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


af8d5205848007dbd5fca12ce2e7dc85bb90d3e6
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27 #include <stddef.h>
28 #include <stdint.h>
29 #include <signal.h>
30 #include <sys/wait.h>
31
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
34 #endif
35 #include "md_u.h"
36 #include "md_p.h"
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char **backup_filep,
43 int verbose)
44 {
45 int i;
46 int *fdlist;
47 struct mdinfo *dev;
48 int err;
49 int disk_count = next_spare + working_disks;
50 char *backup_file = *backup_filep;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61 sprintf(buf, "%d:%d",
62 dev->disk.major,
63 dev->disk.minor);
64 fd = dev_open(buf, O_RDWR);
65
66 if (dev->disk.raid_disk >= 0)
67 fdlist[dev->disk.raid_disk] = fd;
68 else
69 fdlist[next_spare++] = fd;
70 }
71
72 if (!backup_file) {
73 backup_file = locate_backup(content->sys_name);
74 *backup_filep = backup_file;
75 }
76
77 if (st->ss->external && st->ss->recover_backup)
78 err = st->ss->recover_backup(st, content);
79 else
80 err = Grow_restart(st, content, fdlist, next_spare,
81 backup_file, verbose > 0);
82
83 while (next_spare > 0) {
84 next_spare--;
85 if (fdlist[next_spare] >= 0)
86 close(fdlist[next_spare]);
87 }
88 free(fdlist);
89 if (err) {
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
91 if (!backup_file)
92 pr_err("Possibly you need to specify a --backup-file\n");
93 return 1;
94 }
95
96 dprintf("restore_backup() returns status OK.\n");
97 return 0;
98 }
99
100 int Grow_Add_device(char *devname, int fd, char *newdev)
101 {
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
107 * all other devices.
108 * This means that we need to *find* all other devices.
109 */
110 struct mdinfo info;
111
112 struct stat stb;
113 int nfd, fd2;
114 int d, nd;
115 struct supertype *st = NULL;
116 char *subarray = NULL;
117
118 if (md_get_array_info(fd, &info.array) < 0) {
119 pr_err("cannot get array info for %s\n", devname);
120 return 1;
121 }
122
123 if (info.array.level != -1) {
124 pr_err("can only add devices to linear arrays\n");
125 return 1;
126 }
127
128 st = super_by_fd(fd, &subarray);
129 if (!st) {
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info.array.major_version);
132 return 1;
133 }
134
135 if (subarray) {
136 pr_err("Cannot grow linear sub-arrays yet\n");
137 free(subarray);
138 free(st);
139 return 1;
140 }
141
142 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
143 if (nfd < 0) {
144 pr_err("cannot open %s\n", newdev);
145 free(st);
146 return 1;
147 }
148 fstat(nfd, &stb);
149 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
150 pr_err("%s is not a block device!\n", newdev);
151 close(nfd);
152 free(st);
153 return 1;
154 }
155 /* now check out all the devices and make sure we can read the
156 * superblock */
157 for (d=0 ; d < info.array.raid_disks ; d++) {
158 mdu_disk_info_t disk;
159 char *dv;
160
161 st->ss->free_super(st);
162
163 disk.number = d;
164 if (md_get_disk_info(fd, &disk) < 0) {
165 pr_err("cannot get device detail for device %d\n",
166 d);
167 close(nfd);
168 free(st);
169 return 1;
170 }
171 dv = map_dev(disk.major, disk.minor, 1);
172 if (!dv) {
173 pr_err("cannot find device file for device %d\n",
174 d);
175 close(nfd);
176 free(st);
177 return 1;
178 }
179 fd2 = dev_open(dv, O_RDWR);
180 if (fd2 < 0) {
181 pr_err("cannot open device file %s\n", dv);
182 close(nfd);
183 free(st);
184 return 1;
185 }
186
187 if (st->ss->load_super(st, fd2, NULL)) {
188 pr_err("cannot find super block on %s\n", dv);
189 close(nfd);
190 close(fd2);
191 free(st);
192 return 1;
193 }
194 close(fd2);
195 }
196 /* Ok, looks good. Lets update the superblock and write it out to
197 * newdev.
198 */
199
200 info.disk.number = d;
201 info.disk.major = major(stb.st_rdev);
202 info.disk.minor = minor(stb.st_rdev);
203 info.disk.raid_disk = d;
204 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
205 st->ss->update_super(st, &info, "linear-grow-new", newdev,
206 0, 0, NULL);
207
208 if (st->ss->store_super(st, nfd)) {
209 pr_err("Cannot store new superblock on %s\n",
210 newdev);
211 close(nfd);
212 return 1;
213 }
214 close(nfd);
215
216 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
217 pr_err("Cannot add new disk to this array\n");
218 return 1;
219 }
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
222 */
223
224 if (md_get_array_info(fd, &info.array) < 0) {
225 pr_err("cannot get array info for %s\n", devname);
226 return 1;
227 }
228
229 nd = d;
230 for (d=0 ; d < info.array.raid_disks ; d++) {
231 mdu_disk_info_t disk;
232 char *dv;
233
234 disk.number = d;
235 if (md_get_disk_info(fd, &disk) < 0) {
236 pr_err("cannot get device detail for device %d\n",
237 d);
238 return 1;
239 }
240 dv = map_dev(disk.major, disk.minor, 1);
241 if (!dv) {
242 pr_err("cannot find device file for device %d\n",
243 d);
244 return 1;
245 }
246 fd2 = dev_open(dv, O_RDWR);
247 if (fd2 < 0) {
248 pr_err("cannot open device file %s\n", dv);
249 return 1;
250 }
251 if (st->ss->load_super(st, fd2, NULL)) {
252 pr_err("cannot find super block on %s\n", dv);
253 close(fd);
254 return 1;
255 }
256 info.array.raid_disks = nd+1;
257 info.array.nr_disks = nd+1;
258 info.array.active_disks = nd+1;
259 info.array.working_disks = nd+1;
260
261 st->ss->update_super(st, &info, "linear-grow-update", dv,
262 0, 0, NULL);
263
264 if (st->ss->store_super(st, fd2)) {
265 pr_err("Cannot store new superblock on %s\n", dv);
266 close(fd2);
267 return 1;
268 }
269 close(fd2);
270 }
271
272 return 0;
273 }
274
275 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
276 {
277 /*
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
280 * Then add it
281 *
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
284 * to all devices
285 */
286 mdu_bitmap_file_t bmf;
287 mdu_array_info_t array;
288 struct supertype *st;
289 char *subarray = NULL;
290 int major = BITMAP_MAJOR_HI;
291 int vers = md_get_version(fd);
292 unsigned long long bitmapsize, array_size;
293 struct mdinfo *mdi;
294
295 if (vers < 9003) {
296 major = BITMAP_MAJOR_HOSTENDIAN;
297 pr_err("Warning - bitmaps created on this kernel are not portable\n"
298 " between different architectures. Consider upgrading the Linux kernel.\n");
299 }
300
301 /*
302 * We only ever get called if s->bitmap_file is != NULL, so this check
303 * is just here to quiet down static code checkers.
304 */
305 if (!s->bitmap_file)
306 return 1;
307
308 if (strcmp(s->bitmap_file, "clustered") == 0)
309 major = BITMAP_MAJOR_CLUSTERED;
310
311 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
312 if (errno == ENOMEM)
313 pr_err("Memory allocation failure.\n");
314 else
315 pr_err("bitmaps not supported by this kernel.\n");
316 return 1;
317 }
318 if (bmf.pathname[0]) {
319 if (strcmp(s->bitmap_file,"none") == 0) {
320 if (ioctl(fd, SET_BITMAP_FILE, -1) != 0) {
321 pr_err("failed to remove bitmap %s\n",
322 bmf.pathname);
323 return 1;
324 }
325 return 0;
326 }
327 pr_err("%s already has a bitmap (%s)\n",
328 devname, bmf.pathname);
329 return 1;
330 }
331 if (md_get_array_info(fd, &array) != 0) {
332 pr_err("cannot get array status for %s\n", devname);
333 return 1;
334 }
335 if (array.state & (1 << MD_SB_BITMAP_PRESENT)) {
336 if (strcmp(s->bitmap_file, "none")==0) {
337 array.state &= ~(1 << MD_SB_BITMAP_PRESENT);
338 if (md_set_array_info(fd, &array) != 0) {
339 if (array.state & (1 << MD_SB_CLUSTERED))
340 pr_err("failed to remove clustered bitmap.\n");
341 else
342 pr_err("failed to remove internal bitmap.\n");
343 return 1;
344 }
345 return 0;
346 }
347 pr_err("bitmap already present on %s\n", devname);
348 return 1;
349 }
350
351 if (strcmp(s->bitmap_file, "none") == 0) {
352 pr_err("no bitmap found on %s\n", devname);
353 return 1;
354 }
355 if (array.level <= 0) {
356 pr_err("Bitmaps not meaningful with level %s\n",
357 map_num(pers, array.level)?:"of this array");
358 return 1;
359 }
360 bitmapsize = array.size;
361 bitmapsize <<= 1;
362 if (get_dev_size(fd, NULL, &array_size) &&
363 array_size > (0x7fffffffULL << 9)) {
364 /* Array is big enough that we cannot trust array.size
365 * try other approaches
366 */
367 bitmapsize = get_component_size(fd);
368 }
369 if (bitmapsize == 0) {
370 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
371 return 1;
372 }
373
374 if (array.level == 10) {
375 int ncopies;
376
377 ncopies = (array.layout & 255) * ((array.layout >> 8) & 255);
378 bitmapsize = bitmapsize * array.raid_disks / ncopies;
379 }
380
381 st = super_by_fd(fd, &subarray);
382 if (!st) {
383 pr_err("Cannot understand version %d.%d\n",
384 array.major_version, array.minor_version);
385 return 1;
386 }
387 if (subarray) {
388 pr_err("Cannot add bitmaps to sub-arrays yet\n");
389 free(subarray);
390 free(st);
391 return 1;
392 }
393
394 mdi = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY);
395 if (mdi) {
396 if (mdi->consistency_policy == CONSISTENCY_POLICY_PPL) {
397 pr_err("Cannot add bitmap to array with PPL\n");
398 free(mdi);
399 free(st);
400 return 1;
401 }
402 free(mdi);
403 }
404
405 if (strcmp(s->bitmap_file, "internal") == 0 ||
406 strcmp(s->bitmap_file, "clustered") == 0) {
407 int rv;
408 int d;
409 int offset_setable = 0;
410 if (st->ss->add_internal_bitmap == NULL) {
411 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
412 return 1;
413 }
414 st->nodes = c->nodes;
415 st->cluster_name = c->homecluster;
416 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
417 if (mdi)
418 offset_setable = 1;
419 for (d = 0; d < st->max_devs; d++) {
420 mdu_disk_info_t disk;
421 char *dv;
422 int fd2;
423
424 disk.number = d;
425 if (md_get_disk_info(fd, &disk) < 0)
426 continue;
427 if (disk.major == 0 && disk.minor == 0)
428 continue;
429 if ((disk.state & (1 << MD_DISK_SYNC)) == 0)
430 continue;
431 dv = map_dev(disk.major, disk.minor, 1);
432 if (!dv)
433 continue;
434 fd2 = dev_open(dv, O_RDWR);
435 if (fd2 < 0)
436 continue;
437 rv = st->ss->load_super(st, fd2, NULL);
438 if (!rv) {
439 rv = st->ss->add_internal_bitmap(
440 st, &s->bitmap_chunk, c->delay,
441 s->write_behind, bitmapsize,
442 offset_setable, major);
443 if (!rv) {
444 st->ss->write_bitmap(st, fd2,
445 NodeNumUpdate);
446 } else {
447 pr_err("failed to create internal bitmap - chunksize problem.\n");
448 }
449 } else {
450 pr_err("failed to load super-block.\n");
451 }
452 close(fd2);
453 if (rv)
454 return 1;
455 }
456 if (offset_setable) {
457 st->ss->getinfo_super(st, mdi, NULL);
458 sysfs_init(mdi, fd, NULL);
459 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
460 mdi->bitmap_offset);
461 free(mdi);
462 } else {
463 if (strcmp(s->bitmap_file, "clustered") == 0)
464 array.state |= (1 << MD_SB_CLUSTERED);
465 array.state |= (1 << MD_SB_BITMAP_PRESENT);
466 rv = md_set_array_info(fd, &array);
467 }
468 if (rv < 0) {
469 if (errno == EBUSY)
470 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
471 pr_err("failed to set internal bitmap.\n");
472 return 1;
473 }
474 } else {
475 int uuid[4];
476 int bitmap_fd;
477 int d;
478 int max_devs = st->max_devs;
479
480 /* try to load a superblock */
481 for (d = 0; d < max_devs; d++) {
482 mdu_disk_info_t disk;
483 char *dv;
484 int fd2;
485 disk.number = d;
486 if (md_get_disk_info(fd, &disk) < 0)
487 continue;
488 if ((disk.major==0 && disk.minor == 0) ||
489 (disk.state & (1 << MD_DISK_REMOVED)))
490 continue;
491 dv = map_dev(disk.major, disk.minor, 1);
492 if (!dv)
493 continue;
494 fd2 = dev_open(dv, O_RDONLY);
495 if (fd2 >= 0) {
496 if (st->ss->load_super(st, fd2, NULL) == 0) {
497 close(fd2);
498 st->ss->uuid_from_super(st, uuid);
499 break;
500 }
501 close(fd2);
502 }
503 }
504 if (d == max_devs) {
505 pr_err("cannot find UUID for array!\n");
506 return 1;
507 }
508 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid,
509 s->bitmap_chunk, c->delay, s->write_behind,
510 bitmapsize, major)) {
511 return 1;
512 }
513 bitmap_fd = open(s->bitmap_file, O_RDWR);
514 if (bitmap_fd < 0) {
515 pr_err("weird: %s cannot be opened\n", s->bitmap_file);
516 return 1;
517 }
518 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
519 int err = errno;
520 if (errno == EBUSY)
521 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
522 pr_err("Cannot set bitmap file for %s: %s\n",
523 devname, strerror(err));
524 return 1;
525 }
526 }
527
528 return 0;
529 }
530
531 int Grow_consistency_policy(char *devname, int fd, struct context *c, struct shape *s)
532 {
533 struct supertype *st;
534 struct mdinfo *sra;
535 struct mdinfo *sd;
536 char *subarray = NULL;
537 int ret = 0;
538 char container_dev[PATH_MAX];
539
540 if (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
541 s->consistency_policy != CONSISTENCY_POLICY_PPL) {
542 pr_err("Operation not supported for consistency policy %s\n",
543 map_num(consistency_policies, s->consistency_policy));
544 return 1;
545 }
546
547 st = super_by_fd(fd, &subarray);
548 if (!st)
549 return 1;
550
551 sra = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY|GET_LEVEL|
552 GET_DEVS|GET_STATE);
553 if (!sra) {
554 ret = 1;
555 goto free_st;
556 }
557
558 if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
559 !st->ss->write_init_ppl) {
560 pr_err("%s metadata does not support PPL\n", st->ss->name);
561 ret = 1;
562 goto free_info;
563 }
564
565 if (sra->array.level != 5) {
566 pr_err("Operation not supported for array level %d\n",
567 sra->array.level);
568 ret = 1;
569 goto free_info;
570 }
571
572 if (sra->consistency_policy == (unsigned)s->consistency_policy) {
573 pr_err("Consistency policy is already %s\n",
574 map_num(consistency_policies, s->consistency_policy));
575 ret = 1;
576 goto free_info;
577 } else if (sra->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
578 sra->consistency_policy != CONSISTENCY_POLICY_PPL) {
579 pr_err("Current consistency policy is %s, cannot change to %s\n",
580 map_num(consistency_policies, sra->consistency_policy),
581 map_num(consistency_policies, s->consistency_policy));
582 ret = 1;
583 goto free_info;
584 }
585
586 if (subarray) {
587 char *update;
588
589 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
590 update = "ppl";
591 else
592 update = "no-ppl";
593
594 sprintf(container_dev, "/dev/%s", st->container_devnm);
595
596 ret = Update_subarray(container_dev, subarray, update, NULL,
597 c->verbose);
598 if (ret)
599 goto free_info;
600 }
601
602 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
603 struct mdinfo info;
604
605 if (subarray) {
606 struct mdinfo *mdi;
607 int cfd;
608
609 cfd = open(container_dev, O_RDWR|O_EXCL);
610 if (cfd < 0) {
611 pr_err("Failed to open %s\n", container_dev);
612 ret = 1;
613 goto free_info;
614 }
615
616 ret = st->ss->load_container(st, cfd, st->container_devnm);
617 close(cfd);
618
619 if (ret) {
620 pr_err("Cannot read superblock for %s\n",
621 container_dev);
622 goto free_info;
623 }
624
625 mdi = st->ss->container_content(st, subarray);
626 info = *mdi;
627 free(mdi);
628 }
629
630 for (sd = sra->devs; sd; sd = sd->next) {
631 int dfd;
632 char *devpath;
633
634 if ((sd->disk.state & (1 << MD_DISK_SYNC)) == 0)
635 continue;
636
637 devpath = map_dev(sd->disk.major, sd->disk.minor, 0);
638 dfd = dev_open(devpath, O_RDWR);
639 if (dfd < 0) {
640 pr_err("Failed to open %s\n", devpath);
641 ret = 1;
642 goto free_info;
643 }
644
645 if (!subarray) {
646 ret = st->ss->load_super(st, dfd, NULL);
647 if (ret) {
648 pr_err("Failed to load super-block.\n");
649 close(dfd);
650 goto free_info;
651 }
652
653 ret = st->ss->update_super(st, sra, "ppl", devname,
654 c->verbose, 0, NULL);
655 if (ret) {
656 close(dfd);
657 st->ss->free_super(st);
658 goto free_info;
659 }
660 st->ss->getinfo_super(st, &info, NULL);
661 }
662
663 ret |= sysfs_set_num(sra, sd, "ppl_sector", info.ppl_sector);
664 ret |= sysfs_set_num(sra, sd, "ppl_size", info.ppl_size);
665
666 if (ret) {
667 pr_err("Failed to set PPL attributes for %s\n",
668 sd->sys_name);
669 close(dfd);
670 st->ss->free_super(st);
671 goto free_info;
672 }
673
674 ret = st->ss->write_init_ppl(st, &info, dfd);
675 if (ret)
676 pr_err("Failed to write PPL\n");
677
678 close(dfd);
679
680 if (!subarray)
681 st->ss->free_super(st);
682
683 if (ret)
684 goto free_info;
685 }
686 }
687
688 ret = sysfs_set_str(sra, NULL, "consistency_policy",
689 map_num(consistency_policies,
690 s->consistency_policy));
691 if (ret)
692 pr_err("Failed to change array consistency policy\n");
693
694 free_info:
695 sysfs_free(sra);
696 free_st:
697 free(st);
698 free(subarray);
699
700 return ret;
701 }
702
703 /*
704 * When reshaping an array we might need to backup some data.
705 * This is written to all spares with a 'super_block' describing it.
706 * The superblock goes 4K from the end of the used space on the
707 * device.
708 * It if written after the backup is complete.
709 * It has the following structure.
710 */
711
712 static struct mdp_backup_super {
713 char magic[16]; /* md_backup_data-1 or -2 */
714 __u8 set_uuid[16];
715 __u64 mtime;
716 /* start/sizes in 512byte sectors */
717 __u64 devstart; /* address on backup device/file of data */
718 __u64 arraystart;
719 __u64 length;
720 __u32 sb_csum; /* csum of preceeding bytes. */
721 __u32 pad1;
722 __u64 devstart2; /* offset in to data of second section */
723 __u64 arraystart2;
724 __u64 length2;
725 __u32 sb_csum2; /* csum of preceeding bytes. */
726 __u8 pad[512-68-32];
727 } __attribute__((aligned(512))) bsb, bsb2;
728
729 static __u32 bsb_csum(char *buf, int len)
730 {
731 int i;
732 int csum = 0;
733 for (i = 0; i < len; i++)
734 csum = (csum<<3) + buf[0];
735 return __cpu_to_le32(csum);
736 }
737
738 static int check_idle(struct supertype *st)
739 {
740 /* Check that all member arrays for this container, or the
741 * container of this array, are idle
742 */
743 char *container = (st->container_devnm[0]
744 ? st->container_devnm : st->devnm);
745 struct mdstat_ent *ent, *e;
746 int is_idle = 1;
747
748 ent = mdstat_read(0, 0);
749 for (e = ent ; e; e = e->next) {
750 if (!is_container_member(e, container))
751 continue;
752 if (e->percent >= 0) {
753 is_idle = 0;
754 break;
755 }
756 }
757 free_mdstat(ent);
758 return is_idle;
759 }
760
761 static int freeze_container(struct supertype *st)
762 {
763 char *container = (st->container_devnm[0]
764 ? st->container_devnm : st->devnm);
765
766 if (!check_idle(st))
767 return -1;
768
769 if (block_monitor(container, 1)) {
770 pr_err("failed to freeze container\n");
771 return -2;
772 }
773
774 return 1;
775 }
776
777 static void unfreeze_container(struct supertype *st)
778 {
779 char *container = (st->container_devnm[0]
780 ? st->container_devnm : st->devnm);
781
782 unblock_monitor(container, 1);
783 }
784
785 static int freeze(struct supertype *st)
786 {
787 /* Try to freeze resync/rebuild on this array/container.
788 * Return -1 if the array is busy,
789 * return -2 container cannot be frozen,
790 * return 0 if this kernel doesn't support 'frozen'
791 * return 1 if it worked.
792 */
793 if (st->ss->external)
794 return freeze_container(st);
795 else {
796 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
797 int err;
798 char buf[20];
799
800 if (!sra)
801 return -1;
802 /* Need to clear any 'read-auto' status */
803 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
804 strncmp(buf, "read-auto", 9) == 0)
805 sysfs_set_str(sra, NULL, "array_state", "clean");
806
807 err = sysfs_freeze_array(sra);
808 sysfs_free(sra);
809 return err;
810 }
811 }
812
813 static void unfreeze(struct supertype *st)
814 {
815 if (st->ss->external)
816 return unfreeze_container(st);
817 else {
818 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
819 char buf[20];
820
821 if (sra &&
822 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
823 && strcmp(buf, "frozen\n") == 0)
824 sysfs_set_str(sra, NULL, "sync_action", "idle");
825 sysfs_free(sra);
826 }
827 }
828
829 static void wait_reshape(struct mdinfo *sra)
830 {
831 int fd = sysfs_get_fd(sra, NULL, "sync_action");
832 char action[20];
833
834 if (fd < 0)
835 return;
836
837 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
838 strncmp(action, "reshape", 7) == 0)
839 sysfs_wait(fd, NULL);
840 close(fd);
841 }
842
843 static int reshape_super(struct supertype *st, unsigned long long size,
844 int level, int layout, int chunksize, int raid_disks,
845 int delta_disks, char *backup_file, char *dev,
846 int direction, int verbose)
847 {
848 /* nothing extra to check in the native case */
849 if (!st->ss->external)
850 return 0;
851 if (!st->ss->reshape_super ||
852 !st->ss->manage_reshape) {
853 pr_err("%s metadata does not support reshape\n",
854 st->ss->name);
855 return 1;
856 }
857
858 return st->ss->reshape_super(st, size, level, layout, chunksize,
859 raid_disks, delta_disks, backup_file, dev,
860 direction, verbose);
861 }
862
863 static void sync_metadata(struct supertype *st)
864 {
865 if (st->ss->external) {
866 if (st->update_tail) {
867 flush_metadata_updates(st);
868 st->update_tail = &st->updates;
869 } else
870 st->ss->sync_metadata(st);
871 }
872 }
873
874 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
875 {
876 /* when dealing with external metadata subarrays we need to be
877 * prepared to handle EAGAIN. The kernel may need to wait for
878 * mdmon to mark the array active so the kernel can handle
879 * allocations/writeback when preparing the reshape action
880 * (md_allow_write()). We temporarily disable safe_mode_delay
881 * to close a race with the array_state going clean before the
882 * next write to raid_disks / stripe_cache_size
883 */
884 char safe[50];
885 int rc;
886
887 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
888 if (!container ||
889 (strcmp(name, "raid_disks") != 0 &&
890 strcmp(name, "stripe_cache_size") != 0))
891 return sysfs_set_num(sra, NULL, name, n);
892
893 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
894 if (rc <= 0)
895 return -1;
896 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
897 rc = sysfs_set_num(sra, NULL, name, n);
898 if (rc < 0 && errno == EAGAIN) {
899 ping_monitor(container);
900 /* if we get EAGAIN here then the monitor is not active
901 * so stop trying
902 */
903 rc = sysfs_set_num(sra, NULL, name, n);
904 }
905 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
906 return rc;
907 }
908
909 int start_reshape(struct mdinfo *sra, int already_running,
910 int before_data_disks, int data_disks)
911 {
912 int err;
913 unsigned long long sync_max_to_set;
914
915 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
916 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
917 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
918 sra->reshape_progress);
919 if (before_data_disks <= data_disks)
920 sync_max_to_set = sra->reshape_progress / data_disks;
921 else
922 sync_max_to_set = (sra->component_size * data_disks
923 - sra->reshape_progress) / data_disks;
924 if (!already_running)
925 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
926 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
927 if (!already_running && err == 0) {
928 int cnt = 5;
929 do {
930 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
931 if (err)
932 sleep(1);
933 } while (err && errno == EBUSY && cnt-- > 0);
934 }
935 return err;
936 }
937
938 void abort_reshape(struct mdinfo *sra)
939 {
940 sysfs_set_str(sra, NULL, "sync_action", "idle");
941 /*
942 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
943 * suspend_hi to decrease as well as increase.")
944 * you could only increase suspend_{lo,hi} unless the region they
945 * covered was empty. So to reset to 0, you need to push suspend_lo
946 * up past suspend_hi first. So to maximize the chance of mdadm
947 * working on all kernels, we want to keep doing that.
948 */
949 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
950 sysfs_set_num(sra, NULL, "suspend_hi", 0);
951 sysfs_set_num(sra, NULL, "suspend_lo", 0);
952 sysfs_set_num(sra, NULL, "sync_min", 0);
953 // It isn't safe to reset sync_max as we aren't monitoring.
954 // Array really should be stopped at this point.
955 }
956
957 int remove_disks_for_takeover(struct supertype *st,
958 struct mdinfo *sra,
959 int layout)
960 {
961 int nr_of_copies;
962 struct mdinfo *remaining;
963 int slot;
964
965 if (st->ss->external) {
966 int rv = 0;
967 struct mdinfo *arrays = st->ss->container_content(st, NULL);
968 /*
969 * containter_content returns list of arrays in container
970 * If arrays->next is not NULL it means that there are
971 * 2 arrays in container and operation should be blocked
972 */
973 if (arrays) {
974 if (arrays->next)
975 rv = 1;
976 sysfs_free(arrays);
977 if (rv) {
978 pr_err("Error. Cannot perform operation on /dev/%s\n", st->devnm);
979 pr_err("For this operation it MUST be single array in container\n");
980 return rv;
981 }
982 }
983 }
984
985 if (sra->array.level == 10)
986 nr_of_copies = layout & 0xff;
987 else if (sra->array.level == 1)
988 nr_of_copies = sra->array.raid_disks;
989 else
990 return 1;
991
992 remaining = sra->devs;
993 sra->devs = NULL;
994 /* for each 'copy', select one device and remove from the list. */
995 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
996 struct mdinfo **diskp;
997 int found = 0;
998
999 /* Find a working device to keep */
1000 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
1001 struct mdinfo *disk = *diskp;
1002
1003 if (disk->disk.raid_disk < slot)
1004 continue;
1005 if (disk->disk.raid_disk >= slot + nr_of_copies)
1006 continue;
1007 if (disk->disk.state & (1<<MD_DISK_REMOVED))
1008 continue;
1009 if (disk->disk.state & (1<<MD_DISK_FAULTY))
1010 continue;
1011 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
1012 continue;
1013
1014 /* We have found a good disk to use! */
1015 *diskp = disk->next;
1016 disk->next = sra->devs;
1017 sra->devs = disk;
1018 found = 1;
1019 break;
1020 }
1021 if (!found)
1022 break;
1023 }
1024
1025 if (slot < sra->array.raid_disks) {
1026 /* didn't find all slots */
1027 struct mdinfo **e;
1028 e = &remaining;
1029 while (*e)
1030 e = &(*e)->next;
1031 *e = sra->devs;
1032 sra->devs = remaining;
1033 return 1;
1034 }
1035
1036 /* Remove all 'remaining' devices from the array */
1037 while (remaining) {
1038 struct mdinfo *sd = remaining;
1039 remaining = sd->next;
1040
1041 sysfs_set_str(sra, sd, "state", "faulty");
1042 sysfs_set_str(sra, sd, "slot", "none");
1043 /* for external metadata disks should be removed in mdmon */
1044 if (!st->ss->external)
1045 sysfs_set_str(sra, sd, "state", "remove");
1046 sd->disk.state |= (1<<MD_DISK_REMOVED);
1047 sd->disk.state &= ~(1<<MD_DISK_SYNC);
1048 sd->next = sra->devs;
1049 sra->devs = sd;
1050 }
1051 return 0;
1052 }
1053
1054 void reshape_free_fdlist(int *fdlist,
1055 unsigned long long *offsets,
1056 int size)
1057 {
1058 int i;
1059
1060 for (i = 0; i < size; i++)
1061 if (fdlist[i] >= 0)
1062 close(fdlist[i]);
1063
1064 free(fdlist);
1065 free(offsets);
1066 }
1067
1068 int reshape_prepare_fdlist(char *devname,
1069 struct mdinfo *sra,
1070 int raid_disks,
1071 int nrdisks,
1072 unsigned long blocks,
1073 char *backup_file,
1074 int *fdlist,
1075 unsigned long long *offsets)
1076 {
1077 int d = 0;
1078 struct mdinfo *sd;
1079
1080 enable_fds(nrdisks);
1081 for (d = 0; d <= nrdisks; d++)
1082 fdlist[d] = -1;
1083 d = raid_disks;
1084 for (sd = sra->devs; sd; sd = sd->next) {
1085 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1086 continue;
1087 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
1088 sd->disk.raid_disk < raid_disks) {
1089 char *dn = map_dev(sd->disk.major,
1090 sd->disk.minor, 1);
1091 fdlist[sd->disk.raid_disk]
1092 = dev_open(dn, O_RDONLY);
1093 offsets[sd->disk.raid_disk] = sd->data_offset*512;
1094 if (fdlist[sd->disk.raid_disk] < 0) {
1095 pr_err("%s: cannot open component %s\n",
1096 devname, dn ? dn : "-unknown-");
1097 d = -1;
1098 goto release;
1099 }
1100 } else if (backup_file == NULL) {
1101 /* spare */
1102 char *dn = map_dev(sd->disk.major,
1103 sd->disk.minor, 1);
1104 fdlist[d] = dev_open(dn, O_RDWR);
1105 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
1106 if (fdlist[d] < 0) {
1107 pr_err("%s: cannot open component %s\n",
1108 devname, dn ? dn : "-unknown-");
1109 d = -1;
1110 goto release;
1111 }
1112 d++;
1113 }
1114 }
1115 release:
1116 return d;
1117 }
1118
1119 int reshape_open_backup_file(char *backup_file,
1120 int fd,
1121 char *devname,
1122 long blocks,
1123 int *fdlist,
1124 unsigned long long *offsets,
1125 char *sys_name,
1126 int restart)
1127 {
1128 /* Return 1 on success, 0 on any form of failure */
1129 /* need to check backup file is large enough */
1130 char buf[512];
1131 struct stat stb;
1132 unsigned int dev;
1133 int i;
1134
1135 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
1136 S_IRUSR | S_IWUSR);
1137 *offsets = 8 * 512;
1138 if (*fdlist < 0) {
1139 pr_err("%s: cannot create backup file %s: %s\n",
1140 devname, backup_file, strerror(errno));
1141 return 0;
1142 }
1143 /* Guard against backup file being on array device.
1144 * If array is partitioned or if LVM etc is in the
1145 * way this will not notice, but it is better than
1146 * nothing.
1147 */
1148 fstat(*fdlist, &stb);
1149 dev = stb.st_dev;
1150 fstat(fd, &stb);
1151 if (stb.st_rdev == dev) {
1152 pr_err("backup file must NOT be on the array being reshaped.\n");
1153 close(*fdlist);
1154 return 0;
1155 }
1156
1157 memset(buf, 0, 512);
1158 for (i=0; i < blocks + 8 ; i++) {
1159 if (write(*fdlist, buf, 512) != 512) {
1160 pr_err("%s: cannot create backup file %s: %s\n",
1161 devname, backup_file, strerror(errno));
1162 return 0;
1163 }
1164 }
1165 if (fsync(*fdlist) != 0) {
1166 pr_err("%s: cannot create backup file %s: %s\n",
1167 devname, backup_file, strerror(errno));
1168 return 0;
1169 }
1170
1171 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
1172 char *bu = make_backup(sys_name);
1173 if (symlink(backup_file, bu))
1174 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
1175 strerror(errno));
1176 free(bu);
1177 }
1178
1179 return 1;
1180 }
1181
1182 unsigned long compute_backup_blocks(int nchunk, int ochunk,
1183 unsigned int ndata, unsigned int odata)
1184 {
1185 unsigned long a, b, blocks;
1186 /* So how much do we need to backup.
1187 * We need an amount of data which is both a whole number of
1188 * old stripes and a whole number of new stripes.
1189 * So LCM for (chunksize*datadisks).
1190 */
1191 a = (ochunk/512) * odata;
1192 b = (nchunk/512) * ndata;
1193 /* Find GCD */
1194 a = GCD(a, b);
1195 /* LCM == product / GCD */
1196 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
1197
1198 return blocks;
1199 }
1200
1201 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
1202 {
1203 /* Based on the current array state in info->array and
1204 * the changes in info->new_* etc, determine:
1205 * - whether the change is possible
1206 * - Intermediate level/raid_disks/layout
1207 * - whether a restriping reshape is needed
1208 * - number of sectors in minimum change unit. This
1209 * will cover a whole number of stripes in 'before' and
1210 * 'after'.
1211 *
1212 * Return message if the change should be rejected
1213 * NULL if the change can be achieved
1214 *
1215 * This can be called as part of starting a reshape, or
1216 * when assembling an array that is undergoing reshape.
1217 */
1218 int near, far, offset, copies;
1219 int new_disks;
1220 int old_chunk, new_chunk;
1221 /* delta_parity records change in number of devices
1222 * caused by level change
1223 */
1224 int delta_parity = 0;
1225
1226 memset(re, 0, sizeof(*re));
1227
1228 /* If a new level not explicitly given, we assume no-change */
1229 if (info->new_level == UnSet)
1230 info->new_level = info->array.level;
1231
1232 if (info->new_chunk)
1233 switch (info->new_level) {
1234 case 0:
1235 case 4:
1236 case 5:
1237 case 6:
1238 case 10:
1239 /* chunk size is meaningful, must divide component_size
1240 * evenly
1241 */
1242 if (info->component_size % (info->new_chunk/512)) {
1243 unsigned long long shrink = info->component_size;
1244 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1245 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1246 info->new_chunk/1024, info->component_size/2);
1247 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1248 devname, shrink/2);
1249 pr_err("will shrink the array so the given chunk size would work.\n");
1250 return "";
1251 }
1252 break;
1253 default:
1254 return "chunk size not meaningful for this level";
1255 }
1256 else
1257 info->new_chunk = info->array.chunk_size;
1258
1259 switch (info->array.level) {
1260 default:
1261 return "No reshape is possibly for this RAID level";
1262 case LEVEL_LINEAR:
1263 if (info->delta_disks != UnSet)
1264 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1265 else
1266 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1267 case 1:
1268 /* RAID1 can convert to RAID1 with different disks, or
1269 * raid5 with 2 disks, or
1270 * raid0 with 1 disk
1271 */
1272 if (info->new_level > 1 && (info->component_size & 7))
1273 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1274 if (info->new_level == 0) {
1275 if (info->delta_disks != UnSet &&
1276 info->delta_disks != 0)
1277 return "Cannot change number of disks with RAID1->RAID0 conversion";
1278 re->level = 0;
1279 re->before.data_disks = 1;
1280 re->after.data_disks = 1;
1281 return NULL;
1282 }
1283 if (info->new_level == 1) {
1284 if (info->delta_disks == UnSet)
1285 /* Don't know what to do */
1286 return "no change requested for Growing RAID1";
1287 re->level = 1;
1288 return NULL;
1289 }
1290 if (info->array.raid_disks != 2 && info->new_level == 5)
1291 return "Can only convert a 2-device array to RAID5";
1292 if (info->array.raid_disks == 2 && info->new_level == 5) {
1293 re->level = 5;
1294 re->before.data_disks = 1;
1295 if (info->delta_disks != UnSet &&
1296 info->delta_disks != 0)
1297 re->after.data_disks = 1 + info->delta_disks;
1298 else
1299 re->after.data_disks = 1;
1300 if (re->after.data_disks < 1)
1301 return "Number of disks too small for RAID5";
1302
1303 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1304 info->array.chunk_size = 65536;
1305 break;
1306 }
1307 /* Could do some multi-stage conversions, but leave that to
1308 * later.
1309 */
1310 return "Impossibly level change request for RAID1";
1311
1312 case 10:
1313 /* RAID10 can be converted from near mode to
1314 * RAID0 by removing some devices.
1315 * It can also be reshaped if the kernel supports
1316 * new_data_offset.
1317 */
1318 switch (info->new_level) {
1319 case 0:
1320 if ((info->array.layout & ~0xff) != 0x100)
1321 return "Cannot Grow RAID10 with far/offset layout";
1322 /* number of devices must be multiple of number of copies */
1323 if (info->array.raid_disks % (info->array.layout & 0xff))
1324 return "RAID10 layout too complex for Grow operation";
1325
1326 new_disks = (info->array.raid_disks
1327 / (info->array.layout & 0xff));
1328 if (info->delta_disks == UnSet)
1329 info->delta_disks = (new_disks
1330 - info->array.raid_disks);
1331
1332 if (info->delta_disks != new_disks - info->array.raid_disks)
1333 return "New number of raid-devices impossible for RAID10";
1334 if (info->new_chunk &&
1335 info->new_chunk != info->array.chunk_size)
1336 return "Cannot change chunk-size with RAID10 Grow";
1337
1338 /* looks good */
1339 re->level = 0;
1340 re->before.data_disks = new_disks;
1341 re->after.data_disks = re->before.data_disks;
1342 return NULL;
1343
1344 case 10:
1345 near = info->array.layout & 0xff;
1346 far = (info->array.layout >> 8) & 0xff;
1347 offset = info->array.layout & 0x10000;
1348 if (far > 1 && !offset)
1349 return "Cannot reshape RAID10 in far-mode";
1350 copies = near * far;
1351
1352 old_chunk = info->array.chunk_size * far;
1353
1354 if (info->new_layout == UnSet)
1355 info->new_layout = info->array.layout;
1356 else {
1357 near = info->new_layout & 0xff;
1358 far = (info->new_layout >> 8) & 0xff;
1359 offset = info->new_layout & 0x10000;
1360 if (far > 1 && !offset)
1361 return "Cannot reshape RAID10 to far-mode";
1362 if (near * far != copies)
1363 return "Cannot change number of copies when reshaping RAID10";
1364 }
1365 if (info->delta_disks == UnSet)
1366 info->delta_disks = 0;
1367 new_disks = (info->array.raid_disks +
1368 info->delta_disks);
1369
1370 new_chunk = info->new_chunk * far;
1371
1372 re->level = 10;
1373 re->before.layout = info->array.layout;
1374 re->before.data_disks = info->array.raid_disks;
1375 re->after.layout = info->new_layout;
1376 re->after.data_disks = new_disks;
1377 /* For RAID10 we don't do backup but do allow reshape,
1378 * so set backup_blocks to INVALID_SECTORS rather than
1379 * zero.
1380 * And there is no need to synchronise stripes on both
1381 * 'old' and 'new'. So the important
1382 * number is the minimum data_offset difference
1383 * which is the larger of (offset copies * chunk).
1384 */
1385 re->backup_blocks = INVALID_SECTORS;
1386 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1387 if (new_disks < re->before.data_disks &&
1388 info->space_after < re->min_offset_change)
1389 /* Reduce component size by one chunk */
1390 re->new_size = (info->component_size -
1391 re->min_offset_change);
1392 else
1393 re->new_size = info->component_size;
1394 re->new_size = re->new_size * new_disks / copies;
1395 return NULL;
1396
1397 default:
1398 return "RAID10 can only be changed to RAID0";
1399 }
1400 case 0:
1401 /* RAID0 can be converted to RAID10, or to RAID456 */
1402 if (info->new_level == 10) {
1403 if (info->new_layout == UnSet &&
1404 info->delta_disks == UnSet) {
1405 /* Assume near=2 layout */
1406 info->new_layout = 0x102;
1407 info->delta_disks = info->array.raid_disks;
1408 }
1409 if (info->new_layout == UnSet) {
1410 int copies = 1 + (info->delta_disks
1411 / info->array.raid_disks);
1412 if (info->array.raid_disks * (copies-1)
1413 != info->delta_disks)
1414 return "Impossible number of devices for RAID0->RAID10";
1415 info->new_layout = 0x100 + copies;
1416 }
1417 if (info->delta_disks == UnSet) {
1418 int copies = info->new_layout & 0xff;
1419 if (info->new_layout != 0x100 + copies)
1420 return "New layout impossible for RAID0->RAID10";;
1421 info->delta_disks = (copies - 1) *
1422 info->array.raid_disks;
1423 }
1424 if (info->new_chunk &&
1425 info->new_chunk != info->array.chunk_size)
1426 return "Cannot change chunk-size with RAID0->RAID10";
1427 /* looks good */
1428 re->level = 10;
1429 re->before.data_disks = (info->array.raid_disks +
1430 info->delta_disks);
1431 re->after.data_disks = re->before.data_disks;
1432 re->before.layout = info->new_layout;
1433 return NULL;
1434 }
1435
1436 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1437 * a raid4 style layout of the final level.
1438 */
1439 switch (info->new_level) {
1440 case 4:
1441 delta_parity = 1;
1442 case 0:
1443 re->level = 4;
1444 re->before.layout = 0;
1445 break;
1446 case 5:
1447 delta_parity = 1;
1448 re->level = 5;
1449 re->before.layout = ALGORITHM_PARITY_N;
1450 if (info->new_layout == UnSet)
1451 info->new_layout = map_name(r5layout, "default");
1452 break;
1453 case 6:
1454 delta_parity = 2;
1455 re->level = 6;
1456 re->before.layout = ALGORITHM_PARITY_N;
1457 if (info->new_layout == UnSet)
1458 info->new_layout = map_name(r6layout, "default");
1459 break;
1460 default:
1461 return "Impossible level change requested";
1462 }
1463 re->before.data_disks = info->array.raid_disks;
1464 /* determining 'after' layout happens outside this 'switch' */
1465 break;
1466
1467 case 4:
1468 info->array.layout = ALGORITHM_PARITY_N;
1469 case 5:
1470 switch (info->new_level) {
1471 case 0:
1472 delta_parity = -1;
1473 case 4:
1474 re->level = info->array.level;
1475 re->before.data_disks = info->array.raid_disks - 1;
1476 re->before.layout = info->array.layout;
1477 break;
1478 case 5:
1479 re->level = 5;
1480 re->before.data_disks = info->array.raid_disks - 1;
1481 re->before.layout = info->array.layout;
1482 break;
1483 case 6:
1484 delta_parity = 1;
1485 re->level = 6;
1486 re->before.data_disks = info->array.raid_disks - 1;
1487 switch (info->array.layout) {
1488 case ALGORITHM_LEFT_ASYMMETRIC:
1489 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1490 break;
1491 case ALGORITHM_RIGHT_ASYMMETRIC:
1492 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1493 break;
1494 case ALGORITHM_LEFT_SYMMETRIC:
1495 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1496 break;
1497 case ALGORITHM_RIGHT_SYMMETRIC:
1498 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1499 break;
1500 case ALGORITHM_PARITY_0:
1501 re->before.layout = ALGORITHM_PARITY_0_6;
1502 break;
1503 case ALGORITHM_PARITY_N:
1504 re->before.layout = ALGORITHM_PARITY_N_6;
1505 break;
1506 default:
1507 return "Cannot convert an array with this layout";
1508 }
1509 break;
1510 case 1:
1511 if (info->array.raid_disks != 2)
1512 return "Can only convert a 2-device array to RAID1";
1513 if (info->delta_disks != UnSet &&
1514 info->delta_disks != 0)
1515 return "Cannot set raid_disk when converting RAID5->RAID1";
1516 re->level = 1;
1517 info->new_chunk = 0;
1518 return NULL;
1519 default:
1520 return "Impossible level change requested";
1521 }
1522 break;
1523 case 6:
1524 switch (info->new_level) {
1525 case 4:
1526 case 5:
1527 delta_parity = -1;
1528 case 6:
1529 re->level = 6;
1530 re->before.data_disks = info->array.raid_disks - 2;
1531 re->before.layout = info->array.layout;
1532 break;
1533 default:
1534 return "Impossible level change requested";
1535 }
1536 break;
1537 }
1538
1539 /* If we reached here then it looks like a re-stripe is
1540 * happening. We have determined the intermediate level
1541 * and initial raid_disks/layout and stored these in 're'.
1542 *
1543 * We need to deduce the final layout that can be atomically
1544 * converted to the end state.
1545 */
1546 switch (info->new_level) {
1547 case 0:
1548 /* We can only get to RAID0 from RAID4 or RAID5
1549 * with appropriate layout and one extra device
1550 */
1551 if (re->level != 4 && re->level != 5)
1552 return "Cannot covert to RAID0 from this level";
1553
1554 switch (re->level) {
1555 case 4:
1556 re->before.layout = 0;
1557 re->after.layout = 0;
1558 break;
1559 case 5:
1560 re->after.layout = ALGORITHM_PARITY_N;
1561 break;
1562 }
1563 break;
1564
1565 case 4:
1566 /* We can only get to RAID4 from RAID5 */
1567 if (re->level != 4 && re->level != 5)
1568 return "Cannot convert to RAID4 from this level";
1569
1570 switch (re->level) {
1571 case 4:
1572 re->after.layout = 0;
1573 break;
1574 case 5:
1575 re->after.layout = ALGORITHM_PARITY_N;
1576 break;
1577 }
1578 break;
1579
1580 case 5:
1581 /* We get to RAID5 from RAID5 or RAID6 */
1582 if (re->level != 5 && re->level != 6)
1583 return "Cannot convert to RAID5 from this level";
1584
1585 switch (re->level) {
1586 case 5:
1587 if (info->new_layout == UnSet)
1588 re->after.layout = re->before.layout;
1589 else
1590 re->after.layout = info->new_layout;
1591 break;
1592 case 6:
1593 if (info->new_layout == UnSet)
1594 info->new_layout = re->before.layout;
1595
1596 /* after.layout needs to be raid6 version of new_layout */
1597 if (info->new_layout == ALGORITHM_PARITY_N)
1598 re->after.layout = ALGORITHM_PARITY_N;
1599 else {
1600 char layout[40];
1601 char *ls = map_num(r5layout, info->new_layout);
1602 int l;
1603 if (ls) {
1604 /* Current RAID6 layout has a RAID5
1605 * equivalent - good
1606 */
1607 strcat(strcpy(layout, ls), "-6");
1608 l = map_name(r6layout, layout);
1609 if (l == UnSet)
1610 return "Cannot find RAID6 layout to convert to";
1611 } else {
1612 /* Current RAID6 has no equivalent.
1613 * If it is already a '-6' layout we
1614 * can leave it unchanged, else we must
1615 * fail
1616 */
1617 ls = map_num(r6layout, info->new_layout);
1618 if (!ls ||
1619 strcmp(ls+strlen(ls)-2, "-6") != 0)
1620 return "Please specify new layout";
1621 l = info->new_layout;
1622 }
1623 re->after.layout = l;
1624 }
1625 }
1626 break;
1627
1628 case 6:
1629 /* We must already be at level 6 */
1630 if (re->level != 6)
1631 return "Impossible level change";
1632 if (info->new_layout == UnSet)
1633 re->after.layout = info->array.layout;
1634 else
1635 re->after.layout = info->new_layout;
1636 break;
1637 default:
1638 return "Impossible level change requested";
1639 }
1640 if (info->delta_disks == UnSet)
1641 info->delta_disks = delta_parity;
1642
1643 re->after.data_disks =
1644 (re->before.data_disks + info->delta_disks - delta_parity);
1645
1646 switch (re->level) {
1647 case 6:
1648 re->parity = 2;
1649 break;
1650 case 4:
1651 case 5:
1652 re->parity = 1;
1653 break;
1654 default:
1655 re->parity = 0;
1656 break;
1657 }
1658 /* So we have a restripe operation, we need to calculate the number
1659 * of blocks per reshape operation.
1660 */
1661 re->new_size = info->component_size * re->before.data_disks;
1662 if (info->new_chunk == 0)
1663 info->new_chunk = info->array.chunk_size;
1664 if (re->after.data_disks == re->before.data_disks &&
1665 re->after.layout == re->before.layout &&
1666 info->new_chunk == info->array.chunk_size) {
1667 /* Nothing to change, can change level immediately. */
1668 re->level = info->new_level;
1669 re->backup_blocks = 0;
1670 return NULL;
1671 }
1672 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1673 /* chunk and layout changes make no difference */
1674 re->level = info->new_level;
1675 re->backup_blocks = 0;
1676 return NULL;
1677 }
1678
1679 if (re->after.data_disks == re->before.data_disks &&
1680 get_linux_version() < 2006032)
1681 return "in-place reshape is not safe before 2.6.32 - sorry.";
1682
1683 if (re->after.data_disks < re->before.data_disks &&
1684 get_linux_version() < 2006030)
1685 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1686
1687 re->backup_blocks = compute_backup_blocks(
1688 info->new_chunk, info->array.chunk_size,
1689 re->after.data_disks,
1690 re->before.data_disks);
1691 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1692
1693 re->new_size = info->component_size * re->after.data_disks;
1694 return NULL;
1695 }
1696
1697 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1698 char *text_version)
1699 {
1700 struct mdinfo *info;
1701 char *subarray;
1702 int ret_val = -1;
1703
1704 if ((st == NULL) || (sra == NULL))
1705 return ret_val;
1706
1707 if (text_version == NULL)
1708 text_version = sra->text_version;
1709 subarray = strchr(text_version + 1, '/')+1;
1710 info = st->ss->container_content(st, subarray);
1711 if (info) {
1712 unsigned long long current_size = 0;
1713 unsigned long long new_size =
1714 info->custom_array_size/2;
1715
1716 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1717 new_size > current_size) {
1718 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1719 < 0)
1720 dprintf("Error: Cannot set array size");
1721 else {
1722 ret_val = 0;
1723 dprintf("Array size changed");
1724 }
1725 dprintf_cont(" from %llu to %llu.\n",
1726 current_size, new_size);
1727 }
1728 sysfs_free(info);
1729 } else
1730 dprintf("Error: set_array_size(): info pointer in NULL\n");
1731
1732 return ret_val;
1733 }
1734
1735 static int reshape_array(char *container, int fd, char *devname,
1736 struct supertype *st, struct mdinfo *info,
1737 int force, struct mddev_dev *devlist,
1738 unsigned long long data_offset,
1739 char *backup_file, int verbose, int forked,
1740 int restart, int freeze_reshape);
1741 static int reshape_container(char *container, char *devname,
1742 int mdfd,
1743 struct supertype *st,
1744 struct mdinfo *info,
1745 int force,
1746 char *backup_file, int verbose,
1747 int forked, int restart, int freeze_reshape);
1748
1749 int Grow_reshape(char *devname, int fd,
1750 struct mddev_dev *devlist,
1751 unsigned long long data_offset,
1752 struct context *c, struct shape *s)
1753 {
1754 /* Make some changes in the shape of an array.
1755 * The kernel must support the change.
1756 *
1757 * There are three different changes. Each can trigger
1758 * a resync or recovery so we freeze that until we have
1759 * requested everything (if kernel supports freezing - 2.6.30).
1760 * The steps are:
1761 * - change size (i.e. component_size)
1762 * - change level
1763 * - change layout/chunksize/ndisks
1764 *
1765 * The last can require a reshape. It is different on different
1766 * levels so we need to check the level before actioning it.
1767 * Some times the level change needs to be requested after the
1768 * reshape (e.g. raid6->raid5, raid5->raid0)
1769 *
1770 */
1771 struct mdu_array_info_s array;
1772 int rv = 0;
1773 struct supertype *st;
1774 char *subarray = NULL;
1775
1776 int frozen;
1777 int changed = 0;
1778 char *container = NULL;
1779 int cfd = -1;
1780
1781 struct mddev_dev *dv;
1782 int added_disks;
1783
1784 struct mdinfo info;
1785 struct mdinfo *sra;
1786
1787 if (md_get_array_info(fd, &array) < 0) {
1788 pr_err("%s is not an active md array - aborting\n",
1789 devname);
1790 return 1;
1791 }
1792 if (data_offset != INVALID_SECTORS && array.level != 10 &&
1793 (array.level < 4 || array.level > 6)) {
1794 pr_err("--grow --data-offset not yet supported\n");
1795 return 1;
1796 }
1797
1798 if (s->size > 0 &&
1799 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1800 pr_err("cannot change component size at the same time as other changes.\n"
1801 " Change size first, then check data is intact before making other changes.\n");
1802 return 1;
1803 }
1804
1805 if (s->raiddisks && s->raiddisks < array.raid_disks &&
1806 array.level > 1 && get_linux_version() < 2006032 &&
1807 !check_env("MDADM_FORCE_FEWER")) {
1808 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1809 " Please use a newer kernel\n");
1810 return 1;
1811 }
1812
1813 st = super_by_fd(fd, &subarray);
1814 if (!st) {
1815 pr_err("Unable to determine metadata format for %s\n", devname);
1816 return 1;
1817 }
1818 if (s->raiddisks > st->max_devs) {
1819 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1820 return 1;
1821 }
1822 if (s->level == 0 &&
1823 (array.state & (1<<MD_SB_BITMAP_PRESENT)) &&
1824 !(array.state & (1<<MD_SB_CLUSTERED))) {
1825 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
1826 if (md_set_array_info(fd, &array)!= 0) {
1827 pr_err("failed to remove internal bitmap.\n");
1828 return 1;
1829 }
1830 }
1831
1832 /* in the external case we need to check that the requested reshape is
1833 * supported, and perform an initial check that the container holds the
1834 * pre-requisite spare devices (mdmon owns final validation)
1835 */
1836 if (st->ss->external) {
1837 int rv;
1838
1839 if (subarray) {
1840 container = st->container_devnm;
1841 cfd = open_dev_excl(st->container_devnm);
1842 } else {
1843 container = st->devnm;
1844 close(fd);
1845 cfd = open_dev_excl(st->devnm);
1846 fd = cfd;
1847 }
1848 if (cfd < 0) {
1849 pr_err("Unable to open container for %s\n",
1850 devname);
1851 free(subarray);
1852 return 1;
1853 }
1854
1855 rv = st->ss->load_container(st, cfd, NULL);
1856
1857 if (rv) {
1858 pr_err("Cannot read superblock for %s\n",
1859 devname);
1860 free(subarray);
1861 return 1;
1862 }
1863
1864 /* check if operation is supported for metadata handler */
1865 if (st->ss->container_content) {
1866 struct mdinfo *cc = NULL;
1867 struct mdinfo *content = NULL;
1868
1869 cc = st->ss->container_content(st, subarray);
1870 for (content = cc; content ; content = content->next) {
1871 int allow_reshape = 1;
1872
1873 /* check if reshape is allowed based on metadata
1874 * indications stored in content.array.status
1875 */
1876 if (content->array.state &
1877 (1 << MD_SB_BLOCK_VOLUME))
1878 allow_reshape = 0;
1879 if (content->array.state &
1880 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE))
1881 allow_reshape = 0;
1882 if (!allow_reshape) {
1883 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1884 devname, container);
1885 sysfs_free(cc);
1886 free(subarray);
1887 return 1;
1888 }
1889 }
1890 sysfs_free(cc);
1891 }
1892 if (mdmon_running(container))
1893 st->update_tail = &st->updates;
1894 }
1895
1896 added_disks = 0;
1897 for (dv = devlist; dv; dv = dv->next)
1898 added_disks++;
1899 if (s->raiddisks > array.raid_disks &&
1900 array.spare_disks + added_disks < (s->raiddisks - array.raid_disks) &&
1901 !c->force) {
1902 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1903 " Use --force to over-ride this check.\n",
1904 s->raiddisks - array.raid_disks,
1905 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1906 array.spare_disks + added_disks);
1907 return 1;
1908 }
1909
1910 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS |
1911 GET_STATE | GET_VERSION);
1912 if (sra) {
1913 if (st->ss->external && subarray == NULL) {
1914 array.level = LEVEL_CONTAINER;
1915 sra->array.level = LEVEL_CONTAINER;
1916 }
1917 } else {
1918 pr_err("failed to read sysfs parameters for %s\n",
1919 devname);
1920 return 1;
1921 }
1922 frozen = freeze(st);
1923 if (frozen < -1) {
1924 /* freeze() already spewed the reason */
1925 sysfs_free(sra);
1926 return 1;
1927 } else if (frozen < 0) {
1928 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1929 sysfs_free(sra);
1930 return 1;
1931 }
1932
1933 /* ========= set size =============== */
1934 if (s->size > 0 &&
1935 (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1936 unsigned long long orig_size = get_component_size(fd)/2;
1937 unsigned long long min_csize;
1938 struct mdinfo *mdi;
1939 int raid0_takeover = 0;
1940
1941 if (orig_size == 0)
1942 orig_size = (unsigned) array.size;
1943
1944 if (orig_size == 0) {
1945 pr_err("Cannot set device size in this type of array.\n");
1946 rv = 1;
1947 goto release;
1948 }
1949
1950 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1951 devname, APPLY_METADATA_CHANGES,
1952 c->verbose > 0)) {
1953 rv = 1;
1954 goto release;
1955 }
1956 sync_metadata(st);
1957 if (st->ss->external) {
1958 /* metadata can have size limitation
1959 * update size value according to metadata information
1960 */
1961 struct mdinfo *sizeinfo =
1962 st->ss->container_content(st, subarray);
1963 if (sizeinfo) {
1964 unsigned long long new_size =
1965 sizeinfo->custom_array_size/2;
1966 int data_disks = get_data_disks(
1967 sizeinfo->array.level,
1968 sizeinfo->array.layout,
1969 sizeinfo->array.raid_disks);
1970 new_size /= data_disks;
1971 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1972 orig_size, new_size,
1973 new_size * data_disks);
1974 s->size = new_size;
1975 sysfs_free(sizeinfo);
1976 }
1977 }
1978
1979 /* Update the size of each member device in case
1980 * they have been resized. This will never reduce
1981 * below the current used-size. The "size" attribute
1982 * understands '0' to mean 'max'.
1983 */
1984 min_csize = 0;
1985 rv = 0;
1986 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1987 if (sysfs_set_num(sra, mdi, "size",
1988 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1989 /* Probably kernel refusing to let us
1990 * reduce the size - not an error.
1991 */
1992 break;
1993 }
1994 if (array.not_persistent == 0 &&
1995 array.major_version == 0 &&
1996 get_linux_version() < 3001000) {
1997 /* Dangerous to allow size to exceed 2TB */
1998 unsigned long long csize;
1999 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
2000 if (csize >= 2ULL*1024*1024*1024)
2001 csize = 2ULL*1024*1024*1024;
2002 if ((min_csize == 0 || (min_csize
2003 > csize)))
2004 min_csize = csize;
2005 }
2006 }
2007 }
2008 if (rv) {
2009 pr_err("Cannot set size on array members.\n");
2010 goto size_change_error;
2011 }
2012 if (min_csize && s->size > min_csize) {
2013 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2014 rv = 1;
2015 goto size_change_error;
2016 }
2017 if (min_csize && s->size == MAX_SIZE) {
2018 /* Don't let the kernel choose a size - it will get
2019 * it wrong
2020 */
2021 pr_err("Limited v0.90 array to 2TB per device\n");
2022 s->size = min_csize;
2023 }
2024 if (st->ss->external) {
2025 if (sra->array.level == 0) {
2026 rv = sysfs_set_str(sra, NULL, "level",
2027 "raid5");
2028 if (!rv) {
2029 raid0_takeover = 1;
2030 /* get array parameters after takeover
2031 * to change one parameter at time only
2032 */
2033 rv = md_get_array_info(fd, &array);
2034 }
2035 }
2036 /* make sure mdmon is
2037 * aware of the new level */
2038 if (!mdmon_running(st->container_devnm))
2039 start_mdmon(st->container_devnm);
2040 ping_monitor(container);
2041 if (mdmon_running(st->container_devnm) &&
2042 st->update_tail == NULL)
2043 st->update_tail = &st->updates;
2044 }
2045
2046 if (s->size == MAX_SIZE)
2047 s->size = 0;
2048 array.size = s->size;
2049 if (s->size & ~INT32_MAX) {
2050 /* got truncated to 32bit, write to
2051 * component_size instead
2052 */
2053 if (sra)
2054 rv = sysfs_set_num(sra, NULL,
2055 "component_size", s->size);
2056 else
2057 rv = -1;
2058 } else {
2059 rv = md_set_array_info(fd, &array);
2060
2061 /* manage array size when it is managed externally
2062 */
2063 if ((rv == 0) && st->ss->external)
2064 rv = set_array_size(st, sra, sra->text_version);
2065 }
2066
2067 if (raid0_takeover) {
2068 /* do not recync non-existing parity,
2069 * we will drop it anyway
2070 */
2071 sysfs_set_str(sra, NULL, "sync_action", "frozen");
2072 /* go back to raid0, drop parity disk
2073 */
2074 sysfs_set_str(sra, NULL, "level", "raid0");
2075 md_get_array_info(fd, &array);
2076 }
2077
2078 size_change_error:
2079 if (rv != 0) {
2080 int err = errno;
2081
2082 /* restore metadata */
2083 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
2084 UnSet, NULL, devname,
2085 ROLLBACK_METADATA_CHANGES,
2086 c->verbose) == 0)
2087 sync_metadata(st);
2088 pr_err("Cannot set device size for %s: %s\n",
2089 devname, strerror(err));
2090 if (err == EBUSY &&
2091 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2092 cont_err("Bitmap must be removed before size can be changed\n");
2093 rv = 1;
2094 goto release;
2095 }
2096 if (s->assume_clean) {
2097 /* This will fail on kernels older than 3.0 unless
2098 * a backport has been arranged.
2099 */
2100 if (sra == NULL ||
2101 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
2102 pr_err("--assume-clean not supported with --grow on this kernel\n");
2103 }
2104 md_get_array_info(fd, &array);
2105 s->size = get_component_size(fd)/2;
2106 if (s->size == 0)
2107 s->size = array.size;
2108 if (c->verbose >= 0) {
2109 if (s->size == orig_size)
2110 pr_err("component size of %s unchanged at %lluK\n",
2111 devname, s->size);
2112 else
2113 pr_err("component size of %s has been set to %lluK\n",
2114 devname, s->size);
2115 }
2116 changed = 1;
2117 } else if (array.level != LEVEL_CONTAINER) {
2118 s->size = get_component_size(fd)/2;
2119 if (s->size == 0)
2120 s->size = array.size;
2121 }
2122
2123 /* See if there is anything else to do */
2124 if ((s->level == UnSet || s->level == array.level) &&
2125 (s->layout_str == NULL) &&
2126 (s->chunk == 0 || s->chunk == array.chunk_size) &&
2127 data_offset == INVALID_SECTORS &&
2128 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
2129 /* Nothing more to do */
2130 if (!changed && c->verbose >= 0)
2131 pr_err("%s: no change requested\n",
2132 devname);
2133 goto release;
2134 }
2135
2136 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2137 * current implementation assumes that following conditions must be met:
2138 * - RAID10:
2139 * - far_copies == 1
2140 * - near_copies == 2
2141 */
2142 if ((s->level == 0 && array.level == 10 && sra &&
2143 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
2144 (s->level == 0 && array.level == 1 && sra)) {
2145 int err;
2146 err = remove_disks_for_takeover(st, sra, array.layout);
2147 if (err) {
2148 dprintf("Array cannot be reshaped\n");
2149 if (cfd > -1)
2150 close(cfd);
2151 rv = 1;
2152 goto release;
2153 }
2154 /* Make sure mdmon has seen the device removal
2155 * and updated metadata before we continue with
2156 * level change
2157 */
2158 if (container)
2159 ping_monitor(container);
2160 }
2161
2162 memset(&info, 0, sizeof(info));
2163 info.array = array;
2164 sysfs_init(&info, fd, NULL);
2165 strcpy(info.text_version, sra->text_version);
2166 info.component_size = s->size*2;
2167 info.new_level = s->level;
2168 info.new_chunk = s->chunk * 1024;
2169 if (info.array.level == LEVEL_CONTAINER) {
2170 info.delta_disks = UnSet;
2171 info.array.raid_disks = s->raiddisks;
2172 } else if (s->raiddisks)
2173 info.delta_disks = s->raiddisks - info.array.raid_disks;
2174 else
2175 info.delta_disks = UnSet;
2176 if (s->layout_str == NULL) {
2177 info.new_layout = UnSet;
2178 if (info.array.level == 6 &&
2179 (info.new_level == 6 || info.new_level == UnSet) &&
2180 info.array.layout >= 16) {
2181 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
2182 cont_err("during the reshape, please specify --layout=preserve\n");
2183 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2184 rv = 1;
2185 goto release;
2186 }
2187 } else if (strcmp(s->layout_str, "normalise") == 0 ||
2188 strcmp(s->layout_str, "normalize") == 0) {
2189 /* If we have a -6 RAID6 layout, remove the '-6'. */
2190 info.new_layout = UnSet;
2191 if (info.array.level == 6 && info.new_level == UnSet) {
2192 char l[40], *h;
2193 strcpy(l, map_num(r6layout, info.array.layout));
2194 h = strrchr(l, '-');
2195 if (h && strcmp(h, "-6") == 0) {
2196 *h = 0;
2197 info.new_layout = map_name(r6layout, l);
2198 }
2199 } else {
2200 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
2201 rv = 1;
2202 goto release;
2203 }
2204 } else if (strcmp(s->layout_str, "preserve") == 0) {
2205 /* This means that a non-standard RAID6 layout
2206 * is OK.
2207 * In particular:
2208 * - When reshape a RAID6 (e.g. adding a device)
2209 * which is in a non-standard layout, it is OK
2210 * to preserve that layout.
2211 * - When converting a RAID5 to RAID6, leave it in
2212 * the XXX-6 layout, don't re-layout.
2213 */
2214 if (info.array.level == 6 && info.new_level == UnSet)
2215 info.new_layout = info.array.layout;
2216 else if (info.array.level == 5 && info.new_level == 6) {
2217 char l[40];
2218 strcpy(l, map_num(r5layout, info.array.layout));
2219 strcat(l, "-6");
2220 info.new_layout = map_name(r6layout, l);
2221 } else {
2222 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
2223 rv = 1;
2224 goto release;
2225 }
2226 } else {
2227 int l = info.new_level;
2228 if (l == UnSet)
2229 l = info.array.level;
2230 switch (l) {
2231 case 5:
2232 info.new_layout = map_name(r5layout, s->layout_str);
2233 break;
2234 case 6:
2235 info.new_layout = map_name(r6layout, s->layout_str);
2236 break;
2237 case 10:
2238 info.new_layout = parse_layout_10(s->layout_str);
2239 break;
2240 case LEVEL_FAULTY:
2241 info.new_layout = parse_layout_faulty(s->layout_str);
2242 break;
2243 default:
2244 pr_err("layout not meaningful with this level\n");
2245 rv = 1;
2246 goto release;
2247 }
2248 if (info.new_layout == UnSet) {
2249 pr_err("layout %s not understood for this level\n",
2250 s->layout_str);
2251 rv = 1;
2252 goto release;
2253 }
2254 }
2255
2256 if (array.level == LEVEL_FAULTY) {
2257 if (s->level != UnSet && s->level != array.level) {
2258 pr_err("cannot change level of Faulty device\n");
2259 rv =1 ;
2260 }
2261 if (s->chunk) {
2262 pr_err("cannot set chunksize of Faulty device\n");
2263 rv =1 ;
2264 }
2265 if (s->raiddisks && s->raiddisks != 1) {
2266 pr_err("cannot set raid_disks of Faulty device\n");
2267 rv =1 ;
2268 }
2269 if (s->layout_str) {
2270 if (md_get_array_info(fd, &array) != 0) {
2271 dprintf("Cannot get array information.\n");
2272 goto release;
2273 }
2274 array.layout = info.new_layout;
2275 if (md_set_array_info(fd, &array) != 0) {
2276 pr_err("failed to set new layout\n");
2277 rv = 1;
2278 } else if (c->verbose >= 0)
2279 printf("layout for %s set to %d\n",
2280 devname, array.layout);
2281 }
2282 } else if (array.level == LEVEL_CONTAINER) {
2283 /* This change is to be applied to every array in the
2284 * container. This is only needed when the metadata imposes
2285 * restraints of the various arrays in the container.
2286 * Currently we only know that IMSM requires all arrays
2287 * to have the same number of devices so changing the
2288 * number of devices (On-Line Capacity Expansion) must be
2289 * performed at the level of the container
2290 */
2291 if (fd > 0) {
2292 close(fd);
2293 fd = -1;
2294 }
2295 rv = reshape_container(container, devname, -1, st, &info,
2296 c->force, c->backup_file, c->verbose, 0, 0, 0);
2297 frozen = 0;
2298 } else {
2299 /* get spare devices from external metadata
2300 */
2301 if (st->ss->external) {
2302 struct mdinfo *info2;
2303
2304 info2 = st->ss->container_content(st, subarray);
2305 if (info2) {
2306 info.array.spare_disks =
2307 info2->array.spare_disks;
2308 sysfs_free(info2);
2309 }
2310 }
2311
2312 /* Impose these changes on a single array. First
2313 * check that the metadata is OK with the change. */
2314
2315 if (reshape_super(st, 0, info.new_level,
2316 info.new_layout, info.new_chunk,
2317 info.array.raid_disks, info.delta_disks,
2318 c->backup_file, devname, APPLY_METADATA_CHANGES,
2319 c->verbose)) {
2320 rv = 1;
2321 goto release;
2322 }
2323 sync_metadata(st);
2324 rv = reshape_array(container, fd, devname, st, &info, c->force,
2325 devlist, data_offset, c->backup_file, c->verbose,
2326 0, 0, 0);
2327 frozen = 0;
2328 }
2329 release:
2330 sysfs_free(sra);
2331 if (frozen > 0)
2332 unfreeze(st);
2333 return rv;
2334 }
2335
2336 /* verify_reshape_position()
2337 * Function checks if reshape position in metadata is not farther
2338 * than position in md.
2339 * Return value:
2340 * 0 : not valid sysfs entry
2341 * it can be caused by not started reshape, it should be started
2342 * by reshape array or raid0 array is before takeover
2343 * -1 : error, reshape position is obviously wrong
2344 * 1 : success, reshape progress correct or updated
2345 */
2346 static int verify_reshape_position(struct mdinfo *info, int level)
2347 {
2348 int ret_val = 0;
2349 char buf[40];
2350 int rv;
2351
2352 /* read sync_max, failure can mean raid0 array */
2353 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2354
2355 if (rv > 0) {
2356 char *ep;
2357 unsigned long long position = strtoull(buf, &ep, 0);
2358
2359 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2360 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2361 position *= get_data_disks(level,
2362 info->new_layout,
2363 info->array.raid_disks);
2364 if (info->reshape_progress < position) {
2365 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2366 info->reshape_progress, position);
2367 info->reshape_progress = position;
2368 ret_val = 1;
2369 } else if (info->reshape_progress > position) {
2370 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2371 position, info->reshape_progress);
2372 ret_val = -1;
2373 } else {
2374 dprintf("Reshape position in md and metadata are the same;");
2375 ret_val = 1;
2376 }
2377 }
2378 } else if (rv == 0) {
2379 /* for valid sysfs entry, 0-length content
2380 * should be indicated as error
2381 */
2382 ret_val = -1;
2383 }
2384
2385 return ret_val;
2386 }
2387
2388 static unsigned long long choose_offset(unsigned long long lo,
2389 unsigned long long hi,
2390 unsigned long long min,
2391 unsigned long long max)
2392 {
2393 /* Choose a new offset between hi and lo.
2394 * It must be between min and max, but
2395 * we would prefer something near the middle of hi/lo, and also
2396 * prefer to be aligned to a big power of 2.
2397 *
2398 * So we start with the middle, then for each bit,
2399 * starting at '1' and increasing, if it is set, we either
2400 * add it or subtract it if possible, preferring the option
2401 * which is furthest from the boundary.
2402 *
2403 * We stop once we get a 1MB alignment. As units are in sectors,
2404 * 1MB = 2*1024 sectors.
2405 */
2406 unsigned long long choice = (lo + hi) / 2;
2407 unsigned long long bit = 1;
2408
2409 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2410 unsigned long long bigger, smaller;
2411 if (! (bit & choice))
2412 continue;
2413 bigger = choice + bit;
2414 smaller = choice - bit;
2415 if (bigger > max && smaller < min)
2416 break;
2417 if (bigger > max)
2418 choice = smaller;
2419 else if (smaller < min)
2420 choice = bigger;
2421 else if (hi - bigger > smaller - lo)
2422 choice = bigger;
2423 else
2424 choice = smaller;
2425 }
2426 return choice;
2427 }
2428
2429 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2430 char *devname, int delta_disks,
2431 unsigned long long data_offset,
2432 unsigned long long min,
2433 int can_fallback)
2434 {
2435 struct mdinfo *sd;
2436 int dir = 0;
2437 int err = 0;
2438 unsigned long long before, after;
2439
2440 /* Need to find min space before and after so same is used
2441 * on all devices
2442 */
2443 before = UINT64_MAX;
2444 after = UINT64_MAX;
2445 for (sd = sra->devs; sd; sd = sd->next) {
2446 char *dn;
2447 int dfd;
2448 int rv;
2449 struct supertype *st2;
2450 struct mdinfo info2;
2451
2452 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2453 continue;
2454 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2455 dfd = dev_open(dn, O_RDONLY);
2456 if (dfd < 0) {
2457 pr_err("%s: cannot open component %s\n",
2458 devname, dn ? dn : "-unknown-");
2459 goto release;
2460 }
2461 st2 = dup_super(st);
2462 rv = st2->ss->load_super(st2,dfd, NULL);
2463 close(dfd);
2464 if (rv) {
2465 free(st2);
2466 pr_err("%s: cannot get superblock from %s\n",
2467 devname, dn);
2468 goto release;
2469 }
2470 st2->ss->getinfo_super(st2, &info2, NULL);
2471 st2->ss->free_super(st2);
2472 free(st2);
2473 if (info2.space_before == 0 &&
2474 info2.space_after == 0) {
2475 /* Metadata doesn't support data_offset changes */
2476 if (!can_fallback)
2477 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2478 devname);
2479 goto fallback;
2480 }
2481 if (before > info2.space_before)
2482 before = info2.space_before;
2483 if (after > info2.space_after)
2484 after = info2.space_after;
2485
2486 if (data_offset != INVALID_SECTORS) {
2487 if (dir == 0) {
2488 if (info2.data_offset == data_offset) {
2489 pr_err("%s: already has that data_offset\n",
2490 dn);
2491 goto release;
2492 }
2493 if (data_offset < info2.data_offset)
2494 dir = -1;
2495 else
2496 dir = 1;
2497 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2498 (data_offset >= info2.data_offset && dir == -1)) {
2499 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2500 dn);
2501 goto release;
2502 }
2503 }
2504 }
2505 if (before == UINT64_MAX)
2506 /* impossible really, there must be no devices */
2507 return 1;
2508
2509 for (sd = sra->devs; sd; sd = sd->next) {
2510 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2511 unsigned long long new_data_offset;
2512
2513 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2514 continue;
2515 if (delta_disks < 0) {
2516 /* Don't need any space as array is shrinking
2517 * just move data_offset up by min
2518 */
2519 if (data_offset == INVALID_SECTORS)
2520 new_data_offset = sd->data_offset + min;
2521 else {
2522 if (data_offset < sd->data_offset + min) {
2523 pr_err("--data-offset too small for %s\n",
2524 dn);
2525 goto release;
2526 }
2527 new_data_offset = data_offset;
2528 }
2529 } else if (delta_disks > 0) {
2530 /* need space before */
2531 if (before < min) {
2532 if (can_fallback)
2533 goto fallback;
2534 pr_err("Insufficient head-space for reshape on %s\n",
2535 dn);
2536 goto release;
2537 }
2538 if (data_offset == INVALID_SECTORS)
2539 new_data_offset = sd->data_offset - min;
2540 else {
2541 if (data_offset > sd->data_offset - min) {
2542 pr_err("--data-offset too large for %s\n",
2543 dn);
2544 goto release;
2545 }
2546 new_data_offset = data_offset;
2547 }
2548 } else {
2549 if (dir == 0) {
2550 /* can move up or down. If 'data_offset'
2551 * was set we would have already decided,
2552 * so just choose direction with most space.
2553 */
2554 if (before > after)
2555 dir = -1;
2556 else
2557 dir = 1;
2558 }
2559 sysfs_set_str(sra, NULL, "reshape_direction",
2560 dir == 1 ? "backwards" : "forwards");
2561 if (dir > 0) {
2562 /* Increase data offset */
2563 if (after < min) {
2564 if (can_fallback)
2565 goto fallback;
2566 pr_err("Insufficient tail-space for reshape on %s\n",
2567 dn);
2568 goto release;
2569 }
2570 if (data_offset != INVALID_SECTORS &&
2571 data_offset < sd->data_offset + min) {
2572 pr_err("--data-offset too small on %s\n",
2573 dn);
2574 goto release;
2575 }
2576 if (data_offset != INVALID_SECTORS)
2577 new_data_offset = data_offset;
2578 else
2579 new_data_offset = choose_offset(sd->data_offset,
2580 sd->data_offset + after,
2581 sd->data_offset + min,
2582 sd->data_offset + after);
2583 } else {
2584 /* Decrease data offset */
2585 if (before < min) {
2586 if (can_fallback)
2587 goto fallback;
2588 pr_err("insufficient head-room on %s\n",
2589 dn);
2590 goto release;
2591 }
2592 if (data_offset != INVALID_SECTORS &&
2593 data_offset < sd->data_offset - min) {
2594 pr_err("--data-offset too small on %s\n",
2595 dn);
2596 goto release;
2597 }
2598 if (data_offset != INVALID_SECTORS)
2599 new_data_offset = data_offset;
2600 else
2601 new_data_offset = choose_offset(sd->data_offset - before,
2602 sd->data_offset,
2603 sd->data_offset - before,
2604 sd->data_offset - min);
2605 }
2606 }
2607 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2608 if (err < 0 && errno == E2BIG) {
2609 /* try again after increasing data size to max */
2610 err = sysfs_set_num(sra, sd, "size", 0);
2611 if (err < 0 && errno == EINVAL &&
2612 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2613 /* some kernels have a bug where you cannot
2614 * use '0' on spare devices. */
2615 sysfs_set_num(sra, sd, "size",
2616 (sra->component_size + after)/2);
2617 }
2618 err = sysfs_set_num(sra, sd, "new_offset",
2619 new_data_offset);
2620 }
2621 if (err < 0) {
2622 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2623 pr_err("data-offset is too big for %s\n",
2624 dn);
2625 goto release;
2626 }
2627 if (sd == sra->devs &&
2628 (errno == ENOENT || errno == E2BIG))
2629 /* Early kernel, no 'new_offset' file,
2630 * or kernel doesn't like us.
2631 * For RAID5/6 this is not fatal
2632 */
2633 return 1;
2634 pr_err("Cannot set new_offset for %s\n",
2635 dn);
2636 break;
2637 }
2638 }
2639 return err;
2640 release:
2641 return -1;
2642 fallback:
2643 /* Just use a backup file */
2644 return 1;
2645 }
2646
2647 static int raid10_reshape(char *container, int fd, char *devname,
2648 struct supertype *st, struct mdinfo *info,
2649 struct reshape *reshape,
2650 unsigned long long data_offset,
2651 int force, int verbose)
2652 {
2653 /* Changing raid_disks, layout, chunksize or possibly
2654 * just data_offset for a RAID10.
2655 * We must always change data_offset. We change by at least
2656 * ->min_offset_change which is the largest of the old and new
2657 * chunk sizes.
2658 * If raid_disks is increasing, then data_offset must decrease
2659 * by at least this copy size.
2660 * If raid_disks is unchanged, data_offset must increase or
2661 * decrease by at least min_offset_change but preferably by much more.
2662 * We choose half of the available space.
2663 * If raid_disks is decreasing, data_offset must increase by
2664 * at least min_offset_change. To allow of this, component_size
2665 * must be decreased by the same amount.
2666 *
2667 * So we calculate the required minimum and direction, possibly
2668 * reduce the component_size, then iterate through the devices
2669 * and set the new_data_offset.
2670 * If that all works, we set chunk_size, layout, raid_disks, and start
2671 * 'reshape'
2672 */
2673 struct mdinfo *sra;
2674 unsigned long long min;
2675 int err = 0;
2676
2677 sra = sysfs_read(fd, NULL,
2678 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2679 );
2680 if (!sra) {
2681 pr_err("%s: Cannot get array details from sysfs\n",
2682 devname);
2683 goto release;
2684 }
2685 min = reshape->min_offset_change;
2686
2687 if (info->delta_disks)
2688 sysfs_set_str(sra, NULL, "reshape_direction",
2689 info->delta_disks < 0 ? "backwards" : "forwards");
2690 if (info->delta_disks < 0 &&
2691 info->space_after < min) {
2692 int rv = sysfs_set_num(sra, NULL, "component_size",
2693 (sra->component_size -
2694 min)/2);
2695 if (rv) {
2696 pr_err("cannot reduce component size\n");
2697 goto release;
2698 }
2699 }
2700 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2701 min, 0);
2702 if (err == 1) {
2703 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2704 cont_err("supported on this kernel\n");
2705 err = -1;
2706 }
2707 if (err < 0)
2708 goto release;
2709
2710 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2711 err = errno;
2712 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2713 err = errno;
2714 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2715 info->array.raid_disks + info->delta_disks) < 0)
2716 err = errno;
2717 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2718 err = errno;
2719 if (err) {
2720 pr_err("Cannot set array shape for %s\n",
2721 devname);
2722 if (err == EBUSY &&
2723 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2724 cont_err(" Bitmap must be removed before shape can be changed\n");
2725 goto release;
2726 }
2727 sysfs_free(sra);
2728 return 0;
2729 release:
2730 sysfs_free(sra);
2731 return 1;
2732 }
2733
2734 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2735 {
2736 struct mdinfo *sra, *sd;
2737 /* Initialisation to silence compiler warning */
2738 unsigned long long min_space_before = 0, min_space_after = 0;
2739 int first = 1;
2740
2741 sra = sysfs_read(fd, NULL, GET_DEVS);
2742 if (!sra)
2743 return;
2744 for (sd = sra->devs; sd; sd = sd->next) {
2745 char *dn;
2746 int dfd;
2747 struct supertype *st2;
2748 struct mdinfo info2;
2749
2750 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2751 continue;
2752 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2753 dfd = dev_open(dn, O_RDONLY);
2754 if (dfd < 0)
2755 break;
2756 st2 = dup_super(st);
2757 if (st2->ss->load_super(st2,dfd, NULL)) {
2758 close(dfd);
2759 free(st2);
2760 break;
2761 }
2762 close(dfd);
2763 st2->ss->getinfo_super(st2, &info2, NULL);
2764 st2->ss->free_super(st2);
2765 free(st2);
2766 if (first ||
2767 min_space_before > info2.space_before)
2768 min_space_before = info2.space_before;
2769 if (first ||
2770 min_space_after > info2.space_after)
2771 min_space_after = info2.space_after;
2772 first = 0;
2773 }
2774 if (sd == NULL && !first) {
2775 info->space_after = min_space_after;
2776 info->space_before = min_space_before;
2777 }
2778 sysfs_free(sra);
2779 }
2780
2781 static void update_cache_size(char *container, struct mdinfo *sra,
2782 struct mdinfo *info,
2783 int disks, unsigned long long blocks)
2784 {
2785 /* Check that the internal stripe cache is
2786 * large enough, or it won't work.
2787 * It must hold at least 4 stripes of the larger
2788 * chunk size
2789 */
2790 unsigned long cache;
2791 cache = max(info->array.chunk_size, info->new_chunk);
2792 cache *= 4; /* 4 stripes minimum */
2793 cache /= 512; /* convert to sectors */
2794 /* make sure there is room for 'blocks' with a bit to spare */
2795 if (cache < 16 + blocks / disks)
2796 cache = 16 + blocks / disks;
2797 cache /= (4096/512); /* Convert from sectors to pages */
2798
2799 if (sra->cache_size < cache)
2800 subarray_set_num(container, sra, "stripe_cache_size",
2801 cache+1);
2802 }
2803
2804 static int impose_reshape(struct mdinfo *sra,
2805 struct mdinfo *info,
2806 struct supertype *st,
2807 int fd,
2808 int restart,
2809 char *devname, char *container,
2810 struct reshape *reshape)
2811 {
2812 struct mdu_array_info_s array;
2813
2814 sra->new_chunk = info->new_chunk;
2815
2816 if (restart) {
2817 /* for external metadata checkpoint saved by mdmon can be lost
2818 * or missed /due to e.g. crash/. Check if md is not during
2819 * restart farther than metadata points to.
2820 * If so, this means metadata information is obsolete.
2821 */
2822 if (st->ss->external)
2823 verify_reshape_position(info, reshape->level);
2824 sra->reshape_progress = info->reshape_progress;
2825 } else {
2826 sra->reshape_progress = 0;
2827 if (reshape->after.data_disks < reshape->before.data_disks)
2828 /* start from the end of the new array */
2829 sra->reshape_progress = (sra->component_size
2830 * reshape->after.data_disks);
2831 }
2832
2833 md_get_array_info(fd, &array);
2834 if (info->array.chunk_size == info->new_chunk &&
2835 reshape->before.layout == reshape->after.layout &&
2836 st->ss->external == 0) {
2837 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2838 array.raid_disks = reshape->after.data_disks + reshape->parity;
2839 if (!restart && md_set_array_info(fd, &array) != 0) {
2840 int err = errno;
2841
2842 pr_err("Cannot set device shape for %s: %s\n",
2843 devname, strerror(errno));
2844
2845 if (err == EBUSY &&
2846 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2847 cont_err("Bitmap must be removed before shape can be changed\n");
2848
2849 goto release;
2850 }
2851 } else if (!restart) {
2852 /* set them all just in case some old 'new_*' value
2853 * persists from some earlier problem.
2854 */
2855 int err = 0;
2856 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2857 err = errno;
2858 if (!err && sysfs_set_num(sra, NULL, "layout",
2859 reshape->after.layout) < 0)
2860 err = errno;
2861 if (!err && subarray_set_num(container, sra, "raid_disks",
2862 reshape->after.data_disks +
2863 reshape->parity) < 0)
2864 err = errno;
2865 if (err) {
2866 pr_err("Cannot set device shape for %s\n",
2867 devname);
2868
2869 if (err == EBUSY &&
2870 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2871 cont_err("Bitmap must be removed before shape can be changed\n");
2872 goto release;
2873 }
2874 }
2875 return 0;
2876 release:
2877 return -1;
2878 }
2879
2880 static int impose_level(int fd, int level, char *devname, int verbose)
2881 {
2882 char *c;
2883 struct mdu_array_info_s array;
2884 struct mdinfo info;
2885 sysfs_init(&info, fd, NULL);
2886
2887 md_get_array_info(fd, &array);
2888 if (level == 0 &&
2889 (array.level >= 4 && array.level <= 6)) {
2890 /* To convert to RAID0 we need to fail and
2891 * remove any non-data devices. */
2892 int found = 0;
2893 int d;
2894 int data_disks = array.raid_disks - 1;
2895 if (array.level == 6)
2896 data_disks -= 1;
2897 if (array.level == 5 &&
2898 array.layout != ALGORITHM_PARITY_N)
2899 return -1;
2900 if (array.level == 6 &&
2901 array.layout != ALGORITHM_PARITY_N_6)
2902 return -1;
2903 sysfs_set_str(&info, NULL,"sync_action", "idle");
2904 /* First remove any spares so no recovery starts */
2905 for (d = 0, found = 0;
2906 d < MAX_DISKS && found < array.nr_disks;
2907 d++) {
2908 mdu_disk_info_t disk;
2909 disk.number = d;
2910 if (md_get_disk_info(fd, &disk) < 0)
2911 continue;
2912 if (disk.major == 0 && disk.minor == 0)
2913 continue;
2914 found++;
2915 if ((disk.state & (1 << MD_DISK_ACTIVE))
2916 && disk.raid_disk < data_disks)
2917 /* keep this */
2918 continue;
2919 ioctl(fd, HOT_REMOVE_DISK,
2920 makedev(disk.major, disk.minor));
2921 }
2922 /* Now fail anything left */
2923 md_get_array_info(fd, &array);
2924 for (d = 0, found = 0;
2925 d < MAX_DISKS && found < array.nr_disks;
2926 d++) {
2927 mdu_disk_info_t disk;
2928 disk.number = d;
2929 if (md_get_disk_info(fd, &disk) < 0)
2930 continue;
2931 if (disk.major == 0 && disk.minor == 0)
2932 continue;
2933 found++;
2934 if ((disk.state & (1 << MD_DISK_ACTIVE))
2935 && disk.raid_disk < data_disks)
2936 /* keep this */
2937 continue;
2938 ioctl(fd, SET_DISK_FAULTY,
2939 makedev(disk.major, disk.minor));
2940 hot_remove_disk(fd, makedev(disk.major, disk.minor), 1);
2941 }
2942 }
2943 c = map_num(pers, level);
2944 if (c) {
2945 int err = sysfs_set_str(&info, NULL, "level", c);
2946 if (err) {
2947 err = errno;
2948 pr_err("%s: could not set level to %s\n",
2949 devname, c);
2950 if (err == EBUSY &&
2951 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2952 cont_err("Bitmap must be removed before level can be changed\n");
2953 return err;
2954 }
2955 if (verbose >= 0)
2956 pr_err("level of %s changed to %s\n",
2957 devname, c);
2958 }
2959 return 0;
2960 }
2961
2962 int sigterm = 0;
2963 static void catch_term(int sig)
2964 {
2965 sigterm = 1;
2966 }
2967
2968 static int continue_via_systemd(char *devnm)
2969 {
2970 int skipped, i, pid, status;
2971 char pathbuf[1024];
2972 /* In a systemd/udev world, it is best to get systemd to
2973 * run "mdadm --grow --continue" rather than running in the
2974 * background.
2975 */
2976 switch(fork()) {
2977 case 0:
2978 /* FIXME yuk. CLOSE_EXEC?? */
2979 skipped = 0;
2980 for (i = 3; skipped < 20; i++)
2981 if (close(i) < 0)
2982 skipped++;
2983 else
2984 skipped = 0;
2985
2986 /* Don't want to see error messages from
2987 * systemctl. If the service doesn't exist,
2988 * we fork ourselves.
2989 */
2990 close(2);
2991 open("/dev/null", O_WRONLY);
2992 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2993 devnm);
2994 status = execl("/usr/bin/systemctl", "systemctl",
2995 "start",
2996 pathbuf, NULL);
2997 status = execl("/bin/systemctl", "systemctl", "start",
2998 pathbuf, NULL);
2999 exit(1);
3000 case -1: /* Just do it ourselves. */
3001 break;
3002 default: /* parent - good */
3003 pid = wait(&status);
3004 if (pid >= 0 && status == 0)
3005 return 1;
3006 }
3007 return 0;
3008 }
3009
3010 static int reshape_array(char *container, int fd, char *devname,
3011 struct supertype *st, struct mdinfo *info,
3012 int force, struct mddev_dev *devlist,
3013 unsigned long long data_offset,
3014 char *backup_file, int verbose, int forked,
3015 int restart, int freeze_reshape)
3016 {
3017 struct reshape reshape;
3018 int spares_needed;
3019 char *msg;
3020 int orig_level = UnSet;
3021 int odisks;
3022 int delayed;
3023
3024 struct mdu_array_info_s array;
3025 char *c;
3026
3027 struct mddev_dev *dv;
3028 int added_disks;
3029
3030 int *fdlist = NULL;
3031 unsigned long long *offsets = NULL;
3032 int d;
3033 int nrdisks;
3034 int err;
3035 unsigned long blocks;
3036 unsigned long long array_size;
3037 int done;
3038 struct mdinfo *sra = NULL;
3039 char buf[20];
3040
3041 /* when reshaping a RAID0, the component_size might be zero.
3042 * So try to fix that up.
3043 */
3044 if (md_get_array_info(fd, &array) != 0) {
3045 dprintf("Cannot get array information.\n");
3046 goto release;
3047 }
3048 if (array.level == 0 && info->component_size == 0) {
3049 get_dev_size(fd, NULL, &array_size);
3050 info->component_size = array_size / array.raid_disks;
3051 }
3052
3053 if (array.level == 10)
3054 /* Need space_after info */
3055 get_space_after(fd, st, info);
3056
3057 if (info->reshape_active) {
3058 int new_level = info->new_level;
3059 info->new_level = UnSet;
3060 if (info->delta_disks > 0)
3061 info->array.raid_disks -= info->delta_disks;
3062 msg = analyse_change(devname, info, &reshape);
3063 info->new_level = new_level;
3064 if (info->delta_disks > 0)
3065 info->array.raid_disks += info->delta_disks;
3066 if (!restart)
3067 /* Make sure the array isn't read-only */
3068 ioctl(fd, RESTART_ARRAY_RW, 0);
3069 } else
3070 msg = analyse_change(devname, info, &reshape);
3071 if (msg) {
3072 /* if msg == "", error has already been printed */
3073 if (msg[0])
3074 pr_err("%s\n", msg);
3075 goto release;
3076 }
3077 if (restart &&
3078 (reshape.level != info->array.level ||
3079 reshape.before.layout != info->array.layout ||
3080 reshape.before.data_disks + reshape.parity
3081 != info->array.raid_disks - max(0, info->delta_disks))) {
3082 pr_err("reshape info is not in native format - cannot continue.\n");
3083 goto release;
3084 }
3085
3086 if (st->ss->external && restart && (info->reshape_progress == 0) &&
3087 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
3088 (strncmp(buf, "reshape", 7) == 0))) {
3089 /* When reshape is restarted from '0', very begin of array
3090 * it is possible that for external metadata reshape and array
3091 * configuration doesn't happen.
3092 * Check if md has the same opinion, and reshape is restarted
3093 * from 0. If so, this is regular reshape start after reshape
3094 * switch in metadata to next array only.
3095 */
3096 if ((verify_reshape_position(info, reshape.level) >= 0) &&
3097 (info->reshape_progress == 0))
3098 restart = 0;
3099 }
3100 if (restart) {
3101 /* reshape already started. just skip to monitoring the reshape */
3102 if (reshape.backup_blocks == 0)
3103 return 0;
3104 if (restart & RESHAPE_NO_BACKUP)
3105 return 0;
3106
3107 /* Need 'sra' down at 'started:' */
3108 sra = sysfs_read(fd, NULL,
3109 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3110 GET_CACHE);
3111 if (!sra) {
3112 pr_err("%s: Cannot get array details from sysfs\n",
3113 devname);
3114 goto release;
3115 }
3116
3117 if (!backup_file)
3118 backup_file = locate_backup(sra->sys_name);
3119
3120 goto started;
3121 }
3122 /* The container is frozen but the array may not be.
3123 * So freeze the array so spares don't get put to the wrong use
3124 * FIXME there should probably be a cleaner separation between
3125 * freeze_array and freeze_container.
3126 */
3127 sysfs_freeze_array(info);
3128 /* Check we have enough spares to not be degraded */
3129 added_disks = 0;
3130 for (dv = devlist; dv ; dv=dv->next)
3131 added_disks++;
3132 spares_needed = max(reshape.before.data_disks,
3133 reshape.after.data_disks)
3134 + reshape.parity - array.raid_disks;
3135
3136 if (!force &&
3137 info->new_level > 1 && info->array.level > 1 &&
3138 spares_needed > info->array.spare_disks + added_disks) {
3139 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3140 " Use --force to over-ride this check.\n",
3141 spares_needed,
3142 spares_needed == 1 ? "" : "s",
3143 info->array.spare_disks + added_disks);
3144 goto release;
3145 }
3146 /* Check we have enough spares to not fail */
3147 spares_needed = max(reshape.before.data_disks,
3148 reshape.after.data_disks)
3149 - array.raid_disks;
3150 if ((info->new_level > 1 || info->new_level == 0) &&
3151 spares_needed > info->array.spare_disks +added_disks) {
3152 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3153 spares_needed,
3154 spares_needed == 1 ? "" : "s",
3155 info->array.spare_disks + added_disks);
3156 goto release;
3157 }
3158
3159 if (reshape.level != array.level) {
3160 int err = impose_level(fd, reshape.level, devname, verbose);
3161 if (err)
3162 goto release;
3163 info->new_layout = UnSet; /* after level change,
3164 * layout is meaningless */
3165 orig_level = array.level;
3166 sysfs_freeze_array(info);
3167
3168 if (reshape.level > 0 && st->ss->external) {
3169 /* make sure mdmon is aware of the new level */
3170 if (mdmon_running(container))
3171 flush_mdmon(container);
3172
3173 if (!mdmon_running(container))
3174 start_mdmon(container);
3175 ping_monitor(container);
3176 if (mdmon_running(container) &&
3177 st->update_tail == NULL)
3178 st->update_tail = &st->updates;
3179 }
3180 }
3181 /* ->reshape_super might have chosen some spares from the
3182 * container that it wants to be part of the new array.
3183 * We can collect them with ->container_content and give
3184 * them to the kernel.
3185 */
3186 if (st->ss->reshape_super && st->ss->container_content) {
3187 char *subarray = strchr(info->text_version+1, '/')+1;
3188 struct mdinfo *info2 =
3189 st->ss->container_content(st, subarray);
3190 struct mdinfo *d;
3191
3192 if (info2) {
3193 sysfs_init(info2, fd, st->devnm);
3194 /* When increasing number of devices, we need to set
3195 * new raid_disks before adding these, or they might
3196 * be rejected.
3197 */
3198 if (reshape.backup_blocks &&
3199 reshape.after.data_disks > reshape.before.data_disks)
3200 subarray_set_num(container, info2, "raid_disks",
3201 reshape.after.data_disks +
3202 reshape.parity);
3203 for (d = info2->devs; d; d = d->next) {
3204 if (d->disk.state == 0 &&
3205 d->disk.raid_disk >= 0) {
3206 /* This is a spare that wants to
3207 * be part of the array.
3208 */
3209 add_disk(fd, st, info2, d);
3210 }
3211 }
3212 sysfs_free(info2);
3213 }
3214 }
3215 /* We might have been given some devices to add to the
3216 * array. Now that the array has been changed to the right
3217 * level and frozen, we can safely add them.
3218 */
3219 if (devlist) {
3220 if (Manage_subdevs(devname, fd, devlist, verbose,
3221 0, NULL, 0))
3222 goto release;
3223 }
3224
3225 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3226 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3227 if (reshape.backup_blocks == 0) {
3228 /* No restriping needed, but we might need to impose
3229 * some more changes: layout, raid_disks, chunk_size
3230 */
3231 /* read current array info */
3232 if (md_get_array_info(fd, &array) != 0) {
3233 dprintf("Cannot get array information.\n");
3234 goto release;
3235 }
3236 /* compare current array info with new values and if
3237 * it is different update them to new */
3238 if (info->new_layout != UnSet &&
3239 info->new_layout != array.layout) {
3240 array.layout = info->new_layout;
3241 if (md_set_array_info(fd, &array) != 0) {
3242 pr_err("failed to set new layout\n");
3243 goto release;
3244 } else if (verbose >= 0)
3245 printf("layout for %s set to %d\n",
3246 devname, array.layout);
3247 }
3248 if (info->delta_disks != UnSet &&
3249 info->delta_disks != 0 &&
3250 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3251 array.raid_disks += info->delta_disks;
3252 if (md_set_array_info(fd, &array) != 0) {
3253 pr_err("failed to set raid disks\n");
3254 goto release;
3255 } else if (verbose >= 0) {
3256 printf("raid_disks for %s set to %d\n",
3257 devname, array.raid_disks);
3258 }
3259 }
3260 if (info->new_chunk != 0 &&
3261 info->new_chunk != array.chunk_size) {
3262 if (sysfs_set_num(info, NULL,
3263 "chunk_size", info->new_chunk) != 0) {
3264 pr_err("failed to set chunk size\n");
3265 goto release;
3266 } else if (verbose >= 0)
3267 printf("chunk size for %s set to %d\n",
3268 devname, array.chunk_size);
3269 }
3270 unfreeze(st);
3271 return 0;
3272 }
3273
3274 /*
3275 * There are three possibilities.
3276 * 1/ The array will shrink.
3277 * We need to ensure the reshape will pause before reaching
3278 * the 'critical section'. We also need to fork and wait for
3279 * that to happen. When it does we
3280 * suspend/backup/complete/unfreeze
3281 *
3282 * 2/ The array will not change size.
3283 * This requires that we keep a backup of a sliding window
3284 * so that we can restore data after a crash. So we need
3285 * to fork and monitor progress.
3286 * In future we will allow the data_offset to change, so
3287 * a sliding backup becomes unnecessary.
3288 *
3289 * 3/ The array will grow. This is relatively easy.
3290 * However the kernel's restripe routines will cheerfully
3291 * overwrite some early data before it is safe. So we
3292 * need to make a backup of the early parts of the array
3293 * and be ready to restore it if rebuild aborts very early.
3294 * For externally managed metadata, we still need a forked
3295 * child to monitor the reshape and suspend IO over the region
3296 * that is being reshaped.
3297 *
3298 * We backup data by writing it to one spare, or to a
3299 * file which was given on command line.
3300 *
3301 * In each case, we first make sure that storage is available
3302 * for the required backup.
3303 * Then we:
3304 * - request the shape change.
3305 * - fork to handle backup etc.
3306 */
3307 /* Check that we can hold all the data */
3308 get_dev_size(fd, NULL, &array_size);
3309 if (reshape.new_size < (array_size/512)) {
3310 pr_err("this change will reduce the size of the array.\n"
3311 " use --grow --array-size first to truncate array.\n"
3312 " e.g. mdadm --grow %s --array-size %llu\n",
3313 devname, reshape.new_size/2);
3314 goto release;
3315 }
3316
3317 if (array.level == 10) {
3318 /* Reshaping RAID10 does not require any data backup by
3319 * user-space. Instead it requires that the data_offset
3320 * is changed to avoid the need for backup.
3321 * So this is handled very separately
3322 */
3323 if (restart)
3324 /* Nothing to do. */
3325 return 0;
3326 return raid10_reshape(container, fd, devname, st, info,
3327 &reshape, data_offset,
3328 force, verbose);
3329 }
3330 sra = sysfs_read(fd, NULL,
3331 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3332 GET_CACHE);
3333 if (!sra) {
3334 pr_err("%s: Cannot get array details from sysfs\n",
3335 devname);
3336 goto release;
3337 }
3338
3339 if (!backup_file)
3340 switch(set_new_data_offset(sra, st, devname,
3341 reshape.after.data_disks - reshape.before.data_disks,
3342 data_offset,
3343 reshape.min_offset_change, 1)) {
3344 case -1:
3345 goto release;
3346 case 0:
3347 /* Updated data_offset, so it's easy now */
3348 update_cache_size(container, sra, info,
3349 min(reshape.before.data_disks,
3350 reshape.after.data_disks),
3351 reshape.backup_blocks);
3352
3353 /* Right, everything seems fine. Let's kick things off.
3354 */
3355 sync_metadata(st);
3356
3357 if (impose_reshape(sra, info, st, fd, restart,
3358 devname, container, &reshape) < 0)
3359 goto release;
3360 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3361 struct mdinfo *sd;
3362 if (errno != EINVAL) {
3363 pr_err("Failed to initiate reshape!\n");
3364 goto release;
3365 }
3366 /* revert data_offset and try the old way */
3367 for (sd = sra->devs; sd; sd = sd->next) {
3368 sysfs_set_num(sra, sd, "new_offset",
3369 sd->data_offset);
3370 sysfs_set_str(sra, NULL, "reshape_direction",
3371 "forwards");
3372 }
3373 break;
3374 }
3375 if (info->new_level == reshape.level)
3376 return 0;
3377 /* need to adjust level when reshape completes */
3378 switch(fork()) {
3379 case -1: /* ignore error, but don't wait */
3380 return 0;
3381 default: /* parent */
3382 return 0;
3383 case 0:
3384 map_fork();
3385 break;
3386 }
3387 close(fd);
3388 wait_reshape(sra);
3389 fd = open_dev(sra->sys_name);
3390 if (fd >= 0)
3391 impose_level(fd, info->new_level, devname, verbose);
3392 return 0;
3393 case 1: /* Couldn't set data_offset, try the old way */
3394 if (data_offset != INVALID_SECTORS) {
3395 pr_err("Cannot update data_offset on this array\n");
3396 goto release;
3397 }
3398 break;
3399 }
3400
3401 started:
3402 /* Decide how many blocks (sectors) for a reshape
3403 * unit. The number we have so far is just a minimum
3404 */
3405 blocks = reshape.backup_blocks;
3406 if (reshape.before.data_disks ==
3407 reshape.after.data_disks) {
3408 /* Make 'blocks' bigger for better throughput, but
3409 * not so big that we reject it below.
3410 * Try for 16 megabytes
3411 */
3412 while (blocks * 32 < sra->component_size &&
3413 blocks < 16*1024*2)
3414 blocks *= 2;
3415 } else
3416 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3417
3418 if (blocks >= sra->component_size/2) {
3419 pr_err("%s: Something wrong - reshape aborted\n",
3420 devname);
3421 goto release;
3422 }
3423
3424 /* Now we need to open all these devices so we can read/write.
3425 */
3426 nrdisks = max(reshape.before.data_disks,
3427 reshape.after.data_disks) + reshape.parity
3428 + sra->array.spare_disks;
3429 fdlist = xcalloc((1+nrdisks), sizeof(int));
3430 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3431
3432 odisks = reshape.before.data_disks + reshape.parity;
3433 d = reshape_prepare_fdlist(devname, sra, odisks,
3434 nrdisks, blocks, backup_file,
3435 fdlist, offsets);
3436 if (d < odisks) {
3437 goto release;
3438 }
3439 if ((st->ss->manage_reshape == NULL) ||
3440 (st->ss->recover_backup == NULL)) {
3441 if (backup_file == NULL) {
3442 if (reshape.after.data_disks <=
3443 reshape.before.data_disks) {
3444 pr_err("%s: Cannot grow - need backup-file\n",
3445 devname);
3446 pr_err(" Please provide one with \"--backup=...\"\n");
3447 goto release;
3448 } else if (d == odisks) {
3449 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3450 goto release;
3451 }
3452 } else {
3453 if (!reshape_open_backup_file(backup_file, fd, devname,
3454 (signed)blocks,
3455 fdlist+d, offsets+d,
3456 sra->sys_name,
3457 restart)) {
3458 goto release;
3459 }
3460 d++;
3461 }
3462 }
3463
3464 update_cache_size(container, sra, info,
3465 min(reshape.before.data_disks, reshape.after.data_disks),
3466 blocks);
3467
3468 /* Right, everything seems fine. Let's kick things off.
3469 * If only changing raid_disks, use ioctl, else use
3470 * sysfs.
3471 */
3472 sync_metadata(st);
3473
3474 if (impose_reshape(sra, info, st, fd, restart,
3475 devname, container, &reshape) < 0)
3476 goto release;
3477
3478 err = start_reshape(sra, restart, reshape.before.data_disks,
3479 reshape.after.data_disks);
3480 if (err) {
3481 pr_err("Cannot %s reshape for %s\n",
3482 restart ? "continue" : "start",
3483 devname);
3484 goto release;
3485 }
3486 if (restart)
3487 sysfs_set_str(sra, NULL, "array_state", "active");
3488 if (freeze_reshape) {
3489 free(fdlist);
3490 free(offsets);
3491 sysfs_free(sra);
3492 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3493 sra->reshape_progress);
3494 return 1;
3495 }
3496
3497 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3498 if (continue_via_systemd(container ?: sra->sys_name)) {
3499 free(fdlist);
3500 free(offsets);
3501 sysfs_free(sra);
3502 return 0;
3503 }
3504
3505 /* Now we just need to kick off the reshape and watch, while
3506 * handling backups of the data...
3507 * This is all done by a forked background process.
3508 */
3509 switch(forked ? 0 : fork()) {
3510 case -1:
3511 pr_err("Cannot run child to monitor reshape: %s\n",
3512 strerror(errno));
3513 abort_reshape(sra);
3514 goto release;
3515 default:
3516 free(fdlist);
3517 free(offsets);
3518 sysfs_free(sra);
3519 return 0;
3520 case 0:
3521 map_fork();
3522 break;
3523 }
3524
3525 /* If another array on the same devices is busy, the
3526 * reshape will wait for them. This would mean that
3527 * the first section that we suspend will stay suspended
3528 * for a long time. So check on that possibility
3529 * by looking for "DELAYED" in /proc/mdstat, and if found,
3530 * wait a while
3531 */
3532 do {
3533 struct mdstat_ent *mds, *m;
3534 delayed = 0;
3535 mds = mdstat_read(1, 0);
3536 for (m = mds; m; m = m->next)
3537 if (strcmp(m->devnm, sra->sys_name) == 0) {
3538 if (m->resync &&
3539 m->percent == RESYNC_DELAYED)
3540 delayed = 1;
3541 if (m->resync == 0)
3542 /* Haven't started the reshape thread
3543 * yet, wait a bit
3544 */
3545 delayed = 2;
3546 break;
3547 }
3548 free_mdstat(mds);
3549 if (delayed == 1 && get_linux_version() < 3007000) {
3550 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3551 " You might experience problems until other reshapes complete.\n");
3552 delayed = 0;
3553 }
3554 if (delayed)
3555 mdstat_wait(30 - (delayed-1) * 25);
3556 } while (delayed);
3557 mdstat_close();
3558 close(fd);
3559 if (check_env("MDADM_GROW_VERIFY"))
3560 fd = open(devname, O_RDONLY | O_DIRECT);
3561 else
3562 fd = -1;
3563 mlockall(MCL_FUTURE);
3564
3565 signal(SIGTERM, catch_term);
3566
3567 if (st->ss->external) {
3568 /* metadata handler takes it from here */
3569 done = st->ss->manage_reshape(
3570 fd, sra, &reshape, st, blocks,
3571 fdlist, offsets,
3572 d - odisks, fdlist+odisks,
3573 offsets+odisks);
3574 } else
3575 done = child_monitor(
3576 fd, sra, &reshape, st, blocks,
3577 fdlist, offsets,
3578 d - odisks, fdlist+odisks,
3579 offsets+odisks);
3580
3581 free(fdlist);
3582 free(offsets);
3583
3584 if (backup_file && done) {
3585 char *bul;
3586 bul = make_backup(sra->sys_name);
3587 if (bul) {
3588 char buf[1024];
3589 int l = readlink(bul, buf, sizeof(buf) - 1);
3590 if (l > 0) {
3591 buf[l]=0;
3592 unlink(buf);
3593 }
3594 unlink(bul);
3595 free(bul);
3596 }
3597 unlink(backup_file);
3598 }
3599 if (!done) {
3600 abort_reshape(sra);
3601 goto out;
3602 }
3603
3604 if (!st->ss->external &&
3605 !(reshape.before.data_disks != reshape.after.data_disks
3606 && info->custom_array_size) &&
3607 info->new_level == reshape.level &&
3608 !forked) {
3609 /* no need to wait for the reshape to finish as
3610 * there is nothing more to do.
3611 */
3612 sysfs_free(sra);
3613 exit(0);
3614 }
3615 wait_reshape(sra);
3616
3617 if (st->ss->external) {
3618 /* Re-load the metadata as much could have changed */
3619 int cfd = open_dev(st->container_devnm);
3620 if (cfd >= 0) {
3621 flush_mdmon(container);
3622 st->ss->free_super(st);
3623 st->ss->load_container(st, cfd, container);
3624 close(cfd);
3625 }
3626 }
3627
3628 /* set new array size if required customer_array_size is used
3629 * by this metadata.
3630 */
3631 if (reshape.before.data_disks !=
3632 reshape.after.data_disks &&
3633 info->custom_array_size)
3634 set_array_size(st, info, info->text_version);
3635
3636 if (info->new_level != reshape.level) {
3637 if (fd < 0)
3638 fd = open(devname, O_RDONLY);
3639 impose_level(fd, info->new_level, devname, verbose);
3640 close(fd);
3641 if (info->new_level == 0)
3642 st->update_tail = NULL;
3643 }
3644 out:
3645 sysfs_free(sra);
3646 if (forked)
3647 return 0;
3648 unfreeze(st);
3649 exit(0);
3650
3651 release:
3652 free(fdlist);
3653 free(offsets);
3654 if (orig_level != UnSet && sra) {
3655 c = map_num(pers, orig_level);
3656 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3657 pr_err("aborting level change\n");
3658 }
3659 sysfs_free(sra);
3660 if (!forked)
3661 unfreeze(st);
3662 return 1;
3663 }
3664
3665 /* mdfd handle is passed to be closed in child process (after fork).
3666 */
3667 int reshape_container(char *container, char *devname,
3668 int mdfd,
3669 struct supertype *st,
3670 struct mdinfo *info,
3671 int force,
3672 char *backup_file, int verbose,
3673 int forked, int restart, int freeze_reshape)
3674 {
3675 struct mdinfo *cc = NULL;
3676 int rv = restart;
3677 char last_devnm[32] = "";
3678
3679 /* component_size is not meaningful for a container,
3680 * so pass '0' meaning 'no change'
3681 */
3682 if (!restart &&
3683 reshape_super(st, 0, info->new_level,
3684 info->new_layout, info->new_chunk,
3685 info->array.raid_disks, info->delta_disks,
3686 backup_file, devname, APPLY_METADATA_CHANGES,
3687 verbose)) {
3688 unfreeze(st);
3689 return 1;
3690 }
3691
3692 sync_metadata(st);
3693
3694 /* ping monitor to be sure that update is on disk
3695 */
3696 ping_monitor(container);
3697
3698 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3699 if (continue_via_systemd(container))
3700 return 0;
3701
3702 switch (forked ? 0 : fork()) {
3703 case -1: /* error */
3704 perror("Cannot fork to complete reshape\n");
3705 unfreeze(st);
3706 return 1;
3707 default: /* parent */
3708 if (!freeze_reshape)
3709 printf("%s: multi-array reshape continues in background\n", Name);
3710 return 0;
3711 case 0: /* child */
3712 map_fork();
3713 break;
3714 }
3715
3716 /* close unused handle in child process
3717 */
3718 if (mdfd > -1)
3719 close(mdfd);
3720
3721 while(1) {
3722 /* For each member array with reshape_active,
3723 * we need to perform the reshape.
3724 * We pick the first array that needs reshaping and
3725 * reshape it. reshape_array() will re-read the metadata
3726 * so the next time through a different array should be
3727 * ready for reshape.
3728 * It is possible that the 'different' array will not
3729 * be assembled yet. In that case we simple exit.
3730 * When it is assembled, the mdadm which assembles it
3731 * will take over the reshape.
3732 */
3733 struct mdinfo *content;
3734 int fd;
3735 struct mdstat_ent *mdstat;
3736 char *adev;
3737 dev_t devid;
3738
3739 sysfs_free(cc);
3740
3741 cc = st->ss->container_content(st, NULL);
3742
3743 for (content = cc; content ; content = content->next) {
3744 char *subarray;
3745 if (!content->reshape_active)
3746 continue;
3747
3748 subarray = strchr(content->text_version+1, '/')+1;
3749 mdstat = mdstat_by_subdev(subarray, container);
3750 if (!mdstat)
3751 continue;
3752 if (mdstat->active == 0) {
3753 pr_err("Skipping inactive array %s.\n",
3754 mdstat->devnm);
3755 free_mdstat(mdstat);
3756 mdstat = NULL;
3757 continue;
3758 }
3759 break;
3760 }
3761 if (!content)
3762 break;
3763
3764 devid = devnm2devid(mdstat->devnm);
3765 adev = map_dev(major(devid), minor(devid), 0);
3766 if (!adev)
3767 adev = content->text_version;
3768
3769 fd = open_dev(mdstat->devnm);
3770 if (fd < 0) {
3771 pr_err("Device %s cannot be opened for reshape.\n", adev);
3772 break;
3773 }
3774
3775 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3776 /* Do not allow for multiple reshape_array() calls for
3777 * the same array.
3778 * It can happen when reshape_array() returns without
3779 * error, when reshape is not finished (wrong reshape
3780 * starting/continuation conditions). Mdmon doesn't
3781 * switch to next array in container and reentry
3782 * conditions for the same array occur.
3783 * This is possibly interim until the behaviour of
3784 * reshape_array is resolved().
3785 */
3786 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3787 close(fd);
3788 break;
3789 }
3790 strcpy(last_devnm, mdstat->devnm);
3791
3792 sysfs_init(content, fd, mdstat->devnm);
3793
3794 if (mdmon_running(container))
3795 flush_mdmon(container);
3796
3797 rv = reshape_array(container, fd, adev, st,
3798 content, force, NULL, INVALID_SECTORS,
3799 backup_file, verbose, 1, restart,
3800 freeze_reshape);
3801 close(fd);
3802
3803 if (freeze_reshape) {
3804 sysfs_free(cc);
3805 exit(0);
3806 }
3807
3808 restart = 0;
3809 if (rv)
3810 break;
3811
3812 if (mdmon_running(container))
3813 flush_mdmon(container);
3814 }
3815 if (!rv)
3816 unfreeze(st);
3817 sysfs_free(cc);
3818 exit(0);
3819 }
3820
3821 /*
3822 * We run a child process in the background which performs the following
3823 * steps:
3824 * - wait for resync to reach a certain point
3825 * - suspend io to the following section
3826 * - backup that section
3827 * - allow resync to proceed further
3828 * - resume io
3829 * - discard the backup.
3830 *
3831 * When are combined in slightly different ways in the three cases.
3832 * Grow:
3833 * - suspend/backup/allow/wait/resume/discard
3834 * Shrink:
3835 * - allow/wait/suspend/backup/allow/wait/resume/discard
3836 * same-size:
3837 * - wait/resume/discard/suspend/backup/allow
3838 *
3839 * suspend/backup/allow always come together
3840 * wait/resume/discard do too.
3841 * For the same-size case we have two backups to improve flow.
3842 *
3843 */
3844
3845 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3846 unsigned long long backup_point,
3847 unsigned long long wait_point,
3848 unsigned long long *suspend_point,
3849 unsigned long long *reshape_completed, int *frozen)
3850 {
3851 /* This function is called repeatedly by the reshape manager.
3852 * It determines how much progress can safely be made and allows
3853 * that progress.
3854 * - 'info' identifies the array and particularly records in
3855 * ->reshape_progress the metadata's knowledge of progress
3856 * This is a sector offset from the start of the array
3857 * of the next array block to be relocated. This number
3858 * may increase from 0 or decrease from array_size, depending
3859 * on the type of reshape that is happening.
3860 * Note that in contrast, 'sync_completed' is a block count of the
3861 * reshape so far. It gives the distance between the start point
3862 * (head or tail of device) and the next place that data will be
3863 * written. It always increases.
3864 * - 'reshape' is the structure created by analyse_change
3865 * - 'backup_point' shows how much the metadata manager has backed-up
3866 * data. For reshapes with increasing progress, it is the next address
3867 * to be backed up, previous addresses have been backed-up. For
3868 * decreasing progress, it is the earliest address that has been
3869 * backed up - later address are also backed up.
3870 * So addresses between reshape_progress and backup_point are
3871 * backed up providing those are in the 'correct' order.
3872 * - 'wait_point' is an array address. When reshape_completed
3873 * passes this point, progress_reshape should return. It might
3874 * return earlier if it determines that ->reshape_progress needs
3875 * to be updated or further backup is needed.
3876 * - suspend_point is maintained by progress_reshape and the caller
3877 * should not touch it except to initialise to zero.
3878 * It is an array address and it only increases in 2.6.37 and earlier.
3879 * This makes it difficult to handle reducing reshapes with
3880 * external metadata.
3881 * However: it is similar to backup_point in that it records the
3882 * other end of a suspended region from reshape_progress.
3883 * it is moved to extend the region that is safe to backup and/or
3884 * reshape
3885 * - reshape_completed is read from sysfs and returned. The caller
3886 * should copy this into ->reshape_progress when it has reason to
3887 * believe that the metadata knows this, and any backup outside this
3888 * has been erased.
3889 *
3890 * Return value is:
3891 * 1 if more data from backup_point - but only as far as suspend_point,
3892 * should be backed up
3893 * 0 if things are progressing smoothly
3894 * -1 if the reshape is finished because it is all done,
3895 * -2 if the reshape is finished due to an error.
3896 */
3897
3898 int advancing = (reshape->after.data_disks
3899 >= reshape->before.data_disks);
3900 unsigned long long need_backup; /* All data between start of array and
3901 * here will at some point need to
3902 * be backed up.
3903 */
3904 unsigned long long read_offset, write_offset;
3905 unsigned long long write_range;
3906 unsigned long long max_progress, target, completed;
3907 unsigned long long array_size = (info->component_size
3908 * reshape->before.data_disks);
3909 int fd;
3910 char buf[20];
3911
3912 /* First, we unsuspend any region that is now known to be safe.
3913 * If suspend_point is on the 'wrong' side of reshape_progress, then
3914 * we don't have or need suspension at the moment. This is true for
3915 * native metadata when we don't need to back-up.
3916 */
3917 if (advancing) {
3918 if (info->reshape_progress <= *suspend_point)
3919 sysfs_set_num(info, NULL, "suspend_lo",
3920 info->reshape_progress);
3921 } else {
3922 /* Note: this won't work in 2.6.37 and before.
3923 * Something somewhere should make sure we don't need it!
3924 */
3925 if (info->reshape_progress >= *suspend_point)
3926 sysfs_set_num(info, NULL, "suspend_hi",
3927 info->reshape_progress);
3928 }
3929
3930 /* Now work out how far it is safe to progress.
3931 * If the read_offset for ->reshape_progress is less than
3932 * 'blocks' beyond the write_offset, we can only progress as far
3933 * as a backup.
3934 * Otherwise we can progress until the write_offset for the new location
3935 * reaches (within 'blocks' of) the read_offset at the current location.
3936 * However that region must be suspended unless we are using native
3937 * metadata.
3938 * If we need to suspend more, we limit it to 128M per device, which is
3939 * rather arbitrary and should be some time-based calculation.
3940 */
3941 read_offset = info->reshape_progress / reshape->before.data_disks;
3942 write_offset = info->reshape_progress / reshape->after.data_disks;
3943 write_range = info->new_chunk/512;
3944 if (reshape->before.data_disks == reshape->after.data_disks)
3945 need_backup = array_size;
3946 else
3947 need_backup = reshape->backup_blocks;
3948 if (advancing) {
3949 if (read_offset < write_offset + write_range)
3950 max_progress = backup_point;
3951 else
3952 max_progress =
3953 read_offset *
3954 reshape->after.data_disks;
3955 } else {
3956 if (read_offset > write_offset - write_range)
3957 /* Can only progress as far as has been backed up,
3958 * which must be suspended */
3959 max_progress = backup_point;
3960 else if (info->reshape_progress <= need_backup)
3961 max_progress = backup_point;
3962 else {
3963 if (info->array.major_version >= 0)
3964 /* Can progress until backup is needed */
3965 max_progress = need_backup;
3966 else {
3967 /* Can progress until metadata update is required */
3968 max_progress =
3969 read_offset *
3970 reshape->after.data_disks;
3971 /* but data must be suspended */
3972 if (max_progress < *suspend_point)
3973 max_progress = *suspend_point;
3974 }
3975 }
3976 }
3977
3978 /* We know it is safe to progress to 'max_progress' providing
3979 * it is suspended or we are using native metadata.
3980 * Consider extending suspend_point 128M per device if it
3981 * is less than 64M per device beyond reshape_progress.
3982 * But always do a multiple of 'blocks'
3983 * FIXME this is too big - it takes to long to complete
3984 * this much.
3985 */
3986 target = 64*1024*2 * min(reshape->before.data_disks,
3987 reshape->after.data_disks);
3988 target /= reshape->backup_blocks;
3989 if (target < 2)
3990 target = 2;
3991 target *= reshape->backup_blocks;
3992
3993 /* For externally managed metadata we always need to suspend IO to
3994 * the area being reshaped so we regularly push suspend_point forward.
3995 * For native metadata we only need the suspend if we are going to do
3996 * a backup.
3997 */
3998 if (advancing) {
3999 if ((need_backup > info->reshape_progress
4000 || info->array.major_version < 0) &&
4001 *suspend_point < info->reshape_progress + target) {
4002 if (need_backup < *suspend_point + 2 * target)
4003 *suspend_point = need_backup;
4004 else if (*suspend_point + 2 * target < array_size)
4005 *suspend_point += 2 * target;
4006 else
4007 *suspend_point = array_size;
4008 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
4009 if (max_progress > *suspend_point)
4010 max_progress = *suspend_point;
4011 }
4012 } else {
4013 if (info->array.major_version >= 0) {
4014 /* Only need to suspend when about to backup */
4015 if (info->reshape_progress < need_backup * 2 &&
4016 *suspend_point > 0) {
4017 *suspend_point = 0;
4018 sysfs_set_num(info, NULL, "suspend_lo", 0);
4019 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
4020 }
4021 } else {
4022 /* Need to suspend continually */
4023 if (info->reshape_progress < *suspend_point)
4024 *suspend_point = info->reshape_progress;
4025 if (*suspend_point + target < info->reshape_progress)
4026 /* No need to move suspend region yet */;
4027 else {
4028 if (*suspend_point >= 2 * target)
4029 *suspend_point -= 2 * target;
4030 else
4031 *suspend_point = 0;
4032 sysfs_set_num(info, NULL, "suspend_lo",
4033 *suspend_point);
4034 }
4035 if (max_progress < *suspend_point)
4036 max_progress = *suspend_point;
4037 }
4038 }
4039
4040 /* now set sync_max to allow that progress. sync_max, like
4041 * sync_completed is a count of sectors written per device, so
4042 * we find the difference between max_progress and the start point,
4043 * and divide that by after.data_disks to get a sync_max
4044 * number.
4045 * At the same time we convert wait_point to a similar number
4046 * for comparing against sync_completed.
4047 */
4048 /* scale down max_progress to per_disk */
4049 max_progress /= reshape->after.data_disks;
4050 /* Round to chunk size as some kernels give an erroneously high number */
4051 max_progress /= info->new_chunk/512;
4052 max_progress *= info->new_chunk/512;
4053 /* And round to old chunk size as the kernel wants that */
4054 max_progress /= info->array.chunk_size/512;
4055 max_progress *= info->array.chunk_size/512;
4056 /* Limit progress to the whole device */
4057 if (max_progress > info->component_size)
4058 max_progress = info->component_size;
4059 wait_point /= reshape->after.data_disks;
4060 if (!advancing) {
4061 /* switch from 'device offset' to 'processed block count' */
4062 max_progress = info->component_size - max_progress;
4063 wait_point = info->component_size - wait_point;
4064 }
4065
4066 if (!*frozen)
4067 sysfs_set_num(info, NULL, "sync_max", max_progress);
4068
4069 /* Now wait. If we have already reached the point that we were
4070 * asked to wait to, don't wait at all, else wait for any change.
4071 * We need to select on 'sync_completed' as that is the place that
4072 * notifications happen, but we are really interested in
4073 * 'reshape_position'
4074 */
4075 fd = sysfs_get_fd(info, NULL, "sync_completed");
4076 if (fd < 0)
4077 goto check_progress;
4078
4079 if (sysfs_fd_get_ll(fd, &completed) < 0)
4080 goto check_progress;
4081
4082 while (completed < max_progress && completed < wait_point) {
4083 /* Check that sync_action is still 'reshape' to avoid
4084 * waiting forever on a dead array
4085 */
4086 char action[20];
4087 if (sysfs_get_str(info, NULL, "sync_action",
4088 action, 20) <= 0 ||
4089 strncmp(action, "reshape", 7) != 0)
4090 break;
4091 /* Some kernels reset 'sync_completed' to zero
4092 * before setting 'sync_action' to 'idle'.
4093 * So we need these extra tests.
4094 */
4095 if (completed == 0 && advancing
4096 && strncmp(action, "idle", 4) == 0
4097 && info->reshape_progress > 0)
4098 break;
4099 if (completed == 0 && !advancing
4100 && strncmp(action, "idle", 4) == 0
4101 && info->reshape_progress < (info->component_size
4102 * reshape->after.data_disks))
4103 break;
4104 sysfs_wait(fd, NULL);
4105 if (sysfs_fd_get_ll(fd, &completed) < 0)
4106 goto check_progress;
4107 }
4108 /* Some kernels reset 'sync_completed' to zero,
4109 * we need to have real point we are in md.
4110 * So in that case, read 'reshape_position' from sysfs.
4111 */
4112 if (completed == 0) {
4113 unsigned long long reshapep;
4114 char action[20];
4115 if (sysfs_get_str(info, NULL, "sync_action",
4116 action, 20) > 0 &&
4117 strncmp(action, "idle", 4) == 0 &&
4118 sysfs_get_ll(info, NULL,
4119 "reshape_position", &reshapep) == 0)
4120 *reshape_completed = reshapep;
4121 } else {
4122 /* some kernels can give an incorrectly high
4123 * 'completed' number, so round down */
4124 completed /= (info->new_chunk/512);
4125 completed *= (info->new_chunk/512);
4126 /* Convert 'completed' back in to a 'progress' number */
4127 completed *= reshape->after.data_disks;
4128 if (!advancing)
4129 completed = (info->component_size
4130 * reshape->after.data_disks
4131 - completed);
4132 *reshape_completed = completed;
4133 }
4134
4135 close(fd);
4136
4137 /* We return the need_backup flag. Caller will decide
4138 * how much - a multiple of ->backup_blocks up to *suspend_point
4139 */
4140 if (advancing)
4141 return need_backup > info->reshape_progress;
4142 else
4143 return need_backup >= info->reshape_progress;
4144
4145 check_progress:
4146 /* if we couldn't read a number from sync_completed, then
4147 * either the reshape did complete, or it aborted.
4148 * We can tell which by checking for 'none' in reshape_position.
4149 * If it did abort, then it might immediately restart if it
4150 * it was just a device failure that leaves us degraded but
4151 * functioning.
4152 */
4153 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
4154 || strncmp(buf, "none", 4) != 0) {
4155 /* The abort might only be temporary. Wait up to 10
4156 * seconds for fd to contain a valid number again.
4157 */
4158 int wait = 10000;
4159 int rv = -2;
4160 unsigned long long new_sync_max;
4161 while (fd >= 0 && rv < 0 && wait > 0) {
4162 if (sysfs_wait(fd, &wait) != 1)
4163 break;
4164 switch (sysfs_fd_get_ll(fd, &completed)) {
4165 case 0:
4166 /* all good again */
4167 rv = 1;
4168 /* If "sync_max" is no longer max_progress
4169 * we need to freeze things
4170 */
4171 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
4172 *frozen = (new_sync_max != max_progress);
4173 break;
4174 case -2: /* read error - abort */
4175 wait = 0;
4176 break;
4177 }
4178 }
4179 if (fd >= 0)
4180 close(fd);
4181 return rv; /* abort */
4182 } else {
4183 /* Maybe racing with array shutdown - check state */
4184 if (fd >= 0)
4185 close(fd);
4186 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
4187 || strncmp(buf, "inactive", 8) == 0
4188 || strncmp(buf, "clear",5) == 0)
4189 return -2; /* abort */
4190 return -1; /* complete */
4191 }
4192 }
4193
4194 /* FIXME return status is never checked */
4195 static int grow_backup(struct mdinfo *sra,
4196 unsigned long long offset, /* per device */
4197 unsigned long stripes, /* per device, in old chunks */
4198 int *sources, unsigned long long *offsets,
4199 int disks, int chunk, int level, int layout,
4200 int dests, int *destfd, unsigned long long *destoffsets,
4201 int part, int *degraded,
4202 char *buf)
4203 {
4204 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4205 * to storage 'destfd' (offset 'destoffsets'), after first
4206 * suspending IO. Then allow resync to continue
4207 * over the suspended section.
4208 * Use part 'part' of the backup-super-block.
4209 */
4210 int odata = disks;
4211 int rv = 0;
4212 int i;
4213 unsigned long long ll;
4214 int new_degraded;
4215 //printf("offset %llu\n", offset);
4216 if (level >= 4)
4217 odata--;
4218 if (level == 6)
4219 odata--;
4220
4221 /* Check that array hasn't become degraded, else we might backup the wrong data */
4222 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4223 return -1; /* FIXME this error is ignored */
4224 new_degraded = (int)ll;
4225 if (new_degraded != *degraded) {
4226 /* check each device to ensure it is still working */
4227 struct mdinfo *sd;
4228 for (sd = sra->devs ; sd ; sd = sd->next) {
4229 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4230 continue;
4231 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4232 char sbuf[100];
4233
4234 if (sysfs_get_str(sra, sd, "state",
4235 sbuf, sizeof(sbuf)) < 0 ||
4236 strstr(sbuf, "faulty") ||
4237 strstr(sbuf, "in_sync") == NULL) {
4238 /* this device is dead */
4239 sd->disk.state = (1<<MD_DISK_FAULTY);
4240 if (sd->disk.raid_disk >= 0 &&
4241 sources[sd->disk.raid_disk] >= 0) {
4242 close(sources[sd->disk.raid_disk]);
4243 sources[sd->disk.raid_disk] = -1;
4244 }
4245 }
4246 }
4247 }
4248 *degraded = new_degraded;
4249 }
4250 if (part) {
4251 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4252 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4253 } else {
4254 bsb.arraystart = __cpu_to_le64(offset * odata);
4255 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4256 }
4257 if (part)
4258 bsb.magic[15] = '2';
4259 for (i = 0; i < dests; i++)
4260 if (part)
4261 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4262 else
4263 lseek64(destfd[i], destoffsets[i], 0);
4264
4265 rv = save_stripes(sources, offsets,
4266 disks, chunk, level, layout,
4267 dests, destfd,
4268 offset*512*odata, stripes * chunk * odata,
4269 buf);
4270
4271 if (rv)
4272 return rv;
4273 bsb.mtime = __cpu_to_le64(time(0));
4274 for (i = 0; i < dests; i++) {
4275 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4276
4277 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4278 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4279 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4280 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4281
4282 rv = -1;
4283 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4284 != destoffsets[i] - 4096)
4285 break;
4286 if (write(destfd[i], &bsb, 512) != 512)
4287 break;
4288 if (destoffsets[i] > 4096) {
4289 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4290 destoffsets[i]+stripes*chunk*odata)
4291 break;
4292 if (write(destfd[i], &bsb, 512) != 512)
4293 break;
4294 }
4295 fsync(destfd[i]);
4296 rv = 0;
4297 }
4298
4299 return rv;
4300 }
4301
4302 /* in 2.6.30, the value reported by sync_completed can be
4303 * less that it should be by one stripe.
4304 * This only happens when reshape hits sync_max and pauses.
4305 * So allow wait_backup to either extent sync_max further
4306 * than strictly necessary, or return before the
4307 * sync has got quite as far as we would really like.
4308 * This is what 'blocks2' is for.
4309 * The various caller give appropriate values so that
4310 * every works.
4311 */
4312 /* FIXME return value is often ignored */
4313 static int forget_backup(int dests, int *destfd,
4314 unsigned long long *destoffsets,
4315 int part)
4316 {
4317 /*
4318 * Erase backup 'part' (which is 0 or 1)
4319 */
4320 int i;
4321 int rv;
4322
4323 if (part) {
4324 bsb.arraystart2 = __cpu_to_le64(0);
4325 bsb.length2 = __cpu_to_le64(0);
4326 } else {
4327 bsb.arraystart = __cpu_to_le64(0);
4328 bsb.length = __cpu_to_le64(0);
4329 }
4330 bsb.mtime = __cpu_to_le64(time(0));
4331 rv = 0;
4332 for (i = 0; i < dests; i++) {
4333 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4334 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4335 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4336 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4337 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4338 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4339 destoffsets[i]-4096)
4340 rv = -1;
4341 if (rv == 0 &&
4342 write(destfd[i], &bsb, 512) != 512)
4343 rv = -1;
4344 fsync(destfd[i]);
4345 }
4346 return rv;
4347 }
4348
4349 static void fail(char *msg)
4350 {
4351 int rv;
4352 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4353 rv |= (write(2, "\n", 1) != 1);
4354 exit(rv ? 1 : 2);
4355 }
4356
4357 static char *abuf, *bbuf;
4358 static unsigned long long abuflen;
4359 static void validate(int afd, int bfd, unsigned long long offset)
4360 {
4361 /* check that the data in the backup against the array.
4362 * This is only used for regression testing and should not
4363 * be used while the array is active
4364 */
4365 if (afd < 0)
4366 return;
4367 lseek64(bfd, offset - 4096, 0);
4368 if (read(bfd, &bsb2, 512) != 512)
4369 fail("cannot read bsb");
4370 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4371 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4372 fail("first csum bad");
4373 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4374 fail("magic is bad");
4375 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4376 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4377 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4378 fail("second csum bad");
4379
4380 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4381 fail("devstart is wrong");
4382
4383 if (bsb2.length) {
4384 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4385
4386 if (abuflen < len) {
4387 free(abuf);
4388 free(bbuf);
4389 abuflen = len;
4390 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4391 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4392 abuflen = 0;
4393 /* just stop validating on mem-alloc failure */
4394 return;
4395 }
4396 }
4397
4398 lseek64(bfd, offset, 0);
4399 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4400 //printf("len %llu\n", len);
4401 fail("read first backup failed");
4402 }
4403 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4404 if ((unsigned long long)read(afd, abuf, len) != len)
4405 fail("read first from array failed");
4406 if (memcmp(bbuf, abuf, len) != 0) {
4407 #if 0
4408 int i;
4409 printf("offset=%llu len=%llu\n",
4410 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4411 for (i=0; i<len; i++)
4412 if (bbuf[i] != abuf[i]) {
4413 printf("first diff byte %d\n", i);
4414 break;
4415 }
4416 #endif
4417 fail("data1 compare failed");
4418 }
4419 }
4420 if (bsb2.length2) {
4421 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4422
4423 if (abuflen < len) {
4424 free(abuf);
4425 free(bbuf);
4426 abuflen = len;
4427 abuf = xmalloc(abuflen);
4428 bbuf = xmalloc(abuflen);
4429 }
4430
4431 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4432 if ((unsigned long long)read(bfd, bbuf, len) != len)
4433 fail("read second backup failed");
4434 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4435 if ((unsigned long long)read(afd, abuf, len) != len)
4436 fail("read second from array failed");
4437 if (memcmp(bbuf, abuf, len) != 0)
4438 fail("data2 compare failed");
4439 }
4440 }
4441
4442 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4443 struct supertype *st, unsigned long blocks,
4444 int *fds, unsigned long long *offsets,
4445 int dests, int *destfd, unsigned long long *destoffsets)
4446 {
4447 /* Monitor a reshape where backup is being performed using
4448 * 'native' mechanism - either to a backup file, or
4449 * to some space in a spare.
4450 */
4451 char *buf;
4452 int degraded = -1;
4453 unsigned long long speed;
4454 unsigned long long suspend_point, array_size;
4455 unsigned long long backup_point, wait_point;
4456 unsigned long long reshape_completed;
4457 int done = 0;
4458 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4459 int part = 0; /* The next part of the backup area to fill. It may already
4460 * be full, so we need to check */
4461 int level = reshape->level;
4462 int layout = reshape->before.layout;
4463 int data = reshape->before.data_disks;
4464 int disks = reshape->before.data_disks + reshape->parity;
4465 int chunk = sra->array.chunk_size;
4466 struct mdinfo *sd;
4467 unsigned long stripes;
4468 int uuid[4];
4469 int frozen = 0;
4470
4471 /* set up the backup-super-block. This requires the
4472 * uuid from the array.
4473 */
4474 /* Find a superblock */
4475 for (sd = sra->devs; sd; sd = sd->next) {
4476 char *dn;
4477 int devfd;
4478 int ok;
4479 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4480 continue;
4481 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4482 devfd = dev_open(dn, O_RDONLY);
4483 if (devfd < 0)
4484 continue;
4485 ok = st->ss->load_super(st, devfd, NULL);
4486 close(devfd);
4487 if (ok == 0)
4488 break;
4489 }
4490 if (!sd) {
4491 pr_err("Cannot find a superblock\n");
4492 return 0;
4493 }
4494
4495 memset(&bsb, 0, 512);
4496 memcpy(bsb.magic, "md_backup_data-1", 16);
4497 st->ss->uuid_from_super(st, uuid);
4498 memcpy(bsb.set_uuid, uuid, 16);
4499 bsb.mtime = __cpu_to_le64(time(0));
4500 bsb.devstart2 = blocks;
4501
4502 stripes = blocks / (sra->array.chunk_size/512) /
4503 reshape->before.data_disks;
4504
4505 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4506 /* Don't start the 'reshape' */
4507 return 0;
4508 if (reshape->before.data_disks == reshape->after.data_disks) {
4509 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4510 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4511 }
4512
4513 if (increasing) {
4514 array_size = sra->component_size * reshape->after.data_disks;
4515 backup_point = sra->reshape_progress;
4516 suspend_point = 0;
4517 } else {
4518 array_size = sra->component_size * reshape->before.data_disks;
4519 backup_point = reshape->backup_blocks;
4520 suspend_point = array_size;
4521 }
4522
4523 while (!done) {
4524 int rv;
4525
4526 /* Want to return as soon the oldest backup slot can
4527 * be released as that allows us to start backing up
4528 * some more, providing suspend_point has been
4529 * advanced, which it should have.
4530 */
4531 if (increasing) {
4532 wait_point = array_size;
4533 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4534 wait_point = (__le64_to_cpu(bsb.arraystart) +
4535 __le64_to_cpu(bsb.length));
4536 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4537 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4538 __le64_to_cpu(bsb.length2));
4539 } else {
4540 wait_point = 0;
4541 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4542 wait_point = __le64_to_cpu(bsb.arraystart);
4543 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4544 wait_point = __le64_to_cpu(bsb.arraystart2);
4545 }
4546
4547 reshape_completed = sra->reshape_progress;
4548 rv = progress_reshape(sra, reshape,
4549 backup_point, wait_point,
4550 &suspend_point, &reshape_completed,
4551 &frozen);
4552 /* external metadata would need to ping_monitor here */
4553 sra->reshape_progress = reshape_completed;
4554
4555 /* Clear any backup region that is before 'here' */
4556 if (increasing) {
4557 if (__le64_to_cpu(bsb.length) > 0 &&
4558 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4559 __le64_to_cpu(bsb.length)))
4560 forget_backup(dests, destfd,
4561 destoffsets, 0);
4562 if (__le64_to_cpu(bsb.length2) > 0 &&
4563 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4564 __le64_to_cpu(bsb.length2)))
4565 forget_backup(dests, destfd,
4566 destoffsets, 1);
4567 } else {
4568 if (__le64_to_cpu(bsb.length) > 0 &&
4569 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4570 forget_backup(dests, destfd,
4571 destoffsets, 0);
4572 if (__le64_to_cpu(bsb.length2) > 0 &&
4573 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4574 forget_backup(dests, destfd,
4575 destoffsets, 1);
4576 }
4577 if (sigterm)
4578 rv = -2;
4579 if (rv < 0) {
4580 if (rv == -1)
4581 done = 1;
4582 break;
4583 }
4584 if (rv == 0 && increasing && !st->ss->external) {
4585 /* No longer need to monitor this reshape */
4586 sysfs_set_str(sra, NULL, "sync_max", "max");
4587 done = 1;
4588 break;
4589 }
4590
4591 while (rv) {
4592 unsigned long long offset;
4593 unsigned long actual_stripes;
4594 /* Need to backup some data.
4595 * If 'part' is not used and the desired
4596 * backup size is suspended, do a backup,
4597 * then consider the next part.
4598 */
4599 /* Check that 'part' is unused */
4600 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4601 break;
4602 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4603 break;
4604
4605 offset = backup_point / data;
4606 actual_stripes = stripes;
4607 if (increasing) {
4608 if (offset + actual_stripes * (chunk/512) >
4609 sra->component_size)
4610 actual_stripes = ((sra->component_size - offset)
4611 / (chunk/512));
4612 if (offset + actual_stripes * (chunk/512) >
4613 suspend_point/data)
4614 break;
4615 } else {
4616 if (offset < actual_stripes * (chunk/512))
4617 actual_stripes = offset / (chunk/512);
4618 offset -= actual_stripes * (chunk/512);
4619 if (offset < suspend_point/data)
4620 break;
4621 }
4622 if (actual_stripes == 0)
4623 break;
4624 grow_backup(sra, offset, actual_stripes,
4625 fds, offsets,
4626 disks, chunk, level, layout,
4627 dests, destfd, destoffsets,
4628 part, &degraded, buf);
4629 validate(afd, destfd[0], destoffsets[0]);
4630 /* record where 'part' is up to */
4631 part = !part;
4632 if (increasing)
4633 backup_point += actual_stripes * (chunk/512) * data;
4634 else
4635 backup_point -= actual_stripes * (chunk/512) * data;
4636 }
4637 }
4638
4639 /* FIXME maybe call progress_reshape one more time instead */
4640 /* remove any remaining suspension */
4641 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4642 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4643 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4644 sysfs_set_num(sra, NULL, "sync_min", 0);
4645
4646 if (reshape->before.data_disks == reshape->after.data_disks)
4647 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4648 free(buf);
4649 return done;
4650 }
4651
4652 /*
4653 * If any spare contains md_back_data-1 which is recent wrt mtime,
4654 * write that data into the array and update the super blocks with
4655 * the new reshape_progress
4656 */
4657 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4658 char *backup_file, int verbose)
4659 {
4660 int i, j;
4661 int old_disks;
4662 unsigned long long *offsets;
4663 unsigned long long nstripe, ostripe;
4664 int ndata, odata;
4665
4666 odata = info->array.raid_disks - info->delta_disks - 1;
4667 if (info->array.level == 6) odata--; /* number of data disks */
4668 ndata = info->array.raid_disks - 1;
4669 if (info->new_level == 6) ndata--;
4670
4671 old_disks = info->array.raid_disks - info->delta_disks;
4672
4673 if (info->delta_disks <= 0)
4674 /* Didn't grow, so the backup file must have
4675 * been used
4676 */
4677 old_disks = cnt;
4678 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4679 struct mdinfo dinfo;
4680 int fd;
4681 int bsbsize;
4682 char *devname, namebuf[20];
4683 unsigned long long lo, hi;
4684
4685 /* This was a spare and may have some saved data on it.
4686 * Load the superblock, find and load the
4687 * backup_super_block.
4688 * If either fail, go on to next device.
4689 * If the backup contains no new info, just return
4690 * else restore data and update all superblocks
4691 */
4692 if (i == old_disks-1) {
4693 fd = open(backup_file, O_RDONLY);
4694 if (fd<0) {
4695 pr_err("backup file %s inaccessible: %s\n",
4696 backup_file, strerror(errno));
4697 continue;
4698 }
4699 devname = backup_file;
4700 } else {
4701 fd = fdlist[i];
4702 if (fd < 0)
4703 continue;
4704 if (st->ss->load_super(st, fd, NULL))
4705 continue;
4706
4707 st->ss->getinfo_super(st, &dinfo, NULL);
4708 st->ss->free_super(st);
4709
4710 if (lseek64(fd,
4711 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4712 0) < 0) {
4713 pr_err("Cannot seek on device %d\n", i);
4714 continue; /* Cannot seek */
4715 }
4716 sprintf(namebuf, "device-%d", i);
4717 devname = namebuf;
4718 }
4719 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4720 if (verbose)
4721 pr_err("Cannot read from %s\n", devname);
4722 continue; /* Cannot read */
4723 }
4724 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4725 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4726 if (verbose)
4727 pr_err("No backup metadata on %s\n", devname);
4728 continue;
4729 }
4730 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4731 if (verbose)
4732 pr_err("Bad backup-metadata checksum on %s\n", devname);
4733 continue; /* bad checksum */
4734 }
4735 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4736 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4737 if (verbose)
4738 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4739 continue; /* Bad second checksum */
4740 }
4741 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4742 if (verbose)
4743 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4744 continue; /* Wrong uuid */
4745 }
4746
4747 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4748 * sometimes they aren't... So allow considerable flexability in matching, and allow
4749 * this test to be overridden by an environment variable.
4750 */
4751 if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) ||
4752 time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) {
4753 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4754 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4755 (unsigned long)__le64_to_cpu(bsb.mtime),
4756 (unsigned long)info->array.utime);
4757 } else {
4758 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4759 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4760 continue; /* time stamp is too bad */
4761 }
4762 }
4763
4764 if (bsb.magic[15] == '1') {
4765 if (bsb.length == 0)
4766 continue;
4767 if (info->delta_disks >= 0) {
4768 /* reshape_progress is increasing */
4769 if (__le64_to_cpu(bsb.arraystart)
4770 + __le64_to_cpu(bsb.length)
4771 < info->reshape_progress) {
4772 nonew:
4773 if (verbose)
4774 pr_err("backup-metadata found on %s but is not needed\n", devname);
4775 continue; /* No new data here */
4776 }
4777 } else {
4778 /* reshape_progress is decreasing */
4779 if (__le64_to_cpu(bsb.arraystart) >=
4780 info->reshape_progress)
4781 goto nonew; /* No new data here */
4782 }
4783 } else {
4784 if (bsb.length == 0 && bsb.length2 == 0)
4785 continue;
4786 if (info->delta_disks >= 0) {
4787 /* reshape_progress is increasing */
4788 if ((__le64_to_cpu(bsb.arraystart)
4789 + __le64_to_cpu(bsb.length)
4790 < info->reshape_progress)
4791 &&
4792 (__le64_to_cpu(bsb.arraystart2)
4793 + __le64_to_cpu(bsb.length2)
4794 < info->reshape_progress))
4795 goto nonew; /* No new data here */
4796 } else {
4797 /* reshape_progress is decreasing */
4798 if (__le64_to_cpu(bsb.arraystart) >=
4799 info->reshape_progress &&
4800 __le64_to_cpu(bsb.arraystart2) >=
4801 info->reshape_progress)
4802 goto nonew; /* No new data here */
4803 }
4804 }
4805 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4806 second_fail:
4807 if (verbose)
4808 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4809 devname);
4810 continue; /* Cannot seek */
4811 }
4812 /* There should be a duplicate backup superblock 4k before here */
4813 if (lseek64(fd, -4096, 1) < 0 ||
4814 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4815 goto second_fail; /* Cannot find leading superblock */
4816 if (bsb.magic[15] == '1')
4817 bsbsize = offsetof(struct mdp_backup_super, pad1);
4818 else
4819 bsbsize = offsetof(struct mdp_backup_super, pad);
4820 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4821 goto second_fail; /* Cannot find leading superblock */
4822
4823 /* Now need the data offsets for all devices. */
4824 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4825 for(j=0; j<info->array.raid_disks; j++) {
4826 if (fdlist[j] < 0)
4827 continue;
4828 if (st->ss->load_super(st, fdlist[j], NULL))
4829 /* FIXME should be this be an error */
4830 continue;
4831 st->ss->getinfo_super(st, &dinfo, NULL);
4832 st->ss->free_super(st);
4833 offsets[j] = dinfo.data_offset * 512;
4834 }
4835 printf("%s: restoring critical section\n", Name);
4836
4837 if (restore_stripes(fdlist, offsets,
4838 info->array.raid_disks,
4839 info->new_chunk,
4840 info->new_level,
4841 info->new_layout,
4842 fd, __le64_to_cpu(bsb.devstart)*512,
4843 __le64_to_cpu(bsb.arraystart)*512,
4844 __le64_to_cpu(bsb.length)*512, NULL)) {
4845 /* didn't succeed, so giveup */
4846 if (verbose)
4847 pr_err("Error restoring backup from %s\n",
4848 devname);
4849 free(offsets);
4850 return 1;
4851 }
4852
4853 if (bsb.magic[15] == '2' &&
4854 restore_stripes(fdlist, offsets,
4855 info->array.raid_disks,
4856 info->new_chunk,
4857 info->new_level,
4858 info->new_layout,
4859 fd, __le64_to_cpu(bsb.devstart)*512 +
4860 __le64_to_cpu(bsb.devstart2)*512,
4861 __le64_to_cpu(bsb.arraystart2)*512,
4862 __le64_to_cpu(bsb.length2)*512, NULL)) {
4863 /* didn't succeed, so giveup */
4864 if (verbose)
4865 pr_err("Error restoring second backup from %s\n",
4866 devname);
4867 free(offsets);
4868 return 1;
4869 }
4870
4871 free(offsets);
4872
4873 /* Ok, so the data is restored. Let's update those superblocks. */
4874
4875 lo = hi = 0;
4876 if (bsb.length) {
4877 lo = __le64_to_cpu(bsb.arraystart);
4878 hi = lo + __le64_to_cpu(bsb.length);
4879 }
4880 if (bsb.magic[15] == '2' && bsb.length2) {
4881 unsigned long long lo1, hi1;
4882 lo1 = __le64_to_cpu(bsb.arraystart2);
4883 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4884 if (lo == hi) {
4885 lo = lo1;
4886 hi = hi1;
4887 } else if (lo < lo1)
4888 hi = hi1;
4889 else
4890 lo = lo1;
4891 }
4892 if (lo < hi &&
4893 (info->reshape_progress < lo ||
4894 info->reshape_progress > hi))
4895 /* backup does not affect reshape_progress*/ ;
4896 else if (info->delta_disks >= 0) {
4897 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4898 __le64_to_cpu(bsb.length);
4899 if (bsb.magic[15] == '2') {
4900 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4901 __le64_to_cpu(bsb.length2);
4902 if (p2 > info->reshape_progress)
4903 info->reshape_progress = p2;
4904 }
4905 } else {
4906 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4907 if (bsb.magic[15] == '2') {
4908 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4909 if (p2 < info->reshape_progress)
4910 info->reshape_progress = p2;
4911 }
4912 }
4913 for (j=0; j<info->array.raid_disks; j++) {
4914 if (fdlist[j] < 0)
4915 continue;
4916 if (st->ss->load_super(st, fdlist[j], NULL))
4917 continue;
4918 st->ss->getinfo_super(st, &dinfo, NULL);
4919 dinfo.reshape_progress = info->reshape_progress;
4920 st->ss->update_super(st, &dinfo,
4921 "_reshape_progress",
4922 NULL,0, 0, NULL);
4923 st->ss->store_super(st, fdlist[j]);
4924 st->ss->free_super(st);
4925 }
4926 return 0;
4927 }
4928 /* Didn't find any backup data, try to see if any
4929 * was needed.
4930 */
4931 if (info->delta_disks < 0) {
4932 /* When shrinking, the critical section is at the end.
4933 * So see if we are before the critical section.
4934 */
4935 unsigned long long first_block;
4936 nstripe = ostripe = 0;
4937 first_block = 0;
4938 while (ostripe >= nstripe) {
4939 ostripe += info->array.chunk_size / 512;
4940 first_block = ostripe * odata;
4941 nstripe = first_block / ndata / (info->new_chunk/512) *
4942 (info->new_chunk/512);
4943 }
4944
4945 if (info->reshape_progress >= first_block)
4946 return 0;
4947 }
4948 if (info->delta_disks > 0) {
4949 /* See if we are beyond the critical section. */
4950 unsigned long long last_block;
4951 nstripe = ostripe = 0;
4952 last_block = 0;
4953 while (nstripe >= ostripe) {
4954 nstripe += info->new_chunk / 512;
4955 last_block = nstripe * ndata;
4956 ostripe = last_block / odata / (info->array.chunk_size/512) *
4957 (info->array.chunk_size/512);
4958 }
4959
4960 if (info->reshape_progress >= last_block)
4961 return 0;
4962 }
4963 /* needed to recover critical section! */
4964 if (verbose)
4965 pr_err("Failed to find backup of critical section\n");
4966 return 1;
4967 }
4968
4969 int Grow_continue_command(char *devname, int fd,
4970 char *backup_file, int verbose)
4971 {
4972 int ret_val = 0;
4973 struct supertype *st = NULL;
4974 struct mdinfo *content = NULL;
4975 struct mdinfo array;
4976 char *subarray = NULL;
4977 struct mdinfo *cc = NULL;
4978 struct mdstat_ent *mdstat = NULL;
4979 int cfd = -1;
4980 int fd2;
4981
4982 dprintf("Grow continue from command line called for %s\n",
4983 devname);
4984
4985 st = super_by_fd(fd, &subarray);
4986 if (!st || !st->ss) {
4987 pr_err("Unable to determine metadata format for %s\n",
4988 devname);
4989 return 1;
4990 }
4991 dprintf("Grow continue is run for ");
4992 if (st->ss->external == 0) {
4993 int d;
4994 int cnt = 5;
4995 dprintf_cont("native array (%s)\n", devname);
4996 if (md_get_array_info(fd, &array.array) < 0) {
4997 pr_err("%s is not an active md array - aborting\n",
4998 devname);
4999 ret_val = 1;
5000 goto Grow_continue_command_exit;
5001 }
5002 content = &array;
5003 /* Need to load a superblock.
5004 * FIXME we should really get what we need from
5005 * sysfs
5006 */
5007 do {
5008 for (d = 0; d < MAX_DISKS; d++) {
5009 mdu_disk_info_t disk;
5010 char *dv;
5011 int err;
5012 disk.number = d;
5013 if (md_get_disk_info(fd, &disk) < 0)
5014 continue;
5015 if (disk.major == 0 && disk.minor == 0)
5016 continue;
5017 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
5018 continue;
5019 dv = map_dev(disk.major, disk.minor, 1);
5020 if (!dv)
5021 continue;
5022 fd2 = dev_open(dv, O_RDONLY);
5023 if (fd2 < 0)
5024 continue;
5025 err = st->ss->load_super(st, fd2, NULL);
5026 close(fd2);
5027 if (err)
5028 continue;
5029 break;
5030 }
5031 if (d == MAX_DISKS) {
5032 pr_err("Unable to load metadata for %s\n",
5033 devname);
5034 ret_val = 1;
5035 goto Grow_continue_command_exit;
5036 }
5037 st->ss->getinfo_super(st, content, NULL);
5038 if (!content->reshape_active)
5039 sleep(3);
5040 else
5041 break;
5042 } while (cnt-- > 0);
5043 } else {
5044 char *container;
5045
5046 if (subarray) {
5047 dprintf_cont("subarray (%s)\n", subarray);
5048 container = st->container_devnm;
5049 cfd = open_dev_excl(st->container_devnm);
5050 } else {
5051 container = st->devnm;
5052 close(fd);
5053 cfd = open_dev_excl(st->devnm);
5054 dprintf_cont("container (%s)\n", container);
5055 fd = cfd;
5056 }
5057 if (cfd < 0) {
5058 pr_err("Unable to open container for %s\n", devname);
5059 ret_val = 1;
5060 goto Grow_continue_command_exit;
5061 }
5062
5063 /* find in container array under reshape
5064 */
5065 ret_val = st->ss->load_container(st, cfd, NULL);
5066 if (ret_val) {
5067 pr_err("Cannot read superblock for %s\n",
5068 devname);
5069 ret_val = 1;
5070 goto Grow_continue_command_exit;
5071 }
5072
5073 cc = st->ss->container_content(st, subarray);
5074 for (content = cc; content ; content = content->next) {
5075 char *array;
5076 int allow_reshape = 1;
5077
5078 if (content->reshape_active == 0)
5079 continue;
5080 /* The decision about array or container wide
5081 * reshape is taken in Grow_continue based
5082 * content->reshape_active state, therefore we
5083 * need to check_reshape based on
5084 * reshape_active and subarray name
5085 */
5086 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
5087 allow_reshape = 0;
5088 if (content->reshape_active == CONTAINER_RESHAPE &&
5089 (content->array.state
5090 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
5091 allow_reshape = 0;
5092
5093 if (!allow_reshape) {
5094 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5095 devname, container);
5096 ret_val = 1;
5097 goto Grow_continue_command_exit;
5098 }
5099
5100 array = strchr(content->text_version+1, '/')+1;
5101 mdstat = mdstat_by_subdev(array, container);
5102 if (!mdstat)
5103 continue;
5104 if (mdstat->active == 0) {
5105 pr_err("Skipping inactive array %s.\n",
5106 mdstat->devnm);
5107 free_mdstat(mdstat);
5108 mdstat = NULL;
5109 continue;
5110 }
5111 break;
5112 }
5113 if (!content) {
5114 pr_err("Unable to determine reshaped array for %s\n", devname);
5115 ret_val = 1;
5116 goto Grow_continue_command_exit;
5117 }
5118 fd2 = open_dev(mdstat->devnm);
5119 if (fd2 < 0) {
5120 pr_err("cannot open (%s)\n", mdstat->devnm);
5121 ret_val = 1;
5122 goto Grow_continue_command_exit;
5123 }
5124
5125 sysfs_init(content, fd2, mdstat->devnm);
5126
5127 close(fd2);
5128
5129 /* start mdmon in case it is not running
5130 */
5131 if (!mdmon_running(container))
5132 start_mdmon(container);
5133 ping_monitor(container);
5134
5135 if (mdmon_running(container))
5136 st->update_tail = &st->updates;
5137 else {
5138 pr_err("No mdmon found. Grow cannot continue.\n");
5139 ret_val = 1;
5140 goto Grow_continue_command_exit;
5141 }
5142 }
5143
5144 /* verify that array under reshape is started from
5145 * correct position
5146 */
5147 if (verify_reshape_position(content, content->array.level) < 0) {
5148 ret_val = 1;
5149 goto Grow_continue_command_exit;
5150 }
5151
5152 /* continue reshape
5153 */
5154 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
5155
5156 Grow_continue_command_exit:
5157 if (cfd > -1)
5158 close(cfd);
5159 st->ss->free_super(st);
5160 free_mdstat(mdstat);
5161 sysfs_free(cc);
5162 free(subarray);
5163
5164 return ret_val;
5165 }
5166
5167 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
5168 char *backup_file, int forked, int freeze_reshape)
5169 {
5170 int ret_val = 2;
5171
5172 if (!info->reshape_active)
5173 return ret_val;
5174
5175 if (st->ss->external) {
5176 int cfd = open_dev(st->container_devnm);
5177
5178 if (cfd < 0)
5179 return 1;
5180
5181 st->ss->load_container(st, cfd, st->container_devnm);
5182 close(cfd);
5183 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
5184 st, info, 0, backup_file,
5185 0, forked,
5186 1 | info->reshape_active,
5187 freeze_reshape);
5188 } else
5189 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
5190 NULL, INVALID_SECTORS,
5191 backup_file, 0, forked,
5192 1 | info->reshape_active,
5193 freeze_reshape);
5194
5195 return ret_val;
5196 }
5197
5198 char *make_backup(char *name)
5199 {
5200 char *base = "backup_file-";
5201 int len;
5202 char *fname;
5203
5204 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
5205 fname = xmalloc(len);
5206 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
5207 return fname;
5208 }
5209
5210 char *locate_backup(char *name)
5211 {
5212 char *fl = make_backup(name);
5213 struct stat stb;
5214
5215 if (stat(fl, &stb) == 0 &&
5216 S_ISREG(stb.st_mode))
5217 return fl;
5218
5219 free(fl);
5220 return NULL;
5221 }
Unnamed repository; edit this file 'description' to name the repository.