]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
projects / thirdparty / mdadm.git / blob
? search:


f4bd301a193f39d4826fbc4cfa305d01b8ce142e
[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 struct stat stb;
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 fstat(nfd, &stb);
149 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
150 pr_err("%s is not a block device!\n", newdev);
151 close(nfd);
152 free(st);
153 return 1;
154 }
155 /* now check out all the devices and make sure we can read the
156 * superblock */
157 for (d=0 ; d < info.array.raid_disks ; d++) {
158 mdu_disk_info_t disk;
159 char *dv;
160
161 st->ss->free_super(st);
162
163 disk.number = d;
164 if (md_get_disk_info(fd, &disk) < 0) {
165 pr_err("cannot get device detail for device %d\n",
166 d);
167 close(nfd);
168 free(st);
169 return 1;
170 }
171 dv = map_dev(disk.major, disk.minor, 1);
172 if (!dv) {
173 pr_err("cannot find device file for device %d\n",
174 d);
175 close(nfd);
176 free(st);
177 return 1;
178 }
179 fd2 = dev_open(dv, O_RDWR);
180 if (fd2 < 0) {
181 pr_err("cannot open device file %s\n", dv);
182 close(nfd);
183 free(st);
184 return 1;
185 }
186
187 if (st->ss->load_super(st, fd2, NULL)) {
188 pr_err("cannot find super block on %s\n", dv);
189 close(nfd);
190 close(fd2);
191 free(st);
192 return 1;
193 }
194 close(fd2);
195 }
196 /* Ok, looks good. Lets update the superblock and write it out to
197 * newdev.
198 */
199
200 info.disk.number = d;
201 info.disk.major = major(stb.st_rdev);
202 info.disk.minor = minor(stb.st_rdev);
203 info.disk.raid_disk = d;
204 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
205 st->ss->update_super(st, &info, "linear-grow-new", newdev,
206 0, 0, NULL);
207
208 if (st->ss->store_super(st, nfd)) {
209 pr_err("Cannot store new superblock on %s\n",
210 newdev);
211 close(nfd);
212 return 1;
213 }
214 close(nfd);
215
216 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
217 pr_err("Cannot add new disk to this array\n");
218 return 1;
219 }
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
222 */
223
224 if (md_get_array_info(fd, &info.array) < 0) {
225 pr_err("cannot get array info for %s\n", devname);
226 return 1;
227 }
228
229 nd = d;
230 for (d=0 ; d < info.array.raid_disks ; d++) {
231 mdu_disk_info_t disk;
232 char *dv;
233
234 disk.number = d;
235 if (md_get_disk_info(fd, &disk) < 0) {
236 pr_err("cannot get device detail for device %d\n",
237 d);
238 return 1;
239 }
240 dv = map_dev(disk.major, disk.minor, 1);
241 if (!dv) {
242 pr_err("cannot find device file for device %d\n",
243 d);
244 return 1;
245 }
246 fd2 = dev_open(dv, O_RDWR);
247 if (fd2 < 0) {
248 pr_err("cannot open device file %s\n", dv);
249 return 1;
250 }
251 if (st->ss->load_super(st, fd2, NULL)) {
252 pr_err("cannot find super block on %s\n", dv);
253 close(fd);
254 return 1;
255 }
256 info.array.raid_disks = nd+1;
257 info.array.nr_disks = nd+1;
258 info.array.active_disks = nd+1;
259 info.array.working_disks = nd+1;
260
261 st->ss->update_super(st, &info, "linear-grow-update", dv,
262 0, 0, NULL);
263
264 if (st->ss->store_super(st, fd2)) {
265 pr_err("Cannot store new superblock on %s\n", dv);
266 close(fd2);
267 return 1;
268 }
269 close(fd2);
270 }
271
272 return 0;
273 }
274
275 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
276 {
277 /*
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
280 * Then add it
281 *
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
284 * to all devices
285 */
286 mdu_bitmap_file_t bmf;
287 mdu_array_info_t array;
288 struct supertype *st;
289 char *subarray = NULL;
290 int major = BITMAP_MAJOR_HI;
291 unsigned long long bitmapsize, array_size;
292 struct mdinfo *mdi;
293
294 /*
295 * We only ever get called if s->bitmap_file is != NULL, so this check
296 * is just here to quiet down static code checkers.
297 */
298 if (!s->bitmap_file)
299 return 1;
300
301 if (strcmp(s->bitmap_file, "clustered") == 0)
302 major = BITMAP_MAJOR_CLUSTERED;
303
304 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
305 if (errno == ENOMEM)
306 pr_err("Memory allocation failure.\n");
307 else
308 pr_err("bitmaps not supported by this kernel.\n");
309 return 1;
310 }
311 if (bmf.pathname[0]) {
312 if (strcmp(s->bitmap_file,"none") == 0) {
313 if (ioctl(fd, SET_BITMAP_FILE, -1) != 0) {
314 pr_err("failed to remove bitmap %s\n",
315 bmf.pathname);
316 return 1;
317 }
318 return 0;
319 }
320 pr_err("%s already has a bitmap (%s)\n",
321 devname, bmf.pathname);
322 return 1;
323 }
324 if (md_get_array_info(fd, &array) != 0) {
325 pr_err("cannot get array status for %s\n", devname);
326 return 1;
327 }
328 if (array.state & (1 << MD_SB_BITMAP_PRESENT)) {
329 if (strcmp(s->bitmap_file, "none")==0) {
330 array.state &= ~(1 << MD_SB_BITMAP_PRESENT);
331 if (md_set_array_info(fd, &array) != 0) {
332 if (array.state & (1 << MD_SB_CLUSTERED))
333 pr_err("failed to remove clustered bitmap.\n");
334 else
335 pr_err("failed to remove internal bitmap.\n");
336 return 1;
337 }
338 return 0;
339 }
340 pr_err("bitmap already present on %s\n", devname);
341 return 1;
342 }
343
344 if (strcmp(s->bitmap_file, "none") == 0) {
345 pr_err("no bitmap found on %s\n", devname);
346 return 1;
347 }
348 if (array.level <= 0) {
349 pr_err("Bitmaps not meaningful with level %s\n",
350 map_num(pers, array.level)?:"of this array");
351 return 1;
352 }
353 bitmapsize = array.size;
354 bitmapsize <<= 1;
355 if (get_dev_size(fd, NULL, &array_size) &&
356 array_size > (0x7fffffffULL << 9)) {
357 /* Array is big enough that we cannot trust array.size
358 * try other approaches
359 */
360 bitmapsize = get_component_size(fd);
361 }
362 if (bitmapsize == 0) {
363 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
364 return 1;
365 }
366
367 if (array.level == 10) {
368 int ncopies;
369
370 ncopies = (array.layout & 255) * ((array.layout >> 8) & 255);
371 bitmapsize = bitmapsize * array.raid_disks / ncopies;
372 }
373
374 st = super_by_fd(fd, &subarray);
375 if (!st) {
376 pr_err("Cannot understand version %d.%d\n",
377 array.major_version, array.minor_version);
378 return 1;
379 }
380 if (subarray) {
381 pr_err("Cannot add bitmaps to sub-arrays yet\n");
382 free(subarray);
383 free(st);
384 return 1;
385 }
386
387 mdi = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY);
388 if (mdi) {
389 if (mdi->consistency_policy == CONSISTENCY_POLICY_PPL) {
390 pr_err("Cannot add bitmap to array with PPL\n");
391 free(mdi);
392 free(st);
393 return 1;
394 }
395 free(mdi);
396 }
397
398 if (strcmp(s->bitmap_file, "internal") == 0 ||
399 strcmp(s->bitmap_file, "clustered") == 0) {
400 int rv;
401 int d;
402 int offset_setable = 0;
403 if (st->ss->add_internal_bitmap == NULL) {
404 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
405 return 1;
406 }
407 st->nodes = c->nodes;
408 st->cluster_name = c->homecluster;
409 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
410 if (mdi)
411 offset_setable = 1;
412 for (d = 0; d < st->max_devs; d++) {
413 mdu_disk_info_t disk;
414 char *dv;
415 int fd2;
416
417 disk.number = d;
418 if (md_get_disk_info(fd, &disk) < 0)
419 continue;
420 if (disk.major == 0 && disk.minor == 0)
421 continue;
422 if ((disk.state & (1 << MD_DISK_SYNC)) == 0)
423 continue;
424 dv = map_dev(disk.major, disk.minor, 1);
425 if (!dv)
426 continue;
427 fd2 = dev_open(dv, O_RDWR);
428 if (fd2 < 0)
429 continue;
430 rv = st->ss->load_super(st, fd2, NULL);
431 if (!rv) {
432 rv = st->ss->add_internal_bitmap(
433 st, &s->bitmap_chunk, c->delay,
434 s->write_behind, bitmapsize,
435 offset_setable, major);
436 if (!rv) {
437 st->ss->write_bitmap(st, fd2,
438 NodeNumUpdate);
439 } else {
440 pr_err("failed to create internal bitmap - chunksize problem.\n");
441 }
442 } else {
443 pr_err("failed to load super-block.\n");
444 }
445 close(fd2);
446 if (rv)
447 return 1;
448 }
449 if (offset_setable) {
450 st->ss->getinfo_super(st, mdi, NULL);
451 if (sysfs_init(mdi, fd, NULL)) {
452 pr_err("failed to intialize sysfs.\n");
453 free(mdi);
454 }
455 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
456 mdi->bitmap_offset);
457 free(mdi);
458 } else {
459 if (strcmp(s->bitmap_file, "clustered") == 0)
460 array.state |= (1 << MD_SB_CLUSTERED);
461 array.state |= (1 << MD_SB_BITMAP_PRESENT);
462 rv = md_set_array_info(fd, &array);
463 }
464 if (rv < 0) {
465 if (errno == EBUSY)
466 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
467 pr_err("failed to set internal bitmap.\n");
468 return 1;
469 }
470 } else {
471 int uuid[4];
472 int bitmap_fd;
473 int d;
474 int max_devs = st->max_devs;
475
476 /* try to load a superblock */
477 for (d = 0; d < max_devs; d++) {
478 mdu_disk_info_t disk;
479 char *dv;
480 int fd2;
481 disk.number = d;
482 if (md_get_disk_info(fd, &disk) < 0)
483 continue;
484 if ((disk.major==0 && disk.minor == 0) ||
485 (disk.state & (1 << MD_DISK_REMOVED)))
486 continue;
487 dv = map_dev(disk.major, disk.minor, 1);
488 if (!dv)
489 continue;
490 fd2 = dev_open(dv, O_RDONLY);
491 if (fd2 >= 0) {
492 if (st->ss->load_super(st, fd2, NULL) == 0) {
493 close(fd2);
494 st->ss->uuid_from_super(st, uuid);
495 break;
496 }
497 close(fd2);
498 }
499 }
500 if (d == max_devs) {
501 pr_err("cannot find UUID for array!\n");
502 return 1;
503 }
504 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid,
505 s->bitmap_chunk, c->delay, s->write_behind,
506 bitmapsize, major)) {
507 return 1;
508 }
509 bitmap_fd = open(s->bitmap_file, O_RDWR);
510 if (bitmap_fd < 0) {
511 pr_err("weird: %s cannot be opened\n", s->bitmap_file);
512 return 1;
513 }
514 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
515 int err = errno;
516 if (errno == EBUSY)
517 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
518 pr_err("Cannot set bitmap file for %s: %s\n",
519 devname, strerror(err));
520 return 1;
521 }
522 }
523
524 return 0;
525 }
526
527 int Grow_consistency_policy(char *devname, int fd, struct context *c, struct shape *s)
528 {
529 struct supertype *st;
530 struct mdinfo *sra;
531 struct mdinfo *sd;
532 char *subarray = NULL;
533 int ret = 0;
534 char container_dev[PATH_MAX];
535
536 if (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
537 s->consistency_policy != CONSISTENCY_POLICY_PPL) {
538 pr_err("Operation not supported for consistency policy %s\n",
539 map_num(consistency_policies, s->consistency_policy));
540 return 1;
541 }
542
543 st = super_by_fd(fd, &subarray);
544 if (!st)
545 return 1;
546
547 sra = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY|GET_LEVEL|
548 GET_DEVS|GET_STATE);
549 if (!sra) {
550 ret = 1;
551 goto free_st;
552 }
553
554 if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
555 !st->ss->write_init_ppl) {
556 pr_err("%s metadata does not support PPL\n", st->ss->name);
557 ret = 1;
558 goto free_info;
559 }
560
561 if (sra->array.level != 5) {
562 pr_err("Operation not supported for array level %d\n",
563 sra->array.level);
564 ret = 1;
565 goto free_info;
566 }
567
568 if (sra->consistency_policy == (unsigned)s->consistency_policy) {
569 pr_err("Consistency policy is already %s\n",
570 map_num(consistency_policies, s->consistency_policy));
571 ret = 1;
572 goto free_info;
573 } else if (sra->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
574 sra->consistency_policy != CONSISTENCY_POLICY_PPL) {
575 pr_err("Current consistency policy is %s, cannot change to %s\n",
576 map_num(consistency_policies, sra->consistency_policy),
577 map_num(consistency_policies, s->consistency_policy));
578 ret = 1;
579 goto free_info;
580 }
581
582 if (subarray) {
583 char *update;
584
585 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
586 update = "ppl";
587 else
588 update = "no-ppl";
589
590 sprintf(container_dev, "/dev/%s", st->container_devnm);
591
592 ret = Update_subarray(container_dev, subarray, update, NULL,
593 c->verbose);
594 if (ret)
595 goto free_info;
596 }
597
598 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
599 struct mdinfo info;
600
601 if (subarray) {
602 struct mdinfo *mdi;
603 int cfd;
604
605 cfd = open(container_dev, O_RDWR|O_EXCL);
606 if (cfd < 0) {
607 pr_err("Failed to open %s\n", container_dev);
608 ret = 1;
609 goto free_info;
610 }
611
612 ret = st->ss->load_container(st, cfd, st->container_devnm);
613 close(cfd);
614
615 if (ret) {
616 pr_err("Cannot read superblock for %s\n",
617 container_dev);
618 goto free_info;
619 }
620
621 mdi = st->ss->container_content(st, subarray);
622 info = *mdi;
623 free(mdi);
624 }
625
626 for (sd = sra->devs; sd; sd = sd->next) {
627 int dfd;
628 char *devpath;
629
630 if ((sd->disk.state & (1 << MD_DISK_SYNC)) == 0)
631 continue;
632
633 devpath = map_dev(sd->disk.major, sd->disk.minor, 0);
634 dfd = dev_open(devpath, O_RDWR);
635 if (dfd < 0) {
636 pr_err("Failed to open %s\n", devpath);
637 ret = 1;
638 goto free_info;
639 }
640
641 if (!subarray) {
642 ret = st->ss->load_super(st, dfd, NULL);
643 if (ret) {
644 pr_err("Failed to load super-block.\n");
645 close(dfd);
646 goto free_info;
647 }
648
649 ret = st->ss->update_super(st, sra, "ppl", devname,
650 c->verbose, 0, NULL);
651 if (ret) {
652 close(dfd);
653 st->ss->free_super(st);
654 goto free_info;
655 }
656 st->ss->getinfo_super(st, &info, NULL);
657 }
658
659 ret |= sysfs_set_num(sra, sd, "ppl_sector", info.ppl_sector);
660 ret |= sysfs_set_num(sra, sd, "ppl_size", info.ppl_size);
661
662 if (ret) {
663 pr_err("Failed to set PPL attributes for %s\n",
664 sd->sys_name);
665 close(dfd);
666 st->ss->free_super(st);
667 goto free_info;
668 }
669
670 ret = st->ss->write_init_ppl(st, &info, dfd);
671 if (ret)
672 pr_err("Failed to write PPL\n");
673
674 close(dfd);
675
676 if (!subarray)
677 st->ss->free_super(st);
678
679 if (ret)
680 goto free_info;
681 }
682 }
683
684 ret = sysfs_set_str(sra, NULL, "consistency_policy",
685 map_num(consistency_policies,
686 s->consistency_policy));
687 if (ret)
688 pr_err("Failed to change array consistency policy\n");
689
690 free_info:
691 sysfs_free(sra);
692 free_st:
693 free(st);
694 free(subarray);
695
696 return ret;
697 }
698
699 /*
700 * When reshaping an array we might need to backup some data.
701 * This is written to all spares with a 'super_block' describing it.
702 * The superblock goes 4K from the end of the used space on the
703 * device.
704 * It if written after the backup is complete.
705 * It has the following structure.
706 */
707
708 static struct mdp_backup_super {
709 char magic[16]; /* md_backup_data-1 or -2 */
710 __u8 set_uuid[16];
711 __u64 mtime;
712 /* start/sizes in 512byte sectors */
713 __u64 devstart; /* address on backup device/file of data */
714 __u64 arraystart;
715 __u64 length;
716 __u32 sb_csum; /* csum of preceeding bytes. */
717 __u32 pad1;
718 __u64 devstart2; /* offset in to data of second section */
719 __u64 arraystart2;
720 __u64 length2;
721 __u32 sb_csum2; /* csum of preceeding bytes. */
722 __u8 pad[512-68-32];
723 } __attribute__((aligned(512))) bsb, bsb2;
724
725 static __u32 bsb_csum(char *buf, int len)
726 {
727 int i;
728 int csum = 0;
729 for (i = 0; i < len; i++)
730 csum = (csum<<3) + buf[0];
731 return __cpu_to_le32(csum);
732 }
733
734 static int check_idle(struct supertype *st)
735 {
736 /* Check that all member arrays for this container, or the
737 * container of this array, are idle
738 */
739 char *container = (st->container_devnm[0]
740 ? st->container_devnm : st->devnm);
741 struct mdstat_ent *ent, *e;
742 int is_idle = 1;
743
744 ent = mdstat_read(0, 0);
745 for (e = ent ; e; e = e->next) {
746 if (!is_container_member(e, container))
747 continue;
748 if (e->percent >= 0) {
749 is_idle = 0;
750 break;
751 }
752 }
753 free_mdstat(ent);
754 return is_idle;
755 }
756
757 static int freeze_container(struct supertype *st)
758 {
759 char *container = (st->container_devnm[0]
760 ? st->container_devnm : st->devnm);
761
762 if (!check_idle(st))
763 return -1;
764
765 if (block_monitor(container, 1)) {
766 pr_err("failed to freeze container\n");
767 return -2;
768 }
769
770 return 1;
771 }
772
773 static void unfreeze_container(struct supertype *st)
774 {
775 char *container = (st->container_devnm[0]
776 ? st->container_devnm : st->devnm);
777
778 unblock_monitor(container, 1);
779 }
780
781 static int freeze(struct supertype *st)
782 {
783 /* Try to freeze resync/rebuild on this array/container.
784 * Return -1 if the array is busy,
785 * return -2 container cannot be frozen,
786 * return 0 if this kernel doesn't support 'frozen'
787 * return 1 if it worked.
788 */
789 if (st->ss->external)
790 return freeze_container(st);
791 else {
792 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
793 int err;
794 char buf[20];
795
796 if (!sra)
797 return -1;
798 /* Need to clear any 'read-auto' status */
799 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
800 strncmp(buf, "read-auto", 9) == 0)
801 sysfs_set_str(sra, NULL, "array_state", "clean");
802
803 err = sysfs_freeze_array(sra);
804 sysfs_free(sra);
805 return err;
806 }
807 }
808
809 static void unfreeze(struct supertype *st)
810 {
811 if (st->ss->external)
812 return unfreeze_container(st);
813 else {
814 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
815 char buf[20];
816
817 if (sra &&
818 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
819 && strcmp(buf, "frozen\n") == 0)
820 sysfs_set_str(sra, NULL, "sync_action", "idle");
821 sysfs_free(sra);
822 }
823 }
824
825 static void wait_reshape(struct mdinfo *sra)
826 {
827 int fd = sysfs_get_fd(sra, NULL, "sync_action");
828 char action[20];
829
830 if (fd < 0)
831 return;
832
833 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
834 strncmp(action, "reshape", 7) == 0)
835 sysfs_wait(fd, NULL);
836 close(fd);
837 }
838
839 static int reshape_super(struct supertype *st, unsigned long long size,
840 int level, int layout, int chunksize, int raid_disks,
841 int delta_disks, char *backup_file, char *dev,
842 int direction, int verbose)
843 {
844 /* nothing extra to check in the native case */
845 if (!st->ss->external)
846 return 0;
847 if (!st->ss->reshape_super ||
848 !st->ss->manage_reshape) {
849 pr_err("%s metadata does not support reshape\n",
850 st->ss->name);
851 return 1;
852 }
853
854 return st->ss->reshape_super(st, size, level, layout, chunksize,
855 raid_disks, delta_disks, backup_file, dev,
856 direction, verbose);
857 }
858
859 static void sync_metadata(struct supertype *st)
860 {
861 if (st->ss->external) {
862 if (st->update_tail) {
863 flush_metadata_updates(st);
864 st->update_tail = &st->updates;
865 } else
866 st->ss->sync_metadata(st);
867 }
868 }
869
870 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
871 {
872 /* when dealing with external metadata subarrays we need to be
873 * prepared to handle EAGAIN. The kernel may need to wait for
874 * mdmon to mark the array active so the kernel can handle
875 * allocations/writeback when preparing the reshape action
876 * (md_allow_write()). We temporarily disable safe_mode_delay
877 * to close a race with the array_state going clean before the
878 * next write to raid_disks / stripe_cache_size
879 */
880 char safe[50];
881 int rc;
882
883 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
884 if (!container ||
885 (strcmp(name, "raid_disks") != 0 &&
886 strcmp(name, "stripe_cache_size") != 0))
887 return sysfs_set_num(sra, NULL, name, n);
888
889 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
890 if (rc <= 0)
891 return -1;
892 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
893 rc = sysfs_set_num(sra, NULL, name, n);
894 if (rc < 0 && errno == EAGAIN) {
895 ping_monitor(container);
896 /* if we get EAGAIN here then the monitor is not active
897 * so stop trying
898 */
899 rc = sysfs_set_num(sra, NULL, name, n);
900 }
901 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
902 return rc;
903 }
904
905 int start_reshape(struct mdinfo *sra, int already_running,
906 int before_data_disks, int data_disks)
907 {
908 int err;
909 unsigned long long sync_max_to_set;
910
911 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
912 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
913 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
914 sra->reshape_progress);
915 if (before_data_disks <= data_disks)
916 sync_max_to_set = sra->reshape_progress / data_disks;
917 else
918 sync_max_to_set = (sra->component_size * data_disks
919 - sra->reshape_progress) / data_disks;
920 if (!already_running)
921 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
922 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
923 if (!already_running && err == 0) {
924 int cnt = 5;
925 do {
926 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
927 if (err)
928 sleep(1);
929 } while (err && errno == EBUSY && cnt-- > 0);
930 }
931 return err;
932 }
933
934 void abort_reshape(struct mdinfo *sra)
935 {
936 sysfs_set_str(sra, NULL, "sync_action", "idle");
937 /*
938 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
939 * suspend_hi to decrease as well as increase.")
940 * you could only increase suspend_{lo,hi} unless the region they
941 * covered was empty. So to reset to 0, you need to push suspend_lo
942 * up past suspend_hi first. So to maximize the chance of mdadm
943 * working on all kernels, we want to keep doing that.
944 */
945 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
946 sysfs_set_num(sra, NULL, "suspend_hi", 0);
947 sysfs_set_num(sra, NULL, "suspend_lo", 0);
948 sysfs_set_num(sra, NULL, "sync_min", 0);
949 // It isn't safe to reset sync_max as we aren't monitoring.
950 // Array really should be stopped at this point.
951 }
952
953 int remove_disks_for_takeover(struct supertype *st,
954 struct mdinfo *sra,
955 int layout)
956 {
957 int nr_of_copies;
958 struct mdinfo *remaining;
959 int slot;
960
961 if (st->ss->external) {
962 int rv = 0;
963 struct mdinfo *arrays = st->ss->container_content(st, NULL);
964 /*
965 * containter_content returns list of arrays in container
966 * If arrays->next is not NULL it means that there are
967 * 2 arrays in container and operation should be blocked
968 */
969 if (arrays) {
970 if (arrays->next)
971 rv = 1;
972 sysfs_free(arrays);
973 if (rv) {
974 pr_err("Error. Cannot perform operation on /dev/%s\n", st->devnm);
975 pr_err("For this operation it MUST be single array in container\n");
976 return rv;
977 }
978 }
979 }
980
981 if (sra->array.level == 10)
982 nr_of_copies = layout & 0xff;
983 else if (sra->array.level == 1)
984 nr_of_copies = sra->array.raid_disks;
985 else
986 return 1;
987
988 remaining = sra->devs;
989 sra->devs = NULL;
990 /* for each 'copy', select one device and remove from the list. */
991 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
992 struct mdinfo **diskp;
993 int found = 0;
994
995 /* Find a working device to keep */
996 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
997 struct mdinfo *disk = *diskp;
998
999 if (disk->disk.raid_disk < slot)
1000 continue;
1001 if (disk->disk.raid_disk >= slot + nr_of_copies)
1002 continue;
1003 if (disk->disk.state & (1<<MD_DISK_REMOVED))
1004 continue;
1005 if (disk->disk.state & (1<<MD_DISK_FAULTY))
1006 continue;
1007 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
1008 continue;
1009
1010 /* We have found a good disk to use! */
1011 *diskp = disk->next;
1012 disk->next = sra->devs;
1013 sra->devs = disk;
1014 found = 1;
1015 break;
1016 }
1017 if (!found)
1018 break;
1019 }
1020
1021 if (slot < sra->array.raid_disks) {
1022 /* didn't find all slots */
1023 struct mdinfo **e;
1024 e = &remaining;
1025 while (*e)
1026 e = &(*e)->next;
1027 *e = sra->devs;
1028 sra->devs = remaining;
1029 return 1;
1030 }
1031
1032 /* Remove all 'remaining' devices from the array */
1033 while (remaining) {
1034 struct mdinfo *sd = remaining;
1035 remaining = sd->next;
1036
1037 sysfs_set_str(sra, sd, "state", "faulty");
1038 sysfs_set_str(sra, sd, "slot", "none");
1039 /* for external metadata disks should be removed in mdmon */
1040 if (!st->ss->external)
1041 sysfs_set_str(sra, sd, "state", "remove");
1042 sd->disk.state |= (1<<MD_DISK_REMOVED);
1043 sd->disk.state &= ~(1<<MD_DISK_SYNC);
1044 sd->next = sra->devs;
1045 sra->devs = sd;
1046 }
1047 return 0;
1048 }
1049
1050 void reshape_free_fdlist(int *fdlist,
1051 unsigned long long *offsets,
1052 int size)
1053 {
1054 int i;
1055
1056 for (i = 0; i < size; i++)
1057 if (fdlist[i] >= 0)
1058 close(fdlist[i]);
1059
1060 free(fdlist);
1061 free(offsets);
1062 }
1063
1064 int reshape_prepare_fdlist(char *devname,
1065 struct mdinfo *sra,
1066 int raid_disks,
1067 int nrdisks,
1068 unsigned long blocks,
1069 char *backup_file,
1070 int *fdlist,
1071 unsigned long long *offsets)
1072 {
1073 int d = 0;
1074 struct mdinfo *sd;
1075
1076 enable_fds(nrdisks);
1077 for (d = 0; d <= nrdisks; d++)
1078 fdlist[d] = -1;
1079 d = raid_disks;
1080 for (sd = sra->devs; sd; sd = sd->next) {
1081 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1082 continue;
1083 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
1084 sd->disk.raid_disk < raid_disks) {
1085 char *dn = map_dev(sd->disk.major,
1086 sd->disk.minor, 1);
1087 fdlist[sd->disk.raid_disk]
1088 = dev_open(dn, O_RDONLY);
1089 offsets[sd->disk.raid_disk] = sd->data_offset*512;
1090 if (fdlist[sd->disk.raid_disk] < 0) {
1091 pr_err("%s: cannot open component %s\n",
1092 devname, dn ? dn : "-unknown-");
1093 d = -1;
1094 goto release;
1095 }
1096 } else if (backup_file == NULL) {
1097 /* spare */
1098 char *dn = map_dev(sd->disk.major,
1099 sd->disk.minor, 1);
1100 fdlist[d] = dev_open(dn, O_RDWR);
1101 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
1102 if (fdlist[d] < 0) {
1103 pr_err("%s: cannot open component %s\n",
1104 devname, dn ? dn : "-unknown-");
1105 d = -1;
1106 goto release;
1107 }
1108 d++;
1109 }
1110 }
1111 release:
1112 return d;
1113 }
1114
1115 int reshape_open_backup_file(char *backup_file,
1116 int fd,
1117 char *devname,
1118 long blocks,
1119 int *fdlist,
1120 unsigned long long *offsets,
1121 char *sys_name,
1122 int restart)
1123 {
1124 /* Return 1 on success, 0 on any form of failure */
1125 /* need to check backup file is large enough */
1126 char buf[512];
1127 struct stat stb;
1128 unsigned int dev;
1129 int i;
1130
1131 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
1132 S_IRUSR | S_IWUSR);
1133 *offsets = 8 * 512;
1134 if (*fdlist < 0) {
1135 pr_err("%s: cannot create backup file %s: %s\n",
1136 devname, backup_file, strerror(errno));
1137 return 0;
1138 }
1139 /* Guard against backup file being on array device.
1140 * If array is partitioned or if LVM etc is in the
1141 * way this will not notice, but it is better than
1142 * nothing.
1143 */
1144 fstat(*fdlist, &stb);
1145 dev = stb.st_dev;
1146 fstat(fd, &stb);
1147 if (stb.st_rdev == dev) {
1148 pr_err("backup file must NOT be on the array being reshaped.\n");
1149 close(*fdlist);
1150 return 0;
1151 }
1152
1153 memset(buf, 0, 512);
1154 for (i=0; i < blocks + 8 ; i++) {
1155 if (write(*fdlist, buf, 512) != 512) {
1156 pr_err("%s: cannot create backup file %s: %s\n",
1157 devname, backup_file, strerror(errno));
1158 return 0;
1159 }
1160 }
1161 if (fsync(*fdlist) != 0) {
1162 pr_err("%s: cannot create backup file %s: %s\n",
1163 devname, backup_file, strerror(errno));
1164 return 0;
1165 }
1166
1167 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
1168 char *bu = make_backup(sys_name);
1169 if (symlink(backup_file, bu))
1170 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
1171 strerror(errno));
1172 free(bu);
1173 }
1174
1175 return 1;
1176 }
1177
1178 unsigned long compute_backup_blocks(int nchunk, int ochunk,
1179 unsigned int ndata, unsigned int odata)
1180 {
1181 unsigned long a, b, blocks;
1182 /* So how much do we need to backup.
1183 * We need an amount of data which is both a whole number of
1184 * old stripes and a whole number of new stripes.
1185 * So LCM for (chunksize*datadisks).
1186 */
1187 a = (ochunk/512) * odata;
1188 b = (nchunk/512) * ndata;
1189 /* Find GCD */
1190 a = GCD(a, b);
1191 /* LCM == product / GCD */
1192 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
1193
1194 return blocks;
1195 }
1196
1197 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
1198 {
1199 /* Based on the current array state in info->array and
1200 * the changes in info->new_* etc, determine:
1201 * - whether the change is possible
1202 * - Intermediate level/raid_disks/layout
1203 * - whether a restriping reshape is needed
1204 * - number of sectors in minimum change unit. This
1205 * will cover a whole number of stripes in 'before' and
1206 * 'after'.
1207 *
1208 * Return message if the change should be rejected
1209 * NULL if the change can be achieved
1210 *
1211 * This can be called as part of starting a reshape, or
1212 * when assembling an array that is undergoing reshape.
1213 */
1214 int near, far, offset, copies;
1215 int new_disks;
1216 int old_chunk, new_chunk;
1217 /* delta_parity records change in number of devices
1218 * caused by level change
1219 */
1220 int delta_parity = 0;
1221
1222 memset(re, 0, sizeof(*re));
1223
1224 /* If a new level not explicitly given, we assume no-change */
1225 if (info->new_level == UnSet)
1226 info->new_level = info->array.level;
1227
1228 if (info->new_chunk)
1229 switch (info->new_level) {
1230 case 0:
1231 case 4:
1232 case 5:
1233 case 6:
1234 case 10:
1235 /* chunk size is meaningful, must divide component_size
1236 * evenly
1237 */
1238 if (info->component_size % (info->new_chunk/512)) {
1239 unsigned long long shrink = info->component_size;
1240 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1241 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1242 info->new_chunk/1024, info->component_size/2);
1243 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1244 devname, shrink/2);
1245 pr_err("will shrink the array so the given chunk size would work.\n");
1246 return "";
1247 }
1248 break;
1249 default:
1250 return "chunk size not meaningful for this level";
1251 }
1252 else
1253 info->new_chunk = info->array.chunk_size;
1254
1255 switch (info->array.level) {
1256 default:
1257 return "No reshape is possibly for this RAID level";
1258 case LEVEL_LINEAR:
1259 if (info->delta_disks != UnSet)
1260 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1261 else
1262 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1263 case 1:
1264 /* RAID1 can convert to RAID1 with different disks, or
1265 * raid5 with 2 disks, or
1266 * raid0 with 1 disk
1267 */
1268 if (info->new_level > 1 && (info->component_size & 7))
1269 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1270 if (info->new_level == 0) {
1271 if (info->delta_disks != UnSet &&
1272 info->delta_disks != 0)
1273 return "Cannot change number of disks with RAID1->RAID0 conversion";
1274 re->level = 0;
1275 re->before.data_disks = 1;
1276 re->after.data_disks = 1;
1277 return NULL;
1278 }
1279 if (info->new_level == 1) {
1280 if (info->delta_disks == UnSet)
1281 /* Don't know what to do */
1282 return "no change requested for Growing RAID1";
1283 re->level = 1;
1284 return NULL;
1285 }
1286 if (info->array.raid_disks != 2 && info->new_level == 5)
1287 return "Can only convert a 2-device array to RAID5";
1288 if (info->array.raid_disks == 2 && info->new_level == 5) {
1289 re->level = 5;
1290 re->before.data_disks = 1;
1291 if (info->delta_disks != UnSet &&
1292 info->delta_disks != 0)
1293 re->after.data_disks = 1 + info->delta_disks;
1294 else
1295 re->after.data_disks = 1;
1296 if (re->after.data_disks < 1)
1297 return "Number of disks too small for RAID5";
1298
1299 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1300 info->array.chunk_size = 65536;
1301 break;
1302 }
1303 /* Could do some multi-stage conversions, but leave that to
1304 * later.
1305 */
1306 return "Impossibly level change request for RAID1";
1307
1308 case 10:
1309 /* RAID10 can be converted from near mode to
1310 * RAID0 by removing some devices.
1311 * It can also be reshaped if the kernel supports
1312 * new_data_offset.
1313 */
1314 switch (info->new_level) {
1315 case 0:
1316 if ((info->array.layout & ~0xff) != 0x100)
1317 return "Cannot Grow RAID10 with far/offset layout";
1318 /* number of devices must be multiple of number of copies */
1319 if (info->array.raid_disks % (info->array.layout & 0xff))
1320 return "RAID10 layout too complex for Grow operation";
1321
1322 new_disks = (info->array.raid_disks
1323 / (info->array.layout & 0xff));
1324 if (info->delta_disks == UnSet)
1325 info->delta_disks = (new_disks
1326 - info->array.raid_disks);
1327
1328 if (info->delta_disks != new_disks - info->array.raid_disks)
1329 return "New number of raid-devices impossible for RAID10";
1330 if (info->new_chunk &&
1331 info->new_chunk != info->array.chunk_size)
1332 return "Cannot change chunk-size with RAID10 Grow";
1333
1334 /* looks good */
1335 re->level = 0;
1336 re->before.data_disks = new_disks;
1337 re->after.data_disks = re->before.data_disks;
1338 return NULL;
1339
1340 case 10:
1341 near = info->array.layout & 0xff;
1342 far = (info->array.layout >> 8) & 0xff;
1343 offset = info->array.layout & 0x10000;
1344 if (far > 1 && !offset)
1345 return "Cannot reshape RAID10 in far-mode";
1346 copies = near * far;
1347
1348 old_chunk = info->array.chunk_size * far;
1349
1350 if (info->new_layout == UnSet)
1351 info->new_layout = info->array.layout;
1352 else {
1353 near = info->new_layout & 0xff;
1354 far = (info->new_layout >> 8) & 0xff;
1355 offset = info->new_layout & 0x10000;
1356 if (far > 1 && !offset)
1357 return "Cannot reshape RAID10 to far-mode";
1358 if (near * far != copies)
1359 return "Cannot change number of copies when reshaping RAID10";
1360 }
1361 if (info->delta_disks == UnSet)
1362 info->delta_disks = 0;
1363 new_disks = (info->array.raid_disks +
1364 info->delta_disks);
1365
1366 new_chunk = info->new_chunk * far;
1367
1368 re->level = 10;
1369 re->before.layout = info->array.layout;
1370 re->before.data_disks = info->array.raid_disks;
1371 re->after.layout = info->new_layout;
1372 re->after.data_disks = new_disks;
1373 /* For RAID10 we don't do backup but do allow reshape,
1374 * so set backup_blocks to INVALID_SECTORS rather than
1375 * zero.
1376 * And there is no need to synchronise stripes on both
1377 * 'old' and 'new'. So the important
1378 * number is the minimum data_offset difference
1379 * which is the larger of (offset copies * chunk).
1380 */
1381 re->backup_blocks = INVALID_SECTORS;
1382 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1383 if (new_disks < re->before.data_disks &&
1384 info->space_after < re->min_offset_change)
1385 /* Reduce component size by one chunk */
1386 re->new_size = (info->component_size -
1387 re->min_offset_change);
1388 else
1389 re->new_size = info->component_size;
1390 re->new_size = re->new_size * new_disks / copies;
1391 return NULL;
1392
1393 default:
1394 return "RAID10 can only be changed to RAID0";
1395 }
1396 case 0:
1397 /* RAID0 can be converted to RAID10, or to RAID456 */
1398 if (info->new_level == 10) {
1399 if (info->new_layout == UnSet &&
1400 info->delta_disks == UnSet) {
1401 /* Assume near=2 layout */
1402 info->new_layout = 0x102;
1403 info->delta_disks = info->array.raid_disks;
1404 }
1405 if (info->new_layout == UnSet) {
1406 int copies = 1 + (info->delta_disks
1407 / info->array.raid_disks);
1408 if (info->array.raid_disks * (copies-1)
1409 != info->delta_disks)
1410 return "Impossible number of devices for RAID0->RAID10";
1411 info->new_layout = 0x100 + copies;
1412 }
1413 if (info->delta_disks == UnSet) {
1414 int copies = info->new_layout & 0xff;
1415 if (info->new_layout != 0x100 + copies)
1416 return "New layout impossible for RAID0->RAID10";;
1417 info->delta_disks = (copies - 1) *
1418 info->array.raid_disks;
1419 }
1420 if (info->new_chunk &&
1421 info->new_chunk != info->array.chunk_size)
1422 return "Cannot change chunk-size with RAID0->RAID10";
1423 /* looks good */
1424 re->level = 10;
1425 re->before.data_disks = (info->array.raid_disks +
1426 info->delta_disks);
1427 re->after.data_disks = re->before.data_disks;
1428 re->before.layout = info->new_layout;
1429 return NULL;
1430 }
1431
1432 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1433 * a raid4 style layout of the final level.
1434 */
1435 switch (info->new_level) {
1436 case 4:
1437 delta_parity = 1;
1438 case 0:
1439 re->level = 4;
1440 re->before.layout = 0;
1441 break;
1442 case 5:
1443 delta_parity = 1;
1444 re->level = 5;
1445 re->before.layout = ALGORITHM_PARITY_N;
1446 if (info->new_layout == UnSet)
1447 info->new_layout = map_name(r5layout, "default");
1448 break;
1449 case 6:
1450 delta_parity = 2;
1451 re->level = 6;
1452 re->before.layout = ALGORITHM_PARITY_N;
1453 if (info->new_layout == UnSet)
1454 info->new_layout = map_name(r6layout, "default");
1455 break;
1456 default:
1457 return "Impossible level change requested";
1458 }
1459 re->before.data_disks = info->array.raid_disks;
1460 /* determining 'after' layout happens outside this 'switch' */
1461 break;
1462
1463 case 4:
1464 info->array.layout = ALGORITHM_PARITY_N;
1465 case 5:
1466 switch (info->new_level) {
1467 case 0:
1468 delta_parity = -1;
1469 case 4:
1470 re->level = info->array.level;
1471 re->before.data_disks = info->array.raid_disks - 1;
1472 re->before.layout = info->array.layout;
1473 break;
1474 case 5:
1475 re->level = 5;
1476 re->before.data_disks = info->array.raid_disks - 1;
1477 re->before.layout = info->array.layout;
1478 break;
1479 case 6:
1480 delta_parity = 1;
1481 re->level = 6;
1482 re->before.data_disks = info->array.raid_disks - 1;
1483 switch (info->array.layout) {
1484 case ALGORITHM_LEFT_ASYMMETRIC:
1485 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1486 break;
1487 case ALGORITHM_RIGHT_ASYMMETRIC:
1488 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1489 break;
1490 case ALGORITHM_LEFT_SYMMETRIC:
1491 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1492 break;
1493 case ALGORITHM_RIGHT_SYMMETRIC:
1494 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1495 break;
1496 case ALGORITHM_PARITY_0:
1497 re->before.layout = ALGORITHM_PARITY_0_6;
1498 break;
1499 case ALGORITHM_PARITY_N:
1500 re->before.layout = ALGORITHM_PARITY_N_6;
1501 break;
1502 default:
1503 return "Cannot convert an array with this layout";
1504 }
1505 break;
1506 case 1:
1507 if (info->array.raid_disks != 2)
1508 return "Can only convert a 2-device array to RAID1";
1509 if (info->delta_disks != UnSet &&
1510 info->delta_disks != 0)
1511 return "Cannot set raid_disk when converting RAID5->RAID1";
1512 re->level = 1;
1513 info->new_chunk = 0;
1514 return NULL;
1515 default:
1516 return "Impossible level change requested";
1517 }
1518 break;
1519 case 6:
1520 switch (info->new_level) {
1521 case 4:
1522 case 5:
1523 delta_parity = -1;
1524 case 6:
1525 re->level = 6;
1526 re->before.data_disks = info->array.raid_disks - 2;
1527 re->before.layout = info->array.layout;
1528 break;
1529 default:
1530 return "Impossible level change requested";
1531 }
1532 break;
1533 }
1534
1535 /* If we reached here then it looks like a re-stripe is
1536 * happening. We have determined the intermediate level
1537 * and initial raid_disks/layout and stored these in 're'.
1538 *
1539 * We need to deduce the final layout that can be atomically
1540 * converted to the end state.
1541 */
1542 switch (info->new_level) {
1543 case 0:
1544 /* We can only get to RAID0 from RAID4 or RAID5
1545 * with appropriate layout and one extra device
1546 */
1547 if (re->level != 4 && re->level != 5)
1548 return "Cannot covert to RAID0 from this level";
1549
1550 switch (re->level) {
1551 case 4:
1552 re->before.layout = 0;
1553 re->after.layout = 0;
1554 break;
1555 case 5:
1556 re->after.layout = ALGORITHM_PARITY_N;
1557 break;
1558 }
1559 break;
1560
1561 case 4:
1562 /* We can only get to RAID4 from RAID5 */
1563 if (re->level != 4 && re->level != 5)
1564 return "Cannot convert to RAID4 from this level";
1565
1566 switch (re->level) {
1567 case 4:
1568 re->after.layout = 0;
1569 break;
1570 case 5:
1571 re->after.layout = ALGORITHM_PARITY_N;
1572 break;
1573 }
1574 break;
1575
1576 case 5:
1577 /* We get to RAID5 from RAID5 or RAID6 */
1578 if (re->level != 5 && re->level != 6)
1579 return "Cannot convert to RAID5 from this level";
1580
1581 switch (re->level) {
1582 case 5:
1583 if (info->new_layout == UnSet)
1584 re->after.layout = re->before.layout;
1585 else
1586 re->after.layout = info->new_layout;
1587 break;
1588 case 6:
1589 if (info->new_layout == UnSet)
1590 info->new_layout = re->before.layout;
1591
1592 /* after.layout needs to be raid6 version of new_layout */
1593 if (info->new_layout == ALGORITHM_PARITY_N)
1594 re->after.layout = ALGORITHM_PARITY_N;
1595 else {
1596 char layout[40];
1597 char *ls = map_num(r5layout, info->new_layout);
1598 int l;
1599 if (ls) {
1600 /* Current RAID6 layout has a RAID5
1601 * equivalent - good
1602 */
1603 strcat(strcpy(layout, ls), "-6");
1604 l = map_name(r6layout, layout);
1605 if (l == UnSet)
1606 return "Cannot find RAID6 layout to convert to";
1607 } else {
1608 /* Current RAID6 has no equivalent.
1609 * If it is already a '-6' layout we
1610 * can leave it unchanged, else we must
1611 * fail
1612 */
1613 ls = map_num(r6layout, info->new_layout);
1614 if (!ls ||
1615 strcmp(ls+strlen(ls)-2, "-6") != 0)
1616 return "Please specify new layout";
1617 l = info->new_layout;
1618 }
1619 re->after.layout = l;
1620 }
1621 }
1622 break;
1623
1624 case 6:
1625 /* We must already be at level 6 */
1626 if (re->level != 6)
1627 return "Impossible level change";
1628 if (info->new_layout == UnSet)
1629 re->after.layout = info->array.layout;
1630 else
1631 re->after.layout = info->new_layout;
1632 break;
1633 default:
1634 return "Impossible level change requested";
1635 }
1636 if (info->delta_disks == UnSet)
1637 info->delta_disks = delta_parity;
1638
1639 re->after.data_disks =
1640 (re->before.data_disks + info->delta_disks - delta_parity);
1641
1642 switch (re->level) {
1643 case 6:
1644 re->parity = 2;
1645 break;
1646 case 4:
1647 case 5:
1648 re->parity = 1;
1649 break;
1650 default:
1651 re->parity = 0;
1652 break;
1653 }
1654 /* So we have a restripe operation, we need to calculate the number
1655 * of blocks per reshape operation.
1656 */
1657 re->new_size = info->component_size * re->before.data_disks;
1658 if (info->new_chunk == 0)
1659 info->new_chunk = info->array.chunk_size;
1660 if (re->after.data_disks == re->before.data_disks &&
1661 re->after.layout == re->before.layout &&
1662 info->new_chunk == info->array.chunk_size) {
1663 /* Nothing to change, can change level immediately. */
1664 re->level = info->new_level;
1665 re->backup_blocks = 0;
1666 return NULL;
1667 }
1668 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1669 /* chunk and layout changes make no difference */
1670 re->level = info->new_level;
1671 re->backup_blocks = 0;
1672 return NULL;
1673 }
1674
1675 if (re->after.data_disks == re->before.data_disks &&
1676 get_linux_version() < 2006032)
1677 return "in-place reshape is not safe before 2.6.32 - sorry.";
1678
1679 if (re->after.data_disks < re->before.data_disks &&
1680 get_linux_version() < 2006030)
1681 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1682
1683 re->backup_blocks = compute_backup_blocks(
1684 info->new_chunk, info->array.chunk_size,
1685 re->after.data_disks,
1686 re->before.data_disks);
1687 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1688
1689 re->new_size = info->component_size * re->after.data_disks;
1690 return NULL;
1691 }
1692
1693 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1694 char *text_version)
1695 {
1696 struct mdinfo *info;
1697 char *subarray;
1698 int ret_val = -1;
1699
1700 if ((st == NULL) || (sra == NULL))
1701 return ret_val;
1702
1703 if (text_version == NULL)
1704 text_version = sra->text_version;
1705 subarray = strchr(text_version + 1, '/')+1;
1706 info = st->ss->container_content(st, subarray);
1707 if (info) {
1708 unsigned long long current_size = 0;
1709 unsigned long long new_size =
1710 info->custom_array_size/2;
1711
1712 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1713 new_size > current_size) {
1714 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1715 < 0)
1716 dprintf("Error: Cannot set array size");
1717 else {
1718 ret_val = 0;
1719 dprintf("Array size changed");
1720 }
1721 dprintf_cont(" from %llu to %llu.\n",
1722 current_size, new_size);
1723 }
1724 sysfs_free(info);
1725 } else
1726 dprintf("Error: set_array_size(): info pointer in NULL\n");
1727
1728 return ret_val;
1729 }
1730
1731 static int reshape_array(char *container, int fd, char *devname,
1732 struct supertype *st, struct mdinfo *info,
1733 int force, struct mddev_dev *devlist,
1734 unsigned long long data_offset,
1735 char *backup_file, int verbose, int forked,
1736 int restart, int freeze_reshape);
1737 static int reshape_container(char *container, char *devname,
1738 int mdfd,
1739 struct supertype *st,
1740 struct mdinfo *info,
1741 int force,
1742 char *backup_file, int verbose,
1743 int forked, int restart, int freeze_reshape);
1744
1745 int Grow_reshape(char *devname, int fd,
1746 struct mddev_dev *devlist,
1747 unsigned long long data_offset,
1748 struct context *c, struct shape *s)
1749 {
1750 /* Make some changes in the shape of an array.
1751 * The kernel must support the change.
1752 *
1753 * There are three different changes. Each can trigger
1754 * a resync or recovery so we freeze that until we have
1755 * requested everything (if kernel supports freezing - 2.6.30).
1756 * The steps are:
1757 * - change size (i.e. component_size)
1758 * - change level
1759 * - change layout/chunksize/ndisks
1760 *
1761 * The last can require a reshape. It is different on different
1762 * levels so we need to check the level before actioning it.
1763 * Some times the level change needs to be requested after the
1764 * reshape (e.g. raid6->raid5, raid5->raid0)
1765 *
1766 */
1767 struct mdu_array_info_s array;
1768 int rv = 0;
1769 struct supertype *st;
1770 char *subarray = NULL;
1771
1772 int frozen;
1773 int changed = 0;
1774 char *container = NULL;
1775 int cfd = -1;
1776
1777 struct mddev_dev *dv;
1778 int added_disks;
1779
1780 struct mdinfo info;
1781 struct mdinfo *sra;
1782
1783 if (md_get_array_info(fd, &array) < 0) {
1784 pr_err("%s is not an active md array - aborting\n",
1785 devname);
1786 return 1;
1787 }
1788 if (data_offset != INVALID_SECTORS && array.level != 10 &&
1789 (array.level < 4 || array.level > 6)) {
1790 pr_err("--grow --data-offset not yet supported\n");
1791 return 1;
1792 }
1793
1794 if (s->size > 0 &&
1795 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1796 pr_err("cannot change component size at the same time as other changes.\n"
1797 " Change size first, then check data is intact before making other changes.\n");
1798 return 1;
1799 }
1800
1801 if (s->raiddisks && s->raiddisks < array.raid_disks &&
1802 array.level > 1 && get_linux_version() < 2006032 &&
1803 !check_env("MDADM_FORCE_FEWER")) {
1804 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1805 " Please use a newer kernel\n");
1806 return 1;
1807 }
1808
1809 st = super_by_fd(fd, &subarray);
1810 if (!st) {
1811 pr_err("Unable to determine metadata format for %s\n", devname);
1812 return 1;
1813 }
1814 if (s->raiddisks > st->max_devs) {
1815 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1816 return 1;
1817 }
1818 if (s->level == 0 &&
1819 (array.state & (1<<MD_SB_BITMAP_PRESENT)) &&
1820 !(array.state & (1<<MD_SB_CLUSTERED))) {
1821 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
1822 if (md_set_array_info(fd, &array)!= 0) {
1823 pr_err("failed to remove internal bitmap.\n");
1824 return 1;
1825 }
1826 }
1827
1828 /* in the external case we need to check that the requested reshape is
1829 * supported, and perform an initial check that the container holds the
1830 * pre-requisite spare devices (mdmon owns final validation)
1831 */
1832 if (st->ss->external) {
1833 int retval;
1834
1835 if (subarray) {
1836 container = st->container_devnm;
1837 cfd = open_dev_excl(st->container_devnm);
1838 } else {
1839 container = st->devnm;
1840 close(fd);
1841 cfd = open_dev_excl(st->devnm);
1842 fd = cfd;
1843 }
1844 if (cfd < 0) {
1845 pr_err("Unable to open container for %s\n",
1846 devname);
1847 free(subarray);
1848 return 1;
1849 }
1850
1851 retval = st->ss->load_container(st, cfd, NULL);
1852
1853 if (retval) {
1854 pr_err("Cannot read superblock for %s\n",
1855 devname);
1856 free(subarray);
1857 return 1;
1858 }
1859
1860 /* check if operation is supported for metadata handler */
1861 if (st->ss->container_content) {
1862 struct mdinfo *cc = NULL;
1863 struct mdinfo *content = NULL;
1864
1865 cc = st->ss->container_content(st, subarray);
1866 for (content = cc; content ; content = content->next) {
1867 int allow_reshape = 1;
1868
1869 /* check if reshape is allowed based on metadata
1870 * indications stored in content.array.status
1871 */
1872 if (content->array.state &
1873 (1 << MD_SB_BLOCK_VOLUME))
1874 allow_reshape = 0;
1875 if (content->array.state &
1876 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE))
1877 allow_reshape = 0;
1878 if (!allow_reshape) {
1879 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1880 devname, container);
1881 sysfs_free(cc);
1882 free(subarray);
1883 return 1;
1884 }
1885 }
1886 sysfs_free(cc);
1887 }
1888 if (mdmon_running(container))
1889 st->update_tail = &st->updates;
1890 }
1891
1892 added_disks = 0;
1893 for (dv = devlist; dv; dv = dv->next)
1894 added_disks++;
1895 if (s->raiddisks > array.raid_disks &&
1896 array.spare_disks + added_disks < (s->raiddisks - array.raid_disks) &&
1897 !c->force) {
1898 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1899 " Use --force to over-ride this check.\n",
1900 s->raiddisks - array.raid_disks,
1901 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1902 array.spare_disks + added_disks);
1903 return 1;
1904 }
1905
1906 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS |
1907 GET_STATE | GET_VERSION);
1908 if (sra) {
1909 if (st->ss->external && subarray == NULL) {
1910 array.level = LEVEL_CONTAINER;
1911 sra->array.level = LEVEL_CONTAINER;
1912 }
1913 } else {
1914 pr_err("failed to read sysfs parameters for %s\n",
1915 devname);
1916 return 1;
1917 }
1918 frozen = freeze(st);
1919 if (frozen < -1) {
1920 /* freeze() already spewed the reason */
1921 sysfs_free(sra);
1922 return 1;
1923 } else if (frozen < 0) {
1924 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1925 sysfs_free(sra);
1926 return 1;
1927 }
1928
1929 /* ========= set size =============== */
1930 if (s->size > 0 &&
1931 (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1932 unsigned long long orig_size = get_component_size(fd)/2;
1933 unsigned long long min_csize;
1934 struct mdinfo *mdi;
1935 int raid0_takeover = 0;
1936
1937 if (orig_size == 0)
1938 orig_size = (unsigned) array.size;
1939
1940 if (orig_size == 0) {
1941 pr_err("Cannot set device size in this type of array.\n");
1942 rv = 1;
1943 goto release;
1944 }
1945
1946 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1947 devname, APPLY_METADATA_CHANGES,
1948 c->verbose > 0)) {
1949 rv = 1;
1950 goto release;
1951 }
1952 sync_metadata(st);
1953 if (st->ss->external) {
1954 /* metadata can have size limitation
1955 * update size value according to metadata information
1956 */
1957 struct mdinfo *sizeinfo =
1958 st->ss->container_content(st, subarray);
1959 if (sizeinfo) {
1960 unsigned long long new_size =
1961 sizeinfo->custom_array_size/2;
1962 int data_disks = get_data_disks(
1963 sizeinfo->array.level,
1964 sizeinfo->array.layout,
1965 sizeinfo->array.raid_disks);
1966 new_size /= data_disks;
1967 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1968 orig_size, new_size,
1969 new_size * data_disks);
1970 s->size = new_size;
1971 sysfs_free(sizeinfo);
1972 }
1973 }
1974
1975 /* Update the size of each member device in case
1976 * they have been resized. This will never reduce
1977 * below the current used-size. The "size" attribute
1978 * understands '0' to mean 'max'.
1979 */
1980 min_csize = 0;
1981 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1982 if (array.not_persistent == 0 &&
1983 array.major_version == 0 &&
1984 get_linux_version() < 3001000) {
1985 /* Dangerous to allow size to exceed 2TB */
1986 unsigned long long csize;
1987 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1988 if (csize >= 2ULL*1024*1024*1024)
1989 csize = 2ULL*1024*1024*1024;
1990 if ((min_csize == 0 || (min_csize
1991 > csize)))
1992 min_csize = csize;
1993 }
1994 }
1995 }
1996 if (min_csize && s->size > min_csize) {
1997 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1998 rv = 1;
1999 goto size_change_error;
2000 }
2001 if (min_csize && s->size == MAX_SIZE) {
2002 /* Don't let the kernel choose a size - it will get
2003 * it wrong
2004 */
2005 pr_err("Limited v0.90 array to 2TB per device\n");
2006 s->size = min_csize;
2007 }
2008 if (st->ss->external) {
2009 if (sra->array.level == 0) {
2010 rv = sysfs_set_str(sra, NULL, "level",
2011 "raid5");
2012 if (!rv) {
2013 raid0_takeover = 1;
2014 /* get array parameters after takeover
2015 * to change one parameter at time only
2016 */
2017 rv = md_get_array_info(fd, &array);
2018 }
2019 }
2020 /* make sure mdmon is
2021 * aware of the new level */
2022 if (!mdmon_running(st->container_devnm))
2023 start_mdmon(st->container_devnm);
2024 ping_monitor(container);
2025 if (mdmon_running(st->container_devnm) &&
2026 st->update_tail == NULL)
2027 st->update_tail = &st->updates;
2028 }
2029
2030 if (s->size == MAX_SIZE)
2031 s->size = 0;
2032 array.size = s->size;
2033 if (s->size & ~INT32_MAX) {
2034 /* got truncated to 32bit, write to
2035 * component_size instead
2036 */
2037 if (sra)
2038 rv = sysfs_set_num(sra, NULL,
2039 "component_size", s->size);
2040 else
2041 rv = -1;
2042 } else {
2043 rv = md_set_array_info(fd, &array);
2044
2045 /* manage array size when it is managed externally
2046 */
2047 if ((rv == 0) && st->ss->external)
2048 rv = set_array_size(st, sra, sra->text_version);
2049 }
2050
2051 if (raid0_takeover) {
2052 /* do not recync non-existing parity,
2053 * we will drop it anyway
2054 */
2055 sysfs_set_str(sra, NULL, "sync_action", "frozen");
2056 /* go back to raid0, drop parity disk
2057 */
2058 sysfs_set_str(sra, NULL, "level", "raid0");
2059 md_get_array_info(fd, &array);
2060 }
2061
2062 size_change_error:
2063 if (rv != 0) {
2064 int err = errno;
2065
2066 /* restore metadata */
2067 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
2068 UnSet, NULL, devname,
2069 ROLLBACK_METADATA_CHANGES,
2070 c->verbose) == 0)
2071 sync_metadata(st);
2072 pr_err("Cannot set device size for %s: %s\n",
2073 devname, strerror(err));
2074 if (err == EBUSY &&
2075 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2076 cont_err("Bitmap must be removed before size can be changed\n");
2077 rv = 1;
2078 goto release;
2079 }
2080 if (s->assume_clean) {
2081 /* This will fail on kernels older than 3.0 unless
2082 * a backport has been arranged.
2083 */
2084 if (sra == NULL ||
2085 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
2086 pr_err("--assume-clean not supported with --grow on this kernel\n");
2087 }
2088 md_get_array_info(fd, &array);
2089 s->size = get_component_size(fd)/2;
2090 if (s->size == 0)
2091 s->size = array.size;
2092 if (c->verbose >= 0) {
2093 if (s->size == orig_size)
2094 pr_err("component size of %s unchanged at %lluK\n",
2095 devname, s->size);
2096 else
2097 pr_err("component size of %s has been set to %lluK\n",
2098 devname, s->size);
2099 }
2100 changed = 1;
2101 } else if (array.level != LEVEL_CONTAINER) {
2102 s->size = get_component_size(fd)/2;
2103 if (s->size == 0)
2104 s->size = array.size;
2105 }
2106
2107 /* See if there is anything else to do */
2108 if ((s->level == UnSet || s->level == array.level) &&
2109 (s->layout_str == NULL) &&
2110 (s->chunk == 0 || s->chunk == array.chunk_size) &&
2111 data_offset == INVALID_SECTORS &&
2112 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
2113 /* Nothing more to do */
2114 if (!changed && c->verbose >= 0)
2115 pr_err("%s: no change requested\n",
2116 devname);
2117 goto release;
2118 }
2119
2120 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2121 * current implementation assumes that following conditions must be met:
2122 * - RAID10:
2123 * - far_copies == 1
2124 * - near_copies == 2
2125 */
2126 if ((s->level == 0 && array.level == 10 && sra &&
2127 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
2128 (s->level == 0 && array.level == 1 && sra)) {
2129 int err;
2130 err = remove_disks_for_takeover(st, sra, array.layout);
2131 if (err) {
2132 dprintf("Array cannot be reshaped\n");
2133 if (cfd > -1)
2134 close(cfd);
2135 rv = 1;
2136 goto release;
2137 }
2138 /* Make sure mdmon has seen the device removal
2139 * and updated metadata before we continue with
2140 * level change
2141 */
2142 if (container)
2143 ping_monitor(container);
2144 }
2145
2146 memset(&info, 0, sizeof(info));
2147 info.array = array;
2148 if (sysfs_init(&info, fd, NULL)) {
2149 pr_err("failed to intialize sysfs.\n");
2150 rv = 1;
2151 goto release;
2152 }
2153 strcpy(info.text_version, sra->text_version);
2154 info.component_size = s->size*2;
2155 info.new_level = s->level;
2156 info.new_chunk = s->chunk * 1024;
2157 if (info.array.level == LEVEL_CONTAINER) {
2158 info.delta_disks = UnSet;
2159 info.array.raid_disks = s->raiddisks;
2160 } else if (s->raiddisks)
2161 info.delta_disks = s->raiddisks - info.array.raid_disks;
2162 else
2163 info.delta_disks = UnSet;
2164 if (s->layout_str == NULL) {
2165 info.new_layout = UnSet;
2166 if (info.array.level == 6 &&
2167 (info.new_level == 6 || info.new_level == UnSet) &&
2168 info.array.layout >= 16) {
2169 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
2170 cont_err("during the reshape, please specify --layout=preserve\n");
2171 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2172 rv = 1;
2173 goto release;
2174 }
2175 } else if (strcmp(s->layout_str, "normalise") == 0 ||
2176 strcmp(s->layout_str, "normalize") == 0) {
2177 /* If we have a -6 RAID6 layout, remove the '-6'. */
2178 info.new_layout = UnSet;
2179 if (info.array.level == 6 && info.new_level == UnSet) {
2180 char l[40], *h;
2181 strcpy(l, map_num(r6layout, info.array.layout));
2182 h = strrchr(l, '-');
2183 if (h && strcmp(h, "-6") == 0) {
2184 *h = 0;
2185 info.new_layout = map_name(r6layout, l);
2186 }
2187 } else {
2188 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
2189 rv = 1;
2190 goto release;
2191 }
2192 } else if (strcmp(s->layout_str, "preserve") == 0) {
2193 /* This means that a non-standard RAID6 layout
2194 * is OK.
2195 * In particular:
2196 * - When reshape a RAID6 (e.g. adding a device)
2197 * which is in a non-standard layout, it is OK
2198 * to preserve that layout.
2199 * - When converting a RAID5 to RAID6, leave it in
2200 * the XXX-6 layout, don't re-layout.
2201 */
2202 if (info.array.level == 6 && info.new_level == UnSet)
2203 info.new_layout = info.array.layout;
2204 else if (info.array.level == 5 && info.new_level == 6) {
2205 char l[40];
2206 strcpy(l, map_num(r5layout, info.array.layout));
2207 strcat(l, "-6");
2208 info.new_layout = map_name(r6layout, l);
2209 } else {
2210 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
2211 rv = 1;
2212 goto release;
2213 }
2214 } else {
2215 int l = info.new_level;
2216 if (l == UnSet)
2217 l = info.array.level;
2218 switch (l) {
2219 case 5:
2220 info.new_layout = map_name(r5layout, s->layout_str);
2221 break;
2222 case 6:
2223 info.new_layout = map_name(r6layout, s->layout_str);
2224 break;
2225 case 10:
2226 info.new_layout = parse_layout_10(s->layout_str);
2227 break;
2228 case LEVEL_FAULTY:
2229 info.new_layout = parse_layout_faulty(s->layout_str);
2230 break;
2231 default:
2232 pr_err("layout not meaningful with this level\n");
2233 rv = 1;
2234 goto release;
2235 }
2236 if (info.new_layout == UnSet) {
2237 pr_err("layout %s not understood for this level\n",
2238 s->layout_str);
2239 rv = 1;
2240 goto release;
2241 }
2242 }
2243
2244 if (array.level == LEVEL_FAULTY) {
2245 if (s->level != UnSet && s->level != array.level) {
2246 pr_err("cannot change level of Faulty device\n");
2247 rv =1 ;
2248 }
2249 if (s->chunk) {
2250 pr_err("cannot set chunksize of Faulty device\n");
2251 rv =1 ;
2252 }
2253 if (s->raiddisks && s->raiddisks != 1) {
2254 pr_err("cannot set raid_disks of Faulty device\n");
2255 rv =1 ;
2256 }
2257 if (s->layout_str) {
2258 if (md_get_array_info(fd, &array) != 0) {
2259 dprintf("Cannot get array information.\n");
2260 goto release;
2261 }
2262 array.layout = info.new_layout;
2263 if (md_set_array_info(fd, &array) != 0) {
2264 pr_err("failed to set new layout\n");
2265 rv = 1;
2266 } else if (c->verbose >= 0)
2267 printf("layout for %s set to %d\n",
2268 devname, array.layout);
2269 }
2270 } else if (array.level == LEVEL_CONTAINER) {
2271 /* This change is to be applied to every array in the
2272 * container. This is only needed when the metadata imposes
2273 * restraints of the various arrays in the container.
2274 * Currently we only know that IMSM requires all arrays
2275 * to have the same number of devices so changing the
2276 * number of devices (On-Line Capacity Expansion) must be
2277 * performed at the level of the container
2278 */
2279 if (fd > 0) {
2280 close(fd);
2281 fd = -1;
2282 }
2283 rv = reshape_container(container, devname, -1, st, &info,
2284 c->force, c->backup_file, c->verbose, 0, 0, 0);
2285 frozen = 0;
2286 } else {
2287 /* get spare devices from external metadata
2288 */
2289 if (st->ss->external) {
2290 struct mdinfo *info2;
2291
2292 info2 = st->ss->container_content(st, subarray);
2293 if (info2) {
2294 info.array.spare_disks =
2295 info2->array.spare_disks;
2296 sysfs_free(info2);
2297 }
2298 }
2299
2300 /* Impose these changes on a single array. First
2301 * check that the metadata is OK with the change. */
2302
2303 if (reshape_super(st, 0, info.new_level,
2304 info.new_layout, info.new_chunk,
2305 info.array.raid_disks, info.delta_disks,
2306 c->backup_file, devname, APPLY_METADATA_CHANGES,
2307 c->verbose)) {
2308 rv = 1;
2309 goto release;
2310 }
2311 sync_metadata(st);
2312 rv = reshape_array(container, fd, devname, st, &info, c->force,
2313 devlist, data_offset, c->backup_file, c->verbose,
2314 0, 0, 0);
2315 frozen = 0;
2316 }
2317 release:
2318 sysfs_free(sra);
2319 if (frozen > 0)
2320 unfreeze(st);
2321 return rv;
2322 }
2323
2324 /* verify_reshape_position()
2325 * Function checks if reshape position in metadata is not farther
2326 * than position in md.
2327 * Return value:
2328 * 0 : not valid sysfs entry
2329 * it can be caused by not started reshape, it should be started
2330 * by reshape array or raid0 array is before takeover
2331 * -1 : error, reshape position is obviously wrong
2332 * 1 : success, reshape progress correct or updated
2333 */
2334 static int verify_reshape_position(struct mdinfo *info, int level)
2335 {
2336 int ret_val = 0;
2337 char buf[40];
2338 int rv;
2339
2340 /* read sync_max, failure can mean raid0 array */
2341 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2342
2343 if (rv > 0) {
2344 char *ep;
2345 unsigned long long position = strtoull(buf, &ep, 0);
2346
2347 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2348 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2349 position *= get_data_disks(level,
2350 info->new_layout,
2351 info->array.raid_disks);
2352 if (info->reshape_progress < position) {
2353 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2354 info->reshape_progress, position);
2355 info->reshape_progress = position;
2356 ret_val = 1;
2357 } else if (info->reshape_progress > position) {
2358 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2359 position, info->reshape_progress);
2360 ret_val = -1;
2361 } else {
2362 dprintf("Reshape position in md and metadata are the same;");
2363 ret_val = 1;
2364 }
2365 }
2366 } else if (rv == 0) {
2367 /* for valid sysfs entry, 0-length content
2368 * should be indicated as error
2369 */
2370 ret_val = -1;
2371 }
2372
2373 return ret_val;
2374 }
2375
2376 static unsigned long long choose_offset(unsigned long long lo,
2377 unsigned long long hi,
2378 unsigned long long min,
2379 unsigned long long max)
2380 {
2381 /* Choose a new offset between hi and lo.
2382 * It must be between min and max, but
2383 * we would prefer something near the middle of hi/lo, and also
2384 * prefer to be aligned to a big power of 2.
2385 *
2386 * So we start with the middle, then for each bit,
2387 * starting at '1' and increasing, if it is set, we either
2388 * add it or subtract it if possible, preferring the option
2389 * which is furthest from the boundary.
2390 *
2391 * We stop once we get a 1MB alignment. As units are in sectors,
2392 * 1MB = 2*1024 sectors.
2393 */
2394 unsigned long long choice = (lo + hi) / 2;
2395 unsigned long long bit = 1;
2396
2397 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2398 unsigned long long bigger, smaller;
2399 if (! (bit & choice))
2400 continue;
2401 bigger = choice + bit;
2402 smaller = choice - bit;
2403 if (bigger > max && smaller < min)
2404 break;
2405 if (bigger > max)
2406 choice = smaller;
2407 else if (smaller < min)
2408 choice = bigger;
2409 else if (hi - bigger > smaller - lo)
2410 choice = bigger;
2411 else
2412 choice = smaller;
2413 }
2414 return choice;
2415 }
2416
2417 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2418 char *devname, int delta_disks,
2419 unsigned long long data_offset,
2420 unsigned long long min,
2421 int can_fallback)
2422 {
2423 struct mdinfo *sd;
2424 int dir = 0;
2425 int err = 0;
2426 unsigned long long before, after;
2427
2428 /* Need to find min space before and after so same is used
2429 * on all devices
2430 */
2431 before = UINT64_MAX;
2432 after = UINT64_MAX;
2433 for (sd = sra->devs; sd; sd = sd->next) {
2434 char *dn;
2435 int dfd;
2436 int rv;
2437 struct supertype *st2;
2438 struct mdinfo info2;
2439
2440 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2441 continue;
2442 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2443 dfd = dev_open(dn, O_RDONLY);
2444 if (dfd < 0) {
2445 pr_err("%s: cannot open component %s\n",
2446 devname, dn ? dn : "-unknown-");
2447 goto release;
2448 }
2449 st2 = dup_super(st);
2450 rv = st2->ss->load_super(st2,dfd, NULL);
2451 close(dfd);
2452 if (rv) {
2453 free(st2);
2454 pr_err("%s: cannot get superblock from %s\n",
2455 devname, dn);
2456 goto release;
2457 }
2458 st2->ss->getinfo_super(st2, &info2, NULL);
2459 st2->ss->free_super(st2);
2460 free(st2);
2461 if (info2.space_before == 0 &&
2462 info2.space_after == 0) {
2463 /* Metadata doesn't support data_offset changes */
2464 if (!can_fallback)
2465 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2466 devname);
2467 goto fallback;
2468 }
2469 if (before > info2.space_before)
2470 before = info2.space_before;
2471 if (after > info2.space_after)
2472 after = info2.space_after;
2473
2474 if (data_offset != INVALID_SECTORS) {
2475 if (dir == 0) {
2476 if (info2.data_offset == data_offset) {
2477 pr_err("%s: already has that data_offset\n",
2478 dn);
2479 goto release;
2480 }
2481 if (data_offset < info2.data_offset)
2482 dir = -1;
2483 else
2484 dir = 1;
2485 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2486 (data_offset >= info2.data_offset && dir == -1)) {
2487 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2488 dn);
2489 goto release;
2490 }
2491 }
2492 }
2493 if (before == UINT64_MAX)
2494 /* impossible really, there must be no devices */
2495 return 1;
2496
2497 for (sd = sra->devs; sd; sd = sd->next) {
2498 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2499 unsigned long long new_data_offset;
2500
2501 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2502 continue;
2503 if (delta_disks < 0) {
2504 /* Don't need any space as array is shrinking
2505 * just move data_offset up by min
2506 */
2507 if (data_offset == INVALID_SECTORS)
2508 new_data_offset = sd->data_offset + min;
2509 else {
2510 if (data_offset < sd->data_offset + min) {
2511 pr_err("--data-offset too small for %s\n",
2512 dn);
2513 goto release;
2514 }
2515 new_data_offset = data_offset;
2516 }
2517 } else if (delta_disks > 0) {
2518 /* need space before */
2519 if (before < min) {
2520 if (can_fallback)
2521 goto fallback;
2522 pr_err("Insufficient head-space for reshape on %s\n",
2523 dn);
2524 goto release;
2525 }
2526 if (data_offset == INVALID_SECTORS)
2527 new_data_offset = sd->data_offset - min;
2528 else {
2529 if (data_offset > sd->data_offset - min) {
2530 pr_err("--data-offset too large for %s\n",
2531 dn);
2532 goto release;
2533 }
2534 new_data_offset = data_offset;
2535 }
2536 } else {
2537 if (dir == 0) {
2538 /* can move up or down. If 'data_offset'
2539 * was set we would have already decided,
2540 * so just choose direction with most space.
2541 */
2542 if (before > after)
2543 dir = -1;
2544 else
2545 dir = 1;
2546 }
2547 sysfs_set_str(sra, NULL, "reshape_direction",
2548 dir == 1 ? "backwards" : "forwards");
2549 if (dir > 0) {
2550 /* Increase data offset */
2551 if (after < min) {
2552 if (can_fallback)
2553 goto fallback;
2554 pr_err("Insufficient tail-space for reshape on %s\n",
2555 dn);
2556 goto release;
2557 }
2558 if (data_offset != INVALID_SECTORS &&
2559 data_offset < sd->data_offset + min) {
2560 pr_err("--data-offset too small on %s\n",
2561 dn);
2562 goto release;
2563 }
2564 if (data_offset != INVALID_SECTORS)
2565 new_data_offset = data_offset;
2566 else
2567 new_data_offset = choose_offset(sd->data_offset,
2568 sd->data_offset + after,
2569 sd->data_offset + min,
2570 sd->data_offset + after);
2571 } else {
2572 /* Decrease data offset */
2573 if (before < min) {
2574 if (can_fallback)
2575 goto fallback;
2576 pr_err("insufficient head-room 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 - before,
2590 sd->data_offset,
2591 sd->data_offset - before,
2592 sd->data_offset - min);
2593 }
2594 }
2595 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2596 if (err < 0 && errno == E2BIG) {
2597 /* try again after increasing data size to max */
2598 err = sysfs_set_num(sra, sd, "size", 0);
2599 if (err < 0 && errno == EINVAL &&
2600 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2601 /* some kernels have a bug where you cannot
2602 * use '0' on spare devices. */
2603 sysfs_set_num(sra, sd, "size",
2604 (sra->component_size + after)/2);
2605 }
2606 err = sysfs_set_num(sra, sd, "new_offset",
2607 new_data_offset);
2608 }
2609 if (err < 0) {
2610 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2611 pr_err("data-offset is too big for %s\n",
2612 dn);
2613 goto release;
2614 }
2615 if (sd == sra->devs &&
2616 (errno == ENOENT || errno == E2BIG))
2617 /* Early kernel, no 'new_offset' file,
2618 * or kernel doesn't like us.
2619 * For RAID5/6 this is not fatal
2620 */
2621 return 1;
2622 pr_err("Cannot set new_offset for %s\n",
2623 dn);
2624 break;
2625 }
2626 }
2627 return err;
2628 release:
2629 return -1;
2630 fallback:
2631 /* Just use a backup file */
2632 return 1;
2633 }
2634
2635 static int raid10_reshape(char *container, int fd, char *devname,
2636 struct supertype *st, struct mdinfo *info,
2637 struct reshape *reshape,
2638 unsigned long long data_offset,
2639 int force, int verbose)
2640 {
2641 /* Changing raid_disks, layout, chunksize or possibly
2642 * just data_offset for a RAID10.
2643 * We must always change data_offset. We change by at least
2644 * ->min_offset_change which is the largest of the old and new
2645 * chunk sizes.
2646 * If raid_disks is increasing, then data_offset must decrease
2647 * by at least this copy size.
2648 * If raid_disks is unchanged, data_offset must increase or
2649 * decrease by at least min_offset_change but preferably by much more.
2650 * We choose half of the available space.
2651 * If raid_disks is decreasing, data_offset must increase by
2652 * at least min_offset_change. To allow of this, component_size
2653 * must be decreased by the same amount.
2654 *
2655 * So we calculate the required minimum and direction, possibly
2656 * reduce the component_size, then iterate through the devices
2657 * and set the new_data_offset.
2658 * If that all works, we set chunk_size, layout, raid_disks, and start
2659 * 'reshape'
2660 */
2661 struct mdinfo *sra;
2662 unsigned long long min;
2663 int err = 0;
2664
2665 sra = sysfs_read(fd, NULL,
2666 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2667 );
2668 if (!sra) {
2669 pr_err("%s: Cannot get array details from sysfs\n",
2670 devname);
2671 goto release;
2672 }
2673 min = reshape->min_offset_change;
2674
2675 if (info->delta_disks)
2676 sysfs_set_str(sra, NULL, "reshape_direction",
2677 info->delta_disks < 0 ? "backwards" : "forwards");
2678 if (info->delta_disks < 0 &&
2679 info->space_after < min) {
2680 int rv = sysfs_set_num(sra, NULL, "component_size",
2681 (sra->component_size -
2682 min)/2);
2683 if (rv) {
2684 pr_err("cannot reduce component size\n");
2685 goto release;
2686 }
2687 }
2688 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2689 min, 0);
2690 if (err == 1) {
2691 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2692 cont_err("supported on this kernel\n");
2693 err = -1;
2694 }
2695 if (err < 0)
2696 goto release;
2697
2698 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2699 err = errno;
2700 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2701 err = errno;
2702 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2703 info->array.raid_disks + info->delta_disks) < 0)
2704 err = errno;
2705 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2706 err = errno;
2707 if (err) {
2708 pr_err("Cannot set array shape for %s\n",
2709 devname);
2710 if (err == EBUSY &&
2711 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2712 cont_err(" Bitmap must be removed before shape can be changed\n");
2713 goto release;
2714 }
2715 sysfs_free(sra);
2716 return 0;
2717 release:
2718 sysfs_free(sra);
2719 return 1;
2720 }
2721
2722 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2723 {
2724 struct mdinfo *sra, *sd;
2725 /* Initialisation to silence compiler warning */
2726 unsigned long long min_space_before = 0, min_space_after = 0;
2727 int first = 1;
2728
2729 sra = sysfs_read(fd, NULL, GET_DEVS);
2730 if (!sra)
2731 return;
2732 for (sd = sra->devs; sd; sd = sd->next) {
2733 char *dn;
2734 int dfd;
2735 struct supertype *st2;
2736 struct mdinfo info2;
2737
2738 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2739 continue;
2740 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2741 dfd = dev_open(dn, O_RDONLY);
2742 if (dfd < 0)
2743 break;
2744 st2 = dup_super(st);
2745 if (st2->ss->load_super(st2,dfd, NULL)) {
2746 close(dfd);
2747 free(st2);
2748 break;
2749 }
2750 close(dfd);
2751 st2->ss->getinfo_super(st2, &info2, NULL);
2752 st2->ss->free_super(st2);
2753 free(st2);
2754 if (first ||
2755 min_space_before > info2.space_before)
2756 min_space_before = info2.space_before;
2757 if (first ||
2758 min_space_after > info2.space_after)
2759 min_space_after = info2.space_after;
2760 first = 0;
2761 }
2762 if (sd == NULL && !first) {
2763 info->space_after = min_space_after;
2764 info->space_before = min_space_before;
2765 }
2766 sysfs_free(sra);
2767 }
2768
2769 static void update_cache_size(char *container, struct mdinfo *sra,
2770 struct mdinfo *info,
2771 int disks, unsigned long long blocks)
2772 {
2773 /* Check that the internal stripe cache is
2774 * large enough, or it won't work.
2775 * It must hold at least 4 stripes of the larger
2776 * chunk size
2777 */
2778 unsigned long cache;
2779 cache = max(info->array.chunk_size, info->new_chunk);
2780 cache *= 4; /* 4 stripes minimum */
2781 cache /= 512; /* convert to sectors */
2782 /* make sure there is room for 'blocks' with a bit to spare */
2783 if (cache < 16 + blocks / disks)
2784 cache = 16 + blocks / disks;
2785 cache /= (4096/512); /* Convert from sectors to pages */
2786
2787 if (sra->cache_size < cache)
2788 subarray_set_num(container, sra, "stripe_cache_size",
2789 cache+1);
2790 }
2791
2792 static int impose_reshape(struct mdinfo *sra,
2793 struct mdinfo *info,
2794 struct supertype *st,
2795 int fd,
2796 int restart,
2797 char *devname, char *container,
2798 struct reshape *reshape)
2799 {
2800 struct mdu_array_info_s array;
2801
2802 sra->new_chunk = info->new_chunk;
2803
2804 if (restart) {
2805 /* for external metadata checkpoint saved by mdmon can be lost
2806 * or missed /due to e.g. crash/. Check if md is not during
2807 * restart farther than metadata points to.
2808 * If so, this means metadata information is obsolete.
2809 */
2810 if (st->ss->external)
2811 verify_reshape_position(info, reshape->level);
2812 sra->reshape_progress = info->reshape_progress;
2813 } else {
2814 sra->reshape_progress = 0;
2815 if (reshape->after.data_disks < reshape->before.data_disks)
2816 /* start from the end of the new array */
2817 sra->reshape_progress = (sra->component_size
2818 * reshape->after.data_disks);
2819 }
2820
2821 md_get_array_info(fd, &array);
2822 if (info->array.chunk_size == info->new_chunk &&
2823 reshape->before.layout == reshape->after.layout &&
2824 st->ss->external == 0) {
2825 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2826 array.raid_disks = reshape->after.data_disks + reshape->parity;
2827 if (!restart && md_set_array_info(fd, &array) != 0) {
2828 int err = errno;
2829
2830 pr_err("Cannot set device shape for %s: %s\n",
2831 devname, strerror(errno));
2832
2833 if (err == EBUSY &&
2834 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2835 cont_err("Bitmap must be removed before shape can be changed\n");
2836
2837 goto release;
2838 }
2839 } else if (!restart) {
2840 /* set them all just in case some old 'new_*' value
2841 * persists from some earlier problem.
2842 */
2843 int err = 0;
2844 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2845 err = errno;
2846 if (!err && sysfs_set_num(sra, NULL, "layout",
2847 reshape->after.layout) < 0)
2848 err = errno;
2849 if (!err && subarray_set_num(container, sra, "raid_disks",
2850 reshape->after.data_disks +
2851 reshape->parity) < 0)
2852 err = errno;
2853 if (err) {
2854 pr_err("Cannot set device shape for %s\n",
2855 devname);
2856
2857 if (err == EBUSY &&
2858 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2859 cont_err("Bitmap must be removed before shape can be changed\n");
2860 goto release;
2861 }
2862 }
2863 return 0;
2864 release:
2865 return -1;
2866 }
2867
2868 static int impose_level(int fd, int level, char *devname, int verbose)
2869 {
2870 char *c;
2871 struct mdu_array_info_s array;
2872 struct mdinfo info;
2873
2874 if (sysfs_init(&info, fd, NULL)) {
2875 pr_err("failed to intialize sysfs.\n");
2876 return 1;
2877 }
2878
2879 md_get_array_info(fd, &array);
2880 if (level == 0 &&
2881 (array.level >= 4 && array.level <= 6)) {
2882 /* To convert to RAID0 we need to fail and
2883 * remove any non-data devices. */
2884 int found = 0;
2885 int d;
2886 int data_disks = array.raid_disks - 1;
2887 if (array.level == 6)
2888 data_disks -= 1;
2889 if (array.level == 5 &&
2890 array.layout != ALGORITHM_PARITY_N)
2891 return -1;
2892 if (array.level == 6 &&
2893 array.layout != ALGORITHM_PARITY_N_6)
2894 return -1;
2895 sysfs_set_str(&info, NULL,"sync_action", "idle");
2896 /* First remove any spares so no recovery starts */
2897 for (d = 0, found = 0;
2898 d < MAX_DISKS && found < array.nr_disks;
2899 d++) {
2900 mdu_disk_info_t disk;
2901 disk.number = d;
2902 if (md_get_disk_info(fd, &disk) < 0)
2903 continue;
2904 if (disk.major == 0 && disk.minor == 0)
2905 continue;
2906 found++;
2907 if ((disk.state & (1 << MD_DISK_ACTIVE))
2908 && disk.raid_disk < data_disks)
2909 /* keep this */
2910 continue;
2911 ioctl(fd, HOT_REMOVE_DISK,
2912 makedev(disk.major, disk.minor));
2913 }
2914 /* Now fail anything left */
2915 md_get_array_info(fd, &array);
2916 for (d = 0, found = 0;
2917 d < MAX_DISKS && found < array.nr_disks;
2918 d++) {
2919 mdu_disk_info_t disk;
2920 disk.number = d;
2921 if (md_get_disk_info(fd, &disk) < 0)
2922 continue;
2923 if (disk.major == 0 && disk.minor == 0)
2924 continue;
2925 found++;
2926 if ((disk.state & (1 << MD_DISK_ACTIVE))
2927 && disk.raid_disk < data_disks)
2928 /* keep this */
2929 continue;
2930 ioctl(fd, SET_DISK_FAULTY,
2931 makedev(disk.major, disk.minor));
2932 hot_remove_disk(fd, makedev(disk.major, disk.minor), 1);
2933 }
2934 }
2935 c = map_num(pers, level);
2936 if (c) {
2937 int err = sysfs_set_str(&info, NULL, "level", c);
2938 if (err) {
2939 err = errno;
2940 pr_err("%s: could not set level to %s\n",
2941 devname, c);
2942 if (err == EBUSY &&
2943 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2944 cont_err("Bitmap must be removed before level can be changed\n");
2945 return err;
2946 }
2947 if (verbose >= 0)
2948 pr_err("level of %s changed to %s\n",
2949 devname, c);
2950 }
2951 return 0;
2952 }
2953
2954 int sigterm = 0;
2955 static void catch_term(int sig)
2956 {
2957 sigterm = 1;
2958 }
2959
2960 static int continue_via_systemd(char *devnm)
2961 {
2962 int skipped, i, pid, status;
2963 char pathbuf[1024];
2964 /* In a systemd/udev world, it is best to get systemd to
2965 * run "mdadm --grow --continue" rather than running in the
2966 * background.
2967 */
2968 switch(fork()) {
2969 case 0:
2970 /* FIXME yuk. CLOSE_EXEC?? */
2971 skipped = 0;
2972 for (i = 3; skipped < 20; i++)
2973 if (close(i) < 0)
2974 skipped++;
2975 else
2976 skipped = 0;
2977
2978 /* Don't want to see error messages from
2979 * systemctl. If the service doesn't exist,
2980 * we fork ourselves.
2981 */
2982 close(2);
2983 open("/dev/null", O_WRONLY);
2984 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2985 devnm);
2986 status = execl("/usr/bin/systemctl", "systemctl",
2987 "start",
2988 pathbuf, NULL);
2989 status = execl("/bin/systemctl", "systemctl", "start",
2990 pathbuf, NULL);
2991 exit(1);
2992 case -1: /* Just do it ourselves. */
2993 break;
2994 default: /* parent - good */
2995 pid = wait(&status);
2996 if (pid >= 0 && status == 0)
2997 return 1;
2998 }
2999 return 0;
3000 }
3001
3002 static int reshape_array(char *container, int fd, char *devname,
3003 struct supertype *st, struct mdinfo *info,
3004 int force, struct mddev_dev *devlist,
3005 unsigned long long data_offset,
3006 char *backup_file, int verbose, int forked,
3007 int restart, int freeze_reshape)
3008 {
3009 struct reshape reshape;
3010 int spares_needed;
3011 char *msg;
3012 int orig_level = UnSet;
3013 int odisks;
3014 int delayed;
3015
3016 struct mdu_array_info_s array;
3017 char *c;
3018
3019 struct mddev_dev *dv;
3020 int added_disks;
3021
3022 int *fdlist = NULL;
3023 unsigned long long *offsets = NULL;
3024 int d;
3025 int nrdisks;
3026 int err;
3027 unsigned long blocks;
3028 unsigned long long array_size;
3029 int done;
3030 struct mdinfo *sra = NULL;
3031 char buf[20];
3032
3033 /* when reshaping a RAID0, the component_size might be zero.
3034 * So try to fix that up.
3035 */
3036 if (md_get_array_info(fd, &array) != 0) {
3037 dprintf("Cannot get array information.\n");
3038 goto release;
3039 }
3040 if (array.level == 0 && info->component_size == 0) {
3041 get_dev_size(fd, NULL, &array_size);
3042 info->component_size = array_size / array.raid_disks;
3043 }
3044
3045 if (array.level == 10)
3046 /* Need space_after info */
3047 get_space_after(fd, st, info);
3048
3049 if (info->reshape_active) {
3050 int new_level = info->new_level;
3051 info->new_level = UnSet;
3052 if (info->delta_disks > 0)
3053 info->array.raid_disks -= info->delta_disks;
3054 msg = analyse_change(devname, info, &reshape);
3055 info->new_level = new_level;
3056 if (info->delta_disks > 0)
3057 info->array.raid_disks += info->delta_disks;
3058 if (!restart)
3059 /* Make sure the array isn't read-only */
3060 ioctl(fd, RESTART_ARRAY_RW, 0);
3061 } else
3062 msg = analyse_change(devname, info, &reshape);
3063 if (msg) {
3064 /* if msg == "", error has already been printed */
3065 if (msg[0])
3066 pr_err("%s\n", msg);
3067 goto release;
3068 }
3069 if (restart &&
3070 (reshape.level != info->array.level ||
3071 reshape.before.layout != info->array.layout ||
3072 reshape.before.data_disks + reshape.parity
3073 != info->array.raid_disks - max(0, info->delta_disks))) {
3074 pr_err("reshape info is not in native format - cannot continue.\n");
3075 goto release;
3076 }
3077
3078 if (st->ss->external && restart && (info->reshape_progress == 0) &&
3079 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
3080 (strncmp(buf, "reshape", 7) == 0))) {
3081 /* When reshape is restarted from '0', very begin of array
3082 * it is possible that for external metadata reshape and array
3083 * configuration doesn't happen.
3084 * Check if md has the same opinion, and reshape is restarted
3085 * from 0. If so, this is regular reshape start after reshape
3086 * switch in metadata to next array only.
3087 */
3088 if ((verify_reshape_position(info, reshape.level) >= 0) &&
3089 (info->reshape_progress == 0))
3090 restart = 0;
3091 }
3092 if (restart) {
3093 /* reshape already started. just skip to monitoring the reshape */
3094 if (reshape.backup_blocks == 0)
3095 return 0;
3096 if (restart & RESHAPE_NO_BACKUP)
3097 return 0;
3098
3099 /* Need 'sra' down at 'started:' */
3100 sra = sysfs_read(fd, NULL,
3101 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3102 GET_CACHE);
3103 if (!sra) {
3104 pr_err("%s: Cannot get array details from sysfs\n",
3105 devname);
3106 goto release;
3107 }
3108
3109 if (!backup_file)
3110 backup_file = locate_backup(sra->sys_name);
3111
3112 goto started;
3113 }
3114 /* The container is frozen but the array may not be.
3115 * So freeze the array so spares don't get put to the wrong use
3116 * FIXME there should probably be a cleaner separation between
3117 * freeze_array and freeze_container.
3118 */
3119 sysfs_freeze_array(info);
3120 /* Check we have enough spares to not be degraded */
3121 added_disks = 0;
3122 for (dv = devlist; dv ; dv=dv->next)
3123 added_disks++;
3124 spares_needed = max(reshape.before.data_disks,
3125 reshape.after.data_disks)
3126 + reshape.parity - array.raid_disks;
3127
3128 if (!force &&
3129 info->new_level > 1 && info->array.level > 1 &&
3130 spares_needed > info->array.spare_disks + added_disks) {
3131 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3132 " Use --force to over-ride this check.\n",
3133 spares_needed,
3134 spares_needed == 1 ? "" : "s",
3135 info->array.spare_disks + added_disks);
3136 goto release;
3137 }
3138 /* Check we have enough spares to not fail */
3139 spares_needed = max(reshape.before.data_disks,
3140 reshape.after.data_disks)
3141 - array.raid_disks;
3142 if ((info->new_level > 1 || info->new_level == 0) &&
3143 spares_needed > info->array.spare_disks +added_disks) {
3144 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3145 spares_needed,
3146 spares_needed == 1 ? "" : "s",
3147 info->array.spare_disks + added_disks);
3148 goto release;
3149 }
3150
3151 if (reshape.level != array.level) {
3152 int err = impose_level(fd, reshape.level, devname, verbose);
3153 if (err)
3154 goto release;
3155 info->new_layout = UnSet; /* after level change,
3156 * layout is meaningless */
3157 orig_level = array.level;
3158 sysfs_freeze_array(info);
3159
3160 if (reshape.level > 0 && st->ss->external) {
3161 /* make sure mdmon is aware of the new level */
3162 if (mdmon_running(container))
3163 flush_mdmon(container);
3164
3165 if (!mdmon_running(container))
3166 start_mdmon(container);
3167 ping_monitor(container);
3168 if (mdmon_running(container) &&
3169 st->update_tail == NULL)
3170 st->update_tail = &st->updates;
3171 }
3172 }
3173 /* ->reshape_super might have chosen some spares from the
3174 * container that it wants to be part of the new array.
3175 * We can collect them with ->container_content and give
3176 * them to the kernel.
3177 */
3178 if (st->ss->reshape_super && st->ss->container_content) {
3179 char *subarray = strchr(info->text_version+1, '/')+1;
3180 struct mdinfo *info2 =
3181 st->ss->container_content(st, subarray);
3182 struct mdinfo *d;
3183
3184 if (info2) {
3185 if (sysfs_init(info2, fd, st->devnm)) {
3186 pr_err("unable to initialize sysfs for %s",
3187 st->devnm);
3188 free(info2);
3189 goto release;
3190 }
3191 /* When increasing number of devices, we need to set
3192 * new raid_disks before adding these, or they might
3193 * be rejected.
3194 */
3195 if (reshape.backup_blocks &&
3196 reshape.after.data_disks > reshape.before.data_disks)
3197 subarray_set_num(container, info2, "raid_disks",
3198 reshape.after.data_disks +
3199 reshape.parity);
3200 for (d = info2->devs; d; d = d->next) {
3201 if (d->disk.state == 0 &&
3202 d->disk.raid_disk >= 0) {
3203 /* This is a spare that wants to
3204 * be part of the array.
3205 */
3206 add_disk(fd, st, info2, d);
3207 }
3208 }
3209 sysfs_free(info2);
3210 }
3211 }
3212 /* We might have been given some devices to add to the
3213 * array. Now that the array has been changed to the right
3214 * level and frozen, we can safely add them.
3215 */
3216 if (devlist) {
3217 if (Manage_subdevs(devname, fd, devlist, verbose,
3218 0, NULL, 0))
3219 goto release;
3220 }
3221
3222 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3223 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3224 if (reshape.backup_blocks == 0) {
3225 /* No restriping needed, but we might need to impose
3226 * some more changes: layout, raid_disks, chunk_size
3227 */
3228 /* read current array info */
3229 if (md_get_array_info(fd, &array) != 0) {
3230 dprintf("Cannot get array information.\n");
3231 goto release;
3232 }
3233 /* compare current array info with new values and if
3234 * it is different update them to new */
3235 if (info->new_layout != UnSet &&
3236 info->new_layout != array.layout) {
3237 array.layout = info->new_layout;
3238 if (md_set_array_info(fd, &array) != 0) {
3239 pr_err("failed to set new layout\n");
3240 goto release;
3241 } else if (verbose >= 0)
3242 printf("layout for %s set to %d\n",
3243 devname, array.layout);
3244 }
3245 if (info->delta_disks != UnSet &&
3246 info->delta_disks != 0 &&
3247 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3248 array.raid_disks += info->delta_disks;
3249 if (md_set_array_info(fd, &array) != 0) {
3250 pr_err("failed to set raid disks\n");
3251 goto release;
3252 } else if (verbose >= 0) {
3253 printf("raid_disks for %s set to %d\n",
3254 devname, array.raid_disks);
3255 }
3256 }
3257 if (info->new_chunk != 0 &&
3258 info->new_chunk != array.chunk_size) {
3259 if (sysfs_set_num(info, NULL,
3260 "chunk_size", info->new_chunk) != 0) {
3261 pr_err("failed to set chunk size\n");
3262 goto release;
3263 } else if (verbose >= 0)
3264 printf("chunk size for %s set to %d\n",
3265 devname, array.chunk_size);
3266 }
3267 unfreeze(st);
3268 return 0;
3269 }
3270
3271 /*
3272 * There are three possibilities.
3273 * 1/ The array will shrink.
3274 * We need to ensure the reshape will pause before reaching
3275 * the 'critical section'. We also need to fork and wait for
3276 * that to happen. When it does we
3277 * suspend/backup/complete/unfreeze
3278 *
3279 * 2/ The array will not change size.
3280 * This requires that we keep a backup of a sliding window
3281 * so that we can restore data after a crash. So we need
3282 * to fork and monitor progress.
3283 * In future we will allow the data_offset to change, so
3284 * a sliding backup becomes unnecessary.
3285 *
3286 * 3/ The array will grow. This is relatively easy.
3287 * However the kernel's restripe routines will cheerfully
3288 * overwrite some early data before it is safe. So we
3289 * need to make a backup of the early parts of the array
3290 * and be ready to restore it if rebuild aborts very early.
3291 * For externally managed metadata, we still need a forked
3292 * child to monitor the reshape and suspend IO over the region
3293 * that is being reshaped.
3294 *
3295 * We backup data by writing it to one spare, or to a
3296 * file which was given on command line.
3297 *
3298 * In each case, we first make sure that storage is available
3299 * for the required backup.
3300 * Then we:
3301 * - request the shape change.
3302 * - fork to handle backup etc.
3303 */
3304 /* Check that we can hold all the data */
3305 get_dev_size(fd, NULL, &array_size);
3306 if (reshape.new_size < (array_size/512)) {
3307 pr_err("this change will reduce the size of the array.\n"
3308 " use --grow --array-size first to truncate array.\n"
3309 " e.g. mdadm --grow %s --array-size %llu\n",
3310 devname, reshape.new_size/2);
3311 goto release;
3312 }
3313
3314 if (array.level == 10) {
3315 /* Reshaping RAID10 does not require any data backup by
3316 * user-space. Instead it requires that the data_offset
3317 * is changed to avoid the need for backup.
3318 * So this is handled very separately
3319 */
3320 if (restart)
3321 /* Nothing to do. */
3322 return 0;
3323 return raid10_reshape(container, fd, devname, st, info,
3324 &reshape, data_offset,
3325 force, verbose);
3326 }
3327 sra = sysfs_read(fd, NULL,
3328 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3329 GET_CACHE);
3330 if (!sra) {
3331 pr_err("%s: Cannot get array details from sysfs\n",
3332 devname);
3333 goto release;
3334 }
3335
3336 if (!backup_file)
3337 switch(set_new_data_offset(sra, st, devname,
3338 reshape.after.data_disks - reshape.before.data_disks,
3339 data_offset,
3340 reshape.min_offset_change, 1)) {
3341 case -1:
3342 goto release;
3343 case 0:
3344 /* Updated data_offset, so it's easy now */
3345 update_cache_size(container, sra, info,
3346 min(reshape.before.data_disks,
3347 reshape.after.data_disks),
3348 reshape.backup_blocks);
3349
3350 /* Right, everything seems fine. Let's kick things off.
3351 */
3352 sync_metadata(st);
3353
3354 if (impose_reshape(sra, info, st, fd, restart,
3355 devname, container, &reshape) < 0)
3356 goto release;
3357 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3358 struct mdinfo *sd;
3359 if (errno != EINVAL) {
3360 pr_err("Failed to initiate reshape!\n");
3361 goto release;
3362 }
3363 /* revert data_offset and try the old way */
3364 for (sd = sra->devs; sd; sd = sd->next) {
3365 sysfs_set_num(sra, sd, "new_offset",
3366 sd->data_offset);
3367 sysfs_set_str(sra, NULL, "reshape_direction",
3368 "forwards");
3369 }
3370 break;
3371 }
3372 if (info->new_level == reshape.level)
3373 return 0;
3374 /* need to adjust level when reshape completes */
3375 switch(fork()) {
3376 case -1: /* ignore error, but don't wait */
3377 return 0;
3378 default: /* parent */
3379 return 0;
3380 case 0:
3381 map_fork();
3382 break;
3383 }
3384 close(fd);
3385 wait_reshape(sra);
3386 fd = open_dev(sra->sys_name);
3387 if (fd >= 0)
3388 impose_level(fd, info->new_level, devname, verbose);
3389 return 0;
3390 case 1: /* Couldn't set data_offset, try the old way */
3391 if (data_offset != INVALID_SECTORS) {
3392 pr_err("Cannot update data_offset on this array\n");
3393 goto release;
3394 }
3395 break;
3396 }
3397
3398 started:
3399 /* Decide how many blocks (sectors) for a reshape
3400 * unit. The number we have so far is just a minimum
3401 */
3402 blocks = reshape.backup_blocks;
3403 if (reshape.before.data_disks ==
3404 reshape.after.data_disks) {
3405 /* Make 'blocks' bigger for better throughput, but
3406 * not so big that we reject it below.
3407 * Try for 16 megabytes
3408 */
3409 while (blocks * 32 < sra->component_size &&
3410 blocks < 16*1024*2)
3411 blocks *= 2;
3412 } else
3413 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3414
3415 if (blocks >= sra->component_size/2) {
3416 pr_err("%s: Something wrong - reshape aborted\n",
3417 devname);
3418 goto release;
3419 }
3420
3421 /* Now we need to open all these devices so we can read/write.
3422 */
3423 nrdisks = max(reshape.before.data_disks,
3424 reshape.after.data_disks) + reshape.parity
3425 + sra->array.spare_disks;
3426 fdlist = xcalloc((1+nrdisks), sizeof(int));
3427 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3428
3429 odisks = reshape.before.data_disks + reshape.parity;
3430 d = reshape_prepare_fdlist(devname, sra, odisks,
3431 nrdisks, blocks, backup_file,
3432 fdlist, offsets);
3433 if (d < odisks) {
3434 goto release;
3435 }
3436 if ((st->ss->manage_reshape == NULL) ||
3437 (st->ss->recover_backup == NULL)) {
3438 if (backup_file == NULL) {
3439 if (reshape.after.data_disks <=
3440 reshape.before.data_disks) {
3441 pr_err("%s: Cannot grow - need backup-file\n",
3442 devname);
3443 pr_err(" Please provide one with \"--backup=...\"\n");
3444 goto release;
3445 } else if (d == odisks) {
3446 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3447 goto release;
3448 }
3449 } else {
3450 if (!reshape_open_backup_file(backup_file, fd, devname,
3451 (signed)blocks,
3452 fdlist+d, offsets+d,
3453 sra->sys_name,
3454 restart)) {
3455 goto release;
3456 }
3457 d++;
3458 }
3459 }
3460
3461 update_cache_size(container, sra, info,
3462 min(reshape.before.data_disks, reshape.after.data_disks),
3463 blocks);
3464
3465 /* Right, everything seems fine. Let's kick things off.
3466 * If only changing raid_disks, use ioctl, else use
3467 * sysfs.
3468 */
3469 sync_metadata(st);
3470
3471 if (impose_reshape(sra, info, st, fd, restart,
3472 devname, container, &reshape) < 0)
3473 goto release;
3474
3475 err = start_reshape(sra, restart, reshape.before.data_disks,
3476 reshape.after.data_disks);
3477 if (err) {
3478 pr_err("Cannot %s reshape for %s\n",
3479 restart ? "continue" : "start",
3480 devname);
3481 goto release;
3482 }
3483 if (restart)
3484 sysfs_set_str(sra, NULL, "array_state", "active");
3485 if (freeze_reshape) {
3486 free(fdlist);
3487 free(offsets);
3488 sysfs_free(sra);
3489 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3490 sra->reshape_progress);
3491 return 1;
3492 }
3493
3494 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3495 if (continue_via_systemd(container ?: sra->sys_name)) {
3496 free(fdlist);
3497 free(offsets);
3498 sysfs_free(sra);
3499 return 0;
3500 }
3501
3502 /* Now we just need to kick off the reshape and watch, while
3503 * handling backups of the data...
3504 * This is all done by a forked background process.
3505 */
3506 switch(forked ? 0 : fork()) {
3507 case -1:
3508 pr_err("Cannot run child to monitor reshape: %s\n",
3509 strerror(errno));
3510 abort_reshape(sra);
3511 goto release;
3512 default:
3513 free(fdlist);
3514 free(offsets);
3515 sysfs_free(sra);
3516 return 0;
3517 case 0:
3518 map_fork();
3519 break;
3520 }
3521
3522 /* If another array on the same devices is busy, the
3523 * reshape will wait for them. This would mean that
3524 * the first section that we suspend will stay suspended
3525 * for a long time. So check on that possibility
3526 * by looking for "DELAYED" in /proc/mdstat, and if found,
3527 * wait a while
3528 */
3529 do {
3530 struct mdstat_ent *mds, *m;
3531 delayed = 0;
3532 mds = mdstat_read(1, 0);
3533 for (m = mds; m; m = m->next)
3534 if (strcmp(m->devnm, sra->sys_name) == 0) {
3535 if (m->resync &&
3536 m->percent == RESYNC_DELAYED)
3537 delayed = 1;
3538 if (m->resync == 0)
3539 /* Haven't started the reshape thread
3540 * yet, wait a bit
3541 */
3542 delayed = 2;
3543 break;
3544 }
3545 free_mdstat(mds);
3546 if (delayed == 1 && get_linux_version() < 3007000) {
3547 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3548 " You might experience problems until other reshapes complete.\n");
3549 delayed = 0;
3550 }
3551 if (delayed)
3552 mdstat_wait(30 - (delayed-1) * 25);
3553 } while (delayed);
3554 mdstat_close();
3555 close(fd);
3556 if (check_env("MDADM_GROW_VERIFY"))
3557 fd = open(devname, O_RDONLY | O_DIRECT);
3558 else
3559 fd = -1;
3560 mlockall(MCL_FUTURE);
3561
3562 signal(SIGTERM, catch_term);
3563
3564 if (st->ss->external) {
3565 /* metadata handler takes it from here */
3566 done = st->ss->manage_reshape(
3567 fd, sra, &reshape, st, blocks,
3568 fdlist, offsets,
3569 d - odisks, fdlist+odisks,
3570 offsets+odisks);
3571 } else
3572 done = child_monitor(
3573 fd, sra, &reshape, st, blocks,
3574 fdlist, offsets,
3575 d - odisks, fdlist+odisks,
3576 offsets+odisks);
3577
3578 free(fdlist);
3579 free(offsets);
3580
3581 if (backup_file && done) {
3582 char *bul;
3583 bul = make_backup(sra->sys_name);
3584 if (bul) {
3585 char buf[1024];
3586 int l = readlink(bul, buf, sizeof(buf) - 1);
3587 if (l > 0) {
3588 buf[l]=0;
3589 unlink(buf);
3590 }
3591 unlink(bul);
3592 free(bul);
3593 }
3594 unlink(backup_file);
3595 }
3596 if (!done) {
3597 abort_reshape(sra);
3598 goto out;
3599 }
3600
3601 if (!st->ss->external &&
3602 !(reshape.before.data_disks != reshape.after.data_disks
3603 && info->custom_array_size) &&
3604 info->new_level == reshape.level &&
3605 !forked) {
3606 /* no need to wait for the reshape to finish as
3607 * there is nothing more to do.
3608 */
3609 sysfs_free(sra);
3610 exit(0);
3611 }
3612 wait_reshape(sra);
3613
3614 if (st->ss->external) {
3615 /* Re-load the metadata as much could have changed */
3616 int cfd = open_dev(st->container_devnm);
3617 if (cfd >= 0) {
3618 flush_mdmon(container);
3619 st->ss->free_super(st);
3620 st->ss->load_container(st, cfd, container);
3621 close(cfd);
3622 }
3623 }
3624
3625 /* set new array size if required customer_array_size is used
3626 * by this metadata.
3627 */
3628 if (reshape.before.data_disks !=
3629 reshape.after.data_disks &&
3630 info->custom_array_size)
3631 set_array_size(st, info, info->text_version);
3632
3633 if (info->new_level != reshape.level) {
3634 if (fd < 0)
3635 fd = open(devname, O_RDONLY);
3636 impose_level(fd, info->new_level, devname, verbose);
3637 close(fd);
3638 if (info->new_level == 0)
3639 st->update_tail = NULL;
3640 }
3641 out:
3642 sysfs_free(sra);
3643 if (forked)
3644 return 0;
3645 unfreeze(st);
3646 exit(0);
3647
3648 release:
3649 free(fdlist);
3650 free(offsets);
3651 if (orig_level != UnSet && sra) {
3652 c = map_num(pers, orig_level);
3653 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3654 pr_err("aborting level change\n");
3655 }
3656 sysfs_free(sra);
3657 if (!forked)
3658 unfreeze(st);
3659 return 1;
3660 }
3661
3662 /* mdfd handle is passed to be closed in child process (after fork).
3663 */
3664 int reshape_container(char *container, char *devname,
3665 int mdfd,
3666 struct supertype *st,
3667 struct mdinfo *info,
3668 int force,
3669 char *backup_file, int verbose,
3670 int forked, int restart, int freeze_reshape)
3671 {
3672 struct mdinfo *cc = NULL;
3673 int rv = restart;
3674 char last_devnm[32] = "";
3675
3676 /* component_size is not meaningful for a container,
3677 * so pass '0' meaning 'no change'
3678 */
3679 if (!restart &&
3680 reshape_super(st, 0, info->new_level,
3681 info->new_layout, info->new_chunk,
3682 info->array.raid_disks, info->delta_disks,
3683 backup_file, devname, APPLY_METADATA_CHANGES,
3684 verbose)) {
3685 unfreeze(st);
3686 return 1;
3687 }
3688
3689 sync_metadata(st);
3690
3691 /* ping monitor to be sure that update is on disk
3692 */
3693 ping_monitor(container);
3694
3695 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3696 if (continue_via_systemd(container))
3697 return 0;
3698
3699 switch (forked ? 0 : fork()) {
3700 case -1: /* error */
3701 perror("Cannot fork to complete reshape\n");
3702 unfreeze(st);
3703 return 1;
3704 default: /* parent */
3705 if (!freeze_reshape)
3706 printf("%s: multi-array reshape continues in background\n", Name);
3707 return 0;
3708 case 0: /* child */
3709 map_fork();
3710 break;
3711 }
3712
3713 /* close unused handle in child process
3714 */
3715 if (mdfd > -1)
3716 close(mdfd);
3717
3718 while(1) {
3719 /* For each member array with reshape_active,
3720 * we need to perform the reshape.
3721 * We pick the first array that needs reshaping and
3722 * reshape it. reshape_array() will re-read the metadata
3723 * so the next time through a different array should be
3724 * ready for reshape.
3725 * It is possible that the 'different' array will not
3726 * be assembled yet. In that case we simple exit.
3727 * When it is assembled, the mdadm which assembles it
3728 * will take over the reshape.
3729 */
3730 struct mdinfo *content;
3731 int fd;
3732 struct mdstat_ent *mdstat;
3733 char *adev;
3734 dev_t devid;
3735
3736 sysfs_free(cc);
3737
3738 cc = st->ss->container_content(st, NULL);
3739
3740 for (content = cc; content ; content = content->next) {
3741 char *subarray;
3742 if (!content->reshape_active)
3743 continue;
3744
3745 subarray = strchr(content->text_version+1, '/')+1;
3746 mdstat = mdstat_by_subdev(subarray, container);
3747 if (!mdstat)
3748 continue;
3749 if (mdstat->active == 0) {
3750 pr_err("Skipping inactive array %s.\n",
3751 mdstat->devnm);
3752 free_mdstat(mdstat);
3753 mdstat = NULL;
3754 continue;
3755 }
3756 break;
3757 }
3758 if (!content)
3759 break;
3760
3761 devid = devnm2devid(mdstat->devnm);
3762 adev = map_dev(major(devid), minor(devid), 0);
3763 if (!adev)
3764 adev = content->text_version;
3765
3766 fd = open_dev(mdstat->devnm);
3767 if (fd < 0) {
3768 pr_err("Device %s cannot be opened for reshape.\n", adev);
3769 break;
3770 }
3771
3772 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3773 /* Do not allow for multiple reshape_array() calls for
3774 * the same array.
3775 * It can happen when reshape_array() returns without
3776 * error, when reshape is not finished (wrong reshape
3777 * starting/continuation conditions). Mdmon doesn't
3778 * switch to next array in container and reentry
3779 * conditions for the same array occur.
3780 * This is possibly interim until the behaviour of
3781 * reshape_array is resolved().
3782 */
3783 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3784 close(fd);
3785 break;
3786 }
3787 strcpy(last_devnm, mdstat->devnm);
3788
3789 if (sysfs_init(content, fd, mdstat->devnm)) {
3790 pr_err("Unable to initialize sysfs for %s\n",
3791 mdstat->devnm);
3792 rv = 1;
3793 break;
3794 }
3795
3796 if (mdmon_running(container))
3797 flush_mdmon(container);
3798
3799 rv = reshape_array(container, fd, adev, st,
3800 content, force, NULL, INVALID_SECTORS,
3801 backup_file, verbose, 1, restart,
3802 freeze_reshape);
3803 close(fd);
3804
3805 if (freeze_reshape) {
3806 sysfs_free(cc);
3807 exit(0);
3808 }
3809
3810 restart = 0;
3811 if (rv)
3812 break;
3813
3814 if (mdmon_running(container))
3815 flush_mdmon(container);
3816 }
3817 if (!rv)
3818 unfreeze(st);
3819 sysfs_free(cc);
3820 exit(0);
3821 }
3822
3823 /*
3824 * We run a child process in the background which performs the following
3825 * steps:
3826 * - wait for resync to reach a certain point
3827 * - suspend io to the following section
3828 * - backup that section
3829 * - allow resync to proceed further
3830 * - resume io
3831 * - discard the backup.
3832 *
3833 * When are combined in slightly different ways in the three cases.
3834 * Grow:
3835 * - suspend/backup/allow/wait/resume/discard
3836 * Shrink:
3837 * - allow/wait/suspend/backup/allow/wait/resume/discard
3838 * same-size:
3839 * - wait/resume/discard/suspend/backup/allow
3840 *
3841 * suspend/backup/allow always come together
3842 * wait/resume/discard do too.
3843 * For the same-size case we have two backups to improve flow.
3844 *
3845 */
3846
3847 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3848 unsigned long long backup_point,
3849 unsigned long long wait_point,
3850 unsigned long long *suspend_point,
3851 unsigned long long *reshape_completed, int *frozen)
3852 {
3853 /* This function is called repeatedly by the reshape manager.
3854 * It determines how much progress can safely be made and allows
3855 * that progress.
3856 * - 'info' identifies the array and particularly records in
3857 * ->reshape_progress the metadata's knowledge of progress
3858 * This is a sector offset from the start of the array
3859 * of the next array block to be relocated. This number
3860 * may increase from 0 or decrease from array_size, depending
3861 * on the type of reshape that is happening.
3862 * Note that in contrast, 'sync_completed' is a block count of the
3863 * reshape so far. It gives the distance between the start point
3864 * (head or tail of device) and the next place that data will be
3865 * written. It always increases.
3866 * - 'reshape' is the structure created by analyse_change
3867 * - 'backup_point' shows how much the metadata manager has backed-up
3868 * data. For reshapes with increasing progress, it is the next address
3869 * to be backed up, previous addresses have been backed-up. For
3870 * decreasing progress, it is the earliest address that has been
3871 * backed up - later address are also backed up.
3872 * So addresses between reshape_progress and backup_point are
3873 * backed up providing those are in the 'correct' order.
3874 * - 'wait_point' is an array address. When reshape_completed
3875 * passes this point, progress_reshape should return. It might
3876 * return earlier if it determines that ->reshape_progress needs
3877 * to be updated or further backup is needed.
3878 * - suspend_point is maintained by progress_reshape and the caller
3879 * should not touch it except to initialise to zero.
3880 * It is an array address and it only increases in 2.6.37 and earlier.
3881 * This makes it difficult to handle reducing reshapes with
3882 * external metadata.
3883 * However: it is similar to backup_point in that it records the
3884 * other end of a suspended region from reshape_progress.
3885 * it is moved to extend the region that is safe to backup and/or
3886 * reshape
3887 * - reshape_completed is read from sysfs and returned. The caller
3888 * should copy this into ->reshape_progress when it has reason to
3889 * believe that the metadata knows this, and any backup outside this
3890 * has been erased.
3891 *
3892 * Return value is:
3893 * 1 if more data from backup_point - but only as far as suspend_point,
3894 * should be backed up
3895 * 0 if things are progressing smoothly
3896 * -1 if the reshape is finished because it is all done,
3897 * -2 if the reshape is finished due to an error.
3898 */
3899
3900 int advancing = (reshape->after.data_disks
3901 >= reshape->before.data_disks);
3902 unsigned long long need_backup; /* All data between start of array and
3903 * here will at some point need to
3904 * be backed up.
3905 */
3906 unsigned long long read_offset, write_offset;
3907 unsigned long long write_range;
3908 unsigned long long max_progress, target, completed;
3909 unsigned long long array_size = (info->component_size
3910 * reshape->before.data_disks);
3911 int fd;
3912 char buf[20];
3913
3914 /* First, we unsuspend any region that is now known to be safe.
3915 * If suspend_point is on the 'wrong' side of reshape_progress, then
3916 * we don't have or need suspension at the moment. This is true for
3917 * native metadata when we don't need to back-up.
3918 */
3919 if (advancing) {
3920 if (info->reshape_progress <= *suspend_point)
3921 sysfs_set_num(info, NULL, "suspend_lo",
3922 info->reshape_progress);
3923 } else {
3924 /* Note: this won't work in 2.6.37 and before.
3925 * Something somewhere should make sure we don't need it!
3926 */
3927 if (info->reshape_progress >= *suspend_point)
3928 sysfs_set_num(info, NULL, "suspend_hi",
3929 info->reshape_progress);
3930 }
3931
3932 /* Now work out how far it is safe to progress.
3933 * If the read_offset for ->reshape_progress is less than
3934 * 'blocks' beyond the write_offset, we can only progress as far
3935 * as a backup.
3936 * Otherwise we can progress until the write_offset for the new location
3937 * reaches (within 'blocks' of) the read_offset at the current location.
3938 * However that region must be suspended unless we are using native
3939 * metadata.
3940 * If we need to suspend more, we limit it to 128M per device, which is
3941 * rather arbitrary and should be some time-based calculation.
3942 */
3943 read_offset = info->reshape_progress / reshape->before.data_disks;
3944 write_offset = info->reshape_progress / reshape->after.data_disks;
3945 write_range = info->new_chunk/512;
3946 if (reshape->before.data_disks == reshape->after.data_disks)
3947 need_backup = array_size;
3948 else
3949 need_backup = reshape->backup_blocks;
3950 if (advancing) {
3951 if (read_offset < write_offset + write_range)
3952 max_progress = backup_point;
3953 else
3954 max_progress =
3955 read_offset *
3956 reshape->after.data_disks;
3957 } else {
3958 if (read_offset > write_offset - write_range)
3959 /* Can only progress as far as has been backed up,
3960 * which must be suspended */
3961 max_progress = backup_point;
3962 else if (info->reshape_progress <= need_backup)
3963 max_progress = backup_point;
3964 else {
3965 if (info->array.major_version >= 0)
3966 /* Can progress until backup is needed */
3967 max_progress = need_backup;
3968 else {
3969 /* Can progress until metadata update is required */
3970 max_progress =
3971 read_offset *
3972 reshape->after.data_disks;
3973 /* but data must be suspended */
3974 if (max_progress < *suspend_point)
3975 max_progress = *suspend_point;
3976 }
3977 }
3978 }
3979
3980 /* We know it is safe to progress to 'max_progress' providing
3981 * it is suspended or we are using native metadata.
3982 * Consider extending suspend_point 128M per device if it
3983 * is less than 64M per device beyond reshape_progress.
3984 * But always do a multiple of 'blocks'
3985 * FIXME this is too big - it takes to long to complete
3986 * this much.
3987 */
3988 target = 64*1024*2 * min(reshape->before.data_disks,
3989 reshape->after.data_disks);
3990 target /= reshape->backup_blocks;
3991 if (target < 2)
3992 target = 2;
3993 target *= reshape->backup_blocks;
3994
3995 /* For externally managed metadata we always need to suspend IO to
3996 * the area being reshaped so we regularly push suspend_point forward.
3997 * For native metadata we only need the suspend if we are going to do
3998 * a backup.
3999 */
4000 if (advancing) {
4001 if ((need_backup > info->reshape_progress
4002 || info->array.major_version < 0) &&
4003 *suspend_point < info->reshape_progress + target) {
4004 if (need_backup < *suspend_point + 2 * target)
4005 *suspend_point = need_backup;
4006 else if (*suspend_point + 2 * target < array_size)
4007 *suspend_point += 2 * target;
4008 else
4009 *suspend_point = array_size;
4010 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
4011 if (max_progress > *suspend_point)
4012 max_progress = *suspend_point;
4013 }
4014 } else {
4015 if (info->array.major_version >= 0) {
4016 /* Only need to suspend when about to backup */
4017 if (info->reshape_progress < need_backup * 2 &&
4018 *suspend_point > 0) {
4019 *suspend_point = 0;
4020 sysfs_set_num(info, NULL, "suspend_lo", 0);
4021 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
4022 }
4023 } else {
4024 /* Need to suspend continually */
4025 if (info->reshape_progress < *suspend_point)
4026 *suspend_point = info->reshape_progress;
4027 if (*suspend_point + target < info->reshape_progress)
4028 /* No need to move suspend region yet */;
4029 else {
4030 if (*suspend_point >= 2 * target)
4031 *suspend_point -= 2 * target;
4032 else
4033 *suspend_point = 0;
4034 sysfs_set_num(info, NULL, "suspend_lo",
4035 *suspend_point);
4036 }
4037 if (max_progress < *suspend_point)
4038 max_progress = *suspend_point;
4039 }
4040 }
4041
4042 /* now set sync_max to allow that progress. sync_max, like
4043 * sync_completed is a count of sectors written per device, so
4044 * we find the difference between max_progress and the start point,
4045 * and divide that by after.data_disks to get a sync_max
4046 * number.
4047 * At the same time we convert wait_point to a similar number
4048 * for comparing against sync_completed.
4049 */
4050 /* scale down max_progress to per_disk */
4051 max_progress /= reshape->after.data_disks;
4052 /* Round to chunk size as some kernels give an erroneously high number */
4053 max_progress /= info->new_chunk/512;
4054 max_progress *= info->new_chunk/512;
4055 /* And round to old chunk size as the kernel wants that */
4056 max_progress /= info->array.chunk_size/512;
4057 max_progress *= info->array.chunk_size/512;
4058 /* Limit progress to the whole device */
4059 if (max_progress > info->component_size)
4060 max_progress = info->component_size;
4061 wait_point /= reshape->after.data_disks;
4062 if (!advancing) {
4063 /* switch from 'device offset' to 'processed block count' */
4064 max_progress = info->component_size - max_progress;
4065 wait_point = info->component_size - wait_point;
4066 }
4067
4068 if (!*frozen)
4069 sysfs_set_num(info, NULL, "sync_max", max_progress);
4070
4071 /* Now wait. If we have already reached the point that we were
4072 * asked to wait to, don't wait at all, else wait for any change.
4073 * We need to select on 'sync_completed' as that is the place that
4074 * notifications happen, but we are really interested in
4075 * 'reshape_position'
4076 */
4077 fd = sysfs_get_fd(info, NULL, "sync_completed");
4078 if (fd < 0)
4079 goto check_progress;
4080
4081 if (sysfs_fd_get_ll(fd, &completed) < 0)
4082 goto check_progress;
4083
4084 while (completed < max_progress && completed < wait_point) {
4085 /* Check that sync_action is still 'reshape' to avoid
4086 * waiting forever on a dead array
4087 */
4088 char action[20];
4089 if (sysfs_get_str(info, NULL, "sync_action",
4090 action, 20) <= 0 ||
4091 strncmp(action, "reshape", 7) != 0)
4092 break;
4093 /* Some kernels reset 'sync_completed' to zero
4094 * before setting 'sync_action' to 'idle'.
4095 * So we need these extra tests.
4096 */
4097 if (completed == 0 && advancing
4098 && strncmp(action, "idle", 4) == 0
4099 && info->reshape_progress > 0)
4100 break;
4101 if (completed == 0 && !advancing
4102 && strncmp(action, "idle", 4) == 0
4103 && info->reshape_progress < (info->component_size
4104 * reshape->after.data_disks))
4105 break;
4106 sysfs_wait(fd, NULL);
4107 if (sysfs_fd_get_ll(fd, &completed) < 0)
4108 goto check_progress;
4109 }
4110 /* Some kernels reset 'sync_completed' to zero,
4111 * we need to have real point we are in md.
4112 * So in that case, read 'reshape_position' from sysfs.
4113 */
4114 if (completed == 0) {
4115 unsigned long long reshapep;
4116 char action[20];
4117 if (sysfs_get_str(info, NULL, "sync_action",
4118 action, 20) > 0 &&
4119 strncmp(action, "idle", 4) == 0 &&
4120 sysfs_get_ll(info, NULL,
4121 "reshape_position", &reshapep) == 0)
4122 *reshape_completed = reshapep;
4123 } else {
4124 /* some kernels can give an incorrectly high
4125 * 'completed' number, so round down */
4126 completed /= (info->new_chunk/512);
4127 completed *= (info->new_chunk/512);
4128 /* Convert 'completed' back in to a 'progress' number */
4129 completed *= reshape->after.data_disks;
4130 if (!advancing)
4131 completed = (info->component_size
4132 * reshape->after.data_disks
4133 - completed);
4134 *reshape_completed = completed;
4135 }
4136
4137 close(fd);
4138
4139 /* We return the need_backup flag. Caller will decide
4140 * how much - a multiple of ->backup_blocks up to *suspend_point
4141 */
4142 if (advancing)
4143 return need_backup > info->reshape_progress;
4144 else
4145 return need_backup >= info->reshape_progress;
4146
4147 check_progress:
4148 /* if we couldn't read a number from sync_completed, then
4149 * either the reshape did complete, or it aborted.
4150 * We can tell which by checking for 'none' in reshape_position.
4151 * If it did abort, then it might immediately restart if it
4152 * it was just a device failure that leaves us degraded but
4153 * functioning.
4154 */
4155 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
4156 || strncmp(buf, "none", 4) != 0) {
4157 /* The abort might only be temporary. Wait up to 10
4158 * seconds for fd to contain a valid number again.
4159 */
4160 int wait = 10000;
4161 int rv = -2;
4162 unsigned long long new_sync_max;
4163 while (fd >= 0 && rv < 0 && wait > 0) {
4164 if (sysfs_wait(fd, &wait) != 1)
4165 break;
4166 switch (sysfs_fd_get_ll(fd, &completed)) {
4167 case 0:
4168 /* all good again */
4169 rv = 1;
4170 /* If "sync_max" is no longer max_progress
4171 * we need to freeze things
4172 */
4173 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
4174 *frozen = (new_sync_max != max_progress);
4175 break;
4176 case -2: /* read error - abort */
4177 wait = 0;
4178 break;
4179 }
4180 }
4181 if (fd >= 0)
4182 close(fd);
4183 return rv; /* abort */
4184 } else {
4185 /* Maybe racing with array shutdown - check state */
4186 if (fd >= 0)
4187 close(fd);
4188 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
4189 || strncmp(buf, "inactive", 8) == 0
4190 || strncmp(buf, "clear",5) == 0)
4191 return -2; /* abort */
4192 return -1; /* complete */
4193 }
4194 }
4195
4196 /* FIXME return status is never checked */
4197 static int grow_backup(struct mdinfo *sra,
4198 unsigned long long offset, /* per device */
4199 unsigned long stripes, /* per device, in old chunks */
4200 int *sources, unsigned long long *offsets,
4201 int disks, int chunk, int level, int layout,
4202 int dests, int *destfd, unsigned long long *destoffsets,
4203 int part, int *degraded,
4204 char *buf)
4205 {
4206 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4207 * to storage 'destfd' (offset 'destoffsets'), after first
4208 * suspending IO. Then allow resync to continue
4209 * over the suspended section.
4210 * Use part 'part' of the backup-super-block.
4211 */
4212 int odata = disks;
4213 int rv = 0;
4214 int i;
4215 unsigned long long ll;
4216 int new_degraded;
4217 //printf("offset %llu\n", offset);
4218 if (level >= 4)
4219 odata--;
4220 if (level == 6)
4221 odata--;
4222
4223 /* Check that array hasn't become degraded, else we might backup the wrong data */
4224 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4225 return -1; /* FIXME this error is ignored */
4226 new_degraded = (int)ll;
4227 if (new_degraded != *degraded) {
4228 /* check each device to ensure it is still working */
4229 struct mdinfo *sd;
4230 for (sd = sra->devs ; sd ; sd = sd->next) {
4231 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4232 continue;
4233 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4234 char sbuf[100];
4235
4236 if (sysfs_get_str(sra, sd, "state",
4237 sbuf, sizeof(sbuf)) < 0 ||
4238 strstr(sbuf, "faulty") ||
4239 strstr(sbuf, "in_sync") == NULL) {
4240 /* this device is dead */
4241 sd->disk.state = (1<<MD_DISK_FAULTY);
4242 if (sd->disk.raid_disk >= 0 &&
4243 sources[sd->disk.raid_disk] >= 0) {
4244 close(sources[sd->disk.raid_disk]);
4245 sources[sd->disk.raid_disk] = -1;
4246 }
4247 }
4248 }
4249 }
4250 *degraded = new_degraded;
4251 }
4252 if (part) {
4253 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4254 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4255 } else {
4256 bsb.arraystart = __cpu_to_le64(offset * odata);
4257 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4258 }
4259 if (part)
4260 bsb.magic[15] = '2';
4261 for (i = 0; i < dests; i++)
4262 if (part)
4263 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4264 else
4265 lseek64(destfd[i], destoffsets[i], 0);
4266
4267 rv = save_stripes(sources, offsets,
4268 disks, chunk, level, layout,
4269 dests, destfd,
4270 offset*512*odata, stripes * chunk * odata,
4271 buf);
4272
4273 if (rv)
4274 return rv;
4275 bsb.mtime = __cpu_to_le64(time(0));
4276 for (i = 0; i < dests; i++) {
4277 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4278
4279 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4280 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4281 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4282 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4283
4284 rv = -1;
4285 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4286 != destoffsets[i] - 4096)
4287 break;
4288 if (write(destfd[i], &bsb, 512) != 512)
4289 break;
4290 if (destoffsets[i] > 4096) {
4291 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4292 destoffsets[i]+stripes*chunk*odata)
4293 break;
4294 if (write(destfd[i], &bsb, 512) != 512)
4295 break;
4296 }
4297 fsync(destfd[i]);
4298 rv = 0;
4299 }
4300
4301 return rv;
4302 }
4303
4304 /* in 2.6.30, the value reported by sync_completed can be
4305 * less that it should be by one stripe.
4306 * This only happens when reshape hits sync_max and pauses.
4307 * So allow wait_backup to either extent sync_max further
4308 * than strictly necessary, or return before the
4309 * sync has got quite as far as we would really like.
4310 * This is what 'blocks2' is for.
4311 * The various caller give appropriate values so that
4312 * every works.
4313 */
4314 /* FIXME return value is often ignored */
4315 static int forget_backup(int dests, int *destfd,
4316 unsigned long long *destoffsets,
4317 int part)
4318 {
4319 /*
4320 * Erase backup 'part' (which is 0 or 1)
4321 */
4322 int i;
4323 int rv;
4324
4325 if (part) {
4326 bsb.arraystart2 = __cpu_to_le64(0);
4327 bsb.length2 = __cpu_to_le64(0);
4328 } else {
4329 bsb.arraystart = __cpu_to_le64(0);
4330 bsb.length = __cpu_to_le64(0);
4331 }
4332 bsb.mtime = __cpu_to_le64(time(0));
4333 rv = 0;
4334 for (i = 0; i < dests; i++) {
4335 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4336 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4337 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4338 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4339 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4340 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4341 destoffsets[i]-4096)
4342 rv = -1;
4343 if (rv == 0 &&
4344 write(destfd[i], &bsb, 512) != 512)
4345 rv = -1;
4346 fsync(destfd[i]);
4347 }
4348 return rv;
4349 }
4350
4351 static void fail(char *msg)
4352 {
4353 int rv;
4354 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4355 rv |= (write(2, "\n", 1) != 1);
4356 exit(rv ? 1 : 2);
4357 }
4358
4359 static char *abuf, *bbuf;
4360 static unsigned long long abuflen;
4361 static void validate(int afd, int bfd, unsigned long long offset)
4362 {
4363 /* check that the data in the backup against the array.
4364 * This is only used for regression testing and should not
4365 * be used while the array is active
4366 */
4367 if (afd < 0)
4368 return;
4369 lseek64(bfd, offset - 4096, 0);
4370 if (read(bfd, &bsb2, 512) != 512)
4371 fail("cannot read bsb");
4372 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4373 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4374 fail("first csum bad");
4375 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4376 fail("magic is bad");
4377 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4378 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4379 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4380 fail("second csum bad");
4381
4382 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4383 fail("devstart is wrong");
4384
4385 if (bsb2.length) {
4386 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4387
4388 if (abuflen < len) {
4389 free(abuf);
4390 free(bbuf);
4391 abuflen = len;
4392 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4393 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4394 abuflen = 0;
4395 /* just stop validating on mem-alloc failure */
4396 return;
4397 }
4398 }
4399
4400 lseek64(bfd, offset, 0);
4401 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4402 //printf("len %llu\n", len);
4403 fail("read first backup failed");
4404 }
4405 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4406 if ((unsigned long long)read(afd, abuf, len) != len)
4407 fail("read first from array failed");
4408 if (memcmp(bbuf, abuf, len) != 0) {
4409 #if 0
4410 int i;
4411 printf("offset=%llu len=%llu\n",
4412 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4413 for (i=0; i<len; i++)
4414 if (bbuf[i] != abuf[i]) {
4415 printf("first diff byte %d\n", i);
4416 break;
4417 }
4418 #endif
4419 fail("data1 compare failed");
4420 }
4421 }
4422 if (bsb2.length2) {
4423 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4424
4425 if (abuflen < len) {
4426 free(abuf);
4427 free(bbuf);
4428 abuflen = len;
4429 abuf = xmalloc(abuflen);
4430 bbuf = xmalloc(abuflen);
4431 }
4432
4433 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4434 if ((unsigned long long)read(bfd, bbuf, len) != len)
4435 fail("read second backup failed");
4436 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4437 if ((unsigned long long)read(afd, abuf, len) != len)
4438 fail("read second from array failed");
4439 if (memcmp(bbuf, abuf, len) != 0)
4440 fail("data2 compare failed");
4441 }
4442 }
4443
4444 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4445 struct supertype *st, unsigned long blocks,
4446 int *fds, unsigned long long *offsets,
4447 int dests, int *destfd, unsigned long long *destoffsets)
4448 {
4449 /* Monitor a reshape where backup is being performed using
4450 * 'native' mechanism - either to a backup file, or
4451 * to some space in a spare.
4452 */
4453 char *buf;
4454 int degraded = -1;
4455 unsigned long long speed;
4456 unsigned long long suspend_point, array_size;
4457 unsigned long long backup_point, wait_point;
4458 unsigned long long reshape_completed;
4459 int done = 0;
4460 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4461 int part = 0; /* The next part of the backup area to fill. It may already
4462 * be full, so we need to check */
4463 int level = reshape->level;
4464 int layout = reshape->before.layout;
4465 int data = reshape->before.data_disks;
4466 int disks = reshape->before.data_disks + reshape->parity;
4467 int chunk = sra->array.chunk_size;
4468 struct mdinfo *sd;
4469 unsigned long stripes;
4470 int uuid[4];
4471 int frozen = 0;
4472
4473 /* set up the backup-super-block. This requires the
4474 * uuid from the array.
4475 */
4476 /* Find a superblock */
4477 for (sd = sra->devs; sd; sd = sd->next) {
4478 char *dn;
4479 int devfd;
4480 int ok;
4481 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4482 continue;
4483 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4484 devfd = dev_open(dn, O_RDONLY);
4485 if (devfd < 0)
4486 continue;
4487 ok = st->ss->load_super(st, devfd, NULL);
4488 close(devfd);
4489 if (ok == 0)
4490 break;
4491 }
4492 if (!sd) {
4493 pr_err("Cannot find a superblock\n");
4494 return 0;
4495 }
4496
4497 memset(&bsb, 0, 512);
4498 memcpy(bsb.magic, "md_backup_data-1", 16);
4499 st->ss->uuid_from_super(st, uuid);
4500 memcpy(bsb.set_uuid, uuid, 16);
4501 bsb.mtime = __cpu_to_le64(time(0));
4502 bsb.devstart2 = blocks;
4503
4504 stripes = blocks / (sra->array.chunk_size/512) /
4505 reshape->before.data_disks;
4506
4507 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4508 /* Don't start the 'reshape' */
4509 return 0;
4510 if (reshape->before.data_disks == reshape->after.data_disks) {
4511 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4512 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4513 }
4514
4515 if (increasing) {
4516 array_size = sra->component_size * reshape->after.data_disks;
4517 backup_point = sra->reshape_progress;
4518 suspend_point = 0;
4519 } else {
4520 array_size = sra->component_size * reshape->before.data_disks;
4521 backup_point = reshape->backup_blocks;
4522 suspend_point = array_size;
4523 }
4524
4525 while (!done) {
4526 int rv;
4527
4528 /* Want to return as soon the oldest backup slot can
4529 * be released as that allows us to start backing up
4530 * some more, providing suspend_point has been
4531 * advanced, which it should have.
4532 */
4533 if (increasing) {
4534 wait_point = array_size;
4535 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4536 wait_point = (__le64_to_cpu(bsb.arraystart) +
4537 __le64_to_cpu(bsb.length));
4538 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4539 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4540 __le64_to_cpu(bsb.length2));
4541 } else {
4542 wait_point = 0;
4543 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4544 wait_point = __le64_to_cpu(bsb.arraystart);
4545 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4546 wait_point = __le64_to_cpu(bsb.arraystart2);
4547 }
4548
4549 reshape_completed = sra->reshape_progress;
4550 rv = progress_reshape(sra, reshape,
4551 backup_point, wait_point,
4552 &suspend_point, &reshape_completed,
4553 &frozen);
4554 /* external metadata would need to ping_monitor here */
4555 sra->reshape_progress = reshape_completed;
4556
4557 /* Clear any backup region that is before 'here' */
4558 if (increasing) {
4559 if (__le64_to_cpu(bsb.length) > 0 &&
4560 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4561 __le64_to_cpu(bsb.length)))
4562 forget_backup(dests, destfd,
4563 destoffsets, 0);
4564 if (__le64_to_cpu(bsb.length2) > 0 &&
4565 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4566 __le64_to_cpu(bsb.length2)))
4567 forget_backup(dests, destfd,
4568 destoffsets, 1);
4569 } else {
4570 if (__le64_to_cpu(bsb.length) > 0 &&
4571 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4572 forget_backup(dests, destfd,
4573 destoffsets, 0);
4574 if (__le64_to_cpu(bsb.length2) > 0 &&
4575 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4576 forget_backup(dests, destfd,
4577 destoffsets, 1);
4578 }
4579 if (sigterm)
4580 rv = -2;
4581 if (rv < 0) {
4582 if (rv == -1)
4583 done = 1;
4584 break;
4585 }
4586 if (rv == 0 && increasing && !st->ss->external) {
4587 /* No longer need to monitor this reshape */
4588 sysfs_set_str(sra, NULL, "sync_max", "max");
4589 done = 1;
4590 break;
4591 }
4592
4593 while (rv) {
4594 unsigned long long offset;
4595 unsigned long actual_stripes;
4596 /* Need to backup some data.
4597 * If 'part' is not used and the desired
4598 * backup size is suspended, do a backup,
4599 * then consider the next part.
4600 */
4601 /* Check that 'part' is unused */
4602 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4603 break;
4604 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4605 break;
4606
4607 offset = backup_point / data;
4608 actual_stripes = stripes;
4609 if (increasing) {
4610 if (offset + actual_stripes * (chunk/512) >
4611 sra->component_size)
4612 actual_stripes = ((sra->component_size - offset)
4613 / (chunk/512));
4614 if (offset + actual_stripes * (chunk/512) >
4615 suspend_point/data)
4616 break;
4617 } else {
4618 if (offset < actual_stripes * (chunk/512))
4619 actual_stripes = offset / (chunk/512);
4620 offset -= actual_stripes * (chunk/512);
4621 if (offset < suspend_point/data)
4622 break;
4623 }
4624 if (actual_stripes == 0)
4625 break;
4626 grow_backup(sra, offset, actual_stripes,
4627 fds, offsets,
4628 disks, chunk, level, layout,
4629 dests, destfd, destoffsets,
4630 part, &degraded, buf);
4631 validate(afd, destfd[0], destoffsets[0]);
4632 /* record where 'part' is up to */
4633 part = !part;
4634 if (increasing)
4635 backup_point += actual_stripes * (chunk/512) * data;
4636 else
4637 backup_point -= actual_stripes * (chunk/512) * data;
4638 }
4639 }
4640
4641 /* FIXME maybe call progress_reshape one more time instead */
4642 /* remove any remaining suspension */
4643 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4644 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4645 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4646 sysfs_set_num(sra, NULL, "sync_min", 0);
4647
4648 if (reshape->before.data_disks == reshape->after.data_disks)
4649 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4650 free(buf);
4651 return done;
4652 }
4653
4654 /*
4655 * If any spare contains md_back_data-1 which is recent wrt mtime,
4656 * write that data into the array and update the super blocks with
4657 * the new reshape_progress
4658 */
4659 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4660 char *backup_file, int verbose)
4661 {
4662 int i, j;
4663 int old_disks;
4664 unsigned long long *offsets;
4665 unsigned long long nstripe, ostripe;
4666 int ndata, odata;
4667
4668 odata = info->array.raid_disks - info->delta_disks - 1;
4669 if (info->array.level == 6) odata--; /* number of data disks */
4670 ndata = info->array.raid_disks - 1;
4671 if (info->new_level == 6) ndata--;
4672
4673 old_disks = info->array.raid_disks - info->delta_disks;
4674
4675 if (info->delta_disks <= 0)
4676 /* Didn't grow, so the backup file must have
4677 * been used
4678 */
4679 old_disks = cnt;
4680 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4681 struct mdinfo dinfo;
4682 int fd;
4683 int bsbsize;
4684 char *devname, namebuf[20];
4685 unsigned long long lo, hi;
4686
4687 /* This was a spare and may have some saved data on it.
4688 * Load the superblock, find and load the
4689 * backup_super_block.
4690 * If either fail, go on to next device.
4691 * If the backup contains no new info, just return
4692 * else restore data and update all superblocks
4693 */
4694 if (i == old_disks-1) {
4695 fd = open(backup_file, O_RDONLY);
4696 if (fd<0) {
4697 pr_err("backup file %s inaccessible: %s\n",
4698 backup_file, strerror(errno));
4699 continue;
4700 }
4701 devname = backup_file;
4702 } else {
4703 fd = fdlist[i];
4704 if (fd < 0)
4705 continue;
4706 if (st->ss->load_super(st, fd, NULL))
4707 continue;
4708
4709 st->ss->getinfo_super(st, &dinfo, NULL);
4710 st->ss->free_super(st);
4711
4712 if (lseek64(fd,
4713 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4714 0) < 0) {
4715 pr_err("Cannot seek on device %d\n", i);
4716 continue; /* Cannot seek */
4717 }
4718 sprintf(namebuf, "device-%d", i);
4719 devname = namebuf;
4720 }
4721 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4722 if (verbose)
4723 pr_err("Cannot read from %s\n", devname);
4724 continue; /* Cannot read */
4725 }
4726 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4727 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4728 if (verbose)
4729 pr_err("No backup metadata on %s\n", devname);
4730 continue;
4731 }
4732 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4733 if (verbose)
4734 pr_err("Bad backup-metadata checksum on %s\n", devname);
4735 continue; /* bad checksum */
4736 }
4737 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4738 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4739 if (verbose)
4740 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4741 continue; /* Bad second checksum */
4742 }
4743 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4744 if (verbose)
4745 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4746 continue; /* Wrong uuid */
4747 }
4748
4749 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4750 * sometimes they aren't... So allow considerable flexability in matching, and allow
4751 * this test to be overridden by an environment variable.
4752 */
4753 if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) ||
4754 time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) {
4755 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4756 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4757 (unsigned long)__le64_to_cpu(bsb.mtime),
4758 (unsigned long)info->array.utime);
4759 } else {
4760 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4761 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4762 continue; /* time stamp is too bad */
4763 }
4764 }
4765
4766 if (bsb.magic[15] == '1') {
4767 if (bsb.length == 0)
4768 continue;
4769 if (info->delta_disks >= 0) {
4770 /* reshape_progress is increasing */
4771 if (__le64_to_cpu(bsb.arraystart)
4772 + __le64_to_cpu(bsb.length)
4773 < info->reshape_progress) {
4774 nonew:
4775 if (verbose)
4776 pr_err("backup-metadata found on %s but is not needed\n", devname);
4777 continue; /* No new data here */
4778 }
4779 } else {
4780 /* reshape_progress is decreasing */
4781 if (__le64_to_cpu(bsb.arraystart) >=
4782 info->reshape_progress)
4783 goto nonew; /* No new data here */
4784 }
4785 } else {
4786 if (bsb.length == 0 && bsb.length2 == 0)
4787 continue;
4788 if (info->delta_disks >= 0) {
4789 /* reshape_progress is increasing */
4790 if ((__le64_to_cpu(bsb.arraystart)
4791 + __le64_to_cpu(bsb.length)
4792 < info->reshape_progress)
4793 &&
4794 (__le64_to_cpu(bsb.arraystart2)
4795 + __le64_to_cpu(bsb.length2)
4796 < info->reshape_progress))
4797 goto nonew; /* No new data here */
4798 } else {
4799 /* reshape_progress is decreasing */
4800 if (__le64_to_cpu(bsb.arraystart) >=
4801 info->reshape_progress &&
4802 __le64_to_cpu(bsb.arraystart2) >=
4803 info->reshape_progress)
4804 goto nonew; /* No new data here */
4805 }
4806 }
4807 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4808 second_fail:
4809 if (verbose)
4810 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4811 devname);
4812 continue; /* Cannot seek */
4813 }
4814 /* There should be a duplicate backup superblock 4k before here */
4815 if (lseek64(fd, -4096, 1) < 0 ||
4816 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4817 goto second_fail; /* Cannot find leading superblock */
4818 if (bsb.magic[15] == '1')
4819 bsbsize = offsetof(struct mdp_backup_super, pad1);
4820 else
4821 bsbsize = offsetof(struct mdp_backup_super, pad);
4822 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4823 goto second_fail; /* Cannot find leading superblock */
4824
4825 /* Now need the data offsets for all devices. */
4826 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4827 for(j=0; j<info->array.raid_disks; j++) {
4828 if (fdlist[j] < 0)
4829 continue;
4830 if (st->ss->load_super(st, fdlist[j], NULL))
4831 /* FIXME should be this be an error */
4832 continue;
4833 st->ss->getinfo_super(st, &dinfo, NULL);
4834 st->ss->free_super(st);
4835 offsets[j] = dinfo.data_offset * 512;
4836 }
4837 printf("%s: restoring critical section\n", Name);
4838
4839 if (restore_stripes(fdlist, offsets,
4840 info->array.raid_disks,
4841 info->new_chunk,
4842 info->new_level,
4843 info->new_layout,
4844 fd, __le64_to_cpu(bsb.devstart)*512,
4845 __le64_to_cpu(bsb.arraystart)*512,
4846 __le64_to_cpu(bsb.length)*512, NULL)) {
4847 /* didn't succeed, so giveup */
4848 if (verbose)
4849 pr_err("Error restoring backup from %s\n",
4850 devname);
4851 free(offsets);
4852 return 1;
4853 }
4854
4855 if (bsb.magic[15] == '2' &&
4856 restore_stripes(fdlist, offsets,
4857 info->array.raid_disks,
4858 info->new_chunk,
4859 info->new_level,
4860 info->new_layout,
4861 fd, __le64_to_cpu(bsb.devstart)*512 +
4862 __le64_to_cpu(bsb.devstart2)*512,
4863 __le64_to_cpu(bsb.arraystart2)*512,
4864 __le64_to_cpu(bsb.length2)*512, NULL)) {
4865 /* didn't succeed, so giveup */
4866 if (verbose)
4867 pr_err("Error restoring second backup from %s\n",
4868 devname);
4869 free(offsets);
4870 return 1;
4871 }
4872
4873 free(offsets);
4874
4875 /* Ok, so the data is restored. Let's update those superblocks. */
4876
4877 lo = hi = 0;
4878 if (bsb.length) {
4879 lo = __le64_to_cpu(bsb.arraystart);
4880 hi = lo + __le64_to_cpu(bsb.length);
4881 }
4882 if (bsb.magic[15] == '2' && bsb.length2) {
4883 unsigned long long lo1, hi1;
4884 lo1 = __le64_to_cpu(bsb.arraystart2);
4885 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4886 if (lo == hi) {
4887 lo = lo1;
4888 hi = hi1;
4889 } else if (lo < lo1)
4890 hi = hi1;
4891 else
4892 lo = lo1;
4893 }
4894 if (lo < hi &&
4895 (info->reshape_progress < lo ||
4896 info->reshape_progress > hi))
4897 /* backup does not affect reshape_progress*/ ;
4898 else if (info->delta_disks >= 0) {
4899 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4900 __le64_to_cpu(bsb.length);
4901 if (bsb.magic[15] == '2') {
4902 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4903 __le64_to_cpu(bsb.length2);
4904 if (p2 > info->reshape_progress)
4905 info->reshape_progress = p2;
4906 }
4907 } else {
4908 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4909 if (bsb.magic[15] == '2') {
4910 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4911 if (p2 < info->reshape_progress)
4912 info->reshape_progress = p2;
4913 }
4914 }
4915 for (j=0; j<info->array.raid_disks; j++) {
4916 if (fdlist[j] < 0)
4917 continue;
4918 if (st->ss->load_super(st, fdlist[j], NULL))
4919 continue;
4920 st->ss->getinfo_super(st, &dinfo, NULL);
4921 dinfo.reshape_progress = info->reshape_progress;
4922 st->ss->update_super(st, &dinfo,
4923 "_reshape_progress",
4924 NULL,0, 0, NULL);
4925 st->ss->store_super(st, fdlist[j]);
4926 st->ss->free_super(st);
4927 }
4928 return 0;
4929 }
4930 /* Didn't find any backup data, try to see if any
4931 * was needed.
4932 */
4933 if (info->delta_disks < 0) {
4934 /* When shrinking, the critical section is at the end.
4935 * So see if we are before the critical section.
4936 */
4937 unsigned long long first_block;
4938 nstripe = ostripe = 0;
4939 first_block = 0;
4940 while (ostripe >= nstripe) {
4941 ostripe += info->array.chunk_size / 512;
4942 first_block = ostripe * odata;
4943 nstripe = first_block / ndata / (info->new_chunk/512) *
4944 (info->new_chunk/512);
4945 }
4946
4947 if (info->reshape_progress >= first_block)
4948 return 0;
4949 }
4950 if (info->delta_disks > 0) {
4951 /* See if we are beyond the critical section. */
4952 unsigned long long last_block;
4953 nstripe = ostripe = 0;
4954 last_block = 0;
4955 while (nstripe >= ostripe) {
4956 nstripe += info->new_chunk / 512;
4957 last_block = nstripe * ndata;
4958 ostripe = last_block / odata / (info->array.chunk_size/512) *
4959 (info->array.chunk_size/512);
4960 }
4961
4962 if (info->reshape_progress >= last_block)
4963 return 0;
4964 }
4965 /* needed to recover critical section! */
4966 if (verbose)
4967 pr_err("Failed to find backup of critical section\n");
4968 return 1;
4969 }
4970
4971 int Grow_continue_command(char *devname, int fd,
4972 char *backup_file, int verbose)
4973 {
4974 int ret_val = 0;
4975 struct supertype *st = NULL;
4976 struct mdinfo *content = NULL;
4977 struct mdinfo array;
4978 char *subarray = NULL;
4979 struct mdinfo *cc = NULL;
4980 struct mdstat_ent *mdstat = NULL;
4981 int cfd = -1;
4982 int fd2;
4983
4984 dprintf("Grow continue from command line called for %s\n",
4985 devname);
4986
4987 st = super_by_fd(fd, &subarray);
4988 if (!st || !st->ss) {
4989 pr_err("Unable to determine metadata format for %s\n",
4990 devname);
4991 return 1;
4992 }
4993 dprintf("Grow continue is run for ");
4994 if (st->ss->external == 0) {
4995 int d;
4996 int cnt = 5;
4997 dprintf_cont("native array (%s)\n", devname);
4998 if (md_get_array_info(fd, &array.array) < 0) {
4999 pr_err("%s is not an active md array - aborting\n",
5000 devname);
5001 ret_val = 1;
5002 goto Grow_continue_command_exit;
5003 }
5004 content = &array;
5005 sysfs_init(content, fd, NULL);
5006 /* Need to load a superblock.
5007 * FIXME we should really get what we need from
5008 * sysfs
5009 */
5010 do {
5011 for (d = 0; d < MAX_DISKS; d++) {
5012 mdu_disk_info_t disk;
5013 char *dv;
5014 int err;
5015 disk.number = d;
5016 if (md_get_disk_info(fd, &disk) < 0)
5017 continue;
5018 if (disk.major == 0 && disk.minor == 0)
5019 continue;
5020 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
5021 continue;
5022 dv = map_dev(disk.major, disk.minor, 1);
5023 if (!dv)
5024 continue;
5025 fd2 = dev_open(dv, O_RDONLY);
5026 if (fd2 < 0)
5027 continue;
5028 err = st->ss->load_super(st, fd2, NULL);
5029 close(fd2);
5030 if (err)
5031 continue;
5032 break;
5033 }
5034 if (d == MAX_DISKS) {
5035 pr_err("Unable to load metadata for %s\n",
5036 devname);
5037 ret_val = 1;
5038 goto Grow_continue_command_exit;
5039 }
5040 st->ss->getinfo_super(st, content, NULL);
5041 if (!content->reshape_active)
5042 sleep(3);
5043 else
5044 break;
5045 } while (cnt-- > 0);
5046 } else {
5047 char *container;
5048
5049 if (subarray) {
5050 dprintf_cont("subarray (%s)\n", subarray);
5051 container = st->container_devnm;
5052 cfd = open_dev_excl(st->container_devnm);
5053 } else {
5054 container = st->devnm;
5055 close(fd);
5056 cfd = open_dev_excl(st->devnm);
5057 dprintf_cont("container (%s)\n", container);
5058 fd = cfd;
5059 }
5060 if (cfd < 0) {
5061 pr_err("Unable to open container for %s\n", devname);
5062 ret_val = 1;
5063 goto Grow_continue_command_exit;
5064 }
5065
5066 /* find in container array under reshape
5067 */
5068 ret_val = st->ss->load_container(st, cfd, NULL);
5069 if (ret_val) {
5070 pr_err("Cannot read superblock for %s\n",
5071 devname);
5072 ret_val = 1;
5073 goto Grow_continue_command_exit;
5074 }
5075
5076 cc = st->ss->container_content(st, subarray);
5077 for (content = cc; content ; content = content->next) {
5078 char *array_name;
5079 int allow_reshape = 1;
5080
5081 if (content->reshape_active == 0)
5082 continue;
5083 /* The decision about array or container wide
5084 * reshape is taken in Grow_continue based
5085 * content->reshape_active state, therefore we
5086 * need to check_reshape based on
5087 * reshape_active and subarray name
5088 */
5089 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
5090 allow_reshape = 0;
5091 if (content->reshape_active == CONTAINER_RESHAPE &&
5092 (content->array.state
5093 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
5094 allow_reshape = 0;
5095
5096 if (!allow_reshape) {
5097 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5098 devname, container);
5099 ret_val = 1;
5100 goto Grow_continue_command_exit;
5101 }
5102
5103 array_name = strchr(content->text_version+1, '/')+1;
5104 mdstat = mdstat_by_subdev(array_name, container);
5105 if (!mdstat)
5106 continue;
5107 if (mdstat->active == 0) {
5108 pr_err("Skipping inactive array %s.\n",
5109 mdstat->devnm);
5110 free_mdstat(mdstat);
5111 mdstat = NULL;
5112 continue;
5113 }
5114 break;
5115 }
5116 if (!content) {
5117 pr_err("Unable to determine reshaped array for %s\n", devname);
5118 ret_val = 1;
5119 goto Grow_continue_command_exit;
5120 }
5121 fd2 = open_dev(mdstat->devnm);
5122 if (fd2 < 0) {
5123 pr_err("cannot open (%s)\n", mdstat->devnm);
5124 ret_val = 1;
5125 goto Grow_continue_command_exit;
5126 }
5127
5128 if (sysfs_init(content, fd2, mdstat->devnm)) {
5129 pr_err("Unable to initialize sysfs for %s, Grow cannot continue",
5130 mdstat->devnm);
5131 ret_val = 1;
5132 close(fd2);
5133 goto Grow_continue_command_exit;
5134 }
5135
5136 close(fd2);
5137
5138 /* start mdmon in case it is not running
5139 */
5140 if (!mdmon_running(container))
5141 start_mdmon(container);
5142 ping_monitor(container);
5143
5144 if (mdmon_running(container))
5145 st->update_tail = &st->updates;
5146 else {
5147 pr_err("No mdmon found. Grow cannot continue.\n");
5148 ret_val = 1;
5149 goto Grow_continue_command_exit;
5150 }
5151 }
5152
5153 /* verify that array under reshape is started from
5154 * correct position
5155 */
5156 if (verify_reshape_position(content, content->array.level) < 0) {
5157 ret_val = 1;
5158 goto Grow_continue_command_exit;
5159 }
5160
5161 /* continue reshape
5162 */
5163 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
5164
5165 Grow_continue_command_exit:
5166 if (cfd > -1)
5167 close(cfd);
5168 st->ss->free_super(st);
5169 free_mdstat(mdstat);
5170 sysfs_free(cc);
5171 free(subarray);
5172
5173 return ret_val;
5174 }
5175
5176 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
5177 char *backup_file, int forked, int freeze_reshape)
5178 {
5179 int ret_val = 2;
5180
5181 if (!info->reshape_active)
5182 return ret_val;
5183
5184 if (st->ss->external) {
5185 int cfd = open_dev(st->container_devnm);
5186
5187 if (cfd < 0)
5188 return 1;
5189
5190 st->ss->load_container(st, cfd, st->container_devnm);
5191 close(cfd);
5192 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
5193 st, info, 0, backup_file,
5194 0, forked,
5195 1 | info->reshape_active,
5196 freeze_reshape);
5197 } else
5198 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
5199 NULL, INVALID_SECTORS,
5200 backup_file, 0, forked,
5201 1 | info->reshape_active,
5202 freeze_reshape);
5203
5204 return ret_val;
5205 }
5206
5207 char *make_backup(char *name)
5208 {
5209 char *base = "backup_file-";
5210 int len;
5211 char *fname;
5212
5213 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
5214 fname = xmalloc(len);
5215 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
5216 return fname;
5217 }
5218
5219 char *locate_backup(char *name)
5220 {
5221 char *fl = make_backup(name);
5222 struct stat stb;
5223
5224 if (stat(fl, &stb) == 0 &&
5225 S_ISREG(stb.st_mode))
5226 return fl;
5227
5228 free(fl);
5229 return NULL;
5230 }
Unnamed repository; edit this file 'description' to name the repository.