnm : st->devnm);
749 struct mdstat_ent *ent, *e;
750 int is_idle = 1;
751
752 ent = mdstat_read(0, 0);
753 for (e = ent ; e; e = e->next) {
754 if (!is_container_member(e, container))
755 continue;
756 if (e->percent >= 0) {
757 is_idle = 0;
758 break;
759 }
760 }
761 free_mdstat(ent);
762 return is_idle;
763 }
764
765 static int freeze_container(struct supertype *st)
766 {
767 char *container = (st->container_devnm[0]
768 ? st->container_devnm : st->devnm);
769
770 if (!check_idle(st))
771 return -1;
772
773 if (block_monitor(container, 1)) {
774 pr_err("failed to freeze container\n");
775 return -2;
776 }
777
778 return 1;
779 }
780
781 static void unfreeze_container(struct supertype *st)
782 {
783 char *container = (st->container_devnm[0]
784 ? st->container_devnm : st->devnm);
785
786 unblock_monitor(container, 1);
787 }
788
789 static int freeze(struct supertype *st)
790 {
791 /* Try to freeze resync/rebuild on this array/container.
792 * Return -1 if the array is busy,
793 * return -2 container cannot be frozen,
794 * return 0 if this kernel doesn't support 'frozen'
795 * return 1 if it worked.
796 */
797 if (st->ss->external)
798 return freeze_container(st);
799 else {
800 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
801 int err;
802 char buf[20];
803
804 if (!sra)
805 return -1;
806 /* Need to clear any 'read-auto' status */
807 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
808 strncmp(buf, "read-auto", 9) == 0)
809 sysfs_set_str(sra, NULL, "array_state", "clean");
810
811 err = sysfs_freeze_array(sra);
812 sysfs_free(sra);
813 return err;
814 }
815 }
816
817 static void unfreeze(struct supertype *st)
818 {
819 if (st->ss->external)
820 return unfreeze_container(st);
821 else {
822 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
823 char buf[20];
824
825 if (sra &&
826 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 &&
827 strcmp(buf, "frozen\n") == 0)
828 sysfs_set_str(sra, NULL, "sync_action", "idle");
829 sysfs_free(sra);
830 }
831 }
832
833 static void wait_reshape(struct mdinfo *sra)
834 {
835 int fd = sysfs_get_fd(sra, NULL, "sync_action");
836 char action[20];
837
838 if (fd < 0)
839 return;
840
841 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
842 strncmp(action, "reshape", 7) == 0)
843 sysfs_wait(fd, NULL);
844 close(fd);
845 }
846
847 static int reshape_super(struct supertype *st, unsigned long long size,
848 int level, int layout, int chunksize, int raid_disks,
849 int delta_disks, char *backup_file, char *dev,
850 int direction, int verbose)
851 {
852 /* nothing extra to check in the native case */
853 if (!st->ss->external)
854 return 0;
855 if (!st->ss->reshape_super || !st->ss->manage_reshape) {
856 pr_err("%s metadata does not support reshape\n",
857 st->ss->name);
858 return 1;
859 }
860
861 return st->ss->reshape_super(st, size, level, layout, chunksize,
862 raid_disks, delta_disks, backup_file, dev,
863 direction, verbose);
864 }
865
866 static void sync_metadata(struct supertype *st)
867 {
868 if (st->ss->external) {
869 if (st->update_tail) {
870 flush_metadata_updates(st);
871 st->update_tail = &st->updates;
872 } else
873 st->ss->sync_metadata(st);
874 }
875 }
876
877 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
878 {
879 /* when dealing with external metadata subarrays we need to be
880 * prepared to handle EAGAIN. The kernel may need to wait for
881 * mdmon to mark the array active so the kernel can handle
882 * allocations/writeback when preparing the reshape action
883 * (md_allow_write()). We temporarily disable safe_mode_delay
884 * to close a race with the array_state going clean before the
885 * next write to raid_disks / stripe_cache_size
886 */
887 char safe[50];
888 int rc;
889
890 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
891 if (!container ||
892 (strcmp(name, "raid_disks") != 0 &&
893 strcmp(name, "stripe_cache_size") != 0))
894 return sysfs_set_num(sra, NULL, name, n);
895
896 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
897 if (rc <= 0)
898 return -1;
899 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
900 rc = sysfs_set_num(sra, NULL, name, n);
901 if (rc < 0 && errno == EAGAIN) {
902 ping_monitor(container);
903 /* if we get EAGAIN here then the monitor is not active
904 * so stop trying
905 */
906 rc = sysfs_set_num(sra, NULL, name, n);
907 }
908 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
909 return rc;
910 }
911
912 int start_reshape(struct mdinfo *sra, int already_running,
913 int before_data_disks, int data_disks)
914 {
915 int err;
916 unsigned long long sync_max_to_set;
917
918 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
919 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
920 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
921 sra->reshape_progress);
922 if (before_data_disks <= data_disks)
923 sync_max_to_set = sra->reshape_progress / data_disks;
924 else
925 sync_max_to_set = (sra->component_size * data_disks
926 - sra->reshape_progress) / data_disks;
927 if (!already_running)
928 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
929 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
930 if (!already_running && err == 0) {
931 int cnt = 5;
932 do {
933 err = sysfs_set_str(sra, NULL, "sync_action",
934 "reshape");
935 if (err)
936 sleep(1);
937 } while (err && errno == EBUSY && cnt-- > 0);
938 }
939 return err;
940 }
941
942 void abort_reshape(struct mdinfo *sra)
943 {
944 sysfs_set_str(sra, NULL, "sync_action", "idle");
945 /*
946 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
947 * suspend_hi to decrease as well as increase.")
948 * you could only increase suspend_{lo,hi} unless the region they
949 * covered was empty. So to reset to 0, you need to push suspend_lo
950 * up past suspend_hi first. So to maximize the chance of mdadm
951 * working on all kernels, we want to keep doing that.
952 */
953 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
954 sysfs_set_num(sra, NULL, "suspend_hi", 0);
955 sysfs_set_num(sra, NULL, "suspend_lo", 0);
956 sysfs_set_num(sra, NULL, "sync_min", 0);
957 // It isn't safe to reset sync_max as we aren't monitoring.
958 // Array really should be stopped at this point.
959 }
960
961 int remove_disks_for_takeover(struct supertype *st,
962 struct mdinfo *sra,
963 int layout)
964 {
965 int nr_of_copies;
966 struct mdinfo *remaining;
967 int slot;
968
969 if (st->ss->external) {
970 int rv = 0;
971 struct mdinfo *arrays = st->ss->container_content(st, NULL);
972 /*
973 * containter_content returns list of arrays in container
974 * If arrays->next is not NULL it means that there are
975 * 2 arrays in container and operation should be blocked
976 */
977 if (arrays) {
978 if (arrays->next)
979 rv = 1;
980 sysfs_free(arrays);
981 if (rv) {
982 pr_err("Error. Cannot perform operation on /dev/%s\n", st->devnm);
983 pr_err("For this operation it MUST be single array in container\n");
984 return rv;
985 }
986 }
987 }
988
989 if (sra->array.level == 10)
990 nr_of_copies = layout & 0xff;
991 else if (sra->array.level == 1)
992 nr_of_copies = sra->array.raid_disks;
993 else
994 return 1;
995
996 remaining = sra->devs;
997 sra->devs = NULL;
998 /* for each 'copy', select one device and remove from the list. */
999 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
1000 struct mdinfo **diskp;
1001 int found = 0;
1002
1003 /* Find a working device to keep */
1004 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
1005 struct mdinfo *disk = *diskp;
1006
1007 if (disk->disk.raid_disk < slot)
1008 continue;
1009 if (disk->disk.raid_disk >= slot + nr_of_copies)
1010 continue;
1011 if (disk->disk.state & (1<<MD_DISK_REMOVED))
1012 continue;
1013 if (disk->disk.state & (1<<MD_DISK_FAULTY))
1014 continue;
1015 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
1016 continue;
1017
1018 /* We have found a good disk to use! */
1019 *diskp = disk->next;
1020 disk->next = sra->devs;
1021 sra->devs = disk;
1022 found = 1;
1023 break;
1024 }
1025 if (!found)
1026 break;
1027 }
1028
1029 if (slot < sra->array.raid_disks) {
1030 /* didn't find all slots */
1031 struct mdinfo **e;
1032 e = &remaining;
1033 while (*e)
1034 e = &(*e)->next;
1035 *e = sra->devs;
1036 sra->devs = remaining;
1037 return 1;
1038 }
1039
1040 /* Remove all 'remaining' devices from the array */
1041 while (remaining) {
1042 struct mdinfo *sd = remaining;
1043 remaining = sd->next;
1044
1045 sysfs_set_str(sra, sd, "state", "faulty");
1046 sysfs_set_str(sra, sd, "slot", "none");
1047 /* for external metadata disks should be removed in mdmon */
1048 if (!st->ss->external)
1049 sysfs_set_str(sra, sd, "state", "remove");
1050 sd->disk.state |= (1<<MD_DISK_REMOVED);
1051 sd->disk.state &= ~(1<<MD_DISK_SYNC);
1052 sd->next = sra->devs;
1053 sra->devs = sd;
1054 }
1055 return 0;
1056 }
1057
1058 void reshape_free_fdlist(int *fdlist,
1059 unsigned long long *offsets,
1060 int size)
1061 {
1062 int i;
1063
1064 for (i = 0; i < size; i++)
1065 if (fdlist[i] >= 0)
1066 close(fdlist[i]);
1067
1068 free(fdlist);
1069 free(offsets);
1070 }
1071
1072 int reshape_prepare_fdlist(char *devname,
1073 struct mdinfo *sra,
1074 int raid_disks,
1075 int nrdisks,
1076 unsigned long blocks,
1077 char *backup_file,
1078 int *fdlist,
1079 unsigned long long *offsets)
1080 {
1081 int d = 0;
1082 struct mdinfo *sd;
1083
1084 enable_fds(nrdisks);
1085 for (d = 0; d <= nrdisks; d++)
1086 fdlist[d] = -1;
1087 d = raid_disks;
1088 for (sd = sra->devs; sd; sd = sd->next) {
1089 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1090 continue;
1091 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
1092 sd->disk.raid_disk < raid_disks) {
1093 char *dn = map_dev(sd->disk.major, sd->disk.minor, 1);
1094 fdlist[sd->disk.raid_disk] = dev_open(dn, O_RDONLY);
1095 offsets[sd->disk.raid_disk] = sd->data_offset*512;
1096 if (fdlist[sd->disk.raid_disk] < 0) {
1097 pr_err("%s: cannot open component %s\n",
1098 devname, dn ? dn : "-unknown-");
1099 d = -1;
1100 goto release;
1101 }
1102 } else if (backup_file == NULL) {
1103 /* spare */
1104 char *dn = map_dev(sd->disk.major, sd->disk.minor, 1);
1105 fdlist[d] = dev_open(dn, O_RDWR);
1106 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
1107 if (fdlist[d] < 0) {
1108 pr_err("%s: cannot open component %s\n",
1109 devname, dn ? dn : "-unknown-");
1110 d = -1;
1111 goto release;
1112 }
1113 d++;
1114 }
1115 }
1116 release:
1117 return d;
1118 }
1119
1120 int reshape_open_backup_file(char *backup_file,
1121 int fd,
1122 char *devname,
1123 long blocks,
1124 int *fdlist,
1125 unsigned long long *offsets,
1126 char *sys_name,
1127 int restart)
1128 {
1129 /* Return 1 on success, 0 on any form of failure */
1130 /* need to check backup file is large enough */
1131 char buf[512];
1132 struct stat stb;
1133 unsigned int dev;
1134 int i;
1135
1136 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
1137 S_IRUSR | S_IWUSR);
1138 *offsets = 8 * 512;
1139 if (*fdlist < 0) {
1140 pr_err("%s: cannot create backup file %s: %s\n",
1141 devname, backup_file, strerror(errno));
1142 return 0;
1143 }
1144 /* Guard against backup file being on array device.
1145 * If array is partitioned or if LVM etc is in the
1146 * way this will not notice, but it is better than
1147 * nothing.
1148 */
1149 fstat(*fdlist, &stb);
1150 dev = stb.st_dev;
1151 fstat(fd, &stb);
1152 if (stb.st_rdev == dev) {
1153 pr_err("backup file must NOT be on the array being reshaped.\n");
1154 close(*fdlist);
1155 return 0;
1156 }
1157
1158 memset(buf, 0, 512);
1159 for (i=0; i < blocks + 8 ; i++) {
1160 if (write(*fdlist, buf, 512) != 512) {
1161 pr_err("%s: cannot create backup file %s: %s\n",
1162 devname, backup_file, strerror(errno));
1163 return 0;
1164 }
1165 }
1166 if (fsync(*fdlist) != 0) {
1167 pr_err("%s: cannot create backup file %s: %s\n",
1168 devname, backup_file, strerror(errno));
1169 return 0;
1170 }
1171
1172 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
1173 char *bu = make_backup(sys_name);
1174 if (symlink(backup_file, bu))
1175 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
1176 strerror(errno));
1177 free(bu);
1178 }
1179
1180 return 1;
1181 }
1182
1183 unsigned long compute_backup_blocks(int nchunk, int ochunk,
1184 unsigned int ndata, unsigned int odata)
1185 {
1186 unsigned long a, b, blocks;
1187 /* So how much do we need to backup.
1188 * We need an amount of data which is both a whole number of
1189 * old stripes and a whole number of new stripes.
1190 * So LCM for (chunksize*datadisks).
1191 */
1192 a = (ochunk/512) * odata;
1193 b = (nchunk/512) * ndata;
1194 /* Find GCD */
1195 a = GCD(a, b);
1196 /* LCM == product / GCD */
1197 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
1198
1199 return blocks;
1200 }
1201
1202 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
1203 {
1204 /* Based on the current array state in info->array and
1205 * the changes in info->new_* etc, determine:
1206 * - whether the change is possible
1207 * - Intermediate level/raid_disks/layout
1208 * - whether a restriping reshape is needed
1209 * - number of sectors in minimum change unit. This
1210 * will cover a whole number of stripes in 'before' and
1211 * 'after'.
1212 *
1213 * Return message if the change should be rejected
1214 * NULL if the change can be achieved
1215 *
1216 * This can be called as part of starting a reshape, or
1217 * when assembling an array that is undergoing reshape.
1218 */
1219 int near, far, offset, copies;
1220 int new_disks;
1221 int old_chunk, new_chunk;
1222 /* delta_parity records change in number of devices
1223 * caused by level change
1224 */
1225 int delta_parity = 0;
1226
1227 memset(re, 0, sizeof(*re));
1228
1229 /* If a new level not explicitly given, we assume no-change */
1230 if (info->new_level == UnSet)
1231 info->new_level = info->array.level;
1232
1233 if (info->new_chunk)
1234 switch (info->new_level) {
1235 case 0:
1236 case 4:
1237 case 5:
1238 case 6:
1239 case 10:
1240 /* chunk size is meaningful, must divide component_size
1241 * evenly
1242 */
1243 if (info->component_size % (info->new_chunk/512)) {
1244 unsigned long long shrink = info->component_size;
1245 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1246 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1247 info->new_chunk/1024, info->component_size/2);
1248 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1249 devname, shrink/2);
1250 pr_err("will shrink the array so the given chunk size would work.\n");
1251 return "";
1252 }
1253 break;
1254 default:
1255 return "chunk size not meaningful for this level";
1256 }
1257 else
1258 info->new_chunk = info->array.chunk_size;
1259
1260 switch (info->array.level) {
1261 default:
1262 return "No reshape is possibly for this RAID level";
1263 case LEVEL_LINEAR:
1264 if (info->delta_disks != UnSet)
1265 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1266 else
1267 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1268 case 1:
1269 /* RAID1 can convert to RAID1 with different disks, or
1270 * raid5 with 2 disks, or
1271 * raid0 with 1 disk
1272 */
1273 if (info->new_level > 1 && (info->component_size & 7))
1274 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1275 if (info->new_level == 0) {
1276 if (info->delta_disks != UnSet &&
1277 info->delta_disks != 0)
1278 return "Cannot change number of disks with RAID1->RAID0 conversion";
1279 re->level = 0;
1280 re->before.data_disks = 1;
1281 re->after.data_disks = 1;
1282 return NULL;
1283 }
1284 if (info->new_level == 1) {
1285 if (info->delta_disks == UnSet)
1286 /* Don't know what to do */
1287 return "no change requested for Growing RAID1";
1288 re->level = 1;
1289 return NULL;
1290 }
1291 if (info->array.raid_disks != 2 && info->new_level == 5)
1292 return "Can only convert a 2-device array to RAID5";
1293 if (info->array.raid_disks == 2 && info->new_level == 5) {
1294 re->level = 5;
1295 re->before.data_disks = 1;
1296 if (info->delta_disks != UnSet &&
1297 info->delta_disks != 0)
1298 re->after.data_disks = 1 + info->delta_disks;
1299 else
1300 re->after.data_disks = 1;
1301 if (re->after.data_disks < 1)
1302 return "Number of disks too small for RAID5";
1303
1304 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1305 info->array.chunk_size = 65536;
1306 break;
1307 }
1308 /* Could do some multi-stage conversions, but leave that to
1309 * later.
1310 */
1311 return "Impossibly level change request for RAID1";
1312
1313 case 10:
1314 /* RAID10 can be converted from near mode to
1315 * RAID0 by removing some devices.
1316 * It can also be reshaped if the kernel supports
1317 * new_data_offset.
1318 */
1319 switch (info->new_level) {
1320 case 0:
1321 if ((info->array.layout & ~0xff) != 0x100)
1322 return "Cannot Grow RAID10 with far/offset layout";
1323 /*
1324 * number of devices must be multiple of
1325 * number of copies
1326 */
1327 if (info->array.raid_disks %
1328 (info->array.layout & 0xff))
1329 return "RAID10 layout too complex for Grow operation";
1330
1331 new_disks = (info->array.raid_disks /
1332 (info->array.layout & 0xff));
1333 if (info->delta_disks == UnSet)
1334 info->delta_disks = (new_disks
1335 - info->array.raid_disks);
1336
1337 if (info->delta_disks !=
1338 new_disks - info->array.raid_disks)
1339 return "New number of raid-devices impossible for RAID10";
1340 if (info->new_chunk &&
1341 info->new_chunk != info->array.chunk_size)
1342 return "Cannot change chunk-size with RAID10 Grow";
1343
1344 /* looks good */
1345 re->level = 0;
1346 re->before.data_disks = new_disks;
1347 re->after.data_disks = re->before.data_disks;
1348 return NULL;
1349
1350 case 10:
1351 near = info->array.layout & 0xff;
1352 far = (info->array.layout >> 8) & 0xff;
1353 offset = info->array.layout & 0x10000;
1354 if (far > 1 && !offset)
1355 return "Cannot reshape RAID10 in far-mode";
1356 copies = near * far;
1357
1358 old_chunk = info->array.chunk_size * far;
1359
1360 if (info->new_layout == UnSet)
1361 info->new_layout = info->array.layout;
1362 else {
1363 near = info->new_layout & 0xff;
1364 far = (info->new_layout >> 8) & 0xff;
1365 offset = info->new_layout & 0x10000;
1366 if (far > 1 && !offset)
1367 return "Cannot reshape RAID10 to far-mode";
1368 if (near * far != copies)
1369 return "Cannot change number of copies when reshaping RAID10";
1370 }
1371 if (info->delta_disks == UnSet)
1372 info->delta_disks = 0;
1373 new_disks = (info->array.raid_disks +
1374 info->delta_disks);
1375
1376 new_chunk = info->new_chunk * far;
1377
1378 re->level = 10;
1379 re->before.layout = info->array.layout;
1380 re->before.data_disks = info->array.raid_disks;
1381 re->after.layout = info->new_layout;
1382 re->after.data_disks = new_disks;
1383 /* For RAID10 we don't do backup but do allow reshape,
1384 * so set backup_blocks to INVALID_SECTORS rather than
1385 * zero.
1386 * And there is no need to synchronise stripes on both
1387 * 'old' and 'new'. So the important
1388 * number is the minimum data_offset difference
1389 * which is the larger of (offset copies * chunk).
1390 */
1391 re->backup_blocks = INVALID_SECTORS;
1392 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1393 if (new_disks < re->before.data_disks &&
1394 info->space_after < re->min_offset_change)
1395 /* Reduce component size by one chunk */
1396 re->new_size = (info->component_size -
1397 re->min_offset_change);
1398 else
1399 re->new_size = info->component_size;
1400 re->new_size = re->new_size * new_disks / copies;
1401 return NULL;
1402
1403 default:
1404 return "RAID10 can only be changed to RAID0";
1405 }
1406 case 0:
1407 /* RAID0 can be converted to RAID10, or to RAID456 */
1408 if (info->new_level == 10) {
1409 if (info->new_layout == UnSet &&
1410 info->delta_disks == UnSet) {
1411 /* Assume near=2 layout */
1412 info->new_layout = 0x102;
1413 info->delta_disks = info->array.raid_disks;
1414 }
1415 if (info->new_layout == UnSet) {
1416 int copies = 1 + (info->delta_disks
1417 / info->array.raid_disks);
1418 if (info->array.raid_disks * (copies-1) !=
1419 info->delta_disks)
1420 return "Impossible number of devices for RAID0->RAID10";
1421 info->new_layout = 0x100 + copies;
1422 }
1423 if (info->delta_disks == UnSet) {
1424 int copies = info->new_layout & 0xff;
1425 if (info->new_layout != 0x100 + copies)
1426 return "New layout impossible for RAID0->RAID10";;
1427 info->delta_disks = (copies - 1) *
1428 info->array.raid_disks;
1429 }
1430 if (info->new_chunk &&
1431 info->new_chunk != info->array.chunk_size)
1432 return "Cannot change chunk-size with RAID0->RAID10";
1433 /* looks good */
1434 re->level = 10;
1435 re->before.data_disks = (info->array.raid_disks +
1436 info->delta_disks);
1437 re->after.data_disks = re->before.data_disks;
1438 re->before.layout = info->new_layout;
1439 return NULL;
1440 }
1441
1442 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1443 * a raid4 style layout of the final level.
1444 */
1445 switch (info->new_level) {
1446 case 4:
1447 delta_parity = 1;
1448 case 0:
1449 re->level = 4;
1450 re->before.layout = 0;
1451 break;
1452 case 5:
1453 delta_parity = 1;
1454 re->level = 5;
1455 re->before.layout = ALGORITHM_PARITY_N;
1456 if (info->new_layout == UnSet)
1457 info->new_layout = map_name(r5layout, "default");
1458 break;
1459 case 6:
1460 delta_parity = 2;
1461 re->level = 6;
1462 re->before.layout = ALGORITHM_PARITY_N;
1463 if (info->new_layout == UnSet)
1464 info->new_layout = map_name(r6layout, "default");
1465 break;
1466 default:
1467 return "Impossible level change requested";
1468 }
1469 re->before.data_disks = info->array.raid_disks;
1470 /* determining 'after' layout happens outside this 'switch' */
1471 break;
1472
1473 case 4:
1474 info->array.layout = ALGORITHM_PARITY_N;
1475 case 5:
1476 switch (info->new_level) {
1477 case 0:
1478 delta_parity = -1;
1479 case 4:
1480 re->level = info->array.level;
1481 re->before.data_disks = info->array.raid_disks - 1;
1482 re->before.layout = info->array.layout;
1483 break;
1484 case 5:
1485 re->level = 5;
1486 re->before.data_disks = info->array.raid_disks - 1;
1487 re->before.layout = info->array.layout;
1488 break;
1489 case 6:
1490 delta_parity = 1;
1491 re->level = 6;
1492 re->before.data_disks = info->array.raid_disks - 1;
1493 switch (info->array.layout) {
1494 case ALGORITHM_LEFT_ASYMMETRIC:
1495 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1496 break;
1497 case ALGORITHM_RIGHT_ASYMMETRIC:
1498 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1499 break;
1500 case ALGORITHM_LEFT_SYMMETRIC:
1501 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1502 break;
1503 case ALGORITHM_RIGHT_SYMMETRIC:
1504 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1505 break;
1506 case ALGORITHM_PARITY_0:
1507 re->before.layout = ALGORITHM_PARITY_0_6;
1508 break;
1509 case ALGORITHM_PARITY_N:
1510 re->before.layout = ALGORITHM_PARITY_N_6;
1511 break;
1512 default:
1513 return "Cannot convert an array with this layout";
1514 }
1515 break;
1516 case 1:
1517 if (info->array.raid_disks != 2)
1518 return "Can only convert a 2-device array to RAID1";
1519 if (info->delta_disks != UnSet &&
1520 info->delta_disks != 0)
1521 return "Cannot set raid_disk when converting RAID5->RAID1";
1522 re->level = 1;
1523 info->new_chunk = 0;
1524 return NULL;
1525 default:
1526 return "Impossible level change requested";
1527 }
1528 break;
1529 case 6:
1530 switch (info->new_level) {
1531 case 4:
1532 case 5:
1533 delta_parity = -1;
1534 case 6:
1535 re->level = 6;
1536 re->before.data_disks = info->array.raid_disks - 2;
1537 re->before.layout = info->array.layout;
1538 break;
1539 default:
1540 return "Impossible level change requested";
1541 }
1542 break;
1543 }
1544
1545 /* If we reached here then it looks like a re-stripe is
1546 * happening. We have determined the intermediate level
1547 * and initial raid_disks/layout and stored these in 're'.
1548 *
1549 * We need to deduce the final layout that can be atomically
1550 * converted to the end state.
1551 */
1552 switch (info->new_level) {
1553 case 0:
1554 /* We can only get to RAID0 from RAID4 or RAID5
1555 * with appropriate layout and one extra device
1556 */
1557 if (re->level != 4 && re->level != 5)
1558 return "Cannot covert to RAID0 from this level";
1559
1560 switch (re->level) {
1561 case 4:
1562 re->before.layout = 0;
1563 re->after.layout = 0;
1564 break;
1565 case 5:
1566 re->after.layout = ALGORITHM_PARITY_N;
1567 break;
1568 }
1569 break;
1570
1571 case 4:
1572 /* We can only get to RAID4 from RAID5 */
1573 if (re->level != 4 && re->level != 5)
1574 return "Cannot convert to RAID4 from this level";
1575
1576 switch (re->level) {
1577 case 4:
1578 re->after.layout = 0;
1579 break;
1580 case 5:
1581 re->after.layout = ALGORITHM_PARITY_N;
1582 break;
1583 }
1584 break;
1585
1586 case 5:
1587 /* We get to RAID5 from RAID5 or RAID6 */
1588 if (re->level != 5 && re->level != 6)
1589 return "Cannot convert to RAID5 from this level";
1590
1591 switch (re->level) {
1592 case 5:
1593 if (info->new_layout == UnSet)
1594 re->after.layout = re->before.layout;
1595 else
1596 re->after.layout = info->new_layout;
1597 break;
1598 case 6:
1599 if (info->new_layout == UnSet)
1600 info->new_layout = re->before.layout;
1601
1602 /* after.layout needs to be raid6 version of new_layout */
1603 if (info->new_layout == ALGORITHM_PARITY_N)
1604 re->after.layout = ALGORITHM_PARITY_N;
1605 else {
1606 char layout[40];
1607 char *ls = map_num(r5layout, info->new_layout);
1608 int l;
1609 if (ls) {
1610 /* Current RAID6 layout has a RAID5
1611 * equivalent - good
1612 */
1613 strcat(strcpy(layout, ls), "-6");
1614 l = map_name(r6layout, layout);
1615 if (l == UnSet)
1616 return "Cannot find RAID6 layout to convert to";
1617 } else {
1618 /* Current RAID6 has no equivalent.
1619 * If it is already a '-6' layout we
1620 * can leave it unchanged, else we must
1621 * fail
1622 */
1623 ls = map_num(r6layout,
1624 info->new_layout);
1625 if (!ls ||
1626 strcmp(ls+strlen(ls)-2, "-6") != 0)
1627 return "Please specify new layout";
1628 l = info->new_layout;
1629 }
1630 re->after.layout = l;
1631 }
1632 }
1633 break;
1634
1635 case 6:
1636 /* We must already be at level 6 */
1637 if (re->level != 6)
1638 return "Impossible level change";
1639 if (info->new_layout == UnSet)
1640 re->after.layout = info->array.layout;
1641 else
1642 re->after.layout = info->new_layout;
1643 break;
1644 default:
1645 return "Impossible level change requested";
1646 }
1647 if (info->delta_disks == UnSet)
1648 info->delta_disks = delta_parity;
1649
1650 re->after.data_disks =
1651 (re->before.data_disks + info->delta_disks - delta_parity);
1652
1653 switch (re->level) {
1654 case 6:
1655 re->parity = 2;
1656 break;
1657 case 4:
1658 case 5:
1659 re->parity = 1;
1660 break;
1661 default:
1662 re->parity = 0;
1663 break;
1664 }
1665 /* So we have a restripe operation, we need to calculate the number
1666 * of blocks per reshape operation.
1667 */
1668 re->new_size = info->component_size * re->before.data_disks;
1669 if (info->new_chunk == 0)
1670 info->new_chunk = info->array.chunk_size;
1671 if (re->after.data_disks == re->before.data_disks &&
1672 re->after.layout == re->before.layout &&
1673 info->new_chunk == info->array.chunk_size) {
1674 /* Nothing to change, can change level immediately. */
1675 re->level = info->new_level;
1676 re->backup_blocks = 0;
1677 return NULL;
1678 }
1679 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1680 /* chunk and layout changes make no difference */
1681 re->level = info->new_level;
1682 re->backup_blocks = 0;
1683 return NULL;
1684 }
1685
1686 if (re->after.data_disks == re->before.data_disks &&
1687 get_linux_version() < 2006032)
1688 return "in-place reshape is not safe before 2.6.32 - sorry.";
1689
1690 if (re->after.data_disks < re->before.data_disks &&
1691 get_linux_version() < 2006030)
1692 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1693
1694 re->backup_blocks = compute_backup_blocks(
1695 info->new_chunk, info->array.chunk_size,
1696 re->after.data_disks, re->before.data_disks);
1697 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1698
1699 re->new_size = info->component_size * re->after.data_disks;
1700 return NULL;
1701 }
1702
1703 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1704 char *text_version)
1705 {
1706 struct mdinfo *info;
1707 char *subarray;
1708 int ret_val = -1;
1709
1710 if ((st == NULL) || (sra == NULL))
1711 return ret_val;
1712
1713 if (text_version == NULL)
1714 text_version = sra->text_version;
1715 subarray = strchr(text_version + 1, '/')+1;
1716 info = st->ss->container_content(st, subarray);
1717 if (info) {
1718 unsigned long long current_size = 0;
1719 unsigned long long new_size = info->custom_array_size/2;
1720
1721 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1722 new_size > current_size) {
1723 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1724 < 0)
1725 dprintf("Error: Cannot set array size");
1726 else {
1727 ret_val = 0;
1728 dprintf("Array size changed");
1729 }
1730 dprintf_cont(" from %llu to %llu.\n",
1731 current_size, new_size);
1732 }
1733 sysfs_free(info);
1734 } else
1735 dprintf("Error: set_array_size(): info pointer in NULL\n");
1736
1737 return ret_val;
1738 }
1739
1740 static int reshape_array(char *container, int fd, char *devname,
1741 struct supertype *st, struct mdinfo *info,
1742 int force, struct mddev_dev *devlist,
1743 unsigned long long data_offset,
1744 char *backup_file, int verbose, int forked,
1745 int restart, int freeze_reshape);
1746 static int reshape_container(char *container, char *devname,
1747 int mdfd,
1748 struct supertype *st,
1749 struct mdinfo *info,
1750 int force,
1751 char *backup_file, int verbose,
1752 int forked, int restart, int freeze_reshape);
1753
1754 int Grow_reshape(char *devname, int fd,
1755 struct mddev_dev *devlist,
1756 unsigned long long data_offset,
1757 struct context *c, struct shape *s)
1758 {
1759 /* Make some changes in the shape of an array.
1760 * The kernel must support the change.
1761 *
1762 * There are three different changes. Each can trigger
1763 * a resync or recovery so we freeze that until we have
1764 * requested everything (if kernel supports freezing - 2.6.30).
1765 * The steps are:
1766 * - change size (i.e. component_size)
1767 * - change level
1768 * - change layout/chunksize/ndisks
1769 *
1770 * The last can require a reshape. It is different on different
1771 * levels so we need to check the level before actioning it.
1772 * Some times the level change needs to be requested after the
1773 * reshape (e.g. raid6->raid5, raid5->raid0)
1774 *
1775 */
1776 struct mdu_array_info_s array;
1777 int rv = 0;
1778 struct supertype *st;
1779 char *subarray = NULL;
1780
1781 int frozen;
1782 int changed = 0;
1783 char *container = NULL;
1784 int cfd = -1;
1785
1786 struct mddev_dev *dv;
1787 int added_disks;
1788
1789 struct mdinfo info;
1790 struct mdinfo *sra;
1791
1792 if (md_get_array_info(fd, &array) < 0) {
1793 pr_err("%s is not an active md array - aborting\n",
1794 devname);
1795 return 1;
1796 }
1797 if (data_offset != INVALID_SECTORS && array.level != 10 &&
1798 (array.level < 4 || array.level > 6)) {
1799 pr_err("--grow --data-offset not yet supported\n");
1800 return 1;
1801 }
1802
1803 if (s->size > 0 &&
1804 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1805 pr_err("cannot change component size at the same time as other changes.\n"
1806 " Change size first, then check data is intact before making other changes.\n");
1807 return 1;
1808 }
1809
1810 if (s->raiddisks && s->raiddisks < array.raid_disks &&
1811 array.level > 1 && get_linux_version() < 2006032 &&
1812 !check_env("MDADM_FORCE_FEWER")) {
1813 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1814 " Please use a newer kernel\n");
1815 return 1;
1816 }
1817
1818 if (array.level > 1 && s->size > 1 &&
1819 (array.chunk_size / 1024) > (int)s->size) {
1820 pr_err("component size must be larger than chunk size.\n");
1821 return 1;
1822 }
1823
1824 st = super_by_fd(fd, &subarray);
1825 if (!st) {
1826 pr_err("Unable to determine metadata format for %s\n", devname);
1827 return 1;
1828 }
1829 if (s->raiddisks > st->max_devs) {
1830 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1831 return 1;
1832 }
1833 if (s->level == 0 &&
1834 (array.state & (1<<MD_SB_BITMAP_PRESENT)) &&
1835 !(array.state & (1<<MD_SB_CLUSTERED))) {
1836 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
1837 if (md_set_array_info(fd, &array)!= 0) {
1838 pr_err("failed to remove internal bitmap.\n");
1839 return 1;
1840 }
1841 }
1842
1843 /* in the external case we need to check that the requested reshape is
1844 * supported, and perform an initial check that the container holds the
1845 * pre-requisite spare devices (mdmon owns final validation)
1846 */
1847 if (st->ss->external) {
1848 int retval;
1849
1850 if (subarray) {
1851 container = st->container_devnm;
1852 cfd = open_dev_excl(st->container_devnm);
1853 } else {
1854 container = st->devnm;
1855 close(fd);
1856 cfd = open_dev_excl(st->devnm);
1857 fd = cfd;
1858 }
1859 if (cfd < 0) {
1860 pr_err("Unable to open container for %s\n", devname);
1861 free(subarray);
1862 return 1;
1863 }
1864
1865 retval = st->ss->load_container(st, cfd, NULL);
1866
1867 if (retval) {
1868 pr_err("Cannot read superblock for %s\n", devname);
1869 free(subarray);
1870 return 1;
1871 }
1872
1873 /* check if operation is supported for metadata handler */
1874 if (st->ss->container_content) {
1875 struct mdinfo *cc = NULL;
1876 struct mdinfo *content = NULL;
1877
1878 cc = st->ss->container_content(st, subarray);
1879 for (content = cc; content ; content = content->next) {
1880 int allow_reshape = 1;
1881
1882 /* check if reshape is allowed based on metadata
1883 * indications stored in content.array.status
1884 */
1885 if (content->array.state &
1886 (1 << MD_SB_BLOCK_VOLUME))
1887 allow_reshape = 0;
1888 if (content->array.state &
1889 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE))
1890 allow_reshape = 0;
1891 if (!allow_reshape) {
1892 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1893 devname, container);
1894 sysfs_free(cc);
1895 free(subarray);
1896 return 1;
1897 }
1898 if (content->consistency_policy ==
1899 CONSISTENCY_POLICY_PPL) {
1900 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1901 sysfs_free(cc);
1902 free(subarray);
1903 return 1;
1904 }
1905 }
1906 sysfs_free(cc);
1907 }
1908 if (mdmon_running(container))
1909 st->update_tail = &st->updates;
1910 }
1911
1912 added_disks = 0;
1913 for (dv = devlist; dv; dv = dv->next)
1914 added_disks++;
1915 if (s->raiddisks > array.raid_disks &&
1916 array.spare_disks + added_disks <
1917 (s->raiddisks - array.raid_disks) &&
1918 !c->force) {
1919 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1920 " Use --force to over-ride this check.\n",
1921 s->raiddisks - array.raid_disks,
1922 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1923 array.spare_disks + added_disks);
1924 return 1;
1925 }
1926
1927 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS |
1928 GET_STATE | GET_VERSION);
1929 if (sra) {
1930 if (st->ss->external && subarray == NULL) {
1931 array.level = LEVEL_CONTAINER;
1932 sra->array.level = LEVEL_CONTAINER;
1933 }
1934 } else {
1935 pr_err("failed to read sysfs parameters for %s\n",
1936 devname);
1937 return 1;
1938 }
1939 frozen = freeze(st);
1940 if (frozen < -1) {
1941 /* freeze() already spewed the reason */
1942 sysfs_free(sra);
1943 return 1;
1944 } else if (frozen < 0) {
1945 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1946 sysfs_free(sra);
1947 return 1;
1948 }
1949
1950 /* ========= set size =============== */
1951 if (s->size > 0 &&
1952 (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1953 unsigned long long orig_size = get_component_size(fd)/2;
1954 unsigned long long min_csize;
1955 struct mdinfo *mdi;
1956 int raid0_takeover = 0;
1957
1958 if (orig_size == 0)
1959 orig_size = (unsigned) array.size;
1960
1961 if (orig_size == 0) {
1962 pr_err("Cannot set device size in this type of array.\n");
1963 rv = 1;
1964 goto release;
1965 }
1966
1967 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1968 devname, APPLY_METADATA_CHANGES,
1969 c->verbose > 0)) {
1970 rv = 1;
1971 goto release;
1972 }
1973 sync_metadata(st);
1974 if (st->ss->external) {
1975 /* metadata can have size limitation
1976 * update size value according to metadata information
1977 */
1978 struct mdinfo *sizeinfo =
1979 st->ss->container_content(st, subarray);
1980 if (sizeinfo) {
1981 unsigned long long new_size =
1982 sizeinfo->custom_array_size/2;
1983 int data_disks = get_data_disks(
1984 sizeinfo->array.level,
1985 sizeinfo->array.layout,
1986 sizeinfo->array.raid_disks);
1987 new_size /= data_disks;
1988 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1989 orig_size, new_size,
1990 new_size * data_disks);
1991 s->size = new_size;
1992 sysfs_free(sizeinfo);
1993 }
1994 }
1995
1996 /* Update the size of each member device in case
1997 * they have been resized. This will never reduce
1998 * below the current used-size. The "size" attribute
1999 * understands '0' to mean 'max'.
2000 */
2001 min_csize = 0;
2002 for (mdi = sra->devs; mdi; mdi = mdi->next) {
2003 sysfs_set_num(sra, mdi, "size",
2004 s->size == MAX_SIZE ? 0 : s->size);
2005 if (array.not_persistent == 0 &&
2006 array.major_version == 0 &&
2007 get_linux_version() < 3001000) {
2008 /* Dangerous to allow size to exceed 2TB */
2009 unsigned long long csize;
2010 if (sysfs_get_ll(sra, mdi, "size",
2011 &csize) == 0) {
2012 if (csize >= 2ULL*1024*1024*1024)
2013 csize = 2ULL*1024*1024*1024;
2014 if ((min_csize == 0 ||
2015 (min_csize > csize)))
2016 min_csize = csize;
2017 }
2018 }
2019 }
2020 if (min_csize && s->size > min_csize) {
2021 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2022 rv = 1;
2023 goto size_change_error;
2024 }
2025 if (min_csize && s->size == MAX_SIZE) {
2026 /* Don't let the kernel choose a size - it will get
2027 * it wrong
2028 */
2029 pr_err("Limited v0.90 array to 2TB per device\n");
2030 s->size = min_csize;
2031 }
2032 if (st->ss->external) {
2033 if (sra->array.level == 0) {
2034 rv = sysfs_set_str(sra, NULL, "level", "raid5");
2035 if (!rv) {
2036 raid0_takeover = 1;
2037 /* get array parameters after takeover
2038 * to change one parameter at time only
2039 */
2040 rv = md_get_array_info(fd, &array);
2041 }
2042 }
2043 /* make sure mdmon is
2044 * aware of the new level */
2045 if (!mdmon_running(st->container_devnm))
2046 start_mdmon(st->container_devnm);
2047 ping_monitor(container);
2048 if (mdmon_running(st->container_devnm) &&
2049 st->update_tail == NULL)
2050 st->update_tail = &st->updates;
2051 }
2052
2053 if (s->size == MAX_SIZE)
2054 s->size = 0;
2055 array.size = s->size;
2056 if (s->size & ~INT32_MAX) {
2057 /* got truncated to 32bit, write to
2058 * component_size instead
2059 */
2060 if (sra)
2061 rv = sysfs_set_num(sra, NULL,
2062 "component_size", s->size);
2063 else
2064 rv = -1;
2065 } else {
2066 rv = md_set_array_info(fd, &array);
2067
2068 /* manage array size when it is managed externally
2069 */
2070 if ((rv == 0) && st->ss->external)
2071 rv = set_array_size(st, sra, sra->text_version);
2072 }
2073
2074 if (raid0_takeover) {
2075 /* do not recync non-existing parity,
2076 * we will drop it anyway
2077 */
2078 sysfs_set_str(sra, NULL, "sync_action", "frozen");
2079 /* go back to raid0, drop parity disk
2080 */
2081 sysfs_set_str(sra, NULL, "level", "raid0");
2082 md_get_array_info(fd, &array);
2083 }
2084
2085 size_change_error:
2086 if (rv != 0) {
2087 int err = errno;
2088
2089 /* restore metadata */
2090 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
2091 UnSet, NULL, devname,
2092 ROLLBACK_METADATA_CHANGES,
2093 c->verbose) == 0)
2094 sync_metadata(st);
2095 pr_err("Cannot set device size for %s: %s\n",
2096 devname, strerror(err));
2097 if (err == EBUSY &&
2098 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2099 cont_err("Bitmap must be removed before size can be changed\n");
2100 rv = 1;
2101 goto release;
2102 }
2103 if (s->assume_clean) {
2104 /* This will fail on kernels older than 3.0 unless
2105 * a backport has been arranged.
2106 */
2107 if (sra == NULL ||
2108 sysfs_set_str(sra, NULL, "resync_start",
2109 "none") < 0)
2110 pr_err("--assume-clean not supported with --grow on this kernel\n");
2111 }
2112 md_get_array_info(fd, &array);
2113 s->size = get_component_size(fd)/2;
2114 if (s->size == 0)
2115 s->size = array.size;
2116 if (c->verbose >= 0) {
2117 if (s->size == orig_size)
2118 pr_err("component size of %s unchanged at %lluK\n",
2119 devname, s->size);
2120 else
2121 pr_err("component size of %s has been set to %lluK\n",
2122 devname, s->size);
2123 }
2124 changed = 1;
2125 } else if (array.level != LEVEL_CONTAINER) {
2126 s->size = get_component_size(fd)/2;
2127 if (s->size == 0)
2128 s->size = array.size;
2129 }
2130
2131 /* See if there is anything else to do */
2132 if ((s->level == UnSet || s->level == array.level) &&
2133 (s->layout_str == NULL) &&
2134 (s->chunk == 0 || s->chunk == array.chunk_size) &&
2135 data_offset == INVALID_SECTORS &&
2136 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
2137 /* Nothing more to do */
2138 if (!changed && c->verbose >= 0)
2139 pr_err("%s: no change requested\n", devname);
2140 goto release;
2141 }
2142
2143 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2144 * current implementation assumes that following conditions must be met:
2145 * - RAID10:
2146 * - far_copies == 1
2147 * - near_copies == 2
2148 */
2149 if ((s->level == 0 && array.level == 10 && sra &&
2150 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
2151 (s->level == 0 && array.level == 1 && sra)) {
2152 int err;
2153
2154 err = remove_disks_for_takeover(st, sra, array.layout);
2155 if (err) {
2156 dprintf("Array cannot be reshaped\n");
2157 if (cfd > -1)
2158 close(cfd);
2159 rv = 1;
2160 goto release;
2161 }
2162 /* Make sure mdmon has seen the device removal
2163 * and updated metadata before we continue with
2164 * level change
2165 */
2166 if (container)
2167 ping_monitor(container);
2168 }
2169
2170 memset(&info, 0, sizeof(info));
2171 info.array = array;
2172 if (sysfs_init(&info, fd, NULL)) {
2173 pr_err("failed to intialize sysfs.\n");
2174 rv = 1;
2175 goto release;
2176 }
2177 strcpy(info.text_version, sra->text_version);
2178 info.component_size = s->size*2;
2179 info.new_level = s->level;
2180 info.new_chunk = s->chunk * 1024;
2181 if (info.array.level == LEVEL_CONTAINER) {
2182 info.delta_disks = UnSet;
2183 info.array.raid_disks = s->raiddisks;
2184 } else if (s->raiddisks)
2185 info.delta_disks = s->raiddisks - info.array.raid_disks;
2186 else
2187 info.delta_disks = UnSet;
2188 if (s->layout_str == NULL) {
2189 info.new_layout = UnSet;
2190 if (info.array.level == 6 &&
2191 (info.new_level == 6 || info.new_level == UnSet) &&
2192 info.array.layout >= 16) {
2193 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
2194 cont_err("during the reshape, please specify --layout=preserve\n");
2195 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2196 rv = 1;
2197 goto release;
2198 }
2199 } else if (strcmp(s->layout_str, "normalise") == 0 ||
2200 strcmp(s->layout_str, "normalize") == 0) {
2201 /* If we have a -6 RAID6 layout, remove the '-6'. */
2202 info.new_layout = UnSet;
2203 if (info.array.level == 6 && info.new_level == UnSet) {
2204 char l[40], *h;
2205 strcpy(l, map_num(r6layout, info.array.layout));
2206 h = strrchr(l, '-');
2207 if (h && strcmp(h, "-6") == 0) {
2208 *h = 0;
2209 info.new_layout = map_name(r6layout, l);
2210 }
2211 } else {
2212 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
2213 rv = 1;
2214 goto release;
2215 }
2216 } else if (strcmp(s->layout_str, "preserve") == 0) {
2217 /* This means that a non-standard RAID6 layout
2218 * is OK.
2219 * In particular:
2220 * - When reshape a RAID6 (e.g. adding a device)
2221 * which is in a non-standard layout, it is OK
2222 * to preserve that layout.
2223 * - When converting a RAID5 to RAID6, leave it in
2224 * the XXX-6 layout, don't re-layout.
2225 */
2226 if (info.array.level == 6 && info.new_level == UnSet)
2227 info.new_layout = info.array.layout;
2228 else if (info.array.level == 5 && info.new_level == 6) {
2229 char l[40];
2230 strcpy(l, map_num(r5layout, info.array.layout));
2231 strcat(l, "-6");
2232 info.new_layout = map_name(r6layout, l);
2233 } else {
2234 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
2235 rv = 1;
2236 goto release;
2237 }
2238 } else {
2239 int l = info.new_level;
2240 if (l == UnSet)
2241 l = info.array.level;
2242 switch (l) {
2243 case 5:
2244 info.new_layout = map_name(r5layout, s->layout_str);
2245 break;
2246 case 6:
2247 info.new_layout = map_name(r6layout, s->layout_str);
2248 break;
2249 case 10:
2250 info.new_layout = parse_layout_10(s->layout_str);
2251 break;
2252 case LEVEL_FAULTY:
2253 info.new_layout = parse_layout_faulty(s->layout_str);
2254 break;
2255 default:
2256 pr_err("layout not meaningful with this level\n");
2257 rv = 1;
2258 goto release;
2259 }
2260 if (info.new_layout == UnSet) {
2261 pr_err("layout %s not understood for this level\n",
2262 s->layout_str);
2263 rv = 1;
2264 goto release;
2265 }
2266 }
2267
2268 if (array.level == LEVEL_FAULTY) {
2269 if (s->level != UnSet && s->level != array.level) {
2270 pr_err("cannot change level of Faulty device\n");
2271 rv =1 ;
2272 }
2273 if (s->chunk) {
2274 pr_err("cannot set chunksize of Faulty device\n");
2275 rv =1 ;
2276 }
2277 if (s->raiddisks && s->raiddisks != 1) {
2278 pr_err("cannot set raid_disks of Faulty device\n");
2279 rv =1 ;
2280 }
2281 if (s->layout_str) {
2282 if (md_get_array_info(fd, &array) != 0) {
2283 dprintf("Cannot get array information.\n");
2284 goto release;
2285 }
2286 array.layout = info.new_layout;
2287 if (md_set_array_info(fd, &array) != 0) {
2288 pr_err("failed to set new layout\n");
2289 rv = 1;
2290 } else if (c->verbose >= 0)
2291 printf("layout for %s set to %d\n",
2292 devname, array.layout);
2293 }
2294 } else if (array.level == LEVEL_CONTAINER) {
2295 /* This change is to be applied to every array in the
2296 * container. This is only needed when the metadata imposes
2297 * restraints of the various arrays in the container.
2298 * Currently we only know that IMSM requires all arrays
2299 * to have the same number of devices so changing the
2300 * number of devices (On-Line Capacity Expansion) must be
2301 * performed at the level of the container
2302 */
2303 if (fd > 0) {
2304 close(fd);
2305 fd = -1;
2306 }
2307 rv = reshape_container(container, devname, -1, st, &info,
2308 c->force, c->backup_file, c->verbose,
2309 0, 0, 0);
2310 frozen = 0;
2311 } else {
2312 /* get spare devices from external metadata
2313 */
2314 if (st->ss->external) {
2315 struct mdinfo *info2;
2316
2317 info2 = st->ss->container_content(st, subarray);
2318 if (info2) {
2319 info.array.spare_disks =
2320 info2->array.spare_disks;
2321 sysfs_free(info2);
2322 }
2323 }
2324
2325 /* Impose these changes on a single array. First
2326 * check that the metadata is OK with the change. */
2327
2328 if (reshape_super(st, 0, info.new_level,
2329 info.new_layout, info.new_chunk,
2330 info.array.raid_disks, info.delta_disks,
2331 c->backup_file, devname,
2332 APPLY_METADATA_CHANGES, c->verbose)) {
2333 rv = 1;
2334 goto release;
2335 }
2336 sync_metadata(st);
2337 rv = reshape_array(container, fd, devname, st, &info, c->force,
2338 devlist, data_offset, c->backup_file,
2339 c->verbose, 0, 0, 0);
2340 frozen = 0;
2341 }
2342 release:
2343 sysfs_free(sra);
2344 if (frozen > 0)
2345 unfreeze(st);
2346 return rv;
2347 }
2348
2349 /* verify_reshape_position()
2350 * Function checks if reshape position in metadata is not farther
2351 * than position in md.
2352 * Return value:
2353 * 0 : not valid sysfs entry
2354 * it can be caused by not started reshape, it should be started
2355 * by reshape array or raid0 array is before takeover
2356 * -1 : error, reshape position is obviously wrong
2357 * 1 : success, reshape progress correct or updated
2358 */
2359 static int verify_reshape_position(struct mdinfo *info, int level)
2360 {
2361 int ret_val = 0;
2362 char buf[40];
2363 int rv;
2364
2365 /* read sync_max, failure can mean raid0 array */
2366 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2367
2368 if (rv > 0) {
2369 char *ep;
2370 unsigned long long position = strtoull(buf, &ep, 0);
2371
2372 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2373 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2374 position *= get_data_disks(level,
2375 info->new_layout,
2376 info->array.raid_disks);
2377 if (info->reshape_progress < position) {
2378 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2379 info->reshape_progress, position);
2380 info->reshape_progress = position;
2381 ret_val = 1;
2382 } else if (info->reshape_progress > position) {
2383 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2384 position, info->reshape_progress);
2385 ret_val = -1;
2386 } else {
2387 dprintf("Reshape position in md and metadata are the same;");
2388 ret_val = 1;
2389 }
2390 }
2391 } else if (rv == 0) {
2392 /* for valid sysfs entry, 0-length content
2393 * should be indicated as error
2394 */
2395 ret_val = -1;
2396 }
2397
2398 return ret_val;
2399 }
2400
2401 static unsigned long long choose_offset(unsigned long long lo,
2402 unsigned long long hi,
2403 unsigned long long min,
2404 unsigned long long max)
2405 {
2406 /* Choose a new offset between hi and lo.
2407 * It must be between min and max, but
2408 * we would prefer something near the middle of hi/lo, and also
2409 * prefer to be aligned to a big power of 2.
2410 *
2411 * So we start with the middle, then for each bit,
2412 * starting at '1' and increasing, if it is set, we either
2413 * add it or subtract it if possible, preferring the option
2414 * which is furthest from the boundary.
2415 *
2416 * We stop once we get a 1MB alignment. As units are in sectors,
2417 * 1MB = 2*1024 sectors.
2418 */
2419 unsigned long long choice = (lo + hi) / 2;
2420 unsigned long long bit = 1;
2421
2422 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2423 unsigned long long bigger, smaller;
2424 if (! (bit & choice))
2425 continue;
2426 bigger = choice + bit;
2427 smaller = choice - bit;
2428 if (bigger > max && smaller < min)
2429 break;
2430 if (bigger > max)
2431 choice = smaller;
2432 else if (smaller < min)
2433 choice = bigger;
2434 else if (hi - bigger > smaller - lo)
2435 choice = bigger;
2436 else
2437 choice = smaller;
2438 }
2439 return choice;
2440 }
2441
2442 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2443 char *devname, int delta_disks,
2444 unsigned long long data_offset,
2445 unsigned long long min,
2446 int can_fallback)
2447 {
2448 struct mdinfo *sd;
2449 int dir = 0;
2450 int err = 0;
2451 unsigned long long before, after;
2452
2453 /* Need to find min space before and after so same is used
2454 * on all devices
2455 */
2456 before = UINT64_MAX;
2457 after = UINT64_MAX;
2458 for (sd = sra->devs; sd; sd = sd->next) {
2459 char *dn;
2460 int dfd;
2461 int rv;
2462 struct supertype *st2;
2463 struct mdinfo info2;
2464
2465 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2466 continue;
2467 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2468 dfd = dev_open(dn, O_RDONLY);
2469 if (dfd < 0) {
2470 pr_err("%s: cannot open component %s\n",
2471 devname, dn ? dn : "-unknown-");
2472 goto release;
2473 }
2474 st2 = dup_super(st);
2475 rv = st2->ss->load_super(st2,dfd, NULL);
2476 close(dfd);
2477 if (rv) {
2478 free(st2);
2479 pr_err("%s: cannot get superblock from %s\n",
2480 devname, dn);
2481 goto release;
2482 }
2483 st2->ss->getinfo_super(st2, &info2, NULL);
2484 st2->ss->free_super(st2);
2485 free(st2);
2486 if (info2.space_before == 0 &&
2487 info2.space_after == 0) {
2488 /* Metadata doesn't support data_offset changes */
2489 if (!can_fallback)
2490 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2491 devname);
2492 goto fallback;
2493 }
2494 if (before > info2.space_before)
2495 before = info2.space_before;
2496 if (after > info2.space_after)
2497 after = info2.space_after;
2498
2499 if (data_offset != INVALID_SECTORS) {
2500 if (dir == 0) {
2501 if (info2.data_offset == data_offset) {
2502 pr_err("%s: already has that data_offset\n",
2503 dn);
2504 goto release;
2505 }
2506 if (data_offset < info2.data_offset)
2507 dir = -1;
2508 else
2509 dir = 1;
2510 } else if ((data_offset <= info2.data_offset &&
2511 dir == 1) ||
2512 (data_offset >= info2.data_offset &&
2513 dir == -1)) {
2514 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2515 dn);
2516 goto release;
2517 }
2518 }
2519 }
2520 if (before == UINT64_MAX)
2521 /* impossible really, there must be no devices */
2522 return 1;
2523
2524 for (sd = sra->devs; sd; sd = sd->next) {
2525 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2526 unsigned long long new_data_offset;
2527
2528 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2529 continue;
2530 if (delta_disks < 0) {
2531 /* Don't need any space as array is shrinking
2532 * just move data_offset up by min
2533 */
2534 if (data_offset == INVALID_SECTORS)
2535 new_data_offset = sd->data_offset + min;
2536 else {
2537 if (data_offset < sd->data_offset + min) {
2538 pr_err("--data-offset too small for %s\n",
2539 dn);
2540 goto release;
2541 }
2542 new_data_offset = data_offset;
2543 }
2544 } else if (delta_disks > 0) {
2545 /* need space before */
2546 if (before < min) {
2547 if (can_fallback)
2548 goto fallback;
2549 pr_err("Insufficient head-space for reshape on %s\n",
2550 dn);
2551 goto release;
2552 }
2553 if (data_offset == INVALID_SECTORS)
2554 new_data_offset = sd->data_offset - min;
2555 else {
2556 if (data_offset > sd->data_offset - min) {
2557 pr_err("--data-offset too large for %s\n",
2558 dn);
2559 goto release;
2560 }
2561 new_data_offset = data_offset;
2562 }
2563 } else {
2564 if (dir == 0) {
2565 /* can move up or down. If 'data_offset'
2566 * was set we would have already decided,
2567 * so just choose direction with most space.
2568 */
2569 if (before > after)
2570 dir = -1;
2571 else
2572 dir = 1;
2573 }
2574 sysfs_set_str(sra, NULL, "reshape_direction",
2575 dir == 1 ? "backwards" : "forwards");
2576 if (dir > 0) {
2577 /* Increase data offset */
2578 if (after < min) {
2579 if (can_fallback)
2580 goto fallback;
2581 pr_err("Insufficient tail-space for reshape on %s\n",
2582 dn);
2583 goto release;
2584 }
2585 if (data_offset != INVALID_SECTORS &&
2586 data_offset < sd->data_offset + min) {
2587 pr_err("--data-offset too small on %s\n",
2588 dn);
2589 goto release;
2590 }
2591 if (data_offset != INVALID_SECTORS)
2592 new_data_offset = data_offset;
2593 else
2594 new_data_offset = choose_offset(sd->data_offset,
2595 sd->data_offset + after,
2596 sd->data_offset + min,
2597 sd->data_offset + after);
2598 } else {
2599 /* Decrease data offset */
2600 if (before < min) {
2601 if (can_fallback)
2602 goto fallback;
2603 pr_err("insufficient head-room on %s\n",
2604 dn);
2605 goto release;
2606 }
2607 if (data_offset != INVALID_SECTORS &&
2608 data_offset < sd->data_offset - min) {
2609 pr_err("--data-offset too small on %s\n",
2610 dn);
2611 goto release;
2612 }
2613 if (data_offset != INVALID_SECTORS)
2614 new_data_offset = data_offset;
2615 else
2616 new_data_offset = choose_offset(sd->data_offset - before,
2617 sd->data_offset,
2618 sd->data_offset - before,
2619 sd->data_offset - min);
2620 }
2621 }
2622 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2623 if (err < 0 && errno == E2BIG) {
2624 /* try again after increasing data size to max */
2625 err = sysfs_set_num(sra, sd, "size", 0);
2626 if (err < 0 && errno == EINVAL &&
2627 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2628 /* some kernels have a bug where you cannot
2629 * use '0' on spare devices. */
2630 sysfs_set_num(sra, sd, "size",
2631 (sra->component_size + after)/2);
2632 }
2633 err = sysfs_set_num(sra, sd, "new_offset",
2634 new_data_offset);
2635 }
2636 if (err < 0) {
2637 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2638 pr_err("data-offset is too big for %s\n", dn);
2639 goto release;
2640 }
2641 if (sd == sra->devs &&
2642 (errno == ENOENT || errno == E2BIG))
2643 /* Early kernel, no 'new_offset' file,
2644 * or kernel doesn't like us.
2645 * For RAID5/6 this is not fatal
2646 */
2647 return 1;
2648 pr_err("Cannot set new_offset for %s\n", dn);
2649 break;
2650 }
2651 }
2652 return err;
2653 release:
2654 return -1;
2655 fallback:
2656 /* Just use a backup file */
2657 return 1;
2658 }
2659
2660 static int raid10_reshape(char *container, int fd, char *devname,
2661 struct supertype *st, struct mdinfo *info,
2662 struct reshape *reshape,
2663 unsigned long long data_offset,
2664 int force, int verbose)
2665 {
2666 /* Changing raid_disks, layout, chunksize or possibly
2667 * just data_offset for a RAID10.
2668 * We must always change data_offset. We change by at least
2669 * ->min_offset_change which is the largest of the old and new
2670 * chunk sizes.
2671 * If raid_disks is increasing, then data_offset must decrease
2672 * by at least this copy size.
2673 * If raid_disks is unchanged, data_offset must increase or
2674 * decrease by at least min_offset_change but preferably by much more.
2675 * We choose half of the available space.
2676 * If raid_disks is decreasing, data_offset must increase by
2677 * at least min_offset_change. To allow of this, component_size
2678 * must be decreased by the same amount.
2679 *
2680 * So we calculate the required minimum and direction, possibly
2681 * reduce the component_size, then iterate through the devices
2682 * and set the new_data_offset.
2683 * If that all works, we set chunk_size, layout, raid_disks, and start
2684 * 'reshape'
2685 */
2686 struct mdinfo *sra;
2687 unsigned long long min;
2688 int err = 0;
2689
2690 sra = sysfs_read(fd, NULL,
2691 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2692 );
2693 if (!sra) {
2694 pr_err("%s: Cannot get array details from sysfs\n", devname);
2695 goto release;
2696 }
2697 min = reshape->min_offset_change;
2698
2699 if (info->delta_disks)
2700 sysfs_set_str(sra, NULL, "reshape_direction",
2701 info->delta_disks < 0 ? "backwards" : "forwards");
2702 if (info->delta_disks < 0 && info->space_after < min) {
2703 int rv = sysfs_set_num(sra, NULL, "component_size",
2704 (sra->component_size - min)/2);
2705 if (rv) {
2706 pr_err("cannot reduce component size\n");
2707 goto release;
2708 }
2709 }
2710 err = set_new_data_offset(sra, st, devname, info->delta_disks,
2711 data_offset, min, 0);
2712 if (err == 1) {
2713 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2714 cont_err("supported on this kernel\n");
2715 err = -1;
2716 }
2717 if (err < 0)
2718 goto release;
2719
2720 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2721 err = errno;
2722 if (!err && sysfs_set_num(sra, NULL, "layout",
2723 reshape->after.layout) < 0)
2724 err = errno;
2725 if (!err &&
2726 sysfs_set_num(sra, NULL, "raid_disks",
2727 info->array.raid_disks + info->delta_disks) < 0)
2728 err = errno;
2729 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2730 err = errno;
2731 if (err) {
2732 pr_err("Cannot set array shape for %s\n",
2733 devname);
2734 if (err == EBUSY &&
2735 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2736 cont_err(" Bitmap must be removed before shape can be changed\n");
2737 goto release;
2738 }
2739 sysfs_free(sra);
2740 return 0;
2741 release:
2742 sysfs_free(sra);
2743 return 1;
2744 }
2745
2746 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2747 {
2748 struct mdinfo *sra, *sd;
2749 /* Initialisation to silence compiler warning */
2750 unsigned long long min_space_before = 0, min_space_after = 0;
2751 int first = 1;
2752
2753 sra = sysfs_read(fd, NULL, GET_DEVS);
2754 if (!sra)
2755 return;
2756 for (sd = sra->devs; sd; sd = sd->next) {
2757 char *dn;
2758 int dfd;
2759 struct supertype *st2;
2760 struct mdinfo info2;
2761
2762 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2763 continue;
2764 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2765 dfd = dev_open(dn, O_RDONLY);
2766 if (dfd < 0)
2767 break;
2768 st2 = dup_super(st);
2769 if (st2->ss->load_super(st2,dfd, NULL)) {
2770 close(dfd);
2771 free(st2);
2772 break;
2773 }
2774 close(dfd);
2775 st2->ss->getinfo_super(st2, &info2, NULL);
2776 st2->ss->free_super(st2);
2777 free(st2);
2778 if (first ||
2779 min_space_before > info2.space_before)
2780 min_space_before = info2.space_before;
2781 if (first ||
2782 min_space_after > info2.space_after)
2783 min_space_after = info2.space_after;
2784 first = 0;
2785 }
2786 if (sd == NULL && !first) {
2787 info->space_after = min_space_after;
2788 info->space_before = min_space_before;
2789 }
2790 sysfs_free(sra);
2791 }
2792
2793 static void update_cache_size(char *container, struct mdinfo *sra,
2794 struct mdinfo *info,
2795 int disks, unsigned long long blocks)
2796 {
2797 /* Check that the internal stripe cache is
2798 * large enough, or it won't work.
2799 * It must hold at least 4 stripes of the larger
2800 * chunk size
2801 */
2802 unsigned long cache;
2803 cache = max(info->array.chunk_size, info->new_chunk);
2804 cache *= 4; /* 4 stripes minimum */
2805 cache /= 512; /* convert to sectors */
2806 /* make sure there is room for 'blocks' with a bit to spare */
2807 if (cache < 16 + blocks / disks)
2808 cache = 16 + blocks / disks;
2809 cache /= (4096/512); /* Convert from sectors to pages */
2810
2811 if (sra->cache_size < cache)
2812 subarray_set_num(container, sra, "stripe_cache_size",
2813 cache+1);
2814 }
2815
2816 static int impose_reshape(struct mdinfo *sra,
2817 struct mdinfo *info,
2818 struct supertype *st,
2819 int fd,
2820 int restart,
2821 char *devname, char *container,
2822 struct reshape *reshape)
2823 {
2824 struct mdu_array_info_s array;
2825
2826 sra->new_chunk = info->new_chunk;
2827
2828 if (restart) {
2829 /* for external metadata checkpoint saved by mdmon can be lost
2830 * or missed /due to e.g. crash/. Check if md is not during
2831 * restart farther than metadata points to.
2832 * If so, this means metadata information is obsolete.
2833 */
2834 if (st->ss->external)
2835 verify_reshape_position(info, reshape->level);
2836 sra->reshape_progress = info->reshape_progress;
2837 } else {
2838 sra->reshape_progress = 0;
2839 if (reshape->after.data_disks < reshape->before.data_disks)
2840 /* start from the end of the new array */
2841 sra->reshape_progress = (sra->component_size
2842 * reshape->after.data_disks);
2843 }
2844
2845 md_get_array_info(fd, &array);
2846 if (info->array.chunk_size == info->new_chunk &&
2847 reshape->before.layout == reshape->after.layout &&
2848 st->ss->external == 0) {
2849 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2850 array.raid_disks = reshape->after.data_disks + reshape->parity;
2851 if (!restart && md_set_array_info(fd, &array) != 0) {
2852 int err = errno;
2853
2854 pr_err("Cannot set device shape for %s: %s\n",
2855 devname, strerror(errno));
2856
2857 if (err == EBUSY &&
2858 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2859 cont_err("Bitmap must be removed before shape can be changed\n");
2860
2861 goto release;
2862 }
2863 } else if (!restart) {
2864 /* set them all just in case some old 'new_*' value
2865 * persists from some earlier problem.
2866 */
2867 int err = 0;
2868 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2869 err = errno;
2870 if (!err && sysfs_set_num(sra, NULL, "layout",
2871 reshape->after.layout) < 0)
2872 err = errno;
2873 if (!err && subarray_set_num(container, sra, "raid_disks",
2874 reshape->after.data_disks +
2875 reshape->parity) < 0)
2876 err = errno;
2877 if (err) {
2878 pr_err("Cannot set device shape for %s\n", devname);
2879
2880 if (err == EBUSY &&
2881 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2882 cont_err("Bitmap must be removed before shape can be changed\n");
2883 goto release;
2884 }
2885 }
2886 return 0;
2887 release:
2888 return -1;
2889 }
2890
2891 static int impose_level(int fd, int level, char *devname, int verbose)
2892 {
2893 char *c;
2894 struct mdu_array_info_s array;
2895 struct mdinfo info;
2896
2897 if (sysfs_init(&info, fd, NULL)) {
2898 pr_err("failed to intialize sysfs.\n");
2899 return 1;
2900 }
2901
2902 md_get_array_info(fd, &array);
2903 if (level == 0 && (array.level >= 4 && array.level <= 6)) {
2904 /* To convert to RAID0 we need to fail and
2905 * remove any non-data devices. */
2906 int found = 0;
2907 int d;
2908 int data_disks = array.raid_disks - 1;
2909 if (array.level == 6)
2910 data_disks -= 1;
2911 if (array.level == 5 && array.layout != ALGORITHM_PARITY_N)
2912 return -1;
2913 if (array.level == 6 && array.layout != ALGORITHM_PARITY_N_6)
2914 return -1;
2915 sysfs_set_str(&info, NULL,"sync_action", "idle");
2916 /* First remove any spares so no recovery starts */
2917 for (d = 0, found = 0;
2918 d < MAX_DISKS && found < array.nr_disks; d++) {
2919 mdu_disk_info_t disk;
2920 disk.number = d;
2921 if (md_get_disk_info(fd, &disk) < 0)
2922 continue;
2923 if (disk.major == 0 && disk.minor == 0)
2924 continue;
2925 found++;
2926 if ((disk.state & (1 << MD_DISK_ACTIVE)) &&
2927 disk.raid_disk < data_disks)
2928 /* keep this */
2929 continue;
2930 ioctl(fd, HOT_REMOVE_DISK,
2931 makedev(disk.major, disk.minor));
2932 }
2933 /* Now fail anything left */
2934 md_get_array_info(fd, &array);
2935 for (d = 0, found = 0;
2936 d < MAX_DISKS && found < array.nr_disks; d++) {
2937 mdu_disk_info_t disk;
2938 disk.number = d;
2939 if (md_get_disk_info(fd, &disk) < 0)
2940 continue;
2941 if (disk.major == 0 && disk.minor == 0)
2942 continue;
2943 found++;
2944 if ((disk.state & (1 << MD_DISK_ACTIVE)) &&
2945 disk.raid_disk < data_disks)
2946 /* keep this */
2947 continue;
2948 ioctl(fd, SET_DISK_FAULTY,
2949 makedev(disk.major, disk.minor));
2950 hot_remove_disk(fd, makedev(disk.major, disk.minor), 1);
2951 }
2952 }
2953 c = map_num(pers, level);
2954 if (c) {
2955 int err = sysfs_set_str(&info, NULL, "level", c);
2956 if (err) {
2957 err = errno;
2958 pr_err("%s: could not set level to %s\n",
2959 devname, c);
2960 if (err == EBUSY &&
2961 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2962 cont_err("Bitmap must be removed before level can be changed\n");
2963 return err;
2964 }
2965 if (verbose >= 0)
2966 pr_err("level of %s changed to %s\n", devname, c);
2967 }
2968 return 0;
2969 }
2970
2971 int sigterm = 0;
2972 static void catch_term(int sig)
2973 {
2974 sigterm = 1;
2975 }
2976
2977 static int continue_via_systemd(char *devnm)
2978 {
2979 int skipped, i, pid, status;
2980 char pathbuf[1024];
2981 /* In a systemd/udev world, it is best to get systemd to
2982 * run "mdadm --grow --continue" rather than running in the
2983 * background.
2984 */
2985 switch(fork()) {
2986 case 0:
2987 /* FIXME yuk. CLOSE_EXEC?? */
2988 skipped = 0;
2989 for (i = 3; skipped < 20; i++)
2990 if (close(i) < 0)
2991 skipped++;
2992 else
2993 skipped = 0;
2994
2995 /* Don't want to see error messages from
2996 * systemctl. If the service doesn't exist,
2997 * we fork ourselves.
2998 */
2999 close(2);
3000 open("/dev/null", O_WRONLY);
3001 snprintf(pathbuf, sizeof(pathbuf),
3002 "mdadm-grow-continue@%s.service", devnm);
3003 status = execl("/usr/bin/systemctl", "systemctl", "restart",
3004 pathbuf, NULL);
3005 status = execl("/bin/systemctl", "systemctl", "restart",
3006 pathbuf, NULL);
3007 exit(1);
3008 case -1: /* Just do it ourselves. */
3009 break;
3010 default: /* parent - good */
3011 pid = wait(&status);
3012 if (pid >= 0 && status == 0)
3013 return 1;
3014 }
3015 return 0;
3016 }
3017
3018 static int reshape_array(char *container, int fd, char *devname,
3019 struct supertype *st, struct mdinfo *info,
3020 int force, struct mddev_dev *devlist,
3021 unsigned long long data_offset,
3022 char *backup_file, int verbose, int forked,
3023 int restart, int freeze_reshape)
3024 {
3025 struct reshape reshape;
3026 int spares_needed;
3027 char *msg;
3028 int orig_level = UnSet;
3029 int odisks;
3030 int delayed;
3031
3032 struct mdu_array_info_s array;
3033 char *c;
3034
3035 struct mddev_dev *dv;
3036 int added_disks;
3037
3038 int *fdlist = NULL;
3039 unsigned long long *offsets = NULL;
3040 int d;
3041 int nrdisks;
3042 int err;
3043 unsigned long blocks;
3044 unsigned long long array_size;
3045 int done;
3046 struct mdinfo *sra = NULL;
3047 char buf[20];
3048
3049 /* when reshaping a RAID0, the component_size might be zero.
3050 * So try to fix that up.
3051 */
3052 if (md_get_array_info(fd, &array) != 0) {
3053 dprintf("Cannot get array information.\n");
3054 goto release;
3055 }
3056 if (array.level == 0 && info->component_size == 0) {
3057 get_dev_size(fd, NULL, &array_size);
3058 info->component_size = array_size / array.raid_disks;
3059 }
3060
3061 if (array.level == 10)
3062 /* Need space_after info */
3063 get_space_after(fd, st, info);
3064
3065 if (info->reshape_active) {
3066 int new_level = info->new_level;
3067 info->new_level = UnSet;
3068 if (info->delta_disks > 0)
3069 info->array.raid_disks -= info->delta_disks;
3070 msg = analyse_change(devname, info, &reshape);
3071 info->new_level = new_level;
3072 if (info->delta_disks > 0)
3073 info->array.raid_disks += info->delta_disks;
3074 if (!restart)
3075 /* Make sure the array isn't read-only */
3076 ioctl(fd, RESTART_ARRAY_RW, 0);
3077 } else
3078 msg = analyse_change(devname, info, &reshape);
3079 if (msg) {
3080 /* if msg == "", error has already been printed */
3081 if (msg[0])
3082 pr_err("%s\n", msg);
3083 goto release;
3084 }
3085 if (restart && (reshape.level != info->array.level ||
3086 reshape.before.layout != info->array.layout ||
3087 reshape.before.data_disks + reshape.parity !=
3088 info->array.raid_disks - max(0, info->delta_disks))) {
3089 pr_err("reshape info is not in native format - cannot continue.\n");
3090 goto release;
3091 }
3092
3093 if (st->ss->external && restart && (info->reshape_progress == 0) &&
3094 !((sysfs_get_str(info, NULL, "sync_action",
3095 buf, sizeof(buf)) > 0) &&
3096 (strncmp(buf, "reshape", 7) == 0))) {
3097 /* When reshape is restarted from '0', very begin of array
3098 * it is possible that for external metadata reshape and array
3099 * configuration doesn't happen.
3100 * Check if md has the same opinion, and reshape is restarted
3101 * from 0. If so, this is regular reshape start after reshape
3102 * switch in metadata to next array only.
3103 */
3104 if ((verify_reshape_position(info, reshape.level) >= 0) &&
3105 (info->reshape_progress == 0))
3106 restart = 0;
3107 }
3108 if (restart) {
3109 /*
3110 * reshape already started. just skip to monitoring
3111 * the reshape
3112 */
3113 if (reshape.backup_blocks == 0)
3114 return 0;
3115 if (restart & RESHAPE_NO_BACKUP)
3116 return 0;
3117
3118 /* Need 'sra' down at 'started:' */
3119 sra = sysfs_read(fd, NULL,
3120 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|
3121 GET_CHUNK|GET_CACHE);
3122 if (!sra) {
3123 pr_err("%s: Cannot get array details from sysfs\n",
3124 devname);
3125 goto release;
3126 }
3127
3128 if (!backup_file)
3129 backup_file = locate_backup(sra->sys_name);
3130
3131 goto started;
3132 }
3133 /* The container is frozen but the array may not be.
3134 * So freeze the array so spares don't get put to the wrong use
3135 * FIXME there should probably be a cleaner separation between
3136 * freeze_array and freeze_container.
3137 */
3138 sysfs_freeze_array(info);
3139 /* Check we have enough spares to not be degraded */
3140 added_disks = 0;
3141 for (dv = devlist; dv ; dv=dv->next)
3142 added_disks++;
3143 spares_needed = max(reshape.before.data_disks,
3144 reshape.after.data_disks) +
3145 reshape.parity - array.raid_disks;
3146
3147 if (!force && info->new_level > 1 && info->array.level > 1 &&
3148 spares_needed > info->array.spare_disks + added_disks) {
3149 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3150 " Use --force to over-ride this check.\n",
3151 spares_needed,
3152 spares_needed == 1 ? "" : "s",
3153 info->array.spare_disks + added_disks);
3154 goto release;
3155 }
3156 /* Check we have enough spares to not fail */
3157 spares_needed = max(reshape.before.data_disks,
3158 reshape.after.data_disks)
3159 - array.raid_disks;
3160 if ((info->new_level > 1 || info->new_level == 0) &&
3161 spares_needed > info->array.spare_disks +added_disks) {
3162 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3163 spares_needed, spares_needed == 1 ? "" : "s",
3164 info->array.spare_disks + added_disks);
3165 goto release;
3166 }
3167
3168 if (reshape.level != array.level) {
3169 int err = impose_level(fd, reshape.level, devname, verbose);
3170 if (err)
3171 goto release;
3172 info->new_layout = UnSet; /* after level change,
3173 * layout is meaningless */
3174 orig_level = array.level;
3175 sysfs_freeze_array(info);
3176
3177 if (reshape.level > 0 && st->ss->external) {
3178 /* make sure mdmon is aware of the new level */
3179 if (mdmon_running(container))
3180 flush_mdmon(container);
3181
3182 if (!mdmon_running(container))
3183 start_mdmon(container);
3184 ping_monitor(container);
3185 if (mdmon_running(container) && st->update_tail == NULL)
3186 st->update_tail = &st->updates;
3187 }
3188 }
3189 /* ->reshape_super might have chosen some spares from the
3190 * container that it wants to be part of the new array.
3191 * We can collect them with ->container_content and give
3192 * them to the kernel.
3193 */
3194 if (st->ss->reshape_super && st->ss->container_content) {
3195 char *subarray = strchr(info->text_version+1, '/')+1;
3196 struct mdinfo *info2 =
3197 st->ss->container_content(st, subarray);
3198 struct mdinfo *d;
3199
3200 if (info2) {
3201 if (sysfs_init(info2, fd, st->devnm)) {
3202 pr_err("unable to initialize sysfs for %s\n",
3203 st->devnm);
3204 free(info2);
3205 goto release;
3206 }
3207 /* When increasing number of devices, we need to set
3208 * new raid_disks before adding these, or they might
3209 * be rejected.
3210 */
3211 if (reshape.backup_blocks &&
3212 reshape.after.data_disks >
3213 reshape.before.data_disks)
3214 subarray_set_num(container, info2, "raid_disks",
3215 reshape.after.data_disks +
3216 reshape.parity);
3217 for (d = info2->devs; d; d = d->next) {
3218 if (d->disk.state == 0 &&
3219 d->disk.raid_disk >= 0) {
3220 /* This is a spare that wants to
3221 * be part of the array.
3222 */
3223 add_disk(fd, st, info2, d);
3224 }
3225 }
3226 sysfs_free(info2);
3227 }
3228 }
3229 /* We might have been given some devices to add to the
3230 * array. Now that the array has been changed to the right
3231 * level and frozen, we can safely add them.
3232 */
3233 if (devlist) {
3234 if (Manage_subdevs(devname, fd, devlist, verbose, 0, NULL, 0))
3235 goto release;
3236 }
3237
3238 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3239 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3240 if (reshape.backup_blocks == 0) {
3241 /* No restriping needed, but we might need to impose
3242 * some more changes: layout, raid_disks, chunk_size
3243 */
3244 /* read current array info */
3245 if (md_get_array_info(fd, &array) != 0) {
3246 dprintf("Cannot get array information.\n");
3247 goto release;
3248 }
3249 /* compare current array info with new values and if
3250 * it is different update them to new */
3251 if (info->new_layout != UnSet &&
3252 info->new_layout != array.layout) {
3253 array.layout = info->new_layout;
3254 if (md_set_array_info(fd, &array) != 0) {
3255 pr_err("failed to set new layout\n");
3256 goto release;
3257 } else if (verbose >= 0)
3258 printf("layout for %s set to %d\n",
3259 devname, array.layout);
3260 }
3261 if (info->delta_disks != UnSet && info->delta_disks != 0 &&
3262 array.raid_disks !=
3263 (info->array.raid_disks + info->delta_disks)) {
3264 array.raid_disks += info->delta_disks;
3265 if (md_set_array_info(fd, &array) != 0) {
3266 pr_err("failed to set raid disks\n");
3267 goto release;
3268 } else if (verbose >= 0) {
3269 printf("raid_disks for %s set to %d\n",
3270 devname, array.raid_disks);
3271 }
3272 }
3273 if (info->new_chunk != 0 &&
3274 info->new_chunk != array.chunk_size) {
3275 if (sysfs_set_num(info, NULL,
3276 "chunk_size", info->new_chunk) != 0) {
3277 pr_err("failed to set chunk size\n");
3278 goto release;
3279 } else if (verbose >= 0)
3280 printf("chunk size for %s set to %d\n",
3281 devname, array.chunk_size);
3282 }
3283 unfreeze(st);
3284 return 0;
3285 }
3286
3287 /*
3288 * There are three possibilities.
3289 * 1/ The array will shrink.
3290 * We need to ensure the reshape will pause before reaching
3291 * the 'critical section'. We also need to fork and wait for
3292 * that to happen. When it does we
3293 * suspend/backup/complete/unfreeze
3294 *
3295 * 2/ The array will not change size.
3296 * This requires that we keep a backup of a sliding window
3297 * so that we can restore data after a crash. So we need
3298 * to fork and monitor progress.
3299 * In future we will allow the data_offset to change, so
3300 * a sliding backup becomes unnecessary.
3301 *
3302 * 3/ The array will grow. This is relatively easy.
3303 * However the kernel's restripe routines will cheerfully
3304 * overwrite some early data before it is safe. So we
3305 * need to make a backup of the early parts of the array
3306 * and be ready to restore it if rebuild aborts very early.
3307 * For externally managed metadata, we still need a forked
3308 * child to monitor the reshape and suspend IO over the region
3309 * that is being reshaped.
3310 *
3311 * We backup data by writing it to one spare, or to a
3312 * file which was given on command line.
3313 *
3314 * In each case, we first make sure that storage is available
3315 * for the required backup.
3316 * Then we:
3317 * - request the shape change.
3318 * - fork to handle backup etc.
3319 */
3320 /* Check that we can hold all the data */
3321 get_dev_size(fd, NULL, &array_size);
3322 if (reshape.new_size < (array_size/512)) {
3323 pr_err("this change will reduce the size of the array.\n"
3324 " use --grow --array-size first to truncate array.\n"
3325 " e.g. mdadm --grow %s --array-size %llu\n",
3326 devname, reshape.new_size/2);
3327 goto release;
3328 }
3329
3330 if (array.level == 10) {
3331 /* Reshaping RAID10 does not require any data backup by
3332 * user-space. Instead it requires that the data_offset
3333 * is changed to avoid the need for backup.
3334 * So this is handled very separately
3335 */
3336 if (restart)
3337 /* Nothing to do. */
3338 return 0;
3339 return raid10_reshape(container, fd, devname, st, info,
3340 &reshape, data_offset, force, verbose);
3341 }
3342 sra = sysfs_read(fd, NULL,
3343 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3344 GET_CACHE);
3345 if (!sra) {
3346 pr_err("%s: Cannot get array details from sysfs\n",
3347 devname);
3348 goto release;
3349 }
3350
3351 if (!backup_file)
3352 switch(set_new_data_offset(sra, st, devname,
3353 reshape.after.data_disks - reshape.before.data_disks,
3354 data_offset,
3355 reshape.min_offset_change, 1)) {
3356 case -1:
3357 goto release;
3358 case 0:
3359 /* Updated data_offset, so it's easy now */
3360 update_cache_size(container, sra, info,
3361 min(reshape.before.data_disks,
3362 reshape.after.data_disks),
3363 reshape.backup_blocks);
3364
3365 /* Right, everything seems fine. Let's kick things off.
3366 */
3367 sync_metadata(st);
3368
3369 if (impose_reshape(sra, info, st, fd, restart,
3370 devname, container, &reshape) < 0)
3371 goto release;
3372 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3373 struct mdinfo *sd;
3374 if (errno != EINVAL) {
3375 pr_err("Failed to initiate reshape!\n");
3376 goto release;
3377 }
3378 /* revert data_offset and try the old way */
3379 for (sd = sra->devs; sd; sd = sd->next) {
3380 sysfs_set_num(sra, sd, "new_offset",
3381 sd->data_offset);
3382 sysfs_set_str(sra, NULL, "reshape_direction",
3383 "forwards");
3384 }
3385 break;
3386 }
3387 if (info->new_level == reshape.level)
3388 return 0;
3389 /* need to adjust level when reshape completes */
3390 switch(fork()) {
3391 case -1: /* ignore error, but don't wait */
3392 return 0;
3393 default: /* parent */
3394 return 0;
3395 case 0:
3396 map_fork();
3397 break;
3398 }
3399 close(fd);
3400 wait_reshape(sra);
3401 fd = open_dev(sra->sys_name);
3402 if (fd >= 0)
3403 impose_level(fd, info->new_level, devname, verbose);
3404 return 0;
3405 case 1: /* Couldn't set data_offset, try the old way */
3406 if (data_offset != INVALID_SECTORS) {
3407 pr_err("Cannot update data_offset on this array\n");
3408 goto release;
3409 }
3410 break;
3411 }
3412
3413 started:
3414 /* Decide how many blocks (sectors) for a reshape
3415 * unit. The number we have so far is just a minimum
3416 */
3417 blocks = reshape.backup_blocks;
3418 if (reshape.before.data_disks ==
3419 reshape.after.data_disks) {
3420 /* Make 'blocks' bigger for better throughput, but
3421 * not so big that we reject it below.
3422 * Try for 16 megabytes
3423 */
3424 while (blocks * 32 < sra->component_size && blocks < 16*1024*2)
3425 blocks *= 2;
3426 } else
3427 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3428
3429 if (blocks >= sra->component_size/2) {
3430 pr_err("%s: Something wrong - reshape aborted\n", devname);
3431 goto release;
3432 }
3433
3434 /* Now we need to open all these devices so we can read/write.
3435 */
3436 nrdisks = max(reshape.before.data_disks,
3437 reshape.after.data_disks) + reshape.parity
3438 + sra->array.spare_disks;
3439 fdlist = xcalloc((1+nrdisks), sizeof(int));
3440 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3441
3442 odisks = reshape.before.data_disks + reshape.parity;
3443 d = reshape_prepare_fdlist(devname, sra, odisks, nrdisks, blocks,
3444 backup_file, fdlist, offsets);
3445 if (d < odisks) {
3446 goto release;
3447 }
3448 if ((st->ss->manage_reshape == NULL) ||
3449 (st->ss->recover_backup == NULL)) {
3450 if (backup_file == NULL) {
3451 if (reshape.after.data_disks <=
3452 reshape.before.data_disks) {
3453 pr_err("%s: Cannot grow - need backup-file\n",
3454 devname);
3455 pr_err(" Please provide one with \"--backup=...\"\n");
3456 goto release;
3457 } else if (d == odisks) {
3458 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3459 goto release;
3460 }
3461 } else {
3462 if (!reshape_open_backup_file(backup_file, fd, devname,
3463 (signed)blocks,
3464 fdlist+d, offsets+d,
3465 sra->sys_name, restart)) {
3466 goto release;
3467 }
3468 d++;
3469 }
3470 }
3471
3472 update_cache_size(container, sra, info,
3473 min(reshape.before.data_disks,
3474 reshape.after.data_disks), blocks);
3475
3476 /* Right, everything seems fine. Let's kick things off.
3477 * If only changing raid_disks, use ioctl, else use
3478 * sysfs.
3479 */
3480 sync_metadata(st);
3481
3482 if (impose_reshape(sra, info, st, fd, restart,
3483 devname, container, &reshape) < 0)
3484 goto release;
3485
3486 err = start_reshape(sra, restart, reshape.before.data_disks,
3487 reshape.after.data_disks);
3488 if (err) {
3489 pr_err("Cannot %s reshape for %s\n",
3490 restart ? "continue" : "start", devname);
3491 goto release;
3492 }
3493 if (restart)
3494 sysfs_set_str(sra, NULL, "array_state", "active");
3495 if (freeze_reshape) {
3496 free(fdlist);
3497 free(offsets);
3498 sysfs_free(sra);
3499 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3500 sra->reshape_progress);
3501 return 1;
3502 }
3503
3504 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3505 if (continue_via_systemd(container ?: sra->sys_name)) {
3506 free(fdlist);
3507 free(offsets);
3508 sysfs_free(sra);
3509 return 0;
3510 }
3511
3512 /* Now we just need to kick off the reshape and watch, while
3513 * handling backups of the data...
3514 * This is all done by a forked background process.
3515 */
3516 switch(forked ? 0 : fork()) {
3517 case -1:
3518 pr_err("Cannot run child to monitor reshape: %s\n",
3519 strerror(errno));
3520 abort_reshape(sra);
3521 goto release;
3522 default:
3523 free(fdlist);
3524 free(offsets);
3525 sysfs_free(sra);
3526 return 0;
3527 case 0:
3528 map_fork();
3529 break;
3530 }
3531
3532 /* If another array on the same devices is busy, the
3533 * reshape will wait for them. This would mean that
3534 * the first section that we suspend will stay suspended
3535 * for a long time. So check on that possibility
3536 * by looking for "DELAYED" in /proc/mdstat, and if found,
3537 * wait a while
3538 */
3539 do {
3540 struct mdstat_ent *mds, *m;
3541 delayed = 0;
3542 mds = mdstat_read(1, 0);
3543 for (m = mds; m; m = m->next)
3544 if (strcmp(m->devnm, sra->sys_name) == 0) {
3545 if (m->resync && m->percent == RESYNC_DELAYED)
3546 delayed = 1;
3547 if (m->resync == 0)
3548 /* Haven't started the reshape thread
3549 * yet, wait a bit
3550 */
3551 delayed = 2;
3552 break;
3553 }
3554 free_mdstat(mds);
3555 if (delayed == 1 && get_linux_version() < 3007000) {
3556 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3557 " You might experience problems until other reshapes complete.\n");
3558 delayed = 0;
3559 }
3560 if (delayed)
3561 mdstat_wait(30 - (delayed-1) * 25);
3562 } while (delayed);
3563 mdstat_close();
3564 close(fd);
3565 if (check_env("MDADM_GROW_VERIFY"))
3566 fd = open(devname, O_RDONLY | O_DIRECT);
3567 else
3568 fd = -1;
3569 mlockall(MCL_FUTURE);
3570
3571 signal(SIGTERM, catch_term);
3572
3573 if (st->ss->external) {
3574 /* metadata handler takes it from here */
3575 done = st->ss->manage_reshape(
3576 fd, sra, &reshape, st, blocks,
3577 fdlist, offsets, d - odisks, fdlist + odisks,
3578 offsets + odisks);
3579 } else
3580 done = child_monitor(
3581 fd, sra, &reshape, st, blocks, fdlist, offsets,
3582 d - odisks, fdlist + odisks, offsets + odisks);
3583
3584 free(fdlist);
3585 free(offsets);
3586
3587 if (backup_file && done) {
3588 char *bul;
3589 bul = make_backup(sra->sys_name);
3590 if (bul) {
3591 char buf[1024];
3592 int l = readlink(bul, buf, sizeof(buf) - 1);
3593 if (l > 0) {
3594 buf[l]=0;
3595 unlink(buf);
3596 }
3597 unlink(bul);
3598 free(bul);
3599 }
3600 unlink(backup_file);
3601 }
3602 if (!done) {
3603 abort_reshape(sra);
3604 goto out;
3605 }
3606
3607 if (!st->ss->external &&
3608 !(reshape.before.data_disks != reshape.after.data_disks &&
3609 info->custom_array_size) && info->new_level == reshape.level &&
3610 !forked) {
3611 /* no need to wait for the reshape to finish as
3612 * there is nothing more to do.
3613 */
3614 sysfs_free(sra);
3615 exit(0);
3616 }
3617 wait_reshape(sra);
3618
3619 if (st->ss->external) {
3620 /* Re-load the metadata as much could have changed */
3621 int cfd = open_dev(st->container_devnm);
3622 if (cfd >= 0) {
3623 flush_mdmon(container);
3624 st->ss->free_super(st);
3625 st->ss->load_container(st, cfd, container);
3626 close(cfd);
3627 }
3628 }
3629
3630 /* set new array size if required customer_array_size is used
3631 * by this metadata.
3632 */
3633 if (reshape.before.data_disks != reshape.after.data_disks &&
3634 info->custom_array_size)
3635 set_array_size(st, info, info->text_version);
3636
3637 if (info->new_level != reshape.level) {
3638 if (fd < 0)
3639 fd = open(devname, O_RDONLY);
3640 impose_level(fd, info->new_level, devname, verbose);
3641 close(fd);
3642 if (info->new_level == 0)
3643 st->update_tail = NULL;
3644 }
3645 out:
3646 sysfs_free(sra);
3647 if (forked)
3648 return 0;
3649 unfreeze(st);
3650 exit(0);
3651
3652 release:
3653 free(fdlist);
3654 free(offsets);
3655 if (orig_level != UnSet && sra) {
3656 c = map_num(pers, orig_level);
3657 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3658 pr_err("aborting level change\n");
3659 }
3660 sysfs_free(sra);
3661 if (!forked)
3662 unfreeze(st);
3663 return 1;
3664 }
3665
3666 /* mdfd handle is passed to be closed in child process (after fork).
3667 */
3668 int reshape_container(char *container, char *devname,
3669 int mdfd,
3670 struct supertype *st,
3671 struct mdinfo *info,
3672 int force,
3673 char *backup_file, int verbose,
3674 int forked, int restart, int freeze_reshape)
3675 {
3676 struct mdinfo *cc = NULL;
3677 int rv = restart;
3678 char last_devnm[32] = "";
3679
3680 /* component_size is not meaningful for a container,
3681 * so pass '0' meaning 'no change'
3682 */
3683 if (!restart &&
3684 reshape_super(st, 0, info->new_level,
3685 info->new_layout, info->new_chunk,
3686 info->array.raid_disks, info->delta_disks,
3687 backup_file, devname, APPLY_METADATA_CHANGES,
3688 verbose)) {
3689 unfreeze(st);
3690 return 1;
3691 }
3692
3693 sync_metadata(st);
3694
3695 /* ping monitor to be sure that update is on disk
3696 */
3697 ping_monitor(container);
3698
3699 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3700 if (continue_via_systemd(container))
3701 return 0;
3702
3703 switch (forked ? 0 : fork()) {
3704 case -1: /* error */
3705 perror("Cannot fork to complete reshape\n");
3706 unfreeze(st);
3707 return 1;
3708 default: /* parent */
3709 if (!freeze_reshape)
3710 printf("%s: multi-array reshape continues in background\n", Name);
3711 return 0;
3712 case 0: /* child */
3713 map_fork();
3714 break;
3715 }
3716
3717 /* close unused handle in child process
3718 */
3719 if (mdfd > -1)
3720 close(mdfd);
3721
3722 while(1) {
3723 /* For each member array with reshape_active,
3724 * we need to perform the reshape.
3725 * We pick the first array that needs reshaping and
3726 * reshape it. reshape_array() will re-read the metadata
3727 * so the next time through a different array should be
3728 * ready for reshape.
3729 * It is possible that the 'different' array will not
3730 * be assembled yet. In that case we simple exit.
3731 * When it is assembled, the mdadm which assembles it
3732 * will take over the reshape.
3733 */
3734 struct mdinfo *content;
3735 int fd;
3736 struct mdstat_ent *mdstat;
3737 char *adev;
3738 dev_t devid;
3739
3740 sysfs_free(cc);
3741
3742 cc = st->ss->container_content(st, NULL);
3743
3744 for (content = cc; content ; content = content->next) {
3745 char *subarray;
3746 if (!content->reshape_active)
3747 continue;
3748
3749 subarray = strchr(content->text_version+1, '/')+1;
3750 mdstat = mdstat_by_subdev(subarray, container);
3751 if (!mdstat)
3752 continue;
3753 if (mdstat->active == 0) {
3754 pr_err("Skipping inactive array %s.\n",
3755 mdstat->devnm);
3756 free_mdstat(mdstat);
3757 mdstat = NULL;
3758 continue;
3759 }
3760 break;
3761 }
3762 if (!content)
3763 break;
3764
3765 devid = devnm2devid(mdstat->devnm);
3766 adev = map_dev(major(devid), minor(devid), 0);
3767 if (!adev)
3768 adev = content->text_version;
3769
3770 fd = open_dev(mdstat->devnm);
3771 if (fd < 0) {
3772 pr_err("Device %s cannot be opened for reshape.\n",
3773 adev);
3774 break;
3775 }
3776
3777 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3778 /* Do not allow for multiple reshape_array() calls for
3779 * the same array.
3780 * It can happen when reshape_array() returns without
3781 * error, when reshape is not finished (wrong reshape
3782 * starting/continuation conditions). Mdmon doesn't
3783 * switch to next array in container and reentry
3784 * conditions for the same array occur.
3785 * This is possibly interim until the behaviour of
3786 * reshape_array is resolved().
3787 */
3788 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3789 close(fd);
3790 break;
3791 }
3792 strcpy(last_devnm, mdstat->devnm);
3793
3794 if (sysfs_init(content, fd, mdstat->devnm)) {
3795 pr_err("Unable to initialize sysfs for %s\n",
3796 mdstat->devnm);
3797 rv = 1;
3798 break;
3799 }
3800
3801 if (mdmon_running(container))
3802 flush_mdmon(container);
3803
3804 rv = reshape_array(container, fd, adev, st,
3805 content, force, NULL, INVALID_SECTORS,
3806 backup_file, verbose, 1, restart,
3807 freeze_reshape);
3808 close(fd);
3809
3810 if (freeze_reshape) {
3811 sysfs_free(cc);
3812 exit(0);
3813 }
3814
3815 restart = 0;
3816 if (rv)
3817 break;
3818
3819 if (mdmon_running(container))
3820 flush_mdmon(container);
3821 }
3822 if (!rv)
3823 unfreeze(st);
3824 sysfs_free(cc);
3825 exit(0);
3826 }
3827
3828 /*
3829 * We run a child process in the background which performs the following
3830 * steps:
3831 * - wait for resync to reach a certain point
3832 * - suspend io to the following section
3833 * - backup that section
3834 * - allow resync to proceed further
3835 * - resume io
3836 * - discard the backup.
3837 *
3838 * When are combined in slightly different ways in the three cases.
3839 * Grow:
3840 * - suspend/backup/allow/wait/resume/discard
3841 * Shrink:
3842 * - allow/wait/suspend/backup/allow/wait/resume/discard
3843 * same-size:
3844 * - wait/resume/discard/suspend/backup/allow
3845 *
3846 * suspend/backup/allow always come together
3847 * wait/resume/discard do too.
3848 * For the same-size case we have two backups to improve flow.
3849 *
3850 */
3851
3852 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3853 unsigned long long backup_point,
3854 unsigned long long wait_point,
3855 unsigned long long *suspend_point,
3856 unsigned long long *reshape_completed, int *frozen)
3857 {
3858 /* This function is called repeatedly by the reshape manager.
3859 * It determines how much progress can safely be made and allows
3860 * that progress.
3861 * - 'info' identifies the array and particularly records in
3862 * ->reshape_progress the metadata's knowledge of progress
3863 * This is a sector offset from the start of the array
3864 * of the next array block to be relocated. This number
3865 * may increase from 0 or decrease from array_size, depending
3866 * on the type of reshape that is happening.
3867 * Note that in contrast, 'sync_completed' is a block count of the
3868 * reshape so far. It gives the distance between the start point
3869 * (head or tail of device) and the next place that data will be
3870 * written. It always increases.
3871 * - 'reshape' is the structure created by analyse_change
3872 * - 'backup_point' shows how much the metadata manager has backed-up
3873 * data. For reshapes with increasing progress, it is the next address
3874 * to be backed up, previous addresses have been backed-up. For
3875 * decreasing progress, it is the earliest address that has been
3876 * backed up - later address are also backed up.
3877 * So addresses between reshape_progress and backup_point are
3878 * backed up providing those are in the 'correct' order.
3879 * - 'wait_point' is an array address. When reshape_completed
3880 * passes this point, progress_reshape should return. It might
3881 * return earlier if it determines that ->reshape_progress needs
3882 * to be updated or further backup is needed.
3883 * - suspend_point is maintained by progress_reshape and the caller
3884 * should not touch it except to initialise to zero.
3885 * It is an array address and it only increases in 2.6.37 and earlier.
3886 * This makes it difficult to handle reducing reshapes with
3887 * external metadata.
3888 * However: it is similar to backup_point in that it records the
3889 * other end of a suspended region from reshape_progress.
3890 * it is moved to extend the region that is safe to backup and/or
3891 * reshape
3892 * - reshape_completed is read from sysfs and returned. The caller
3893 * should copy this into ->reshape_progress when it has reason to
3894 * believe that the metadata knows this, and any backup outside this
3895 * has been erased.
3896 *
3897 * Return value is:
3898 * 1 if more data from backup_point - but only as far as suspend_point,
3899 * should be backed up
3900 * 0 if things are progressing smoothly
3901 * -1 if the reshape is finished because it is all done,
3902 * -2 if the reshape is finished due to an error.
3903 */
3904
3905 int advancing = (reshape->after.data_disks
3906 >= reshape->before.data_disks);
3907 unsigned long long need_backup; /* All data between start of array and
3908 * here will at some point need to
3909 * be backed up.
3910 */
3911 unsigned long long read_offset, write_offset;
3912 unsigned long long write_range;
3913 unsigned long long max_progress, target, completed;
3914 unsigned long long array_size = (info->component_size
3915 * reshape->before.data_disks);
3916 int fd;
3917 char buf[20];
3918
3919 /* First, we unsuspend any region that is now known to be safe.
3920 * If suspend_point is on the 'wrong' side of reshape_progress, then
3921 * we don't have or need suspension at the moment. This is true for
3922 * native metadata when we don't need to back-up.
3923 */
3924 if (advancing) {
3925 if (info->reshape_progress <= *suspend_point)
3926 sysfs_set_num(info, NULL, "suspend_lo",
3927 info->reshape_progress);
3928 } else {
3929 /* Note: this won't work in 2.6.37 and before.
3930 * Something somewhere should make sure we don't need it!
3931 */
3932 if (info->reshape_progress >= *suspend_point)
3933 sysfs_set_num(info, NULL, "suspend_hi",
3934 info->reshape_progress);
3935 }
3936
3937 /* Now work out how far it is safe to progress.
3938 * If the read_offset for ->reshape_progress is less than
3939 * 'blocks' beyond the write_offset, we can only progress as far
3940 * as a backup.
3941 * Otherwise we can progress until the write_offset for the new location
3942 * reaches (within 'blocks' of) the read_offset at the current location.
3943 * However that region must be suspended unless we are using native
3944 * metadata.
3945 * If we need to suspend more, we limit it to 128M per device, which is
3946 * rather arbitrary and should be some time-based calculation.
3947 */
3948 read_offset = info->reshape_progress / reshape->before.data_disks;
3949 write_offset = info->reshape_progress / reshape->after.data_disks;
3950 write_range = info->new_chunk/512;
3951 if (reshape->before.data_disks == reshape->after.data_disks)
3952 need_backup = array_size;
3953 else
3954 need_backup = reshape->backup_blocks;
3955 if (advancing) {
3956 if (read_offset < write_offset + write_range)
3957 max_progress = backup_point;
3958 else
3959 max_progress =
3960 read_offset * reshape->after.data_disks;
3961 } else {
3962 if (read_offset > write_offset - write_range)
3963 /* Can only progress as far as has been backed up,
3964 * which must be suspended */
3965 max_progress = backup_point;
3966 else if (info->reshape_progress <= need_backup)
3967 max_progress = backup_point;
3968 else {
3969 if (info->array.major_version >= 0)
3970 /* Can progress until backup is needed */
3971 max_progress = need_backup;
3972 else {
3973 /* Can progress until metadata update is required */
3974 max_progress =
3975 read_offset * reshape->after.data_disks;
3976 /* but data must be suspended */
3977 if (max_progress < *suspend_point)
3978 max_progress = *suspend_point;
3979 }
3980 }
3981 }
3982
3983 /* We know it is safe to progress to 'max_progress' providing
3984 * it is suspended or we are using native metadata.
3985 * Consider extending suspend_point 128M per device if it
3986 * is less than 64M per device beyond reshape_progress.
3987 * But always do a multiple of 'blocks'
3988 * FIXME this is too big - it takes to long to complete
3989 * this much.
3990 */
3991 target = 64*1024*2 * min(reshape->before.data_disks,
3992 reshape->after.data_disks);
3993 target /= reshape->backup_blocks;
3994 if (target < 2)
3995 target = 2;
3996 target *= reshape->backup_blocks;
3997
3998 /* For externally managed metadata we always need to suspend IO to
3999 * the area being reshaped so we regularly push suspend_point forward.
4000 * For native metadata we only need the suspend if we are going to do
4001 * a backup.
4002 */
4003 if (advancing) {
4004 if ((need_backup > info->reshape_progress ||
4005 info->array.major_version < 0) &&
4006 *suspend_point < info->reshape_progress + target) {
4007 if (need_backup < *suspend_point + 2 * target)
4008 *suspend_point = need_backup;
4009 else if (*suspend_point + 2 * target < array_size)
4010 *suspend_point += 2 * target;
4011 else
4012 *suspend_point = array_size;
4013 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
4014 if (max_progress > *suspend_point)
4015 max_progress = *suspend_point;
4016 }
4017 } else {
4018 if (info->array.major_version >= 0) {
4019 /* Only need to suspend when about to backup */
4020 if (info->reshape_progress < need_backup * 2 &&
4021 *suspend_point > 0) {
4022 *suspend_point = 0;
4023 sysfs_set_num(info, NULL, "suspend_lo", 0);
4024 sysfs_set_num(info, NULL, "suspend_hi",
4025 need_backup);
4026 }
4027 } else {
4028 /* Need to suspend continually */
4029 if (info->reshape_progress < *suspend_point)
4030 *suspend_point = info->reshape_progress;
4031 if (*suspend_point + target < info->reshape_progress)
4032 /* No need to move suspend region yet */;
4033 else {
4034 if (*suspend_point >= 2 * target)
4035 *suspend_point -= 2 * target;
4036 else
4037 *suspend_point = 0;
4038 sysfs_set_num(info, NULL, "suspend_lo",
4039 *suspend_point);
4040 }
4041 if (max_progress < *suspend_point)
4042 max_progress = *suspend_point;
4043 }
4044 }
4045
4046 /* now set sync_max to allow that progress. sync_max, like
4047 * sync_completed is a count of sectors written per device, so
4048 * we find the difference between max_progress and the start point,
4049 * and divide that by after.data_disks to get a sync_max
4050 * number.
4051 * At the same time we convert wait_point to a similar number
4052 * for comparing against sync_completed.
4053 */
4054 /* scale down max_progress to per_disk */
4055 max_progress /= reshape->after.data_disks;
4056 /*
4057 * Round to chunk size as some kernels give an erroneously
4058 * high number
4059 */
4060 max_progress /= info->new_chunk/512;
4061 max_progress *= info->new_chunk/512;
4062 /* And round to old chunk size as the kernel wants that */
4063 max_progress /= info->array.chunk_size/512;
4064 max_progress *= info->array.chunk_size/512;
4065 /* Limit progress to the whole device */
4066 if (max_progress > info->component_size)
4067 max_progress = info->component_size;
4068 wait_point /= reshape->after.data_disks;
4069 if (!advancing) {
4070 /* switch from 'device offset' to 'processed block count' */
4071 max_progress = info->component_size - max_progress;
4072 wait_point = info->component_size - wait_point;
4073 }
4074
4075 if (!*frozen)
4076 sysfs_set_num(info, NULL, "sync_max", max_progress);
4077
4078 /* Now wait. If we have already reached the point that we were
4079 * asked to wait to, don't wait at all, else wait for any change.
4080 * We need to select on 'sync_completed' as that is the place that
4081 * notifications happen, but we are really interested in
4082 * 'reshape_position'
4083 */
4084 fd = sysfs_get_fd(info, NULL, "sync_completed");
4085 if (fd < 0)
4086 goto check_progress;
4087
4088 if (sysfs_fd_get_ll(fd, &completed) < 0)
4089 goto check_progress;
4090
4091 while (completed < max_progress && completed < wait_point) {
4092 /* Check that sync_action is still 'reshape' to avoid
4093 * waiting forever on a dead array
4094 */
4095 char action[20];
4096 if (sysfs_get_str(info, NULL, "sync_action", action, 20) <= 0 ||
4097 strncmp(action, "reshape", 7) != 0)
4098 break;
4099 /* Some kernels reset 'sync_completed' to zero
4100 * before setting 'sync_action' to 'idle'.
4101 * So we need these extra tests.
4102 */
4103 if (completed == 0 && advancing &&
4104 strncmp(action, "idle", 4) == 0 &&
4105 info->reshape_progress > 0)
4106 break;
4107 if (completed == 0 && !advancing &&
4108 strncmp(action, "idle", 4) == 0 &&
4109 info->reshape_progress <
4110 (info->component_size * reshape->after.data_disks))
4111 break;
4112 sysfs_wait(fd, NULL);
4113 if (sysfs_fd_get_ll(fd, &completed) < 0)
4114 goto check_progress;
4115 }
4116 /* Some kernels reset 'sync_completed' to zero,
4117 * we need to have real point we are in md.
4118 * So in that case, read 'reshape_position' from sysfs.
4119 */
4120 if (completed == 0) {
4121 unsigned long long reshapep;
4122 char action[20];
4123 if (sysfs_get_str(info, NULL, "sync_action", action, 20) > 0 &&
4124 strncmp(action, "idle", 4) == 0 &&
4125 sysfs_get_ll(info, NULL,
4126 "reshape_position", &reshapep) == 0)
4127 *reshape_completed = reshapep;
4128 } else {
4129 /* some kernels can give an incorrectly high
4130 * 'completed' number, so round down */
4131 completed /= (info->new_chunk/512);
4132 completed *= (info->new_chunk/512);
4133 /* Convert 'completed' back in to a 'progress' number */
4134 completed *= reshape->after.data_disks;
4135 if (!advancing)
4136 completed = (info->component_size
4137 * reshape->after.data_disks
4138 - completed);
4139 *reshape_completed = completed;
4140 }
4141
4142 close(fd);
4143
4144 /* We return the need_backup flag. Caller will decide
4145 * how much - a multiple of ->backup_blocks up to *suspend_point
4146 */
4147 if (advancing)
4148 return need_backup > info->reshape_progress;
4149 else
4150 return need_backup >= info->reshape_progress;
4151
4152 check_progress:
4153 /* if we couldn't read a number from sync_completed, then
4154 * either the reshape did complete, or it aborted.
4155 * We can tell which by checking for 'none' in reshape_position.
4156 * If it did abort, then it might immediately restart if it
4157 * it was just a device failure that leaves us degraded but
4158 * functioning.
4159 */
4160 if (sysfs_get_str(info, NULL, "reshape_position", buf,
4161 sizeof(buf)) < 0 || strncmp(buf, "none", 4) != 0) {
4162 /* The abort might only be temporary. Wait up to 10
4163 * seconds for fd to contain a valid number again.
4164 */
4165 int wait = 10000;
4166 int rv = -2;
4167 unsigned long long new_sync_max;
4168 while (fd >= 0 && rv < 0 && wait > 0) {
4169 if (sysfs_wait(fd, &wait) != 1)
4170 break;
4171 switch (sysfs_fd_get_ll(fd, &completed)) {
4172 case 0:
4173 /* all good again */
4174 rv = 1;
4175 /* If "sync_max" is no longer max_progress
4176 * we need to freeze things
4177 */
4178 sysfs_get_ll(info, NULL, "sync_max",
4179 &new_sync_max);
4180 *frozen = (new_sync_max != max_progress);
4181 break;
4182 case -2: /* read error - abort */
4183 wait = 0;
4184 break;
4185 }
4186 }
4187 if (fd >= 0)
4188 close(fd);
4189 return rv; /* abort */
4190 } else {
4191 /* Maybe racing with array shutdown - check state */
4192 if (fd >= 0)
4193 close(fd);
4194 if (sysfs_get_str(info, NULL, "array_state", buf,
4195 sizeof(buf)) < 0 ||
4196 strncmp(buf, "inactive", 8) == 0 ||
4197 strncmp(buf, "clear",5) == 0)
4198 return -2; /* abort */
4199 return -1; /* complete */
4200 }
4201 }
4202
4203 /* FIXME return status is never checked */
4204 static int grow_backup(struct mdinfo *sra,
4205 unsigned long long offset, /* per device */
4206 unsigned long stripes, /* per device, in old chunks */
4207 int *sources, unsigned long long *offsets,
4208 int disks, int chunk, int level, int layout,
4209 int dests, int *destfd, unsigned long long *destoffsets,
4210 int part, int *degraded,
4211 char *buf)
4212 {
4213 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4214 * to storage 'destfd' (offset 'destoffsets'), after first
4215 * suspending IO. Then allow resync to continue
4216 * over the suspended section.
4217 * Use part 'part' of the backup-super-block.
4218 */
4219 int odata = disks;
4220 int rv = 0;
4221 int i;
4222 unsigned long long ll;
4223 int new_degraded;
4224 //printf("offset %llu\n", offset);
4225 if (level >= 4)
4226 odata--;
4227 if (level == 6)
4228 odata--;
4229
4230 /* Check that array hasn't become degraded, else we might backup the wrong data */
4231 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4232 return -1; /* FIXME this error is ignored */
4233 new_degraded = (int)ll;
4234 if (new_degraded != *degraded) {
4235 /* check each device to ensure it is still working */
4236 struct mdinfo *sd;
4237 for (sd = sra->devs ; sd ; sd = sd->next) {
4238 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4239 continue;
4240 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4241 char sbuf[100];
4242
4243 if (sysfs_get_str(sra, sd, "state",
4244 sbuf, sizeof(sbuf)) < 0 ||
4245 strstr(sbuf, "faulty") ||
4246 strstr(sbuf, "in_sync") == NULL) {
4247 /* this device is dead */
4248 sd->disk.state = (1<<MD_DISK_FAULTY);
4249 if (sd->disk.raid_disk >= 0 &&
4250 sources[sd->disk.raid_disk] >= 0) {
4251 close(sources[sd->disk.raid_disk]);
4252 sources[sd->disk.raid_disk] = -1;
4253 }
4254 }
4255 }
4256 }
4257 *degraded = new_degraded;
4258 }
4259 if (part) {
4260 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4261 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4262 } else {
4263 bsb.arraystart = __cpu_to_le64(offset * odata);
4264 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4265 }
4266 if (part)
4267 bsb.magic[15] = '2';
4268 for (i = 0; i < dests; i++)
4269 if (part)
4270 lseek64(destfd[i], destoffsets[i] +
4271 __le64_to_cpu(bsb.devstart2)*512, 0);
4272 else
4273 lseek64(destfd[i], destoffsets[i], 0);
4274
4275 rv = save_stripes(sources, offsets, disks, chunk, level, layout,
4276 dests, destfd, offset * 512 * odata,
4277 stripes * chunk * odata, buf);
4278
4279 if (rv)
4280 return rv;
4281 bsb.mtime = __cpu_to_le64(time(0));
4282 for (i = 0; i < dests; i++) {
4283 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4284
4285 bsb.sb_csum = bsb_csum((char*)&bsb,
4286 ((char*)&bsb.sb_csum)-((char*)&bsb));
4287 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4288 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4289 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4290
4291 rv = -1;
4292 if ((unsigned long long)lseek64(destfd[i],
4293 destoffsets[i] - 4096, 0) !=
4294 destoffsets[i] - 4096)
4295 break;
4296 if (write(destfd[i], &bsb, 512) != 512)
4297 break;
4298 if (destoffsets[i] > 4096) {
4299 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4300 destoffsets[i]+stripes*chunk*odata)
4301 break;
4302 if (write(destfd[i], &bsb, 512) != 512)
4303 break;
4304 }
4305 fsync(destfd[i]);
4306 rv = 0;
4307 }
4308
4309 return rv;
4310 }
4311
4312 /* in 2.6.30, the value reported by sync_completed can be
4313 * less that it should be by one stripe.
4314 * This only happens when reshape hits sync_max and pauses.
4315 * So allow wait_backup to either extent sync_max further
4316 * than strictly necessary, or return before the
4317 * sync has got quite as far as we would really like.
4318 * This is what 'blocks2' is for.
4319 * The various caller give appropriate values so that
4320 * every works.
4321 */
4322 /* FIXME return value is often ignored */
4323 static int forget_backup(int dests, int *destfd,
4324 unsigned long long *destoffsets,
4325 int part)
4326 {
4327 /*
4328 * Erase backup 'part' (which is 0 or 1)
4329 */
4330 int i;
4331 int rv;
4332
4333 if (part) {
4334 bsb.arraystart2 = __cpu_to_le64(0);
4335 bsb.length2 = __cpu_to_le64(0);
4336 } else {
4337 bsb.arraystart = __cpu_to_le64(0);
4338 bsb.length = __cpu_to_le64(0);
4339 }
4340 bsb.mtime = __cpu_to_le64(time(0));
4341 rv = 0;
4342 for (i = 0; i < dests; i++) {
4343 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4344 bsb.sb_csum = bsb_csum((char*)&bsb,
4345 ((char*)&bsb.sb_csum)-((char*)&bsb));
4346 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4347 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4348 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4349 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4350 destoffsets[i]-4096)
4351 rv = -1;
4352 if (rv == 0 && write(destfd[i], &bsb, 512) != 512)
4353 rv = -1;
4354 fsync(destfd[i]);
4355 }
4356 return rv;
4357 }
4358
4359 static void fail(char *msg)
4360 {
4361 int rv;
4362 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4363 rv |= (write(2, "\n", 1) != 1);
4364 exit(rv ? 1 : 2);
4365 }
4366
4367 static char *abuf, *bbuf;
4368 static unsigned long long abuflen;
4369 static void validate(int afd, int bfd, unsigned long long offset)
4370 {
4371 /* check that the data in the backup against the array.
4372 * This is only used for regression testing and should not
4373 * be used while the array is active
4374 */
4375 if (afd < 0)
4376 return;
4377 lseek64(bfd, offset - 4096, 0);
4378 if (read(bfd, &bsb2, 512) != 512)
4379 fail("cannot read bsb");
4380 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4381 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4382 fail("first csum bad");
4383 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4384 fail("magic is bad");
4385 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4386 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4387 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4388 fail("second csum bad");
4389
4390 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4391 fail("devstart is wrong");
4392
4393 if (bsb2.length) {
4394 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4395
4396 if (abuflen < len) {
4397 free(abuf);
4398 free(bbuf);
4399 abuflen = len;
4400 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4401 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4402 abuflen = 0;
4403 /* just stop validating on mem-alloc failure */
4404 return;
4405 }
4406 }
4407
4408 lseek64(bfd, offset, 0);
4409 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4410 //printf("len %llu\n", len);
4411 fail("read first backup failed");
4412 }
4413 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4414 if ((unsigned long long)read(afd, abuf, len) != len)
4415 fail("read first from array failed");
4416 if (memcmp(bbuf, abuf, len) != 0) {
4417 #if 0
4418 int i;
4419 printf("offset=%llu len=%llu\n",
4420 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4421 for (i=0; i<len; i++)
4422 if (bbuf[i] != abuf[i]) {
4423 printf("first diff byte %d\n", i);
4424 break;
4425 }
4426 #endif
4427 fail("data1 compare failed");
4428 }
4429 }
4430 if (bsb2.length2) {
4431 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4432
4433 if (abuflen < len) {
4434 free(abuf);
4435 free(bbuf);
4436 abuflen = len;
4437 abuf = xmalloc(abuflen);
4438 bbuf = xmalloc(abuflen);
4439 }
4440
4441 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4442 if ((unsigned long long)read(bfd, bbuf, len) != len)
4443 fail("read second backup failed");
4444 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4445 if ((unsigned long long)read(afd, abuf, len) != len)
4446 fail("read second from array failed");
4447 if (memcmp(bbuf, abuf, len) != 0)
4448 fail("data2 compare failed");
4449 }
4450 }
4451
4452 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4453 struct supertype *st, unsigned long blocks,
4454 int *fds, unsigned long long *offsets,
4455 int dests, int *destfd, unsigned long long *destoffsets)
4456 {
4457 /* Monitor a reshape where backup is being performed using
4458 * 'native' mechanism - either to a backup file, or
4459 * to some space in a spare.
4460 */
4461 char *buf;
4462 int degraded = -1;
4463 unsigned long long speed;
4464 unsigned long long suspend_point, array_size;
4465 unsigned long long backup_point, wait_point;
4466 unsigned long long reshape_completed;
4467 int done = 0;
4468 int increasing = reshape->after.data_disks >=
4469 reshape->before.data_disks;
4470 int part = 0; /* The next part of the backup area to fill. It
4471 * may already be full, so we need to check */
4472 int level = reshape->level;
4473 int layout = reshape->before.layout;
4474 int data = reshape->before.data_disks;
4475 int disks = reshape->before.data_disks + reshape->parity;
4476 int chunk = sra->array.chunk_size;
4477 struct mdinfo *sd;
4478 unsigned long stripes;
4479 int uuid[4];
4480 int frozen = 0;
4481
4482 /* set up the backup-super-block. This requires the
4483 * uuid from the array.
4484 */
4485 /* Find a superblock */
4486 for (sd = sra->devs; sd; sd = sd->next) {
4487 char *dn;
4488 int devfd;
4489 int ok;
4490 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4491 continue;
4492 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4493 devfd = dev_open(dn, O_RDONLY);
4494 if (devfd < 0)
4495 continue;
4496 ok = st->ss->load_super(st, devfd, NULL);
4497 close(devfd);
4498 if (ok == 0)
4499 break;
4500 }
4501 if (!sd) {
4502 pr_err("Cannot find a superblock\n");
4503 return 0;
4504 }
4505
4506 memset(&bsb, 0, 512);
4507 memcpy(bsb.magic, "md_backup_data-1", 16);
4508 st->ss->uuid_from_super(st, uuid);
4509 memcpy(bsb.set_uuid, uuid, 16);
4510 bsb.mtime = __cpu_to_le64(time(0));
4511 bsb.devstart2 = blocks;
4512
4513 stripes = blocks / (sra->array.chunk_size/512) /
4514 reshape->before.data_disks;
4515
4516 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4517 /* Don't start the 'reshape' */
4518 return 0;
4519 if (reshape->before.data_disks == reshape->after.data_disks) {
4520 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4521 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4522 }
4523
4524 if (increasing) {
4525 array_size = sra->component_size * reshape->after.data_disks;
4526 backup_point = sra->reshape_progress;
4527 suspend_point = 0;
4528 } else {
4529 array_size = sra->component_size * reshape->before.data_disks;
4530 backup_point = reshape->backup_blocks;
4531 suspend_point = array_size;
4532 }
4533
4534 while (!done) {
4535 int rv;
4536
4537 /* Want to return as soon the oldest backup slot can
4538 * be released as that allows us to start backing up
4539 * some more, providing suspend_point has been
4540 * advanced, which it should have.
4541 */
4542 if (increasing) {
4543 wait_point = array_size;
4544 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4545 wait_point = (__le64_to_cpu(bsb.arraystart) +
4546 __le64_to_cpu(bsb.length));
4547 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4548 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4549 __le64_to_cpu(bsb.length2));
4550 } else {
4551 wait_point = 0;
4552 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4553 wait_point = __le64_to_cpu(bsb.arraystart);
4554 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4555 wait_point = __le64_to_cpu(bsb.arraystart2);
4556 }
4557
4558 reshape_completed = sra->reshape_progress;
4559 rv = progress_reshape(sra, reshape,
4560 backup_point, wait_point,
4561 &suspend_point, &reshape_completed,
4562 &frozen);
4563 /* external metadata would need to ping_monitor here */
4564 sra->reshape_progress = reshape_completed;
4565
4566 /* Clear any backup region that is before 'here' */
4567 if (increasing) {
4568 if (__le64_to_cpu(bsb.length) > 0 &&
4569 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4570 __le64_to_cpu(bsb.length)))
4571 forget_backup(dests, destfd,
4572 destoffsets, 0);
4573 if (__le64_to_cpu(bsb.length2) > 0 &&
4574 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4575 __le64_to_cpu(bsb.length2)))
4576 forget_backup(dests, destfd,
4577 destoffsets, 1);
4578 } else {
4579 if (__le64_to_cpu(bsb.length) > 0 &&
4580 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4581 forget_backup(dests, destfd,
4582 destoffsets, 0);
4583 if (__le64_to_cpu(bsb.length2) > 0 &&
4584 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4585 forget_backup(dests, destfd,
4586 destoffsets, 1);
4587 }
4588 if (sigterm)
4589 rv = -2;
4590 if (rv < 0) {
4591 if (rv == -1)
4592 done = 1;
4593 break;
4594 }
4595 if (rv == 0 && increasing && !st->ss->external) {
4596 /* No longer need to monitor this reshape */
4597 sysfs_set_str(sra, NULL, "sync_max", "max");
4598 done = 1;
4599 break;
4600 }
4601
4602 while (rv) {
4603 unsigned long long offset;
4604 unsigned long actual_stripes;
4605 /* Need to backup some data.
4606 * If 'part' is not used and the desired
4607 * backup size is suspended, do a backup,
4608 * then consider the next part.
4609 */
4610 /* Check that 'part' is unused */
4611 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4612 break;
4613 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4614 break;
4615
4616 offset = backup_point / data;
4617 actual_stripes = stripes;
4618 if (increasing) {
4619 if (offset + actual_stripes * (chunk/512) >
4620 sra->component_size)
4621 actual_stripes = ((sra->component_size - offset)
4622 / (chunk/512));
4623 if (offset + actual_stripes * (chunk/512) >
4624 suspend_point/data)
4625 break;
4626 } else {
4627 if (offset < actual_stripes * (chunk/512))
4628 actual_stripes = offset / (chunk/512);
4629 offset -= actual_stripes * (chunk/512);
4630 if (offset < suspend_point/data)
4631 break;
4632 }
4633 if (actual_stripes == 0)
4634 break;
4635 grow_backup(sra, offset, actual_stripes, fds, offsets,
4636 disks, chunk, level, layout, dests, destfd,
4637 destoffsets, part, &degraded, buf);
4638 validate(afd, destfd[0], destoffsets[0]);
4639 /* record where 'part' is up to */
4640 part = !part;
4641 if (increasing)
4642 backup_point += actual_stripes * (chunk/512) * data;
4643 else
4644 backup_point -= actual_stripes * (chunk/512) * data;
4645 }
4646 }
4647
4648 /* FIXME maybe call progress_reshape one more time instead */
4649 /* remove any remaining suspension */
4650 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4651 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4652 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4653 sysfs_set_num(sra, NULL, "sync_min", 0);
4654
4655 if (reshape->before.data_disks == reshape->after.data_disks)
4656 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4657 free(buf);
4658 return done;
4659 }
4660
4661 /*
4662 * If any spare contains md_back_data-1 which is recent wrt mtime,
4663 * write that data into the array and update the super blocks with
4664 * the new reshape_progress
4665 */
4666 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist,
4667 int cnt, char *backup_file, int verbose)
4668 {
4669 int i, j;
4670 int old_disks;
4671 unsigned long long *offsets;
4672 unsigned long long nstripe, ostripe;
4673 int ndata, odata;
4674
4675 odata = info->array.raid_disks - info->delta_disks - 1;
4676 if (info->array.level == 6)
4677 odata--; /* number of data disks */
4678 ndata = info->array.raid_disks - 1;
4679 if (info->new_level == 6)
4680 ndata--;
4681
4682 old_disks = info->array.raid_disks - info->delta_disks;
4683
4684 if (info->delta_disks <= 0)
4685 /* Didn't grow, so the backup file must have
4686 * been used
4687 */
4688 old_disks = cnt;
4689 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4690 struct mdinfo dinfo;
4691 int fd;
4692 int bsbsize;
4693 char *devname, namebuf[20];
4694 unsigned long long lo, hi;
4695
4696 /* This was a spare and may have some saved data on it.
4697 * Load the superblock, find and load the
4698 * backup_super_block.
4699 * If either fail, go on to next device.
4700 * If the backup contains no new info, just return
4701 * else restore data and update all superblocks
4702 */
4703 if (i == old_disks-1) {
4704 fd = open(backup_file, O_RDONLY);
4705 if (fd<0) {
4706 pr_err("backup file %s inaccessible: %s\n",
4707 backup_file, strerror(errno));
4708 continue;
4709 }
4710 devname = backup_file;
4711 } else {
4712 fd = fdlist[i];
4713 if (fd < 0)
4714 continue;
4715 if (st->ss->load_super(st, fd, NULL))
4716 continue;
4717
4718 st->ss->getinfo_super(st, &dinfo, NULL);
4719 st->ss->free_super(st);
4720
4721 if (lseek64(fd,
4722 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4723 0) < 0) {
4724 pr_err("Cannot seek on device %d\n", i);
4725 continue; /* Cannot seek */
4726 }
4727 sprintf(namebuf, "device-%d", i);
4728 devname = namebuf;
4729 }
4730 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4731 if (verbose)
4732 pr_err("Cannot read from %s\n", devname);
4733 continue; /* Cannot read */
4734 }
4735 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4736 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4737 if (verbose)
4738 pr_err("No backup metadata on %s\n", devname);
4739 continue;
4740 }
4741 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4742 if (verbose)
4743 pr_err("Bad backup-metadata checksum on %s\n",
4744 devname);
4745 continue; /* bad checksum */
4746 }
4747 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4748 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4749 if (verbose)
4750 pr_err("Bad backup-metadata checksum2 on %s\n",
4751 devname);
4752 continue; /* Bad second checksum */
4753 }
4754 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4755 if (verbose)
4756 pr_err("Wrong uuid on backup-metadata on %s\n",
4757 devname);
4758 continue; /* Wrong uuid */
4759 }
4760
4761 /*
4762 * array utime and backup-mtime should be updated at
4763 * much the same time, but it seems that sometimes
4764 * they aren't... So allow considerable flexability in
4765 * matching, and allow this test to be overridden by
4766 * an environment variable.
4767 */
4768 if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) ||
4769 time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) {
4770 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4771 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4772 (unsigned long)__le64_to_cpu(bsb.mtime),
4773 (unsigned long)info->array.utime);
4774 } else {
4775 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4776 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4777 continue; /* time stamp is too bad */
4778 }
4779 }
4780
4781 if (bsb.magic[15] == '1') {
4782 if (bsb.length == 0)
4783 continue;
4784 if (info->delta_disks >= 0) {
4785 /* reshape_progress is increasing */
4786 if (__le64_to_cpu(bsb.arraystart)
4787 + __le64_to_cpu(bsb.length)
4788 < info->reshape_progress) {
4789 nonew:
4790 if (verbose)
4791 pr_err("backup-metadata found on %s but is not needed\n", devname);
4792 continue; /* No new data here */
4793 }
4794 } else {
4795 /* reshape_progress is decreasing */
4796 if (__le64_to_cpu(bsb.arraystart) >=
4797 info->reshape_progress)
4798 goto nonew; /* No new data here */
4799 }
4800 } else {
4801 if (bsb.length == 0 && bsb.length2 == 0)
4802 continue;
4803 if (info->delta_disks >= 0) {
4804 /* reshape_progress is increasing */
4805 if ((__le64_to_cpu(bsb.arraystart)
4806 + __le64_to_cpu(bsb.length)
4807 < info->reshape_progress) &&
4808 (__le64_to_cpu(bsb.arraystart2)
4809 + __le64_to_cpu(bsb.length2)
4810 < info->reshape_progress))
4811 goto nonew; /* No new data here */
4812 } else {
4813 /* reshape_progress is decreasing */
4814 if (__le64_to_cpu(bsb.arraystart) >=
4815 info->reshape_progress &&
4816 __le64_to_cpu(bsb.arraystart2) >=
4817 info->reshape_progress)
4818 goto nonew; /* No new data here */
4819 }
4820 }
4821 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4822 second_fail:
4823 if (verbose)
4824 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4825 devname);
4826 continue; /* Cannot seek */
4827 }
4828 /* There should be a duplicate backup superblock 4k before here */
4829 if (lseek64(fd, -4096, 1) < 0 ||
4830 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4831 goto second_fail; /* Cannot find leading superblock */
4832 if (bsb.magic[15] == '1')
4833 bsbsize = offsetof(struct mdp_backup_super, pad1);
4834 else
4835 bsbsize = offsetof(struct mdp_backup_super, pad);
4836 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4837 goto second_fail; /* Cannot find leading superblock */
4838
4839 /* Now need the data offsets for all devices. */
4840 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4841 for(j=0; j<info->array.raid_disks; j++) {
4842 if (fdlist[j] < 0)
4843 continue;
4844 if (st->ss->load_super(st, fdlist[j], NULL))
4845 /* FIXME should be this be an error */
4846 continue;
4847 st->ss->getinfo_super(st, &dinfo, NULL);
4848 st->ss->free_super(st);
4849 offsets[j] = dinfo.data_offset * 512;
4850 }
4851 printf("%s: restoring critical section\n", Name);
4852
4853 if (restore_stripes(fdlist, offsets, info->array.raid_disks,
4854 info->new_chunk, info->new_level,
4855 info->new_layout, fd,
4856 __le64_to_cpu(bsb.devstart)*512,
4857 __le64_to_cpu(bsb.arraystart)*512,
4858 __le64_to_cpu(bsb.length)*512, NULL)) {
4859 /* didn't succeed, so giveup */
4860 if (verbose)
4861 pr_err("Error restoring backup from %s\n",
4862 devname);
4863 free(offsets);
4864 return 1;
4865 }
4866
4867 if (bsb.magic[15] == '2' &&
4868 restore_stripes(fdlist, offsets, info->array.raid_disks,
4869 info->new_chunk, info->new_level,
4870 info->new_layout, fd,
4871 __le64_to_cpu(bsb.devstart)*512 +
4872 __le64_to_cpu(bsb.devstart2)*512,
4873 __le64_to_cpu(bsb.arraystart2)*512,
4874 __le64_to_cpu(bsb.length2)*512, NULL)) {
4875 /* didn't succeed, so giveup */
4876 if (verbose)
4877 pr_err("Error restoring second backup from %s\n",
4878 devname);
4879 free(offsets);
4880 return 1;
4881 }
4882
4883 free(offsets);
4884
4885 /* Ok, so the data is restored. Let's update those superblocks. */
4886
4887 lo = hi = 0;
4888 if (bsb.length) {
4889 lo = __le64_to_cpu(bsb.arraystart);
4890 hi = lo + __le64_to_cpu(bsb.length);
4891 }
4892 if (bsb.magic[15] == '2' && bsb.length2) {
4893 unsigned long long lo1, hi1;
4894 lo1 = __le64_to_cpu(bsb.arraystart2);
4895 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4896 if (lo == hi) {
4897 lo = lo1;
4898 hi = hi1;
4899 } else if (lo < lo1)
4900 hi = hi1;
4901 else
4902 lo = lo1;
4903 }
4904 if (lo < hi && (info->reshape_progress < lo ||
4905 info->reshape_progress > hi))
4906 /* backup does not affect reshape_progress*/ ;
4907 else if (info->delta_disks >= 0) {
4908 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4909 __le64_to_cpu(bsb.length);
4910 if (bsb.magic[15] == '2') {
4911 unsigned long long p2;
4912
4913 p2 = __le64_to_cpu(bsb.arraystart2) +
4914 __le64_to_cpu(bsb.length2);
4915 if (p2 > info->reshape_progress)
4916 info->reshape_progress = p2;
4917 }
4918 } else {
4919 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4920 if (bsb.magic[15] == '2') {
4921 unsigned long long p2;
4922
4923 p2 = __le64_to_cpu(bsb.arraystart2);
4924 if (p2 < info->reshape_progress)
4925 info->reshape_progress = p2;
4926 }
4927 }
4928 for (j=0; j<info->array.raid_disks; j++) {
4929 if (fdlist[j] < 0)
4930 continue;
4931 if (st->ss->load_super(st, fdlist[j], NULL))
4932 continue;
4933 st->ss->getinfo_super(st, &dinfo, NULL);
4934 dinfo.reshape_progress = info->reshape_progress;
4935 st->ss->update_super(st, &dinfo, "_reshape_progress",
4936 NULL,0, 0, NULL);
4937 st->ss->store_super(st, fdlist[j]);
4938 st->ss->free_super(st);
4939 }
4940 return 0;
4941 }
4942 /* Didn't find any backup data, try to see if any
4943 * was needed.
4944 */
4945 if (info->delta_disks < 0) {
4946 /* When shrinking, the critical section is at the end.
4947 * So see if we are before the critical section.
4948 */
4949 unsigned long long first_block;
4950 nstripe = ostripe = 0;
4951 first_block = 0;
4952 while (ostripe >= nstripe) {
4953 ostripe += info->array.chunk_size / 512;
4954 first_block = ostripe * odata;
4955 nstripe = first_block / ndata / (info->new_chunk/512) *
4956 (info->new_chunk/512);
4957 }
4958
4959 if (info->reshape_progress >= first_block)
4960 return 0;
4961 }
4962 if (info->delta_disks > 0) {
4963 /* See if we are beyond the critical section. */
4964 unsigned long long last_block;
4965 nstripe = ostripe = 0;
4966 last_block = 0;
4967 while (nstripe >= ostripe) {
4968 nstripe += info->new_chunk / 512;
4969 last_block = nstripe * ndata;
4970 ostripe = last_block / odata / (info->array.chunk_size/512) *
4971 (info->array.chunk_size/512);
4972 }
4973
4974 if (info->reshape_progress >= last_block)
4975 return 0;
4976 }
4977 /* needed to recover critical section! */
4978 if (verbose)
4979 pr_err("Failed to find backup of critical section\n");
4980 return 1;
4981 }
4982
4983 int Grow_continue_command(char *devname, int fd,
4984 char *backup_file, int verbose)
4985 {
4986 int ret_val = 0;
4987 struct supertype *st = NULL;
4988 struct mdinfo *content = NULL;
4989 struct mdinfo array;
4990 char *subarray = NULL;
4991 struct mdinfo *cc = NULL;
4992 struct mdstat_ent *mdstat = NULL;
4993 int cfd = -1;
4994 int fd2;
4995
4996 dprintf("Grow continue from command line called for %s\n", devname);
4997
4998 st = super_by_fd(fd, &subarray);
4999 if (!st || !st->ss) {
5000 pr_err("Unable to determine metadata format for %s\n", devname);
5001 return 1;
5002 }
5003 dprintf("Grow continue is run for ");
5004 if (st->ss->external == 0) {
5005 int d;
5006 int cnt = 5;
5007 dprintf_cont("native array (%s)\n", devname);
5008 if (md_get_array_info(fd, &array.array) < 0) {
5009 pr_err("%s is not an active md array - aborting\n",
5010 devname);
5011 ret_val = 1;
5012 goto Grow_continue_command_exit;
5013 }
5014 content = &array;
5015 sysfs_init(content, fd, NULL);
5016 /* Need to load a superblock.
5017 * FIXME we should really get what we need from
5018 * sysfs
5019 */
5020 do {
5021 for (d = 0; d < MAX_DISKS; d++) {
5022 mdu_disk_info_t disk;
5023 char *dv;
5024 int err;
5025 disk.number = d;
5026 if (md_get_disk_info(fd, &disk) < 0)
5027 continue;
5028 if (disk.major == 0 && disk.minor == 0)
5029 continue;
5030 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
5031 continue;
5032 dv = map_dev(disk.major, disk.minor, 1);
5033 if (!dv)
5034 continue;
5035 fd2 = dev_open(dv, O_RDONLY);
5036 if (fd2 < 0)
5037 continue;
5038 err = st->ss->load_super(st, fd2, NULL);
5039 close(fd2);
5040 if (err)
5041 continue;
5042 break;
5043 }
5044 if (d == MAX_DISKS) {
5045 pr_err("Unable to load metadata for %s\n",
5046 devname);
5047 ret_val = 1;
5048 goto Grow_continue_command_exit;
5049 }
5050 st->ss->getinfo_super(st, content, NULL);
5051 if (!content->reshape_active)
5052 sleep(3);
5053 else
5054 break;
5055 } while (cnt-- > 0);
5056 } else {
5057 char *container;
5058
5059 if (subarray) {
5060 dprintf_cont("subarray (%s)\n", subarray);
5061 container = st->container_devnm;
5062 cfd = open_dev_excl(st->container_devnm);
5063 } else {
5064 container = st->devnm;
5065 close(fd);
5066 cfd = open_dev_excl(st->devnm);
5067 dprintf_cont("container (%s)\n", container);
5068 fd = cfd;
5069 }
5070 if (cfd < 0) {
5071 pr_err("Unable to open container for %s\n", devname);
5072 ret_val = 1;
5073 goto Grow_continue_command_exit;
5074 }
5075
5076 /* find in container array under reshape
5077 */
5078 ret_val = st->ss->load_container(st, cfd, NULL);
5079 if (ret_val) {
5080 pr_err("Cannot read superblock for %s\n", devname);
5081 ret_val = 1;
5082 goto Grow_continue_command_exit;
5083 }
5084
5085 cc = st->ss->container_content(st, subarray);
5086 for (content = cc; content ; content = content->next) {
5087 char *array_name;
5088 int allow_reshape = 1;
5089
5090 if (content->reshape_active == 0)
5091 continue;
5092 /* The decision about array or container wide
5093 * reshape is taken in Grow_continue based
5094 * content->reshape_active state, therefore we
5095 * need to check_reshape based on
5096 * reshape_active and subarray name
5097 */
5098 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
5099 allow_reshape = 0;
5100 if (content->reshape_active == CONTAINER_RESHAPE &&
5101 (content->array.state
5102 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
5103 allow_reshape = 0;
5104
5105 if (!allow_reshape) {
5106 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5107 devname, container);
5108 ret_val = 1;
5109 goto Grow_continue_command_exit;
5110 }
5111
5112 array_name = strchr(content->text_version+1, '/')+1;
5113 mdstat = mdstat_by_subdev(array_name, container);
5114 if (!mdstat)
5115 continue;
5116 if (mdstat->active == 0) {
5117 pr_err("Skipping inactive array %s.\n",
5118 mdstat->devnm);
5119 free_mdstat(mdstat);
5120 mdstat = NULL;
5121 continue;
5122 }
5123 break;
5124 }
5125 if (!content) {
5126 pr_err("Unable to determine reshaped array for %s\n", devname);
5127 ret_val = 1;
5128 goto Grow_continue_command_exit;
5129 }
5130 fd2 = open_dev(mdstat->devnm);
5131 if (fd2 < 0) {
5132 pr_err("cannot open (%s)\n", mdstat->devnm);
5133 ret_val = 1;
5134 goto Grow_continue_command_exit;
5135 }
5136
5137 if (sysfs_init(content, fd2, mdstat->devnm)) {
5138 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5139 mdstat->devnm);
5140 ret_val = 1;
5141 close(fd2);
5142 goto Grow_continue_command_exit;
5143 }
5144
5145 close(fd2);
5146
5147 /* start mdmon in case it is not running
5148 */
5149 if (!mdmon_running(container))
5150 start_mdmon(container);
5151 ping_monitor(container);
5152
5153 if (mdmon_running(container))
5154 st->update_tail = &st->updates;
5155 else {
5156 pr_err("No mdmon found. Grow cannot continue.\n");
5157 ret_val = 1;
5158 goto Grow_continue_command_exit;
5159 }
5160 }
5161
5162 /* verify that array under reshape is started from
5163 * correct position
5164 */
5165 if (verify_reshape_position(content, content->array.level) < 0) {
5166 ret_val = 1;
5167 goto Grow_continue_command_exit;
5168 }
5169
5170 /* continue reshape
5171 */
5172 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
5173
5174 Grow_continue_command_exit:
5175 if (cfd > -1)
5176 close(cfd);
5177 st->ss->free_super(st);
5178 free_mdstat(mdstat);
5179 sysfs_free(cc);
5180 free(subarray);
5181
5182 return ret_val;
5183 }
5184
5185 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
5186 char *backup_file, int forked, int freeze_reshape)
5187 {
5188 int ret_val = 2;
5189
5190 if (!info->reshape_active)
5191 return ret_val;
5192
5193 if (st->ss->external) {
5194 int cfd = open_dev(st->container_devnm);
5195
5196 if (cfd < 0)
5197 return 1;
5198
5199 st->ss->load_container(st, cfd, st->container_devnm);
5200 close(cfd);
5201 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
5202 st, info, 0, backup_file, 0,
5203 forked, 1 | info->reshape_active,
5204 freeze_reshape);
5205 } else
5206 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
5207 NULL, INVALID_SECTORS, backup_file,
5208 0, forked, 1 | info->reshape_active,
5209 freeze_reshape);
5210
5211 return ret_val;
5212 }
5213
5214 char *make_backup(char *name)
5215 {
5216 char *base = "backup_file-";
5217 int len;
5218 char *fname;
5219
5220 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
5221 fname = xmalloc(len);
5222 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
5223 return fname;
5224 }
5225
5226 char *locate_backup(char *name)
5227 {
5228 char *fl = make_backup(name);
5229 struct stat stb;
5230
5231 if (stat(fl, &stb) == 0 && S_ISREG(stb.st_mode))
5232 return fl;
5233
5234 free(fl);
5235 return NULL;
5236 }
Unnamed repository; edit this file 'description' to name the repository.