]> 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
Grow: fix switching on PPL during recovery
[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 sprintf(buf, "%d:%d",
62 dev->disk.major,
63 dev->disk.minor);
64 fd = dev_open(buf, O_RDWR);
65
66 if (dev->disk.raid_disk >= 0)
67 fdlist[dev->disk.raid_disk] = fd;
68 else
69 fdlist[next_spare++] = fd;
70 }
71
72 if (!backup_file) {
73 backup_file = locate_backup(content->sys_name);
74 *backup_filep = backup_file;
75 }
76
77 if (st->ss->external && st->ss->recover_backup)
78 err = st->ss->recover_backup(st, content);
79 else
80 err = Grow_restart(st, content, fdlist, next_spare,
81 backup_file, verbose > 0);
82
83 while (next_spare > 0) {
84 next_spare--;
85 if (fdlist[next_spare] >= 0)
86 close(fdlist[next_spare]);
87 }
88 free(fdlist);
89 if (err) {
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
91 if (!backup_file)
92 pr_err("Possibly you need to specify a --backup-file\n");
93 return 1;
94 }
95
96 dprintf("restore_backup() returns status OK.\n");
97 return 0;
98 }
99
100 int Grow_Add_device(char *devname, int fd, char *newdev)
101 {
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
107 * all other devices.
108 * This means that we need to *find* all other devices.
109 */
110 struct mdinfo info;
111
112 dev_t rdev;
113 int nfd, fd2;
114 int d, nd;
115 struct supertype *st = NULL;
116 char *subarray = NULL;
117
118 if (md_get_array_info(fd, &info.array) < 0) {
119 pr_err("cannot get array info for %s\n", devname);
120 return 1;
121 }
122
123 if (info.array.level != -1) {
124 pr_err("can only add devices to linear arrays\n");
125 return 1;
126 }
127
128 st = super_by_fd(fd, &subarray);
129 if (!st) {
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info.array.major_version);
132 return 1;
133 }
134
135 if (subarray) {
136 pr_err("Cannot grow linear sub-arrays yet\n");
137 free(subarray);
138 free(st);
139 return 1;
140 }
141
142 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
143 if (nfd < 0) {
144 pr_err("cannot open %s\n", newdev);
145 free(st);
146 return 1;
147 }
148 if (!fstat_is_blkdev(nfd, newdev, &rdev)) {
149 close(nfd);
150 free(st);
151 return 1;
152 }
153 /* now check out all the devices and make sure we can read the
154 * superblock */
155 for (d=0 ; d < info.array.raid_disks ; d++) {
156 mdu_disk_info_t disk;
157 char *dv;
158
159 st->ss->free_super(st);
160
161 disk.number = d;
162 if (md_get_disk_info(fd, &disk) < 0) {
163 pr_err("cannot get device detail for device %d\n",
164 d);
165 close(nfd);
166 free(st);
167 return 1;
168 }
169 dv = map_dev(disk.major, disk.minor, 1);
170 if (!dv) {
171 pr_err("cannot find device file for device %d\n",
172 d);
173 close(nfd);
174 free(st);
175 return 1;
176 }
177 fd2 = dev_open(dv, O_RDWR);
178 if (fd2 < 0) {
179 pr_err("cannot open device file %s\n", dv);
180 close(nfd);
181 free(st);
182 return 1;
183 }
184
185 if (st->ss->load_super(st, fd2, NULL)) {
186 pr_err("cannot find super block on %s\n", dv);
187 close(nfd);
188 close(fd2);
189 free(st);
190 return 1;
191 }
192 close(fd2);
193 }
194 /* Ok, looks good. Lets update the superblock and write it out to
195 * newdev.
196 */
197
198 info.disk.number = d;
199 info.disk.major = major(rdev);
200 info.disk.minor = minor(rdev);
201 info.disk.raid_disk = d;
202 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
203 st->ss->update_super(st, &info, "linear-grow-new", newdev,
204 0, 0, NULL);
205
206 if (st->ss->store_super(st, nfd)) {
207 pr_err("Cannot store new superblock on %s\n",
208 newdev);
209 close(nfd);
210 return 1;
211 }
212 close(nfd);
213
214 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
215 pr_err("Cannot add new disk to this array\n");
216 return 1;
217 }
218 /* Well, that seems to have worked.
219 * Now go through and update all superblocks
220 */
221
222 if (md_get_array_info(fd, &info.array) < 0) {
223 pr_err("cannot get array info for %s\n", devname);
224 return 1;
225 }
226
227 nd = d;
228 for (d=0 ; d < info.array.raid_disks ; d++) {
229 mdu_disk_info_t disk;
230 char *dv;
231
232 disk.number = d;
233 if (md_get_disk_info(fd, &disk) < 0) {
234 pr_err("cannot get device detail for device %d\n",
235 d);
236 return 1;
237 }
238 dv = map_dev(disk.major, disk.minor, 1);
239 if (!dv) {
240 pr_err("cannot find device file for device %d\n",
241 d);
242 return 1;
243 }
244 fd2 = dev_open(dv, O_RDWR);
245 if (fd2 < 0) {
246 pr_err("cannot open device file %s\n", dv);
247 return 1;
248 }
249 if (st->ss->load_super(st, fd2, NULL)) {
250 pr_err("cannot find super block on %s\n", dv);
251 close(fd);
252 return 1;
253 }
254 info.array.raid_disks = nd+1;
255 info.array.nr_disks = nd+1;
256 info.array.active_disks = nd+1;
257 info.array.working_disks = nd+1;
258
259 st->ss->update_super(st, &info, "linear-grow-update", dv,
260 0, 0, NULL);
261
262 if (st->ss->store_super(st, fd2)) {
263 pr_err("Cannot store new superblock on %s\n", dv);
264 close(fd2);
265 return 1;
266 }
267 close(fd2);
268 }
269
270 return 0;
271 }
272
273 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
274 {
275 /*
276 * First check that array doesn't have a bitmap
277 * Then create the bitmap
278 * Then add it
279 *
280 * For internal bitmaps, we need to check the version,
281 * find all the active devices, and write the bitmap block
282 * to all devices
283 */
284 mdu_bitmap_file_t bmf;
285 mdu_array_info_t array;
286 struct supertype *st;
287 char *subarray = NULL;
288 int major = BITMAP_MAJOR_HI;
289 unsigned long long bitmapsize, array_size;
290 struct mdinfo *mdi;
291
292 /*
293 * We only ever get called if s->bitmap_file is != NULL, so this check
294 * is just here to quiet down static code checkers.
295 */
296 if (!s->bitmap_file)
297 return 1;
298
299 if (strcmp(s->bitmap_file, "clustered") == 0)
300 major = BITMAP_MAJOR_CLUSTERED;
301
302 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
303 if (errno == ENOMEM)
304 pr_err("Memory allocation failure.\n");
305 else
306 pr_err("bitmaps not supported by this kernel.\n");
307 return 1;
308 }
309 if (bmf.pathname[0]) {
310 if (strcmp(s->bitmap_file,"none") == 0) {
311 if (ioctl(fd, SET_BITMAP_FILE, -1) != 0) {
312 pr_err("failed to remove bitmap %s\n",
313 bmf.pathname);
314 return 1;
315 }
316 return 0;
317 }
318 pr_err("%s already has a bitmap (%s)\n",
319 devname, bmf.pathname);
320 return 1;
321 }
322 if (md_get_array_info(fd, &array) != 0) {
323 pr_err("cannot get array status for %s\n", devname);
324 return 1;
325 }
326 if (array.state & (1 << MD_SB_BITMAP_PRESENT)) {
327 if (strcmp(s->bitmap_file, "none")==0) {
328 array.state &= ~(1 << MD_SB_BITMAP_PRESENT);
329 if (md_set_array_info(fd, &array) != 0) {
330 if (array.state & (1 << MD_SB_CLUSTERED))
331 pr_err("failed to remove clustered bitmap.\n");
332 else
333 pr_err("failed to remove internal bitmap.\n");
334 return 1;
335 }
336 return 0;
337 }
338 pr_err("bitmap already present on %s\n", devname);
339 return 1;
340 }
341
342 if (strcmp(s->bitmap_file, "none") == 0) {
343 pr_err("no bitmap found on %s\n", devname);
344 return 1;
345 }
346 if (array.level <= 0) {
347 pr_err("Bitmaps not meaningful with level %s\n",
348 map_num(pers, array.level)?:"of this array");
349 return 1;
350 }
351 bitmapsize = array.size;
352 bitmapsize <<= 1;
353 if (get_dev_size(fd, NULL, &array_size) &&
354 array_size > (0x7fffffffULL << 9)) {
355 /* Array is big enough that we cannot trust array.size
356 * try other approaches
357 */
358 bitmapsize = get_component_size(fd);
359 }
360 if (bitmapsize == 0) {
361 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
362 return 1;
363 }
364
365 if (array.level == 10) {
366 int ncopies;
367
368 ncopies = (array.layout & 255) * ((array.layout >> 8) & 255);
369 bitmapsize = bitmapsize * array.raid_disks / ncopies;
370 }
371
372 st = super_by_fd(fd, &subarray);
373 if (!st) {
374 pr_err("Cannot understand version %d.%d\n",
375 array.major_version, array.minor_version);
376 return 1;
377 }
378 if (subarray) {
379 pr_err("Cannot add bitmaps to sub-arrays yet\n");
380 free(subarray);
381 free(st);
382 return 1;
383 }
384
385 mdi = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY);
386 if (mdi) {
387 if (mdi->consistency_policy == CONSISTENCY_POLICY_PPL) {
388 pr_err("Cannot add bitmap to array with PPL\n");
389 free(mdi);
390 free(st);
391 return 1;
392 }
393 free(mdi);
394 }
395
396 if (strcmp(s->bitmap_file, "internal") == 0 ||
397 strcmp(s->bitmap_file, "clustered") == 0) {
398 int rv;
399 int d;
400 int offset_setable = 0;
401 if (st->ss->add_internal_bitmap == NULL) {
402 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
403 return 1;
404 }
405 st->nodes = c->nodes;
406 st->cluster_name = c->homecluster;
407 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
408 if (mdi)
409 offset_setable = 1;
410 for (d = 0; d < st->max_devs; d++) {
411 mdu_disk_info_t disk;
412 char *dv;
413 int fd2;
414
415 disk.number = d;
416 if (md_get_disk_info(fd, &disk) < 0)
417 continue;
418 if (disk.major == 0 && disk.minor == 0)
419 continue;
420 if ((disk.state & (1 << MD_DISK_SYNC)) == 0)
421 continue;
422 dv = map_dev(disk.major, disk.minor, 1);
423 if (!dv)
424 continue;
425 fd2 = dev_open(dv, O_RDWR);
426 if (fd2 < 0)
427 continue;
428 rv = st->ss->load_super(st, fd2, NULL);
429 if (!rv) {
430 rv = st->ss->add_internal_bitmap(
431 st, &s->bitmap_chunk, c->delay,
432 s->write_behind, bitmapsize,
433 offset_setable, major);
434 if (!rv) {
435 st->ss->write_bitmap(st, fd2,
436 NodeNumUpdate);
437 } else {
438 pr_err("failed to create internal bitmap - chunksize problem.\n");
439 }
440 } else {
441 pr_err("failed to load super-block.\n");
442 }
443 close(fd2);
444 if (rv)
445 return 1;
446 }
447 if (offset_setable) {
448 st->ss->getinfo_super(st, mdi, NULL);
449 if (sysfs_init(mdi, fd, NULL)) {
450 pr_err("failed to intialize sysfs.\n");
451 free(mdi);
452 }
453 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
454 mdi->bitmap_offset);
455 free(mdi);
456 } else {
457 if (strcmp(s->bitmap_file, "clustered") == 0)
458 array.state |= (1 << MD_SB_CLUSTERED);
459 array.state |= (1 << MD_SB_BITMAP_PRESENT);
460 rv = md_set_array_info(fd, &array);
461 }
462 if (rv < 0) {
463 if (errno == EBUSY)
464 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
465 pr_err("failed to set internal bitmap.\n");
466 return 1;
467 }
468 } else {
469 int uuid[4];
470 int bitmap_fd;
471 int d;
472 int max_devs = st->max_devs;
473
474 /* try to load a superblock */
475 for (d = 0; d < max_devs; d++) {
476 mdu_disk_info_t disk;
477 char *dv;
478 int fd2;
479 disk.number = d;
480 if (md_get_disk_info(fd, &disk) < 0)
481 continue;
482 if ((disk.major==0 && disk.minor == 0) ||
483 (disk.state & (1 << MD_DISK_REMOVED)))
484 continue;
485 dv = map_dev(disk.major, disk.minor, 1);
486 if (!dv)
487 continue;
488 fd2 = dev_open(dv, O_RDONLY);
489 if (fd2 >= 0) {
490 if (st->ss->load_super(st, fd2, NULL) == 0) {
491 close(fd2);
492 st->ss->uuid_from_super(st, uuid);
493 break;
494 }
495 close(fd2);
496 }
497 }
498 if (d == max_devs) {
499 pr_err("cannot find UUID for array!\n");
500 return 1;
501 }
502 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid,
503 s->bitmap_chunk, c->delay, s->write_behind,
504 bitmapsize, major)) {
505 return 1;
506 }
507 bitmap_fd = open(s->bitmap_file, O_RDWR);
508 if (bitmap_fd < 0) {
509 pr_err("weird: %s cannot be opened\n", s->bitmap_file);
510 return 1;
511 }
512 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
513 int err = errno;
514 if (errno == EBUSY)
515 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
516 pr_err("Cannot set bitmap file for %s: %s\n",
517 devname, strerror(err));
518 return 1;
519 }
520 }
521
522 return 0;
523 }
524
525 int Grow_consistency_policy(char *devname, int fd, struct context *c, struct shape *s)
526 {
527 struct supertype *st;
528 struct mdinfo *sra;
529 struct mdinfo *sd;
530 char *subarray = NULL;
531 int ret = 0;
532 char container_dev[PATH_MAX];
533 char buf[20];
534
535 if (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
536 s->consistency_policy != CONSISTENCY_POLICY_PPL) {
537 pr_err("Operation not supported for consistency policy %s\n",
538 map_num(consistency_policies, s->consistency_policy));
539 return 1;
540 }
541
542 st = super_by_fd(fd, &subarray);
543 if (!st)
544 return 1;
545
546 sra = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY|GET_LEVEL|
547 GET_DEVS|GET_STATE);
548 if (!sra) {
549 ret = 1;
550 goto free_st;
551 }
552
553 if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
554 !st->ss->write_init_ppl) {
555 pr_err("%s metadata does not support PPL\n", st->ss->name);
556 ret = 1;
557 goto free_info;
558 }
559
560 if (sra->array.level != 5) {
561 pr_err("Operation not supported for array level %d\n",
562 sra->array.level);
563 ret = 1;
564 goto free_info;
565 }
566
567 if (sra->consistency_policy == (unsigned)s->consistency_policy) {
568 pr_err("Consistency policy is already %s\n",
569 map_num(consistency_policies, s->consistency_policy));
570 ret = 1;
571 goto free_info;
572 } else if (sra->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
573 sra->consistency_policy != CONSISTENCY_POLICY_PPL) {
574 pr_err("Current consistency policy is %s, cannot change to %s\n",
575 map_num(consistency_policies, sra->consistency_policy),
576 map_num(consistency_policies, s->consistency_policy));
577 ret = 1;
578 goto free_info;
579 }
580
581 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
582 if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0) {
583 ret = 1;
584 goto free_info;
585 } else if (strcmp(buf, "reshape\n") == 0) {
586 pr_err("PPL cannot be enabled when reshape is in progress\n");
587 ret = 1;
588 goto free_info;
589 }
590 }
591
592 if (subarray) {
593 char *update;
594
595 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
596 update = "ppl";
597 else
598 update = "no-ppl";
599
600 sprintf(container_dev, "/dev/%s", st->container_devnm);
601
602 ret = Update_subarray(container_dev, subarray, update, NULL,
603 c->verbose);
604 if (ret)
605 goto free_info;
606 }
607
608 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
609 struct mdinfo info;
610
611 if (subarray) {
612 struct mdinfo *mdi;
613 int cfd;
614
615 cfd = open(container_dev, O_RDWR|O_EXCL);
616 if (cfd < 0) {
617 pr_err("Failed to open %s\n", container_dev);
618 ret = 1;
619 goto free_info;
620 }
621
622 ret = st->ss->load_container(st, cfd, st->container_devnm);
623 close(cfd);
624
625 if (ret) {
626 pr_err("Cannot read superblock for %s\n",
627 container_dev);
628 goto free_info;
629 }
630
631 mdi = st->ss->container_content(st, subarray);
632 info = *mdi;
633 free(mdi);
634 }
635
636 for (sd = sra->devs; sd; sd = sd->next) {
637 int dfd;
638 char *devpath;
639
640 devpath = map_dev(sd->disk.major, sd->disk.minor, 0);
641 dfd = dev_open(devpath, O_RDWR);
642 if (dfd < 0) {
643 pr_err("Failed to open %s\n", devpath);
644 ret = 1;
645 goto free_info;
646 }
647
648 if (!subarray) {
649 ret = st->ss->load_super(st, dfd, NULL);
650 if (ret) {
651 pr_err("Failed to load super-block.\n");
652 close(dfd);
653 goto free_info;
654 }
655
656 ret = st->ss->update_super(st, sra, "ppl", devname,
657 c->verbose, 0, NULL);
658 if (ret) {
659 close(dfd);
660 st->ss->free_super(st);
661 goto free_info;
662 }
663 st->ss->getinfo_super(st, &info, NULL);
664 }
665
666 ret |= sysfs_set_num(sra, sd, "ppl_sector", info.ppl_sector);
667 ret |= sysfs_set_num(sra, sd, "ppl_size", info.ppl_size);
668
669 if (ret) {
670 pr_err("Failed to set PPL attributes for %s\n",
671 sd->sys_name);
672 close(dfd);
673 st->ss->free_super(st);
674 goto free_info;
675 }
676
677 ret = st->ss->write_init_ppl(st, &info, dfd);
678 if (ret)
679 pr_err("Failed to write PPL\n");
680
681 close(dfd);
682
683 if (!subarray)
684 st->ss->free_super(st);
685
686 if (ret)
687 goto free_info;
688 }
689 }
690
691 ret = sysfs_set_str(sra, NULL, "consistency_policy",
692 map_num(consistency_policies,
693 s->consistency_policy));
694 if (ret)
695 pr_err("Failed to change array consistency policy\n");
696
697 free_info:
698 sysfs_free(sra);
699 free_st:
700 free(st);
701 free(subarray);
702
703 return ret;
704 }
705
706 /*
707 * When reshaping an array we might need to backup some data.
708 * This is written to all spares with a 'super_block' describing it.
709 * The superblock goes 4K from the end of the used space on the
710 * device.
711 * It if written after the backup is complete.
712 * It has the following structure.
713 */
714
715 static struct mdp_backup_super {
716 char magic[16]; /* md_backup_data-1 or -2 */
717 __u8 set_uuid[16];
718 __u64 mtime;
719 /* start/sizes in 512byte sectors */
720 __u64 devstart; /* address on backup device/file of data */
721 __u64 arraystart;
722 __u64 length;
723 __u32 sb_csum; /* csum of preceeding bytes. */
724 __u32 pad1;
725 __u64 devstart2; /* offset in to data of second section */
726 __u64 arraystart2;
727 __u64 length2;
728 __u32 sb_csum2; /* csum of preceeding bytes. */
729 __u8 pad[512-68-32];
730 } __attribute__((aligned(512))) bsb, bsb2;
731
732 static __u32 bsb_csum(char *buf, int len)
733 {
734 int i;
735 int csum = 0;
736 for (i = 0; i < len; i++)
737 csum = (csum<<3) + buf[0];
738 return __cpu_to_le32(csum);
739 }
740
741 static int check_idle(struct supertype *st)
742 {
743 /* Check that all member arrays for this container, or the
744 * container of this array, are idle
745 */
746 char *container = (st->container_devnm[0]
747 ? st->container_devnm : st->devnm);
748 struct mdstat_ent *ent, *e;
749 int is_idle = 1;
750
751 ent = mdstat_read(0, 0);
752 for (e = ent ; e; e = e->next) {
753 if (!is_container_member(e, container))
754 continue;
755 if (e->percent >= 0) {
756 is_idle = 0;
757 break;
758 }
759 }
760 free_mdstat(ent);
761 return is_idle;
762 }
763
764 static int freeze_container(struct supertype *st)
765 {
766 char *container = (st->container_devnm[0]
767 ? st->container_devnm : st->devnm);
768
769 if (!check_idle(st))
770 return -1;
771
772 if (block_monitor(container, 1)) {
773 pr_err("failed to freeze container\n");
774 return -2;
775 }
776
777 return 1;
778 }
779
780 static void unfreeze_container(struct supertype *st)
781 {
782 char *container = (st->container_devnm[0]
783 ? st->container_devnm : st->devnm);
784
785 unblock_monitor(container, 1);
786 }
787
788 static int freeze(struct supertype *st)
789 {
790 /* Try to freeze resync/rebuild on this array/container.
791 * Return -1 if the array is busy,
792 * return -2 container cannot be frozen,
793 * return 0 if this kernel doesn't support 'frozen'
794 * return 1 if it worked.
795 */
796 if (st->ss->external)
797 return freeze_container(st);
798 else {
799 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
800 int err;
801 char buf[20];
802
803 if (!sra)
804 return -1;
805 /* Need to clear any 'read-auto' status */
806 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
807 strncmp(buf, "read-auto", 9) == 0)
808 sysfs_set_str(sra, NULL, "array_state", "clean");
809
810 err = sysfs_freeze_array(sra);
811 sysfs_free(sra);
812 return err;
813 }
814 }
815
816 static void unfreeze(struct supertype *st)
817 {
818 if (st->ss->external)
819 return unfreeze_container(st);
820 else {
821 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
822 char buf[20];
823
824 if (sra &&
825 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 &&
826 strcmp(buf, "frozen\n") == 0)
827 sysfs_set_str(sra, NULL, "sync_action", "idle");
828 sysfs_free(sra);
829 }
830 }
831
832 static void wait_reshape(struct mdinfo *sra)
833 {
834 int fd = sysfs_get_fd(sra, NULL, "sync_action");
835 char action[20];
836
837 if (fd < 0)
838 return;
839
840 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
841 strncmp(action, "reshape", 7) == 0)
842 sysfs_wait(fd, NULL);
843 close(fd);
844 }
845
846 static int reshape_super(struct supertype *st, unsigned long long size,
847 int level, int layout, int chunksize, int raid_disks,
848 int delta_disks, char *backup_file, char *dev,
849 int direction, int verbose)
850 {
851 /* nothing extra to check in the native case */
852 if (!st->ss->external)
853 return 0;
854 if (!st->ss->reshape_super ||
855 !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", "reshape");
934 if (err)
935 sleep(1);
936 } while (err && errno == EBUSY && cnt-- > 0);
937 }
938 return err;
939 }
940
941 void abort_reshape(struct mdinfo *sra)
942 {
943 sysfs_set_str(sra, NULL, "sync_action", "idle");
944 /*
945 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
946 * suspend_hi to decrease as well as increase.")
947 * you could only increase suspend_{lo,hi} unless the region they
948 * covered was empty. So to reset to 0, you need to push suspend_lo
949 * up past suspend_hi first. So to maximize the chance of mdadm
950 * working on all kernels, we want to keep doing that.
951 */
952 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
953 sysfs_set_num(sra, NULL, "suspend_hi", 0);
954 sysfs_set_num(sra, NULL, "suspend_lo", 0);
955 sysfs_set_num(sra, NULL, "sync_min", 0);
956 // It isn't safe to reset sync_max as we aren't monitoring.
957 // Array really should be stopped at this point.
958 }
959
960 int remove_disks_for_takeover(struct supertype *st,
961 struct mdinfo *sra,
962 int layout)
963 {
964 int nr_of_copies;
965 struct mdinfo *remaining;
966 int slot;
967
968 if (st->ss->external) {
969 int rv = 0;
970 struct mdinfo *arrays = st->ss->container_content(st, NULL);
971 /*
972 * containter_content returns list of arrays in container
973 * If arrays->next is not NULL it means that there are
974 * 2 arrays in container and operation should be blocked
975 */
976 if (arrays) {
977 if (arrays->next)
978 rv = 1;
979 sysfs_free(arrays);
980 if (rv) {
981 pr_err("Error. Cannot perform operation on /dev/%s\n", st->devnm);
982 pr_err("For this operation it MUST be single array in container\n");
983 return rv;
984 }
985 }
986 }
987
988 if (sra->array.level == 10)
989 nr_of_copies = layout & 0xff;
990 else if (sra->array.level == 1)
991 nr_of_copies = sra->array.raid_disks;
992 else
993 return 1;
994
995 remaining = sra->devs;
996 sra->devs = NULL;
997 /* for each 'copy', select one device and remove from the list. */
998 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
999 struct mdinfo **diskp;
1000 int found = 0;
1001
1002 /* Find a working device to keep */
1003 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
1004 struct mdinfo *disk = *diskp;
1005
1006 if (disk->disk.raid_disk < slot)
1007 continue;
1008 if (disk->disk.raid_disk >= slot + nr_of_copies)
1009 continue;
1010 if (disk->disk.state & (1<<MD_DISK_REMOVED))
1011 continue;
1012 if (disk->disk.state & (1<<MD_DISK_FAULTY))
1013 continue;
1014 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
1015 continue;
1016
1017 /* We have found a good disk to use! */
1018 *diskp = disk->next;
1019 disk->next = sra->devs;
1020 sra->devs = disk;
1021 found = 1;
1022 break;
1023 }
1024 if (!found)
1025 break;
1026 }
1027
1028 if (slot < sra->array.raid_disks) {
1029 /* didn't find all slots */
1030 struct mdinfo **e;
1031 e = &remaining;
1032 while (*e)
1033 e = &(*e)->next;
1034 *e = sra->devs;
1035 sra->devs = remaining;
1036 return 1;
1037 }
1038
1039 /* Remove all 'remaining' devices from the array */
1040 while (remaining) {
1041 struct mdinfo *sd = remaining;
1042 remaining = sd->next;
1043
1044 sysfs_set_str(sra, sd, "state", "faulty");
1045 sysfs_set_str(sra, sd, "slot", "none");
1046 /* for external metadata disks should be removed in mdmon */
1047 if (!st->ss->external)
1048 sysfs_set_str(sra, sd, "state", "remove");
1049 sd->disk.state |= (1<<MD_DISK_REMOVED);
1050 sd->disk.state &= ~(1<<MD_DISK_SYNC);
1051 sd->next = sra->devs;
1052 sra->devs = sd;
1053 }
1054 return 0;
1055 }
1056
1057 void reshape_free_fdlist(int *fdlist,
1058 unsigned long long *offsets,
1059 int size)
1060 {
1061 int i;
1062
1063 for (i = 0; i < size; i++)
1064 if (fdlist[i] >= 0)
1065 close(fdlist[i]);
1066
1067 free(fdlist);
1068 free(offsets);
1069 }
1070
1071 int reshape_prepare_fdlist(char *devname,
1072 struct mdinfo *sra,
1073 int raid_disks,
1074 int nrdisks,
1075 unsigned long blocks,
1076 char *backup_file,
1077 int *fdlist,
1078 unsigned long long *offsets)
1079 {
1080 int d = 0;
1081 struct mdinfo *sd;
1082
1083 enable_fds(nrdisks);
1084 for (d = 0; d <= nrdisks; d++)
1085 fdlist[d] = -1;
1086 d = raid_disks;
1087 for (sd = sra->devs; sd; sd = sd->next) {
1088 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1089 continue;
1090 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
1091 sd->disk.raid_disk < raid_disks) {
1092 char *dn = map_dev(sd->disk.major,
1093 sd->disk.minor, 1);
1094 fdlist[sd->disk.raid_disk]
1095 = dev_open(dn, O_RDONLY);
1096 offsets[sd->disk.raid_disk] = sd->data_offset*512;
1097 if (fdlist[sd->disk.raid_disk] < 0) {
1098 pr_err("%s: cannot open component %s\n",
1099 devname, dn ? dn : "-unknown-");
1100 d = -1;
1101 goto release;
1102 }
1103 } else if (backup_file == NULL) {
1104 /* spare */
1105 char *dn = map_dev(sd->disk.major,
1106 sd->disk.minor, 1);
1107 fdlist[d] = dev_open(dn, O_RDWR);
1108 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
1109 if (fdlist[d] < 0) {
1110 pr_err("%s: cannot open component %s\n",
1111 devname, dn ? dn : "-unknown-");
1112 d = -1;
1113 goto release;
1114 }
1115 d++;
1116 }
1117 }
1118 release:
1119 return d;
1120 }
1121
1122 int reshape_open_backup_file(char *backup_file,
1123 int fd,
1124 char *devname,
1125 long blocks,
1126 int *fdlist,
1127 unsigned long long *offsets,
1128 char *sys_name,
1129 int restart)
1130 {
1131 /* Return 1 on success, 0 on any form of failure */
1132 /* need to check backup file is large enough */
1133 char buf[512];
1134 struct stat stb;
1135 unsigned int dev;
1136 int i;
1137
1138 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
1139 S_IRUSR | S_IWUSR);
1140 *offsets = 8 * 512;
1141 if (*fdlist < 0) {
1142 pr_err("%s: cannot create backup file %s: %s\n",
1143 devname, backup_file, strerror(errno));
1144 return 0;
1145 }
1146 /* Guard against backup file being on array device.
1147 * If array is partitioned or if LVM etc is in the
1148 * way this will not notice, but it is better than
1149 * nothing.
1150 */
1151 fstat(*fdlist, &stb);
1152 dev = stb.st_dev;
1153 fstat(fd, &stb);
1154 if (stb.st_rdev == dev) {
1155 pr_err("backup file must NOT be on the array being reshaped.\n");
1156 close(*fdlist);
1157 return 0;
1158 }
1159
1160 memset(buf, 0, 512);
1161 for (i=0; i < blocks + 8 ; i++) {
1162 if (write(*fdlist, buf, 512) != 512) {
1163 pr_err("%s: cannot create backup file %s: %s\n",
1164 devname, backup_file, strerror(errno));
1165 return 0;
1166 }
1167 }
1168 if (fsync(*fdlist) != 0) {
1169 pr_err("%s: cannot create backup file %s: %s\n",
1170 devname, backup_file, strerror(errno));
1171 return 0;
1172 }
1173
1174 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
1175 char *bu = make_backup(sys_name);
1176 if (symlink(backup_file, bu))
1177 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
1178 strerror(errno));
1179 free(bu);
1180 }
1181
1182 return 1;
1183 }
1184
1185 unsigned long compute_backup_blocks(int nchunk, int ochunk,
1186 unsigned int ndata, unsigned int odata)
1187 {
1188 unsigned long a, b, blocks;
1189 /* So how much do we need to backup.
1190 * We need an amount of data which is both a whole number of
1191 * old stripes and a whole number of new stripes.
1192 * So LCM for (chunksize*datadisks).
1193 */
1194 a = (ochunk/512) * odata;
1195 b = (nchunk/512) * ndata;
1196 /* Find GCD */
1197 a = GCD(a, b);
1198 /* LCM == product / GCD */
1199 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
1200
1201 return blocks;
1202 }
1203
1204 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
1205 {
1206 /* Based on the current array state in info->array and
1207 * the changes in info->new_* etc, determine:
1208 * - whether the change is possible
1209 * - Intermediate level/raid_disks/layout
1210 * - whether a restriping reshape is needed
1211 * - number of sectors in minimum change unit. This
1212 * will cover a whole number of stripes in 'before' and
1213 * 'after'.
1214 *
1215 * Return message if the change should be rejected
1216 * NULL if the change can be achieved
1217 *
1218 * This can be called as part of starting a reshape, or
1219 * when assembling an array that is undergoing reshape.
1220 */
1221 int near, far, offset, copies;
1222 int new_disks;
1223 int old_chunk, new_chunk;
1224 /* delta_parity records change in number of devices
1225 * caused by level change
1226 */
1227 int delta_parity = 0;
1228
1229 memset(re, 0, sizeof(*re));
1230
1231 /* If a new level not explicitly given, we assume no-change */
1232 if (info->new_level == UnSet)
1233 info->new_level = info->array.level;
1234
1235 if (info->new_chunk)
1236 switch (info->new_level) {
1237 case 0:
1238 case 4:
1239 case 5:
1240 case 6:
1241 case 10:
1242 /* chunk size is meaningful, must divide component_size
1243 * evenly
1244 */
1245 if (info->component_size % (info->new_chunk/512)) {
1246 unsigned long long shrink = info->component_size;
1247 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1248 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1249 info->new_chunk/1024, info->component_size/2);
1250 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1251 devname, shrink/2);
1252 pr_err("will shrink the array so the given chunk size would work.\n");
1253 return "";
1254 }
1255 break;
1256 default:
1257 return "chunk size not meaningful for this level";
1258 }
1259 else
1260 info->new_chunk = info->array.chunk_size;
1261
1262 switch (info->array.level) {
1263 default:
1264 return "No reshape is possibly for this RAID level";
1265 case LEVEL_LINEAR:
1266 if (info->delta_disks != UnSet)
1267 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1268 else
1269 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1270 case 1:
1271 /* RAID1 can convert to RAID1 with different disks, or
1272 * raid5 with 2 disks, or
1273 * raid0 with 1 disk
1274 */
1275 if (info->new_level > 1 && (info->component_size & 7))
1276 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1277 if (info->new_level == 0) {
1278 if (info->delta_disks != UnSet &&
1279 info->delta_disks != 0)
1280 return "Cannot change number of disks with RAID1->RAID0 conversion";
1281 re->level = 0;
1282 re->before.data_disks = 1;
1283 re->after.data_disks = 1;
1284 return NULL;
1285 }
1286 if (info->new_level == 1) {
1287 if (info->delta_disks == UnSet)
1288 /* Don't know what to do */
1289 return "no change requested for Growing RAID1";
1290 re->level = 1;
1291 return NULL;
1292 }
1293 if (info->array.raid_disks != 2 && info->new_level == 5)
1294 return "Can only convert a 2-device array to RAID5";
1295 if (info->array.raid_disks == 2 && info->new_level == 5) {
1296 re->level = 5;
1297 re->before.data_disks = 1;
1298 if (info->delta_disks != UnSet &&
1299 info->delta_disks != 0)
1300 re->after.data_disks = 1 + info->delta_disks;
1301 else
1302 re->after.data_disks = 1;
1303 if (re->after.data_disks < 1)
1304 return "Number of disks too small for RAID5";
1305
1306 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1307 info->array.chunk_size = 65536;
1308 break;
1309 }
1310 /* Could do some multi-stage conversions, but leave that to
1311 * later.
1312 */
1313 return "Impossibly level change request for RAID1";
1314
1315 case 10:
1316 /* RAID10 can be converted from near mode to
1317 * RAID0 by removing some devices.
1318 * It can also be reshaped if the kernel supports
1319 * new_data_offset.
1320 */
1321 switch (info->new_level) {
1322 case 0:
1323 if ((info->array.layout & ~0xff) != 0x100)
1324 return "Cannot Grow RAID10 with far/offset layout";
1325 /* number of devices must be multiple of number of copies */
1326 if (info->array.raid_disks % (info->array.layout & 0xff))
1327 return "RAID10 layout too complex for Grow operation";
1328
1329 new_disks = (info->array.raid_disks
1330 / (info->array.layout & 0xff));
1331 if (info->delta_disks == UnSet)
1332 info->delta_disks = (new_disks
1333 - info->array.raid_disks);
1334
1335 if (info->delta_disks != new_disks - info->array.raid_disks)
1336 return "New number of raid-devices impossible for RAID10";
1337 if (info->new_chunk &&
1338 info->new_chunk != info->array.chunk_size)
1339 return "Cannot change chunk-size with RAID10 Grow";
1340
1341 /* looks good */
1342 re->level = 0;
1343 re->before.data_disks = new_disks;
1344 re->after.data_disks = re->before.data_disks;
1345 return NULL;
1346
1347 case 10:
1348 near = info->array.layout & 0xff;
1349 far = (info->array.layout >> 8) & 0xff;
1350 offset = info->array.layout & 0x10000;
1351 if (far > 1 && !offset)
1352 return "Cannot reshape RAID10 in far-mode";
1353 copies = near * far;
1354
1355 old_chunk = info->array.chunk_size * far;
1356
1357 if (info->new_layout == UnSet)
1358 info->new_layout = info->array.layout;
1359 else {
1360 near = info->new_layout & 0xff;
1361 far = (info->new_layout >> 8) & 0xff;
1362 offset = info->new_layout & 0x10000;
1363 if (far > 1 && !offset)
1364 return "Cannot reshape RAID10 to far-mode";
1365 if (near * far != copies)
1366 return "Cannot change number of copies when reshaping RAID10";
1367 }
1368 if (info->delta_disks == UnSet)
1369 info->delta_disks = 0;
1370 new_disks = (info->array.raid_disks +
1371 info->delta_disks);
1372
1373 new_chunk = info->new_chunk * far;
1374
1375 re->level = 10;
1376 re->before.layout = info->array.layout;
1377 re->before.data_disks = info->array.raid_disks;
1378 re->after.layout = info->new_layout;
1379 re->after.data_disks = new_disks;
1380 /* For RAID10 we don't do backup but do allow reshape,
1381 * so set backup_blocks to INVALID_SECTORS rather than
1382 * zero.
1383 * And there is no need to synchronise stripes on both
1384 * 'old' and 'new'. So the important
1385 * number is the minimum data_offset difference
1386 * which is the larger of (offset copies * chunk).
1387 */
1388 re->backup_blocks = INVALID_SECTORS;
1389 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1390 if (new_disks < re->before.data_disks &&
1391 info->space_after < re->min_offset_change)
1392 /* Reduce component size by one chunk */
1393 re->new_size = (info->component_size -
1394 re->min_offset_change);
1395 else
1396 re->new_size = info->component_size;
1397 re->new_size = re->new_size * new_disks / copies;
1398 return NULL;
1399
1400 default:
1401 return "RAID10 can only be changed to RAID0";
1402 }
1403 case 0:
1404 /* RAID0 can be converted to RAID10, or to RAID456 */
1405 if (info->new_level == 10) {
1406 if (info->new_layout == UnSet &&
1407 info->delta_disks == UnSet) {
1408 /* Assume near=2 layout */
1409 info->new_layout = 0x102;
1410 info->delta_disks = info->array.raid_disks;
1411 }
1412 if (info->new_layout == UnSet) {
1413 int copies = 1 + (info->delta_disks
1414 / info->array.raid_disks);
1415 if (info->array.raid_disks * (copies-1) !=
1416 info->delta_disks)
1417 return "Impossible number of devices for RAID0->RAID10";
1418 info->new_layout = 0x100 + copies;
1419 }
1420 if (info->delta_disks == UnSet) {
1421 int copies = info->new_layout & 0xff;
1422 if (info->new_layout != 0x100 + copies)
1423 return "New layout impossible for RAID0->RAID10";;
1424 info->delta_disks = (copies - 1) *
1425 info->array.raid_disks;
1426 }
1427 if (info->new_chunk &&
1428 info->new_chunk != info->array.chunk_size)
1429 return "Cannot change chunk-size with RAID0->RAID10";
1430 /* looks good */
1431 re->level = 10;
1432 re->before.data_disks = (info->array.raid_disks +
1433 info->delta_disks);
1434 re->after.data_disks = re->before.data_disks;
1435 re->before.layout = info->new_layout;
1436 return NULL;
1437 }
1438
1439 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1440 * a raid4 style layout of the final level.
1441 */
1442 switch (info->new_level) {
1443 case 4:
1444 delta_parity = 1;
1445 case 0:
1446 re->level = 4;
1447 re->before.layout = 0;
1448 break;
1449 case 5:
1450 delta_parity = 1;
1451 re->level = 5;
1452 re->before.layout = ALGORITHM_PARITY_N;
1453 if (info->new_layout == UnSet)
1454 info->new_layout = map_name(r5layout, "default");
1455 break;
1456 case 6:
1457 delta_parity = 2;
1458 re->level = 6;
1459 re->before.layout = ALGORITHM_PARITY_N;
1460 if (info->new_layout == UnSet)
1461 info->new_layout = map_name(r6layout, "default");
1462 break;
1463 default:
1464 return "Impossible level change requested";
1465 }
1466 re->before.data_disks = info->array.raid_disks;
1467 /* determining 'after' layout happens outside this 'switch' */
1468 break;
1469
1470 case 4:
1471 info->array.layout = ALGORITHM_PARITY_N;
1472 case 5:
1473 switch (info->new_level) {
1474 case 0:
1475 delta_parity = -1;
1476 case 4:
1477 re->level = info->array.level;
1478 re->before.data_disks = info->array.raid_disks - 1;
1479 re->before.layout = info->array.layout;
1480 break;
1481 case 5:
1482 re->level = 5;
1483 re->before.data_disks = info->array.raid_disks - 1;
1484 re->before.layout = info->array.layout;
1485 break;
1486 case 6:
1487 delta_parity = 1;
1488 re->level = 6;
1489 re->before.data_disks = info->array.raid_disks - 1;
1490 switch (info->array.layout) {
1491 case ALGORITHM_LEFT_ASYMMETRIC:
1492 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1493 break;
1494 case ALGORITHM_RIGHT_ASYMMETRIC:
1495 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1496 break;
1497 case ALGORITHM_LEFT_SYMMETRIC:
1498 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1499 break;
1500 case ALGORITHM_RIGHT_SYMMETRIC:
1501 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1502 break;
1503 case ALGORITHM_PARITY_0:
1504 re->before.layout = ALGORITHM_PARITY_0_6;
1505 break;
1506 case ALGORITHM_PARITY_N:
1507 re->before.layout = ALGORITHM_PARITY_N_6;
1508 break;
1509 default:
1510 return "Cannot convert an array with this layout";
1511 }
1512 break;
1513 case 1:
1514 if (info->array.raid_disks != 2)
1515 return "Can only convert a 2-device array to RAID1";
1516 if (info->delta_disks != UnSet &&
1517 info->delta_disks != 0)
1518 return "Cannot set raid_disk when converting RAID5->RAID1";
1519 re->level = 1;
1520 info->new_chunk = 0;
1521 return NULL;
1522 default:
1523 return "Impossible level change requested";
1524 }
1525 break;
1526 case 6:
1527 switch (info->new_level) {
1528 case 4:
1529 case 5:
1530 delta_parity = -1;
1531 case 6:
1532 re->level = 6;
1533 re->before.data_disks = info->array.raid_disks - 2;
1534 re->before.layout = info->array.layout;
1535 break;
1536 default:
1537 return "Impossible level change requested";
1538 }
1539 break;
1540 }
1541
1542 /* If we reached here then it looks like a re-stripe is
1543 * happening. We have determined the intermediate level
1544 * and initial raid_disks/layout and stored these in 're'.
1545 *
1546 * We need to deduce the final layout that can be atomically
1547 * converted to the end state.
1548 */
1549 switch (info->new_level) {
1550 case 0:
1551 /* We can only get to RAID0 from RAID4 or RAID5
1552 * with appropriate layout and one extra device
1553 */
1554 if (re->level != 4 && re->level != 5)
1555 return "Cannot covert to RAID0 from this level";
1556
1557 switch (re->level) {
1558 case 4:
1559 re->before.layout = 0;
1560 re->after.layout = 0;
1561 break;
1562 case 5:
1563 re->after.layout = ALGORITHM_PARITY_N;
1564 break;
1565 }
1566 break;
1567
1568 case 4:
1569 /* We can only get to RAID4 from RAID5 */
1570 if (re->level != 4 && re->level != 5)
1571 return "Cannot convert to RAID4 from this level";
1572
1573 switch (re->level) {
1574 case 4:
1575 re->after.layout = 0;
1576 break;
1577 case 5:
1578 re->after.layout = ALGORITHM_PARITY_N;
1579 break;
1580 }
1581 break;
1582
1583 case 5:
1584 /* We get to RAID5 from RAID5 or RAID6 */
1585 if (re->level != 5 && re->level != 6)
1586 return "Cannot convert to RAID5 from this level";
1587
1588 switch (re->level) {
1589 case 5:
1590 if (info->new_layout == UnSet)
1591 re->after.layout = re->before.layout;
1592 else
1593 re->after.layout = info->new_layout;
1594 break;
1595 case 6:
1596 if (info->new_layout == UnSet)
1597 info->new_layout = re->before.layout;
1598
1599 /* after.layout needs to be raid6 version of new_layout */
1600 if (info->new_layout == ALGORITHM_PARITY_N)
1601 re->after.layout = ALGORITHM_PARITY_N;
1602 else {
1603 char layout[40];
1604 char *ls = map_num(r5layout, info->new_layout);
1605 int l;
1606 if (ls) {
1607 /* Current RAID6 layout has a RAID5
1608 * equivalent - good
1609 */
1610 strcat(strcpy(layout, ls), "-6");
1611 l = map_name(r6layout, layout);
1612 if (l == UnSet)
1613 return "Cannot find RAID6 layout to convert to";
1614 } else {
1615 /* Current RAID6 has no equivalent.
1616 * If it is already a '-6' layout we
1617 * can leave it unchanged, else we must
1618 * fail
1619 */
1620 ls = map_num(r6layout, info->new_layout);
1621 if (!ls ||
1622 strcmp(ls+strlen(ls)-2, "-6") != 0)
1623 return "Please specify new layout";
1624 l = info->new_layout;
1625 }
1626 re->after.layout = l;
1627 }
1628 }
1629 break;
1630
1631 case 6:
1632 /* We must already be at level 6 */
1633 if (re->level != 6)
1634 return "Impossible level change";
1635 if (info->new_layout == UnSet)
1636 re->after.layout = info->array.layout;
1637 else
1638 re->after.layout = info->new_layout;
1639 break;
1640 default:
1641 return "Impossible level change requested";
1642 }
1643 if (info->delta_disks == UnSet)
1644 info->delta_disks = delta_parity;
1645
1646 re->after.data_disks =
1647 (re->before.data_disks + info->delta_disks - delta_parity);
1648
1649 switch (re->level) {
1650 case 6:
1651 re->parity = 2;
1652 break;
1653 case 4:
1654 case 5:
1655 re->parity = 1;
1656 break;
1657 default:
1658 re->parity = 0;
1659 break;
1660 }
1661 /* So we have a restripe operation, we need to calculate the number
1662 * of blocks per reshape operation.
1663 */
1664 re->new_size = info->component_size * re->before.data_disks;
1665 if (info->new_chunk == 0)
1666 info->new_chunk = info->array.chunk_size;
1667 if (re->after.data_disks == re->before.data_disks &&
1668 re->after.layout == re->before.layout &&
1669 info->new_chunk == info->array.chunk_size) {
1670 /* Nothing to change, can change level immediately. */
1671 re->level = info->new_level;
1672 re->backup_blocks = 0;
1673 return NULL;
1674 }
1675 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1676 /* chunk and layout changes make no difference */
1677 re->level = info->new_level;
1678 re->backup_blocks = 0;
1679 return NULL;
1680 }
1681
1682 if (re->after.data_disks == re->before.data_disks &&
1683 get_linux_version() < 2006032)
1684 return "in-place reshape is not safe before 2.6.32 - sorry.";
1685
1686 if (re->after.data_disks < re->before.data_disks &&
1687 get_linux_version() < 2006030)
1688 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1689
1690 re->backup_blocks = compute_backup_blocks(
1691 info->new_chunk, info->array.chunk_size,
1692 re->after.data_disks,
1693 re->before.data_disks);
1694 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1695
1696 re->new_size = info->component_size * re->after.data_disks;
1697 return NULL;
1698 }
1699
1700 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1701 char *text_version)
1702 {
1703 struct mdinfo *info;
1704 char *subarray;
1705 int ret_val = -1;
1706
1707 if ((st == NULL) || (sra == NULL))
1708 return ret_val;
1709
1710 if (text_version == NULL)
1711 text_version = sra->text_version;
1712 subarray = strchr(text_version + 1, '/')+1;
1713 info = st->ss->container_content(st, subarray);
1714 if (info) {
1715 unsigned long long current_size = 0;
1716 unsigned long long new_size =
1717 info->custom_array_size/2;
1718
1719 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1720 new_size > current_size) {
1721 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1722 < 0)
1723 dprintf("Error: Cannot set array size");
1724 else {
1725 ret_val = 0;
1726 dprintf("Array size changed");
1727 }
1728 dprintf_cont(" from %llu to %llu.\n",
1729 current_size, new_size);
1730 }
1731 sysfs_free(info);
1732 } else
1733 dprintf("Error: set_array_size(): info pointer in NULL\n");
1734
1735 return ret_val;
1736 }
1737
1738 static int reshape_array(char *container, int fd, char *devname,
1739 struct supertype *st, struct mdinfo *info,
1740 int force, struct mddev_dev *devlist,
1741 unsigned long long data_offset,
1742 char *backup_file, int verbose, int forked,
1743 int restart, int freeze_reshape);
1744 static int reshape_container(char *container, char *devname,
1745 int mdfd,
1746 struct supertype *st,
1747 struct mdinfo *info,
1748 int force,
1749 char *backup_file, int verbose,
1750 int forked, int restart, int freeze_reshape);
1751
1752 int Grow_reshape(char *devname, int fd,
1753 struct mddev_dev *devlist,
1754 unsigned long long data_offset,
1755 struct context *c, struct shape *s)
1756 {
1757 /* Make some changes in the shape of an array.
1758 * The kernel must support the change.
1759 *
1760 * There are three different changes. Each can trigger
1761 * a resync or recovery so we freeze that until we have
1762 * requested everything (if kernel supports freezing - 2.6.30).
1763 * The steps are:
1764 * - change size (i.e. component_size)
1765 * - change level
1766 * - change layout/chunksize/ndisks
1767 *
1768 * The last can require a reshape. It is different on different
1769 * levels so we need to check the level before actioning it.
1770 * Some times the level change needs to be requested after the
1771 * reshape (e.g. raid6->raid5, raid5->raid0)
1772 *
1773 */
1774 struct mdu_array_info_s array;
1775 int rv = 0;
1776 struct supertype *st;
1777 char *subarray = NULL;
1778
1779 int frozen;
1780 int changed = 0;
1781 char *container = NULL;
1782 int cfd = -1;
1783
1784 struct mddev_dev *dv;
1785 int added_disks;
1786
1787 struct mdinfo info;
1788 struct mdinfo *sra;
1789
1790 if (md_get_array_info(fd, &array) < 0) {
1791 pr_err("%s is not an active md array - aborting\n",
1792 devname);
1793 return 1;
1794 }
1795 if (data_offset != INVALID_SECTORS && array.level != 10 &&
1796 (array.level < 4 || array.level > 6)) {
1797 pr_err("--grow --data-offset not yet supported\n");
1798 return 1;
1799 }
1800
1801 if (s->size > 0 &&
1802 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1803 pr_err("cannot change component size at the same time as other changes.\n"
1804 " Change size first, then check data is intact before making other changes.\n");
1805 return 1;
1806 }
1807
1808 if (s->raiddisks && s->raiddisks < array.raid_disks &&
1809 array.level > 1 && get_linux_version() < 2006032 &&
1810 !check_env("MDADM_FORCE_FEWER")) {
1811 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1812 " Please use a newer kernel\n");
1813 return 1;
1814 }
1815
1816 if (array.level > 1 &&
1817 (array.chunk_size / 1024) > (int)s->size) {
1818 pr_err("component size must be larger than chunk size.\n");
1819 return 1;
1820 }
1821
1822 st = super_by_fd(fd, &subarray);
1823 if (!st) {
1824 pr_err("Unable to determine metadata format for %s\n", devname);
1825 return 1;
1826 }
1827 if (s->raiddisks > st->max_devs) {
1828 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1829 return 1;
1830 }
1831 if (s->level == 0 &&
1832 (array.state & (1<<MD_SB_BITMAP_PRESENT)) &&
1833 !(array.state & (1<<MD_SB_CLUSTERED))) {
1834 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
1835 if (md_set_array_info(fd, &array)!= 0) {
1836 pr_err("failed to remove internal bitmap.\n");
1837 return 1;
1838 }
1839 }
1840
1841 /* in the external case we need to check that the requested reshape is
1842 * supported, and perform an initial check that the container holds the
1843 * pre-requisite spare devices (mdmon owns final validation)
1844 */
1845 if (st->ss->external) {
1846 int retval;
1847
1848 if (subarray) {
1849 container = st->container_devnm;
1850 cfd = open_dev_excl(st->container_devnm);
1851 } else {
1852 container = st->devnm;
1853 close(fd);
1854 cfd = open_dev_excl(st->devnm);
1855 fd = cfd;
1856 }
1857 if (cfd < 0) {
1858 pr_err("Unable to open container for %s\n",
1859 devname);
1860 free(subarray);
1861 return 1;
1862 }
1863
1864 retval = st->ss->load_container(st, cfd, NULL);
1865
1866 if (retval) {
1867 pr_err("Cannot read superblock for %s\n",
1868 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 < (s->raiddisks - array.raid_disks) &&
1917 !c->force) {
1918 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1919 " Use --force to over-ride this check.\n",
1920 s->raiddisks - array.raid_disks,
1921 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1922 array.spare_disks + added_disks);
1923 return 1;
1924 }
1925
1926 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS |
1927 GET_STATE | GET_VERSION);
1928 if (sra) {
1929 if (st->ss->external && subarray == NULL) {
1930 array.level = LEVEL_CONTAINER;
1931 sra->array.level = LEVEL_CONTAINER;
1932 }
1933 } else {
1934 pr_err("failed to read sysfs parameters for %s\n",
1935 devname);
1936 return 1;
1937 }
1938 frozen = freeze(st);
1939 if (frozen < -1) {
1940 /* freeze() already spewed the reason */
1941 sysfs_free(sra);
1942 return 1;
1943 } else if (frozen < 0) {
1944 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1945 sysfs_free(sra);
1946 return 1;
1947 }
1948
1949 /* ========= set size =============== */
1950 if (s->size > 0 &&
1951 (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1952 unsigned long long orig_size = get_component_size(fd)/2;
1953 unsigned long long min_csize;
1954 struct mdinfo *mdi;
1955 int raid0_takeover = 0;
1956
1957 if (orig_size == 0)
1958 orig_size = (unsigned) array.size;
1959
1960 if (orig_size == 0) {
1961 pr_err("Cannot set device size in this type of array.\n");
1962 rv = 1;
1963 goto release;
1964 }
1965
1966 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1967 devname, APPLY_METADATA_CHANGES,
1968 c->verbose > 0)) {
1969 rv = 1;
1970 goto release;
1971 }
1972 sync_metadata(st);
1973 if (st->ss->external) {
1974 /* metadata can have size limitation
1975 * update size value according to metadata information
1976 */
1977 struct mdinfo *sizeinfo =
1978 st->ss->container_content(st, subarray);
1979 if (sizeinfo) {
1980 unsigned long long new_size =
1981 sizeinfo->custom_array_size/2;
1982 int data_disks = get_data_disks(
1983 sizeinfo->array.level,
1984 sizeinfo->array.layout,
1985 sizeinfo->array.raid_disks);
1986 new_size /= data_disks;
1987 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1988 orig_size, new_size,
1989 new_size * data_disks);
1990 s->size = new_size;
1991 sysfs_free(sizeinfo);
1992 }
1993 }
1994
1995 /* Update the size of each member device in case
1996 * they have been resized. This will never reduce
1997 * below the current used-size. The "size" attribute
1998 * understands '0' to mean 'max'.
1999 */
2000 min_csize = 0;
2001 for (mdi = sra->devs; mdi; mdi = mdi->next) {
2002 sysfs_set_num(sra, mdi, "size", s->size == MAX_SIZE ? 0
2003 : s->size);
2004 if (array.not_persistent == 0 &&
2005 array.major_version == 0 &&
2006 get_linux_version() < 3001000) {
2007 /* Dangerous to allow size to exceed 2TB */
2008 unsigned long long csize;
2009 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
2010 if (csize >= 2ULL*1024*1024*1024)
2011 csize = 2ULL*1024*1024*1024;
2012 if ((min_csize == 0 || (min_csize
2013 > csize)))
2014 min_csize = csize;
2015 }
2016 }
2017 }
2018 if (min_csize && s->size > min_csize) {
2019 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2020 rv = 1;
2021 goto size_change_error;
2022 }
2023 if (min_csize && s->size == MAX_SIZE) {
2024 /* Don't let the kernel choose a size - it will get
2025 * it wrong
2026 */
2027 pr_err("Limited v0.90 array to 2TB per device\n");
2028 s->size = min_csize;
2029 }
2030 if (st->ss->external) {
2031 if (sra->array.level == 0) {
2032 rv = sysfs_set_str(sra, NULL, "level",
2033 "raid5");
2034 if (!rv) {
2035 raid0_takeover = 1;
2036 /* get array parameters after takeover
2037 * to change one parameter at time only
2038 */
2039 rv = md_get_array_info(fd, &array);
2040 }
2041 }
2042 /* make sure mdmon is
2043 * aware of the new level */
2044 if (!mdmon_running(st->container_devnm))
2045 start_mdmon(st->container_devnm);
2046 ping_monitor(container);
2047 if (mdmon_running(st->container_devnm) &&
2048 st->update_tail == NULL)
2049 st->update_tail = &st->updates;
2050 }
2051
2052 if (s->size == MAX_SIZE)
2053 s->size = 0;
2054 array.size = s->size;
2055 if (s->size & ~INT32_MAX) {
2056 /* got truncated to 32bit, write to
2057 * component_size instead
2058 */
2059 if (sra)
2060 rv = sysfs_set_num(sra, NULL,
2061 "component_size", s->size);
2062 else
2063 rv = -1;
2064 } else {
2065 rv = md_set_array_info(fd, &array);
2066
2067 /* manage array size when it is managed externally
2068 */
2069 if ((rv == 0) && st->ss->external)
2070 rv = set_array_size(st, sra, sra->text_version);
2071 }
2072
2073 if (raid0_takeover) {
2074 /* do not recync non-existing parity,
2075 * we will drop it anyway
2076 */
2077 sysfs_set_str(sra, NULL, "sync_action", "frozen");
2078 /* go back to raid0, drop parity disk
2079 */
2080 sysfs_set_str(sra, NULL, "level", "raid0");
2081 md_get_array_info(fd, &array);
2082 }
2083
2084 size_change_error:
2085 if (rv != 0) {
2086 int err = errno;
2087
2088 /* restore metadata */
2089 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
2090 UnSet, NULL, devname,
2091 ROLLBACK_METADATA_CHANGES,
2092 c->verbose) == 0)
2093 sync_metadata(st);
2094 pr_err("Cannot set device size for %s: %s\n",
2095 devname, strerror(err));
2096 if (err == EBUSY &&
2097 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2098 cont_err("Bitmap must be removed before size can be changed\n");
2099 rv = 1;
2100 goto release;
2101 }
2102 if (s->assume_clean) {
2103 /* This will fail on kernels older than 3.0 unless
2104 * a backport has been arranged.
2105 */
2106 if (sra == NULL ||
2107 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
2108 pr_err("--assume-clean not supported with --grow on this kernel\n");
2109 }
2110 md_get_array_info(fd, &array);
2111 s->size = get_component_size(fd)/2;
2112 if (s->size == 0)
2113 s->size = array.size;
2114 if (c->verbose >= 0) {
2115 if (s->size == orig_size)
2116 pr_err("component size of %s unchanged at %lluK\n",
2117 devname, s->size);
2118 else
2119 pr_err("component size of %s has been set to %lluK\n",
2120 devname, s->size);
2121 }
2122 changed = 1;
2123 } else if (array.level != LEVEL_CONTAINER) {
2124 s->size = get_component_size(fd)/2;
2125 if (s->size == 0)
2126 s->size = array.size;
2127 }
2128
2129 /* See if there is anything else to do */
2130 if ((s->level == UnSet || s->level == array.level) &&
2131 (s->layout_str == NULL) &&
2132 (s->chunk == 0 || s->chunk == array.chunk_size) &&
2133 data_offset == INVALID_SECTORS &&
2134 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
2135 /* Nothing more to do */
2136 if (!changed && c->verbose >= 0)
2137 pr_err("%s: no change requested\n",
2138 devname);
2139 goto release;
2140 }
2141
2142 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2143 * current implementation assumes that following conditions must be met:
2144 * - RAID10:
2145 * - far_copies == 1
2146 * - near_copies == 2
2147 */
2148 if ((s->level == 0 && array.level == 10 && sra &&
2149 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
2150 (s->level == 0 && array.level == 1 && sra)) {
2151 int err;
2152 err = remove_disks_for_takeover(st, sra, array.layout);
2153 if (err) {
2154 dprintf("Array cannot be reshaped\n");
2155 if (cfd > -1)
2156 close(cfd);
2157 rv = 1;
2158 goto release;
2159 }
2160 /* Make sure mdmon has seen the device removal
2161 * and updated metadata before we continue with
2162 * level change
2163 */
2164 if (container)
2165 ping_monitor(container);
2166 }
2167
2168 memset(&info, 0, sizeof(info));
2169 info.array = array;
2170 if (sysfs_init(&info, fd, NULL)) {
2171 pr_err("failed to intialize sysfs.\n");
2172 rv = 1;
2173 goto release;
2174 }
2175 strcpy(info.text_version, sra->text_version);
2176 info.component_size = s->size*2;
2177 info.new_level = s->level;
2178 info.new_chunk = s->chunk * 1024;
2179 if (info.array.level == LEVEL_CONTAINER) {
2180 info.delta_disks = UnSet;
2181 info.array.raid_disks = s->raiddisks;
2182 } else if (s->raiddisks)
2183 info.delta_disks = s->raiddisks - info.array.raid_disks;
2184 else
2185 info.delta_disks = UnSet;
2186 if (s->layout_str == NULL) {
2187 info.new_layout = UnSet;
2188 if (info.array.level == 6 &&
2189 (info.new_level == 6 || info.new_level == UnSet) &&
2190 info.array.layout >= 16) {
2191 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
2192 cont_err("during the reshape, please specify --layout=preserve\n");
2193 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2194 rv = 1;
2195 goto release;
2196 }
2197 } else if (strcmp(s->layout_str, "normalise") == 0 ||
2198 strcmp(s->layout_str, "normalize") == 0) {
2199 /* If we have a -6 RAID6 layout, remove the '-6'. */
2200 info.new_layout = UnSet;
2201 if (info.array.level == 6 && info.new_level == UnSet) {
2202 char l[40], *h;
2203 strcpy(l, map_num(r6layout, info.array.layout));
2204 h = strrchr(l, '-');
2205 if (h && strcmp(h, "-6") == 0) {
2206 *h = 0;
2207 info.new_layout = map_name(r6layout, l);
2208 }
2209 } else {
2210 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
2211 rv = 1;
2212 goto release;
2213 }
2214 } else if (strcmp(s->layout_str, "preserve") == 0) {
2215 /* This means that a non-standard RAID6 layout
2216 * is OK.
2217 * In particular:
2218 * - When reshape a RAID6 (e.g. adding a device)
2219 * which is in a non-standard layout, it is OK
2220 * to preserve that layout.
2221 * - When converting a RAID5 to RAID6, leave it in
2222 * the XXX-6 layout, don't re-layout.
2223 */
2224 if (info.array.level == 6 && info.new_level == UnSet)
2225 info.new_layout = info.array.layout;
2226 else if (info.array.level == 5 && info.new_level == 6) {
2227 char l[40];
2228 strcpy(l, map_num(r5layout, info.array.layout));
2229 strcat(l, "-6");
2230 info.new_layout = map_name(r6layout, l);
2231 } else {
2232 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
2233 rv = 1;
2234 goto release;
2235 }
2236 } else {
2237 int l = info.new_level;
2238 if (l == UnSet)
2239 l = info.array.level;
2240 switch (l) {
2241 case 5:
2242 info.new_layout = map_name(r5layout, s->layout_str);
2243 break;
2244 case 6:
2245 info.new_layout = map_name(r6layout, s->layout_str);
2246 break;
2247 case 10:
2248 info.new_layout = parse_layout_10(s->layout_str);
2249 break;
2250 case LEVEL_FAULTY:
2251 info.new_layout = parse_layout_faulty(s->layout_str);
2252 break;
2253 default:
2254 pr_err("layout not meaningful with this level\n");
2255 rv = 1;
2256 goto release;
2257 }
2258 if (info.new_layout == UnSet) {
2259 pr_err("layout %s not understood for this level\n",
2260 s->layout_str);
2261 rv = 1;
2262 goto release;
2263 }
2264 }
2265
2266 if (array.level == LEVEL_FAULTY) {
2267 if (s->level != UnSet && s->level != array.level) {
2268 pr_err("cannot change level of Faulty device\n");
2269 rv =1 ;
2270 }
2271 if (s->chunk) {
2272 pr_err("cannot set chunksize of Faulty device\n");
2273 rv =1 ;
2274 }
2275 if (s->raiddisks && s->raiddisks != 1) {
2276 pr_err("cannot set raid_disks of Faulty device\n");
2277 rv =1 ;
2278 }
2279 if (s->layout_str) {
2280 if (md_get_array_info(fd, &array) != 0) {
2281 dprintf("Cannot get array information.\n");
2282 goto release;
2283 }
2284 array.layout = info.new_layout;
2285 if (md_set_array_info(fd, &array) != 0) {
2286 pr_err("failed to set new layout\n");
2287 rv = 1;
2288 } else if (c->verbose >= 0)
2289 printf("layout for %s set to %d\n",
2290 devname, array.layout);
2291 }
2292 } else if (array.level == LEVEL_CONTAINER) {
2293 /* This change is to be applied to every array in the
2294 * container. This is only needed when the metadata imposes
2295 * restraints of the various arrays in the container.
2296 * Currently we only know that IMSM requires all arrays
2297 * to have the same number of devices so changing the
2298 * number of devices (On-Line Capacity Expansion) must be
2299 * performed at the level of the container
2300 */
2301 if (fd > 0) {
2302 close(fd);
2303 fd = -1;
2304 }
2305 rv = reshape_container(container, devname, -1, st, &info,
2306 c->force, c->backup_file, c->verbose, 0, 0, 0);
2307 frozen = 0;
2308 } else {
2309 /* get spare devices from external metadata
2310 */
2311 if (st->ss->external) {
2312 struct mdinfo *info2;
2313
2314 info2 = st->ss->container_content(st, subarray);
2315 if (info2) {
2316 info.array.spare_disks =
2317 info2->array.spare_disks;
2318 sysfs_free(info2);
2319 }
2320 }
2321
2322 /* Impose these changes on a single array. First
2323 * check that the metadata is OK with the change. */
2324
2325 if (reshape_super(st, 0, info.new_level,
2326 info.new_layout, info.new_chunk,
2327 info.array.raid_disks, info.delta_disks,
2328 c->backup_file, devname, APPLY_METADATA_CHANGES,
2329 c->verbose)) {
2330 rv = 1;
2331 goto release;
2332 }
2333 sync_metadata(st);
2334 rv = reshape_array(container, fd, devname, st, &info, c->force,
2335 devlist, data_offset, c->backup_file, c->verbose,
2336 0, 0, 0);
2337 frozen = 0;
2338 }
2339 release:
2340 sysfs_free(sra);
2341 if (frozen > 0)
2342 unfreeze(st);
2343 return rv;
2344 }
2345
2346 /* verify_reshape_position()
2347 * Function checks if reshape position in metadata is not farther
2348 * than position in md.
2349 * Return value:
2350 * 0 : not valid sysfs entry
2351 * it can be caused by not started reshape, it should be started
2352 * by reshape array or raid0 array is before takeover
2353 * -1 : error, reshape position is obviously wrong
2354 * 1 : success, reshape progress correct or updated
2355 */
2356 static int verify_reshape_position(struct mdinfo *info, int level)
2357 {
2358 int ret_val = 0;
2359 char buf[40];
2360 int rv;
2361
2362 /* read sync_max, failure can mean raid0 array */
2363 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2364
2365 if (rv > 0) {
2366 char *ep;
2367 unsigned long long position = strtoull(buf, &ep, 0);
2368
2369 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2370 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2371 position *= get_data_disks(level,
2372 info->new_layout,
2373 info->array.raid_disks);
2374 if (info->reshape_progress < position) {
2375 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2376 info->reshape_progress, position);
2377 info->reshape_progress = position;
2378 ret_val = 1;
2379 } else if (info->reshape_progress > position) {
2380 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2381 position, info->reshape_progress);
2382 ret_val = -1;
2383 } else {
2384 dprintf("Reshape position in md and metadata are the same;");
2385 ret_val = 1;
2386 }
2387 }
2388 } else if (rv == 0) {
2389 /* for valid sysfs entry, 0-length content
2390 * should be indicated as error
2391 */
2392 ret_val = -1;
2393 }
2394
2395 return ret_val;
2396 }
2397
2398 static unsigned long long choose_offset(unsigned long long lo,
2399 unsigned long long hi,
2400 unsigned long long min,
2401 unsigned long long max)
2402 {
2403 /* Choose a new offset between hi and lo.
2404 * It must be between min and max, but
2405 * we would prefer something near the middle of hi/lo, and also
2406 * prefer to be aligned to a big power of 2.
2407 *
2408 * So we start with the middle, then for each bit,
2409 * starting at '1' and increasing, if it is set, we either
2410 * add it or subtract it if possible, preferring the option
2411 * which is furthest from the boundary.
2412 *
2413 * We stop once we get a 1MB alignment. As units are in sectors,
2414 * 1MB = 2*1024 sectors.
2415 */
2416 unsigned long long choice = (lo + hi) / 2;
2417 unsigned long long bit = 1;
2418
2419 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2420 unsigned long long bigger, smaller;
2421 if (! (bit & choice))
2422 continue;
2423 bigger = choice + bit;
2424 smaller = choice - bit;
2425 if (bigger > max && smaller < min)
2426 break;
2427 if (bigger > max)
2428 choice = smaller;
2429 else if (smaller < min)
2430 choice = bigger;
2431 else if (hi - bigger > smaller - lo)
2432 choice = bigger;
2433 else
2434 choice = smaller;
2435 }
2436 return choice;
2437 }
2438
2439 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2440 char *devname, int delta_disks,
2441 unsigned long long data_offset,
2442 unsigned long long min,
2443 int can_fallback)
2444 {
2445 struct mdinfo *sd;
2446 int dir = 0;
2447 int err = 0;
2448 unsigned long long before, after;
2449
2450 /* Need to find min space before and after so same is used
2451 * on all devices
2452 */
2453 before = UINT64_MAX;
2454 after = UINT64_MAX;
2455 for (sd = sra->devs; sd; sd = sd->next) {
2456 char *dn;
2457 int dfd;
2458 int rv;
2459 struct supertype *st2;
2460 struct mdinfo info2;
2461
2462 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2463 continue;
2464 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2465 dfd = dev_open(dn, O_RDONLY);
2466 if (dfd < 0) {
2467 pr_err("%s: cannot open component %s\n",
2468 devname, dn ? dn : "-unknown-");
2469 goto release;
2470 }
2471 st2 = dup_super(st);
2472 rv = st2->ss->load_super(st2,dfd, NULL);
2473 close(dfd);
2474 if (rv) {
2475 free(st2);
2476 pr_err("%s: cannot get superblock from %s\n",
2477 devname, dn);
2478 goto release;
2479 }
2480 st2->ss->getinfo_super(st2, &info2, NULL);
2481 st2->ss->free_super(st2);
2482 free(st2);
2483 if (info2.space_before == 0 &&
2484 info2.space_after == 0) {
2485 /* Metadata doesn't support data_offset changes */
2486 if (!can_fallback)
2487 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2488 devname);
2489 goto fallback;
2490 }
2491 if (before > info2.space_before)
2492 before = info2.space_before;
2493 if (after > info2.space_after)
2494 after = info2.space_after;
2495
2496 if (data_offset != INVALID_SECTORS) {
2497 if (dir == 0) {
2498 if (info2.data_offset == data_offset) {
2499 pr_err("%s: already has that data_offset\n",
2500 dn);
2501 goto release;
2502 }
2503 if (data_offset < info2.data_offset)
2504 dir = -1;
2505 else
2506 dir = 1;
2507 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2508 (data_offset >= info2.data_offset && dir == -1)) {
2509 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2510 dn);
2511 goto release;
2512 }
2513 }
2514 }
2515 if (before == UINT64_MAX)
2516 /* impossible really, there must be no devices */
2517 return 1;
2518
2519 for (sd = sra->devs; sd; sd = sd->next) {
2520 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2521 unsigned long long new_data_offset;
2522
2523 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2524 continue;
2525 if (delta_disks < 0) {
2526 /* Don't need any space as array is shrinking
2527 * just move data_offset up by min
2528 */
2529 if (data_offset == INVALID_SECTORS)
2530 new_data_offset = sd->data_offset + min;
2531 else {
2532 if (data_offset < sd->data_offset + min) {
2533 pr_err("--data-offset too small for %s\n",
2534 dn);
2535 goto release;
2536 }
2537 new_data_offset = data_offset;
2538 }
2539 } else if (delta_disks > 0) {
2540 /* need space before */
2541 if (before < min) {
2542 if (can_fallback)
2543 goto fallback;
2544 pr_err("Insufficient head-space for reshape on %s\n",
2545 dn);
2546 goto release;
2547 }
2548 if (data_offset == INVALID_SECTORS)
2549 new_data_offset = sd->data_offset - min;
2550 else {
2551 if (data_offset > sd->data_offset - min) {
2552 pr_err("--data-offset too large for %s\n",
2553 dn);
2554 goto release;
2555 }
2556 new_data_offset = data_offset;
2557 }
2558 } else {
2559 if (dir == 0) {
2560 /* can move up or down. If 'data_offset'
2561 * was set we would have already decided,
2562 * so just choose direction with most space.
2563 */
2564 if (before > after)
2565 dir = -1;
2566 else
2567 dir = 1;
2568 }
2569 sysfs_set_str(sra, NULL, "reshape_direction",
2570 dir == 1 ? "backwards" : "forwards");
2571 if (dir > 0) {
2572 /* Increase data offset */
2573 if (after < min) {
2574 if (can_fallback)
2575 goto fallback;
2576 pr_err("Insufficient tail-space for reshape on %s\n",
2577 dn);
2578 goto release;
2579 }
2580 if (data_offset != INVALID_SECTORS &&
2581 data_offset < sd->data_offset + min) {
2582 pr_err("--data-offset too small on %s\n",
2583 dn);
2584 goto release;
2585 }
2586 if (data_offset != INVALID_SECTORS)
2587 new_data_offset = data_offset;
2588 else
2589 new_data_offset = choose_offset(sd->data_offset,
2590 sd->data_offset + after,
2591 sd->data_offset + min,
2592 sd->data_offset + after);
2593 } else {
2594 /* Decrease data offset */
2595 if (before < min) {
2596 if (can_fallback)
2597 goto fallback;
2598 pr_err("insufficient head-room on %s\n",
2599 dn);
2600 goto release;
2601 }
2602 if (data_offset != INVALID_SECTORS &&
2603 data_offset < sd->data_offset - min) {
2604 pr_err("--data-offset too small on %s\n",
2605 dn);
2606 goto release;
2607 }
2608 if (data_offset != INVALID_SECTORS)
2609 new_data_offset = data_offset;
2610 else
2611 new_data_offset = choose_offset(sd->data_offset - before,
2612 sd->data_offset,
2613 sd->data_offset - before,
2614 sd->data_offset - min);
2615 }
2616 }
2617 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2618 if (err < 0 && errno == E2BIG) {
2619 /* try again after increasing data size to max */
2620 err = sysfs_set_num(sra, sd, "size", 0);
2621 if (err < 0 && errno == EINVAL &&
2622 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2623 /* some kernels have a bug where you cannot
2624 * use '0' on spare devices. */
2625 sysfs_set_num(sra, sd, "size",
2626 (sra->component_size + after)/2);
2627 }
2628 err = sysfs_set_num(sra, sd, "new_offset",
2629 new_data_offset);
2630 }
2631 if (err < 0) {
2632 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2633 pr_err("data-offset is too big for %s\n",
2634 dn);
2635 goto release;
2636 }
2637 if (sd == sra->devs &&
2638 (errno == ENOENT || errno == E2BIG))
2639 /* Early kernel, no 'new_offset' file,
2640 * or kernel doesn't like us.
2641 * For RAID5/6 this is not fatal
2642 */
2643 return 1;
2644 pr_err("Cannot set new_offset for %s\n",
2645 dn);
2646 break;
2647 }
2648 }
2649 return err;
2650 release:
2651 return -1;
2652 fallback:
2653 /* Just use a backup file */
2654 return 1;
2655 }
2656
2657 static int raid10_reshape(char *container, int fd, char *devname,
2658 struct supertype *st, struct mdinfo *info,
2659 struct reshape *reshape,
2660 unsigned long long data_offset,
2661 int force, int verbose)
2662 {
2663 /* Changing raid_disks, layout, chunksize or possibly
2664 * just data_offset for a RAID10.
2665 * We must always change data_offset. We change by at least
2666 * ->min_offset_change which is the largest of the old and new
2667 * chunk sizes.
2668 * If raid_disks is increasing, then data_offset must decrease
2669 * by at least this copy size.
2670 * If raid_disks is unchanged, data_offset must increase or
2671 * decrease by at least min_offset_change but preferably by much more.
2672 * We choose half of the available space.
2673 * If raid_disks is decreasing, data_offset must increase by
2674 * at least min_offset_change. To allow of this, component_size
2675 * must be decreased by the same amount.
2676 *
2677 * So we calculate the required minimum and direction, possibly
2678 * reduce the component_size, then iterate through the devices
2679 * and set the new_data_offset.
2680 * If that all works, we set chunk_size, layout, raid_disks, and start
2681 * 'reshape'
2682 */
2683 struct mdinfo *sra;
2684 unsigned long long min;
2685 int err = 0;
2686
2687 sra = sysfs_read(fd, NULL,
2688 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2689 );
2690 if (!sra) {
2691 pr_err("%s: Cannot get array details from sysfs\n",
2692 devname);
2693 goto release;
2694 }
2695 min = reshape->min_offset_change;
2696
2697 if (info->delta_disks)
2698 sysfs_set_str(sra, NULL, "reshape_direction",
2699 info->delta_disks < 0 ? "backwards" : "forwards");
2700 if (info->delta_disks < 0 &&
2701 info->space_after < min) {
2702 int rv = sysfs_set_num(sra, NULL, "component_size",
2703 (sra->component_size -
2704 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, data_offset,
2711 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", reshape->after.layout) < 0)
2723 err = errno;
2724 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2725 info->array.raid_disks + info->delta_disks) < 0)
2726 err = errno;
2727 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2728 err = errno;
2729 if (err) {
2730 pr_err("Cannot set array shape for %s\n",
2731 devname);
2732 if (err == EBUSY &&
2733 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2734 cont_err(" Bitmap must be removed before shape can be changed\n");
2735 goto release;
2736 }
2737 sysfs_free(sra);
2738 return 0;
2739 release:
2740 sysfs_free(sra);
2741 return 1;
2742 }
2743
2744 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2745 {
2746 struct mdinfo *sra, *sd;
2747 /* Initialisation to silence compiler warning */
2748 unsigned long long min_space_before = 0, min_space_after = 0;
2749 int first = 1;
2750
2751 sra = sysfs_read(fd, NULL, GET_DEVS);
2752 if (!sra)
2753 return;
2754 for (sd = sra->devs; sd; sd = sd->next) {
2755 char *dn;
2756 int dfd;
2757 struct supertype *st2;
2758 struct mdinfo info2;
2759
2760 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2761 continue;
2762 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2763 dfd = dev_open(dn, O_RDONLY);
2764 if (dfd < 0)
2765 break;
2766 st2 = dup_super(st);
2767 if (st2->ss->load_super(st2,dfd, NULL)) {
2768 close(dfd);
2769 free(st2);
2770 break;
2771 }
2772 close(dfd);
2773 st2->ss->getinfo_super(st2, &info2, NULL);
2774 st2->ss->free_super(st2);
2775 free(st2);
2776 if (first ||
2777 min_space_before > info2.space_before)
2778 min_space_before = info2.space_before;
2779 if (first ||
2780 min_space_after > info2.space_after)
2781 min_space_after = info2.space_after;
2782 first = 0;
2783 }
2784 if (sd == NULL && !first) {
2785 info->space_after = min_space_after;
2786 info->space_before = min_space_before;
2787 }
2788 sysfs_free(sra);
2789 }
2790
2791 static void update_cache_size(char *container, struct mdinfo *sra,
2792 struct mdinfo *info,
2793 int disks, unsigned long long blocks)
2794 {
2795 /* Check that the internal stripe cache is
2796 * large enough, or it won't work.
2797 * It must hold at least 4 stripes of the larger
2798 * chunk size
2799 */
2800 unsigned long cache;
2801 cache = max(info->array.chunk_size, info->new_chunk);
2802 cache *= 4; /* 4 stripes minimum */
2803 cache /= 512; /* convert to sectors */
2804 /* make sure there is room for 'blocks' with a bit to spare */
2805 if (cache < 16 + blocks / disks)
2806 cache = 16 + blocks / disks;
2807 cache /= (4096/512); /* Convert from sectors to pages */
2808
2809 if (sra->cache_size < cache)
2810 subarray_set_num(container, sra, "stripe_cache_size",
2811 cache+1);
2812 }
2813
2814 static int impose_reshape(struct mdinfo *sra,
2815 struct mdinfo *info,
2816 struct supertype *st,
2817 int fd,
2818 int restart,
2819 char *devname, char *container,
2820 struct reshape *reshape)
2821 {
2822 struct mdu_array_info_s array;
2823
2824 sra->new_chunk = info->new_chunk;
2825
2826 if (restart) {
2827 /* for external metadata checkpoint saved by mdmon can be lost
2828 * or missed /due to e.g. crash/. Check if md is not during
2829 * restart farther than metadata points to.
2830 * If so, this means metadata information is obsolete.
2831 */
2832 if (st->ss->external)
2833 verify_reshape_position(info, reshape->level);
2834 sra->reshape_progress = info->reshape_progress;
2835 } else {
2836 sra->reshape_progress = 0;
2837 if (reshape->after.data_disks < reshape->before.data_disks)
2838 /* start from the end of the new array */
2839 sra->reshape_progress = (sra->component_size
2840 * reshape->after.data_disks);
2841 }
2842
2843 md_get_array_info(fd, &array);
2844 if (info->array.chunk_size == info->new_chunk &&
2845 reshape->before.layout == reshape->after.layout &&
2846 st->ss->external == 0) {
2847 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2848 array.raid_disks = reshape->after.data_disks + reshape->parity;
2849 if (!restart && md_set_array_info(fd, &array) != 0) {
2850 int err = errno;
2851
2852 pr_err("Cannot set device shape for %s: %s\n",
2853 devname, strerror(errno));
2854
2855 if (err == EBUSY &&
2856 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2857 cont_err("Bitmap must be removed before shape can be changed\n");
2858
2859 goto release;
2860 }
2861 } else if (!restart) {
2862 /* set them all just in case some old 'new_*' value
2863 * persists from some earlier problem.
2864 */
2865 int err = 0;
2866 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2867 err = errno;
2868 if (!err && sysfs_set_num(sra, NULL, "layout",
2869 reshape->after.layout) < 0)
2870 err = errno;
2871 if (!err && subarray_set_num(container, sra, "raid_disks",
2872 reshape->after.data_disks +
2873 reshape->parity) < 0)
2874 err = errno;
2875 if (err) {
2876 pr_err("Cannot set device shape for %s\n",
2877 devname);
2878
2879 if (err == EBUSY &&
2880 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2881 cont_err("Bitmap must be removed before shape can be changed\n");
2882 goto release;
2883 }
2884 }
2885 return 0;
2886 release:
2887 return -1;
2888 }
2889
2890 static int impose_level(int fd, int level, char *devname, int verbose)
2891 {
2892 char *c;
2893 struct mdu_array_info_s array;
2894 struct mdinfo info;
2895
2896 if (sysfs_init(&info, fd, NULL)) {
2897 pr_err("failed to intialize sysfs.\n");
2898 return 1;
2899 }
2900
2901 md_get_array_info(fd, &array);
2902 if (level == 0 &&
2903 (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 &&
2912 array.layout != ALGORITHM_PARITY_N)
2913 return -1;
2914 if (array.level == 6 &&
2915 array.layout != ALGORITHM_PARITY_N_6)
2916 return -1;
2917 sysfs_set_str(&info, NULL,"sync_action", "idle");
2918 /* First remove any spares so no recovery starts */
2919 for (d = 0, found = 0;
2920 d < MAX_DISKS && found < array.nr_disks;
2921 d++) {
2922 mdu_disk_info_t disk;
2923 disk.number = d;
2924 if (md_get_disk_info(fd, &disk) < 0)
2925 continue;
2926 if (disk.major == 0 && disk.minor == 0)
2927 continue;
2928 found++;
2929 if ((disk.state & (1 << MD_DISK_ACTIVE)) &&
2930 disk.raid_disk < data_disks)
2931 /* keep this */
2932 continue;
2933 ioctl(fd, HOT_REMOVE_DISK,
2934 makedev(disk.major, disk.minor));
2935 }
2936 /* Now fail anything left */
2937 md_get_array_info(fd, &array);
2938 for (d = 0, found = 0;
2939 d < MAX_DISKS && found < array.nr_disks;
2940 d++) {
2941 mdu_disk_info_t disk;
2942 disk.number = d;
2943 if (md_get_disk_info(fd, &disk) < 0)
2944 continue;
2945 if (disk.major == 0 && disk.minor == 0)
2946 continue;
2947 found++;
2948 if ((disk.state & (1 << MD_DISK_ACTIVE)) &&
2949 disk.raid_disk < data_disks)
2950 /* keep this */
2951 continue;
2952 ioctl(fd, SET_DISK_FAULTY,
2953 makedev(disk.major, disk.minor));
2954 hot_remove_disk(fd, makedev(disk.major, disk.minor), 1);
2955 }
2956 }
2957 c = map_num(pers, level);
2958 if (c) {
2959 int err = sysfs_set_str(&info, NULL, "level", c);
2960 if (err) {
2961 err = errno;
2962 pr_err("%s: could not set level to %s\n",
2963 devname, c);
2964 if (err == EBUSY &&
2965 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2966 cont_err("Bitmap must be removed before level can be changed\n");
2967 return err;
2968 }
2969 if (verbose >= 0)
2970 pr_err("level of %s changed to %s\n",
2971 devname, c);
2972 }
2973 return 0;
2974 }
2975
2976 int sigterm = 0;
2977 static void catch_term(int sig)
2978 {
2979 sigterm = 1;
2980 }
2981
2982 static int continue_via_systemd(char *devnm)
2983 {
2984 int skipped, i, pid, status;
2985 char pathbuf[1024];
2986 /* In a systemd/udev world, it is best to get systemd to
2987 * run "mdadm --grow --continue" rather than running in the
2988 * background.
2989 */
2990 switch(fork()) {
2991 case 0:
2992 /* FIXME yuk. CLOSE_EXEC?? */
2993 skipped = 0;
2994 for (i = 3; skipped < 20; i++)
2995 if (close(i) < 0)
2996 skipped++;
2997 else
2998 skipped = 0;
2999
3000 /* Don't want to see error messages from
3001 * systemctl. If the service doesn't exist,
3002 * we fork ourselves.
3003 */
3004 close(2);
3005 open("/dev/null", O_WRONLY);
3006 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
3007 devnm);
3008 status = execl("/usr/bin/systemctl", "systemctl",
3009 "restart",
3010 pathbuf, NULL);
3011 status = execl("/bin/systemctl", "systemctl", "restart",
3012 pathbuf, NULL);
3013 exit(1);
3014 case -1: /* Just do it ourselves. */
3015 break;
3016 default: /* parent - good */
3017 pid = wait(&status);
3018 if (pid >= 0 && status == 0)
3019 return 1;
3020 }
3021 return 0;
3022 }
3023
3024 static int reshape_array(char *container, int fd, char *devname,
3025 struct supertype *st, struct mdinfo *info,
3026 int force, struct mddev_dev *devlist,
3027 unsigned long long data_offset,
3028 char *backup_file, int verbose, int forked,
3029 int restart, int freeze_reshape)
3030 {
3031 struct reshape reshape;
3032 int spares_needed;
3033 char *msg;
3034 int orig_level = UnSet;
3035 int odisks;
3036 int delayed;
3037
3038 struct mdu_array_info_s array;
3039 char *c;
3040
3041 struct mddev_dev *dv;
3042 int added_disks;
3043
3044 int *fdlist = NULL;
3045 unsigned long long *offsets = NULL;
3046 int d;
3047 int nrdisks;
3048 int err;
3049 unsigned long blocks;
3050 unsigned long long array_size;
3051 int done;
3052 struct mdinfo *sra = NULL;
3053 char buf[20];
3054
3055 /* when reshaping a RAID0, the component_size might be zero.
3056 * So try to fix that up.
3057 */
3058 if (md_get_array_info(fd, &array) != 0) {
3059 dprintf("Cannot get array information.\n");
3060 goto release;
3061 }
3062 if (array.level == 0 && info->component_size == 0) {
3063 get_dev_size(fd, NULL, &array_size);
3064 info->component_size = array_size / array.raid_disks;
3065 }
3066
3067 if (array.level == 10)
3068 /* Need space_after info */
3069 get_space_after(fd, st, info);
3070
3071 if (info->reshape_active) {
3072 int new_level = info->new_level;
3073 info->new_level = UnSet;
3074 if (info->delta_disks > 0)
3075 info->array.raid_disks -= info->delta_disks;
3076 msg = analyse_change(devname, info, &reshape);
3077 info->new_level = new_level;
3078 if (info->delta_disks > 0)
3079 info->array.raid_disks += info->delta_disks;
3080 if (!restart)
3081 /* Make sure the array isn't read-only */
3082 ioctl(fd, RESTART_ARRAY_RW, 0);
3083 } else
3084 msg = analyse_change(devname, info, &reshape);
3085 if (msg) {
3086 /* if msg == "", error has already been printed */
3087 if (msg[0])
3088 pr_err("%s\n", msg);
3089 goto release;
3090 }
3091 if (restart &&
3092 (reshape.level != info->array.level ||
3093 reshape.before.layout != info->array.layout ||
3094 reshape.before.data_disks + reshape.parity !=
3095 info->array.raid_disks - max(0, info->delta_disks))) {
3096 pr_err("reshape info is not in native format - cannot continue.\n");
3097 goto release;
3098 }
3099
3100 if (st->ss->external && restart && (info->reshape_progress == 0) &&
3101 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
3102 (strncmp(buf, "reshape", 7) == 0))) {
3103 /* When reshape is restarted from '0', very begin of array
3104 * it is possible that for external metadata reshape and array
3105 * configuration doesn't happen.
3106 * Check if md has the same opinion, and reshape is restarted
3107 * from 0. If so, this is regular reshape start after reshape
3108 * switch in metadata to next array only.
3109 */
3110 if ((verify_reshape_position(info, reshape.level) >= 0) &&
3111 (info->reshape_progress == 0))
3112 restart = 0;
3113 }
3114 if (restart) {
3115 /* reshape already started. just skip to monitoring the reshape */
3116 if (reshape.backup_blocks == 0)
3117 return 0;
3118 if (restart & RESHAPE_NO_BACKUP)
3119 return 0;
3120
3121 /* Need 'sra' down at 'started:' */
3122 sra = sysfs_read(fd, NULL,
3123 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3124 GET_CACHE);
3125 if (!sra) {
3126 pr_err("%s: Cannot get array details from sysfs\n",
3127 devname);
3128 goto release;
3129 }
3130
3131 if (!backup_file)
3132 backup_file = locate_backup(sra->sys_name);
3133
3134 goto started;
3135 }
3136 /* The container is frozen but the array may not be.
3137 * So freeze the array so spares don't get put to the wrong use
3138 * FIXME there should probably be a cleaner separation between
3139 * freeze_array and freeze_container.
3140 */
3141 sysfs_freeze_array(info);
3142 /* Check we have enough spares to not be degraded */
3143 added_disks = 0;
3144 for (dv = devlist; dv ; dv=dv->next)
3145 added_disks++;
3146 spares_needed = max(reshape.before.data_disks,
3147 reshape.after.data_disks)
3148 + reshape.parity - array.raid_disks;
3149
3150 if (!force &&
3151 info->new_level > 1 && info->array.level > 1 &&
3152 spares_needed > info->array.spare_disks + added_disks) {
3153 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3154 " Use --force to over-ride this check.\n",
3155 spares_needed,
3156 spares_needed == 1 ? "" : "s",
3157 info->array.spare_disks + added_disks);
3158 goto release;
3159 }
3160 /* Check we have enough spares to not fail */
3161 spares_needed = max(reshape.before.data_disks,
3162 reshape.after.data_disks)
3163 - array.raid_disks;
3164 if ((info->new_level > 1 || info->new_level == 0) &&
3165 spares_needed > info->array.spare_disks +added_disks) {
3166 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3167 spares_needed,
3168 spares_needed == 1 ? "" : "s",
3169 info->array.spare_disks + added_disks);
3170 goto release;
3171 }
3172
3173 if (reshape.level != array.level) {
3174 int err = impose_level(fd, reshape.level, devname, verbose);
3175 if (err)
3176 goto release;
3177 info->new_layout = UnSet; /* after level change,
3178 * layout is meaningless */
3179 orig_level = array.level;
3180 sysfs_freeze_array(info);
3181
3182 if (reshape.level > 0 && st->ss->external) {
3183 /* make sure mdmon is aware of the new level */
3184 if (mdmon_running(container))
3185 flush_mdmon(container);
3186
3187 if (!mdmon_running(container))
3188 start_mdmon(container);
3189 ping_monitor(container);
3190 if (mdmon_running(container) &&
3191 st->update_tail == NULL)
3192 st->update_tail = &st->updates;
3193 }
3194 }
3195 /* ->reshape_super might have chosen some spares from the
3196 * container that it wants to be part of the new array.
3197 * We can collect them with ->container_content and give
3198 * them to the kernel.
3199 */
3200 if (st->ss->reshape_super && st->ss->container_content) {
3201 char *subarray = strchr(info->text_version+1, '/')+1;
3202 struct mdinfo *info2 =
3203 st->ss->container_content(st, subarray);
3204 struct mdinfo *d;
3205
3206 if (info2) {
3207 if (sysfs_init(info2, fd, st->devnm)) {
3208 pr_err("unable to initialize sysfs for %s\n",
3209 st->devnm);
3210 free(info2);
3211 goto release;
3212 }
3213 /* When increasing number of devices, we need to set
3214 * new raid_disks before adding these, or they might
3215 * be rejected.
3216 */
3217 if (reshape.backup_blocks &&
3218 reshape.after.data_disks > reshape.before.data_disks)
3219 subarray_set_num(container, info2, "raid_disks",
3220 reshape.after.data_disks +
3221 reshape.parity);
3222 for (d = info2->devs; d; d = d->next) {
3223 if (d->disk.state == 0 &&
3224 d->disk.raid_disk >= 0) {
3225 /* This is a spare that wants to
3226 * be part of the array.
3227 */
3228 add_disk(fd, st, info2, d);
3229 }
3230 }
3231 sysfs_free(info2);
3232 }
3233 }
3234 /* We might have been given some devices to add to the
3235 * array. Now that the array has been changed to the right
3236 * level and frozen, we can safely add them.
3237 */
3238 if (devlist) {
3239 if (Manage_subdevs(devname, fd, devlist, verbose,
3240 0, NULL, 0))
3241 goto release;
3242 }
3243
3244 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3245 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3246 if (reshape.backup_blocks == 0) {
3247 /* No restriping needed, but we might need to impose
3248 * some more changes: layout, raid_disks, chunk_size
3249 */
3250 /* read current array info */
3251 if (md_get_array_info(fd, &array) != 0) {
3252 dprintf("Cannot get array information.\n");
3253 goto release;
3254 }
3255 /* compare current array info with new values and if
3256 * it is different update them to new */
3257 if (info->new_layout != UnSet &&
3258 info->new_layout != array.layout) {
3259 array.layout = info->new_layout;
3260 if (md_set_array_info(fd, &array) != 0) {
3261 pr_err("failed to set new layout\n");
3262 goto release;
3263 } else if (verbose >= 0)
3264 printf("layout for %s set to %d\n",
3265 devname, array.layout);
3266 }
3267 if (info->delta_disks != UnSet &&
3268 info->delta_disks != 0 &&
3269 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3270 array.raid_disks += info->delta_disks;
3271 if (md_set_array_info(fd, &array) != 0) {
3272 pr_err("failed to set raid disks\n");
3273 goto release;
3274 } else if (verbose >= 0) {
3275 printf("raid_disks for %s set to %d\n",
3276 devname, array.raid_disks);
3277 }
3278 }
3279 if (info->new_chunk != 0 &&
3280 info->new_chunk != array.chunk_size) {
3281 if (sysfs_set_num(info, NULL,
3282 "chunk_size", info->new_chunk) != 0) {
3283 pr_err("failed to set chunk size\n");
3284 goto release;
3285 } else if (verbose >= 0)
3286 printf("chunk size for %s set to %d\n",
3287 devname, array.chunk_size);
3288 }
3289 unfreeze(st);
3290 return 0;
3291 }
3292
3293 /*
3294 * There are three possibilities.
3295 * 1/ The array will shrink.
3296 * We need to ensure the reshape will pause before reaching
3297 * the 'critical section'. We also need to fork and wait for
3298 * that to happen. When it does we
3299 * suspend/backup/complete/unfreeze
3300 *
3301 * 2/ The array will not change size.
3302 * This requires that we keep a backup of a sliding window
3303 * so that we can restore data after a crash. So we need
3304 * to fork and monitor progress.
3305 * In future we will allow the data_offset to change, so
3306 * a sliding backup becomes unnecessary.
3307 *
3308 * 3/ The array will grow. This is relatively easy.
3309 * However the kernel's restripe routines will cheerfully
3310 * overwrite some early data before it is safe. So we
3311 * need to make a backup of the early parts of the array
3312 * and be ready to restore it if rebuild aborts very early.
3313 * For externally managed metadata, we still need a forked
3314 * child to monitor the reshape and suspend IO over the region
3315 * that is being reshaped.
3316 *
3317 * We backup data by writing it to one spare, or to a
3318 * file which was given on command line.
3319 *
3320 * In each case, we first make sure that storage is available
3321 * for the required backup.
3322 * Then we:
3323 * - request the shape change.
3324 * - fork to handle backup etc.
3325 */
3326 /* Check that we can hold all the data */
3327 get_dev_size(fd, NULL, &array_size);
3328 if (reshape.new_size < (array_size/512)) {
3329 pr_err("this change will reduce the size of the array.\n"
3330 " use --grow --array-size first to truncate array.\n"
3331 " e.g. mdadm --grow %s --array-size %llu\n",
3332 devname, reshape.new_size/2);
3333 goto release;
3334 }
3335
3336 if (array.level == 10) {
3337 /* Reshaping RAID10 does not require any data backup by
3338 * user-space. Instead it requires that the data_offset
3339 * is changed to avoid the need for backup.
3340 * So this is handled very separately
3341 */
3342 if (restart)
3343 /* Nothing to do. */
3344 return 0;
3345 return raid10_reshape(container, fd, devname, st, info,
3346 &reshape, data_offset,
3347 force, verbose);
3348 }
3349 sra = sysfs_read(fd, NULL,
3350 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3351 GET_CACHE);
3352 if (!sra) {
3353 pr_err("%s: Cannot get array details from sysfs\n",
3354 devname);
3355 goto release;
3356 }
3357
3358 if (!backup_file)
3359 switch(set_new_data_offset(sra, st, devname,
3360 reshape.after.data_disks - reshape.before.data_disks,
3361 data_offset,
3362 reshape.min_offset_change, 1)) {
3363 case -1:
3364 goto release;
3365 case 0:
3366 /* Updated data_offset, so it's easy now */
3367 update_cache_size(container, sra, info,
3368 min(reshape.before.data_disks,
3369 reshape.after.data_disks),
3370 reshape.backup_blocks);
3371
3372 /* Right, everything seems fine. Let's kick things off.
3373 */
3374 sync_metadata(st);
3375
3376 if (impose_reshape(sra, info, st, fd, restart,
3377 devname, container, &reshape) < 0)
3378 goto release;
3379 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3380 struct mdinfo *sd;
3381 if (errno != EINVAL) {
3382 pr_err("Failed to initiate reshape!\n");
3383 goto release;
3384 }
3385 /* revert data_offset and try the old way */
3386 for (sd = sra->devs; sd; sd = sd->next) {
3387 sysfs_set_num(sra, sd, "new_offset",
3388 sd->data_offset);
3389 sysfs_set_str(sra, NULL, "reshape_direction",
3390 "forwards");
3391 }
3392 break;
3393 }
3394 if (info->new_level == reshape.level)
3395 return 0;
3396 /* need to adjust level when reshape completes */
3397 switch(fork()) {
3398 case -1: /* ignore error, but don't wait */
3399 return 0;
3400 default: /* parent */
3401 return 0;
3402 case 0:
3403 map_fork();
3404 break;
3405 }
3406 close(fd);
3407 wait_reshape(sra);
3408 fd = open_dev(sra->sys_name);
3409 if (fd >= 0)
3410 impose_level(fd, info->new_level, devname, verbose);
3411 return 0;
3412 case 1: /* Couldn't set data_offset, try the old way */
3413 if (data_offset != INVALID_SECTORS) {
3414 pr_err("Cannot update data_offset on this array\n");
3415 goto release;
3416 }
3417 break;
3418 }
3419
3420 started:
3421 /* Decide how many blocks (sectors) for a reshape
3422 * unit. The number we have so far is just a minimum
3423 */
3424 blocks = reshape.backup_blocks;
3425 if (reshape.before.data_disks ==
3426 reshape.after.data_disks) {
3427 /* Make 'blocks' bigger for better throughput, but
3428 * not so big that we reject it below.
3429 * Try for 16 megabytes
3430 */
3431 while (blocks * 32 < sra->component_size &&
3432 blocks < 16*1024*2)
3433 blocks *= 2;
3434 } else
3435 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3436
3437 if (blocks >= sra->component_size/2) {
3438 pr_err("%s: Something wrong - reshape aborted\n",
3439 devname);
3440 goto release;
3441 }
3442
3443 /* Now we need to open all these devices so we can read/write.
3444 */
3445 nrdisks = max(reshape.before.data_disks,
3446 reshape.after.data_disks) + reshape.parity
3447 + sra->array.spare_disks;
3448 fdlist = xcalloc((1+nrdisks), sizeof(int));
3449 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3450
3451 odisks = reshape.before.data_disks + reshape.parity;
3452 d = reshape_prepare_fdlist(devname, sra, odisks,
3453 nrdisks, blocks, backup_file,
3454 fdlist, offsets);
3455 if (d < odisks) {
3456 goto release;
3457 }
3458 if ((st->ss->manage_reshape == NULL) ||
3459 (st->ss->recover_backup == NULL)) {
3460 if (backup_file == NULL) {
3461 if (reshape.after.data_disks <=
3462 reshape.before.data_disks) {
3463 pr_err("%s: Cannot grow - need backup-file\n",
3464 devname);
3465 pr_err(" Please provide one with \"--backup=...\"\n");
3466 goto release;
3467 } else if (d == odisks) {
3468 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3469 goto release;
3470 }
3471 } else {
3472 if (!reshape_open_backup_file(backup_file, fd, devname,
3473 (signed)blocks,
3474 fdlist+d, offsets+d,
3475 sra->sys_name,
3476 restart)) {
3477 goto release;
3478 }
3479 d++;
3480 }
3481 }
3482
3483 update_cache_size(container, sra, info,
3484 min(reshape.before.data_disks, reshape.after.data_disks),
3485 blocks);
3486
3487 /* Right, everything seems fine. Let's kick things off.
3488 * If only changing raid_disks, use ioctl, else use
3489 * sysfs.
3490 */
3491 sync_metadata(st);
3492
3493 if (impose_reshape(sra, info, st, fd, restart,
3494 devname, container, &reshape) < 0)
3495 goto release;
3496
3497 err = start_reshape(sra, restart, reshape.before.data_disks,
3498 reshape.after.data_disks);
3499 if (err) {
3500 pr_err("Cannot %s reshape for %s\n",
3501 restart ? "continue" : "start",
3502 devname);
3503 goto release;
3504 }
3505 if (restart)
3506 sysfs_set_str(sra, NULL, "array_state", "active");
3507 if (freeze_reshape) {
3508 free(fdlist);
3509 free(offsets);
3510 sysfs_free(sra);
3511 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3512 sra->reshape_progress);
3513 return 1;
3514 }
3515
3516 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3517 if (continue_via_systemd(container ?: sra->sys_name)) {
3518 free(fdlist);
3519 free(offsets);
3520 sysfs_free(sra);
3521 return 0;
3522 }
3523
3524 /* Now we just need to kick off the reshape and watch, while
3525 * handling backups of the data...
3526 * This is all done by a forked background process.
3527 */
3528 switch(forked ? 0 : fork()) {
3529 case -1:
3530 pr_err("Cannot run child to monitor reshape: %s\n",
3531 strerror(errno));
3532 abort_reshape(sra);
3533 goto release;
3534 default:
3535 free(fdlist);
3536 free(offsets);
3537 sysfs_free(sra);
3538 return 0;
3539 case 0:
3540 map_fork();
3541 break;
3542 }
3543
3544 /* If another array on the same devices is busy, the
3545 * reshape will wait for them. This would mean that
3546 * the first section that we suspend will stay suspended
3547 * for a long time. So check on that possibility
3548 * by looking for "DELAYED" in /proc/mdstat, and if found,
3549 * wait a while
3550 */
3551 do {
3552 struct mdstat_ent *mds, *m;
3553 delayed = 0;
3554 mds = mdstat_read(1, 0);
3555 for (m = mds; m; m = m->next)
3556 if (strcmp(m->devnm, sra->sys_name) == 0) {
3557 if (m->resync &&
3558 m->percent == RESYNC_DELAYED)
3559 delayed = 1;
3560 if (m->resync == 0)
3561 /* Haven't started the reshape thread
3562 * yet, wait a bit
3563 */
3564 delayed = 2;
3565 break;
3566 }
3567 free_mdstat(mds);
3568 if (delayed == 1 && get_linux_version() < 3007000) {
3569 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3570 " You might experience problems until other reshapes complete.\n");
3571 delayed = 0;
3572 }
3573 if (delayed)
3574 mdstat_wait(30 - (delayed-1) * 25);
3575 } while (delayed);
3576 mdstat_close();
3577 close(fd);
3578 if (check_env("MDADM_GROW_VERIFY"))
3579 fd = open(devname, O_RDONLY | O_DIRECT);
3580 else
3581 fd = -1;
3582 mlockall(MCL_FUTURE);
3583
3584 signal(SIGTERM, catch_term);
3585
3586 if (st->ss->external) {
3587 /* metadata handler takes it from here */
3588 done = st->ss->manage_reshape(
3589 fd, sra, &reshape, st, blocks,
3590 fdlist, offsets,
3591 d - odisks, fdlist+odisks,
3592 offsets+odisks);
3593 } else
3594 done = child_monitor(
3595 fd, sra, &reshape, st, blocks,
3596 fdlist, offsets,
3597 d - odisks, fdlist+odisks,
3598 offsets+odisks);
3599
3600 free(fdlist);
3601 free(offsets);
3602
3603 if (backup_file && done) {
3604 char *bul;
3605 bul = make_backup(sra->sys_name);
3606 if (bul) {
3607 char buf[1024];
3608 int l = readlink(bul, buf, sizeof(buf) - 1);
3609 if (l > 0) {
3610 buf[l]=0;
3611 unlink(buf);
3612 }
3613 unlink(bul);
3614 free(bul);
3615 }
3616 unlink(backup_file);
3617 }
3618 if (!done) {
3619 abort_reshape(sra);
3620 goto out;
3621 }
3622
3623 if (!st->ss->external &&
3624 !(reshape.before.data_disks != reshape.after.data_disks &&
3625 info->custom_array_size) && info->new_level == reshape.level &&
3626 !forked) {
3627 /* no need to wait for the reshape to finish as
3628 * there is nothing more to do.
3629 */
3630 sysfs_free(sra);
3631 exit(0);
3632 }
3633 wait_reshape(sra);
3634
3635 if (st->ss->external) {
3636 /* Re-load the metadata as much could have changed */
3637 int cfd = open_dev(st->container_devnm);
3638 if (cfd >= 0) {
3639 flush_mdmon(container);
3640 st->ss->free_super(st);
3641 st->ss->load_container(st, cfd, container);
3642 close(cfd);
3643 }
3644 }
3645
3646 /* set new array size if required customer_array_size is used
3647 * by this metadata.
3648 */
3649 if (reshape.before.data_disks !=
3650 reshape.after.data_disks &&
3651 info->custom_array_size)
3652 set_array_size(st, info, info->text_version);
3653
3654 if (info->new_level != reshape.level) {
3655 if (fd < 0)
3656 fd = open(devname, O_RDONLY);
3657 impose_level(fd, info->new_level, devname, verbose);
3658 close(fd);
3659 if (info->new_level == 0)
3660 st->update_tail = NULL;
3661 }
3662 out:
3663 sysfs_free(sra);
3664 if (forked)
3665 return 0;
3666 unfreeze(st);
3667 exit(0);
3668
3669 release:
3670 free(fdlist);
3671 free(offsets);
3672 if (orig_level != UnSet && sra) {
3673 c = map_num(pers, orig_level);
3674 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3675 pr_err("aborting level change\n");
3676 }
3677 sysfs_free(sra);
3678 if (!forked)
3679 unfreeze(st);
3680 return 1;
3681 }
3682
3683 /* mdfd handle is passed to be closed in child process (after fork).
3684 */
3685 int reshape_container(char *container, char *devname,
3686 int mdfd,
3687 struct supertype *st,
3688 struct mdinfo *info,
3689 int force,
3690 char *backup_file, int verbose,
3691 int forked, int restart, int freeze_reshape)
3692 {
3693 struct mdinfo *cc = NULL;
3694 int rv = restart;
3695 char last_devnm[32] = "";
3696
3697 /* component_size is not meaningful for a container,
3698 * so pass '0' meaning 'no change'
3699 */
3700 if (!restart &&
3701 reshape_super(st, 0, info->new_level,
3702 info->new_layout, info->new_chunk,
3703 info->array.raid_disks, info->delta_disks,
3704 backup_file, devname, APPLY_METADATA_CHANGES,
3705 verbose)) {
3706 unfreeze(st);
3707 return 1;
3708 }
3709
3710 sync_metadata(st);
3711
3712 /* ping monitor to be sure that update is on disk
3713 */
3714 ping_monitor(container);
3715
3716 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3717 if (continue_via_systemd(container))
3718 return 0;
3719
3720 switch (forked ? 0 : fork()) {
3721 case -1: /* error */
3722 perror("Cannot fork to complete reshape\n");
3723 unfreeze(st);
3724 return 1;
3725 default: /* parent */
3726 if (!freeze_reshape)
3727 printf("%s: multi-array reshape continues in background\n", Name);
3728 return 0;
3729 case 0: /* child */
3730 map_fork();
3731 break;
3732 }
3733
3734 /* close unused handle in child process
3735 */
3736 if (mdfd > -1)
3737 close(mdfd);
3738
3739 while(1) {
3740 /* For each member array with reshape_active,
3741 * we need to perform the reshape.
3742 * We pick the first array that needs reshaping and
3743 * reshape it. reshape_array() will re-read the metadata
3744 * so the next time through a different array should be
3745 * ready for reshape.
3746 * It is possible that the 'different' array will not
3747 * be assembled yet. In that case we simple exit.
3748 * When it is assembled, the mdadm which assembles it
3749 * will take over the reshape.
3750 */
3751 struct mdinfo *content;
3752 int fd;
3753 struct mdstat_ent *mdstat;
3754 char *adev;
3755 dev_t devid;
3756
3757 sysfs_free(cc);
3758
3759 cc = st->ss->container_content(st, NULL);
3760
3761 for (content = cc; content ; content = content->next) {
3762 char *subarray;
3763 if (!content->reshape_active)
3764 continue;
3765
3766 subarray = strchr(content->text_version+1, '/')+1;
3767 mdstat = mdstat_by_subdev(subarray, container);
3768 if (!mdstat)
3769 continue;
3770 if (mdstat->active == 0) {
3771 pr_err("Skipping inactive array %s.\n",
3772 mdstat->devnm);
3773 free_mdstat(mdstat);
3774 mdstat = NULL;
3775 continue;
3776 }
3777 break;
3778 }
3779 if (!content)
3780 break;
3781
3782 devid = devnm2devid(mdstat->devnm);
3783 adev = map_dev(major(devid), minor(devid), 0);
3784 if (!adev)
3785 adev = content->text_version;
3786
3787 fd = open_dev(mdstat->devnm);
3788 if (fd < 0) {
3789 pr_err("Device %s cannot be opened for reshape.\n", adev);
3790 break;
3791 }
3792
3793 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3794 /* Do not allow for multiple reshape_array() calls for
3795 * the same array.
3796 * It can happen when reshape_array() returns without
3797 * error, when reshape is not finished (wrong reshape
3798 * starting/continuation conditions). Mdmon doesn't
3799 * switch to next array in container and reentry
3800 * conditions for the same array occur.
3801 * This is possibly interim until the behaviour of
3802 * reshape_array is resolved().
3803 */
3804 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3805 close(fd);
3806 break;
3807 }
3808 strcpy(last_devnm, mdstat->devnm);
3809
3810 if (sysfs_init(content, fd, mdstat->devnm)) {
3811 pr_err("Unable to initialize sysfs for %s\n",
3812 mdstat->devnm);
3813 rv = 1;
3814 break;
3815 }
3816
3817 if (mdmon_running(container))
3818 flush_mdmon(container);
3819
3820 rv = reshape_array(container, fd, adev, st,
3821 content, force, NULL, INVALID_SECTORS,
3822 backup_file, verbose, 1, restart,
3823 freeze_reshape);
3824 close(fd);
3825
3826 if (freeze_reshape) {
3827 sysfs_free(cc);
3828 exit(0);
3829 }
3830
3831 restart = 0;
3832 if (rv)
3833 break;
3834
3835 if (mdmon_running(container))
3836 flush_mdmon(container);
3837 }
3838 if (!rv)
3839 unfreeze(st);
3840 sysfs_free(cc);
3841 exit(0);
3842 }
3843
3844 /*
3845 * We run a child process in the background which performs the following
3846 * steps:
3847 * - wait for resync to reach a certain point
3848 * - suspend io to the following section
3849 * - backup that section
3850 * - allow resync to proceed further
3851 * - resume io
3852 * - discard the backup.
3853 *
3854 * When are combined in slightly different ways in the three cases.
3855 * Grow:
3856 * - suspend/backup/allow/wait/resume/discard
3857 * Shrink:
3858 * - allow/wait/suspend/backup/allow/wait/resume/discard
3859 * same-size:
3860 * - wait/resume/discard/suspend/backup/allow
3861 *
3862 * suspend/backup/allow always come together
3863 * wait/resume/discard do too.
3864 * For the same-size case we have two backups to improve flow.
3865 *
3866 */
3867
3868 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3869 unsigned long long backup_point,
3870 unsigned long long wait_point,
3871 unsigned long long *suspend_point,
3872 unsigned long long *reshape_completed, int *frozen)
3873 {
3874 /* This function is called repeatedly by the reshape manager.
3875 * It determines how much progress can safely be made and allows
3876 * that progress.
3877 * - 'info' identifies the array and particularly records in
3878 * ->reshape_progress the metadata's knowledge of progress
3879 * This is a sector offset from the start of the array
3880 * of the next array block to be relocated. This number
3881 * may increase from 0 or decrease from array_size, depending
3882 * on the type of reshape that is happening.
3883 * Note that in contrast, 'sync_completed' is a block count of the
3884 * reshape so far. It gives the distance between the start point
3885 * (head or tail of device) and the next place that data will be
3886 * written. It always increases.
3887 * - 'reshape' is the structure created by analyse_change
3888 * - 'backup_point' shows how much the metadata manager has backed-up
3889 * data. For reshapes with increasing progress, it is the next address
3890 * to be backed up, previous addresses have been backed-up. For
3891 * decreasing progress, it is the earliest address that has been
3892 * backed up - later address are also backed up.
3893 * So addresses between reshape_progress and backup_point are
3894 * backed up providing those are in the 'correct' order.
3895 * - 'wait_point' is an array address. When reshape_completed
3896 * passes this point, progress_reshape should return. It might
3897 * return earlier if it determines that ->reshape_progress needs
3898 * to be updated or further backup is needed.
3899 * - suspend_point is maintained by progress_reshape and the caller
3900 * should not touch it except to initialise to zero.
3901 * It is an array address and it only increases in 2.6.37 and earlier.
3902 * This makes it difficult to handle reducing reshapes with
3903 * external metadata.
3904 * However: it is similar to backup_point in that it records the
3905 * other end of a suspended region from reshape_progress.
3906 * it is moved to extend the region that is safe to backup and/or
3907 * reshape
3908 * - reshape_completed is read from sysfs and returned. The caller
3909 * should copy this into ->reshape_progress when it has reason to
3910 * believe that the metadata knows this, and any backup outside this
3911 * has been erased.
3912 *
3913 * Return value is:
3914 * 1 if more data from backup_point - but only as far as suspend_point,
3915 * should be backed up
3916 * 0 if things are progressing smoothly
3917 * -1 if the reshape is finished because it is all done,
3918 * -2 if the reshape is finished due to an error.
3919 */
3920
3921 int advancing = (reshape->after.data_disks
3922 >= reshape->before.data_disks);
3923 unsigned long long need_backup; /* All data between start of array and
3924 * here will at some point need to
3925 * be backed up.
3926 */
3927 unsigned long long read_offset, write_offset;
3928 unsigned long long write_range;
3929 unsigned long long max_progress, target, completed;
3930 unsigned long long array_size = (info->component_size
3931 * reshape->before.data_disks);
3932 int fd;
3933 char buf[20];
3934
3935 /* First, we unsuspend any region that is now known to be safe.
3936 * If suspend_point is on the 'wrong' side of reshape_progress, then
3937 * we don't have or need suspension at the moment. This is true for
3938 * native metadata when we don't need to back-up.
3939 */
3940 if (advancing) {
3941 if (info->reshape_progress <= *suspend_point)
3942 sysfs_set_num(info, NULL, "suspend_lo",
3943 info->reshape_progress);
3944 } else {
3945 /* Note: this won't work in 2.6.37 and before.
3946 * Something somewhere should make sure we don't need it!
3947 */
3948 if (info->reshape_progress >= *suspend_point)
3949 sysfs_set_num(info, NULL, "suspend_hi",
3950 info->reshape_progress);
3951 }
3952
3953 /* Now work out how far it is safe to progress.
3954 * If the read_offset for ->reshape_progress is less than
3955 * 'blocks' beyond the write_offset, we can only progress as far
3956 * as a backup.
3957 * Otherwise we can progress until the write_offset for the new location
3958 * reaches (within 'blocks' of) the read_offset at the current location.
3959 * However that region must be suspended unless we are using native
3960 * metadata.
3961 * If we need to suspend more, we limit it to 128M per device, which is
3962 * rather arbitrary and should be some time-based calculation.
3963 */
3964 read_offset = info->reshape_progress / reshape->before.data_disks;
3965 write_offset = info->reshape_progress / reshape->after.data_disks;
3966 write_range = info->new_chunk/512;
3967 if (reshape->before.data_disks == reshape->after.data_disks)
3968 need_backup = array_size;
3969 else
3970 need_backup = reshape->backup_blocks;
3971 if (advancing) {
3972 if (read_offset < write_offset + write_range)
3973 max_progress = backup_point;
3974 else
3975 max_progress =
3976 read_offset *
3977 reshape->after.data_disks;
3978 } else {
3979 if (read_offset > write_offset - write_range)
3980 /* Can only progress as far as has been backed up,
3981 * which must be suspended */
3982 max_progress = backup_point;
3983 else if (info->reshape_progress <= need_backup)
3984 max_progress = backup_point;
3985 else {
3986 if (info->array.major_version >= 0)
3987 /* Can progress until backup is needed */
3988 max_progress = need_backup;
3989 else {
3990 /* Can progress until metadata update is required */
3991 max_progress =
3992 read_offset *
3993 reshape->after.data_disks;
3994 /* but data must be suspended */
3995 if (max_progress < *suspend_point)
3996 max_progress = *suspend_point;
3997 }
3998 }
3999 }
4000
4001 /* We know it is safe to progress to 'max_progress' providing
4002 * it is suspended or we are using native metadata.
4003 * Consider extending suspend_point 128M per device if it
4004 * is less than 64M per device beyond reshape_progress.
4005 * But always do a multiple of 'blocks'
4006 * FIXME this is too big - it takes to long to complete
4007 * this much.
4008 */
4009 target = 64*1024*2 * min(reshape->before.data_disks,
4010 reshape->after.data_disks);
4011 target /= reshape->backup_blocks;
4012 if (target < 2)
4013 target = 2;
4014 target *= reshape->backup_blocks;
4015
4016 /* For externally managed metadata we always need to suspend IO to
4017 * the area being reshaped so we regularly push suspend_point forward.
4018 * For native metadata we only need the suspend if we are going to do
4019 * a backup.
4020 */
4021 if (advancing) {
4022 if ((need_backup > info->reshape_progress ||
4023 info->array.major_version < 0) &&
4024 *suspend_point < info->reshape_progress + target) {
4025 if (need_backup < *suspend_point + 2 * target)
4026 *suspend_point = need_backup;
4027 else if (*suspend_point + 2 * target < array_size)
4028 *suspend_point += 2 * target;
4029 else
4030 *suspend_point = array_size;
4031 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
4032 if (max_progress > *suspend_point)
4033 max_progress = *suspend_point;
4034 }
4035 } else {
4036 if (info->array.major_version >= 0) {
4037 /* Only need to suspend when about to backup */
4038 if (info->reshape_progress < need_backup * 2 &&
4039 *suspend_point > 0) {
4040 *suspend_point = 0;
4041 sysfs_set_num(info, NULL, "suspend_lo", 0);
4042 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
4043 }
4044 } else {
4045 /* Need to suspend continually */
4046 if (info->reshape_progress < *suspend_point)
4047 *suspend_point = info->reshape_progress;
4048 if (*suspend_point + target < info->reshape_progress)
4049 /* No need to move suspend region yet */;
4050 else {
4051 if (*suspend_point >= 2 * target)
4052 *suspend_point -= 2 * target;
4053 else
4054 *suspend_point = 0;
4055 sysfs_set_num(info, NULL, "suspend_lo",
4056 *suspend_point);
4057 }
4058 if (max_progress < *suspend_point)
4059 max_progress = *suspend_point;
4060 }
4061 }
4062
4063 /* now set sync_max to allow that progress. sync_max, like
4064 * sync_completed is a count of sectors written per device, so
4065 * we find the difference between max_progress and the start point,
4066 * and divide that by after.data_disks to get a sync_max
4067 * number.
4068 * At the same time we convert wait_point to a similar number
4069 * for comparing against sync_completed.
4070 */
4071 /* scale down max_progress to per_disk */
4072 max_progress /= reshape->after.data_disks;
4073 /* Round to chunk size as some kernels give an erroneously high number */
4074 max_progress /= info->new_chunk/512;
4075 max_progress *= info->new_chunk/512;
4076 /* And round to old chunk size as the kernel wants that */
4077 max_progress /= info->array.chunk_size/512;
4078 max_progress *= info->array.chunk_size/512;
4079 /* Limit progress to the whole device */
4080 if (max_progress > info->component_size)
4081 max_progress = info->component_size;
4082 wait_point /= reshape->after.data_disks;
4083 if (!advancing) {
4084 /* switch from 'device offset' to 'processed block count' */
4085 max_progress = info->component_size - max_progress;
4086 wait_point = info->component_size - wait_point;
4087 }
4088
4089 if (!*frozen)
4090 sysfs_set_num(info, NULL, "sync_max", max_progress);
4091
4092 /* Now wait. If we have already reached the point that we were
4093 * asked to wait to, don't wait at all, else wait for any change.
4094 * We need to select on 'sync_completed' as that is the place that
4095 * notifications happen, but we are really interested in
4096 * 'reshape_position'
4097 */
4098 fd = sysfs_get_fd(info, NULL, "sync_completed");
4099 if (fd < 0)
4100 goto check_progress;
4101
4102 if (sysfs_fd_get_ll(fd, &completed) < 0)
4103 goto check_progress;
4104
4105 while (completed < max_progress && completed < wait_point) {
4106 /* Check that sync_action is still 'reshape' to avoid
4107 * waiting forever on a dead array
4108 */
4109 char action[20];
4110 if (sysfs_get_str(info, NULL, "sync_action",
4111 action, 20) <= 0 ||
4112 strncmp(action, "reshape", 7) != 0)
4113 break;
4114 /* Some kernels reset 'sync_completed' to zero
4115 * before setting 'sync_action' to 'idle'.
4116 * So we need these extra tests.
4117 */
4118 if (completed == 0 && advancing &&
4119 strncmp(action, "idle", 4) == 0 &&
4120 info->reshape_progress > 0)
4121 break;
4122 if (completed == 0 && !advancing &&
4123 strncmp(action, "idle", 4) == 0 &&
4124 info->reshape_progress < (info->component_size
4125 * reshape->after.data_disks))
4126 break;
4127 sysfs_wait(fd, NULL);
4128 if (sysfs_fd_get_ll(fd, &completed) < 0)
4129 goto check_progress;
4130 }
4131 /* Some kernels reset 'sync_completed' to zero,
4132 * we need to have real point we are in md.
4133 * So in that case, read 'reshape_position' from sysfs.
4134 */
4135 if (completed == 0) {
4136 unsigned long long reshapep;
4137 char action[20];
4138 if (sysfs_get_str(info, NULL, "sync_action",
4139 action, 20) > 0 &&
4140 strncmp(action, "idle", 4) == 0 &&
4141 sysfs_get_ll(info, NULL,
4142 "reshape_position", &reshapep) == 0)
4143 *reshape_completed = reshapep;
4144 } else {
4145 /* some kernels can give an incorrectly high
4146 * 'completed' number, so round down */
4147 completed /= (info->new_chunk/512);
4148 completed *= (info->new_chunk/512);
4149 /* Convert 'completed' back in to a 'progress' number */
4150 completed *= reshape->after.data_disks;
4151 if (!advancing)
4152 completed = (info->component_size
4153 * reshape->after.data_disks
4154 - completed);
4155 *reshape_completed = completed;
4156 }
4157
4158 close(fd);
4159
4160 /* We return the need_backup flag. Caller will decide
4161 * how much - a multiple of ->backup_blocks up to *suspend_point
4162 */
4163 if (advancing)
4164 return need_backup > info->reshape_progress;
4165 else
4166 return need_backup >= info->reshape_progress;
4167
4168 check_progress:
4169 /* if we couldn't read a number from sync_completed, then
4170 * either the reshape did complete, or it aborted.
4171 * We can tell which by checking for 'none' in reshape_position.
4172 * If it did abort, then it might immediately restart if it
4173 * it was just a device failure that leaves us degraded but
4174 * functioning.
4175 */
4176 if (sysfs_get_str(info, NULL, "reshape_position", buf,
4177 sizeof(buf)) < 0 ||
4178 strncmp(buf, "none", 4) != 0) {
4179 /* The abort might only be temporary. Wait up to 10
4180 * seconds for fd to contain a valid number again.
4181 */
4182 int wait = 10000;
4183 int rv = -2;
4184 unsigned long long new_sync_max;
4185 while (fd >= 0 && rv < 0 && wait > 0) {
4186 if (sysfs_wait(fd, &wait) != 1)
4187 break;
4188 switch (sysfs_fd_get_ll(fd, &completed)) {
4189 case 0:
4190 /* all good again */
4191 rv = 1;
4192 /* If "sync_max" is no longer max_progress
4193 * we need to freeze things
4194 */
4195 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
4196 *frozen = (new_sync_max != max_progress);
4197 break;
4198 case -2: /* read error - abort */
4199 wait = 0;
4200 break;
4201 }
4202 }
4203 if (fd >= 0)
4204 close(fd);
4205 return rv; /* abort */
4206 } else {
4207 /* Maybe racing with array shutdown - check state */
4208 if (fd >= 0)
4209 close(fd);
4210 if (sysfs_get_str(info, NULL, "array_state", buf,
4211 sizeof(buf)) < 0 ||
4212 strncmp(buf, "inactive", 8) == 0 ||
4213 strncmp(buf, "clear",5) == 0)
4214 return -2; /* abort */
4215 return -1; /* complete */
4216 }
4217 }
4218
4219 /* FIXME return status is never checked */
4220 static int grow_backup(struct mdinfo *sra,
4221 unsigned long long offset, /* per device */
4222 unsigned long stripes, /* per device, in old chunks */
4223 int *sources, unsigned long long *offsets,
4224 int disks, int chunk, int level, int layout,
4225 int dests, int *destfd, unsigned long long *destoffsets,
4226 int part, int *degraded,
4227 char *buf)
4228 {
4229 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4230 * to storage 'destfd' (offset 'destoffsets'), after first
4231 * suspending IO. Then allow resync to continue
4232 * over the suspended section.
4233 * Use part 'part' of the backup-super-block.
4234 */
4235 int odata = disks;
4236 int rv = 0;
4237 int i;
4238 unsigned long long ll;
4239 int new_degraded;
4240 //printf("offset %llu\n", offset);
4241 if (level >= 4)
4242 odata--;
4243 if (level == 6)
4244 odata--;
4245
4246 /* Check that array hasn't become degraded, else we might backup the wrong data */
4247 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4248 return -1; /* FIXME this error is ignored */
4249 new_degraded = (int)ll;
4250 if (new_degraded != *degraded) {
4251 /* check each device to ensure it is still working */
4252 struct mdinfo *sd;
4253 for (sd = sra->devs ; sd ; sd = sd->next) {
4254 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4255 continue;
4256 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4257 char sbuf[100];
4258
4259 if (sysfs_get_str(sra, sd, "state",
4260 sbuf, sizeof(sbuf)) < 0 ||
4261 strstr(sbuf, "faulty") ||
4262 strstr(sbuf, "in_sync") == NULL) {
4263 /* this device is dead */
4264 sd->disk.state = (1<<MD_DISK_FAULTY);
4265 if (sd->disk.raid_disk >= 0 &&
4266 sources[sd->disk.raid_disk] >= 0) {
4267 close(sources[sd->disk.raid_disk]);
4268 sources[sd->disk.raid_disk] = -1;
4269 }
4270 }
4271 }
4272 }
4273 *degraded = new_degraded;
4274 }
4275 if (part) {
4276 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4277 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4278 } else {
4279 bsb.arraystart = __cpu_to_le64(offset * odata);
4280 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4281 }
4282 if (part)
4283 bsb.magic[15] = '2';
4284 for (i = 0; i < dests; i++)
4285 if (part)
4286 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4287 else
4288 lseek64(destfd[i], destoffsets[i], 0);
4289
4290 rv = save_stripes(sources, offsets,
4291 disks, chunk, level, layout,
4292 dests, destfd,
4293 offset*512*odata, stripes * chunk * odata,
4294 buf);
4295
4296 if (rv)
4297 return rv;
4298 bsb.mtime = __cpu_to_le64(time(0));
4299 for (i = 0; i < dests; i++) {
4300 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4301
4302 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4303 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4304 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4305 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4306
4307 rv = -1;
4308 if ((unsigned long long)lseek64(destfd[i],
4309 destoffsets[i] - 4096, 0) !=
4310 destoffsets[i] - 4096)
4311 break;
4312 if (write(destfd[i], &bsb, 512) != 512)
4313 break;
4314 if (destoffsets[i] > 4096) {
4315 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4316 destoffsets[i]+stripes*chunk*odata)
4317 break;
4318 if (write(destfd[i], &bsb, 512) != 512)
4319 break;
4320 }
4321 fsync(destfd[i]);
4322 rv = 0;
4323 }
4324
4325 return rv;
4326 }
4327
4328 /* in 2.6.30, the value reported by sync_completed can be
4329 * less that it should be by one stripe.
4330 * This only happens when reshape hits sync_max and pauses.
4331 * So allow wait_backup to either extent sync_max further
4332 * than strictly necessary, or return before the
4333 * sync has got quite as far as we would really like.
4334 * This is what 'blocks2' is for.
4335 * The various caller give appropriate values so that
4336 * every works.
4337 */
4338 /* FIXME return value is often ignored */
4339 static int forget_backup(int dests, int *destfd,
4340 unsigned long long *destoffsets,
4341 int part)
4342 {
4343 /*
4344 * Erase backup 'part' (which is 0 or 1)
4345 */
4346 int i;
4347 int rv;
4348
4349 if (part) {
4350 bsb.arraystart2 = __cpu_to_le64(0);
4351 bsb.length2 = __cpu_to_le64(0);
4352 } else {
4353 bsb.arraystart = __cpu_to_le64(0);
4354 bsb.length = __cpu_to_le64(0);
4355 }
4356 bsb.mtime = __cpu_to_le64(time(0));
4357 rv = 0;
4358 for (i = 0; i < dests; i++) {
4359 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4360 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4361 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4362 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4363 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4364 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4365 destoffsets[i]-4096)
4366 rv = -1;
4367 if (rv == 0 &&
4368 write(destfd[i], &bsb, 512) != 512)
4369 rv = -1;
4370 fsync(destfd[i]);
4371 }
4372 return rv;
4373 }
4374
4375 static void fail(char *msg)
4376 {
4377 int rv;
4378 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4379 rv |= (write(2, "\n", 1) != 1);
4380 exit(rv ? 1 : 2);
4381 }
4382
4383 static char *abuf, *bbuf;
4384 static unsigned long long abuflen;
4385 static void validate(int afd, int bfd, unsigned long long offset)
4386 {
4387 /* check that the data in the backup against the array.
4388 * This is only used for regression testing and should not
4389 * be used while the array is active
4390 */
4391 if (afd < 0)
4392 return;
4393 lseek64(bfd, offset - 4096, 0);
4394 if (read(bfd, &bsb2, 512) != 512)
4395 fail("cannot read bsb");
4396 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4397 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4398 fail("first csum bad");
4399 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4400 fail("magic is bad");
4401 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4402 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4403 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4404 fail("second csum bad");
4405
4406 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4407 fail("devstart is wrong");
4408
4409 if (bsb2.length) {
4410 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4411
4412 if (abuflen < len) {
4413 free(abuf);
4414 free(bbuf);
4415 abuflen = len;
4416 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4417 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4418 abuflen = 0;
4419 /* just stop validating on mem-alloc failure */
4420 return;
4421 }
4422 }
4423
4424 lseek64(bfd, offset, 0);
4425 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4426 //printf("len %llu\n", len);
4427 fail("read first backup failed");
4428 }
4429 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4430 if ((unsigned long long)read(afd, abuf, len) != len)
4431 fail("read first from array failed");
4432 if (memcmp(bbuf, abuf, len) != 0) {
4433 #if 0
4434 int i;
4435 printf("offset=%llu len=%llu\n",
4436 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4437 for (i=0; i<len; i++)
4438 if (bbuf[i] != abuf[i]) {
4439 printf("first diff byte %d\n", i);
4440 break;
4441 }
4442 #endif
4443 fail("data1 compare failed");
4444 }
4445 }
4446 if (bsb2.length2) {
4447 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4448
4449 if (abuflen < len) {
4450 free(abuf);
4451 free(bbuf);
4452 abuflen = len;
4453 abuf = xmalloc(abuflen);
4454 bbuf = xmalloc(abuflen);
4455 }
4456
4457 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4458 if ((unsigned long long)read(bfd, bbuf, len) != len)
4459 fail("read second backup failed");
4460 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4461 if ((unsigned long long)read(afd, abuf, len) != len)
4462 fail("read second from array failed");
4463 if (memcmp(bbuf, abuf, len) != 0)
4464 fail("data2 compare failed");
4465 }
4466 }
4467
4468 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4469 struct supertype *st, unsigned long blocks,
4470 int *fds, unsigned long long *offsets,
4471 int dests, int *destfd, unsigned long long *destoffsets)
4472 {
4473 /* Monitor a reshape where backup is being performed using
4474 * 'native' mechanism - either to a backup file, or
4475 * to some space in a spare.
4476 */
4477 char *buf;
4478 int degraded = -1;
4479 unsigned long long speed;
4480 unsigned long long suspend_point, array_size;
4481 unsigned long long backup_point, wait_point;
4482 unsigned long long reshape_completed;
4483 int done = 0;
4484 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4485 int part = 0; /* The next part of the backup area to fill. It may already
4486 * be full, so we need to check */
4487 int level = reshape->level;
4488 int layout = reshape->before.layout;
4489 int data = reshape->before.data_disks;
4490 int disks = reshape->before.data_disks + reshape->parity;
4491 int chunk = sra->array.chunk_size;
4492 struct mdinfo *sd;
4493 unsigned long stripes;
4494 int uuid[4];
4495 int frozen = 0;
4496
4497 /* set up the backup-super-block. This requires the
4498 * uuid from the array.
4499 */
4500 /* Find a superblock */
4501 for (sd = sra->devs; sd; sd = sd->next) {
4502 char *dn;
4503 int devfd;
4504 int ok;
4505 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4506 continue;
4507 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4508 devfd = dev_open(dn, O_RDONLY);
4509 if (devfd < 0)
4510 continue;
4511 ok = st->ss->load_super(st, devfd, NULL);
4512 close(devfd);
4513 if (ok == 0)
4514 break;
4515 }
4516 if (!sd) {
4517 pr_err("Cannot find a superblock\n");
4518 return 0;
4519 }
4520
4521 memset(&bsb, 0, 512);
4522 memcpy(bsb.magic, "md_backup_data-1", 16);
4523 st->ss->uuid_from_super(st, uuid);
4524 memcpy(bsb.set_uuid, uuid, 16);
4525 bsb.mtime = __cpu_to_le64(time(0));
4526 bsb.devstart2 = blocks;
4527
4528 stripes = blocks / (sra->array.chunk_size/512) /
4529 reshape->before.data_disks;
4530
4531 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4532 /* Don't start the 'reshape' */
4533 return 0;
4534 if (reshape->before.data_disks == reshape->after.data_disks) {
4535 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4536 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4537 }
4538
4539 if (increasing) {
4540 array_size = sra->component_size * reshape->after.data_disks;
4541 backup_point = sra->reshape_progress;
4542 suspend_point = 0;
4543 } else {
4544 array_size = sra->component_size * reshape->before.data_disks;
4545 backup_point = reshape->backup_blocks;
4546 suspend_point = array_size;
4547 }
4548
4549 while (!done) {
4550 int rv;
4551
4552 /* Want to return as soon the oldest backup slot can
4553 * be released as that allows us to start backing up
4554 * some more, providing suspend_point has been
4555 * advanced, which it should have.
4556 */
4557 if (increasing) {
4558 wait_point = array_size;
4559 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4560 wait_point = (__le64_to_cpu(bsb.arraystart) +
4561 __le64_to_cpu(bsb.length));
4562 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4563 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4564 __le64_to_cpu(bsb.length2));
4565 } else {
4566 wait_point = 0;
4567 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4568 wait_point = __le64_to_cpu(bsb.arraystart);
4569 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4570 wait_point = __le64_to_cpu(bsb.arraystart2);
4571 }
4572
4573 reshape_completed = sra->reshape_progress;
4574 rv = progress_reshape(sra, reshape,
4575 backup_point, wait_point,
4576 &suspend_point, &reshape_completed,
4577 &frozen);
4578 /* external metadata would need to ping_monitor here */
4579 sra->reshape_progress = reshape_completed;
4580
4581 /* Clear any backup region that is before 'here' */
4582 if (increasing) {
4583 if (__le64_to_cpu(bsb.length) > 0 &&
4584 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4585 __le64_to_cpu(bsb.length)))
4586 forget_backup(dests, destfd,
4587 destoffsets, 0);
4588 if (__le64_to_cpu(bsb.length2) > 0 &&
4589 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4590 __le64_to_cpu(bsb.length2)))
4591 forget_backup(dests, destfd,
4592 destoffsets, 1);
4593 } else {
4594 if (__le64_to_cpu(bsb.length) > 0 &&
4595 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4596 forget_backup(dests, destfd,
4597 destoffsets, 0);
4598 if (__le64_to_cpu(bsb.length2) > 0 &&
4599 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4600 forget_backup(dests, destfd,
4601 destoffsets, 1);
4602 }
4603 if (sigterm)
4604 rv = -2;
4605 if (rv < 0) {
4606 if (rv == -1)
4607 done = 1;
4608 break;
4609 }
4610 if (rv == 0 && increasing && !st->ss->external) {
4611 /* No longer need to monitor this reshape */
4612 sysfs_set_str(sra, NULL, "sync_max", "max");
4613 done = 1;
4614 break;
4615 }
4616
4617 while (rv) {
4618 unsigned long long offset;
4619 unsigned long actual_stripes;
4620 /* Need to backup some data.
4621 * If 'part' is not used and the desired
4622 * backup size is suspended, do a backup,
4623 * then consider the next part.
4624 */
4625 /* Check that 'part' is unused */
4626 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4627 break;
4628 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4629 break;
4630
4631 offset = backup_point / data;
4632 actual_stripes = stripes;
4633 if (increasing) {
4634 if (offset + actual_stripes * (chunk/512) >
4635 sra->component_size)
4636 actual_stripes = ((sra->component_size - offset)
4637 / (chunk/512));
4638 if (offset + actual_stripes * (chunk/512) >
4639 suspend_point/data)
4640 break;
4641 } else {
4642 if (offset < actual_stripes * (chunk/512))
4643 actual_stripes = offset / (chunk/512);
4644 offset -= actual_stripes * (chunk/512);
4645 if (offset < suspend_point/data)
4646 break;
4647 }
4648 if (actual_stripes == 0)
4649 break;
4650 grow_backup(sra, offset, actual_stripes,
4651 fds, offsets,
4652 disks, chunk, level, layout,
4653 dests, destfd, destoffsets,
4654 part, &degraded, buf);
4655 validate(afd, destfd[0], destoffsets[0]);
4656 /* record where 'part' is up to */
4657 part = !part;
4658 if (increasing)
4659 backup_point += actual_stripes * (chunk/512) * data;
4660 else
4661 backup_point -= actual_stripes * (chunk/512) * data;
4662 }
4663 }
4664
4665 /* FIXME maybe call progress_reshape one more time instead */
4666 /* remove any remaining suspension */
4667 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4668 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4669 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4670 sysfs_set_num(sra, NULL, "sync_min", 0);
4671
4672 if (reshape->before.data_disks == reshape->after.data_disks)
4673 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4674 free(buf);
4675 return done;
4676 }
4677
4678 /*
4679 * If any spare contains md_back_data-1 which is recent wrt mtime,
4680 * write that data into the array and update the super blocks with
4681 * the new reshape_progress
4682 */
4683 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4684 char *backup_file, int verbose)
4685 {
4686 int i, j;
4687 int old_disks;
4688 unsigned long long *offsets;
4689 unsigned long long nstripe, ostripe;
4690 int ndata, odata;
4691
4692 odata = info->array.raid_disks - info->delta_disks - 1;
4693 if (info->array.level == 6) odata--; /* number of data disks */
4694 ndata = info->array.raid_disks - 1;
4695 if (info->new_level == 6) ndata--;
4696
4697 old_disks = info->array.raid_disks - info->delta_disks;
4698
4699 if (info->delta_disks <= 0)
4700 /* Didn't grow, so the backup file must have
4701 * been used
4702 */
4703 old_disks = cnt;
4704 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4705 struct mdinfo dinfo;
4706 int fd;
4707 int bsbsize;
4708 char *devname, namebuf[20];
4709 unsigned long long lo, hi;
4710
4711 /* This was a spare and may have some saved data on it.
4712 * Load the superblock, find and load the
4713 * backup_super_block.
4714 * If either fail, go on to next device.
4715 * If the backup contains no new info, just return
4716 * else restore data and update all superblocks
4717 */
4718 if (i == old_disks-1) {
4719 fd = open(backup_file, O_RDONLY);
4720 if (fd<0) {
4721 pr_err("backup file %s inaccessible: %s\n",
4722 backup_file, strerror(errno));
4723 continue;
4724 }
4725 devname = backup_file;
4726 } else {
4727 fd = fdlist[i];
4728 if (fd < 0)
4729 continue;
4730 if (st->ss->load_super(st, fd, NULL))
4731 continue;
4732
4733 st->ss->getinfo_super(st, &dinfo, NULL);
4734 st->ss->free_super(st);
4735
4736 if (lseek64(fd,
4737 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4738 0) < 0) {
4739 pr_err("Cannot seek on device %d\n", i);
4740 continue; /* Cannot seek */
4741 }
4742 sprintf(namebuf, "device-%d", i);
4743 devname = namebuf;
4744 }
4745 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4746 if (verbose)
4747 pr_err("Cannot read from %s\n", devname);
4748 continue; /* Cannot read */
4749 }
4750 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4751 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4752 if (verbose)
4753 pr_err("No backup metadata on %s\n", devname);
4754 continue;
4755 }
4756 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4757 if (verbose)
4758 pr_err("Bad backup-metadata checksum on %s\n", devname);
4759 continue; /* bad checksum */
4760 }
4761 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4762 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4763 if (verbose)
4764 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4765 continue; /* Bad second checksum */
4766 }
4767 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4768 if (verbose)
4769 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4770 continue; /* Wrong uuid */
4771 }
4772
4773 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4774 * sometimes they aren't... So allow considerable flexability in matching, and allow
4775 * this test to be overridden by an environment variable.
4776 */
4777 if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) ||
4778 time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) {
4779 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4780 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4781 (unsigned long)__le64_to_cpu(bsb.mtime),
4782 (unsigned long)info->array.utime);
4783 } else {
4784 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4785 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4786 continue; /* time stamp is too bad */
4787 }
4788 }
4789
4790 if (bsb.magic[15] == '1') {
4791 if (bsb.length == 0)
4792 continue;
4793 if (info->delta_disks >= 0) {
4794 /* reshape_progress is increasing */
4795 if (__le64_to_cpu(bsb.arraystart)
4796 + __le64_to_cpu(bsb.length)
4797 < info->reshape_progress) {
4798 nonew:
4799 if (verbose)
4800 pr_err("backup-metadata found on %s but is not needed\n", devname);
4801 continue; /* No new data here */
4802 }
4803 } else {
4804 /* reshape_progress is decreasing */
4805 if (__le64_to_cpu(bsb.arraystart) >=
4806 info->reshape_progress)
4807 goto nonew; /* No new data here */
4808 }
4809 } else {
4810 if (bsb.length == 0 && bsb.length2 == 0)
4811 continue;
4812 if (info->delta_disks >= 0) {
4813 /* reshape_progress is increasing */
4814 if ((__le64_to_cpu(bsb.arraystart)
4815 + __le64_to_cpu(bsb.length)
4816 < info->reshape_progress) &&
4817 (__le64_to_cpu(bsb.arraystart2)
4818 + __le64_to_cpu(bsb.length2)
4819 < info->reshape_progress))
4820 goto nonew; /* No new data here */
4821 } else {
4822 /* reshape_progress is decreasing */
4823 if (__le64_to_cpu(bsb.arraystart) >=
4824 info->reshape_progress &&
4825 __le64_to_cpu(bsb.arraystart2) >=
4826 info->reshape_progress)
4827 goto nonew; /* No new data here */
4828 }
4829 }
4830 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4831 second_fail:
4832 if (verbose)
4833 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4834 devname);
4835 continue; /* Cannot seek */
4836 }
4837 /* There should be a duplicate backup superblock 4k before here */
4838 if (lseek64(fd, -4096, 1) < 0 ||
4839 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4840 goto second_fail; /* Cannot find leading superblock */
4841 if (bsb.magic[15] == '1')
4842 bsbsize = offsetof(struct mdp_backup_super, pad1);
4843 else
4844 bsbsize = offsetof(struct mdp_backup_super, pad);
4845 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4846 goto second_fail; /* Cannot find leading superblock */
4847
4848 /* Now need the data offsets for all devices. */
4849 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4850 for(j=0; j<info->array.raid_disks; j++) {
4851 if (fdlist[j] < 0)
4852 continue;
4853 if (st->ss->load_super(st, fdlist[j], NULL))
4854 /* FIXME should be this be an error */
4855 continue;
4856 st->ss->getinfo_super(st, &dinfo, NULL);
4857 st->ss->free_super(st);
4858 offsets[j] = dinfo.data_offset * 512;
4859 }
4860 printf("%s: restoring critical section\n", Name);
4861
4862 if (restore_stripes(fdlist, offsets,
4863 info->array.raid_disks,
4864 info->new_chunk,
4865 info->new_level,
4866 info->new_layout,
4867 fd, __le64_to_cpu(bsb.devstart)*512,
4868 __le64_to_cpu(bsb.arraystart)*512,
4869 __le64_to_cpu(bsb.length)*512, NULL)) {
4870 /* didn't succeed, so giveup */
4871 if (verbose)
4872 pr_err("Error restoring backup from %s\n",
4873 devname);
4874 free(offsets);
4875 return 1;
4876 }
4877
4878 if (bsb.magic[15] == '2' &&
4879 restore_stripes(fdlist, offsets,
4880 info->array.raid_disks,
4881 info->new_chunk,
4882 info->new_level,
4883 info->new_layout,
4884 fd, __le64_to_cpu(bsb.devstart)*512 +
4885 __le64_to_cpu(bsb.devstart2)*512,
4886 __le64_to_cpu(bsb.arraystart2)*512,
4887 __le64_to_cpu(bsb.length2)*512, NULL)) {
4888 /* didn't succeed, so giveup */
4889 if (verbose)
4890 pr_err("Error restoring second backup from %s\n",
4891 devname);
4892 free(offsets);
4893 return 1;
4894 }
4895
4896 free(offsets);
4897
4898 /* Ok, so the data is restored. Let's update those superblocks. */
4899
4900 lo = hi = 0;
4901 if (bsb.length) {
4902 lo = __le64_to_cpu(bsb.arraystart);
4903 hi = lo + __le64_to_cpu(bsb.length);
4904 }
4905 if (bsb.magic[15] == '2' && bsb.length2) {
4906 unsigned long long lo1, hi1;
4907 lo1 = __le64_to_cpu(bsb.arraystart2);
4908 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4909 if (lo == hi) {
4910 lo = lo1;
4911 hi = hi1;
4912 } else if (lo < lo1)
4913 hi = hi1;
4914 else
4915 lo = lo1;
4916 }
4917 if (lo < hi &&
4918 (info->reshape_progress < lo ||
4919 info->reshape_progress > hi))
4920 /* backup does not affect reshape_progress*/ ;
4921 else if (info->delta_disks >= 0) {
4922 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4923 __le64_to_cpu(bsb.length);
4924 if (bsb.magic[15] == '2') {
4925 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4926 __le64_to_cpu(bsb.length2);
4927 if (p2 > info->reshape_progress)
4928 info->reshape_progress = p2;
4929 }
4930 } else {
4931 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4932 if (bsb.magic[15] == '2') {
4933 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4934 if (p2 < info->reshape_progress)
4935 info->reshape_progress = p2;
4936 }
4937 }
4938 for (j=0; j<info->array.raid_disks; j++) {
4939 if (fdlist[j] < 0)
4940 continue;
4941 if (st->ss->load_super(st, fdlist[j], NULL))
4942 continue;
4943 st->ss->getinfo_super(st, &dinfo, NULL);
4944 dinfo.reshape_progress = info->reshape_progress;
4945 st->ss->update_super(st, &dinfo,
4946 "_reshape_progress",
4947 NULL,0, 0, NULL);
4948 st->ss->store_super(st, fdlist[j]);
4949 st->ss->free_super(st);
4950 }
4951 return 0;
4952 }
4953 /* Didn't find any backup data, try to see if any
4954 * was needed.
4955 */
4956 if (info->delta_disks < 0) {
4957 /* When shrinking, the critical section is at the end.
4958 * So see if we are before the critical section.
4959 */
4960 unsigned long long first_block;
4961 nstripe = ostripe = 0;
4962 first_block = 0;
4963 while (ostripe >= nstripe) {
4964 ostripe += info->array.chunk_size / 512;
4965 first_block = ostripe * odata;
4966 nstripe = first_block / ndata / (info->new_chunk/512) *
4967 (info->new_chunk/512);
4968 }
4969
4970 if (info->reshape_progress >= first_block)
4971 return 0;
4972 }
4973 if (info->delta_disks > 0) {
4974 /* See if we are beyond the critical section. */
4975 unsigned long long last_block;
4976 nstripe = ostripe = 0;
4977 last_block = 0;
4978 while (nstripe >= ostripe) {
4979 nstripe += info->new_chunk / 512;
4980 last_block = nstripe * ndata;
4981 ostripe = last_block / odata / (info->array.chunk_size/512) *
4982 (info->array.chunk_size/512);
4983 }
4984
4985 if (info->reshape_progress >= last_block)
4986 return 0;
4987 }
4988 /* needed to recover critical section! */
4989 if (verbose)
4990 pr_err("Failed to find backup of critical section\n");
4991 return 1;
4992 }
4993
4994 int Grow_continue_command(char *devname, int fd,
4995 char *backup_file, int verbose)
4996 {
4997 int ret_val = 0;
4998 struct supertype *st = NULL;
4999 struct mdinfo *content = NULL;
5000 struct mdinfo array;
5001 char *subarray = NULL;
5002 struct mdinfo *cc = NULL;
5003 struct mdstat_ent *mdstat = NULL;
5004 int cfd = -1;
5005 int fd2;
5006
5007 dprintf("Grow continue from command line called for %s\n",
5008 devname);
5009
5010 st = super_by_fd(fd, &subarray);
5011 if (!st || !st->ss) {
5012 pr_err("Unable to determine metadata format for %s\n",
5013 devname);
5014 return 1;
5015 }
5016 dprintf("Grow continue is run for ");
5017 if (st->ss->external == 0) {
5018 int d;
5019 int cnt = 5;
5020 dprintf_cont("native array (%s)\n", devname);
5021 if (md_get_array_info(fd, &array.array) < 0) {
5022 pr_err("%s is not an active md array - aborting\n",
5023 devname);
5024 ret_val = 1;
5025 goto Grow_continue_command_exit;
5026 }
5027 content = &array;
5028 sysfs_init(content, fd, NULL);
5029 /* Need to load a superblock.
5030 * FIXME we should really get what we need from
5031 * sysfs
5032 */
5033 do {
5034 for (d = 0; d < MAX_DISKS; d++) {
5035 mdu_disk_info_t disk;
5036 char *dv;
5037 int err;
5038 disk.number = d;
5039 if (md_get_disk_info(fd, &disk) < 0)
5040 continue;
5041 if (disk.major == 0 && disk.minor == 0)
5042 continue;
5043 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
5044 continue;
5045 dv = map_dev(disk.major, disk.minor, 1);
5046 if (!dv)
5047 continue;
5048 fd2 = dev_open(dv, O_RDONLY);
5049 if (fd2 < 0)
5050 continue;
5051 err = st->ss->load_super(st, fd2, NULL);
5052 close(fd2);
5053 if (err)
5054 continue;
5055 break;
5056 }
5057 if (d == MAX_DISKS) {
5058 pr_err("Unable to load metadata for %s\n",
5059 devname);
5060 ret_val = 1;
5061 goto Grow_continue_command_exit;
5062 }
5063 st->ss->getinfo_super(st, content, NULL);
5064 if (!content->reshape_active)
5065 sleep(3);
5066 else
5067 break;
5068 } while (cnt-- > 0);
5069 } else {
5070 char *container;
5071
5072 if (subarray) {
5073 dprintf_cont("subarray (%s)\n", subarray);
5074 container = st->container_devnm;
5075 cfd = open_dev_excl(st->container_devnm);
5076 } else {
5077 container = st->devnm;
5078 close(fd);
5079 cfd = open_dev_excl(st->devnm);
5080 dprintf_cont("container (%s)\n", container);
5081 fd = cfd;
5082 }
5083 if (cfd < 0) {
5084 pr_err("Unable to open container for %s\n", devname);
5085 ret_val = 1;
5086 goto Grow_continue_command_exit;
5087 }
5088
5089 /* find in container array under reshape
5090 */
5091 ret_val = st->ss->load_container(st, cfd, NULL);
5092 if (ret_val) {
5093 pr_err("Cannot read superblock for %s\n",
5094 devname);
5095 ret_val = 1;
5096 goto Grow_continue_command_exit;
5097 }
5098
5099 cc = st->ss->container_content(st, subarray);
5100 for (content = cc; content ; content = content->next) {
5101 char *array_name;
5102 int allow_reshape = 1;
5103
5104 if (content->reshape_active == 0)
5105 continue;
5106 /* The decision about array or container wide
5107 * reshape is taken in Grow_continue based
5108 * content->reshape_active state, therefore we
5109 * need to check_reshape based on
5110 * reshape_active and subarray name
5111 */
5112 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
5113 allow_reshape = 0;
5114 if (content->reshape_active == CONTAINER_RESHAPE &&
5115 (content->array.state
5116 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
5117 allow_reshape = 0;
5118
5119 if (!allow_reshape) {
5120 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5121 devname, container);
5122 ret_val = 1;
5123 goto Grow_continue_command_exit;
5124 }
5125
5126 array_name = strchr(content->text_version+1, '/')+1;
5127 mdstat = mdstat_by_subdev(array_name, container);
5128 if (!mdstat)
5129 continue;
5130 if (mdstat->active == 0) {
5131 pr_err("Skipping inactive array %s.\n",
5132 mdstat->devnm);
5133 free_mdstat(mdstat);
5134 mdstat = NULL;
5135 continue;
5136 }
5137 break;
5138 }
5139 if (!content) {
5140 pr_err("Unable to determine reshaped array for %s\n", devname);
5141 ret_val = 1;
5142 goto Grow_continue_command_exit;
5143 }
5144 fd2 = open_dev(mdstat->devnm);
5145 if (fd2 < 0) {
5146 pr_err("cannot open (%s)\n", mdstat->devnm);
5147 ret_val = 1;
5148 goto Grow_continue_command_exit;
5149 }
5150
5151 if (sysfs_init(content, fd2, mdstat->devnm)) {
5152 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5153 mdstat->devnm);
5154 ret_val = 1;
5155 close(fd2);
5156 goto Grow_continue_command_exit;
5157 }
5158
5159 close(fd2);
5160
5161 /* start mdmon in case it is not running
5162 */
5163 if (!mdmon_running(container))
5164 start_mdmon(container);
5165 ping_monitor(container);
5166
5167 if (mdmon_running(container))
5168 st->update_tail = &st->updates;
5169 else {
5170 pr_err("No mdmon found. Grow cannot continue.\n");
5171 ret_val = 1;
5172 goto Grow_continue_command_exit;
5173 }
5174 }
5175
5176 /* verify that array under reshape is started from
5177 * correct position
5178 */
5179 if (verify_reshape_position(content, content->array.level) < 0) {
5180 ret_val = 1;
5181 goto Grow_continue_command_exit;
5182 }
5183
5184 /* continue reshape
5185 */
5186 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
5187
5188 Grow_continue_command_exit:
5189 if (cfd > -1)
5190 close(cfd);
5191 st->ss->free_super(st);
5192 free_mdstat(mdstat);
5193 sysfs_free(cc);
5194 free(subarray);
5195
5196 return ret_val;
5197 }
5198
5199 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
5200 char *backup_file, int forked, int freeze_reshape)
5201 {
5202 int ret_val = 2;
5203
5204 if (!info->reshape_active)
5205 return ret_val;
5206
5207 if (st->ss->external) {
5208 int cfd = open_dev(st->container_devnm);
5209
5210 if (cfd < 0)
5211 return 1;
5212
5213 st->ss->load_container(st, cfd, st->container_devnm);
5214 close(cfd);
5215 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
5216 st, info, 0, backup_file,
5217 0, forked,
5218 1 | info->reshape_active,
5219 freeze_reshape);
5220 } else
5221 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
5222 NULL, INVALID_SECTORS,
5223 backup_file, 0, forked,
5224 1 | info->reshape_active,
5225 freeze_reshape);
5226
5227 return ret_val;
5228 }
5229
5230 char *make_backup(char *name)
5231 {
5232 char *base = "backup_file-";
5233 int len;
5234 char *fname;
5235
5236 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
5237 fname = xmalloc(len);
5238 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
5239 return fname;
5240 }
5241
5242 char *locate_backup(char *name)
5243 {
5244 char *fl = make_backup(name);
5245 struct stat stb;
5246
5247 if (stat(fl, &stb) == 0 &&
5248 S_ISREG(stb.st_mode))
5249 return fl;
5250
5251 free(fl);
5252 return NULL;
5253 }
Unnamed repository; edit this file 'description' to name the repository.