]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Add one sanity check for missing device
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27 #include <stddef.h>
28 #include <stdint.h>
29 #include <signal.h>
30 #include <sys/wait.h>
31
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
34 #endif
35 #include "md_u.h"
36 #include "md_p.h"
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char **backup_filep,
43 int verbose)
44 {
45 int i;
46 int *fdlist;
47 struct mdinfo *dev;
48 int err;
49 int disk_count = next_spare + working_disks;
50 char *backup_file = *backup_filep;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61
62 sprintf(buf, "%d:%d", dev->disk.major, dev->disk.minor);
63 fd = dev_open(buf, O_RDWR);
64
65 if (dev->disk.raid_disk >= 0)
66 fdlist[dev->disk.raid_disk] = fd;
67 else
68 fdlist[next_spare++] = fd;
69 }
70
71 if (!backup_file) {
72 backup_file = locate_backup(content->sys_name);
73 *backup_filep = backup_file;
74 }
75
76 if (st->ss->external && st->ss->recover_backup)
77 err = st->ss->recover_backup(st, content);
78 else
79 err = Grow_restart(st, content, fdlist, next_spare,
80 backup_file, verbose > 0);
81
82 while (next_spare > 0) {
83 next_spare--;
84 if (fdlist[next_spare] >= 0)
85 close(fdlist[next_spare]);
86 }
87 free(fdlist);
88 if (err) {
89 pr_err("Failed to restore critical section for reshape - sorry.\n");
90 if (!backup_file)
91 pr_err("Possibly you need to specify a --backup-file\n");
92 return 1;
93 }
94
95 dprintf("restore_backup() returns status OK.\n");
96 return 0;
97 }
98
99 int Grow_Add_device(char *devname, int fd, char *newdev)
100 {
101 /* Add a device to an active array.
102 * Currently, just extend a linear array.
103 * This requires writing a new superblock on the
104 * new device, calling the kernel to add the device,
105 * and if that succeeds, update the superblock on
106 * all other devices.
107 * This means that we need to *find* all other devices.
108 */
109 struct mdinfo info;
110
111 dev_t rdev;
112 int nfd, fd2;
113 int d, nd;
114 struct supertype *st = NULL;
115 char *subarray = NULL;
116
117 if (md_get_array_info(fd, &info.array) < 0) {
118 pr_err("cannot get array info for %s\n", devname);
119 return 1;
120 }
121
122 if (info.array.level != -1) {
123 pr_err("can only add devices to linear arrays\n");
124 return 1;
125 }
126
127 st = super_by_fd(fd, &subarray);
128 if (!st) {
129 pr_err("cannot handle arrays with superblock version %d\n",
130 info.array.major_version);
131 return 1;
132 }
133
134 if (subarray) {
135 pr_err("Cannot grow linear sub-arrays yet\n");
136 free(subarray);
137 free(st);
138 return 1;
139 }
140
141 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
142 if (nfd < 0) {
143 pr_err("cannot open %s\n", newdev);
144 free(st);
145 return 1;
146 }
147 if (!fstat_is_blkdev(nfd, newdev, &rdev)) {
148 close(nfd);
149 free(st);
150 return 1;
151 }
152 /* now check out all the devices and make sure we can read the
153 * superblock */
154 for (d=0 ; d < info.array.raid_disks ; d++) {
155 mdu_disk_info_t disk;
156 char *dv;
157
158 st->ss->free_super(st);
159
160 disk.number = d;
161 if (md_get_disk_info(fd, &disk) < 0) {
162 pr_err("cannot get device detail for device %d\n", d);
163 close(nfd);
164 free(st);
165 return 1;
166 }
167 dv = map_dev(disk.major, disk.minor, 1);
168 if (!dv) {
169 pr_err("cannot find device file for device %d\n", d);
170 close(nfd);
171 free(st);
172 return 1;
173 }
174 fd2 = dev_open(dv, O_RDWR);
175 if (fd2 < 0) {
176 pr_err("cannot open device file %s\n", dv);
177 close(nfd);
178 free(st);
179 return 1;
180 }
181
182 if (st->ss->load_super(st, fd2, NULL)) {
183 pr_err("cannot find super block on %s\n", dv);
184 close(nfd);
185 close(fd2);
186 free(st);
187 return 1;
188 }
189 close(fd2);
190 }
191 /* Ok, looks good. Lets update the superblock and write it out to
192 * newdev.
193 */
194
195 info.disk.number = d;
196 info.disk.major = major(rdev);
197 info.disk.minor = minor(rdev);
198 info.disk.raid_disk = d;
199 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
200 st->ss->update_super(st, &info, "linear-grow-new", newdev, 0, 0, NULL);
201
202 if (st->ss->store_super(st, nfd)) {
203 pr_err("Cannot store new superblock on %s\n", newdev);
204 close(nfd);
205 return 1;
206 }
207 close(nfd);
208
209 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
210 pr_err("Cannot add new disk to this array\n");
211 return 1;
212 }
213 /* Well, that seems to have worked.
214 * Now go through and update all superblocks
215 */
216
217 if (md_get_array_info(fd, &info.array) < 0) {
218 pr_err("cannot get array info for %s\n", devname);
219 return 1;
220 }
221
222 nd = d;
223 for (d=0 ; d < info.array.raid_disks ; d++) {
224 mdu_disk_info_t disk;
225 char *dv;
226
227 disk.number = d;
228 if (md_get_disk_info(fd, &disk) < 0) {
229 pr_err("cannot get device detail for device %d\n", d);
230 return 1;
231 }
232 dv = map_dev(disk.major, disk.minor, 1);
233 if (!dv) {
234 pr_err("cannot find device file for device %d\n", d);
235 return 1;
236 }
237 fd2 = dev_open(dv, O_RDWR);
238 if (fd2 < 0) {
239 pr_err("cannot open device file %s\n", dv);
240 return 1;
241 }
242 if (st->ss->load_super(st, fd2, NULL)) {
243 pr_err("cannot find super block on %s\n", dv);
244 close(fd);
245 return 1;
246 }
247 info.array.raid_disks = nd+1;
248 info.array.nr_disks = nd+1;
249 info.array.active_disks = nd+1;
250 info.array.working_disks = nd+1;
251
252 st->ss->update_super(st, &info, "linear-grow-update", dv,
253 0, 0, NULL);
254
255 if (st->ss->store_super(st, fd2)) {
256 pr_err("Cannot store new superblock on %s\n", dv);
257 close(fd2);
258 return 1;
259 }
260 close(fd2);
261 }
262
263 return 0;
264 }
265
266 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
267 {
268 /*
269 * First check that array doesn't have a bitmap
270 * Then create the bitmap
271 * Then add it
272 *
273 * For internal bitmaps, we need to check the version,
274 * find all the active devices, and write the bitmap block
275 * to all devices
276 */
277 mdu_bitmap_file_t bmf;
278 mdu_array_info_t array;
279 struct supertype *st;
280 char *subarray = NULL;
281 int major = BITMAP_MAJOR_HI;
282 unsigned long long bitmapsize, array_size;
283 struct mdinfo *mdi;
284
285 /*
286 * We only ever get called if s->bitmap_file is != NULL, so this check
287 * is just here to quiet down static code checkers.
288 */
289 if (!s->bitmap_file)
290 return 1;
291
292 if (strcmp(s->bitmap_file, "clustered") == 0)
293 major = BITMAP_MAJOR_CLUSTERED;
294
295 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
296 if (errno == ENOMEM)
297 pr_err("Memory allocation failure.\n");
298 else
299 pr_err("bitmaps not supported by this kernel.\n");
300 return 1;
301 }
302 if (bmf.pathname[0]) {
303 if (strcmp(s->bitmap_file,"none") == 0) {
304 if (ioctl(fd, SET_BITMAP_FILE, -1) != 0) {
305 pr_err("failed to remove bitmap %s\n",
306 bmf.pathname);
307 return 1;
308 }
309 return 0;
310 }
311 pr_err("%s already has a bitmap (%s)\n", devname, bmf.pathname);
312 return 1;
313 }
314 if (md_get_array_info(fd, &array) != 0) {
315 pr_err("cannot get array status for %s\n", devname);
316 return 1;
317 }
318 if (array.state & (1 << MD_SB_BITMAP_PRESENT)) {
319 if (strcmp(s->bitmap_file, "none")==0) {
320 array.state &= ~(1 << MD_SB_BITMAP_PRESENT);
321 if (md_set_array_info(fd, &array) != 0) {
322 if (array.state & (1 << MD_SB_CLUSTERED))
323 pr_err("failed to remove clustered bitmap.\n");
324 else
325 pr_err("failed to remove internal bitmap.\n");
326 return 1;
327 }
328 return 0;
329 }
330 pr_err("bitmap already present on %s\n", devname);
331 return 1;
332 }
333
334 if (strcmp(s->bitmap_file, "none") == 0) {
335 pr_err("no bitmap found on %s\n", devname);
336 return 1;
337 }
338 if (array.level <= 0) {
339 pr_err("Bitmaps not meaningful with level %s\n",
340 map_num(pers, array.level)?:"of this array");
341 return 1;
342 }
343 bitmapsize = array.size;
344 bitmapsize <<= 1;
345 if (get_dev_size(fd, NULL, &array_size) &&
346 array_size > (0x7fffffffULL << 9)) {
347 /* Array is big enough that we cannot trust array.size
348 * try other approaches
349 */
350 bitmapsize = get_component_size(fd);
351 }
352 if (bitmapsize == 0) {
353 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
354 return 1;
355 }
356
357 if (array.level == 10) {
358 int ncopies;
359
360 ncopies = (array.layout & 255) * ((array.layout >> 8) & 255);
361 bitmapsize = bitmapsize * array.raid_disks / ncopies;
362
363 if (strcmp(s->bitmap_file, "clustered") == 0 &&
364 !is_near_layout_10(array.layout)) {
365 pr_err("only near layout is supported with clustered raid10\n");
366 return 1;
367 }
368 }
369
370 st = super_by_fd(fd, &subarray);
371 if (!st) {
372 pr_err("Cannot understand version %d.%d\n",
373 array.major_version, array.minor_version);
374 return 1;
375 }
376 if (subarray) {
377 pr_err("Cannot add bitmaps to sub-arrays yet\n");
378 free(subarray);
379 free(st);
380 return 1;
381 }
382
383 mdi = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY);
384 if (mdi) {
385 if (mdi->consistency_policy == CONSISTENCY_POLICY_PPL) {
386 pr_err("Cannot add bitmap to array with PPL\n");
387 free(mdi);
388 free(st);
389 return 1;
390 }
391 free(mdi);
392 }
393
394 if (strcmp(s->bitmap_file, "internal") == 0 ||
395 strcmp(s->bitmap_file, "clustered") == 0) {
396 int rv;
397 int d;
398 int offset_setable = 0;
399 if (st->ss->add_internal_bitmap == NULL) {
400 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
401 return 1;
402 }
403 st->nodes = c->nodes;
404 st->cluster_name = c->homecluster;
405 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
406 if (mdi)
407 offset_setable = 1;
408 for (d = 0; d < st->max_devs; d++) {
409 mdu_disk_info_t disk;
410 char *dv;
411 int fd2;
412
413 disk.number = d;
414 if (md_get_disk_info(fd, &disk) < 0)
415 continue;
416 if (disk.major == 0 && disk.minor == 0)
417 continue;
418 if ((disk.state & (1 << MD_DISK_SYNC)) == 0)
419 continue;
420 dv = map_dev(disk.major, disk.minor, 1);
421 if (!dv)
422 continue;
423 fd2 = dev_open(dv, O_RDWR);
424 if (fd2 < 0)
425 continue;
426 rv = st->ss->load_super(st, fd2, NULL);
427 if (!rv) {
428 rv = st->ss->add_internal_bitmap(
429 st, &s->bitmap_chunk, c->delay,
430 s->write_behind, bitmapsize,
431 offset_setable, major);
432 if (!rv) {
433 st->ss->write_bitmap(st, fd2,
434 NodeNumUpdate);
435 } else {
436 pr_err("failed to create internal bitmap - chunksize problem.\n");
437 }
438 } else {
439 pr_err("failed to load super-block.\n");
440 }
441 close(fd2);
442 if (rv)
443 return 1;
444 }
445 if (offset_setable) {
446 st->ss->getinfo_super(st, mdi, NULL);
447 if (sysfs_init(mdi, fd, NULL)) {
448 pr_err("failed to intialize sysfs.\n");
449 free(mdi);
450 }
451 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
452 mdi->bitmap_offset);
453 free(mdi);
454 } else {
455 if (strcmp(s->bitmap_file, "clustered") == 0)
456 array.state |= (1 << MD_SB_CLUSTERED);
457 array.state |= (1 << MD_SB_BITMAP_PRESENT);
458 rv = md_set_array_info(fd, &array);
459 }
460 if (rv < 0) {
461 if (errno == EBUSY)
462 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
463 pr_err("failed to set internal bitmap.\n");
464 return 1;
465 }
466 } else {
467 int uuid[4];
468 int bitmap_fd;
469 int d;
470 int max_devs = st->max_devs;
471
472 /* try to load a superblock */
473 for (d = 0; d < max_devs; d++) {
474 mdu_disk_info_t disk;
475 char *dv;
476 int fd2;
477 disk.number = d;
478 if (md_get_disk_info(fd, &disk) < 0)
479 continue;
480 if ((disk.major==0 && disk.minor == 0) ||
481 (disk.state & (1 << MD_DISK_REMOVED)))
482 continue;
483 dv = map_dev(disk.major, disk.minor, 1);
484 if (!dv)
485 continue;
486 fd2 = dev_open(dv, O_RDONLY);
487 if (fd2 >= 0) {
488 if (st->ss->load_super(st, fd2, NULL) == 0) {
489 close(fd2);
490 st->ss->uuid_from_super(st, uuid);
491 break;
492 }
493 close(fd2);
494 }
495 }
496 if (d == max_devs) {
497 pr_err("cannot find UUID for array!\n");
498 return 1;
499 }
500 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid,
501 s->bitmap_chunk, c->delay, s->write_behind,
502 bitmapsize, major)) {
503 return 1;
504 }
505 bitmap_fd = open(s->bitmap_file, O_RDWR);
506 if (bitmap_fd < 0) {
507 pr_err("weird: %s cannot be opened\n", s->bitmap_file);
508 return 1;
509 }
510 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
511 int err = errno;
512 if (errno == EBUSY)
513 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
514 pr_err("Cannot set bitmap file for %s: %s\n",
515 devname, strerror(err));
516 return 1;
517 }
518 }
519
520 return 0;
521 }
522
523 int Grow_consistency_policy(char *devname, int fd, struct context *c, struct shape *s)
524 {
525 struct supertype *st;
526 struct mdinfo *sra;
527 struct mdinfo *sd;
528 char *subarray = NULL;
529 int ret = 0;
530 char container_dev[PATH_MAX];
531 char buf[20];
532
533 if (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
534 s->consistency_policy != CONSISTENCY_POLICY_PPL) {
535 pr_err("Operation not supported for consistency policy %s\n",
536 map_num(consistency_policies, s->consistency_policy));
537 return 1;
538 }
539
540 st = super_by_fd(fd, &subarray);
541 if (!st)
542 return 1;
543
544 sra = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY|GET_LEVEL|
545 GET_DEVS|GET_STATE);
546 if (!sra) {
547 ret = 1;
548 goto free_st;
549 }
550
551 if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
552 !st->ss->write_init_ppl) {
553 pr_err("%s metadata does not support PPL\n", st->ss->name);
554 ret = 1;
555 goto free_info;
556 }
557
558 if (sra->array.level != 5) {
559 pr_err("Operation not supported for array level %d\n",
560 sra->array.level);
561 ret = 1;
562 goto free_info;
563 }
564
565 if (sra->consistency_policy == (unsigned)s->consistency_policy) {
566 pr_err("Consistency policy is already %s\n",
567 map_num(consistency_policies, s->consistency_policy));
568 ret = 1;
569 goto free_info;
570 } else if (sra->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
571 sra->consistency_policy != CONSISTENCY_POLICY_PPL) {
572 pr_err("Current consistency policy is %s, cannot change to %s\n",
573 map_num(consistency_policies, sra->consistency_policy),
574 map_num(consistency_policies, s->consistency_policy));
575 ret = 1;
576 goto free_info;
577 }
578
579 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
580 if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0) {
581 ret = 1;
582 goto free_info;
583 } else if (strcmp(buf, "reshape\n") == 0) {
584 pr_err("PPL cannot be enabled when reshape is in progress\n");
585 ret = 1;
586 goto free_info;
587 }
588 }
589
590 if (subarray) {
591 char *update;
592
593 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
594 update = "ppl";
595 else
596 update = "no-ppl";
597
598 sprintf(container_dev, "/dev/%s", st->container_devnm);
599
600 ret = Update_subarray(container_dev, subarray, update, NULL,
601 c->verbose);
602 if (ret)
603 goto free_info;
604 }
605
606 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
607 struct mdinfo info;
608
609 if (subarray) {
610 struct mdinfo *mdi;
611 int cfd;
612
613 cfd = open(container_dev, O_RDWR|O_EXCL);
614 if (cfd < 0) {
615 pr_err("Failed to open %s\n", container_dev);
616 ret = 1;
617 goto free_info;
618 }
619
620 ret = st->ss->load_container(st, cfd, st->container_devnm);
621 close(cfd);
622
623 if (ret) {
624 pr_err("Cannot read superblock for %s\n",
625 container_dev);
626 goto free_info;
627 }
628
629 mdi = st->ss->container_content(st, subarray);
630 info = *mdi;
631 free(mdi);
632 }
633
634 for (sd = sra->devs; sd; sd = sd->next) {
635 int dfd;
636 char *devpath;
637
638 devpath = map_dev(sd->disk.major, sd->disk.minor, 0);
639 dfd = dev_open(devpath, O_RDWR);
640 if (dfd < 0) {
641 pr_err("Failed to open %s\n", devpath);
642 ret = 1;
643 goto free_info;
644 }
645
646 if (!subarray) {
647 ret = st->ss->load_super(st, dfd, NULL);
648 if (ret) {
649 pr_err("Failed to load super-block.\n");
650 close(dfd);
651 goto free_info;
652 }
653
654 ret = st->ss->update_super(st, sra, "ppl",
655 devname,
656 c->verbose, 0, NULL);
657 if (ret) {
658 close(dfd);
659 st->ss->free_super(st);
660 goto free_info;
661 }
662 st->ss->getinfo_super(st, &info, NULL);
663 }
664
665 ret |= sysfs_set_num(sra, sd, "ppl_sector",
666 info.ppl_sector);
667 ret |= sysfs_set_num(sra, sd, "ppl_size",
668 info.ppl_size);
669
670 if (ret) {
671 pr_err("Failed to set PPL attributes for %s\n",
672 sd->sys_name);
673 close(dfd);
674 st->ss->free_super(st);
675 goto free_info;
676 }
677
678 ret = st->ss->write_init_ppl(st, &info, dfd);
679 if (ret)
680 pr_err("Failed to write PPL\n");
681
682 close(dfd);
683
684 if (!subarray)
685 st->ss->free_super(st);
686
687 if (ret)
688 goto free_info;
689 }
690 }
691
692 ret = sysfs_set_str(sra, NULL, "consistency_policy",
693 map_num(consistency_policies,
694 s->consistency_policy));
695 if (ret)
696 pr_err("Failed to change array consistency policy\n");
697
698 free_info:
699 sysfs_free(sra);
700 free_st:
701 free(st);
702 free(subarray);
703
704 return ret;
705 }
706
707 /*
708 * When reshaping an array we might need to backup some data.
709 * This is written to all spares with a 'super_block' describing it.
710 * The superblock goes 4K from the end of the used space on the
711 * device.
712 * It if written after the backup is complete.
713 * It has the following structure.
714 */
715
716 static struct mdp_backup_super {
717 char magic[16]; /* md_backup_data-1 or -2 */
718 __u8 set_uuid[16];
719 __u64 mtime;
720 /* start/sizes in 512byte sectors */
721 __u64 devstart; /* address on backup device/file of data */
722 __u64 arraystart;
723 __u64 length;
724 __u32 sb_csum; /* csum of preceeding bytes. */
725 __u32 pad1;
726 __u64 devstart2; /* offset in to data of second section */
727 __u64 arraystart2;
728 __u64 length2;
729 __u32 sb_csum2; /* csum of preceeding bytes. */
730 __u8 pad[512-68-32];
731 } __attribute__((aligned(512))) bsb, bsb2;
732
733 static __u32 bsb_csum(char *buf, int len)
734 {
735 int i;
736 int csum = 0;
737 for (i = 0; i < len; i++)
738 csum = (csum<<3) + buf[0];
739 return __cpu_to_le32(csum);
740 }
741
742 static int check_idle(struct supertype *st)
743 {
744 /* Check that all member arrays for this container, or the
745 * container of this array, are idle
746 */
747 char *container = (st->container_devnm[0]
748 ? st->container_devnm : st->devnm);
749 struct mdstat_ent *ent, *e;
750 int is_idle = 1;
751
752 ent = mdstat_read(0, 0);
753 for (e = ent ; e; e = e->next) {
754 if (!is_container_member(e, container))
755 continue;
756 if (e->percent >= 0) {
757 is_idle = 0;
758 break;
759 }
760 }
761 free_mdstat(ent);
762 return is_idle;
763 }
764
765 static int freeze_container(struct supertype *st)
766 {
767 char *container = (st->container_devnm[0]
768 ? st->container_devnm : st->devnm);
769
770 if (!check_idle(st))
771 return -1;
772
773 if (block_monitor(container, 1)) {
774 pr_err("failed to freeze container\n");
775 return -2;
776 }
777
778 return 1;
779 }
780
781 static void unfreeze_container(struct supertype *st)
782 {
783 char *container = (st->container_devnm[0]
784 ? st->container_devnm : st->devnm);
785
786 unblock_monitor(container, 1);
787 }
788
789 static int freeze(struct supertype *st)
790 {
791 /* Try to freeze resync/rebuild on this array/container.
792 * Return -1 if the array is busy,
793 * return -2 container cannot be frozen,
794 * return 0 if this kernel doesn't support 'frozen'
795 * return 1 if it worked.
796 */
797 if (st->ss->external)
798 return freeze_container(st);
799 else {
800 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
801 int err;
802 char buf[20];
803
804 if (!sra)
805 return -1;
806 /* Need to clear any 'read-auto' status */
807 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
808 strncmp(buf, "read-auto", 9) == 0)
809 sysfs_set_str(sra, NULL, "array_state", "clean");
810
811 err = sysfs_freeze_array(sra);
812 sysfs_free(sra);
813 return err;
814 }
815 }
816
817 static void unfreeze(struct supertype *st)
818 {
819 if (st->ss->external)
820 return unfreeze_container(st);
821 else {
822 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
823 char buf[20];
824
825 if (sra &&
826 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 &&
827 strcmp(buf, "frozen\n") == 0)
828 sysfs_set_str(sra, NULL, "sync_action", "idle");
829 sysfs_free(sra);
830 }
831 }
832
833 static void wait_reshape(struct mdinfo *sra)
834 {
835 int fd = sysfs_get_fd(sra, NULL, "sync_action");
836 char action[20];
837
838 if (fd < 0)
839 return;
840
841 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
842 strncmp(action, "reshape", 7) == 0)
843 sysfs_wait(fd, NULL);
844 close(fd);
845 }
846
847 static int reshape_super(struct supertype *st, unsigned long long size,
848 int level, int layout, int chunksize, int raid_disks,
849 int delta_disks, char *backup_file, char *dev,
850 int direction, int verbose)
851 {
852 /* nothing extra to check in the native case */
853 if (!st->ss->external)
854 return 0;
855 if (!st->ss->reshape_super || !st->ss->manage_reshape) {
856 pr_err("%s metadata does not support reshape\n",
857 st->ss->name);
858 return 1;
859 }
860
861 return st->ss->reshape_super(st, size, level, layout, chunksize,
862 raid_disks, delta_disks, backup_file, dev,
863 direction, verbose);
864 }
865
866 static void sync_metadata(struct supertype *st)
867 {
868 if (st->ss->external) {
869 if (st->update_tail) {
870 flush_metadata_updates(st);
871 st->update_tail = &st->updates;
872 } else
873 st->ss->sync_metadata(st);
874 }
875 }
876
877 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
878 {
879 /* when dealing with external metadata subarrays we need to be
880 * prepared to handle EAGAIN. The kernel may need to wait for
881 * mdmon to mark the array active so the kernel can handle
882 * allocations/writeback when preparing the reshape action
883 * (md_allow_write()). We temporarily disable safe_mode_delay
884 * to close a race with the array_state going clean before the
885 * next write to raid_disks / stripe_cache_size
886 */
887 char safe[50];
888 int rc;
889
890 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
891 if (!container ||
892 (strcmp(name, "raid_disks") != 0 &&
893 strcmp(name, "stripe_cache_size") != 0))
894 return sysfs_set_num(sra, NULL, name, n);
895
896 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
897 if (rc <= 0)
898 return -1;
899 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
900 rc = sysfs_set_num(sra, NULL, name, n);
901 if (rc < 0 && errno == EAGAIN) {
902 ping_monitor(container);
903 /* if we get EAGAIN here then the monitor is not active
904 * so stop trying
905 */
906 rc = sysfs_set_num(sra, NULL, name, n);
907 }
908 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
909 return rc;
910 }
911
912 int start_reshape(struct mdinfo *sra, int already_running,
913 int before_data_disks, int data_disks)
914 {
915 int err;
916 unsigned long long sync_max_to_set;
917
918 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
919 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
920 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
921 sra->reshape_progress);
922 if (before_data_disks <= data_disks)
923 sync_max_to_set = sra->reshape_progress / data_disks;
924 else
925 sync_max_to_set = (sra->component_size * data_disks
926 - sra->reshape_progress) / data_disks;
927 if (!already_running)
928 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
929 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
930 if (!already_running && err == 0) {
931 int cnt = 5;
932 do {
933 err = sysfs_set_str(sra, NULL, "sync_action",
934 "reshape");
935 if (err)
936 sleep(1);
937 } while (err && errno == EBUSY && cnt-- > 0);
938 }
939 return err;
940 }
941
942 void abort_reshape(struct mdinfo *sra)
943 {
944 sysfs_set_str(sra, NULL, "sync_action", "idle");
945 /*
946 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
947 * suspend_hi to decrease as well as increase.")
948 * you could only increase suspend_{lo,hi} unless the region they
949 * covered was empty. So to reset to 0, you need to push suspend_lo
950 * up past suspend_hi first. So to maximize the chance of mdadm
951 * working on all kernels, we want to keep doing that.
952 */
953 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
954 sysfs_set_num(sra, NULL, "suspend_hi", 0);
955 sysfs_set_num(sra, NULL, "suspend_lo", 0);
956 sysfs_set_num(sra, NULL, "sync_min", 0);
957 // It isn't safe to reset sync_max as we aren't monitoring.
958 // Array really should be stopped at this point.
959 }
960
961 int remove_disks_for_takeover(struct supertype *st,
962 struct mdinfo *sra,
963 int layout)
964 {
965 int nr_of_copies;
966 struct mdinfo *remaining;
967 int slot;
968
969 if (st->ss->external) {
970 int rv = 0;
971 struct mdinfo *arrays = st->ss->container_content(st, NULL);
972 /*
973 * containter_content returns list of arrays in container
974 * If arrays->next is not NULL it means that there are
975 * 2 arrays in container and operation should be blocked
976 */
977 if (arrays) {
978 if (arrays->next)
979 rv = 1;
980 sysfs_free(arrays);
981 if (rv) {
982 pr_err("Error. Cannot perform operation on /dev/%s\n", st->devnm);
983 pr_err("For this operation it MUST be single array in container\n");
984 return rv;
985 }
986 }
987 }
988
989 if (sra->array.level == 10)
990 nr_of_copies = layout & 0xff;
991 else if (sra->array.level == 1)
992 nr_of_copies = sra->array.raid_disks;
993 else
994 return 1;
995
996 remaining = sra->devs;
997 sra->devs = NULL;
998 /* for each 'copy', select one device and remove from the list. */
999 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
1000 struct mdinfo **diskp;
1001 int found = 0;
1002
1003 /* Find a working device to keep */
1004 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
1005 struct mdinfo *disk = *diskp;
1006
1007 if (disk->disk.raid_disk < slot)
1008 continue;
1009 if (disk->disk.raid_disk >= slot + nr_of_copies)
1010 continue;
1011 if (disk->disk.state & (1<<MD_DISK_REMOVED))
1012 continue;
1013 if (disk->disk.state & (1<<MD_DISK_FAULTY))
1014 continue;
1015 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
1016 continue;
1017
1018 /* We have found a good disk to use! */
1019 *diskp = disk->next;
1020 disk->next = sra->devs;
1021 sra->devs = disk;
1022 found = 1;
1023 break;
1024 }
1025 if (!found)
1026 break;
1027 }
1028
1029 if (slot < sra->array.raid_disks) {
1030 /* didn't find all slots */
1031 struct mdinfo **e;
1032 e = &remaining;
1033 while (*e)
1034 e = &(*e)->next;
1035 *e = sra->devs;
1036 sra->devs = remaining;
1037 return 1;
1038 }
1039
1040 /* Remove all 'remaining' devices from the array */
1041 while (remaining) {
1042 struct mdinfo *sd = remaining;
1043 remaining = sd->next;
1044
1045 sysfs_set_str(sra, sd, "state", "faulty");
1046 sysfs_set_str(sra, sd, "slot", "none");
1047 /* for external metadata disks should be removed in mdmon */
1048 if (!st->ss->external)
1049 sysfs_set_str(sra, sd, "state", "remove");
1050 sd->disk.state |= (1<<MD_DISK_REMOVED);
1051 sd->disk.state &= ~(1<<MD_DISK_SYNC);
1052 sd->next = sra->devs;
1053 sra->devs = sd;
1054 }
1055 return 0;
1056 }
1057
1058 void reshape_free_fdlist(int *fdlist,
1059 unsigned long long *offsets,
1060 int size)
1061 {
1062 int i;
1063
1064 for (i = 0; i < size; i++)
1065 if (fdlist[i] >= 0)
1066 close(fdlist[i]);
1067
1068 free(fdlist);
1069 free(offsets);
1070 }
1071
1072 int reshape_prepare_fdlist(char *devname,
1073 struct mdinfo *sra,
1074 int raid_disks,
1075 int nrdisks,
1076 unsigned long blocks,
1077 char *backup_file,
1078 int *fdlist,
1079 unsigned long long *offsets)
1080 {
1081 int d = 0;
1082 struct mdinfo *sd;
1083
1084 enable_fds(nrdisks);
1085 for (d = 0; d <= nrdisks; d++)
1086 fdlist[d] = -1;
1087 d = raid_disks;
1088 for (sd = sra->devs; sd; sd = sd->next) {
1089 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1090 continue;
1091 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
1092 sd->disk.raid_disk < raid_disks) {
1093 char *dn = map_dev(sd->disk.major, sd->disk.minor, 1);
1094 fdlist[sd->disk.raid_disk] = dev_open(dn, O_RDONLY);
1095 offsets[sd->disk.raid_disk] = sd->data_offset*512;
1096 if (fdlist[sd->disk.raid_disk] < 0) {
1097 pr_err("%s: cannot open component %s\n",
1098 devname, dn ? dn : "-unknown-");
1099 d = -1;
1100 goto release;
1101 }
1102 } else if (backup_file == NULL) {
1103 /* spare */
1104 char *dn = map_dev(sd->disk.major, sd->disk.minor, 1);
1105 fdlist[d] = dev_open(dn, O_RDWR);
1106 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
1107 if (fdlist[d] < 0) {
1108 pr_err("%s: cannot open component %s\n",
1109 devname, dn ? dn : "-unknown-");
1110 d = -1;
1111 goto release;
1112 }
1113 d++;
1114 }
1115 }
1116 release:
1117 return d;
1118 }
1119
1120 int reshape_open_backup_file(char *backup_file,
1121 int fd,
1122 char *devname,
1123 long blocks,
1124 int *fdlist,
1125 unsigned long long *offsets,
1126 char *sys_name,
1127 int restart)
1128 {
1129 /* Return 1 on success, 0 on any form of failure */
1130 /* need to check backup file is large enough */
1131 char buf[512];
1132 struct stat stb;
1133 unsigned int dev;
1134 int i;
1135
1136 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
1137 S_IRUSR | S_IWUSR);
1138 *offsets = 8 * 512;
1139 if (*fdlist < 0) {
1140 pr_err("%s: cannot create backup file %s: %s\n",
1141 devname, backup_file, strerror(errno));
1142 return 0;
1143 }
1144 /* Guard against backup file being on array device.
1145 * If array is partitioned or if LVM etc is in the
1146 * way this will not notice, but it is better than
1147 * nothing.
1148 */
1149 fstat(*fdlist, &stb);
1150 dev = stb.st_dev;
1151 fstat(fd, &stb);
1152 if (stb.st_rdev == dev) {
1153 pr_err("backup file must NOT be on the array being reshaped.\n");
1154 close(*fdlist);
1155 return 0;
1156 }
1157
1158 memset(buf, 0, 512);
1159 for (i=0; i < blocks + 8 ; i++) {
1160 if (write(*fdlist, buf, 512) != 512) {
1161 pr_err("%s: cannot create backup file %s: %s\n",
1162 devname, backup_file, strerror(errno));
1163 return 0;
1164 }
1165 }
1166 if (fsync(*fdlist) != 0) {
1167 pr_err("%s: cannot create backup file %s: %s\n",
1168 devname, backup_file, strerror(errno));
1169 return 0;
1170 }
1171
1172 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
1173 char *bu = make_backup(sys_name);
1174 if (symlink(backup_file, bu))
1175 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
1176 strerror(errno));
1177 free(bu);
1178 }
1179
1180 return 1;
1181 }
1182
1183 unsigned long compute_backup_blocks(int nchunk, int ochunk,
1184 unsigned int ndata, unsigned int odata)
1185 {
1186 unsigned long a, b, blocks;
1187 /* So how much do we need to backup.
1188 * We need an amount of data which is both a whole number of
1189 * old stripes and a whole number of new stripes.
1190 * So LCM for (chunksize*datadisks).
1191 */
1192 a = (ochunk/512) * odata;
1193 b = (nchunk/512) * ndata;
1194 /* Find GCD */
1195 a = GCD(a, b);
1196 /* LCM == product / GCD */
1197 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
1198
1199 return blocks;
1200 }
1201
1202 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
1203 {
1204 /* Based on the current array state in info->array and
1205 * the changes in info->new_* etc, determine:
1206 * - whether the change is possible
1207 * - Intermediate level/raid_disks/layout
1208 * - whether a restriping reshape is needed
1209 * - number of sectors in minimum change unit. This
1210 * will cover a whole number of stripes in 'before' and
1211 * 'after'.
1212 *
1213 * Return message if the change should be rejected
1214 * NULL if the change can be achieved
1215 *
1216 * This can be called as part of starting a reshape, or
1217 * when assembling an array that is undergoing reshape.
1218 */
1219 int near, far, offset, copies;
1220 int new_disks;
1221 int old_chunk, new_chunk;
1222 /* delta_parity records change in number of devices
1223 * caused by level change
1224 */
1225 int delta_parity = 0;
1226
1227 memset(re, 0, sizeof(*re));
1228
1229 /* If a new level not explicitly given, we assume no-change */
1230 if (info->new_level == UnSet)
1231 info->new_level = info->array.level;
1232
1233 if (info->new_chunk)
1234 switch (info->new_level) {
1235 case 0:
1236 case 4:
1237 case 5:
1238 case 6:
1239 case 10:
1240 /* chunk size is meaningful, must divide component_size
1241 * evenly
1242 */
1243 if (info->component_size % (info->new_chunk/512)) {
1244 unsigned long long shrink = info->component_size;
1245 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1246 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1247 info->new_chunk/1024, info->component_size/2);
1248 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1249 devname, shrink/2);
1250 pr_err("will shrink the array so the given chunk size would work.\n");
1251 return "";
1252 }
1253 break;
1254 default:
1255 return "chunk size not meaningful for this level";
1256 }
1257 else
1258 info->new_chunk = info->array.chunk_size;
1259
1260 switch (info->array.level) {
1261 default:
1262 return "No reshape is possibly for this RAID level";
1263 case LEVEL_LINEAR:
1264 if (info->delta_disks != UnSet)
1265 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1266 else
1267 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1268 case 1:
1269 /* RAID1 can convert to RAID1 with different disks, or
1270 * raid5 with 2 disks, or
1271 * raid0 with 1 disk
1272 */
1273 if (info->new_level > 1 && (info->component_size & 7))
1274 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1275 if (info->new_level == 0) {
1276 if (info->delta_disks != UnSet &&
1277 info->delta_disks != 0)
1278 return "Cannot change number of disks with RAID1->RAID0 conversion";
1279 re->level = 0;
1280 re->before.data_disks = 1;
1281 re->after.data_disks = 1;
1282 return NULL;
1283 }
1284 if (info->new_level == 1) {
1285 if (info->delta_disks == UnSet)
1286 /* Don't know what to do */
1287 return "no change requested for Growing RAID1";
1288 re->level = 1;
1289 return NULL;
1290 }
1291 if (info->array.raid_disks != 2 && info->new_level == 5)
1292 return "Can only convert a 2-device array to RAID5";
1293 if (info->array.raid_disks == 2 && info->new_level == 5) {
1294 re->level = 5;
1295 re->before.data_disks = 1;
1296 if (info->delta_disks != UnSet &&
1297 info->delta_disks != 0)
1298 re->after.data_disks = 1 + info->delta_disks;
1299 else
1300 re->after.data_disks = 1;
1301 if (re->after.data_disks < 1)
1302 return "Number of disks too small for RAID5";
1303
1304 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1305 info->array.chunk_size = 65536;
1306 break;
1307 }
1308 /* Could do some multi-stage conversions, but leave that to
1309 * later.
1310 */
1311 return "Impossibly level change request for RAID1";
1312
1313 case 10:
1314 /* RAID10 can be converted from near mode to
1315 * RAID0 by removing some devices.
1316 * It can also be reshaped if the kernel supports
1317 * new_data_offset.
1318 */
1319 switch (info->new_level) {
1320 case 0:
1321 if ((info->array.layout & ~0xff) != 0x100)
1322 return "Cannot Grow RAID10 with far/offset layout";
1323 /*
1324 * number of devices must be multiple of
1325 * number of copies
1326 */
1327 if (info->array.raid_disks %
1328 (info->array.layout & 0xff))
1329 return "RAID10 layout too complex for Grow operation";
1330
1331 new_disks = (info->array.raid_disks /
1332 (info->array.layout & 0xff));
1333 if (info->delta_disks == UnSet)
1334 info->delta_disks = (new_disks
1335 - info->array.raid_disks);
1336
1337 if (info->delta_disks !=
1338 new_disks - info->array.raid_disks)
1339 return "New number of raid-devices impossible for RAID10";
1340 if (info->new_chunk &&
1341 info->new_chunk != info->array.chunk_size)
1342 return "Cannot change chunk-size with RAID10 Grow";
1343
1344 /* looks good */
1345 re->level = 0;
1346 re->before.data_disks = new_disks;
1347 re->after.data_disks = re->before.data_disks;
1348 return NULL;
1349
1350 case 10:
1351 near = info->array.layout & 0xff;
1352 far = (info->array.layout >> 8) & 0xff;
1353 offset = info->array.layout & 0x10000;
1354 if (far > 1 && !offset)
1355 return "Cannot reshape RAID10 in far-mode";
1356 copies = near * far;
1357
1358 old_chunk = info->array.chunk_size * far;
1359
1360 if (info->new_layout == UnSet)
1361 info->new_layout = info->array.layout;
1362 else {
1363 near = info->new_layout & 0xff;
1364 far = (info->new_layout >> 8) & 0xff;
1365 offset = info->new_layout & 0x10000;
1366 if (far > 1 && !offset)
1367 return "Cannot reshape RAID10 to far-mode";
1368 if (near * far != copies)
1369 return "Cannot change number of copies when reshaping RAID10";
1370 }
1371 if (info->delta_disks == UnSet)
1372 info->delta_disks = 0;
1373 new_disks = (info->array.raid_disks +
1374 info->delta_disks);
1375
1376 new_chunk = info->new_chunk * far;
1377
1378 re->level = 10;
1379 re->before.layout = info->array.layout;
1380 re->before.data_disks = info->array.raid_disks;
1381 re->after.layout = info->new_layout;
1382 re->after.data_disks = new_disks;
1383 /* For RAID10 we don't do backup but do allow reshape,
1384 * so set backup_blocks to INVALID_SECTORS rather than
1385 * zero.
1386 * And there is no need to synchronise stripes on both
1387 * 'old' and 'new'. So the important
1388 * number is the minimum data_offset difference
1389 * which is the larger of (offset copies * chunk).
1390 */
1391 re->backup_blocks = INVALID_SECTORS;
1392 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1393 if (new_disks < re->before.data_disks &&
1394 info->space_after < re->min_offset_change)
1395 /* Reduce component size by one chunk */
1396 re->new_size = (info->component_size -
1397 re->min_offset_change);
1398 else
1399 re->new_size = info->component_size;
1400 re->new_size = re->new_size * new_disks / copies;
1401 return NULL;
1402
1403 default:
1404 return "RAID10 can only be changed to RAID0";
1405 }
1406 case 0:
1407 /* RAID0 can be converted to RAID10, or to RAID456 */
1408 if (info->new_level == 10) {
1409 if (info->new_layout == UnSet &&
1410 info->delta_disks == UnSet) {
1411 /* Assume near=2 layout */
1412 info->new_layout = 0x102;
1413 info->delta_disks = info->array.raid_disks;
1414 }
1415 if (info->new_layout == UnSet) {
1416 int copies = 1 + (info->delta_disks
1417 / info->array.raid_disks);
1418 if (info->array.raid_disks * (copies-1) !=
1419 info->delta_disks)
1420 return "Impossible number of devices for RAID0->RAID10";
1421 info->new_layout = 0x100 + copies;
1422 }
1423 if (info->delta_disks == UnSet) {
1424 int copies = info->new_layout & 0xff;
1425 if (info->new_layout != 0x100 + copies)
1426 return "New layout impossible for RAID0->RAID10";;
1427 info->delta_disks = (copies - 1) *
1428 info->array.raid_disks;
1429 }
1430 if (info->new_chunk &&
1431 info->new_chunk != info->array.chunk_size)
1432 return "Cannot change chunk-size with RAID0->RAID10";
1433 /* looks good */
1434 re->level = 10;
1435 re->before.data_disks = (info->array.raid_disks +
1436 info->delta_disks);
1437 re->after.data_disks = re->before.data_disks;
1438 re->before.layout = info->new_layout;
1439 return NULL;
1440 }
1441
1442 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1443 * a raid4 style layout of the final level.
1444 */
1445 switch (info->new_level) {
1446 case 4:
1447 delta_parity = 1;
1448 case 0:
1449 re->level = 4;
1450 re->before.layout = 0;
1451 break;
1452 case 5:
1453 delta_parity = 1;
1454 re->level = 5;
1455 re->before.layout = ALGORITHM_PARITY_N;
1456 if (info->new_layout == UnSet)
1457 info->new_layout = map_name(r5layout, "default");
1458 break;
1459 case 6:
1460 delta_parity = 2;
1461 re->level = 6;
1462 re->before.layout = ALGORITHM_PARITY_N;
1463 if (info->new_layout == UnSet)
1464 info->new_layout = map_name(r6layout, "default");
1465 break;
1466 default:
1467 return "Impossible level change requested";
1468 }
1469 re->before.data_disks = info->array.raid_disks;
1470 /* determining 'after' layout happens outside this 'switch' */
1471 break;
1472
1473 case 4:
1474 info->array.layout = ALGORITHM_PARITY_N;
1475 case 5:
1476 switch (info->new_level) {
1477 case 0:
1478 delta_parity = -1;
1479 case 4:
1480 re->level = info->array.level;
1481 re->before.data_disks = info->array.raid_disks - 1;
1482 re->before.layout = info->array.layout;
1483 break;
1484 case 5:
1485 re->level = 5;
1486 re->before.data_disks = info->array.raid_disks - 1;
1487 re->before.layout = info->array.layout;
1488 break;
1489 case 6:
1490 delta_parity = 1;
1491 re->level = 6;
1492 re->before.data_disks = info->array.raid_disks - 1;
1493 switch (info->array.layout) {
1494 case ALGORITHM_LEFT_ASYMMETRIC:
1495 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1496 break;
1497 case ALGORITHM_RIGHT_ASYMMETRIC:
1498 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1499 break;
1500 case ALGORITHM_LEFT_SYMMETRIC:
1501 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1502 break;
1503 case ALGORITHM_RIGHT_SYMMETRIC:
1504 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1505 break;
1506 case ALGORITHM_PARITY_0:
1507 re->before.layout = ALGORITHM_PARITY_0_6;
1508 break;
1509 case ALGORITHM_PARITY_N:
1510 re->before.layout = ALGORITHM_PARITY_N_6;
1511 break;
1512 default:
1513 return "Cannot convert an array with this layout";
1514 }
1515 break;
1516 case 1:
1517 if (info->array.raid_disks != 2)
1518 return "Can only convert a 2-device array to RAID1";
1519 if (info->delta_disks != UnSet &&
1520 info->delta_disks != 0)
1521 return "Cannot set raid_disk when converting RAID5->RAID1";
1522 re->level = 1;
1523 info->new_chunk = 0;
1524 return NULL;
1525 default:
1526 return "Impossible level change requested";
1527 }
1528 break;
1529 case 6:
1530 switch (info->new_level) {
1531 case 4:
1532 case 5:
1533 delta_parity = -1;
1534 case 6:
1535 re->level = 6;
1536 re->before.data_disks = info->array.raid_disks - 2;
1537 re->before.layout = info->array.layout;
1538 break;
1539 default:
1540 return "Impossible level change requested";
1541 }
1542 break;
1543 }
1544
1545 /* If we reached here then it looks like a re-stripe is
1546 * happening. We have determined the intermediate level
1547 * and initial raid_disks/layout and stored these in 're'.
1548 *
1549 * We need to deduce the final layout that can be atomically
1550 * converted to the end state.
1551 */
1552 switch (info->new_level) {
1553 case 0:
1554 /* We can only get to RAID0 from RAID4 or RAID5
1555 * with appropriate layout and one extra device
1556 */
1557 if (re->level != 4 && re->level != 5)
1558 return "Cannot covert to RAID0 from this level";
1559
1560 switch (re->level) {
1561 case 4:
1562 re->before.layout = 0;
1563 re->after.layout = 0;
1564 break;
1565 case 5:
1566 re->after.layout = ALGORITHM_PARITY_N;
1567 break;
1568 }
1569 break;
1570
1571 case 4:
1572 /* We can only get to RAID4 from RAID5 */
1573 if (re->level != 4 && re->level != 5)
1574 return "Cannot convert to RAID4 from this level";
1575
1576 switch (re->level) {
1577 case 4:
1578 re->after.layout = 0;
1579 break;
1580 case 5:
1581 re->after.layout = ALGORITHM_PARITY_N;
1582 break;
1583 }
1584 break;
1585
1586 case 5:
1587 /* We get to RAID5 from RAID5 or RAID6 */
1588 if (re->level != 5 && re->level != 6)
1589 return "Cannot convert to RAID5 from this level";
1590
1591 switch (re->level) {
1592 case 5:
1593 if (info->new_layout == UnSet)
1594 re->after.layout = re->before.layout;
1595 else
1596 re->after.layout = info->new_layout;
1597 break;
1598 case 6:
1599 if (info->new_layout == UnSet)
1600 info->new_layout = re->before.layout;
1601
1602 /* after.layout needs to be raid6 version of new_layout */
1603 if (info->new_layout == ALGORITHM_PARITY_N)
1604 re->after.layout = ALGORITHM_PARITY_N;
1605 else {
1606 char layout[40];
1607 char *ls = map_num(r5layout, info->new_layout);
1608 int l;
1609 if (ls) {
1610 /* Current RAID6 layout has a RAID5
1611 * equivalent - good
1612 */
1613 strcat(strcpy(layout, ls), "-6");
1614 l = map_name(r6layout, layout);
1615 if (l == UnSet)
1616 return "Cannot find RAID6 layout to convert to";
1617 } else {
1618 /* Current RAID6 has no equivalent.
1619 * If it is already a '-6' layout we
1620 * can leave it unchanged, else we must
1621 * fail
1622 */
1623 ls = map_num(r6layout,
1624 info->new_layout);
1625 if (!ls ||
1626 strcmp(ls+strlen(ls)-2, "-6") != 0)
1627 return "Please specify new layout";
1628 l = info->new_layout;
1629 }
1630 re->after.layout = l;
1631 }
1632 }
1633 break;
1634
1635 case 6:
1636 /* We must already be at level 6 */
1637 if (re->level != 6)
1638 return "Impossible level change";
1639 if (info->new_layout == UnSet)
1640 re->after.layout = info->array.layout;
1641 else
1642 re->after.layout = info->new_layout;
1643 break;
1644 default:
1645 return "Impossible level change requested";
1646 }
1647 if (info->delta_disks == UnSet)
1648 info->delta_disks = delta_parity;
1649
1650 re->after.data_disks =
1651 (re->before.data_disks + info->delta_disks - delta_parity);
1652
1653 switch (re->level) {
1654 case 6:
1655 re->parity = 2;
1656 break;
1657 case 4:
1658 case 5:
1659 re->parity = 1;
1660 break;
1661 default:
1662 re->parity = 0;
1663 break;
1664 }
1665 /* So we have a restripe operation, we need to calculate the number
1666 * of blocks per reshape operation.
1667 */
1668 re->new_size = info->component_size * re->before.data_disks;
1669 if (info->new_chunk == 0)
1670 info->new_chunk = info->array.chunk_size;
1671 if (re->after.data_disks == re->before.data_disks &&
1672 re->after.layout == re->before.layout &&
1673 info->new_chunk == info->array.chunk_size) {
1674 /* Nothing to change, can change level immediately. */
1675 re->level = info->new_level;
1676 re->backup_blocks = 0;
1677 return NULL;
1678 }
1679 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1680 /* chunk and layout changes make no difference */
1681 re->level = info->new_level;
1682 re->backup_blocks = 0;
1683 return NULL;
1684 }
1685
1686 if (re->after.data_disks == re->before.data_disks &&
1687 get_linux_version() < 2006032)
1688 return "in-place reshape is not safe before 2.6.32 - sorry.";
1689
1690 if (re->after.data_disks < re->before.data_disks &&
1691 get_linux_version() < 2006030)
1692 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1693
1694 re->backup_blocks = compute_backup_blocks(
1695 info->new_chunk, info->array.chunk_size,
1696 re->after.data_disks, re->before.data_disks);
1697 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1698
1699 re->new_size = info->component_size * re->after.data_disks;
1700 return NULL;
1701 }
1702
1703 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1704 char *text_version)
1705 {
1706 struct mdinfo *info;
1707 char *subarray;
1708 int ret_val = -1;
1709
1710 if ((st == NULL) || (sra == NULL))
1711 return ret_val;
1712
1713 if (text_version == NULL)
1714 text_version = sra->text_version;
1715 subarray = strchr(text_version + 1, '/')+1;
1716 info = st->ss->container_content(st, subarray);
1717 if (info) {
1718 unsigned long long current_size = 0;
1719 unsigned long long new_size = info->custom_array_size/2;
1720
1721 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1722 new_size > current_size) {
1723 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1724 < 0)
1725 dprintf("Error: Cannot set array size");
1726 else {
1727 ret_val = 0;
1728 dprintf("Array size changed");
1729 }
1730 dprintf_cont(" from %llu to %llu.\n",
1731 current_size, new_size);
1732 }
1733 sysfs_free(info);
1734 } else
1735 dprintf("Error: set_array_size(): info pointer in NULL\n");
1736
1737 return ret_val;
1738 }
1739
1740 static int reshape_array(char *container, int fd, char *devname,
1741 struct supertype *st, struct mdinfo *info,
1742 int force, struct mddev_dev *devlist,
1743 unsigned long long data_offset,
1744 char *backup_file, int verbose, int forked,
1745 int restart, int freeze_reshape);
1746 static int reshape_container(char *container, char *devname,
1747 int mdfd,
1748 struct supertype *st,
1749 struct mdinfo *info,
1750 int force,
1751 char *backup_file, int verbose,
1752 int forked, int restart, int freeze_reshape);
1753
1754 int Grow_reshape(char *devname, int fd,
1755 struct mddev_dev *devlist,
1756 unsigned long long data_offset,
1757 struct context *c, struct shape *s)
1758 {
1759 /* Make some changes in the shape of an array.
1760 * The kernel must support the change.
1761 *
1762 * There are three different changes. Each can trigger
1763 * a resync or recovery so we freeze that until we have
1764 * requested everything (if kernel supports freezing - 2.6.30).
1765 * The steps are:
1766 * - change size (i.e. component_size)
1767 * - change level
1768 * - change layout/chunksize/ndisks
1769 *
1770 * The last can require a reshape. It is different on different
1771 * levels so we need to check the level before actioning it.
1772 * Some times the level change needs to be requested after the
1773 * reshape (e.g. raid6->raid5, raid5->raid0)
1774 *
1775 */
1776 struct mdu_array_info_s array;
1777 int rv = 0;
1778 struct supertype *st;
1779 char *subarray = NULL;
1780
1781 int frozen;
1782 int changed = 0;
1783 char *container = NULL;
1784 int cfd = -1;
1785
1786 struct mddev_dev *dv;
1787 int added_disks;
1788
1789 struct mdinfo info;
1790 struct mdinfo *sra;
1791
1792 if (md_get_array_info(fd, &array) < 0) {
1793 pr_err("%s is not an active md array - aborting\n",
1794 devname);
1795 return 1;
1796 }
1797 if (s->level != UnSet && s->chunk) {
1798 pr_err("Cannot change array level in the same operation as changing chunk size.\n");
1799 return 1;
1800 }
1801
1802 if (data_offset != INVALID_SECTORS && array.level != 10 &&
1803 (array.level < 4 || array.level > 6)) {
1804 pr_err("--grow --data-offset not yet supported\n");
1805 return 1;
1806 }
1807
1808 if (s->size > 0 &&
1809 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1810 pr_err("cannot change component size at the same time as other changes.\n"
1811 " Change size first, then check data is intact before making other changes.\n");
1812 return 1;
1813 }
1814
1815 if (s->raiddisks && s->raiddisks < array.raid_disks &&
1816 array.level > 1 && get_linux_version() < 2006032 &&
1817 !check_env("MDADM_FORCE_FEWER")) {
1818 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1819 " Please use a newer kernel\n");
1820 return 1;
1821 }
1822
1823 if (array.level > 1 && s->size > 1 &&
1824 (array.chunk_size / 1024) > (int)s->size) {
1825 pr_err("component size must be larger than chunk size.\n");
1826 return 1;
1827 }
1828
1829 st = super_by_fd(fd, &subarray);
1830 if (!st) {
1831 pr_err("Unable to determine metadata format for %s\n", devname);
1832 return 1;
1833 }
1834 if (s->raiddisks > st->max_devs) {
1835 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1836 return 1;
1837 }
1838 if (s->level == 0 &&
1839 (array.state & (1<<MD_SB_BITMAP_PRESENT)) &&
1840 !(array.state & (1<<MD_SB_CLUSTERED))) {
1841 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
1842 if (md_set_array_info(fd, &array)!= 0) {
1843 pr_err("failed to remove internal bitmap.\n");
1844 return 1;
1845 }
1846 }
1847
1848 /* in the external case we need to check that the requested reshape is
1849 * supported, and perform an initial check that the container holds the
1850 * pre-requisite spare devices (mdmon owns final validation)
1851 */
1852 if (st->ss->external) {
1853 int retval;
1854
1855 if (subarray) {
1856 container = st->container_devnm;
1857 cfd = open_dev_excl(st->container_devnm);
1858 } else {
1859 container = st->devnm;
1860 close(fd);
1861 cfd = open_dev_excl(st->devnm);
1862 fd = cfd;
1863 }
1864 if (cfd < 0) {
1865 pr_err("Unable to open container for %s\n", devname);
1866 free(subarray);
1867 return 1;
1868 }
1869
1870 retval = st->ss->load_container(st, cfd, NULL);
1871
1872 if (retval) {
1873 pr_err("Cannot read superblock for %s\n", devname);
1874 free(subarray);
1875 return 1;
1876 }
1877
1878 /* check if operation is supported for metadata handler */
1879 if (st->ss->container_content) {
1880 struct mdinfo *cc = NULL;
1881 struct mdinfo *content = NULL;
1882
1883 cc = st->ss->container_content(st, subarray);
1884 for (content = cc; content ; content = content->next) {
1885 int allow_reshape = 1;
1886
1887 /* check if reshape is allowed based on metadata
1888 * indications stored in content.array.status
1889 */
1890 if (content->array.state &
1891 (1 << MD_SB_BLOCK_VOLUME))
1892 allow_reshape = 0;
1893 if (content->array.state &
1894 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE))
1895 allow_reshape = 0;
1896 if (!allow_reshape) {
1897 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1898 devname, container);
1899 sysfs_free(cc);
1900 free(subarray);
1901 return 1;
1902 }
1903 if (content->consistency_policy ==
1904 CONSISTENCY_POLICY_PPL) {
1905 pr_err("Operation not supported when ppl consistency policy is enabled\n");
1906 sysfs_free(cc);
1907 free(subarray);
1908 return 1;
1909 }
1910 }
1911 sysfs_free(cc);
1912 }
1913 if (mdmon_running(container))
1914 st->update_tail = &st->updates;
1915 }
1916
1917 added_disks = 0;
1918 for (dv = devlist; dv; dv = dv->next)
1919 added_disks++;
1920 if (s->raiddisks > array.raid_disks &&
1921 array.spare_disks + added_disks <
1922 (s->raiddisks - array.raid_disks) &&
1923 !c->force) {
1924 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1925 " Use --force to over-ride this check.\n",
1926 s->raiddisks - array.raid_disks,
1927 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1928 array.spare_disks + added_disks);
1929 return 1;
1930 }
1931
1932 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS |
1933 GET_STATE | GET_VERSION);
1934 if (sra) {
1935 if (st->ss->external && subarray == NULL) {
1936 array.level = LEVEL_CONTAINER;
1937 sra->array.level = LEVEL_CONTAINER;
1938 }
1939 } else {
1940 pr_err("failed to read sysfs parameters for %s\n",
1941 devname);
1942 return 1;
1943 }
1944 frozen = freeze(st);
1945 if (frozen < -1) {
1946 /* freeze() already spewed the reason */
1947 sysfs_free(sra);
1948 return 1;
1949 } else if (frozen < 0) {
1950 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1951 sysfs_free(sra);
1952 return 1;
1953 }
1954
1955 /* ========= set size =============== */
1956 if (s->size > 0 &&
1957 (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1958 unsigned long long orig_size = get_component_size(fd)/2;
1959 unsigned long long min_csize;
1960 struct mdinfo *mdi;
1961 int raid0_takeover = 0;
1962
1963 if (orig_size == 0)
1964 orig_size = (unsigned) array.size;
1965
1966 if (orig_size == 0) {
1967 pr_err("Cannot set device size in this type of array.\n");
1968 rv = 1;
1969 goto release;
1970 }
1971
1972 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1973 devname, APPLY_METADATA_CHANGES,
1974 c->verbose > 0)) {
1975 rv = 1;
1976 goto release;
1977 }
1978 sync_metadata(st);
1979 if (st->ss->external) {
1980 /* metadata can have size limitation
1981 * update size value according to metadata information
1982 */
1983 struct mdinfo *sizeinfo =
1984 st->ss->container_content(st, subarray);
1985 if (sizeinfo) {
1986 unsigned long long new_size =
1987 sizeinfo->custom_array_size/2;
1988 int data_disks = get_data_disks(
1989 sizeinfo->array.level,
1990 sizeinfo->array.layout,
1991 sizeinfo->array.raid_disks);
1992 new_size /= data_disks;
1993 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1994 orig_size, new_size,
1995 new_size * data_disks);
1996 s->size = new_size;
1997 sysfs_free(sizeinfo);
1998 }
1999 }
2000
2001 /* Update the size of each member device in case
2002 * they have been resized. This will never reduce
2003 * below the current used-size. The "size" attribute
2004 * understands '0' to mean 'max'.
2005 */
2006 min_csize = 0;
2007 for (mdi = sra->devs; mdi; mdi = mdi->next) {
2008 sysfs_set_num(sra, mdi, "size",
2009 s->size == MAX_SIZE ? 0 : s->size);
2010 if (array.not_persistent == 0 &&
2011 array.major_version == 0 &&
2012 get_linux_version() < 3001000) {
2013 /* Dangerous to allow size to exceed 2TB */
2014 unsigned long long csize;
2015 if (sysfs_get_ll(sra, mdi, "size",
2016 &csize) == 0) {
2017 if (csize >= 2ULL*1024*1024*1024)
2018 csize = 2ULL*1024*1024*1024;
2019 if ((min_csize == 0 ||
2020 (min_csize > csize)))
2021 min_csize = csize;
2022 }
2023 }
2024 }
2025 if (min_csize && s->size > min_csize) {
2026 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2027 rv = 1;
2028 goto size_change_error;
2029 }
2030 if (min_csize && s->size == MAX_SIZE) {
2031 /* Don't let the kernel choose a size - it will get
2032 * it wrong
2033 */
2034 pr_err("Limited v0.90 array to 2TB per device\n");
2035 s->size = min_csize;
2036 }
2037 if (st->ss->external) {
2038 if (sra->array.level == 0) {
2039 rv = sysfs_set_str(sra, NULL, "level", "raid5");
2040 if (!rv) {
2041 raid0_takeover = 1;
2042 /* get array parameters after takeover
2043 * to change one parameter at time only
2044 */
2045 rv = md_get_array_info(fd, &array);
2046 }
2047 }
2048 /* make sure mdmon is
2049 * aware of the new level */
2050 if (!mdmon_running(st->container_devnm))
2051 start_mdmon(st->container_devnm);
2052 ping_monitor(container);
2053 if (mdmon_running(st->container_devnm) &&
2054 st->update_tail == NULL)
2055 st->update_tail = &st->updates;
2056 }
2057
2058 if (s->size == MAX_SIZE)
2059 s->size = 0;
2060 array.size = s->size;
2061 if (s->size & ~INT32_MAX) {
2062 /* got truncated to 32bit, write to
2063 * component_size instead
2064 */
2065 if (sra)
2066 rv = sysfs_set_num(sra, NULL,
2067 "component_size", s->size);
2068 else
2069 rv = -1;
2070 } else {
2071 rv = md_set_array_info(fd, &array);
2072
2073 /* manage array size when it is managed externally
2074 */
2075 if ((rv == 0) && st->ss->external)
2076 rv = set_array_size(st, sra, sra->text_version);
2077 }
2078
2079 if (raid0_takeover) {
2080 /* do not recync non-existing parity,
2081 * we will drop it anyway
2082 */
2083 sysfs_set_str(sra, NULL, "sync_action", "frozen");
2084 /* go back to raid0, drop parity disk
2085 */
2086 sysfs_set_str(sra, NULL, "level", "raid0");
2087 md_get_array_info(fd, &array);
2088 }
2089
2090 size_change_error:
2091 if (rv != 0) {
2092 int err = errno;
2093
2094 /* restore metadata */
2095 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
2096 UnSet, NULL, devname,
2097 ROLLBACK_METADATA_CHANGES,
2098 c->verbose) == 0)
2099 sync_metadata(st);
2100 pr_err("Cannot set device size for %s: %s\n",
2101 devname, strerror(err));
2102 if (err == EBUSY &&
2103 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2104 cont_err("Bitmap must be removed before size can be changed\n");
2105 rv = 1;
2106 goto release;
2107 }
2108 if (s->assume_clean) {
2109 /* This will fail on kernels older than 3.0 unless
2110 * a backport has been arranged.
2111 */
2112 if (sra == NULL ||
2113 sysfs_set_str(sra, NULL, "resync_start",
2114 "none") < 0)
2115 pr_err("--assume-clean not supported with --grow on this kernel\n");
2116 }
2117 md_get_array_info(fd, &array);
2118 s->size = get_component_size(fd)/2;
2119 if (s->size == 0)
2120 s->size = array.size;
2121 if (c->verbose >= 0) {
2122 if (s->size == orig_size)
2123 pr_err("component size of %s unchanged at %lluK\n",
2124 devname, s->size);
2125 else
2126 pr_err("component size of %s has been set to %lluK\n",
2127 devname, s->size);
2128 }
2129 changed = 1;
2130 } else if (array.level != LEVEL_CONTAINER) {
2131 s->size = get_component_size(fd)/2;
2132 if (s->size == 0)
2133 s->size = array.size;
2134 }
2135
2136 /* See if there is anything else to do */
2137 if ((s->level == UnSet || s->level == array.level) &&
2138 (s->layout_str == NULL) &&
2139 (s->chunk == 0 || s->chunk == array.chunk_size) &&
2140 data_offset == INVALID_SECTORS &&
2141 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
2142 /* Nothing more to do */
2143 if (!changed && c->verbose >= 0)
2144 pr_err("%s: no change requested\n", devname);
2145 goto release;
2146 }
2147
2148 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2149 * current implementation assumes that following conditions must be met:
2150 * - RAID10:
2151 * - far_copies == 1
2152 * - near_copies == 2
2153 */
2154 if ((s->level == 0 && array.level == 10 && sra &&
2155 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
2156 (s->level == 0 && array.level == 1 && sra)) {
2157 int err;
2158
2159 err = remove_disks_for_takeover(st, sra, array.layout);
2160 if (err) {
2161 dprintf("Array cannot be reshaped\n");
2162 if (cfd > -1)
2163 close(cfd);
2164 rv = 1;
2165 goto release;
2166 }
2167 /* Make sure mdmon has seen the device removal
2168 * and updated metadata before we continue with
2169 * level change
2170 */
2171 if (container)
2172 ping_monitor(container);
2173 }
2174
2175 memset(&info, 0, sizeof(info));
2176 info.array = array;
2177 if (sysfs_init(&info, fd, NULL)) {
2178 pr_err("failed to intialize sysfs.\n");
2179 rv = 1;
2180 goto release;
2181 }
2182 strcpy(info.text_version, sra->text_version);
2183 info.component_size = s->size*2;
2184 info.new_level = s->level;
2185 info.new_chunk = s->chunk * 1024;
2186 if (info.array.level == LEVEL_CONTAINER) {
2187 info.delta_disks = UnSet;
2188 info.array.raid_disks = s->raiddisks;
2189 } else if (s->raiddisks)
2190 info.delta_disks = s->raiddisks - info.array.raid_disks;
2191 else
2192 info.delta_disks = UnSet;
2193 if (s->layout_str == NULL) {
2194 info.new_layout = UnSet;
2195 if (info.array.level == 6 &&
2196 (info.new_level == 6 || info.new_level == UnSet) &&
2197 info.array.layout >= 16) {
2198 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
2199 cont_err("during the reshape, please specify --layout=preserve\n");
2200 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2201 rv = 1;
2202 goto release;
2203 }
2204 } else if (strcmp(s->layout_str, "normalise") == 0 ||
2205 strcmp(s->layout_str, "normalize") == 0) {
2206 /* If we have a -6 RAID6 layout, remove the '-6'. */
2207 info.new_layout = UnSet;
2208 if (info.array.level == 6 && info.new_level == UnSet) {
2209 char l[40], *h;
2210 strcpy(l, map_num(r6layout, info.array.layout));
2211 h = strrchr(l, '-');
2212 if (h && strcmp(h, "-6") == 0) {
2213 *h = 0;
2214 info.new_layout = map_name(r6layout, l);
2215 }
2216 } else {
2217 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
2218 rv = 1;
2219 goto release;
2220 }
2221 } else if (strcmp(s->layout_str, "preserve") == 0) {
2222 /* This means that a non-standard RAID6 layout
2223 * is OK.
2224 * In particular:
2225 * - When reshape a RAID6 (e.g. adding a device)
2226 * which is in a non-standard layout, it is OK
2227 * to preserve that layout.
2228 * - When converting a RAID5 to RAID6, leave it in
2229 * the XXX-6 layout, don't re-layout.
2230 */
2231 if (info.array.level == 6 && info.new_level == UnSet)
2232 info.new_layout = info.array.layout;
2233 else if (info.array.level == 5 && info.new_level == 6) {
2234 char l[40];
2235 strcpy(l, map_num(r5layout, info.array.layout));
2236 strcat(l, "-6");
2237 info.new_layout = map_name(r6layout, l);
2238 } else {
2239 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
2240 rv = 1;
2241 goto release;
2242 }
2243 } else {
2244 int l = info.new_level;
2245 if (l == UnSet)
2246 l = info.array.level;
2247 switch (l) {
2248 case 5:
2249 info.new_layout = map_name(r5layout, s->layout_str);
2250 break;
2251 case 6:
2252 info.new_layout = map_name(r6layout, s->layout_str);
2253 break;
2254 case 10:
2255 info.new_layout = parse_layout_10(s->layout_str);
2256 break;
2257 case LEVEL_FAULTY:
2258 info.new_layout = parse_layout_faulty(s->layout_str);
2259 break;
2260 default:
2261 pr_err("layout not meaningful with this level\n");
2262 rv = 1;
2263 goto release;
2264 }
2265 if (info.new_layout == UnSet) {
2266 pr_err("layout %s not understood for this level\n",
2267 s->layout_str);
2268 rv = 1;
2269 goto release;
2270 }
2271 }
2272
2273 if (array.level == LEVEL_FAULTY) {
2274 if (s->level != UnSet && s->level != array.level) {
2275 pr_err("cannot change level of Faulty device\n");
2276 rv =1 ;
2277 }
2278 if (s->chunk) {
2279 pr_err("cannot set chunksize of Faulty device\n");
2280 rv =1 ;
2281 }
2282 if (s->raiddisks && s->raiddisks != 1) {
2283 pr_err("cannot set raid_disks of Faulty device\n");
2284 rv =1 ;
2285 }
2286 if (s->layout_str) {
2287 if (md_get_array_info(fd, &array) != 0) {
2288 dprintf("Cannot get array information.\n");
2289 goto release;
2290 }
2291 array.layout = info.new_layout;
2292 if (md_set_array_info(fd, &array) != 0) {
2293 pr_err("failed to set new layout\n");
2294 rv = 1;
2295 } else if (c->verbose >= 0)
2296 printf("layout for %s set to %d\n",
2297 devname, array.layout);
2298 }
2299 } else if (array.level == LEVEL_CONTAINER) {
2300 /* This change is to be applied to every array in the
2301 * container. This is only needed when the metadata imposes
2302 * restraints of the various arrays in the container.
2303 * Currently we only know that IMSM requires all arrays
2304 * to have the same number of devices so changing the
2305 * number of devices (On-Line Capacity Expansion) must be
2306 * performed at the level of the container
2307 */
2308 if (fd > 0) {
2309 close(fd);
2310 fd = -1;
2311 }
2312 rv = reshape_container(container, devname, -1, st, &info,
2313 c->force, c->backup_file, c->verbose,
2314 0, 0, 0);
2315 frozen = 0;
2316 } else {
2317 /* get spare devices from external metadata
2318 */
2319 if (st->ss->external) {
2320 struct mdinfo *info2;
2321
2322 info2 = st->ss->container_content(st, subarray);
2323 if (info2) {
2324 info.array.spare_disks =
2325 info2->array.spare_disks;
2326 sysfs_free(info2);
2327 }
2328 }
2329
2330 /* Impose these changes on a single array. First
2331 * check that the metadata is OK with the change. */
2332
2333 if (reshape_super(st, 0, info.new_level,
2334 info.new_layout, info.new_chunk,
2335 info.array.raid_disks, info.delta_disks,
2336 c->backup_file, devname,
2337 APPLY_METADATA_CHANGES, c->verbose)) {
2338 rv = 1;
2339 goto release;
2340 }
2341 sync_metadata(st);
2342 rv = reshape_array(container, fd, devname, st, &info, c->force,
2343 devlist, data_offset, c->backup_file,
2344 c->verbose, 0, 0, 0);
2345 frozen = 0;
2346 }
2347 release:
2348 sysfs_free(sra);
2349 if (frozen > 0)
2350 unfreeze(st);
2351 return rv;
2352 }
2353
2354 /* verify_reshape_position()
2355 * Function checks if reshape position in metadata is not farther
2356 * than position in md.
2357 * Return value:
2358 * 0 : not valid sysfs entry
2359 * it can be caused by not started reshape, it should be started
2360 * by reshape array or raid0 array is before takeover
2361 * -1 : error, reshape position is obviously wrong
2362 * 1 : success, reshape progress correct or updated
2363 */
2364 static int verify_reshape_position(struct mdinfo *info, int level)
2365 {
2366 int ret_val = 0;
2367 char buf[40];
2368 int rv;
2369
2370 /* read sync_max, failure can mean raid0 array */
2371 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2372
2373 if (rv > 0) {
2374 char *ep;
2375 unsigned long long position = strtoull(buf, &ep, 0);
2376
2377 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2378 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2379 position *= get_data_disks(level,
2380 info->new_layout,
2381 info->array.raid_disks);
2382 if (info->reshape_progress < position) {
2383 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2384 info->reshape_progress, position);
2385 info->reshape_progress = position;
2386 ret_val = 1;
2387 } else if (info->reshape_progress > position) {
2388 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2389 position, info->reshape_progress);
2390 ret_val = -1;
2391 } else {
2392 dprintf("Reshape position in md and metadata are the same;");
2393 ret_val = 1;
2394 }
2395 }
2396 } else if (rv == 0) {
2397 /* for valid sysfs entry, 0-length content
2398 * should be indicated as error
2399 */
2400 ret_val = -1;
2401 }
2402
2403 return ret_val;
2404 }
2405
2406 static unsigned long long choose_offset(unsigned long long lo,
2407 unsigned long long hi,
2408 unsigned long long min,
2409 unsigned long long max)
2410 {
2411 /* Choose a new offset between hi and lo.
2412 * It must be between min and max, but
2413 * we would prefer something near the middle of hi/lo, and also
2414 * prefer to be aligned to a big power of 2.
2415 *
2416 * So we start with the middle, then for each bit,
2417 * starting at '1' and increasing, if it is set, we either
2418 * add it or subtract it if possible, preferring the option
2419 * which is furthest from the boundary.
2420 *
2421 * We stop once we get a 1MB alignment. As units are in sectors,
2422 * 1MB = 2*1024 sectors.
2423 */
2424 unsigned long long choice = (lo + hi) / 2;
2425 unsigned long long bit = 1;
2426
2427 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2428 unsigned long long bigger, smaller;
2429 if (! (bit & choice))
2430 continue;
2431 bigger = choice + bit;
2432 smaller = choice - bit;
2433 if (bigger > max && smaller < min)
2434 break;
2435 if (bigger > max)
2436 choice = smaller;
2437 else if (smaller < min)
2438 choice = bigger;
2439 else if (hi - bigger > smaller - lo)
2440 choice = bigger;
2441 else
2442 choice = smaller;
2443 }
2444 return choice;
2445 }
2446
2447 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2448 char *devname, int delta_disks,
2449 unsigned long long data_offset,
2450 unsigned long long min,
2451 int can_fallback)
2452 {
2453 struct mdinfo *sd;
2454 int dir = 0;
2455 int err = 0;
2456 unsigned long long before, after;
2457
2458 /* Need to find min space before and after so same is used
2459 * on all devices
2460 */
2461 before = UINT64_MAX;
2462 after = UINT64_MAX;
2463 for (sd = sra->devs; sd; sd = sd->next) {
2464 char *dn;
2465 int dfd;
2466 int rv;
2467 struct supertype *st2;
2468 struct mdinfo info2;
2469
2470 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2471 continue;
2472 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2473 dfd = dev_open(dn, O_RDONLY);
2474 if (dfd < 0) {
2475 pr_err("%s: cannot open component %s\n",
2476 devname, dn ? dn : "-unknown-");
2477 goto release;
2478 }
2479 st2 = dup_super(st);
2480 rv = st2->ss->load_super(st2,dfd, NULL);
2481 close(dfd);
2482 if (rv) {
2483 free(st2);
2484 pr_err("%s: cannot get superblock from %s\n",
2485 devname, dn);
2486 goto release;
2487 }
2488 st2->ss->getinfo_super(st2, &info2, NULL);
2489 st2->ss->free_super(st2);
2490 free(st2);
2491 if (info2.space_before == 0 &&
2492 info2.space_after == 0) {
2493 /* Metadata doesn't support data_offset changes */
2494 if (!can_fallback)
2495 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2496 devname);
2497 goto fallback;
2498 }
2499 if (before > info2.space_before)
2500 before = info2.space_before;
2501 if (after > info2.space_after)
2502 after = info2.space_after;
2503
2504 if (data_offset != INVALID_SECTORS) {
2505 if (dir == 0) {
2506 if (info2.data_offset == data_offset) {
2507 pr_err("%s: already has that data_offset\n",
2508 dn);
2509 goto release;
2510 }
2511 if (data_offset < info2.data_offset)
2512 dir = -1;
2513 else
2514 dir = 1;
2515 } else if ((data_offset <= info2.data_offset &&
2516 dir == 1) ||
2517 (data_offset >= info2.data_offset &&
2518 dir == -1)) {
2519 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2520 dn);
2521 goto release;
2522 }
2523 }
2524 }
2525 if (before == UINT64_MAX)
2526 /* impossible really, there must be no devices */
2527 return 1;
2528
2529 for (sd = sra->devs; sd; sd = sd->next) {
2530 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2531 unsigned long long new_data_offset;
2532
2533 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2534 continue;
2535 if (delta_disks < 0) {
2536 /* Don't need any space as array is shrinking
2537 * just move data_offset up by min
2538 */
2539 if (data_offset == INVALID_SECTORS)
2540 new_data_offset = sd->data_offset + min;
2541 else {
2542 if (data_offset < sd->data_offset + min) {
2543 pr_err("--data-offset too small for %s\n",
2544 dn);
2545 goto release;
2546 }
2547 new_data_offset = data_offset;
2548 }
2549 } else if (delta_disks > 0) {
2550 /* need space before */
2551 if (before < min) {
2552 if (can_fallback)
2553 goto fallback;
2554 pr_err("Insufficient head-space for reshape on %s\n",
2555 dn);
2556 goto release;
2557 }
2558 if (data_offset == INVALID_SECTORS)
2559 new_data_offset = sd->data_offset - min;
2560 else {
2561 if (data_offset > sd->data_offset - min) {
2562 pr_err("--data-offset too large for %s\n",
2563 dn);
2564 goto release;
2565 }
2566 new_data_offset = data_offset;
2567 }
2568 } else {
2569 if (dir == 0) {
2570 /* can move up or down. If 'data_offset'
2571 * was set we would have already decided,
2572 * so just choose direction with most space.
2573 */
2574 if (before > after)
2575 dir = -1;
2576 else
2577 dir = 1;
2578 }
2579 sysfs_set_str(sra, NULL, "reshape_direction",
2580 dir == 1 ? "backwards" : "forwards");
2581 if (dir > 0) {
2582 /* Increase data offset */
2583 if (after < min) {
2584 if (can_fallback)
2585 goto fallback;
2586 pr_err("Insufficient tail-space for reshape on %s\n",
2587 dn);
2588 goto release;
2589 }
2590 if (data_offset != INVALID_SECTORS &&
2591 data_offset < sd->data_offset + min) {
2592 pr_err("--data-offset too small on %s\n",
2593 dn);
2594 goto release;
2595 }
2596 if (data_offset != INVALID_SECTORS)
2597 new_data_offset = data_offset;
2598 else
2599 new_data_offset = choose_offset(sd->data_offset,
2600 sd->data_offset + after,
2601 sd->data_offset + min,
2602 sd->data_offset + after);
2603 } else {
2604 /* Decrease data offset */
2605 if (before < min) {
2606 if (can_fallback)
2607 goto fallback;
2608 pr_err("insufficient head-room on %s\n",
2609 dn);
2610 goto release;
2611 }
2612 if (data_offset != INVALID_SECTORS &&
2613 data_offset < sd->data_offset - min) {
2614 pr_err("--data-offset too small on %s\n",
2615 dn);
2616 goto release;
2617 }
2618 if (data_offset != INVALID_SECTORS)
2619 new_data_offset = data_offset;
2620 else
2621 new_data_offset = choose_offset(sd->data_offset - before,
2622 sd->data_offset,
2623 sd->data_offset - before,
2624 sd->data_offset - min);
2625 }
2626 }
2627 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2628 if (err < 0 && errno == E2BIG) {
2629 /* try again after increasing data size to max */
2630 err = sysfs_set_num(sra, sd, "size", 0);
2631 if (err < 0 && errno == EINVAL &&
2632 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2633 /* some kernels have a bug where you cannot
2634 * use '0' on spare devices. */
2635 sysfs_set_num(sra, sd, "size",
2636 (sra->component_size + after)/2);
2637 }
2638 err = sysfs_set_num(sra, sd, "new_offset",
2639 new_data_offset);
2640 }
2641 if (err < 0) {
2642 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2643 pr_err("data-offset is too big for %s\n", dn);
2644 goto release;
2645 }
2646 if (sd == sra->devs &&
2647 (errno == ENOENT || errno == E2BIG))
2648 /* Early kernel, no 'new_offset' file,
2649 * or kernel doesn't like us.
2650 * For RAID5/6 this is not fatal
2651 */
2652 return 1;
2653 pr_err("Cannot set new_offset for %s\n", dn);
2654 break;
2655 }
2656 }
2657 return err;
2658 release:
2659 return -1;
2660 fallback:
2661 /* Just use a backup file */
2662 return 1;
2663 }
2664
2665 static int raid10_reshape(char *container, int fd, char *devname,
2666 struct supertype *st, struct mdinfo *info,
2667 struct reshape *reshape,
2668 unsigned long long data_offset,
2669 int force, int verbose)
2670 {
2671 /* Changing raid_disks, layout, chunksize or possibly
2672 * just data_offset for a RAID10.
2673 * We must always change data_offset. We change by at least
2674 * ->min_offset_change which is the largest of the old and new
2675 * chunk sizes.
2676 * If raid_disks is increasing, then data_offset must decrease
2677 * by at least this copy size.
2678 * If raid_disks is unchanged, data_offset must increase or
2679 * decrease by at least min_offset_change but preferably by much more.
2680 * We choose half of the available space.
2681 * If raid_disks is decreasing, data_offset must increase by
2682 * at least min_offset_change. To allow of this, component_size
2683 * must be decreased by the same amount.
2684 *
2685 * So we calculate the required minimum and direction, possibly
2686 * reduce the component_size, then iterate through the devices
2687 * and set the new_data_offset.
2688 * If that all works, we set chunk_size, layout, raid_disks, and start
2689 * 'reshape'
2690 */
2691 struct mdinfo *sra;
2692 unsigned long long min;
2693 int err = 0;
2694
2695 sra = sysfs_read(fd, NULL,
2696 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2697 );
2698 if (!sra) {
2699 pr_err("%s: Cannot get array details from sysfs\n", devname);
2700 goto release;
2701 }
2702 min = reshape->min_offset_change;
2703
2704 if (info->delta_disks)
2705 sysfs_set_str(sra, NULL, "reshape_direction",
2706 info->delta_disks < 0 ? "backwards" : "forwards");
2707 if (info->delta_disks < 0 && info->space_after < min) {
2708 int rv = sysfs_set_num(sra, NULL, "component_size",
2709 (sra->component_size - min)/2);
2710 if (rv) {
2711 pr_err("cannot reduce component size\n");
2712 goto release;
2713 }
2714 }
2715 err = set_new_data_offset(sra, st, devname, info->delta_disks,
2716 data_offset, min, 0);
2717 if (err == 1) {
2718 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2719 cont_err("supported on this kernel\n");
2720 err = -1;
2721 }
2722 if (err < 0)
2723 goto release;
2724
2725 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2726 err = errno;
2727 if (!err && sysfs_set_num(sra, NULL, "layout",
2728 reshape->after.layout) < 0)
2729 err = errno;
2730 if (!err &&
2731 sysfs_set_num(sra, NULL, "raid_disks",
2732 info->array.raid_disks + info->delta_disks) < 0)
2733 err = errno;
2734 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2735 err = errno;
2736 if (err) {
2737 pr_err("Cannot set array shape for %s\n",
2738 devname);
2739 if (err == EBUSY &&
2740 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2741 cont_err(" Bitmap must be removed before shape can be changed\n");
2742 goto release;
2743 }
2744 sysfs_free(sra);
2745 return 0;
2746 release:
2747 sysfs_free(sra);
2748 return 1;
2749 }
2750
2751 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2752 {
2753 struct mdinfo *sra, *sd;
2754 /* Initialisation to silence compiler warning */
2755 unsigned long long min_space_before = 0, min_space_after = 0;
2756 int first = 1;
2757
2758 sra = sysfs_read(fd, NULL, GET_DEVS);
2759 if (!sra)
2760 return;
2761 for (sd = sra->devs; sd; sd = sd->next) {
2762 char *dn;
2763 int dfd;
2764 struct supertype *st2;
2765 struct mdinfo info2;
2766
2767 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2768 continue;
2769 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2770 dfd = dev_open(dn, O_RDONLY);
2771 if (dfd < 0)
2772 break;
2773 st2 = dup_super(st);
2774 if (st2->ss->load_super(st2,dfd, NULL)) {
2775 close(dfd);
2776 free(st2);
2777 break;
2778 }
2779 close(dfd);
2780 st2->ss->getinfo_super(st2, &info2, NULL);
2781 st2->ss->free_super(st2);
2782 free(st2);
2783 if (first ||
2784 min_space_before > info2.space_before)
2785 min_space_before = info2.space_before;
2786 if (first ||
2787 min_space_after > info2.space_after)
2788 min_space_after = info2.space_after;
2789 first = 0;
2790 }
2791 if (sd == NULL && !first) {
2792 info->space_after = min_space_after;
2793 info->space_before = min_space_before;
2794 }
2795 sysfs_free(sra);
2796 }
2797
2798 static void update_cache_size(char *container, struct mdinfo *sra,
2799 struct mdinfo *info,
2800 int disks, unsigned long long blocks)
2801 {
2802 /* Check that the internal stripe cache is
2803 * large enough, or it won't work.
2804 * It must hold at least 4 stripes of the larger
2805 * chunk size
2806 */
2807 unsigned long cache;
2808 cache = max(info->array.chunk_size, info->new_chunk);
2809 cache *= 4; /* 4 stripes minimum */
2810 cache /= 512; /* convert to sectors */
2811 /* make sure there is room for 'blocks' with a bit to spare */
2812 if (cache < 16 + blocks / disks)
2813 cache = 16 + blocks / disks;
2814 cache /= (4096/512); /* Convert from sectors to pages */
2815
2816 if (sra->cache_size < cache)
2817 subarray_set_num(container, sra, "stripe_cache_size",
2818 cache+1);
2819 }
2820
2821 static int impose_reshape(struct mdinfo *sra,
2822 struct mdinfo *info,
2823 struct supertype *st,
2824 int fd,
2825 int restart,
2826 char *devname, char *container,
2827 struct reshape *reshape)
2828 {
2829 struct mdu_array_info_s array;
2830
2831 sra->new_chunk = info->new_chunk;
2832
2833 if (restart) {
2834 /* for external metadata checkpoint saved by mdmon can be lost
2835 * or missed /due to e.g. crash/. Check if md is not during
2836 * restart farther than metadata points to.
2837 * If so, this means metadata information is obsolete.
2838 */
2839 if (st->ss->external)
2840 verify_reshape_position(info, reshape->level);
2841 sra->reshape_progress = info->reshape_progress;
2842 } else {
2843 sra->reshape_progress = 0;
2844 if (reshape->after.data_disks < reshape->before.data_disks)
2845 /* start from the end of the new array */
2846 sra->reshape_progress = (sra->component_size
2847 * reshape->after.data_disks);
2848 }
2849
2850 md_get_array_info(fd, &array);
2851 if (info->array.chunk_size == info->new_chunk &&
2852 reshape->before.layout == reshape->after.layout &&
2853 st->ss->external == 0) {
2854 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2855 array.raid_disks = reshape->after.data_disks + reshape->parity;
2856 if (!restart && md_set_array_info(fd, &array) != 0) {
2857 int err = errno;
2858
2859 pr_err("Cannot set device shape for %s: %s\n",
2860 devname, strerror(errno));
2861
2862 if (err == EBUSY &&
2863 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2864 cont_err("Bitmap must be removed before shape can be changed\n");
2865
2866 goto release;
2867 }
2868 } else if (!restart) {
2869 /* set them all just in case some old 'new_*' value
2870 * persists from some earlier problem.
2871 */
2872 int err = 0;
2873 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2874 err = errno;
2875 if (!err && sysfs_set_num(sra, NULL, "layout",
2876 reshape->after.layout) < 0)
2877 err = errno;
2878 if (!err && subarray_set_num(container, sra, "raid_disks",
2879 reshape->after.data_disks +
2880 reshape->parity) < 0)
2881 err = errno;
2882 if (err) {
2883 pr_err("Cannot set device shape for %s\n", devname);
2884
2885 if (err == EBUSY &&
2886 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2887 cont_err("Bitmap must be removed before shape can be changed\n");
2888 goto release;
2889 }
2890 }
2891 return 0;
2892 release:
2893 return -1;
2894 }
2895
2896 static int impose_level(int fd, int level, char *devname, int verbose)
2897 {
2898 char *c;
2899 struct mdu_array_info_s array;
2900 struct mdinfo info;
2901
2902 if (sysfs_init(&info, fd, NULL)) {
2903 pr_err("failed to intialize sysfs.\n");
2904 return 1;
2905 }
2906
2907 md_get_array_info(fd, &array);
2908 if (level == 0 && (array.level >= 4 && array.level <= 6)) {
2909 /* To convert to RAID0 we need to fail and
2910 * remove any non-data devices. */
2911 int found = 0;
2912 int d;
2913 int data_disks = array.raid_disks - 1;
2914 if (array.level == 6)
2915 data_disks -= 1;
2916 if (array.level == 5 && array.layout != ALGORITHM_PARITY_N)
2917 return -1;
2918 if (array.level == 6 && array.layout != ALGORITHM_PARITY_N_6)
2919 return -1;
2920 sysfs_set_str(&info, NULL,"sync_action", "idle");
2921 /* First remove any spares so no recovery starts */
2922 for (d = 0, found = 0;
2923 d < MAX_DISKS && found < array.nr_disks; d++) {
2924 mdu_disk_info_t disk;
2925 disk.number = d;
2926 if (md_get_disk_info(fd, &disk) < 0)
2927 continue;
2928 if (disk.major == 0 && disk.minor == 0)
2929 continue;
2930 found++;
2931 if ((disk.state & (1 << MD_DISK_ACTIVE)) &&
2932 disk.raid_disk < data_disks)
2933 /* keep this */
2934 continue;
2935 ioctl(fd, HOT_REMOVE_DISK,
2936 makedev(disk.major, disk.minor));
2937 }
2938 /* Now fail anything left */
2939 md_get_array_info(fd, &array);
2940 for (d = 0, found = 0;
2941 d < MAX_DISKS && found < array.nr_disks; d++) {
2942 mdu_disk_info_t disk;
2943 disk.number = d;
2944 if (md_get_disk_info(fd, &disk) < 0)
2945 continue;
2946 if (disk.major == 0 && disk.minor == 0)
2947 continue;
2948 found++;
2949 if ((disk.state & (1 << MD_DISK_ACTIVE)) &&
2950 disk.raid_disk < data_disks)
2951 /* keep this */
2952 continue;
2953 ioctl(fd, SET_DISK_FAULTY,
2954 makedev(disk.major, disk.minor));
2955 hot_remove_disk(fd, makedev(disk.major, disk.minor), 1);
2956 }
2957 }
2958 c = map_num(pers, level);
2959 if (c) {
2960 int err = sysfs_set_str(&info, NULL, "level", c);
2961 if (err) {
2962 err = errno;
2963 pr_err("%s: could not set level to %s\n",
2964 devname, c);
2965 if (err == EBUSY &&
2966 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2967 cont_err("Bitmap must be removed before level can be changed\n");
2968 return err;
2969 }
2970 if (verbose >= 0)
2971 pr_err("level of %s changed to %s\n", devname, c);
2972 }
2973 return 0;
2974 }
2975
2976 int sigterm = 0;
2977 static void catch_term(int sig)
2978 {
2979 sigterm = 1;
2980 }
2981
2982 static int continue_via_systemd(char *devnm)
2983 {
2984 int skipped, i, pid, status;
2985 char pathbuf[1024];
2986 /* In a systemd/udev world, it is best to get systemd to
2987 * run "mdadm --grow --continue" rather than running in the
2988 * background.
2989 */
2990 switch(fork()) {
2991 case 0:
2992 /* FIXME yuk. CLOSE_EXEC?? */
2993 skipped = 0;
2994 for (i = 3; skipped < 20; i++)
2995 if (close(i) < 0)
2996 skipped++;
2997 else
2998 skipped = 0;
2999
3000 /* Don't want to see error messages from
3001 * systemctl. If the service doesn't exist,
3002 * we fork ourselves.
3003 */
3004 close(2);
3005 open("/dev/null", O_WRONLY);
3006 snprintf(pathbuf, sizeof(pathbuf),
3007 "mdadm-grow-continue@%s.service", devnm);
3008 status = execl("/usr/bin/systemctl", "systemctl", "restart",
3009 pathbuf, NULL);
3010 status = execl("/bin/systemctl", "systemctl", "restart",
3011 pathbuf, NULL);
3012 exit(1);
3013 case -1: /* Just do it ourselves. */
3014 break;
3015 default: /* parent - good */
3016 pid = wait(&status);
3017 if (pid >= 0 && status == 0)
3018 return 1;
3019 }
3020 return 0;
3021 }
3022
3023 static int reshape_array(char *container, int fd, char *devname,
3024 struct supertype *st, struct mdinfo *info,
3025 int force, struct mddev_dev *devlist,
3026 unsigned long long data_offset,
3027 char *backup_file, int verbose, int forked,
3028 int restart, int freeze_reshape)
3029 {
3030 struct reshape reshape;
3031 int spares_needed;
3032 char *msg;
3033 int orig_level = UnSet;
3034 int odisks;
3035 int delayed;
3036
3037 struct mdu_array_info_s array;
3038 char *c;
3039
3040 struct mddev_dev *dv;
3041 int added_disks;
3042
3043 int *fdlist = NULL;
3044 unsigned long long *offsets = NULL;
3045 int d;
3046 int nrdisks;
3047 int err;
3048 unsigned long blocks;
3049 unsigned long long array_size;
3050 int done;
3051 struct mdinfo *sra = NULL;
3052 char buf[20];
3053
3054 /* when reshaping a RAID0, the component_size might be zero.
3055 * So try to fix that up.
3056 */
3057 if (md_get_array_info(fd, &array) != 0) {
3058 dprintf("Cannot get array information.\n");
3059 goto release;
3060 }
3061 if (array.level == 0 && info->component_size == 0) {
3062 get_dev_size(fd, NULL, &array_size);
3063 info->component_size = array_size / array.raid_disks;
3064 }
3065
3066 if (array.level == 10)
3067 /* Need space_after info */
3068 get_space_after(fd, st, info);
3069
3070 if (info->reshape_active) {
3071 int new_level = info->new_level;
3072 info->new_level = UnSet;
3073 if (info->delta_disks > 0)
3074 info->array.raid_disks -= info->delta_disks;
3075 msg = analyse_change(devname, info, &reshape);
3076 info->new_level = new_level;
3077 if (info->delta_disks > 0)
3078 info->array.raid_disks += info->delta_disks;
3079 if (!restart)
3080 /* Make sure the array isn't read-only */
3081 ioctl(fd, RESTART_ARRAY_RW, 0);
3082 } else
3083 msg = analyse_change(devname, info, &reshape);
3084 if (msg) {
3085 /* if msg == "", error has already been printed */
3086 if (msg[0])
3087 pr_err("%s\n", msg);
3088 goto release;
3089 }
3090 if (restart && (reshape.level != info->array.level ||
3091 reshape.before.layout != info->array.layout ||
3092 reshape.before.data_disks + reshape.parity !=
3093 info->array.raid_disks - max(0, info->delta_disks))) {
3094 pr_err("reshape info is not in native format - cannot continue.\n");
3095 goto release;
3096 }
3097
3098 if (st->ss->external && restart && (info->reshape_progress == 0) &&
3099 !((sysfs_get_str(info, NULL, "sync_action",
3100 buf, sizeof(buf)) > 0) &&
3101 (strncmp(buf, "reshape", 7) == 0))) {
3102 /* When reshape is restarted from '0', very begin of array
3103 * it is possible that for external metadata reshape and array
3104 * configuration doesn't happen.
3105 * Check if md has the same opinion, and reshape is restarted
3106 * from 0. If so, this is regular reshape start after reshape
3107 * switch in metadata to next array only.
3108 */
3109 if ((verify_reshape_position(info, reshape.level) >= 0) &&
3110 (info->reshape_progress == 0))
3111 restart = 0;
3112 }
3113 if (restart) {
3114 /*
3115 * reshape already started. just skip to monitoring
3116 * the reshape
3117 */
3118 if (reshape.backup_blocks == 0)
3119 return 0;
3120 if (restart & RESHAPE_NO_BACKUP)
3121 return 0;
3122
3123 /* Need 'sra' down at 'started:' */
3124 sra = sysfs_read(fd, NULL,
3125 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|
3126 GET_CHUNK|GET_CACHE);
3127 if (!sra) {
3128 pr_err("%s: Cannot get array details from sysfs\n",
3129 devname);
3130 goto release;
3131 }
3132
3133 if (!backup_file)
3134 backup_file = locate_backup(sra->sys_name);
3135
3136 goto started;
3137 }
3138 /* The container is frozen but the array may not be.
3139 * So freeze the array so spares don't get put to the wrong use
3140 * FIXME there should probably be a cleaner separation between
3141 * freeze_array and freeze_container.
3142 */
3143 sysfs_freeze_array(info);
3144 /* Check we have enough spares to not be degraded */
3145 added_disks = 0;
3146 for (dv = devlist; dv ; dv=dv->next)
3147 added_disks++;
3148 spares_needed = max(reshape.before.data_disks,
3149 reshape.after.data_disks) +
3150 reshape.parity - array.raid_disks;
3151
3152 if (!force && info->new_level > 1 && info->array.level > 1 &&
3153 spares_needed > info->array.spare_disks + added_disks) {
3154 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3155 " Use --force to over-ride this check.\n",
3156 spares_needed,
3157 spares_needed == 1 ? "" : "s",
3158 info->array.spare_disks + added_disks);
3159 goto release;
3160 }
3161 /* Check we have enough spares to not fail */
3162 spares_needed = max(reshape.before.data_disks,
3163 reshape.after.data_disks)
3164 - array.raid_disks;
3165 if ((info->new_level > 1 || info->new_level == 0) &&
3166 spares_needed > info->array.spare_disks +added_disks) {
3167 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3168 spares_needed, spares_needed == 1 ? "" : "s",
3169 info->array.spare_disks + added_disks);
3170 goto release;
3171 }
3172
3173 if (reshape.level != array.level) {
3174 int err = impose_level(fd, reshape.level, devname, verbose);
3175 if (err)
3176 goto release;
3177 info->new_layout = UnSet; /* after level change,
3178 * layout is meaningless */
3179 orig_level = array.level;
3180 sysfs_freeze_array(info);
3181
3182 if (reshape.level > 0 && st->ss->external) {
3183 /* make sure mdmon is aware of the new level */
3184 if (mdmon_running(container))
3185 flush_mdmon(container);
3186
3187 if (!mdmon_running(container))
3188 start_mdmon(container);
3189 ping_monitor(container);
3190 if (mdmon_running(container) && st->update_tail == NULL)
3191 st->update_tail = &st->updates;
3192 }
3193 }
3194 /* ->reshape_super might have chosen some spares from the
3195 * container that it wants to be part of the new array.
3196 * We can collect them with ->container_content and give
3197 * them to the kernel.
3198 */
3199 if (st->ss->reshape_super && st->ss->container_content) {
3200 char *subarray = strchr(info->text_version+1, '/')+1;
3201 struct mdinfo *info2 =
3202 st->ss->container_content(st, subarray);
3203 struct mdinfo *d;
3204
3205 if (info2) {
3206 if (sysfs_init(info2, fd, st->devnm)) {
3207 pr_err("unable to initialize sysfs for %s\n",
3208 st->devnm);
3209 free(info2);
3210 goto release;
3211 }
3212 /* When increasing number of devices, we need to set
3213 * new raid_disks before adding these, or they might
3214 * be rejected.
3215 */
3216 if (reshape.backup_blocks &&
3217 reshape.after.data_disks >
3218 reshape.before.data_disks)
3219 subarray_set_num(container, info2, "raid_disks",
3220 reshape.after.data_disks +
3221 reshape.parity);
3222 for (d = info2->devs; d; d = d->next) {
3223 if (d->disk.state == 0 &&
3224 d->disk.raid_disk >= 0) {
3225 /* This is a spare that wants to
3226 * be part of the array.
3227 */
3228 add_disk(fd, st, info2, d);
3229 }
3230 }
3231 sysfs_free(info2);
3232 }
3233 }
3234 /* We might have been given some devices to add to the
3235 * array. Now that the array has been changed to the right
3236 * level and frozen, we can safely add them.
3237 */
3238 if (devlist) {
3239 if (Manage_subdevs(devname, fd, devlist, verbose, 0, NULL, 0))
3240 goto release;
3241 }
3242
3243 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3244 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3245 if (reshape.backup_blocks == 0) {
3246 /* No restriping needed, but we might need to impose
3247 * some more changes: layout, raid_disks, chunk_size
3248 */
3249 /* read current array info */
3250 if (md_get_array_info(fd, &array) != 0) {
3251 dprintf("Cannot get array information.\n");
3252 goto release;
3253 }
3254 /* compare current array info with new values and if
3255 * it is different update them to new */
3256 if (info->new_layout != UnSet &&
3257 info->new_layout != array.layout) {
3258 array.layout = info->new_layout;
3259 if (md_set_array_info(fd, &array) != 0) {
3260 pr_err("failed to set new layout\n");
3261 goto release;
3262 } else if (verbose >= 0)
3263 printf("layout for %s set to %d\n",
3264 devname, array.layout);
3265 }
3266 if (info->delta_disks != UnSet && info->delta_disks != 0 &&
3267 array.raid_disks !=
3268 (info->array.raid_disks + info->delta_disks)) {
3269 array.raid_disks += info->delta_disks;
3270 if (md_set_array_info(fd, &array) != 0) {
3271 pr_err("failed to set raid disks\n");
3272 goto release;
3273 } else if (verbose >= 0) {
3274 printf("raid_disks for %s set to %d\n",
3275 devname, array.raid_disks);
3276 }
3277 }
3278 if (info->new_chunk != 0 &&
3279 info->new_chunk != array.chunk_size) {
3280 if (sysfs_set_num(info, NULL,
3281 "chunk_size", info->new_chunk) != 0) {
3282 pr_err("failed to set chunk size\n");
3283 goto release;
3284 } else if (verbose >= 0)
3285 printf("chunk size for %s set to %d\n",
3286 devname, array.chunk_size);
3287 }
3288 unfreeze(st);
3289 return 0;
3290 }
3291
3292 /*
3293 * There are three possibilities.
3294 * 1/ The array will shrink.
3295 * We need to ensure the reshape will pause before reaching
3296 * the 'critical section'. We also need to fork and wait for
3297 * that to happen. When it does we
3298 * suspend/backup/complete/unfreeze
3299 *
3300 * 2/ The array will not change size.
3301 * This requires that we keep a backup of a sliding window
3302 * so that we can restore data after a crash. So we need
3303 * to fork and monitor progress.
3304 * In future we will allow the data_offset to change, so
3305 * a sliding backup becomes unnecessary.
3306 *
3307 * 3/ The array will grow. This is relatively easy.
3308 * However the kernel's restripe routines will cheerfully
3309 * overwrite some early data before it is safe. So we
3310 * need to make a backup of the early parts of the array
3311 * and be ready to restore it if rebuild aborts very early.
3312 * For externally managed metadata, we still need a forked
3313 * child to monitor the reshape and suspend IO over the region
3314 * that is being reshaped.
3315 *
3316 * We backup data by writing it to one spare, or to a
3317 * file which was given on command line.
3318 *
3319 * In each case, we first make sure that storage is available
3320 * for the required backup.
3321 * Then we:
3322 * - request the shape change.
3323 * - fork to handle backup etc.
3324 */
3325 /* Check that we can hold all the data */
3326 get_dev_size(fd, NULL, &array_size);
3327 if (reshape.new_size < (array_size/512)) {
3328 pr_err("this change will reduce the size of the array.\n"
3329 " use --grow --array-size first to truncate array.\n"
3330 " e.g. mdadm --grow %s --array-size %llu\n",
3331 devname, reshape.new_size/2);
3332 goto release;
3333 }
3334
3335 if (array.level == 10) {
3336 /* Reshaping RAID10 does not require any data backup by
3337 * user-space. Instead it requires that the data_offset
3338 * is changed to avoid the need for backup.
3339 * So this is handled very separately
3340 */
3341 if (restart)
3342 /* Nothing to do. */
3343 return 0;
3344 return raid10_reshape(container, fd, devname, st, info,
3345 &reshape, data_offset, force, verbose);
3346 }
3347 sra = sysfs_read(fd, NULL,
3348 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3349 GET_CACHE);
3350 if (!sra) {
3351 pr_err("%s: Cannot get array details from sysfs\n",
3352 devname);
3353 goto release;
3354 }
3355
3356 if (!backup_file)
3357 switch(set_new_data_offset(sra, st, devname,
3358 reshape.after.data_disks - reshape.before.data_disks,
3359 data_offset,
3360 reshape.min_offset_change, 1)) {
3361 case -1:
3362 goto release;
3363 case 0:
3364 /* Updated data_offset, so it's easy now */
3365 update_cache_size(container, sra, info,
3366 min(reshape.before.data_disks,
3367 reshape.after.data_disks),
3368 reshape.backup_blocks);
3369
3370 /* Right, everything seems fine. Let's kick things off.
3371 */
3372 sync_metadata(st);
3373
3374 if (impose_reshape(sra, info, st, fd, restart,
3375 devname, container, &reshape) < 0)
3376 goto release;
3377 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3378 struct mdinfo *sd;
3379 if (errno != EINVAL) {
3380 pr_err("Failed to initiate reshape!\n");
3381 goto release;
3382 }
3383 /* revert data_offset and try the old way */
3384 for (sd = sra->devs; sd; sd = sd->next) {
3385 sysfs_set_num(sra, sd, "new_offset",
3386 sd->data_offset);
3387 sysfs_set_str(sra, NULL, "reshape_direction",
3388 "forwards");
3389 }
3390 break;
3391 }
3392 if (info->new_level == reshape.level)
3393 return 0;
3394 /* need to adjust level when reshape completes */
3395 switch(fork()) {
3396 case -1: /* ignore error, but don't wait */
3397 return 0;
3398 default: /* parent */
3399 return 0;
3400 case 0:
3401 map_fork();
3402 break;
3403 }
3404 close(fd);
3405 wait_reshape(sra);
3406 fd = open_dev(sra->sys_name);
3407 if (fd >= 0)
3408 impose_level(fd, info->new_level, devname, verbose);
3409 return 0;
3410 case 1: /* Couldn't set data_offset, try the old way */
3411 if (data_offset != INVALID_SECTORS) {
3412 pr_err("Cannot update data_offset on this array\n");
3413 goto release;
3414 }
3415 break;
3416 }
3417
3418 started:
3419 /* Decide how many blocks (sectors) for a reshape
3420 * unit. The number we have so far is just a minimum
3421 */
3422 blocks = reshape.backup_blocks;
3423 if (reshape.before.data_disks ==
3424 reshape.after.data_disks) {
3425 /* Make 'blocks' bigger for better throughput, but
3426 * not so big that we reject it below.
3427 * Try for 16 megabytes
3428 */
3429 while (blocks * 32 < sra->component_size && blocks < 16*1024*2)
3430 blocks *= 2;
3431 } else
3432 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3433
3434 if (blocks >= sra->component_size/2) {
3435 pr_err("%s: Something wrong - reshape aborted\n", devname);
3436 goto release;
3437 }
3438
3439 /* Now we need to open all these devices so we can read/write.
3440 */
3441 nrdisks = max(reshape.before.data_disks,
3442 reshape.after.data_disks) + reshape.parity
3443 + sra->array.spare_disks;
3444 fdlist = xcalloc((1+nrdisks), sizeof(int));
3445 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3446
3447 odisks = reshape.before.data_disks + reshape.parity;
3448 d = reshape_prepare_fdlist(devname, sra, odisks, nrdisks, blocks,
3449 backup_file, fdlist, offsets);
3450 if (d < odisks) {
3451 goto release;
3452 }
3453 if ((st->ss->manage_reshape == NULL) ||
3454 (st->ss->recover_backup == NULL)) {
3455 if (backup_file == NULL) {
3456 if (reshape.after.data_disks <=
3457 reshape.before.data_disks) {
3458 pr_err("%s: Cannot grow - need backup-file\n",
3459 devname);
3460 pr_err(" Please provide one with \"--backup=...\"\n");
3461 goto release;
3462 } else if (d == odisks) {
3463 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3464 goto release;
3465 }
3466 } else {
3467 if (!reshape_open_backup_file(backup_file, fd, devname,
3468 (signed)blocks,
3469 fdlist+d, offsets+d,
3470 sra->sys_name, restart)) {
3471 goto release;
3472 }
3473 d++;
3474 }
3475 }
3476
3477 update_cache_size(container, sra, info,
3478 min(reshape.before.data_disks,
3479 reshape.after.data_disks), blocks);
3480
3481 /* Right, everything seems fine. Let's kick things off.
3482 * If only changing raid_disks, use ioctl, else use
3483 * sysfs.
3484 */
3485 sync_metadata(st);
3486
3487 if (impose_reshape(sra, info, st, fd, restart,
3488 devname, container, &reshape) < 0)
3489 goto release;
3490
3491 err = start_reshape(sra, restart, reshape.before.data_disks,
3492 reshape.after.data_disks);
3493 if (err) {
3494 pr_err("Cannot %s reshape for %s\n",
3495 restart ? "continue" : "start", devname);
3496 goto release;
3497 }
3498 if (restart)
3499 sysfs_set_str(sra, NULL, "array_state", "active");
3500 if (freeze_reshape) {
3501 free(fdlist);
3502 free(offsets);
3503 sysfs_free(sra);
3504 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3505 sra->reshape_progress);
3506 return 1;
3507 }
3508
3509 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3510 if (continue_via_systemd(container ?: sra->sys_name)) {
3511 free(fdlist);
3512 free(offsets);
3513 sysfs_free(sra);
3514 return 0;
3515 }
3516
3517 /* Now we just need to kick off the reshape and watch, while
3518 * handling backups of the data...
3519 * This is all done by a forked background process.
3520 */
3521 switch(forked ? 0 : fork()) {
3522 case -1:
3523 pr_err("Cannot run child to monitor reshape: %s\n",
3524 strerror(errno));
3525 abort_reshape(sra);
3526 goto release;
3527 default:
3528 free(fdlist);
3529 free(offsets);
3530 sysfs_free(sra);
3531 return 0;
3532 case 0:
3533 map_fork();
3534 break;
3535 }
3536
3537 /* If another array on the same devices is busy, the
3538 * reshape will wait for them. This would mean that
3539 * the first section that we suspend will stay suspended
3540 * for a long time. So check on that possibility
3541 * by looking for "DELAYED" in /proc/mdstat, and if found,
3542 * wait a while
3543 */
3544 do {
3545 struct mdstat_ent *mds, *m;
3546 delayed = 0;
3547 mds = mdstat_read(1, 0);
3548 for (m = mds; m; m = m->next)
3549 if (strcmp(m->devnm, sra->sys_name) == 0) {
3550 if (m->resync && m->percent == RESYNC_DELAYED)
3551 delayed = 1;
3552 if (m->resync == 0)
3553 /* Haven't started the reshape thread
3554 * yet, wait a bit
3555 */
3556 delayed = 2;
3557 break;
3558 }
3559 free_mdstat(mds);
3560 if (delayed == 1 && get_linux_version() < 3007000) {
3561 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3562 " You might experience problems until other reshapes complete.\n");
3563 delayed = 0;
3564 }
3565 if (delayed)
3566 mdstat_wait(30 - (delayed-1) * 25);
3567 } while (delayed);
3568 mdstat_close();
3569 close(fd);
3570 if (check_env("MDADM_GROW_VERIFY"))
3571 fd = open(devname, O_RDONLY | O_DIRECT);
3572 else
3573 fd = -1;
3574 mlockall(MCL_FUTURE);
3575
3576 signal(SIGTERM, catch_term);
3577
3578 if (st->ss->external) {
3579 /* metadata handler takes it from here */
3580 done = st->ss->manage_reshape(
3581 fd, sra, &reshape, st, blocks,
3582 fdlist, offsets, d - odisks, fdlist + odisks,
3583 offsets + odisks);
3584 } else
3585 done = child_monitor(
3586 fd, sra, &reshape, st, blocks, fdlist, offsets,
3587 d - odisks, fdlist + odisks, offsets + odisks);
3588
3589 free(fdlist);
3590 free(offsets);
3591
3592 if (backup_file && done) {
3593 char *bul;
3594 bul = make_backup(sra->sys_name);
3595 if (bul) {
3596 char buf[1024];
3597 int l = readlink(bul, buf, sizeof(buf) - 1);
3598 if (l > 0) {
3599 buf[l]=0;
3600 unlink(buf);
3601 }
3602 unlink(bul);
3603 free(bul);
3604 }
3605 unlink(backup_file);
3606 }
3607 if (!done) {
3608 abort_reshape(sra);
3609 goto out;
3610 }
3611
3612 if (!st->ss->external &&
3613 !(reshape.before.data_disks != reshape.after.data_disks &&
3614 info->custom_array_size) && info->new_level == reshape.level &&
3615 !forked) {
3616 /* no need to wait for the reshape to finish as
3617 * there is nothing more to do.
3618 */
3619 sysfs_free(sra);
3620 exit(0);
3621 }
3622 wait_reshape(sra);
3623
3624 if (st->ss->external) {
3625 /* Re-load the metadata as much could have changed */
3626 int cfd = open_dev(st->container_devnm);
3627 if (cfd >= 0) {
3628 flush_mdmon(container);
3629 st->ss->free_super(st);
3630 st->ss->load_container(st, cfd, container);
3631 close(cfd);
3632 }
3633 }
3634
3635 /* set new array size if required customer_array_size is used
3636 * by this metadata.
3637 */
3638 if (reshape.before.data_disks != reshape.after.data_disks &&
3639 info->custom_array_size)
3640 set_array_size(st, info, info->text_version);
3641
3642 if (info->new_level != reshape.level) {
3643 if (fd < 0)
3644 fd = open(devname, O_RDONLY);
3645 impose_level(fd, info->new_level, devname, verbose);
3646 close(fd);
3647 if (info->new_level == 0)
3648 st->update_tail = NULL;
3649 }
3650 out:
3651 sysfs_free(sra);
3652 if (forked)
3653 return 0;
3654 unfreeze(st);
3655 exit(0);
3656
3657 release:
3658 free(fdlist);
3659 free(offsets);
3660 if (orig_level != UnSet && sra) {
3661 c = map_num(pers, orig_level);
3662 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3663 pr_err("aborting level change\n");
3664 }
3665 sysfs_free(sra);
3666 if (!forked)
3667 unfreeze(st);
3668 return 1;
3669 }
3670
3671 /* mdfd handle is passed to be closed in child process (after fork).
3672 */
3673 int reshape_container(char *container, char *devname,
3674 int mdfd,
3675 struct supertype *st,
3676 struct mdinfo *info,
3677 int force,
3678 char *backup_file, int verbose,
3679 int forked, int restart, int freeze_reshape)
3680 {
3681 struct mdinfo *cc = NULL;
3682 int rv = restart;
3683 char last_devnm[32] = "";
3684
3685 /* component_size is not meaningful for a container,
3686 * so pass '0' meaning 'no change'
3687 */
3688 if (!restart &&
3689 reshape_super(st, 0, info->new_level,
3690 info->new_layout, info->new_chunk,
3691 info->array.raid_disks, info->delta_disks,
3692 backup_file, devname, APPLY_METADATA_CHANGES,
3693 verbose)) {
3694 unfreeze(st);
3695 return 1;
3696 }
3697
3698 sync_metadata(st);
3699
3700 /* ping monitor to be sure that update is on disk
3701 */
3702 ping_monitor(container);
3703
3704 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3705 if (continue_via_systemd(container))
3706 return 0;
3707
3708 switch (forked ? 0 : fork()) {
3709 case -1: /* error */
3710 perror("Cannot fork to complete reshape\n");
3711 unfreeze(st);
3712 return 1;
3713 default: /* parent */
3714 if (!freeze_reshape)
3715 printf("%s: multi-array reshape continues in background\n", Name);
3716 return 0;
3717 case 0: /* child */
3718 map_fork();
3719 break;
3720 }
3721
3722 /* close unused handle in child process
3723 */
3724 if (mdfd > -1)
3725 close(mdfd);
3726
3727 while(1) {
3728 /* For each member array with reshape_active,
3729 * we need to perform the reshape.
3730 * We pick the first array that needs reshaping and
3731 * reshape it. reshape_array() will re-read the metadata
3732 * so the next time through a different array should be
3733 * ready for reshape.
3734 * It is possible that the 'different' array will not
3735 * be assembled yet. In that case we simple exit.
3736 * When it is assembled, the mdadm which assembles it
3737 * will take over the reshape.
3738 */
3739 struct mdinfo *content;
3740 int fd;
3741 struct mdstat_ent *mdstat;
3742 char *adev;
3743 dev_t devid;
3744
3745 sysfs_free(cc);
3746
3747 cc = st->ss->container_content(st, NULL);
3748
3749 for (content = cc; content ; content = content->next) {
3750 char *subarray;
3751 if (!content->reshape_active)
3752 continue;
3753
3754 subarray = strchr(content->text_version+1, '/')+1;
3755 mdstat = mdstat_by_subdev(subarray, container);
3756 if (!mdstat)
3757 continue;
3758 if (mdstat->active == 0) {
3759 pr_err("Skipping inactive array %s.\n",
3760 mdstat->devnm);
3761 free_mdstat(mdstat);
3762 mdstat = NULL;
3763 continue;
3764 }
3765 break;
3766 }
3767 if (!content)
3768 break;
3769
3770 devid = devnm2devid(mdstat->devnm);
3771 adev = map_dev(major(devid), minor(devid), 0);
3772 if (!adev)
3773 adev = content->text_version;
3774
3775 fd = open_dev(mdstat->devnm);
3776 if (fd < 0) {
3777 pr_err("Device %s cannot be opened for reshape.\n",
3778 adev);
3779 break;
3780 }
3781
3782 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3783 /* Do not allow for multiple reshape_array() calls for
3784 * the same array.
3785 * It can happen when reshape_array() returns without
3786 * error, when reshape is not finished (wrong reshape
3787 * starting/continuation conditions). Mdmon doesn't
3788 * switch to next array in container and reentry
3789 * conditions for the same array occur.
3790 * This is possibly interim until the behaviour of
3791 * reshape_array is resolved().
3792 */
3793 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3794 close(fd);
3795 break;
3796 }
3797 strcpy(last_devnm, mdstat->devnm);
3798
3799 if (sysfs_init(content, fd, mdstat->devnm)) {
3800 pr_err("Unable to initialize sysfs for %s\n",
3801 mdstat->devnm);
3802 rv = 1;
3803 break;
3804 }
3805
3806 if (mdmon_running(container))
3807 flush_mdmon(container);
3808
3809 rv = reshape_array(container, fd, adev, st,
3810 content, force, NULL, INVALID_SECTORS,
3811 backup_file, verbose, 1, restart,
3812 freeze_reshape);
3813 close(fd);
3814
3815 if (freeze_reshape) {
3816 sysfs_free(cc);
3817 exit(0);
3818 }
3819
3820 restart = 0;
3821 if (rv)
3822 break;
3823
3824 if (mdmon_running(container))
3825 flush_mdmon(container);
3826 }
3827 if (!rv)
3828 unfreeze(st);
3829 sysfs_free(cc);
3830 exit(0);
3831 }
3832
3833 /*
3834 * We run a child process in the background which performs the following
3835 * steps:
3836 * - wait for resync to reach a certain point
3837 * - suspend io to the following section
3838 * - backup that section
3839 * - allow resync to proceed further
3840 * - resume io
3841 * - discard the backup.
3842 *
3843 * When are combined in slightly different ways in the three cases.
3844 * Grow:
3845 * - suspend/backup/allow/wait/resume/discard
3846 * Shrink:
3847 * - allow/wait/suspend/backup/allow/wait/resume/discard
3848 * same-size:
3849 * - wait/resume/discard/suspend/backup/allow
3850 *
3851 * suspend/backup/allow always come together
3852 * wait/resume/discard do too.
3853 * For the same-size case we have two backups to improve flow.
3854 *
3855 */
3856
3857 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3858 unsigned long long backup_point,
3859 unsigned long long wait_point,
3860 unsigned long long *suspend_point,
3861 unsigned long long *reshape_completed, int *frozen)
3862 {
3863 /* This function is called repeatedly by the reshape manager.
3864 * It determines how much progress can safely be made and allows
3865 * that progress.
3866 * - 'info' identifies the array and particularly records in
3867 * ->reshape_progress the metadata's knowledge of progress
3868 * This is a sector offset from the start of the array
3869 * of the next array block to be relocated. This number
3870 * may increase from 0 or decrease from array_size, depending
3871 * on the type of reshape that is happening.
3872 * Note that in contrast, 'sync_completed' is a block count of the
3873 * reshape so far. It gives the distance between the start point
3874 * (head or tail of device) and the next place that data will be
3875 * written. It always increases.
3876 * - 'reshape' is the structure created by analyse_change
3877 * - 'backup_point' shows how much the metadata manager has backed-up
3878 * data. For reshapes with increasing progress, it is the next address
3879 * to be backed up, previous addresses have been backed-up. For
3880 * decreasing progress, it is the earliest address that has been
3881 * backed up - later address are also backed up.
3882 * So addresses between reshape_progress and backup_point are
3883 * backed up providing those are in the 'correct' order.
3884 * - 'wait_point' is an array address. When reshape_completed
3885 * passes this point, progress_reshape should return. It might
3886 * return earlier if it determines that ->reshape_progress needs
3887 * to be updated or further backup is needed.
3888 * - suspend_point is maintained by progress_reshape and the caller
3889 * should not touch it except to initialise to zero.
3890 * It is an array address and it only increases in 2.6.37 and earlier.
3891 * This makes it difficult to handle reducing reshapes with
3892 * external metadata.
3893 * However: it is similar to backup_point in that it records the
3894 * other end of a suspended region from reshape_progress.
3895 * it is moved to extend the region that is safe to backup and/or
3896 * reshape
3897 * - reshape_completed is read from sysfs and returned. The caller
3898 * should copy this into ->reshape_progress when it has reason to
3899 * believe that the metadata knows this, and any backup outside this
3900 * has been erased.
3901 *
3902 * Return value is:
3903 * 1 if more data from backup_point - but only as far as suspend_point,
3904 * should be backed up
3905 * 0 if things are progressing smoothly
3906 * -1 if the reshape is finished because it is all done,
3907 * -2 if the reshape is finished due to an error.
3908 */
3909
3910 int advancing = (reshape->after.data_disks
3911 >= reshape->before.data_disks);
3912 unsigned long long need_backup; /* All data between start of array and
3913 * here will at some point need to
3914 * be backed up.
3915 */
3916 unsigned long long read_offset, write_offset;
3917 unsigned long long write_range;
3918 unsigned long long max_progress, target, completed;
3919 unsigned long long array_size = (info->component_size
3920 * reshape->before.data_disks);
3921 int fd;
3922 char buf[20];
3923
3924 /* First, we unsuspend any region that is now known to be safe.
3925 * If suspend_point is on the 'wrong' side of reshape_progress, then
3926 * we don't have or need suspension at the moment. This is true for
3927 * native metadata when we don't need to back-up.
3928 */
3929 if (advancing) {
3930 if (info->reshape_progress <= *suspend_point)
3931 sysfs_set_num(info, NULL, "suspend_lo",
3932 info->reshape_progress);
3933 } else {
3934 /* Note: this won't work in 2.6.37 and before.
3935 * Something somewhere should make sure we don't need it!
3936 */
3937 if (info->reshape_progress >= *suspend_point)
3938 sysfs_set_num(info, NULL, "suspend_hi",
3939 info->reshape_progress);
3940 }
3941
3942 /* Now work out how far it is safe to progress.
3943 * If the read_offset for ->reshape_progress is less than
3944 * 'blocks' beyond the write_offset, we can only progress as far
3945 * as a backup.
3946 * Otherwise we can progress until the write_offset for the new location
3947 * reaches (within 'blocks' of) the read_offset at the current location.
3948 * However that region must be suspended unless we are using native
3949 * metadata.
3950 * If we need to suspend more, we limit it to 128M per device, which is
3951 * rather arbitrary and should be some time-based calculation.
3952 */
3953 read_offset = info->reshape_progress / reshape->before.data_disks;
3954 write_offset = info->reshape_progress / reshape->after.data_disks;
3955 write_range = info->new_chunk/512;
3956 if (reshape->before.data_disks == reshape->after.data_disks)
3957 need_backup = array_size;
3958 else
3959 need_backup = reshape->backup_blocks;
3960 if (advancing) {
3961 if (read_offset < write_offset + write_range)
3962 max_progress = backup_point;
3963 else
3964 max_progress =
3965 read_offset * reshape->after.data_disks;
3966 } else {
3967 if (read_offset > write_offset - write_range)
3968 /* Can only progress as far as has been backed up,
3969 * which must be suspended */
3970 max_progress = backup_point;
3971 else if (info->reshape_progress <= need_backup)
3972 max_progress = backup_point;
3973 else {
3974 if (info->array.major_version >= 0)
3975 /* Can progress until backup is needed */
3976 max_progress = need_backup;
3977 else {
3978 /* Can progress until metadata update is required */
3979 max_progress =
3980 read_offset * reshape->after.data_disks;
3981 /* but data must be suspended */
3982 if (max_progress < *suspend_point)
3983 max_progress = *suspend_point;
3984 }
3985 }
3986 }
3987
3988 /* We know it is safe to progress to 'max_progress' providing
3989 * it is suspended or we are using native metadata.
3990 * Consider extending suspend_point 128M per device if it
3991 * is less than 64M per device beyond reshape_progress.
3992 * But always do a multiple of 'blocks'
3993 * FIXME this is too big - it takes to long to complete
3994 * this much.
3995 */
3996 target = 64*1024*2 * min(reshape->before.data_disks,
3997 reshape->after.data_disks);
3998 target /= reshape->backup_blocks;
3999 if (target < 2)
4000 target = 2;
4001 target *= reshape->backup_blocks;
4002
4003 /* For externally managed metadata we always need to suspend IO to
4004 * the area being reshaped so we regularly push suspend_point forward.
4005 * For native metadata we only need the suspend if we are going to do
4006 * a backup.
4007 */
4008 if (advancing) {
4009 if ((need_backup > info->reshape_progress ||
4010 info->array.major_version < 0) &&
4011 *suspend_point < info->reshape_progress + target) {
4012 if (need_backup < *suspend_point + 2 * target)
4013 *suspend_point = need_backup;
4014 else if (*suspend_point + 2 * target < array_size)
4015 *suspend_point += 2 * target;
4016 else
4017 *suspend_point = array_size;
4018 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
4019 if (max_progress > *suspend_point)
4020 max_progress = *suspend_point;
4021 }
4022 } else {
4023 if (info->array.major_version >= 0) {
4024 /* Only need to suspend when about to backup */
4025 if (info->reshape_progress < need_backup * 2 &&
4026 *suspend_point > 0) {
4027 *suspend_point = 0;
4028 sysfs_set_num(info, NULL, "suspend_lo", 0);
4029 sysfs_set_num(info, NULL, "suspend_hi",
4030 need_backup);
4031 }
4032 } else {
4033 /* Need to suspend continually */
4034 if (info->reshape_progress < *suspend_point)
4035 *suspend_point = info->reshape_progress;
4036 if (*suspend_point + target < info->reshape_progress)
4037 /* No need to move suspend region yet */;
4038 else {
4039 if (*suspend_point >= 2 * target)
4040 *suspend_point -= 2 * target;
4041 else
4042 *suspend_point = 0;
4043 sysfs_set_num(info, NULL, "suspend_lo",
4044 *suspend_point);
4045 }
4046 if (max_progress < *suspend_point)
4047 max_progress = *suspend_point;
4048 }
4049 }
4050
4051 /* now set sync_max to allow that progress. sync_max, like
4052 * sync_completed is a count of sectors written per device, so
4053 * we find the difference between max_progress and the start point,
4054 * and divide that by after.data_disks to get a sync_max
4055 * number.
4056 * At the same time we convert wait_point to a similar number
4057 * for comparing against sync_completed.
4058 */
4059 /* scale down max_progress to per_disk */
4060 max_progress /= reshape->after.data_disks;
4061 /*
4062 * Round to chunk size as some kernels give an erroneously
4063 * high number
4064 */
4065 max_progress /= info->new_chunk/512;
4066 max_progress *= info->new_chunk/512;
4067 /* And round to old chunk size as the kernel wants that */
4068 max_progress /= info->array.chunk_size/512;
4069 max_progress *= info->array.chunk_size/512;
4070 /* Limit progress to the whole device */
4071 if (max_progress > info->component_size)
4072 max_progress = info->component_size;
4073 wait_point /= reshape->after.data_disks;
4074 if (!advancing) {
4075 /* switch from 'device offset' to 'processed block count' */
4076 max_progress = info->component_size - max_progress;
4077 wait_point = info->component_size - wait_point;
4078 }
4079
4080 if (!*frozen)
4081 sysfs_set_num(info, NULL, "sync_max", max_progress);
4082
4083 /* Now wait. If we have already reached the point that we were
4084 * asked to wait to, don't wait at all, else wait for any change.
4085 * We need to select on 'sync_completed' as that is the place that
4086 * notifications happen, but we are really interested in
4087 * 'reshape_position'
4088 */
4089 fd = sysfs_get_fd(info, NULL, "sync_completed");
4090 if (fd < 0)
4091 goto check_progress;
4092
4093 if (sysfs_fd_get_ll(fd, &completed) < 0)
4094 goto check_progress;
4095
4096 while (completed < max_progress && completed < wait_point) {
4097 /* Check that sync_action is still 'reshape' to avoid
4098 * waiting forever on a dead array
4099 */
4100 char action[20];
4101 if (sysfs_get_str(info, NULL, "sync_action", action, 20) <= 0 ||
4102 strncmp(action, "reshape", 7) != 0)
4103 break;
4104 /* Some kernels reset 'sync_completed' to zero
4105 * before setting 'sync_action' to 'idle'.
4106 * So we need these extra tests.
4107 */
4108 if (completed == 0 && advancing &&
4109 strncmp(action, "idle", 4) == 0 &&
4110 info->reshape_progress > 0)
4111 break;
4112 if (completed == 0 && !advancing &&
4113 strncmp(action, "idle", 4) == 0 &&
4114 info->reshape_progress <
4115 (info->component_size * reshape->after.data_disks))
4116 break;
4117 sysfs_wait(fd, NULL);
4118 if (sysfs_fd_get_ll(fd, &completed) < 0)
4119 goto check_progress;
4120 }
4121 /* Some kernels reset 'sync_completed' to zero,
4122 * we need to have real point we are in md.
4123 * So in that case, read 'reshape_position' from sysfs.
4124 */
4125 if (completed == 0) {
4126 unsigned long long reshapep;
4127 char action[20];
4128 if (sysfs_get_str(info, NULL, "sync_action", action, 20) > 0 &&
4129 strncmp(action, "idle", 4) == 0 &&
4130 sysfs_get_ll(info, NULL,
4131 "reshape_position", &reshapep) == 0)
4132 *reshape_completed = reshapep;
4133 } else {
4134 /* some kernels can give an incorrectly high
4135 * 'completed' number, so round down */
4136 completed /= (info->new_chunk/512);
4137 completed *= (info->new_chunk/512);
4138 /* Convert 'completed' back in to a 'progress' number */
4139 completed *= reshape->after.data_disks;
4140 if (!advancing)
4141 completed = (info->component_size
4142 * reshape->after.data_disks
4143 - completed);
4144 *reshape_completed = completed;
4145 }
4146
4147 close(fd);
4148
4149 /* We return the need_backup flag. Caller will decide
4150 * how much - a multiple of ->backup_blocks up to *suspend_point
4151 */
4152 if (advancing)
4153 return need_backup > info->reshape_progress;
4154 else
4155 return need_backup >= info->reshape_progress;
4156
4157 check_progress:
4158 /* if we couldn't read a number from sync_completed, then
4159 * either the reshape did complete, or it aborted.
4160 * We can tell which by checking for 'none' in reshape_position.
4161 * If it did abort, then it might immediately restart if it
4162 * it was just a device failure that leaves us degraded but
4163 * functioning.
4164 */
4165 if (sysfs_get_str(info, NULL, "reshape_position", buf,
4166 sizeof(buf)) < 0 || strncmp(buf, "none", 4) != 0) {
4167 /* The abort might only be temporary. Wait up to 10
4168 * seconds for fd to contain a valid number again.
4169 */
4170 int wait = 10000;
4171 int rv = -2;
4172 unsigned long long new_sync_max;
4173 while (fd >= 0 && rv < 0 && wait > 0) {
4174 if (sysfs_wait(fd, &wait) != 1)
4175 break;
4176 switch (sysfs_fd_get_ll(fd, &completed)) {
4177 case 0:
4178 /* all good again */
4179 rv = 1;
4180 /* If "sync_max" is no longer max_progress
4181 * we need to freeze things
4182 */
4183 sysfs_get_ll(info, NULL, "sync_max",
4184 &new_sync_max);
4185 *frozen = (new_sync_max != max_progress);
4186 break;
4187 case -2: /* read error - abort */
4188 wait = 0;
4189 break;
4190 }
4191 }
4192 if (fd >= 0)
4193 close(fd);
4194 return rv; /* abort */
4195 } else {
4196 /* Maybe racing with array shutdown - check state */
4197 if (fd >= 0)
4198 close(fd);
4199 if (sysfs_get_str(info, NULL, "array_state", buf,
4200 sizeof(buf)) < 0 ||
4201 strncmp(buf, "inactive", 8) == 0 ||
4202 strncmp(buf, "clear",5) == 0)
4203 return -2; /* abort */
4204 return -1; /* complete */
4205 }
4206 }
4207
4208 /* FIXME return status is never checked */
4209 static int grow_backup(struct mdinfo *sra,
4210 unsigned long long offset, /* per device */
4211 unsigned long stripes, /* per device, in old chunks */
4212 int *sources, unsigned long long *offsets,
4213 int disks, int chunk, int level, int layout,
4214 int dests, int *destfd, unsigned long long *destoffsets,
4215 int part, int *degraded,
4216 char *buf)
4217 {
4218 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4219 * to storage 'destfd' (offset 'destoffsets'), after first
4220 * suspending IO. Then allow resync to continue
4221 * over the suspended section.
4222 * Use part 'part' of the backup-super-block.
4223 */
4224 int odata = disks;
4225 int rv = 0;
4226 int i;
4227 unsigned long long ll;
4228 int new_degraded;
4229 //printf("offset %llu\n", offset);
4230 if (level >= 4)
4231 odata--;
4232 if (level == 6)
4233 odata--;
4234
4235 /* Check that array hasn't become degraded, else we might backup the wrong data */
4236 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4237 return -1; /* FIXME this error is ignored */
4238 new_degraded = (int)ll;
4239 if (new_degraded != *degraded) {
4240 /* check each device to ensure it is still working */
4241 struct mdinfo *sd;
4242 for (sd = sra->devs ; sd ; sd = sd->next) {
4243 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4244 continue;
4245 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4246 char sbuf[100];
4247
4248 if (sysfs_get_str(sra, sd, "state",
4249 sbuf, sizeof(sbuf)) < 0 ||
4250 strstr(sbuf, "faulty") ||
4251 strstr(sbuf, "in_sync") == NULL) {
4252 /* this device is dead */
4253 sd->disk.state = (1<<MD_DISK_FAULTY);
4254 if (sd->disk.raid_disk >= 0 &&
4255 sources[sd->disk.raid_disk] >= 0) {
4256 close(sources[sd->disk.raid_disk]);
4257 sources[sd->disk.raid_disk] = -1;
4258 }
4259 }
4260 }
4261 }
4262 *degraded = new_degraded;
4263 }
4264 if (part) {
4265 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4266 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4267 } else {
4268 bsb.arraystart = __cpu_to_le64(offset * odata);
4269 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4270 }
4271 if (part)
4272 bsb.magic[15] = '2';
4273 for (i = 0; i < dests; i++)
4274 if (part)
4275 lseek64(destfd[i], destoffsets[i] +
4276 __le64_to_cpu(bsb.devstart2)*512, 0);
4277 else
4278 lseek64(destfd[i], destoffsets[i], 0);
4279
4280 rv = save_stripes(sources, offsets, disks, chunk, level, layout,
4281 dests, destfd, offset * 512 * odata,
4282 stripes * chunk * odata, buf);
4283
4284 if (rv)
4285 return rv;
4286 bsb.mtime = __cpu_to_le64(time(0));
4287 for (i = 0; i < dests; i++) {
4288 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4289
4290 bsb.sb_csum = bsb_csum((char*)&bsb,
4291 ((char*)&bsb.sb_csum)-((char*)&bsb));
4292 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4293 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4294 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4295
4296 rv = -1;
4297 if ((unsigned long long)lseek64(destfd[i],
4298 destoffsets[i] - 4096, 0) !=
4299 destoffsets[i] - 4096)
4300 break;
4301 if (write(destfd[i], &bsb, 512) != 512)
4302 break;
4303 if (destoffsets[i] > 4096) {
4304 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4305 destoffsets[i]+stripes*chunk*odata)
4306 break;
4307 if (write(destfd[i], &bsb, 512) != 512)
4308 break;
4309 }
4310 fsync(destfd[i]);
4311 rv = 0;
4312 }
4313
4314 return rv;
4315 }
4316
4317 /* in 2.6.30, the value reported by sync_completed can be
4318 * less that it should be by one stripe.
4319 * This only happens when reshape hits sync_max and pauses.
4320 * So allow wait_backup to either extent sync_max further
4321 * than strictly necessary, or return before the
4322 * sync has got quite as far as we would really like.
4323 * This is what 'blocks2' is for.
4324 * The various caller give appropriate values so that
4325 * every works.
4326 */
4327 /* FIXME return value is often ignored */
4328 static int forget_backup(int dests, int *destfd,
4329 unsigned long long *destoffsets,
4330 int part)
4331 {
4332 /*
4333 * Erase backup 'part' (which is 0 or 1)
4334 */
4335 int i;
4336 int rv;
4337
4338 if (part) {
4339 bsb.arraystart2 = __cpu_to_le64(0);
4340 bsb.length2 = __cpu_to_le64(0);
4341 } else {
4342 bsb.arraystart = __cpu_to_le64(0);
4343 bsb.length = __cpu_to_le64(0);
4344 }
4345 bsb.mtime = __cpu_to_le64(time(0));
4346 rv = 0;
4347 for (i = 0; i < dests; i++) {
4348 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4349 bsb.sb_csum = bsb_csum((char*)&bsb,
4350 ((char*)&bsb.sb_csum)-((char*)&bsb));
4351 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4352 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4353 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4354 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4355 destoffsets[i]-4096)
4356 rv = -1;
4357 if (rv == 0 && write(destfd[i], &bsb, 512) != 512)
4358 rv = -1;
4359 fsync(destfd[i]);
4360 }
4361 return rv;
4362 }
4363
4364 static void fail(char *msg)
4365 {
4366 int rv;
4367 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4368 rv |= (write(2, "\n", 1) != 1);
4369 exit(rv ? 1 : 2);
4370 }
4371
4372 static char *abuf, *bbuf;
4373 static unsigned long long abuflen;
4374 static void validate(int afd, int bfd, unsigned long long offset)
4375 {
4376 /* check that the data in the backup against the array.
4377 * This is only used for regression testing and should not
4378 * be used while the array is active
4379 */
4380 if (afd < 0)
4381 return;
4382 lseek64(bfd, offset - 4096, 0);
4383 if (read(bfd, &bsb2, 512) != 512)
4384 fail("cannot read bsb");
4385 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4386 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4387 fail("first csum bad");
4388 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4389 fail("magic is bad");
4390 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4391 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4392 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4393 fail("second csum bad");
4394
4395 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4396 fail("devstart is wrong");
4397
4398 if (bsb2.length) {
4399 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4400
4401 if (abuflen < len) {
4402 free(abuf);
4403 free(bbuf);
4404 abuflen = len;
4405 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4406 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4407 abuflen = 0;
4408 /* just stop validating on mem-alloc failure */
4409 return;
4410 }
4411 }
4412
4413 lseek64(bfd, offset, 0);
4414 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4415 //printf("len %llu\n", len);
4416 fail("read first backup failed");
4417 }
4418 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4419 if ((unsigned long long)read(afd, abuf, len) != len)
4420 fail("read first from array failed");
4421 if (memcmp(bbuf, abuf, len) != 0) {
4422 #if 0
4423 int i;
4424 printf("offset=%llu len=%llu\n",
4425 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4426 for (i=0; i<len; i++)
4427 if (bbuf[i] != abuf[i]) {
4428 printf("first diff byte %d\n", i);
4429 break;
4430 }
4431 #endif
4432 fail("data1 compare failed");
4433 }
4434 }
4435 if (bsb2.length2) {
4436 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4437
4438 if (abuflen < len) {
4439 free(abuf);
4440 free(bbuf);
4441 abuflen = len;
4442 abuf = xmalloc(abuflen);
4443 bbuf = xmalloc(abuflen);
4444 }
4445
4446 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4447 if ((unsigned long long)read(bfd, bbuf, len) != len)
4448 fail("read second backup failed");
4449 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4450 if ((unsigned long long)read(afd, abuf, len) != len)
4451 fail("read second from array failed");
4452 if (memcmp(bbuf, abuf, len) != 0)
4453 fail("data2 compare failed");
4454 }
4455 }
4456
4457 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4458 struct supertype *st, unsigned long blocks,
4459 int *fds, unsigned long long *offsets,
4460 int dests, int *destfd, unsigned long long *destoffsets)
4461 {
4462 /* Monitor a reshape where backup is being performed using
4463 * 'native' mechanism - either to a backup file, or
4464 * to some space in a spare.
4465 */
4466 char *buf;
4467 int degraded = -1;
4468 unsigned long long speed;
4469 unsigned long long suspend_point, array_size;
4470 unsigned long long backup_point, wait_point;
4471 unsigned long long reshape_completed;
4472 int done = 0;
4473 int increasing = reshape->after.data_disks >=
4474 reshape->before.data_disks;
4475 int part = 0; /* The next part of the backup area to fill. It
4476 * may already be full, so we need to check */
4477 int level = reshape->level;
4478 int layout = reshape->before.layout;
4479 int data = reshape->before.data_disks;
4480 int disks = reshape->before.data_disks + reshape->parity;
4481 int chunk = sra->array.chunk_size;
4482 struct mdinfo *sd;
4483 unsigned long stripes;
4484 int uuid[4];
4485 int frozen = 0;
4486
4487 /* set up the backup-super-block. This requires the
4488 * uuid from the array.
4489 */
4490 /* Find a superblock */
4491 for (sd = sra->devs; sd; sd = sd->next) {
4492 char *dn;
4493 int devfd;
4494 int ok;
4495 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4496 continue;
4497 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4498 devfd = dev_open(dn, O_RDONLY);
4499 if (devfd < 0)
4500 continue;
4501 ok = st->ss->load_super(st, devfd, NULL);
4502 close(devfd);
4503 if (ok == 0)
4504 break;
4505 }
4506 if (!sd) {
4507 pr_err("Cannot find a superblock\n");
4508 return 0;
4509 }
4510
4511 memset(&bsb, 0, 512);
4512 memcpy(bsb.magic, "md_backup_data-1", 16);
4513 st->ss->uuid_from_super(st, uuid);
4514 memcpy(bsb.set_uuid, uuid, 16);
4515 bsb.mtime = __cpu_to_le64(time(0));
4516 bsb.devstart2 = blocks;
4517
4518 stripes = blocks / (sra->array.chunk_size/512) /
4519 reshape->before.data_disks;
4520
4521 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4522 /* Don't start the 'reshape' */
4523 return 0;
4524 if (reshape->before.data_disks == reshape->after.data_disks) {
4525 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4526 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4527 }
4528
4529 if (increasing) {
4530 array_size = sra->component_size * reshape->after.data_disks;
4531 backup_point = sra->reshape_progress;
4532 suspend_point = 0;
4533 } else {
4534 array_size = sra->component_size * reshape->before.data_disks;
4535 backup_point = reshape->backup_blocks;
4536 suspend_point = array_size;
4537 }
4538
4539 while (!done) {
4540 int rv;
4541
4542 /* Want to return as soon the oldest backup slot can
4543 * be released as that allows us to start backing up
4544 * some more, providing suspend_point has been
4545 * advanced, which it should have.
4546 */
4547 if (increasing) {
4548 wait_point = array_size;
4549 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4550 wait_point = (__le64_to_cpu(bsb.arraystart) +
4551 __le64_to_cpu(bsb.length));
4552 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4553 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4554 __le64_to_cpu(bsb.length2));
4555 } else {
4556 wait_point = 0;
4557 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4558 wait_point = __le64_to_cpu(bsb.arraystart);
4559 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4560 wait_point = __le64_to_cpu(bsb.arraystart2);
4561 }
4562
4563 reshape_completed = sra->reshape_progress;
4564 rv = progress_reshape(sra, reshape,
4565 backup_point, wait_point,
4566 &suspend_point, &reshape_completed,
4567 &frozen);
4568 /* external metadata would need to ping_monitor here */
4569 sra->reshape_progress = reshape_completed;
4570
4571 /* Clear any backup region that is before 'here' */
4572 if (increasing) {
4573 if (__le64_to_cpu(bsb.length) > 0 &&
4574 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4575 __le64_to_cpu(bsb.length)))
4576 forget_backup(dests, destfd,
4577 destoffsets, 0);
4578 if (__le64_to_cpu(bsb.length2) > 0 &&
4579 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4580 __le64_to_cpu(bsb.length2)))
4581 forget_backup(dests, destfd,
4582 destoffsets, 1);
4583 } else {
4584 if (__le64_to_cpu(bsb.length) > 0 &&
4585 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4586 forget_backup(dests, destfd,
4587 destoffsets, 0);
4588 if (__le64_to_cpu(bsb.length2) > 0 &&
4589 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4590 forget_backup(dests, destfd,
4591 destoffsets, 1);
4592 }
4593 if (sigterm)
4594 rv = -2;
4595 if (rv < 0) {
4596 if (rv == -1)
4597 done = 1;
4598 break;
4599 }
4600 if (rv == 0 && increasing && !st->ss->external) {
4601 /* No longer need to monitor this reshape */
4602 sysfs_set_str(sra, NULL, "sync_max", "max");
4603 done = 1;
4604 break;
4605 }
4606
4607 while (rv) {
4608 unsigned long long offset;
4609 unsigned long actual_stripes;
4610 /* Need to backup some data.
4611 * If 'part' is not used and the desired
4612 * backup size is suspended, do a backup,
4613 * then consider the next part.
4614 */
4615 /* Check that 'part' is unused */
4616 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4617 break;
4618 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4619 break;
4620
4621 offset = backup_point / data;
4622 actual_stripes = stripes;
4623 if (increasing) {
4624 if (offset + actual_stripes * (chunk/512) >
4625 sra->component_size)
4626 actual_stripes = ((sra->component_size - offset)
4627 / (chunk/512));
4628 if (offset + actual_stripes * (chunk/512) >
4629 suspend_point/data)
4630 break;
4631 } else {
4632 if (offset < actual_stripes * (chunk/512))
4633 actual_stripes = offset / (chunk/512);
4634 offset -= actual_stripes * (chunk/512);
4635 if (offset < suspend_point/data)
4636 break;
4637 }
4638 if (actual_stripes == 0)
4639 break;
4640 grow_backup(sra, offset, actual_stripes, fds, offsets,
4641 disks, chunk, level, layout, dests, destfd,
4642 destoffsets, part, &degraded, buf);
4643 validate(afd, destfd[0], destoffsets[0]);
4644 /* record where 'part' is up to */
4645 part = !part;
4646 if (increasing)
4647 backup_point += actual_stripes * (chunk/512) * data;
4648 else
4649 backup_point -= actual_stripes * (chunk/512) * data;
4650 }
4651 }
4652
4653 /* FIXME maybe call progress_reshape one more time instead */
4654 /* remove any remaining suspension */
4655 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4656 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4657 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4658 sysfs_set_num(sra, NULL, "sync_min", 0);
4659
4660 if (reshape->before.data_disks == reshape->after.data_disks)
4661 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4662 free(buf);
4663 return done;
4664 }
4665
4666 /*
4667 * If any spare contains md_back_data-1 which is recent wrt mtime,
4668 * write that data into the array and update the super blocks with
4669 * the new reshape_progress
4670 */
4671 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist,
4672 int cnt, char *backup_file, int verbose)
4673 {
4674 int i, j;
4675 int old_disks;
4676 unsigned long long *offsets;
4677 unsigned long long nstripe, ostripe;
4678 int ndata, odata;
4679
4680 odata = info->array.raid_disks - info->delta_disks - 1;
4681 if (info->array.level == 6)
4682 odata--; /* number of data disks */
4683 ndata = info->array.raid_disks - 1;
4684 if (info->new_level == 6)
4685 ndata--;
4686
4687 old_disks = info->array.raid_disks - info->delta_disks;
4688
4689 if (info->delta_disks <= 0)
4690 /* Didn't grow, so the backup file must have
4691 * been used
4692 */
4693 old_disks = cnt;
4694 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4695 struct mdinfo dinfo;
4696 int fd;
4697 int bsbsize;
4698 char *devname, namebuf[20];
4699 unsigned long long lo, hi;
4700
4701 /* This was a spare and may have some saved data on it.
4702 * Load the superblock, find and load the
4703 * backup_super_block.
4704 * If either fail, go on to next device.
4705 * If the backup contains no new info, just return
4706 * else restore data and update all superblocks
4707 */
4708 if (i == old_disks-1) {
4709 fd = open(backup_file, O_RDONLY);
4710 if (fd<0) {
4711 pr_err("backup file %s inaccessible: %s\n",
4712 backup_file, strerror(errno));
4713 continue;
4714 }
4715 devname = backup_file;
4716 } else {
4717 fd = fdlist[i];
4718 if (fd < 0)
4719 continue;
4720 if (st->ss->load_super(st, fd, NULL))
4721 continue;
4722
4723 st->ss->getinfo_super(st, &dinfo, NULL);
4724 st->ss->free_super(st);
4725
4726 if (lseek64(fd,
4727 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4728 0) < 0) {
4729 pr_err("Cannot seek on device %d\n", i);
4730 continue; /* Cannot seek */
4731 }
4732 sprintf(namebuf, "device-%d", i);
4733 devname = namebuf;
4734 }
4735 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4736 if (verbose)
4737 pr_err("Cannot read from %s\n", devname);
4738 continue; /* Cannot read */
4739 }
4740 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4741 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4742 if (verbose)
4743 pr_err("No backup metadata on %s\n", devname);
4744 continue;
4745 }
4746 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4747 if (verbose)
4748 pr_err("Bad backup-metadata checksum on %s\n",
4749 devname);
4750 continue; /* bad checksum */
4751 }
4752 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4753 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4754 if (verbose)
4755 pr_err("Bad backup-metadata checksum2 on %s\n",
4756 devname);
4757 continue; /* Bad second checksum */
4758 }
4759 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4760 if (verbose)
4761 pr_err("Wrong uuid on backup-metadata on %s\n",
4762 devname);
4763 continue; /* Wrong uuid */
4764 }
4765
4766 /*
4767 * array utime and backup-mtime should be updated at
4768 * much the same time, but it seems that sometimes
4769 * they aren't... So allow considerable flexability in
4770 * matching, and allow this test to be overridden by
4771 * an environment variable.
4772 */
4773 if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) ||
4774 time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) {
4775 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4776 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4777 (unsigned long)__le64_to_cpu(bsb.mtime),
4778 (unsigned long)info->array.utime);
4779 } else {
4780 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4781 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4782 continue; /* time stamp is too bad */
4783 }
4784 }
4785
4786 if (bsb.magic[15] == '1') {
4787 if (bsb.length == 0)
4788 continue;
4789 if (info->delta_disks >= 0) {
4790 /* reshape_progress is increasing */
4791 if (__le64_to_cpu(bsb.arraystart)
4792 + __le64_to_cpu(bsb.length)
4793 < info->reshape_progress) {
4794 nonew:
4795 if (verbose)
4796 pr_err("backup-metadata found on %s but is not needed\n", devname);
4797 continue; /* No new data here */
4798 }
4799 } else {
4800 /* reshape_progress is decreasing */
4801 if (__le64_to_cpu(bsb.arraystart) >=
4802 info->reshape_progress)
4803 goto nonew; /* No new data here */
4804 }
4805 } else {
4806 if (bsb.length == 0 && bsb.length2 == 0)
4807 continue;
4808 if (info->delta_disks >= 0) {
4809 /* reshape_progress is increasing */
4810 if ((__le64_to_cpu(bsb.arraystart)
4811 + __le64_to_cpu(bsb.length)
4812 < info->reshape_progress) &&
4813 (__le64_to_cpu(bsb.arraystart2)
4814 + __le64_to_cpu(bsb.length2)
4815 < info->reshape_progress))
4816 goto nonew; /* No new data here */
4817 } else {
4818 /* reshape_progress is decreasing */
4819 if (__le64_to_cpu(bsb.arraystart) >=
4820 info->reshape_progress &&
4821 __le64_to_cpu(bsb.arraystart2) >=
4822 info->reshape_progress)
4823 goto nonew; /* No new data here */
4824 }
4825 }
4826 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4827 second_fail:
4828 if (verbose)
4829 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4830 devname);
4831 continue; /* Cannot seek */
4832 }
4833 /* There should be a duplicate backup superblock 4k before here */
4834 if (lseek64(fd, -4096, 1) < 0 ||
4835 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4836 goto second_fail; /* Cannot find leading superblock */
4837 if (bsb.magic[15] == '1')
4838 bsbsize = offsetof(struct mdp_backup_super, pad1);
4839 else
4840 bsbsize = offsetof(struct mdp_backup_super, pad);
4841 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4842 goto second_fail; /* Cannot find leading superblock */
4843
4844 /* Now need the data offsets for all devices. */
4845 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4846 for(j=0; j<info->array.raid_disks; j++) {
4847 if (fdlist[j] < 0)
4848 continue;
4849 if (st->ss->load_super(st, fdlist[j], NULL))
4850 /* FIXME should be this be an error */
4851 continue;
4852 st->ss->getinfo_super(st, &dinfo, NULL);
4853 st->ss->free_super(st);
4854 offsets[j] = dinfo.data_offset * 512;
4855 }
4856 printf("%s: restoring critical section\n", Name);
4857
4858 if (restore_stripes(fdlist, offsets, info->array.raid_disks,
4859 info->new_chunk, info->new_level,
4860 info->new_layout, fd,
4861 __le64_to_cpu(bsb.devstart)*512,
4862 __le64_to_cpu(bsb.arraystart)*512,
4863 __le64_to_cpu(bsb.length)*512, NULL)) {
4864 /* didn't succeed, so giveup */
4865 if (verbose)
4866 pr_err("Error restoring backup from %s\n",
4867 devname);
4868 free(offsets);
4869 return 1;
4870 }
4871
4872 if (bsb.magic[15] == '2' &&
4873 restore_stripes(fdlist, offsets, info->array.raid_disks,
4874 info->new_chunk, info->new_level,
4875 info->new_layout, fd,
4876 __le64_to_cpu(bsb.devstart)*512 +
4877 __le64_to_cpu(bsb.devstart2)*512,
4878 __le64_to_cpu(bsb.arraystart2)*512,
4879 __le64_to_cpu(bsb.length2)*512, NULL)) {
4880 /* didn't succeed, so giveup */
4881 if (verbose)
4882 pr_err("Error restoring second backup from %s\n",
4883 devname);
4884 free(offsets);
4885 return 1;
4886 }
4887
4888 free(offsets);
4889
4890 /* Ok, so the data is restored. Let's update those superblocks. */
4891
4892 lo = hi = 0;
4893 if (bsb.length) {
4894 lo = __le64_to_cpu(bsb.arraystart);
4895 hi = lo + __le64_to_cpu(bsb.length);
4896 }
4897 if (bsb.magic[15] == '2' && bsb.length2) {
4898 unsigned long long lo1, hi1;
4899 lo1 = __le64_to_cpu(bsb.arraystart2);
4900 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4901 if (lo == hi) {
4902 lo = lo1;
4903 hi = hi1;
4904 } else if (lo < lo1)
4905 hi = hi1;
4906 else
4907 lo = lo1;
4908 }
4909 if (lo < hi && (info->reshape_progress < lo ||
4910 info->reshape_progress > hi))
4911 /* backup does not affect reshape_progress*/ ;
4912 else if (info->delta_disks >= 0) {
4913 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4914 __le64_to_cpu(bsb.length);
4915 if (bsb.magic[15] == '2') {
4916 unsigned long long p2;
4917
4918 p2 = __le64_to_cpu(bsb.arraystart2) +
4919 __le64_to_cpu(bsb.length2);
4920 if (p2 > info->reshape_progress)
4921 info->reshape_progress = p2;
4922 }
4923 } else {
4924 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4925 if (bsb.magic[15] == '2') {
4926 unsigned long long p2;
4927
4928 p2 = __le64_to_cpu(bsb.arraystart2);
4929 if (p2 < info->reshape_progress)
4930 info->reshape_progress = p2;
4931 }
4932 }
4933 for (j=0; j<info->array.raid_disks; j++) {
4934 if (fdlist[j] < 0)
4935 continue;
4936 if (st->ss->load_super(st, fdlist[j], NULL))
4937 continue;
4938 st->ss->getinfo_super(st, &dinfo, NULL);
4939 dinfo.reshape_progress = info->reshape_progress;
4940 st->ss->update_super(st, &dinfo, "_reshape_progress",
4941 NULL,0, 0, NULL);
4942 st->ss->store_super(st, fdlist[j]);
4943 st->ss->free_super(st);
4944 }
4945 return 0;
4946 }
4947 /* Didn't find any backup data, try to see if any
4948 * was needed.
4949 */
4950 if (info->delta_disks < 0) {
4951 /* When shrinking, the critical section is at the end.
4952 * So see if we are before the critical section.
4953 */
4954 unsigned long long first_block;
4955 nstripe = ostripe = 0;
4956 first_block = 0;
4957 while (ostripe >= nstripe) {
4958 ostripe += info->array.chunk_size / 512;
4959 first_block = ostripe * odata;
4960 nstripe = first_block / ndata / (info->new_chunk/512) *
4961 (info->new_chunk/512);
4962 }
4963
4964 if (info->reshape_progress >= first_block)
4965 return 0;
4966 }
4967 if (info->delta_disks > 0) {
4968 /* See if we are beyond the critical section. */
4969 unsigned long long last_block;
4970 nstripe = ostripe = 0;
4971 last_block = 0;
4972 while (nstripe >= ostripe) {
4973 nstripe += info->new_chunk / 512;
4974 last_block = nstripe * ndata;
4975 ostripe = last_block / odata / (info->array.chunk_size/512) *
4976 (info->array.chunk_size/512);
4977 }
4978
4979 if (info->reshape_progress >= last_block)
4980 return 0;
4981 }
4982 /* needed to recover critical section! */
4983 if (verbose)
4984 pr_err("Failed to find backup of critical section\n");
4985 return 1;
4986 }
4987
4988 int Grow_continue_command(char *devname, int fd,
4989 char *backup_file, int verbose)
4990 {
4991 int ret_val = 0;
4992 struct supertype *st = NULL;
4993 struct mdinfo *content = NULL;
4994 struct mdinfo array;
4995 char *subarray = NULL;
4996 struct mdinfo *cc = NULL;
4997 struct mdstat_ent *mdstat = NULL;
4998 int cfd = -1;
4999 int fd2;
5000
5001 dprintf("Grow continue from command line called for %s\n", devname);
5002
5003 st = super_by_fd(fd, &subarray);
5004 if (!st || !st->ss) {
5005 pr_err("Unable to determine metadata format for %s\n", devname);
5006 return 1;
5007 }
5008 dprintf("Grow continue is run for ");
5009 if (st->ss->external == 0) {
5010 int d;
5011 int cnt = 5;
5012 dprintf_cont("native array (%s)\n", devname);
5013 if (md_get_array_info(fd, &array.array) < 0) {
5014 pr_err("%s is not an active md array - aborting\n",
5015 devname);
5016 ret_val = 1;
5017 goto Grow_continue_command_exit;
5018 }
5019 content = &array;
5020 sysfs_init(content, fd, NULL);
5021 /* Need to load a superblock.
5022 * FIXME we should really get what we need from
5023 * sysfs
5024 */
5025 do {
5026 for (d = 0; d < MAX_DISKS; d++) {
5027 mdu_disk_info_t disk;
5028 char *dv;
5029 int err;
5030 disk.number = d;
5031 if (md_get_disk_info(fd, &disk) < 0)
5032 continue;
5033 if (disk.major == 0 && disk.minor == 0)
5034 continue;
5035 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
5036 continue;
5037 dv = map_dev(disk.major, disk.minor, 1);
5038 if (!dv)
5039 continue;
5040 fd2 = dev_open(dv, O_RDONLY);
5041 if (fd2 < 0)
5042 continue;
5043 err = st->ss->load_super(st, fd2, NULL);
5044 close(fd2);
5045 if (err)
5046 continue;
5047 break;
5048 }
5049 if (d == MAX_DISKS) {
5050 pr_err("Unable to load metadata for %s\n",
5051 devname);
5052 ret_val = 1;
5053 goto Grow_continue_command_exit;
5054 }
5055 st->ss->getinfo_super(st, content, NULL);
5056 if (!content->reshape_active)
5057 sleep(3);
5058 else
5059 break;
5060 } while (cnt-- > 0);
5061 } else {
5062 char *container;
5063
5064 if (subarray) {
5065 dprintf_cont("subarray (%s)\n", subarray);
5066 container = st->container_devnm;
5067 cfd = open_dev_excl(st->container_devnm);
5068 } else {
5069 container = st->devnm;
5070 close(fd);
5071 cfd = open_dev_excl(st->devnm);
5072 dprintf_cont("container (%s)\n", container);
5073 fd = cfd;
5074 }
5075 if (cfd < 0) {
5076 pr_err("Unable to open container for %s\n", devname);
5077 ret_val = 1;
5078 goto Grow_continue_command_exit;
5079 }
5080
5081 /* find in container array under reshape
5082 */
5083 ret_val = st->ss->load_container(st, cfd, NULL);
5084 if (ret_val) {
5085 pr_err("Cannot read superblock for %s\n", devname);
5086 ret_val = 1;
5087 goto Grow_continue_command_exit;
5088 }
5089
5090 cc = st->ss->container_content(st, subarray);
5091 for (content = cc; content ; content = content->next) {
5092 char *array_name;
5093 int allow_reshape = 1;
5094
5095 if (content->reshape_active == 0)
5096 continue;
5097 /* The decision about array or container wide
5098 * reshape is taken in Grow_continue based
5099 * content->reshape_active state, therefore we
5100 * need to check_reshape based on
5101 * reshape_active and subarray name
5102 */
5103 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
5104 allow_reshape = 0;
5105 if (content->reshape_active == CONTAINER_RESHAPE &&
5106 (content->array.state
5107 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
5108 allow_reshape = 0;
5109
5110 if (!allow_reshape) {
5111 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5112 devname, container);
5113 ret_val = 1;
5114 goto Grow_continue_command_exit;
5115 }
5116
5117 array_name = strchr(content->text_version+1, '/')+1;
5118 mdstat = mdstat_by_subdev(array_name, container);
5119 if (!mdstat)
5120 continue;
5121 if (mdstat->active == 0) {
5122 pr_err("Skipping inactive array %s.\n",
5123 mdstat->devnm);
5124 free_mdstat(mdstat);
5125 mdstat = NULL;
5126 continue;
5127 }
5128 break;
5129 }
5130 if (!content) {
5131 pr_err("Unable to determine reshaped array for %s\n", devname);
5132 ret_val = 1;
5133 goto Grow_continue_command_exit;
5134 }
5135 fd2 = open_dev(mdstat->devnm);
5136 if (fd2 < 0) {
5137 pr_err("cannot open (%s)\n", mdstat->devnm);
5138 ret_val = 1;
5139 goto Grow_continue_command_exit;
5140 }
5141
5142 if (sysfs_init(content, fd2, mdstat->devnm)) {
5143 pr_err("Unable to initialize sysfs for %s, Grow cannot continue.\n",
5144 mdstat->devnm);
5145 ret_val = 1;
5146 close(fd2);
5147 goto Grow_continue_command_exit;
5148 }
5149
5150 close(fd2);
5151
5152 /* start mdmon in case it is not running
5153 */
5154 if (!mdmon_running(container))
5155 start_mdmon(container);
5156 ping_monitor(container);
5157
5158 if (mdmon_running(container))
5159 st->update_tail = &st->updates;
5160 else {
5161 pr_err("No mdmon found. Grow cannot continue.\n");
5162 ret_val = 1;
5163 goto Grow_continue_command_exit;
5164 }
5165 }
5166
5167 /* verify that array under reshape is started from
5168 * correct position
5169 */
5170 if (verify_reshape_position(content, content->array.level) < 0) {
5171 ret_val = 1;
5172 goto Grow_continue_command_exit;
5173 }
5174
5175 /* continue reshape
5176 */
5177 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
5178
5179 Grow_continue_command_exit:
5180 if (cfd > -1)
5181 close(cfd);
5182 st->ss->free_super(st);
5183 free_mdstat(mdstat);
5184 sysfs_free(cc);
5185 free(subarray);
5186
5187 return ret_val;
5188 }
5189
5190 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
5191 char *backup_file, int forked, int freeze_reshape)
5192 {
5193 int ret_val = 2;
5194
5195 if (!info->reshape_active)
5196 return ret_val;
5197
5198 if (st->ss->external) {
5199 int cfd = open_dev(st->container_devnm);
5200
5201 if (cfd < 0)
5202 return 1;
5203
5204 st->ss->load_container(st, cfd, st->container_devnm);
5205 close(cfd);
5206 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
5207 st, info, 0, backup_file, 0,
5208 forked, 1 | info->reshape_active,
5209 freeze_reshape);
5210 } else
5211 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
5212 NULL, INVALID_SECTORS, backup_file,
5213 0, forked, 1 | info->reshape_active,
5214 freeze_reshape);
5215
5216 return ret_val;
5217 }
5218
5219 char *make_backup(char *name)
5220 {
5221 char *base = "backup_file-";
5222 int len;
5223 char *fname;
5224
5225 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
5226 fname = xmalloc(len);
5227 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
5228 return fname;
5229 }
5230
5231 char *locate_backup(char *name)
5232 {
5233 char *fl = make_backup(name);
5234 struct stat stb;
5235
5236 if (stat(fl, &stb) == 0 && S_ISREG(stb.st_mode))
5237 return fl;
5238
5239 free(fl);
5240 return NULL;
5241 }
Unnamed repository; edit this file 'description' to name the repository.