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