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