]> 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
sysfs: Make sysfs_init() return an error code
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27 #include <stddef.h>
28 #include <stdint.h>
29 #include <signal.h>
30 #include <sys/wait.h>
31
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
34 #endif
35 #include "md_u.h"
36 #include "md_p.h"
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char **backup_filep,
43 int verbose)
44 {
45 int i;
46 int *fdlist;
47 struct mdinfo *dev;
48 int err;
49 int disk_count = next_spare + working_disks;
50 char *backup_file = *backup_filep;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61 sprintf(buf, "%d:%d",
62 dev->disk.major,
63 dev->disk.minor);
64 fd = dev_open(buf, O_RDWR);
65
66 if (dev->disk.raid_disk >= 0)
67 fdlist[dev->disk.raid_disk] = fd;
68 else
69 fdlist[next_spare++] = fd;
70 }
71
72 if (!backup_file) {
73 backup_file = locate_backup(content->sys_name);
74 *backup_filep = backup_file;
75 }
76
77 if (st->ss->external && st->ss->recover_backup)
78 err = st->ss->recover_backup(st, content);
79 else
80 err = Grow_restart(st, content, fdlist, next_spare,
81 backup_file, verbose > 0);
82
83 while (next_spare > 0) {
84 next_spare--;
85 if (fdlist[next_spare] >= 0)
86 close(fdlist[next_spare]);
87 }
88 free(fdlist);
89 if (err) {
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
91 if (!backup_file)
92 pr_err("Possibly you need to specify a --backup-file\n");
93 return 1;
94 }
95
96 dprintf("restore_backup() returns status OK.\n");
97 return 0;
98 }
99
100 int Grow_Add_device(char *devname, int fd, char *newdev)
101 {
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
107 * all other devices.
108 * This means that we need to *find* all other devices.
109 */
110 struct mdinfo info;
111
112 struct stat stb;
113 int nfd, fd2;
114 int d, nd;
115 struct supertype *st = NULL;
116 char *subarray = NULL;
117
118 if (md_get_array_info(fd, &info.array) < 0) {
119 pr_err("cannot get array info for %s\n", devname);
120 return 1;
121 }
122
123 if (info.array.level != -1) {
124 pr_err("can only add devices to linear arrays\n");
125 return 1;
126 }
127
128 st = super_by_fd(fd, &subarray);
129 if (!st) {
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info.array.major_version);
132 return 1;
133 }
134
135 if (subarray) {
136 pr_err("Cannot grow linear sub-arrays yet\n");
137 free(subarray);
138 free(st);
139 return 1;
140 }
141
142 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
143 if (nfd < 0) {
144 pr_err("cannot open %s\n", newdev);
145 free(st);
146 return 1;
147 }
148 fstat(nfd, &stb);
149 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
150 pr_err("%s is not a block device!\n", newdev);
151 close(nfd);
152 free(st);
153 return 1;
154 }
155 /* now check out all the devices and make sure we can read the
156 * superblock */
157 for (d=0 ; d < info.array.raid_disks ; d++) {
158 mdu_disk_info_t disk;
159 char *dv;
160
161 st->ss->free_super(st);
162
163 disk.number = d;
164 if (md_get_disk_info(fd, &disk) < 0) {
165 pr_err("cannot get device detail for device %d\n",
166 d);
167 close(nfd);
168 free(st);
169 return 1;
170 }
171 dv = map_dev(disk.major, disk.minor, 1);
172 if (!dv) {
173 pr_err("cannot find device file for device %d\n",
174 d);
175 close(nfd);
176 free(st);
177 return 1;
178 }
179 fd2 = dev_open(dv, O_RDWR);
180 if (fd2 < 0) {
181 pr_err("cannot open device file %s\n", dv);
182 close(nfd);
183 free(st);
184 return 1;
185 }
186
187 if (st->ss->load_super(st, fd2, NULL)) {
188 pr_err("cannot find super block on %s\n", dv);
189 close(nfd);
190 close(fd2);
191 free(st);
192 return 1;
193 }
194 close(fd2);
195 }
196 /* Ok, looks good. Lets update the superblock and write it out to
197 * newdev.
198 */
199
200 info.disk.number = d;
201 info.disk.major = major(stb.st_rdev);
202 info.disk.minor = minor(stb.st_rdev);
203 info.disk.raid_disk = d;
204 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
205 st->ss->update_super(st, &info, "linear-grow-new", newdev,
206 0, 0, NULL);
207
208 if (st->ss->store_super(st, nfd)) {
209 pr_err("Cannot store new superblock on %s\n",
210 newdev);
211 close(nfd);
212 return 1;
213 }
214 close(nfd);
215
216 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
217 pr_err("Cannot add new disk to this array\n");
218 return 1;
219 }
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
222 */
223
224 if (md_get_array_info(fd, &info.array) < 0) {
225 pr_err("cannot get array info for %s\n", devname);
226 return 1;
227 }
228
229 nd = d;
230 for (d=0 ; d < info.array.raid_disks ; d++) {
231 mdu_disk_info_t disk;
232 char *dv;
233
234 disk.number = d;
235 if (md_get_disk_info(fd, &disk) < 0) {
236 pr_err("cannot get device detail for device %d\n",
237 d);
238 return 1;
239 }
240 dv = map_dev(disk.major, disk.minor, 1);
241 if (!dv) {
242 pr_err("cannot find device file for device %d\n",
243 d);
244 return 1;
245 }
246 fd2 = dev_open(dv, O_RDWR);
247 if (fd2 < 0) {
248 pr_err("cannot open device file %s\n", dv);
249 return 1;
250 }
251 if (st->ss->load_super(st, fd2, NULL)) {
252 pr_err("cannot find super block on %s\n", dv);
253 close(fd);
254 return 1;
255 }
256 info.array.raid_disks = nd+1;
257 info.array.nr_disks = nd+1;
258 info.array.active_disks = nd+1;
259 info.array.working_disks = nd+1;
260
261 st->ss->update_super(st, &info, "linear-grow-update", dv,
262 0, 0, NULL);
263
264 if (st->ss->store_super(st, fd2)) {
265 pr_err("Cannot store new superblock on %s\n", dv);
266 close(fd2);
267 return 1;
268 }
269 close(fd2);
270 }
271
272 return 0;
273 }
274
275 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
276 {
277 /*
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
280 * Then add it
281 *
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
284 * to all devices
285 */
286 mdu_bitmap_file_t bmf;
287 mdu_array_info_t array;
288 struct supertype *st;
289 char *subarray = NULL;
290 int major = BITMAP_MAJOR_HI;
291 int vers = md_get_version(fd);
292 unsigned long long bitmapsize, array_size;
293 struct mdinfo *mdi;
294
295 if (vers < 9003) {
296 major = BITMAP_MAJOR_HOSTENDIAN;
297 pr_err("Warning - bitmaps created on this kernel are not portable\n"
298 " between different architectures. Consider upgrading the Linux kernel.\n");
299 }
300
301 /*
302 * We only ever get called if s->bitmap_file is != NULL, so this check
303 * is just here to quiet down static code checkers.
304 */
305 if (!s->bitmap_file)
306 return 1;
307
308 if (strcmp(s->bitmap_file, "clustered") == 0)
309 major = BITMAP_MAJOR_CLUSTERED;
310
311 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
312 if (errno == ENOMEM)
313 pr_err("Memory allocation failure.\n");
314 else
315 pr_err("bitmaps not supported by this kernel.\n");
316 return 1;
317 }
318 if (bmf.pathname[0]) {
319 if (strcmp(s->bitmap_file,"none") == 0) {
320 if (ioctl(fd, SET_BITMAP_FILE, -1) != 0) {
321 pr_err("failed to remove bitmap %s\n",
322 bmf.pathname);
323 return 1;
324 }
325 return 0;
326 }
327 pr_err("%s already has a bitmap (%s)\n",
328 devname, bmf.pathname);
329 return 1;
330 }
331 if (md_get_array_info(fd, &array) != 0) {
332 pr_err("cannot get array status for %s\n", devname);
333 return 1;
334 }
335 if (array.state & (1 << MD_SB_BITMAP_PRESENT)) {
336 if (strcmp(s->bitmap_file, "none")==0) {
337 array.state &= ~(1 << MD_SB_BITMAP_PRESENT);
338 if (md_set_array_info(fd, &array) != 0) {
339 if (array.state & (1 << MD_SB_CLUSTERED))
340 pr_err("failed to remove clustered bitmap.\n");
341 else
342 pr_err("failed to remove internal bitmap.\n");
343 return 1;
344 }
345 return 0;
346 }
347 pr_err("bitmap already present on %s\n", devname);
348 return 1;
349 }
350
351 if (strcmp(s->bitmap_file, "none") == 0) {
352 pr_err("no bitmap found on %s\n", devname);
353 return 1;
354 }
355 if (array.level <= 0) {
356 pr_err("Bitmaps not meaningful with level %s\n",
357 map_num(pers, array.level)?:"of this array");
358 return 1;
359 }
360 bitmapsize = array.size;
361 bitmapsize <<= 1;
362 if (get_dev_size(fd, NULL, &array_size) &&
363 array_size > (0x7fffffffULL << 9)) {
364 /* Array is big enough that we cannot trust array.size
365 * try other approaches
366 */
367 bitmapsize = get_component_size(fd);
368 }
369 if (bitmapsize == 0) {
370 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
371 return 1;
372 }
373
374 if (array.level == 10) {
375 int ncopies;
376
377 ncopies = (array.layout & 255) * ((array.layout >> 8) & 255);
378 bitmapsize = bitmapsize * array.raid_disks / ncopies;
379 }
380
381 st = super_by_fd(fd, &subarray);
382 if (!st) {
383 pr_err("Cannot understand version %d.%d\n",
384 array.major_version, array.minor_version);
385 return 1;
386 }
387 if (subarray) {
388 pr_err("Cannot add bitmaps to sub-arrays yet\n");
389 free(subarray);
390 free(st);
391 return 1;
392 }
393
394 mdi = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY);
395 if (mdi) {
396 if (mdi->consistency_policy == CONSISTENCY_POLICY_PPL) {
397 pr_err("Cannot add bitmap to array with PPL\n");
398 free(mdi);
399 free(st);
400 return 1;
401 }
402 free(mdi);
403 }
404
405 if (strcmp(s->bitmap_file, "internal") == 0 ||
406 strcmp(s->bitmap_file, "clustered") == 0) {
407 int rv;
408 int d;
409 int offset_setable = 0;
410 if (st->ss->add_internal_bitmap == NULL) {
411 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
412 return 1;
413 }
414 st->nodes = c->nodes;
415 st->cluster_name = c->homecluster;
416 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
417 if (mdi)
418 offset_setable = 1;
419 for (d = 0; d < st->max_devs; d++) {
420 mdu_disk_info_t disk;
421 char *dv;
422 int fd2;
423
424 disk.number = d;
425 if (md_get_disk_info(fd, &disk) < 0)
426 continue;
427 if (disk.major == 0 && disk.minor == 0)
428 continue;
429 if ((disk.state & (1 << MD_DISK_SYNC)) == 0)
430 continue;
431 dv = map_dev(disk.major, disk.minor, 1);
432 if (!dv)
433 continue;
434 fd2 = dev_open(dv, O_RDWR);
435 if (fd2 < 0)
436 continue;
437 rv = st->ss->load_super(st, fd2, NULL);
438 if (!rv) {
439 rv = st->ss->add_internal_bitmap(
440 st, &s->bitmap_chunk, c->delay,
441 s->write_behind, bitmapsize,
442 offset_setable, major);
443 if (!rv) {
444 st->ss->write_bitmap(st, fd2,
445 NodeNumUpdate);
446 } else {
447 pr_err("failed to create internal bitmap - chunksize problem.\n");
448 }
449 } else {
450 pr_err("failed to load super-block.\n");
451 }
452 close(fd2);
453 if (rv)
454 return 1;
455 }
456 if (offset_setable) {
457 st->ss->getinfo_super(st, mdi, NULL);
458 if (sysfs_init(mdi, fd, NULL)) {
459 pr_err("failed to intialize sysfs.\n");
460 free(mdi);
461 }
462 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
463 mdi->bitmap_offset);
464 free(mdi);
465 } else {
466 if (strcmp(s->bitmap_file, "clustered") == 0)
467 array.state |= (1 << MD_SB_CLUSTERED);
468 array.state |= (1 << MD_SB_BITMAP_PRESENT);
469 rv = md_set_array_info(fd, &array);
470 }
471 if (rv < 0) {
472 if (errno == EBUSY)
473 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
474 pr_err("failed to set internal bitmap.\n");
475 return 1;
476 }
477 } else {
478 int uuid[4];
479 int bitmap_fd;
480 int d;
481 int max_devs = st->max_devs;
482
483 /* try to load a superblock */
484 for (d = 0; d < max_devs; d++) {
485 mdu_disk_info_t disk;
486 char *dv;
487 int fd2;
488 disk.number = d;
489 if (md_get_disk_info(fd, &disk) < 0)
490 continue;
491 if ((disk.major==0 && disk.minor == 0) ||
492 (disk.state & (1 << MD_DISK_REMOVED)))
493 continue;
494 dv = map_dev(disk.major, disk.minor, 1);
495 if (!dv)
496 continue;
497 fd2 = dev_open(dv, O_RDONLY);
498 if (fd2 >= 0) {
499 if (st->ss->load_super(st, fd2, NULL) == 0) {
500 close(fd2);
501 st->ss->uuid_from_super(st, uuid);
502 break;
503 }
504 close(fd2);
505 }
506 }
507 if (d == max_devs) {
508 pr_err("cannot find UUID for array!\n");
509 return 1;
510 }
511 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid,
512 s->bitmap_chunk, c->delay, s->write_behind,
513 bitmapsize, major)) {
514 return 1;
515 }
516 bitmap_fd = open(s->bitmap_file, O_RDWR);
517 if (bitmap_fd < 0) {
518 pr_err("weird: %s cannot be opened\n", s->bitmap_file);
519 return 1;
520 }
521 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
522 int err = errno;
523 if (errno == EBUSY)
524 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
525 pr_err("Cannot set bitmap file for %s: %s\n",
526 devname, strerror(err));
527 return 1;
528 }
529 }
530
531 return 0;
532 }
533
534 int Grow_consistency_policy(char *devname, int fd, struct context *c, struct shape *s)
535 {
536 struct supertype *st;
537 struct mdinfo *sra;
538 struct mdinfo *sd;
539 char *subarray = NULL;
540 int ret = 0;
541 char container_dev[PATH_MAX];
542
543 if (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
544 s->consistency_policy != CONSISTENCY_POLICY_PPL) {
545 pr_err("Operation not supported for consistency policy %s\n",
546 map_num(consistency_policies, s->consistency_policy));
547 return 1;
548 }
549
550 st = super_by_fd(fd, &subarray);
551 if (!st)
552 return 1;
553
554 sra = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY|GET_LEVEL|
555 GET_DEVS|GET_STATE);
556 if (!sra) {
557 ret = 1;
558 goto free_st;
559 }
560
561 if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
562 !st->ss->write_init_ppl) {
563 pr_err("%s metadata does not support PPL\n", st->ss->name);
564 ret = 1;
565 goto free_info;
566 }
567
568 if (sra->array.level != 5) {
569 pr_err("Operation not supported for array level %d\n",
570 sra->array.level);
571 ret = 1;
572 goto free_info;
573 }
574
575 if (sra->consistency_policy == (unsigned)s->consistency_policy) {
576 pr_err("Consistency policy is already %s\n",
577 map_num(consistency_policies, s->consistency_policy));
578 ret = 1;
579 goto free_info;
580 } else if (sra->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
581 sra->consistency_policy != CONSISTENCY_POLICY_PPL) {
582 pr_err("Current consistency policy is %s, cannot change to %s\n",
583 map_num(consistency_policies, sra->consistency_policy),
584 map_num(consistency_policies, s->consistency_policy));
585 ret = 1;
586 goto free_info;
587 }
588
589 if (subarray) {
590 char *update;
591
592 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
593 update = "ppl";
594 else
595 update = "no-ppl";
596
597 sprintf(container_dev, "/dev/%s", st->container_devnm);
598
599 ret = Update_subarray(container_dev, subarray, update, NULL,
600 c->verbose);
601 if (ret)
602 goto free_info;
603 }
604
605 if (s->consistency_policy == CONSISTENCY_POLICY_PPL) {
606 struct mdinfo info;
607
608 if (subarray) {
609 struct mdinfo *mdi;
610 int cfd;
611
612 cfd = open(container_dev, O_RDWR|O_EXCL);
613 if (cfd < 0) {
614 pr_err("Failed to open %s\n", container_dev);
615 ret = 1;
616 goto free_info;
617 }
618
619 ret = st->ss->load_container(st, cfd, st->container_devnm);
620 close(cfd);
621
622 if (ret) {
623 pr_err("Cannot read superblock for %s\n",
624 container_dev);
625 goto free_info;
626 }
627
628 mdi = st->ss->container_content(st, subarray);
629 info = *mdi;
630 free(mdi);
631 }
632
633 for (sd = sra->devs; sd; sd = sd->next) {
634 int dfd;
635 char *devpath;
636
637 if ((sd->disk.state & (1 << MD_DISK_SYNC)) == 0)
638 continue;
639
640 devpath = map_dev(sd->disk.major, sd->disk.minor, 0);
641 dfd = dev_open(devpath, O_RDWR);
642 if (dfd < 0) {
643 pr_err("Failed to open %s\n", devpath);
644 ret = 1;
645 goto free_info;
646 }
647
648 if (!subarray) {
649 ret = st->ss->load_super(st, dfd, NULL);
650 if (ret) {
651 pr_err("Failed to load super-block.\n");
652 close(dfd);
653 goto free_info;
654 }
655
656 ret = st->ss->update_super(st, sra, "ppl", devname,
657 c->verbose, 0, NULL);
658 if (ret) {
659 close(dfd);
660 st->ss->free_super(st);
661 goto free_info;
662 }
663 st->ss->getinfo_super(st, &info, NULL);
664 }
665
666 ret |= sysfs_set_num(sra, sd, "ppl_sector", info.ppl_sector);
667 ret |= sysfs_set_num(sra, sd, "ppl_size", info.ppl_size);
668
669 if (ret) {
670 pr_err("Failed to set PPL attributes for %s\n",
671 sd->sys_name);
672 close(dfd);
673 st->ss->free_super(st);
674 goto free_info;
675 }
676
677 ret = st->ss->write_init_ppl(st, &info, dfd);
678 if (ret)
679 pr_err("Failed to write PPL\n");
680
681 close(dfd);
682
683 if (!subarray)
684 st->ss->free_super(st);
685
686 if (ret)
687 goto free_info;
688 }
689 }
690
691 ret = sysfs_set_str(sra, NULL, "consistency_policy",
692 map_num(consistency_policies,
693 s->consistency_policy));
694 if (ret)
695 pr_err("Failed to change array consistency policy\n");
696
697 free_info:
698 sysfs_free(sra);
699 free_st:
700 free(st);
701 free(subarray);
702
703 return ret;
704 }
705
706 /*
707 * When reshaping an array we might need to backup some data.
708 * This is written to all spares with a 'super_block' describing it.
709 * The superblock goes 4K from the end of the used space on the
710 * device.
711 * It if written after the backup is complete.
712 * It has the following structure.
713 */
714
715 static struct mdp_backup_super {
716 char magic[16]; /* md_backup_data-1 or -2 */
717 __u8 set_uuid[16];
718 __u64 mtime;
719 /* start/sizes in 512byte sectors */
720 __u64 devstart; /* address on backup device/file of data */
721 __u64 arraystart;
722 __u64 length;
723 __u32 sb_csum; /* csum of preceeding bytes. */
724 __u32 pad1;
725 __u64 devstart2; /* offset in to data of second section */
726 __u64 arraystart2;
727 __u64 length2;
728 __u32 sb_csum2; /* csum of preceeding bytes. */
729 __u8 pad[512-68-32];
730 } __attribute__((aligned(512))) bsb, bsb2;
731
732 static __u32 bsb_csum(char *buf, int len)
733 {
734 int i;
735 int csum = 0;
736 for (i = 0; i < len; i++)
737 csum = (csum<<3) + buf[0];
738 return __cpu_to_le32(csum);
739 }
740
741 static int check_idle(struct supertype *st)
742 {
743 /* Check that all member arrays for this container, or the
744 * container of this array, are idle
745 */
746 char *container = (st->container_devnm[0]
747 ? st->container_devnm : st->devnm);
748 struct mdstat_ent *ent, *e;
749 int is_idle = 1;
750
751 ent = mdstat_read(0, 0);
752 for (e = ent ; e; e = e->next) {
753 if (!is_container_member(e, container))
754 continue;
755 if (e->percent >= 0) {
756 is_idle = 0;
757 break;
758 }
759 }
760 free_mdstat(ent);
761 return is_idle;
762 }
763
764 static int freeze_container(struct supertype *st)
765 {
766 char *container = (st->container_devnm[0]
767 ? st->container_devnm : st->devnm);
768
769 if (!check_idle(st))
770 return -1;
771
772 if (block_monitor(container, 1)) {
773 pr_err("failed to freeze container\n");
774 return -2;
775 }
776
777 return 1;
778 }
779
780 static void unfreeze_container(struct supertype *st)
781 {
782 char *container = (st->container_devnm[0]
783 ? st->container_devnm : st->devnm);
784
785 unblock_monitor(container, 1);
786 }
787
788 static int freeze(struct supertype *st)
789 {
790 /* Try to freeze resync/rebuild on this array/container.
791 * Return -1 if the array is busy,
792 * return -2 container cannot be frozen,
793 * return 0 if this kernel doesn't support 'frozen'
794 * return 1 if it worked.
795 */
796 if (st->ss->external)
797 return freeze_container(st);
798 else {
799 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
800 int err;
801 char buf[20];
802
803 if (!sra)
804 return -1;
805 /* Need to clear any 'read-auto' status */
806 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
807 strncmp(buf, "read-auto", 9) == 0)
808 sysfs_set_str(sra, NULL, "array_state", "clean");
809
810 err = sysfs_freeze_array(sra);
811 sysfs_free(sra);
812 return err;
813 }
814 }
815
816 static void unfreeze(struct supertype *st)
817 {
818 if (st->ss->external)
819 return unfreeze_container(st);
820 else {
821 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
822 char buf[20];
823
824 if (sra &&
825 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
826 && strcmp(buf, "frozen\n") == 0)
827 sysfs_set_str(sra, NULL, "sync_action", "idle");
828 sysfs_free(sra);
829 }
830 }
831
832 static void wait_reshape(struct mdinfo *sra)
833 {
834 int fd = sysfs_get_fd(sra, NULL, "sync_action");
835 char action[20];
836
837 if (fd < 0)
838 return;
839
840 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
841 strncmp(action, "reshape", 7) == 0)
842 sysfs_wait(fd, NULL);
843 close(fd);
844 }
845
846 static int reshape_super(struct supertype *st, unsigned long long size,
847 int level, int layout, int chunksize, int raid_disks,
848 int delta_disks, char *backup_file, char *dev,
849 int direction, int verbose)
850 {
851 /* nothing extra to check in the native case */
852 if (!st->ss->external)
853 return 0;
854 if (!st->ss->reshape_super ||
855 !st->ss->manage_reshape) {
856 pr_err("%s metadata does not support reshape\n",
857 st->ss->name);
858 return 1;
859 }
860
861 return st->ss->reshape_super(st, size, level, layout, chunksize,
862 raid_disks, delta_disks, backup_file, dev,
863 direction, verbose);
864 }
865
866 static void sync_metadata(struct supertype *st)
867 {
868 if (st->ss->external) {
869 if (st->update_tail) {
870 flush_metadata_updates(st);
871 st->update_tail = &st->updates;
872 } else
873 st->ss->sync_metadata(st);
874 }
875 }
876
877 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
878 {
879 /* when dealing with external metadata subarrays we need to be
880 * prepared to handle EAGAIN. The kernel may need to wait for
881 * mdmon to mark the array active so the kernel can handle
882 * allocations/writeback when preparing the reshape action
883 * (md_allow_write()). We temporarily disable safe_mode_delay
884 * to close a race with the array_state going clean before the
885 * next write to raid_disks / stripe_cache_size
886 */
887 char safe[50];
888 int rc;
889
890 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
891 if (!container ||
892 (strcmp(name, "raid_disks") != 0 &&
893 strcmp(name, "stripe_cache_size") != 0))
894 return sysfs_set_num(sra, NULL, name, n);
895
896 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
897 if (rc <= 0)
898 return -1;
899 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
900 rc = sysfs_set_num(sra, NULL, name, n);
901 if (rc < 0 && errno == EAGAIN) {
902 ping_monitor(container);
903 /* if we get EAGAIN here then the monitor is not active
904 * so stop trying
905 */
906 rc = sysfs_set_num(sra, NULL, name, n);
907 }
908 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
909 return rc;
910 }
911
912 int start_reshape(struct mdinfo *sra, int already_running,
913 int before_data_disks, int data_disks)
914 {
915 int err;
916 unsigned long long sync_max_to_set;
917
918 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
919 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
920 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
921 sra->reshape_progress);
922 if (before_data_disks <= data_disks)
923 sync_max_to_set = sra->reshape_progress / data_disks;
924 else
925 sync_max_to_set = (sra->component_size * data_disks
926 - sra->reshape_progress) / data_disks;
927 if (!already_running)
928 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
929 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
930 if (!already_running && err == 0) {
931 int cnt = 5;
932 do {
933 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
934 if (err)
935 sleep(1);
936 } while (err && errno == EBUSY && cnt-- > 0);
937 }
938 return err;
939 }
940
941 void abort_reshape(struct mdinfo *sra)
942 {
943 sysfs_set_str(sra, NULL, "sync_action", "idle");
944 /*
945 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
946 * suspend_hi to decrease as well as increase.")
947 * you could only increase suspend_{lo,hi} unless the region they
948 * covered was empty. So to reset to 0, you need to push suspend_lo
949 * up past suspend_hi first. So to maximize the chance of mdadm
950 * working on all kernels, we want to keep doing that.
951 */
952 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
953 sysfs_set_num(sra, NULL, "suspend_hi", 0);
954 sysfs_set_num(sra, NULL, "suspend_lo", 0);
955 sysfs_set_num(sra, NULL, "sync_min", 0);
956 // It isn't safe to reset sync_max as we aren't monitoring.
957 // Array really should be stopped at this point.
958 }
959
960 int remove_disks_for_takeover(struct supertype *st,
961 struct mdinfo *sra,
962 int layout)
963 {
964 int nr_of_copies;
965 struct mdinfo *remaining;
966 int slot;
967
968 if (st->ss->external) {
969 int rv = 0;
970 struct mdinfo *arrays = st->ss->container_content(st, NULL);
971 /*
972 * containter_content returns list of arrays in container
973 * If arrays->next is not NULL it means that there are
974 * 2 arrays in container and operation should be blocked
975 */
976 if (arrays) {
977 if (arrays->next)
978 rv = 1;
979 sysfs_free(arrays);
980 if (rv) {
981 pr_err("Error. Cannot perform operation on /dev/%s\n", st->devnm);
982 pr_err("For this operation it MUST be single array in container\n");
983 return rv;
984 }
985 }
986 }
987
988 if (sra->array.level == 10)
989 nr_of_copies = layout & 0xff;
990 else if (sra->array.level == 1)
991 nr_of_copies = sra->array.raid_disks;
992 else
993 return 1;
994
995 remaining = sra->devs;
996 sra->devs = NULL;
997 /* for each 'copy', select one device and remove from the list. */
998 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
999 struct mdinfo **diskp;
1000 int found = 0;
1001
1002 /* Find a working device to keep */
1003 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
1004 struct mdinfo *disk = *diskp;
1005
1006 if (disk->disk.raid_disk < slot)
1007 continue;
1008 if (disk->disk.raid_disk >= slot + nr_of_copies)
1009 continue;
1010 if (disk->disk.state & (1<<MD_DISK_REMOVED))
1011 continue;
1012 if (disk->disk.state & (1<<MD_DISK_FAULTY))
1013 continue;
1014 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
1015 continue;
1016
1017 /* We have found a good disk to use! */
1018 *diskp = disk->next;
1019 disk->next = sra->devs;
1020 sra->devs = disk;
1021 found = 1;
1022 break;
1023 }
1024 if (!found)
1025 break;
1026 }
1027
1028 if (slot < sra->array.raid_disks) {
1029 /* didn't find all slots */
1030 struct mdinfo **e;
1031 e = &remaining;
1032 while (*e)
1033 e = &(*e)->next;
1034 *e = sra->devs;
1035 sra->devs = remaining;
1036 return 1;
1037 }
1038
1039 /* Remove all 'remaining' devices from the array */
1040 while (remaining) {
1041 struct mdinfo *sd = remaining;
1042 remaining = sd->next;
1043
1044 sysfs_set_str(sra, sd, "state", "faulty");
1045 sysfs_set_str(sra, sd, "slot", "none");
1046 /* for external metadata disks should be removed in mdmon */
1047 if (!st->ss->external)
1048 sysfs_set_str(sra, sd, "state", "remove");
1049 sd->disk.state |= (1<<MD_DISK_REMOVED);
1050 sd->disk.state &= ~(1<<MD_DISK_SYNC);
1051 sd->next = sra->devs;
1052 sra->devs = sd;
1053 }
1054 return 0;
1055 }
1056
1057 void reshape_free_fdlist(int *fdlist,
1058 unsigned long long *offsets,
1059 int size)
1060 {
1061 int i;
1062
1063 for (i = 0; i < size; i++)
1064 if (fdlist[i] >= 0)
1065 close(fdlist[i]);
1066
1067 free(fdlist);
1068 free(offsets);
1069 }
1070
1071 int reshape_prepare_fdlist(char *devname,
1072 struct mdinfo *sra,
1073 int raid_disks,
1074 int nrdisks,
1075 unsigned long blocks,
1076 char *backup_file,
1077 int *fdlist,
1078 unsigned long long *offsets)
1079 {
1080 int d = 0;
1081 struct mdinfo *sd;
1082
1083 enable_fds(nrdisks);
1084 for (d = 0; d <= nrdisks; d++)
1085 fdlist[d] = -1;
1086 d = raid_disks;
1087 for (sd = sra->devs; sd; sd = sd->next) {
1088 if (sd->disk.state & (1<<MD_DISK_FAULTY))
1089 continue;
1090 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
1091 sd->disk.raid_disk < raid_disks) {
1092 char *dn = map_dev(sd->disk.major,
1093 sd->disk.minor, 1);
1094 fdlist[sd->disk.raid_disk]
1095 = dev_open(dn, O_RDONLY);
1096 offsets[sd->disk.raid_disk] = sd->data_offset*512;
1097 if (fdlist[sd->disk.raid_disk] < 0) {
1098 pr_err("%s: cannot open component %s\n",
1099 devname, dn ? dn : "-unknown-");
1100 d = -1;
1101 goto release;
1102 }
1103 } else if (backup_file == NULL) {
1104 /* spare */
1105 char *dn = map_dev(sd->disk.major,
1106 sd->disk.minor, 1);
1107 fdlist[d] = dev_open(dn, O_RDWR);
1108 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
1109 if (fdlist[d] < 0) {
1110 pr_err("%s: cannot open component %s\n",
1111 devname, dn ? dn : "-unknown-");
1112 d = -1;
1113 goto release;
1114 }
1115 d++;
1116 }
1117 }
1118 release:
1119 return d;
1120 }
1121
1122 int reshape_open_backup_file(char *backup_file,
1123 int fd,
1124 char *devname,
1125 long blocks,
1126 int *fdlist,
1127 unsigned long long *offsets,
1128 char *sys_name,
1129 int restart)
1130 {
1131 /* Return 1 on success, 0 on any form of failure */
1132 /* need to check backup file is large enough */
1133 char buf[512];
1134 struct stat stb;
1135 unsigned int dev;
1136 int i;
1137
1138 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
1139 S_IRUSR | S_IWUSR);
1140 *offsets = 8 * 512;
1141 if (*fdlist < 0) {
1142 pr_err("%s: cannot create backup file %s: %s\n",
1143 devname, backup_file, strerror(errno));
1144 return 0;
1145 }
1146 /* Guard against backup file being on array device.
1147 * If array is partitioned or if LVM etc is in the
1148 * way this will not notice, but it is better than
1149 * nothing.
1150 */
1151 fstat(*fdlist, &stb);
1152 dev = stb.st_dev;
1153 fstat(fd, &stb);
1154 if (stb.st_rdev == dev) {
1155 pr_err("backup file must NOT be on the array being reshaped.\n");
1156 close(*fdlist);
1157 return 0;
1158 }
1159
1160 memset(buf, 0, 512);
1161 for (i=0; i < blocks + 8 ; i++) {
1162 if (write(*fdlist, buf, 512) != 512) {
1163 pr_err("%s: cannot create backup file %s: %s\n",
1164 devname, backup_file, strerror(errno));
1165 return 0;
1166 }
1167 }
1168 if (fsync(*fdlist) != 0) {
1169 pr_err("%s: cannot create backup file %s: %s\n",
1170 devname, backup_file, strerror(errno));
1171 return 0;
1172 }
1173
1174 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
1175 char *bu = make_backup(sys_name);
1176 if (symlink(backup_file, bu))
1177 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
1178 strerror(errno));
1179 free(bu);
1180 }
1181
1182 return 1;
1183 }
1184
1185 unsigned long compute_backup_blocks(int nchunk, int ochunk,
1186 unsigned int ndata, unsigned int odata)
1187 {
1188 unsigned long a, b, blocks;
1189 /* So how much do we need to backup.
1190 * We need an amount of data which is both a whole number of
1191 * old stripes and a whole number of new stripes.
1192 * So LCM for (chunksize*datadisks).
1193 */
1194 a = (ochunk/512) * odata;
1195 b = (nchunk/512) * ndata;
1196 /* Find GCD */
1197 a = GCD(a, b);
1198 /* LCM == product / GCD */
1199 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
1200
1201 return blocks;
1202 }
1203
1204 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
1205 {
1206 /* Based on the current array state in info->array and
1207 * the changes in info->new_* etc, determine:
1208 * - whether the change is possible
1209 * - Intermediate level/raid_disks/layout
1210 * - whether a restriping reshape is needed
1211 * - number of sectors in minimum change unit. This
1212 * will cover a whole number of stripes in 'before' and
1213 * 'after'.
1214 *
1215 * Return message if the change should be rejected
1216 * NULL if the change can be achieved
1217 *
1218 * This can be called as part of starting a reshape, or
1219 * when assembling an array that is undergoing reshape.
1220 */
1221 int near, far, offset, copies;
1222 int new_disks;
1223 int old_chunk, new_chunk;
1224 /* delta_parity records change in number of devices
1225 * caused by level change
1226 */
1227 int delta_parity = 0;
1228
1229 memset(re, 0, sizeof(*re));
1230
1231 /* If a new level not explicitly given, we assume no-change */
1232 if (info->new_level == UnSet)
1233 info->new_level = info->array.level;
1234
1235 if (info->new_chunk)
1236 switch (info->new_level) {
1237 case 0:
1238 case 4:
1239 case 5:
1240 case 6:
1241 case 10:
1242 /* chunk size is meaningful, must divide component_size
1243 * evenly
1244 */
1245 if (info->component_size % (info->new_chunk/512)) {
1246 unsigned long long shrink = info->component_size;
1247 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1248 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1249 info->new_chunk/1024, info->component_size/2);
1250 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1251 devname, shrink/2);
1252 pr_err("will shrink the array so the given chunk size would work.\n");
1253 return "";
1254 }
1255 break;
1256 default:
1257 return "chunk size not meaningful for this level";
1258 }
1259 else
1260 info->new_chunk = info->array.chunk_size;
1261
1262 switch (info->array.level) {
1263 default:
1264 return "No reshape is possibly for this RAID level";
1265 case LEVEL_LINEAR:
1266 if (info->delta_disks != UnSet)
1267 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1268 else
1269 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1270 case 1:
1271 /* RAID1 can convert to RAID1 with different disks, or
1272 * raid5 with 2 disks, or
1273 * raid0 with 1 disk
1274 */
1275 if (info->new_level > 1 && (info->component_size & 7))
1276 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1277 if (info->new_level == 0) {
1278 if (info->delta_disks != UnSet &&
1279 info->delta_disks != 0)
1280 return "Cannot change number of disks with RAID1->RAID0 conversion";
1281 re->level = 0;
1282 re->before.data_disks = 1;
1283 re->after.data_disks = 1;
1284 return NULL;
1285 }
1286 if (info->new_level == 1) {
1287 if (info->delta_disks == UnSet)
1288 /* Don't know what to do */
1289 return "no change requested for Growing RAID1";
1290 re->level = 1;
1291 return NULL;
1292 }
1293 if (info->array.raid_disks != 2 && info->new_level == 5)
1294 return "Can only convert a 2-device array to RAID5";
1295 if (info->array.raid_disks == 2 && info->new_level == 5) {
1296 re->level = 5;
1297 re->before.data_disks = 1;
1298 if (info->delta_disks != UnSet &&
1299 info->delta_disks != 0)
1300 re->after.data_disks = 1 + info->delta_disks;
1301 else
1302 re->after.data_disks = 1;
1303 if (re->after.data_disks < 1)
1304 return "Number of disks too small for RAID5";
1305
1306 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1307 info->array.chunk_size = 65536;
1308 break;
1309 }
1310 /* Could do some multi-stage conversions, but leave that to
1311 * later.
1312 */
1313 return "Impossibly level change request for RAID1";
1314
1315 case 10:
1316 /* RAID10 can be converted from near mode to
1317 * RAID0 by removing some devices.
1318 * It can also be reshaped if the kernel supports
1319 * new_data_offset.
1320 */
1321 switch (info->new_level) {
1322 case 0:
1323 if ((info->array.layout & ~0xff) != 0x100)
1324 return "Cannot Grow RAID10 with far/offset layout";
1325 /* number of devices must be multiple of number of copies */
1326 if (info->array.raid_disks % (info->array.layout & 0xff))
1327 return "RAID10 layout too complex for Grow operation";
1328
1329 new_disks = (info->array.raid_disks
1330 / (info->array.layout & 0xff));
1331 if (info->delta_disks == UnSet)
1332 info->delta_disks = (new_disks
1333 - info->array.raid_disks);
1334
1335 if (info->delta_disks != new_disks - info->array.raid_disks)
1336 return "New number of raid-devices impossible for RAID10";
1337 if (info->new_chunk &&
1338 info->new_chunk != info->array.chunk_size)
1339 return "Cannot change chunk-size with RAID10 Grow";
1340
1341 /* looks good */
1342 re->level = 0;
1343 re->before.data_disks = new_disks;
1344 re->after.data_disks = re->before.data_disks;
1345 return NULL;
1346
1347 case 10:
1348 near = info->array.layout & 0xff;
1349 far = (info->array.layout >> 8) & 0xff;
1350 offset = info->array.layout & 0x10000;
1351 if (far > 1 && !offset)
1352 return "Cannot reshape RAID10 in far-mode";
1353 copies = near * far;
1354
1355 old_chunk = info->array.chunk_size * far;
1356
1357 if (info->new_layout == UnSet)
1358 info->new_layout = info->array.layout;
1359 else {
1360 near = info->new_layout & 0xff;
1361 far = (info->new_layout >> 8) & 0xff;
1362 offset = info->new_layout & 0x10000;
1363 if (far > 1 && !offset)
1364 return "Cannot reshape RAID10 to far-mode";
1365 if (near * far != copies)
1366 return "Cannot change number of copies when reshaping RAID10";
1367 }
1368 if (info->delta_disks == UnSet)
1369 info->delta_disks = 0;
1370 new_disks = (info->array.raid_disks +
1371 info->delta_disks);
1372
1373 new_chunk = info->new_chunk * far;
1374
1375 re->level = 10;
1376 re->before.layout = info->array.layout;
1377 re->before.data_disks = info->array.raid_disks;
1378 re->after.layout = info->new_layout;
1379 re->after.data_disks = new_disks;
1380 /* For RAID10 we don't do backup but do allow reshape,
1381 * so set backup_blocks to INVALID_SECTORS rather than
1382 * zero.
1383 * And there is no need to synchronise stripes on both
1384 * 'old' and 'new'. So the important
1385 * number is the minimum data_offset difference
1386 * which is the larger of (offset copies * chunk).
1387 */
1388 re->backup_blocks = INVALID_SECTORS;
1389 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1390 if (new_disks < re->before.data_disks &&
1391 info->space_after < re->min_offset_change)
1392 /* Reduce component size by one chunk */
1393 re->new_size = (info->component_size -
1394 re->min_offset_change);
1395 else
1396 re->new_size = info->component_size;
1397 re->new_size = re->new_size * new_disks / copies;
1398 return NULL;
1399
1400 default:
1401 return "RAID10 can only be changed to RAID0";
1402 }
1403 case 0:
1404 /* RAID0 can be converted to RAID10, or to RAID456 */
1405 if (info->new_level == 10) {
1406 if (info->new_layout == UnSet &&
1407 info->delta_disks == UnSet) {
1408 /* Assume near=2 layout */
1409 info->new_layout = 0x102;
1410 info->delta_disks = info->array.raid_disks;
1411 }
1412 if (info->new_layout == UnSet) {
1413 int copies = 1 + (info->delta_disks
1414 / info->array.raid_disks);
1415 if (info->array.raid_disks * (copies-1)
1416 != info->delta_disks)
1417 return "Impossible number of devices for RAID0->RAID10";
1418 info->new_layout = 0x100 + copies;
1419 }
1420 if (info->delta_disks == UnSet) {
1421 int copies = info->new_layout & 0xff;
1422 if (info->new_layout != 0x100 + copies)
1423 return "New layout impossible for RAID0->RAID10";;
1424 info->delta_disks = (copies - 1) *
1425 info->array.raid_disks;
1426 }
1427 if (info->new_chunk &&
1428 info->new_chunk != info->array.chunk_size)
1429 return "Cannot change chunk-size with RAID0->RAID10";
1430 /* looks good */
1431 re->level = 10;
1432 re->before.data_disks = (info->array.raid_disks +
1433 info->delta_disks);
1434 re->after.data_disks = re->before.data_disks;
1435 re->before.layout = info->new_layout;
1436 return NULL;
1437 }
1438
1439 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1440 * a raid4 style layout of the final level.
1441 */
1442 switch (info->new_level) {
1443 case 4:
1444 delta_parity = 1;
1445 case 0:
1446 re->level = 4;
1447 re->before.layout = 0;
1448 break;
1449 case 5:
1450 delta_parity = 1;
1451 re->level = 5;
1452 re->before.layout = ALGORITHM_PARITY_N;
1453 if (info->new_layout == UnSet)
1454 info->new_layout = map_name(r5layout, "default");
1455 break;
1456 case 6:
1457 delta_parity = 2;
1458 re->level = 6;
1459 re->before.layout = ALGORITHM_PARITY_N;
1460 if (info->new_layout == UnSet)
1461 info->new_layout = map_name(r6layout, "default");
1462 break;
1463 default:
1464 return "Impossible level change requested";
1465 }
1466 re->before.data_disks = info->array.raid_disks;
1467 /* determining 'after' layout happens outside this 'switch' */
1468 break;
1469
1470 case 4:
1471 info->array.layout = ALGORITHM_PARITY_N;
1472 case 5:
1473 switch (info->new_level) {
1474 case 0:
1475 delta_parity = -1;
1476 case 4:
1477 re->level = info->array.level;
1478 re->before.data_disks = info->array.raid_disks - 1;
1479 re->before.layout = info->array.layout;
1480 break;
1481 case 5:
1482 re->level = 5;
1483 re->before.data_disks = info->array.raid_disks - 1;
1484 re->before.layout = info->array.layout;
1485 break;
1486 case 6:
1487 delta_parity = 1;
1488 re->level = 6;
1489 re->before.data_disks = info->array.raid_disks - 1;
1490 switch (info->array.layout) {
1491 case ALGORITHM_LEFT_ASYMMETRIC:
1492 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1493 break;
1494 case ALGORITHM_RIGHT_ASYMMETRIC:
1495 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1496 break;
1497 case ALGORITHM_LEFT_SYMMETRIC:
1498 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1499 break;
1500 case ALGORITHM_RIGHT_SYMMETRIC:
1501 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1502 break;
1503 case ALGORITHM_PARITY_0:
1504 re->before.layout = ALGORITHM_PARITY_0_6;
1505 break;
1506 case ALGORITHM_PARITY_N:
1507 re->before.layout = ALGORITHM_PARITY_N_6;
1508 break;
1509 default:
1510 return "Cannot convert an array with this layout";
1511 }
1512 break;
1513 case 1:
1514 if (info->array.raid_disks != 2)
1515 return "Can only convert a 2-device array to RAID1";
1516 if (info->delta_disks != UnSet &&
1517 info->delta_disks != 0)
1518 return "Cannot set raid_disk when converting RAID5->RAID1";
1519 re->level = 1;
1520 info->new_chunk = 0;
1521 return NULL;
1522 default:
1523 return "Impossible level change requested";
1524 }
1525 break;
1526 case 6:
1527 switch (info->new_level) {
1528 case 4:
1529 case 5:
1530 delta_parity = -1;
1531 case 6:
1532 re->level = 6;
1533 re->before.data_disks = info->array.raid_disks - 2;
1534 re->before.layout = info->array.layout;
1535 break;
1536 default:
1537 return "Impossible level change requested";
1538 }
1539 break;
1540 }
1541
1542 /* If we reached here then it looks like a re-stripe is
1543 * happening. We have determined the intermediate level
1544 * and initial raid_disks/layout and stored these in 're'.
1545 *
1546 * We need to deduce the final layout that can be atomically
1547 * converted to the end state.
1548 */
1549 switch (info->new_level) {
1550 case 0:
1551 /* We can only get to RAID0 from RAID4 or RAID5
1552 * with appropriate layout and one extra device
1553 */
1554 if (re->level != 4 && re->level != 5)
1555 return "Cannot covert to RAID0 from this level";
1556
1557 switch (re->level) {
1558 case 4:
1559 re->before.layout = 0;
1560 re->after.layout = 0;
1561 break;
1562 case 5:
1563 re->after.layout = ALGORITHM_PARITY_N;
1564 break;
1565 }
1566 break;
1567
1568 case 4:
1569 /* We can only get to RAID4 from RAID5 */
1570 if (re->level != 4 && re->level != 5)
1571 return "Cannot convert to RAID4 from this level";
1572
1573 switch (re->level) {
1574 case 4:
1575 re->after.layout = 0;
1576 break;
1577 case 5:
1578 re->after.layout = ALGORITHM_PARITY_N;
1579 break;
1580 }
1581 break;
1582
1583 case 5:
1584 /* We get to RAID5 from RAID5 or RAID6 */
1585 if (re->level != 5 && re->level != 6)
1586 return "Cannot convert to RAID5 from this level";
1587
1588 switch (re->level) {
1589 case 5:
1590 if (info->new_layout == UnSet)
1591 re->after.layout = re->before.layout;
1592 else
1593 re->after.layout = info->new_layout;
1594 break;
1595 case 6:
1596 if (info->new_layout == UnSet)
1597 info->new_layout = re->before.layout;
1598
1599 /* after.layout needs to be raid6 version of new_layout */
1600 if (info->new_layout == ALGORITHM_PARITY_N)
1601 re->after.layout = ALGORITHM_PARITY_N;
1602 else {
1603 char layout[40];
1604 char *ls = map_num(r5layout, info->new_layout);
1605 int l;
1606 if (ls) {
1607 /* Current RAID6 layout has a RAID5
1608 * equivalent - good
1609 */
1610 strcat(strcpy(layout, ls), "-6");
1611 l = map_name(r6layout, layout);
1612 if (l == UnSet)
1613 return "Cannot find RAID6 layout to convert to";
1614 } else {
1615 /* Current RAID6 has no equivalent.
1616 * If it is already a '-6' layout we
1617 * can leave it unchanged, else we must
1618 * fail
1619 */
1620 ls = map_num(r6layout, info->new_layout);
1621 if (!ls ||
1622 strcmp(ls+strlen(ls)-2, "-6") != 0)
1623 return "Please specify new layout";
1624 l = info->new_layout;
1625 }
1626 re->after.layout = l;
1627 }
1628 }
1629 break;
1630
1631 case 6:
1632 /* We must already be at level 6 */
1633 if (re->level != 6)
1634 return "Impossible level change";
1635 if (info->new_layout == UnSet)
1636 re->after.layout = info->array.layout;
1637 else
1638 re->after.layout = info->new_layout;
1639 break;
1640 default:
1641 return "Impossible level change requested";
1642 }
1643 if (info->delta_disks == UnSet)
1644 info->delta_disks = delta_parity;
1645
1646 re->after.data_disks =
1647 (re->before.data_disks + info->delta_disks - delta_parity);
1648
1649 switch (re->level) {
1650 case 6:
1651 re->parity = 2;
1652 break;
1653 case 4:
1654 case 5:
1655 re->parity = 1;
1656 break;
1657 default:
1658 re->parity = 0;
1659 break;
1660 }
1661 /* So we have a restripe operation, we need to calculate the number
1662 * of blocks per reshape operation.
1663 */
1664 re->new_size = info->component_size * re->before.data_disks;
1665 if (info->new_chunk == 0)
1666 info->new_chunk = info->array.chunk_size;
1667 if (re->after.data_disks == re->before.data_disks &&
1668 re->after.layout == re->before.layout &&
1669 info->new_chunk == info->array.chunk_size) {
1670 /* Nothing to change, can change level immediately. */
1671 re->level = info->new_level;
1672 re->backup_blocks = 0;
1673 return NULL;
1674 }
1675 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1676 /* chunk and layout changes make no difference */
1677 re->level = info->new_level;
1678 re->backup_blocks = 0;
1679 return NULL;
1680 }
1681
1682 if (re->after.data_disks == re->before.data_disks &&
1683 get_linux_version() < 2006032)
1684 return "in-place reshape is not safe before 2.6.32 - sorry.";
1685
1686 if (re->after.data_disks < re->before.data_disks &&
1687 get_linux_version() < 2006030)
1688 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1689
1690 re->backup_blocks = compute_backup_blocks(
1691 info->new_chunk, info->array.chunk_size,
1692 re->after.data_disks,
1693 re->before.data_disks);
1694 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1695
1696 re->new_size = info->component_size * re->after.data_disks;
1697 return NULL;
1698 }
1699
1700 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1701 char *text_version)
1702 {
1703 struct mdinfo *info;
1704 char *subarray;
1705 int ret_val = -1;
1706
1707 if ((st == NULL) || (sra == NULL))
1708 return ret_val;
1709
1710 if (text_version == NULL)
1711 text_version = sra->text_version;
1712 subarray = strchr(text_version + 1, '/')+1;
1713 info = st->ss->container_content(st, subarray);
1714 if (info) {
1715 unsigned long long current_size = 0;
1716 unsigned long long new_size =
1717 info->custom_array_size/2;
1718
1719 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1720 new_size > current_size) {
1721 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1722 < 0)
1723 dprintf("Error: Cannot set array size");
1724 else {
1725 ret_val = 0;
1726 dprintf("Array size changed");
1727 }
1728 dprintf_cont(" from %llu to %llu.\n",
1729 current_size, new_size);
1730 }
1731 sysfs_free(info);
1732 } else
1733 dprintf("Error: set_array_size(): info pointer in NULL\n");
1734
1735 return ret_val;
1736 }
1737
1738 static int reshape_array(char *container, int fd, char *devname,
1739 struct supertype *st, struct mdinfo *info,
1740 int force, struct mddev_dev *devlist,
1741 unsigned long long data_offset,
1742 char *backup_file, int verbose, int forked,
1743 int restart, int freeze_reshape);
1744 static int reshape_container(char *container, char *devname,
1745 int mdfd,
1746 struct supertype *st,
1747 struct mdinfo *info,
1748 int force,
1749 char *backup_file, int verbose,
1750 int forked, int restart, int freeze_reshape);
1751
1752 int Grow_reshape(char *devname, int fd,
1753 struct mddev_dev *devlist,
1754 unsigned long long data_offset,
1755 struct context *c, struct shape *s)
1756 {
1757 /* Make some changes in the shape of an array.
1758 * The kernel must support the change.
1759 *
1760 * There are three different changes. Each can trigger
1761 * a resync or recovery so we freeze that until we have
1762 * requested everything (if kernel supports freezing - 2.6.30).
1763 * The steps are:
1764 * - change size (i.e. component_size)
1765 * - change level
1766 * - change layout/chunksize/ndisks
1767 *
1768 * The last can require a reshape. It is different on different
1769 * levels so we need to check the level before actioning it.
1770 * Some times the level change needs to be requested after the
1771 * reshape (e.g. raid6->raid5, raid5->raid0)
1772 *
1773 */
1774 struct mdu_array_info_s array;
1775 int rv = 0;
1776 struct supertype *st;
1777 char *subarray = NULL;
1778
1779 int frozen;
1780 int changed = 0;
1781 char *container = NULL;
1782 int cfd = -1;
1783
1784 struct mddev_dev *dv;
1785 int added_disks;
1786
1787 struct mdinfo info;
1788 struct mdinfo *sra;
1789
1790 if (md_get_array_info(fd, &array) < 0) {
1791 pr_err("%s is not an active md array - aborting\n",
1792 devname);
1793 return 1;
1794 }
1795 if (data_offset != INVALID_SECTORS && array.level != 10 &&
1796 (array.level < 4 || array.level > 6)) {
1797 pr_err("--grow --data-offset not yet supported\n");
1798 return 1;
1799 }
1800
1801 if (s->size > 0 &&
1802 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1803 pr_err("cannot change component size at the same time as other changes.\n"
1804 " Change size first, then check data is intact before making other changes.\n");
1805 return 1;
1806 }
1807
1808 if (s->raiddisks && s->raiddisks < array.raid_disks &&
1809 array.level > 1 && get_linux_version() < 2006032 &&
1810 !check_env("MDADM_FORCE_FEWER")) {
1811 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1812 " Please use a newer kernel\n");
1813 return 1;
1814 }
1815
1816 st = super_by_fd(fd, &subarray);
1817 if (!st) {
1818 pr_err("Unable to determine metadata format for %s\n", devname);
1819 return 1;
1820 }
1821 if (s->raiddisks > st->max_devs) {
1822 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1823 return 1;
1824 }
1825 if (s->level == 0 &&
1826 (array.state & (1<<MD_SB_BITMAP_PRESENT)) &&
1827 !(array.state & (1<<MD_SB_CLUSTERED))) {
1828 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
1829 if (md_set_array_info(fd, &array)!= 0) {
1830 pr_err("failed to remove internal bitmap.\n");
1831 return 1;
1832 }
1833 }
1834
1835 /* in the external case we need to check that the requested reshape is
1836 * supported, and perform an initial check that the container holds the
1837 * pre-requisite spare devices (mdmon owns final validation)
1838 */
1839 if (st->ss->external) {
1840 int retval;
1841
1842 if (subarray) {
1843 container = st->container_devnm;
1844 cfd = open_dev_excl(st->container_devnm);
1845 } else {
1846 container = st->devnm;
1847 close(fd);
1848 cfd = open_dev_excl(st->devnm);
1849 fd = cfd;
1850 }
1851 if (cfd < 0) {
1852 pr_err("Unable to open container for %s\n",
1853 devname);
1854 free(subarray);
1855 return 1;
1856 }
1857
1858 retval = st->ss->load_container(st, cfd, NULL);
1859
1860 if (retval) {
1861 pr_err("Cannot read superblock for %s\n",
1862 devname);
1863 free(subarray);
1864 return 1;
1865 }
1866
1867 /* check if operation is supported for metadata handler */
1868 if (st->ss->container_content) {
1869 struct mdinfo *cc = NULL;
1870 struct mdinfo *content = NULL;
1871
1872 cc = st->ss->container_content(st, subarray);
1873 for (content = cc; content ; content = content->next) {
1874 int allow_reshape = 1;
1875
1876 /* check if reshape is allowed based on metadata
1877 * indications stored in content.array.status
1878 */
1879 if (content->array.state &
1880 (1 << MD_SB_BLOCK_VOLUME))
1881 allow_reshape = 0;
1882 if (content->array.state &
1883 (1 << MD_SB_BLOCK_CONTAINER_RESHAPE))
1884 allow_reshape = 0;
1885 if (!allow_reshape) {
1886 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1887 devname, container);
1888 sysfs_free(cc);
1889 free(subarray);
1890 return 1;
1891 }
1892 }
1893 sysfs_free(cc);
1894 }
1895 if (mdmon_running(container))
1896 st->update_tail = &st->updates;
1897 }
1898
1899 added_disks = 0;
1900 for (dv = devlist; dv; dv = dv->next)
1901 added_disks++;
1902 if (s->raiddisks > array.raid_disks &&
1903 array.spare_disks + added_disks < (s->raiddisks - array.raid_disks) &&
1904 !c->force) {
1905 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1906 " Use --force to over-ride this check.\n",
1907 s->raiddisks - array.raid_disks,
1908 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1909 array.spare_disks + added_disks);
1910 return 1;
1911 }
1912
1913 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS |
1914 GET_STATE | GET_VERSION);
1915 if (sra) {
1916 if (st->ss->external && subarray == NULL) {
1917 array.level = LEVEL_CONTAINER;
1918 sra->array.level = LEVEL_CONTAINER;
1919 }
1920 } else {
1921 pr_err("failed to read sysfs parameters for %s\n",
1922 devname);
1923 return 1;
1924 }
1925 frozen = freeze(st);
1926 if (frozen < -1) {
1927 /* freeze() already spewed the reason */
1928 sysfs_free(sra);
1929 return 1;
1930 } else if (frozen < 0) {
1931 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1932 sysfs_free(sra);
1933 return 1;
1934 }
1935
1936 /* ========= set size =============== */
1937 if (s->size > 0 &&
1938 (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1939 unsigned long long orig_size = get_component_size(fd)/2;
1940 unsigned long long min_csize;
1941 struct mdinfo *mdi;
1942 int raid0_takeover = 0;
1943
1944 if (orig_size == 0)
1945 orig_size = (unsigned) array.size;
1946
1947 if (orig_size == 0) {
1948 pr_err("Cannot set device size in this type of array.\n");
1949 rv = 1;
1950 goto release;
1951 }
1952
1953 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1954 devname, APPLY_METADATA_CHANGES,
1955 c->verbose > 0)) {
1956 rv = 1;
1957 goto release;
1958 }
1959 sync_metadata(st);
1960 if (st->ss->external) {
1961 /* metadata can have size limitation
1962 * update size value according to metadata information
1963 */
1964 struct mdinfo *sizeinfo =
1965 st->ss->container_content(st, subarray);
1966 if (sizeinfo) {
1967 unsigned long long new_size =
1968 sizeinfo->custom_array_size/2;
1969 int data_disks = get_data_disks(
1970 sizeinfo->array.level,
1971 sizeinfo->array.layout,
1972 sizeinfo->array.raid_disks);
1973 new_size /= data_disks;
1974 dprintf("Metadata size correction from %llu to %llu (%llu)\n",
1975 orig_size, new_size,
1976 new_size * data_disks);
1977 s->size = new_size;
1978 sysfs_free(sizeinfo);
1979 }
1980 }
1981
1982 /* Update the size of each member device in case
1983 * they have been resized. This will never reduce
1984 * below the current used-size. The "size" attribute
1985 * understands '0' to mean 'max'.
1986 */
1987 min_csize = 0;
1988 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1989 if (array.not_persistent == 0 &&
1990 array.major_version == 0 &&
1991 get_linux_version() < 3001000) {
1992 /* Dangerous to allow size to exceed 2TB */
1993 unsigned long long csize;
1994 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1995 if (csize >= 2ULL*1024*1024*1024)
1996 csize = 2ULL*1024*1024*1024;
1997 if ((min_csize == 0 || (min_csize
1998 > csize)))
1999 min_csize = csize;
2000 }
2001 }
2002 }
2003 if (min_csize && s->size > min_csize) {
2004 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
2005 rv = 1;
2006 goto size_change_error;
2007 }
2008 if (min_csize && s->size == MAX_SIZE) {
2009 /* Don't let the kernel choose a size - it will get
2010 * it wrong
2011 */
2012 pr_err("Limited v0.90 array to 2TB per device\n");
2013 s->size = min_csize;
2014 }
2015 if (st->ss->external) {
2016 if (sra->array.level == 0) {
2017 rv = sysfs_set_str(sra, NULL, "level",
2018 "raid5");
2019 if (!rv) {
2020 raid0_takeover = 1;
2021 /* get array parameters after takeover
2022 * to change one parameter at time only
2023 */
2024 rv = md_get_array_info(fd, &array);
2025 }
2026 }
2027 /* make sure mdmon is
2028 * aware of the new level */
2029 if (!mdmon_running(st->container_devnm))
2030 start_mdmon(st->container_devnm);
2031 ping_monitor(container);
2032 if (mdmon_running(st->container_devnm) &&
2033 st->update_tail == NULL)
2034 st->update_tail = &st->updates;
2035 }
2036
2037 if (s->size == MAX_SIZE)
2038 s->size = 0;
2039 array.size = s->size;
2040 if (s->size & ~INT32_MAX) {
2041 /* got truncated to 32bit, write to
2042 * component_size instead
2043 */
2044 if (sra)
2045 rv = sysfs_set_num(sra, NULL,
2046 "component_size", s->size);
2047 else
2048 rv = -1;
2049 } else {
2050 rv = md_set_array_info(fd, &array);
2051
2052 /* manage array size when it is managed externally
2053 */
2054 if ((rv == 0) && st->ss->external)
2055 rv = set_array_size(st, sra, sra->text_version);
2056 }
2057
2058 if (raid0_takeover) {
2059 /* do not recync non-existing parity,
2060 * we will drop it anyway
2061 */
2062 sysfs_set_str(sra, NULL, "sync_action", "frozen");
2063 /* go back to raid0, drop parity disk
2064 */
2065 sysfs_set_str(sra, NULL, "level", "raid0");
2066 md_get_array_info(fd, &array);
2067 }
2068
2069 size_change_error:
2070 if (rv != 0) {
2071 int err = errno;
2072
2073 /* restore metadata */
2074 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
2075 UnSet, NULL, devname,
2076 ROLLBACK_METADATA_CHANGES,
2077 c->verbose) == 0)
2078 sync_metadata(st);
2079 pr_err("Cannot set device size for %s: %s\n",
2080 devname, strerror(err));
2081 if (err == EBUSY &&
2082 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2083 cont_err("Bitmap must be removed before size can be changed\n");
2084 rv = 1;
2085 goto release;
2086 }
2087 if (s->assume_clean) {
2088 /* This will fail on kernels older than 3.0 unless
2089 * a backport has been arranged.
2090 */
2091 if (sra == NULL ||
2092 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
2093 pr_err("--assume-clean not supported with --grow on this kernel\n");
2094 }
2095 md_get_array_info(fd, &array);
2096 s->size = get_component_size(fd)/2;
2097 if (s->size == 0)
2098 s->size = array.size;
2099 if (c->verbose >= 0) {
2100 if (s->size == orig_size)
2101 pr_err("component size of %s unchanged at %lluK\n",
2102 devname, s->size);
2103 else
2104 pr_err("component size of %s has been set to %lluK\n",
2105 devname, s->size);
2106 }
2107 changed = 1;
2108 } else if (array.level != LEVEL_CONTAINER) {
2109 s->size = get_component_size(fd)/2;
2110 if (s->size == 0)
2111 s->size = array.size;
2112 }
2113
2114 /* See if there is anything else to do */
2115 if ((s->level == UnSet || s->level == array.level) &&
2116 (s->layout_str == NULL) &&
2117 (s->chunk == 0 || s->chunk == array.chunk_size) &&
2118 data_offset == INVALID_SECTORS &&
2119 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
2120 /* Nothing more to do */
2121 if (!changed && c->verbose >= 0)
2122 pr_err("%s: no change requested\n",
2123 devname);
2124 goto release;
2125 }
2126
2127 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
2128 * current implementation assumes that following conditions must be met:
2129 * - RAID10:
2130 * - far_copies == 1
2131 * - near_copies == 2
2132 */
2133 if ((s->level == 0 && array.level == 10 && sra &&
2134 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
2135 (s->level == 0 && array.level == 1 && sra)) {
2136 int err;
2137 err = remove_disks_for_takeover(st, sra, array.layout);
2138 if (err) {
2139 dprintf("Array cannot be reshaped\n");
2140 if (cfd > -1)
2141 close(cfd);
2142 rv = 1;
2143 goto release;
2144 }
2145 /* Make sure mdmon has seen the device removal
2146 * and updated metadata before we continue with
2147 * level change
2148 */
2149 if (container)
2150 ping_monitor(container);
2151 }
2152
2153 memset(&info, 0, sizeof(info));
2154 info.array = array;
2155 if (sysfs_init(&info, fd, NULL)) {
2156 pr_err("failed to intialize sysfs.\n");
2157 rv = 1;
2158 goto release;
2159 }
2160 strcpy(info.text_version, sra->text_version);
2161 info.component_size = s->size*2;
2162 info.new_level = s->level;
2163 info.new_chunk = s->chunk * 1024;
2164 if (info.array.level == LEVEL_CONTAINER) {
2165 info.delta_disks = UnSet;
2166 info.array.raid_disks = s->raiddisks;
2167 } else if (s->raiddisks)
2168 info.delta_disks = s->raiddisks - info.array.raid_disks;
2169 else
2170 info.delta_disks = UnSet;
2171 if (s->layout_str == NULL) {
2172 info.new_layout = UnSet;
2173 if (info.array.level == 6 &&
2174 (info.new_level == 6 || info.new_level == UnSet) &&
2175 info.array.layout >= 16) {
2176 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
2177 cont_err("during the reshape, please specify --layout=preserve\n");
2178 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2179 rv = 1;
2180 goto release;
2181 }
2182 } else if (strcmp(s->layout_str, "normalise") == 0 ||
2183 strcmp(s->layout_str, "normalize") == 0) {
2184 /* If we have a -6 RAID6 layout, remove the '-6'. */
2185 info.new_layout = UnSet;
2186 if (info.array.level == 6 && info.new_level == UnSet) {
2187 char l[40], *h;
2188 strcpy(l, map_num(r6layout, info.array.layout));
2189 h = strrchr(l, '-');
2190 if (h && strcmp(h, "-6") == 0) {
2191 *h = 0;
2192 info.new_layout = map_name(r6layout, l);
2193 }
2194 } else {
2195 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
2196 rv = 1;
2197 goto release;
2198 }
2199 } else if (strcmp(s->layout_str, "preserve") == 0) {
2200 /* This means that a non-standard RAID6 layout
2201 * is OK.
2202 * In particular:
2203 * - When reshape a RAID6 (e.g. adding a device)
2204 * which is in a non-standard layout, it is OK
2205 * to preserve that layout.
2206 * - When converting a RAID5 to RAID6, leave it in
2207 * the XXX-6 layout, don't re-layout.
2208 */
2209 if (info.array.level == 6 && info.new_level == UnSet)
2210 info.new_layout = info.array.layout;
2211 else if (info.array.level == 5 && info.new_level == 6) {
2212 char l[40];
2213 strcpy(l, map_num(r5layout, info.array.layout));
2214 strcat(l, "-6");
2215 info.new_layout = map_name(r6layout, l);
2216 } else {
2217 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
2218 rv = 1;
2219 goto release;
2220 }
2221 } else {
2222 int l = info.new_level;
2223 if (l == UnSet)
2224 l = info.array.level;
2225 switch (l) {
2226 case 5:
2227 info.new_layout = map_name(r5layout, s->layout_str);
2228 break;
2229 case 6:
2230 info.new_layout = map_name(r6layout, s->layout_str);
2231 break;
2232 case 10:
2233 info.new_layout = parse_layout_10(s->layout_str);
2234 break;
2235 case LEVEL_FAULTY:
2236 info.new_layout = parse_layout_faulty(s->layout_str);
2237 break;
2238 default:
2239 pr_err("layout not meaningful with this level\n");
2240 rv = 1;
2241 goto release;
2242 }
2243 if (info.new_layout == UnSet) {
2244 pr_err("layout %s not understood for this level\n",
2245 s->layout_str);
2246 rv = 1;
2247 goto release;
2248 }
2249 }
2250
2251 if (array.level == LEVEL_FAULTY) {
2252 if (s->level != UnSet && s->level != array.level) {
2253 pr_err("cannot change level of Faulty device\n");
2254 rv =1 ;
2255 }
2256 if (s->chunk) {
2257 pr_err("cannot set chunksize of Faulty device\n");
2258 rv =1 ;
2259 }
2260 if (s->raiddisks && s->raiddisks != 1) {
2261 pr_err("cannot set raid_disks of Faulty device\n");
2262 rv =1 ;
2263 }
2264 if (s->layout_str) {
2265 if (md_get_array_info(fd, &array) != 0) {
2266 dprintf("Cannot get array information.\n");
2267 goto release;
2268 }
2269 array.layout = info.new_layout;
2270 if (md_set_array_info(fd, &array) != 0) {
2271 pr_err("failed to set new layout\n");
2272 rv = 1;
2273 } else if (c->verbose >= 0)
2274 printf("layout for %s set to %d\n",
2275 devname, array.layout);
2276 }
2277 } else if (array.level == LEVEL_CONTAINER) {
2278 /* This change is to be applied to every array in the
2279 * container. This is only needed when the metadata imposes
2280 * restraints of the various arrays in the container.
2281 * Currently we only know that IMSM requires all arrays
2282 * to have the same number of devices so changing the
2283 * number of devices (On-Line Capacity Expansion) must be
2284 * performed at the level of the container
2285 */
2286 if (fd > 0) {
2287 close(fd);
2288 fd = -1;
2289 }
2290 rv = reshape_container(container, devname, -1, st, &info,
2291 c->force, c->backup_file, c->verbose, 0, 0, 0);
2292 frozen = 0;
2293 } else {
2294 /* get spare devices from external metadata
2295 */
2296 if (st->ss->external) {
2297 struct mdinfo *info2;
2298
2299 info2 = st->ss->container_content(st, subarray);
2300 if (info2) {
2301 info.array.spare_disks =
2302 info2->array.spare_disks;
2303 sysfs_free(info2);
2304 }
2305 }
2306
2307 /* Impose these changes on a single array. First
2308 * check that the metadata is OK with the change. */
2309
2310 if (reshape_super(st, 0, info.new_level,
2311 info.new_layout, info.new_chunk,
2312 info.array.raid_disks, info.delta_disks,
2313 c->backup_file, devname, APPLY_METADATA_CHANGES,
2314 c->verbose)) {
2315 rv = 1;
2316 goto release;
2317 }
2318 sync_metadata(st);
2319 rv = reshape_array(container, fd, devname, st, &info, c->force,
2320 devlist, data_offset, c->backup_file, c->verbose,
2321 0, 0, 0);
2322 frozen = 0;
2323 }
2324 release:
2325 sysfs_free(sra);
2326 if (frozen > 0)
2327 unfreeze(st);
2328 return rv;
2329 }
2330
2331 /* verify_reshape_position()
2332 * Function checks if reshape position in metadata is not farther
2333 * than position in md.
2334 * Return value:
2335 * 0 : not valid sysfs entry
2336 * it can be caused by not started reshape, it should be started
2337 * by reshape array or raid0 array is before takeover
2338 * -1 : error, reshape position is obviously wrong
2339 * 1 : success, reshape progress correct or updated
2340 */
2341 static int verify_reshape_position(struct mdinfo *info, int level)
2342 {
2343 int ret_val = 0;
2344 char buf[40];
2345 int rv;
2346
2347 /* read sync_max, failure can mean raid0 array */
2348 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2349
2350 if (rv > 0) {
2351 char *ep;
2352 unsigned long long position = strtoull(buf, &ep, 0);
2353
2354 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2355 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2356 position *= get_data_disks(level,
2357 info->new_layout,
2358 info->array.raid_disks);
2359 if (info->reshape_progress < position) {
2360 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2361 info->reshape_progress, position);
2362 info->reshape_progress = position;
2363 ret_val = 1;
2364 } else if (info->reshape_progress > position) {
2365 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2366 position, info->reshape_progress);
2367 ret_val = -1;
2368 } else {
2369 dprintf("Reshape position in md and metadata are the same;");
2370 ret_val = 1;
2371 }
2372 }
2373 } else if (rv == 0) {
2374 /* for valid sysfs entry, 0-length content
2375 * should be indicated as error
2376 */
2377 ret_val = -1;
2378 }
2379
2380 return ret_val;
2381 }
2382
2383 static unsigned long long choose_offset(unsigned long long lo,
2384 unsigned long long hi,
2385 unsigned long long min,
2386 unsigned long long max)
2387 {
2388 /* Choose a new offset between hi and lo.
2389 * It must be between min and max, but
2390 * we would prefer something near the middle of hi/lo, and also
2391 * prefer to be aligned to a big power of 2.
2392 *
2393 * So we start with the middle, then for each bit,
2394 * starting at '1' and increasing, if it is set, we either
2395 * add it or subtract it if possible, preferring the option
2396 * which is furthest from the boundary.
2397 *
2398 * We stop once we get a 1MB alignment. As units are in sectors,
2399 * 1MB = 2*1024 sectors.
2400 */
2401 unsigned long long choice = (lo + hi) / 2;
2402 unsigned long long bit = 1;
2403
2404 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2405 unsigned long long bigger, smaller;
2406 if (! (bit & choice))
2407 continue;
2408 bigger = choice + bit;
2409 smaller = choice - bit;
2410 if (bigger > max && smaller < min)
2411 break;
2412 if (bigger > max)
2413 choice = smaller;
2414 else if (smaller < min)
2415 choice = bigger;
2416 else if (hi - bigger > smaller - lo)
2417 choice = bigger;
2418 else
2419 choice = smaller;
2420 }
2421 return choice;
2422 }
2423
2424 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2425 char *devname, int delta_disks,
2426 unsigned long long data_offset,
2427 unsigned long long min,
2428 int can_fallback)
2429 {
2430 struct mdinfo *sd;
2431 int dir = 0;
2432 int err = 0;
2433 unsigned long long before, after;
2434
2435 /* Need to find min space before and after so same is used
2436 * on all devices
2437 */
2438 before = UINT64_MAX;
2439 after = UINT64_MAX;
2440 for (sd = sra->devs; sd; sd = sd->next) {
2441 char *dn;
2442 int dfd;
2443 int rv;
2444 struct supertype *st2;
2445 struct mdinfo info2;
2446
2447 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2448 continue;
2449 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2450 dfd = dev_open(dn, O_RDONLY);
2451 if (dfd < 0) {
2452 pr_err("%s: cannot open component %s\n",
2453 devname, dn ? dn : "-unknown-");
2454 goto release;
2455 }
2456 st2 = dup_super(st);
2457 rv = st2->ss->load_super(st2,dfd, NULL);
2458 close(dfd);
2459 if (rv) {
2460 free(st2);
2461 pr_err("%s: cannot get superblock from %s\n",
2462 devname, dn);
2463 goto release;
2464 }
2465 st2->ss->getinfo_super(st2, &info2, NULL);
2466 st2->ss->free_super(st2);
2467 free(st2);
2468 if (info2.space_before == 0 &&
2469 info2.space_after == 0) {
2470 /* Metadata doesn't support data_offset changes */
2471 if (!can_fallback)
2472 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2473 devname);
2474 goto fallback;
2475 }
2476 if (before > info2.space_before)
2477 before = info2.space_before;
2478 if (after > info2.space_after)
2479 after = info2.space_after;
2480
2481 if (data_offset != INVALID_SECTORS) {
2482 if (dir == 0) {
2483 if (info2.data_offset == data_offset) {
2484 pr_err("%s: already has that data_offset\n",
2485 dn);
2486 goto release;
2487 }
2488 if (data_offset < info2.data_offset)
2489 dir = -1;
2490 else
2491 dir = 1;
2492 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2493 (data_offset >= info2.data_offset && dir == -1)) {
2494 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2495 dn);
2496 goto release;
2497 }
2498 }
2499 }
2500 if (before == UINT64_MAX)
2501 /* impossible really, there must be no devices */
2502 return 1;
2503
2504 for (sd = sra->devs; sd; sd = sd->next) {
2505 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2506 unsigned long long new_data_offset;
2507
2508 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2509 continue;
2510 if (delta_disks < 0) {
2511 /* Don't need any space as array is shrinking
2512 * just move data_offset up by min
2513 */
2514 if (data_offset == INVALID_SECTORS)
2515 new_data_offset = sd->data_offset + min;
2516 else {
2517 if (data_offset < sd->data_offset + min) {
2518 pr_err("--data-offset too small for %s\n",
2519 dn);
2520 goto release;
2521 }
2522 new_data_offset = data_offset;
2523 }
2524 } else if (delta_disks > 0) {
2525 /* need space before */
2526 if (before < min) {
2527 if (can_fallback)
2528 goto fallback;
2529 pr_err("Insufficient head-space for reshape on %s\n",
2530 dn);
2531 goto release;
2532 }
2533 if (data_offset == INVALID_SECTORS)
2534 new_data_offset = sd->data_offset - min;
2535 else {
2536 if (data_offset > sd->data_offset - min) {
2537 pr_err("--data-offset too large for %s\n",
2538 dn);
2539 goto release;
2540 }
2541 new_data_offset = data_offset;
2542 }
2543 } else {
2544 if (dir == 0) {
2545 /* can move up or down. If 'data_offset'
2546 * was set we would have already decided,
2547 * so just choose direction with most space.
2548 */
2549 if (before > after)
2550 dir = -1;
2551 else
2552 dir = 1;
2553 }
2554 sysfs_set_str(sra, NULL, "reshape_direction",
2555 dir == 1 ? "backwards" : "forwards");
2556 if (dir > 0) {
2557 /* Increase data offset */
2558 if (after < min) {
2559 if (can_fallback)
2560 goto fallback;
2561 pr_err("Insufficient tail-space for reshape on %s\n",
2562 dn);
2563 goto release;
2564 }
2565 if (data_offset != INVALID_SECTORS &&
2566 data_offset < sd->data_offset + min) {
2567 pr_err("--data-offset too small on %s\n",
2568 dn);
2569 goto release;
2570 }
2571 if (data_offset != INVALID_SECTORS)
2572 new_data_offset = data_offset;
2573 else
2574 new_data_offset = choose_offset(sd->data_offset,
2575 sd->data_offset + after,
2576 sd->data_offset + min,
2577 sd->data_offset + after);
2578 } else {
2579 /* Decrease data offset */
2580 if (before < min) {
2581 if (can_fallback)
2582 goto fallback;
2583 pr_err("insufficient head-room on %s\n",
2584 dn);
2585 goto release;
2586 }
2587 if (data_offset != INVALID_SECTORS &&
2588 data_offset < sd->data_offset - min) {
2589 pr_err("--data-offset too small on %s\n",
2590 dn);
2591 goto release;
2592 }
2593 if (data_offset != INVALID_SECTORS)
2594 new_data_offset = data_offset;
2595 else
2596 new_data_offset = choose_offset(sd->data_offset - before,
2597 sd->data_offset,
2598 sd->data_offset - before,
2599 sd->data_offset - min);
2600 }
2601 }
2602 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2603 if (err < 0 && errno == E2BIG) {
2604 /* try again after increasing data size to max */
2605 err = sysfs_set_num(sra, sd, "size", 0);
2606 if (err < 0 && errno == EINVAL &&
2607 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2608 /* some kernels have a bug where you cannot
2609 * use '0' on spare devices. */
2610 sysfs_set_num(sra, sd, "size",
2611 (sra->component_size + after)/2);
2612 }
2613 err = sysfs_set_num(sra, sd, "new_offset",
2614 new_data_offset);
2615 }
2616 if (err < 0) {
2617 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2618 pr_err("data-offset is too big for %s\n",
2619 dn);
2620 goto release;
2621 }
2622 if (sd == sra->devs &&
2623 (errno == ENOENT || errno == E2BIG))
2624 /* Early kernel, no 'new_offset' file,
2625 * or kernel doesn't like us.
2626 * For RAID5/6 this is not fatal
2627 */
2628 return 1;
2629 pr_err("Cannot set new_offset for %s\n",
2630 dn);
2631 break;
2632 }
2633 }
2634 return err;
2635 release:
2636 return -1;
2637 fallback:
2638 /* Just use a backup file */
2639 return 1;
2640 }
2641
2642 static int raid10_reshape(char *container, int fd, char *devname,
2643 struct supertype *st, struct mdinfo *info,
2644 struct reshape *reshape,
2645 unsigned long long data_offset,
2646 int force, int verbose)
2647 {
2648 /* Changing raid_disks, layout, chunksize or possibly
2649 * just data_offset for a RAID10.
2650 * We must always change data_offset. We change by at least
2651 * ->min_offset_change which is the largest of the old and new
2652 * chunk sizes.
2653 * If raid_disks is increasing, then data_offset must decrease
2654 * by at least this copy size.
2655 * If raid_disks is unchanged, data_offset must increase or
2656 * decrease by at least min_offset_change but preferably by much more.
2657 * We choose half of the available space.
2658 * If raid_disks is decreasing, data_offset must increase by
2659 * at least min_offset_change. To allow of this, component_size
2660 * must be decreased by the same amount.
2661 *
2662 * So we calculate the required minimum and direction, possibly
2663 * reduce the component_size, then iterate through the devices
2664 * and set the new_data_offset.
2665 * If that all works, we set chunk_size, layout, raid_disks, and start
2666 * 'reshape'
2667 */
2668 struct mdinfo *sra;
2669 unsigned long long min;
2670 int err = 0;
2671
2672 sra = sysfs_read(fd, NULL,
2673 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2674 );
2675 if (!sra) {
2676 pr_err("%s: Cannot get array details from sysfs\n",
2677 devname);
2678 goto release;
2679 }
2680 min = reshape->min_offset_change;
2681
2682 if (info->delta_disks)
2683 sysfs_set_str(sra, NULL, "reshape_direction",
2684 info->delta_disks < 0 ? "backwards" : "forwards");
2685 if (info->delta_disks < 0 &&
2686 info->space_after < min) {
2687 int rv = sysfs_set_num(sra, NULL, "component_size",
2688 (sra->component_size -
2689 min)/2);
2690 if (rv) {
2691 pr_err("cannot reduce component size\n");
2692 goto release;
2693 }
2694 }
2695 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2696 min, 0);
2697 if (err == 1) {
2698 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2699 cont_err("supported on this kernel\n");
2700 err = -1;
2701 }
2702 if (err < 0)
2703 goto release;
2704
2705 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2706 err = errno;
2707 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2708 err = errno;
2709 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2710 info->array.raid_disks + info->delta_disks) < 0)
2711 err = errno;
2712 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2713 err = errno;
2714 if (err) {
2715 pr_err("Cannot set array shape for %s\n",
2716 devname);
2717 if (err == EBUSY &&
2718 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2719 cont_err(" Bitmap must be removed before shape can be changed\n");
2720 goto release;
2721 }
2722 sysfs_free(sra);
2723 return 0;
2724 release:
2725 sysfs_free(sra);
2726 return 1;
2727 }
2728
2729 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2730 {
2731 struct mdinfo *sra, *sd;
2732 /* Initialisation to silence compiler warning */
2733 unsigned long long min_space_before = 0, min_space_after = 0;
2734 int first = 1;
2735
2736 sra = sysfs_read(fd, NULL, GET_DEVS);
2737 if (!sra)
2738 return;
2739 for (sd = sra->devs; sd; sd = sd->next) {
2740 char *dn;
2741 int dfd;
2742 struct supertype *st2;
2743 struct mdinfo info2;
2744
2745 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2746 continue;
2747 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2748 dfd = dev_open(dn, O_RDONLY);
2749 if (dfd < 0)
2750 break;
2751 st2 = dup_super(st);
2752 if (st2->ss->load_super(st2,dfd, NULL)) {
2753 close(dfd);
2754 free(st2);
2755 break;
2756 }
2757 close(dfd);
2758 st2->ss->getinfo_super(st2, &info2, NULL);
2759 st2->ss->free_super(st2);
2760 free(st2);
2761 if (first ||
2762 min_space_before > info2.space_before)
2763 min_space_before = info2.space_before;
2764 if (first ||
2765 min_space_after > info2.space_after)
2766 min_space_after = info2.space_after;
2767 first = 0;
2768 }
2769 if (sd == NULL && !first) {
2770 info->space_after = min_space_after;
2771 info->space_before = min_space_before;
2772 }
2773 sysfs_free(sra);
2774 }
2775
2776 static void update_cache_size(char *container, struct mdinfo *sra,
2777 struct mdinfo *info,
2778 int disks, unsigned long long blocks)
2779 {
2780 /* Check that the internal stripe cache is
2781 * large enough, or it won't work.
2782 * It must hold at least 4 stripes of the larger
2783 * chunk size
2784 */
2785 unsigned long cache;
2786 cache = max(info->array.chunk_size, info->new_chunk);
2787 cache *= 4; /* 4 stripes minimum */
2788 cache /= 512; /* convert to sectors */
2789 /* make sure there is room for 'blocks' with a bit to spare */
2790 if (cache < 16 + blocks / disks)
2791 cache = 16 + blocks / disks;
2792 cache /= (4096/512); /* Convert from sectors to pages */
2793
2794 if (sra->cache_size < cache)
2795 subarray_set_num(container, sra, "stripe_cache_size",
2796 cache+1);
2797 }
2798
2799 static int impose_reshape(struct mdinfo *sra,
2800 struct mdinfo *info,
2801 struct supertype *st,
2802 int fd,
2803 int restart,
2804 char *devname, char *container,
2805 struct reshape *reshape)
2806 {
2807 struct mdu_array_info_s array;
2808
2809 sra->new_chunk = info->new_chunk;
2810
2811 if (restart) {
2812 /* for external metadata checkpoint saved by mdmon can be lost
2813 * or missed /due to e.g. crash/. Check if md is not during
2814 * restart farther than metadata points to.
2815 * If so, this means metadata information is obsolete.
2816 */
2817 if (st->ss->external)
2818 verify_reshape_position(info, reshape->level);
2819 sra->reshape_progress = info->reshape_progress;
2820 } else {
2821 sra->reshape_progress = 0;
2822 if (reshape->after.data_disks < reshape->before.data_disks)
2823 /* start from the end of the new array */
2824 sra->reshape_progress = (sra->component_size
2825 * reshape->after.data_disks);
2826 }
2827
2828 md_get_array_info(fd, &array);
2829 if (info->array.chunk_size == info->new_chunk &&
2830 reshape->before.layout == reshape->after.layout &&
2831 st->ss->external == 0) {
2832 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2833 array.raid_disks = reshape->after.data_disks + reshape->parity;
2834 if (!restart && md_set_array_info(fd, &array) != 0) {
2835 int err = errno;
2836
2837 pr_err("Cannot set device shape for %s: %s\n",
2838 devname, strerror(errno));
2839
2840 if (err == EBUSY &&
2841 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2842 cont_err("Bitmap must be removed before shape can be changed\n");
2843
2844 goto release;
2845 }
2846 } else if (!restart) {
2847 /* set them all just in case some old 'new_*' value
2848 * persists from some earlier problem.
2849 */
2850 int err = 0;
2851 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2852 err = errno;
2853 if (!err && sysfs_set_num(sra, NULL, "layout",
2854 reshape->after.layout) < 0)
2855 err = errno;
2856 if (!err && subarray_set_num(container, sra, "raid_disks",
2857 reshape->after.data_disks +
2858 reshape->parity) < 0)
2859 err = errno;
2860 if (err) {
2861 pr_err("Cannot set device shape for %s\n",
2862 devname);
2863
2864 if (err == EBUSY &&
2865 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2866 cont_err("Bitmap must be removed before shape can be changed\n");
2867 goto release;
2868 }
2869 }
2870 return 0;
2871 release:
2872 return -1;
2873 }
2874
2875 static int impose_level(int fd, int level, char *devname, int verbose)
2876 {
2877 char *c;
2878 struct mdu_array_info_s array;
2879 struct mdinfo info;
2880
2881 if (sysfs_init(&info, fd, NULL)) {
2882 pr_err("failed to intialize sysfs.\n");
2883 return 1;
2884 }
2885
2886 md_get_array_info(fd, &array);
2887 if (level == 0 &&
2888 (array.level >= 4 && array.level <= 6)) {
2889 /* To convert to RAID0 we need to fail and
2890 * remove any non-data devices. */
2891 int found = 0;
2892 int d;
2893 int data_disks = array.raid_disks - 1;
2894 if (array.level == 6)
2895 data_disks -= 1;
2896 if (array.level == 5 &&
2897 array.layout != ALGORITHM_PARITY_N)
2898 return -1;
2899 if (array.level == 6 &&
2900 array.layout != ALGORITHM_PARITY_N_6)
2901 return -1;
2902 sysfs_set_str(&info, NULL,"sync_action", "idle");
2903 /* First remove any spares so no recovery starts */
2904 for (d = 0, found = 0;
2905 d < MAX_DISKS && found < array.nr_disks;
2906 d++) {
2907 mdu_disk_info_t disk;
2908 disk.number = d;
2909 if (md_get_disk_info(fd, &disk) < 0)
2910 continue;
2911 if (disk.major == 0 && disk.minor == 0)
2912 continue;
2913 found++;
2914 if ((disk.state & (1 << MD_DISK_ACTIVE))
2915 && disk.raid_disk < data_disks)
2916 /* keep this */
2917 continue;
2918 ioctl(fd, HOT_REMOVE_DISK,
2919 makedev(disk.major, disk.minor));
2920 }
2921 /* Now fail anything left */
2922 md_get_array_info(fd, &array);
2923 for (d = 0, found = 0;
2924 d < MAX_DISKS && found < array.nr_disks;
2925 d++) {
2926 mdu_disk_info_t disk;
2927 disk.number = d;
2928 if (md_get_disk_info(fd, &disk) < 0)
2929 continue;
2930 if (disk.major == 0 && disk.minor == 0)
2931 continue;
2932 found++;
2933 if ((disk.state & (1 << MD_DISK_ACTIVE))
2934 && disk.raid_disk < data_disks)
2935 /* keep this */
2936 continue;
2937 ioctl(fd, SET_DISK_FAULTY,
2938 makedev(disk.major, disk.minor));
2939 hot_remove_disk(fd, makedev(disk.major, disk.minor), 1);
2940 }
2941 }
2942 c = map_num(pers, level);
2943 if (c) {
2944 int err = sysfs_set_str(&info, NULL, "level", c);
2945 if (err) {
2946 err = errno;
2947 pr_err("%s: could not set level to %s\n",
2948 devname, c);
2949 if (err == EBUSY &&
2950 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2951 cont_err("Bitmap must be removed before level can be changed\n");
2952 return err;
2953 }
2954 if (verbose >= 0)
2955 pr_err("level of %s changed to %s\n",
2956 devname, c);
2957 }
2958 return 0;
2959 }
2960
2961 int sigterm = 0;
2962 static void catch_term(int sig)
2963 {
2964 sigterm = 1;
2965 }
2966
2967 static int continue_via_systemd(char *devnm)
2968 {
2969 int skipped, i, pid, status;
2970 char pathbuf[1024];
2971 /* In a systemd/udev world, it is best to get systemd to
2972 * run "mdadm --grow --continue" rather than running in the
2973 * background.
2974 */
2975 switch(fork()) {
2976 case 0:
2977 /* FIXME yuk. CLOSE_EXEC?? */
2978 skipped = 0;
2979 for (i = 3; skipped < 20; i++)
2980 if (close(i) < 0)
2981 skipped++;
2982 else
2983 skipped = 0;
2984
2985 /* Don't want to see error messages from
2986 * systemctl. If the service doesn't exist,
2987 * we fork ourselves.
2988 */
2989 close(2);
2990 open("/dev/null", O_WRONLY);
2991 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2992 devnm);
2993 status = execl("/usr/bin/systemctl", "systemctl",
2994 "start",
2995 pathbuf, NULL);
2996 status = execl("/bin/systemctl", "systemctl", "start",
2997 pathbuf, NULL);
2998 exit(1);
2999 case -1: /* Just do it ourselves. */
3000 break;
3001 default: /* parent - good */
3002 pid = wait(&status);
3003 if (pid >= 0 && status == 0)
3004 return 1;
3005 }
3006 return 0;
3007 }
3008
3009 static int reshape_array(char *container, int fd, char *devname,
3010 struct supertype *st, struct mdinfo *info,
3011 int force, struct mddev_dev *devlist,
3012 unsigned long long data_offset,
3013 char *backup_file, int verbose, int forked,
3014 int restart, int freeze_reshape)
3015 {
3016 struct reshape reshape;
3017 int spares_needed;
3018 char *msg;
3019 int orig_level = UnSet;
3020 int odisks;
3021 int delayed;
3022
3023 struct mdu_array_info_s array;
3024 char *c;
3025
3026 struct mddev_dev *dv;
3027 int added_disks;
3028
3029 int *fdlist = NULL;
3030 unsigned long long *offsets = NULL;
3031 int d;
3032 int nrdisks;
3033 int err;
3034 unsigned long blocks;
3035 unsigned long long array_size;
3036 int done;
3037 struct mdinfo *sra = NULL;
3038 char buf[20];
3039
3040 /* when reshaping a RAID0, the component_size might be zero.
3041 * So try to fix that up.
3042 */
3043 if (md_get_array_info(fd, &array) != 0) {
3044 dprintf("Cannot get array information.\n");
3045 goto release;
3046 }
3047 if (array.level == 0 && info->component_size == 0) {
3048 get_dev_size(fd, NULL, &array_size);
3049 info->component_size = array_size / array.raid_disks;
3050 }
3051
3052 if (array.level == 10)
3053 /* Need space_after info */
3054 get_space_after(fd, st, info);
3055
3056 if (info->reshape_active) {
3057 int new_level = info->new_level;
3058 info->new_level = UnSet;
3059 if (info->delta_disks > 0)
3060 info->array.raid_disks -= info->delta_disks;
3061 msg = analyse_change(devname, info, &reshape);
3062 info->new_level = new_level;
3063 if (info->delta_disks > 0)
3064 info->array.raid_disks += info->delta_disks;
3065 if (!restart)
3066 /* Make sure the array isn't read-only */
3067 ioctl(fd, RESTART_ARRAY_RW, 0);
3068 } else
3069 msg = analyse_change(devname, info, &reshape);
3070 if (msg) {
3071 /* if msg == "", error has already been printed */
3072 if (msg[0])
3073 pr_err("%s\n", msg);
3074 goto release;
3075 }
3076 if (restart &&
3077 (reshape.level != info->array.level ||
3078 reshape.before.layout != info->array.layout ||
3079 reshape.before.data_disks + reshape.parity
3080 != info->array.raid_disks - max(0, info->delta_disks))) {
3081 pr_err("reshape info is not in native format - cannot continue.\n");
3082 goto release;
3083 }
3084
3085 if (st->ss->external && restart && (info->reshape_progress == 0) &&
3086 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
3087 (strncmp(buf, "reshape", 7) == 0))) {
3088 /* When reshape is restarted from '0', very begin of array
3089 * it is possible that for external metadata reshape and array
3090 * configuration doesn't happen.
3091 * Check if md has the same opinion, and reshape is restarted
3092 * from 0. If so, this is regular reshape start after reshape
3093 * switch in metadata to next array only.
3094 */
3095 if ((verify_reshape_position(info, reshape.level) >= 0) &&
3096 (info->reshape_progress == 0))
3097 restart = 0;
3098 }
3099 if (restart) {
3100 /* reshape already started. just skip to monitoring the reshape */
3101 if (reshape.backup_blocks == 0)
3102 return 0;
3103 if (restart & RESHAPE_NO_BACKUP)
3104 return 0;
3105
3106 /* Need 'sra' down at 'started:' */
3107 sra = sysfs_read(fd, NULL,
3108 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3109 GET_CACHE);
3110 if (!sra) {
3111 pr_err("%s: Cannot get array details from sysfs\n",
3112 devname);
3113 goto release;
3114 }
3115
3116 if (!backup_file)
3117 backup_file = locate_backup(sra->sys_name);
3118
3119 goto started;
3120 }
3121 /* The container is frozen but the array may not be.
3122 * So freeze the array so spares don't get put to the wrong use
3123 * FIXME there should probably be a cleaner separation between
3124 * freeze_array and freeze_container.
3125 */
3126 sysfs_freeze_array(info);
3127 /* Check we have enough spares to not be degraded */
3128 added_disks = 0;
3129 for (dv = devlist; dv ; dv=dv->next)
3130 added_disks++;
3131 spares_needed = max(reshape.before.data_disks,
3132 reshape.after.data_disks)
3133 + reshape.parity - array.raid_disks;
3134
3135 if (!force &&
3136 info->new_level > 1 && info->array.level > 1 &&
3137 spares_needed > info->array.spare_disks + added_disks) {
3138 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
3139 " Use --force to over-ride this check.\n",
3140 spares_needed,
3141 spares_needed == 1 ? "" : "s",
3142 info->array.spare_disks + added_disks);
3143 goto release;
3144 }
3145 /* Check we have enough spares to not fail */
3146 spares_needed = max(reshape.before.data_disks,
3147 reshape.after.data_disks)
3148 - array.raid_disks;
3149 if ((info->new_level > 1 || info->new_level == 0) &&
3150 spares_needed > info->array.spare_disks +added_disks) {
3151 pr_err("Need %d spare%s to create working array, and only have %d.\n",
3152 spares_needed,
3153 spares_needed == 1 ? "" : "s",
3154 info->array.spare_disks + added_disks);
3155 goto release;
3156 }
3157
3158 if (reshape.level != array.level) {
3159 int err = impose_level(fd, reshape.level, devname, verbose);
3160 if (err)
3161 goto release;
3162 info->new_layout = UnSet; /* after level change,
3163 * layout is meaningless */
3164 orig_level = array.level;
3165 sysfs_freeze_array(info);
3166
3167 if (reshape.level > 0 && st->ss->external) {
3168 /* make sure mdmon is aware of the new level */
3169 if (mdmon_running(container))
3170 flush_mdmon(container);
3171
3172 if (!mdmon_running(container))
3173 start_mdmon(container);
3174 ping_monitor(container);
3175 if (mdmon_running(container) &&
3176 st->update_tail == NULL)
3177 st->update_tail = &st->updates;
3178 }
3179 }
3180 /* ->reshape_super might have chosen some spares from the
3181 * container that it wants to be part of the new array.
3182 * We can collect them with ->container_content and give
3183 * them to the kernel.
3184 */
3185 if (st->ss->reshape_super && st->ss->container_content) {
3186 char *subarray = strchr(info->text_version+1, '/')+1;
3187 struct mdinfo *info2 =
3188 st->ss->container_content(st, subarray);
3189 struct mdinfo *d;
3190
3191 if (info2) {
3192 if (sysfs_init(info2, fd, st->devnm)) {
3193 pr_err("unable to initialize sysfs for %s",
3194 st->devnm);
3195 free(info2);
3196 goto release;
3197 }
3198 /* When increasing number of devices, we need to set
3199 * new raid_disks before adding these, or they might
3200 * be rejected.
3201 */
3202 if (reshape.backup_blocks &&
3203 reshape.after.data_disks > reshape.before.data_disks)
3204 subarray_set_num(container, info2, "raid_disks",
3205 reshape.after.data_disks +
3206 reshape.parity);
3207 for (d = info2->devs; d; d = d->next) {
3208 if (d->disk.state == 0 &&
3209 d->disk.raid_disk >= 0) {
3210 /* This is a spare that wants to
3211 * be part of the array.
3212 */
3213 add_disk(fd, st, info2, d);
3214 }
3215 }
3216 sysfs_free(info2);
3217 }
3218 }
3219 /* We might have been given some devices to add to the
3220 * array. Now that the array has been changed to the right
3221 * level and frozen, we can safely add them.
3222 */
3223 if (devlist) {
3224 if (Manage_subdevs(devname, fd, devlist, verbose,
3225 0, NULL, 0))
3226 goto release;
3227 }
3228
3229 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3230 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3231 if (reshape.backup_blocks == 0) {
3232 /* No restriping needed, but we might need to impose
3233 * some more changes: layout, raid_disks, chunk_size
3234 */
3235 /* read current array info */
3236 if (md_get_array_info(fd, &array) != 0) {
3237 dprintf("Cannot get array information.\n");
3238 goto release;
3239 }
3240 /* compare current array info with new values and if
3241 * it is different update them to new */
3242 if (info->new_layout != UnSet &&
3243 info->new_layout != array.layout) {
3244 array.layout = info->new_layout;
3245 if (md_set_array_info(fd, &array) != 0) {
3246 pr_err("failed to set new layout\n");
3247 goto release;
3248 } else if (verbose >= 0)
3249 printf("layout for %s set to %d\n",
3250 devname, array.layout);
3251 }
3252 if (info->delta_disks != UnSet &&
3253 info->delta_disks != 0 &&
3254 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3255 array.raid_disks += info->delta_disks;
3256 if (md_set_array_info(fd, &array) != 0) {
3257 pr_err("failed to set raid disks\n");
3258 goto release;
3259 } else if (verbose >= 0) {
3260 printf("raid_disks for %s set to %d\n",
3261 devname, array.raid_disks);
3262 }
3263 }
3264 if (info->new_chunk != 0 &&
3265 info->new_chunk != array.chunk_size) {
3266 if (sysfs_set_num(info, NULL,
3267 "chunk_size", info->new_chunk) != 0) {
3268 pr_err("failed to set chunk size\n");
3269 goto release;
3270 } else if (verbose >= 0)
3271 printf("chunk size for %s set to %d\n",
3272 devname, array.chunk_size);
3273 }
3274 unfreeze(st);
3275 return 0;
3276 }
3277
3278 /*
3279 * There are three possibilities.
3280 * 1/ The array will shrink.
3281 * We need to ensure the reshape will pause before reaching
3282 * the 'critical section'. We also need to fork and wait for
3283 * that to happen. When it does we
3284 * suspend/backup/complete/unfreeze
3285 *
3286 * 2/ The array will not change size.
3287 * This requires that we keep a backup of a sliding window
3288 * so that we can restore data after a crash. So we need
3289 * to fork and monitor progress.
3290 * In future we will allow the data_offset to change, so
3291 * a sliding backup becomes unnecessary.
3292 *
3293 * 3/ The array will grow. This is relatively easy.
3294 * However the kernel's restripe routines will cheerfully
3295 * overwrite some early data before it is safe. So we
3296 * need to make a backup of the early parts of the array
3297 * and be ready to restore it if rebuild aborts very early.
3298 * For externally managed metadata, we still need a forked
3299 * child to monitor the reshape and suspend IO over the region
3300 * that is being reshaped.
3301 *
3302 * We backup data by writing it to one spare, or to a
3303 * file which was given on command line.
3304 *
3305 * In each case, we first make sure that storage is available
3306 * for the required backup.
3307 * Then we:
3308 * - request the shape change.
3309 * - fork to handle backup etc.
3310 */
3311 /* Check that we can hold all the data */
3312 get_dev_size(fd, NULL, &array_size);
3313 if (reshape.new_size < (array_size/512)) {
3314 pr_err("this change will reduce the size of the array.\n"
3315 " use --grow --array-size first to truncate array.\n"
3316 " e.g. mdadm --grow %s --array-size %llu\n",
3317 devname, reshape.new_size/2);
3318 goto release;
3319 }
3320
3321 if (array.level == 10) {
3322 /* Reshaping RAID10 does not require any data backup by
3323 * user-space. Instead it requires that the data_offset
3324 * is changed to avoid the need for backup.
3325 * So this is handled very separately
3326 */
3327 if (restart)
3328 /* Nothing to do. */
3329 return 0;
3330 return raid10_reshape(container, fd, devname, st, info,
3331 &reshape, data_offset,
3332 force, verbose);
3333 }
3334 sra = sysfs_read(fd, NULL,
3335 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3336 GET_CACHE);
3337 if (!sra) {
3338 pr_err("%s: Cannot get array details from sysfs\n",
3339 devname);
3340 goto release;
3341 }
3342
3343 if (!backup_file)
3344 switch(set_new_data_offset(sra, st, devname,
3345 reshape.after.data_disks - reshape.before.data_disks,
3346 data_offset,
3347 reshape.min_offset_change, 1)) {
3348 case -1:
3349 goto release;
3350 case 0:
3351 /* Updated data_offset, so it's easy now */
3352 update_cache_size(container, sra, info,
3353 min(reshape.before.data_disks,
3354 reshape.after.data_disks),
3355 reshape.backup_blocks);
3356
3357 /* Right, everything seems fine. Let's kick things off.
3358 */
3359 sync_metadata(st);
3360
3361 if (impose_reshape(sra, info, st, fd, restart,
3362 devname, container, &reshape) < 0)
3363 goto release;
3364 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3365 struct mdinfo *sd;
3366 if (errno != EINVAL) {
3367 pr_err("Failed to initiate reshape!\n");
3368 goto release;
3369 }
3370 /* revert data_offset and try the old way */
3371 for (sd = sra->devs; sd; sd = sd->next) {
3372 sysfs_set_num(sra, sd, "new_offset",
3373 sd->data_offset);
3374 sysfs_set_str(sra, NULL, "reshape_direction",
3375 "forwards");
3376 }
3377 break;
3378 }
3379 if (info->new_level == reshape.level)
3380 return 0;
3381 /* need to adjust level when reshape completes */
3382 switch(fork()) {
3383 case -1: /* ignore error, but don't wait */
3384 return 0;
3385 default: /* parent */
3386 return 0;
3387 case 0:
3388 map_fork();
3389 break;
3390 }
3391 close(fd);
3392 wait_reshape(sra);
3393 fd = open_dev(sra->sys_name);
3394 if (fd >= 0)
3395 impose_level(fd, info->new_level, devname, verbose);
3396 return 0;
3397 case 1: /* Couldn't set data_offset, try the old way */
3398 if (data_offset != INVALID_SECTORS) {
3399 pr_err("Cannot update data_offset on this array\n");
3400 goto release;
3401 }
3402 break;
3403 }
3404
3405 started:
3406 /* Decide how many blocks (sectors) for a reshape
3407 * unit. The number we have so far is just a minimum
3408 */
3409 blocks = reshape.backup_blocks;
3410 if (reshape.before.data_disks ==
3411 reshape.after.data_disks) {
3412 /* Make 'blocks' bigger for better throughput, but
3413 * not so big that we reject it below.
3414 * Try for 16 megabytes
3415 */
3416 while (blocks * 32 < sra->component_size &&
3417 blocks < 16*1024*2)
3418 blocks *= 2;
3419 } else
3420 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3421
3422 if (blocks >= sra->component_size/2) {
3423 pr_err("%s: Something wrong - reshape aborted\n",
3424 devname);
3425 goto release;
3426 }
3427
3428 /* Now we need to open all these devices so we can read/write.
3429 */
3430 nrdisks = max(reshape.before.data_disks,
3431 reshape.after.data_disks) + reshape.parity
3432 + sra->array.spare_disks;
3433 fdlist = xcalloc((1+nrdisks), sizeof(int));
3434 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3435
3436 odisks = reshape.before.data_disks + reshape.parity;
3437 d = reshape_prepare_fdlist(devname, sra, odisks,
3438 nrdisks, blocks, backup_file,
3439 fdlist, offsets);
3440 if (d < odisks) {
3441 goto release;
3442 }
3443 if ((st->ss->manage_reshape == NULL) ||
3444 (st->ss->recover_backup == NULL)) {
3445 if (backup_file == NULL) {
3446 if (reshape.after.data_disks <=
3447 reshape.before.data_disks) {
3448 pr_err("%s: Cannot grow - need backup-file\n",
3449 devname);
3450 pr_err(" Please provide one with \"--backup=...\"\n");
3451 goto release;
3452 } else if (d == odisks) {
3453 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3454 goto release;
3455 }
3456 } else {
3457 if (!reshape_open_backup_file(backup_file, fd, devname,
3458 (signed)blocks,
3459 fdlist+d, offsets+d,
3460 sra->sys_name,
3461 restart)) {
3462 goto release;
3463 }
3464 d++;
3465 }
3466 }
3467
3468 update_cache_size(container, sra, info,
3469 min(reshape.before.data_disks, reshape.after.data_disks),
3470 blocks);
3471
3472 /* Right, everything seems fine. Let's kick things off.
3473 * If only changing raid_disks, use ioctl, else use
3474 * sysfs.
3475 */
3476 sync_metadata(st);
3477
3478 if (impose_reshape(sra, info, st, fd, restart,
3479 devname, container, &reshape) < 0)
3480 goto release;
3481
3482 err = start_reshape(sra, restart, reshape.before.data_disks,
3483 reshape.after.data_disks);
3484 if (err) {
3485 pr_err("Cannot %s reshape for %s\n",
3486 restart ? "continue" : "start",
3487 devname);
3488 goto release;
3489 }
3490 if (restart)
3491 sysfs_set_str(sra, NULL, "array_state", "active");
3492 if (freeze_reshape) {
3493 free(fdlist);
3494 free(offsets);
3495 sysfs_free(sra);
3496 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3497 sra->reshape_progress);
3498 return 1;
3499 }
3500
3501 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3502 if (continue_via_systemd(container ?: sra->sys_name)) {
3503 free(fdlist);
3504 free(offsets);
3505 sysfs_free(sra);
3506 return 0;
3507 }
3508
3509 /* Now we just need to kick off the reshape and watch, while
3510 * handling backups of the data...
3511 * This is all done by a forked background process.
3512 */
3513 switch(forked ? 0 : fork()) {
3514 case -1:
3515 pr_err("Cannot run child to monitor reshape: %s\n",
3516 strerror(errno));
3517 abort_reshape(sra);
3518 goto release;
3519 default:
3520 free(fdlist);
3521 free(offsets);
3522 sysfs_free(sra);
3523 return 0;
3524 case 0:
3525 map_fork();
3526 break;
3527 }
3528
3529 /* If another array on the same devices is busy, the
3530 * reshape will wait for them. This would mean that
3531 * the first section that we suspend will stay suspended
3532 * for a long time. So check on that possibility
3533 * by looking for "DELAYED" in /proc/mdstat, and if found,
3534 * wait a while
3535 */
3536 do {
3537 struct mdstat_ent *mds, *m;
3538 delayed = 0;
3539 mds = mdstat_read(1, 0);
3540 for (m = mds; m; m = m->next)
3541 if (strcmp(m->devnm, sra->sys_name) == 0) {
3542 if (m->resync &&
3543 m->percent == RESYNC_DELAYED)
3544 delayed = 1;
3545 if (m->resync == 0)
3546 /* Haven't started the reshape thread
3547 * yet, wait a bit
3548 */
3549 delayed = 2;
3550 break;
3551 }
3552 free_mdstat(mds);
3553 if (delayed == 1 && get_linux_version() < 3007000) {
3554 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3555 " You might experience problems until other reshapes complete.\n");
3556 delayed = 0;
3557 }
3558 if (delayed)
3559 mdstat_wait(30 - (delayed-1) * 25);
3560 } while (delayed);
3561 mdstat_close();
3562 close(fd);
3563 if (check_env("MDADM_GROW_VERIFY"))
3564 fd = open(devname, O_RDONLY | O_DIRECT);
3565 else
3566 fd = -1;
3567 mlockall(MCL_FUTURE);
3568
3569 signal(SIGTERM, catch_term);
3570
3571 if (st->ss->external) {
3572 /* metadata handler takes it from here */
3573 done = st->ss->manage_reshape(
3574 fd, sra, &reshape, st, blocks,
3575 fdlist, offsets,
3576 d - odisks, fdlist+odisks,
3577 offsets+odisks);
3578 } else
3579 done = child_monitor(
3580 fd, sra, &reshape, st, blocks,
3581 fdlist, offsets,
3582 d - odisks, fdlist+odisks,
3583 offsets+odisks);
3584
3585 free(fdlist);
3586 free(offsets);
3587
3588 if (backup_file && done) {
3589 char *bul;
3590 bul = make_backup(sra->sys_name);
3591 if (bul) {
3592 char buf[1024];
3593 int l = readlink(bul, buf, sizeof(buf) - 1);
3594 if (l > 0) {
3595 buf[l]=0;
3596 unlink(buf);
3597 }
3598 unlink(bul);
3599 free(bul);
3600 }
3601 unlink(backup_file);
3602 }
3603 if (!done) {
3604 abort_reshape(sra);
3605 goto out;
3606 }
3607
3608 if (!st->ss->external &&
3609 !(reshape.before.data_disks != reshape.after.data_disks
3610 && info->custom_array_size) &&
3611 info->new_level == reshape.level &&
3612 !forked) {
3613 /* no need to wait for the reshape to finish as
3614 * there is nothing more to do.
3615 */
3616 sysfs_free(sra);
3617 exit(0);
3618 }
3619 wait_reshape(sra);
3620
3621 if (st->ss->external) {
3622 /* Re-load the metadata as much could have changed */
3623 int cfd = open_dev(st->container_devnm);
3624 if (cfd >= 0) {
3625 flush_mdmon(container);
3626 st->ss->free_super(st);
3627 st->ss->load_container(st, cfd, container);
3628 close(cfd);
3629 }
3630 }
3631
3632 /* set new array size if required customer_array_size is used
3633 * by this metadata.
3634 */
3635 if (reshape.before.data_disks !=
3636 reshape.after.data_disks &&
3637 info->custom_array_size)
3638 set_array_size(st, info, info->text_version);
3639
3640 if (info->new_level != reshape.level) {
3641 if (fd < 0)
3642 fd = open(devname, O_RDONLY);
3643 impose_level(fd, info->new_level, devname, verbose);
3644 close(fd);
3645 if (info->new_level == 0)
3646 st->update_tail = NULL;
3647 }
3648 out:
3649 sysfs_free(sra);
3650 if (forked)
3651 return 0;
3652 unfreeze(st);
3653 exit(0);
3654
3655 release:
3656 free(fdlist);
3657 free(offsets);
3658 if (orig_level != UnSet && sra) {
3659 c = map_num(pers, orig_level);
3660 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3661 pr_err("aborting level change\n");
3662 }
3663 sysfs_free(sra);
3664 if (!forked)
3665 unfreeze(st);
3666 return 1;
3667 }
3668
3669 /* mdfd handle is passed to be closed in child process (after fork).
3670 */
3671 int reshape_container(char *container, char *devname,
3672 int mdfd,
3673 struct supertype *st,
3674 struct mdinfo *info,
3675 int force,
3676 char *backup_file, int verbose,
3677 int forked, int restart, int freeze_reshape)
3678 {
3679 struct mdinfo *cc = NULL;
3680 int rv = restart;
3681 char last_devnm[32] = "";
3682
3683 /* component_size is not meaningful for a container,
3684 * so pass '0' meaning 'no change'
3685 */
3686 if (!restart &&
3687 reshape_super(st, 0, info->new_level,
3688 info->new_layout, info->new_chunk,
3689 info->array.raid_disks, info->delta_disks,
3690 backup_file, devname, APPLY_METADATA_CHANGES,
3691 verbose)) {
3692 unfreeze(st);
3693 return 1;
3694 }
3695
3696 sync_metadata(st);
3697
3698 /* ping monitor to be sure that update is on disk
3699 */
3700 ping_monitor(container);
3701
3702 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3703 if (continue_via_systemd(container))
3704 return 0;
3705
3706 switch (forked ? 0 : fork()) {
3707 case -1: /* error */
3708 perror("Cannot fork to complete reshape\n");
3709 unfreeze(st);
3710 return 1;
3711 default: /* parent */
3712 if (!freeze_reshape)
3713 printf("%s: multi-array reshape continues in background\n", Name);
3714 return 0;
3715 case 0: /* child */
3716 map_fork();
3717 break;
3718 }
3719
3720 /* close unused handle in child process
3721 */
3722 if (mdfd > -1)
3723 close(mdfd);
3724
3725 while(1) {
3726 /* For each member array with reshape_active,
3727 * we need to perform the reshape.
3728 * We pick the first array that needs reshaping and
3729 * reshape it. reshape_array() will re-read the metadata
3730 * so the next time through a different array should be
3731 * ready for reshape.
3732 * It is possible that the 'different' array will not
3733 * be assembled yet. In that case we simple exit.
3734 * When it is assembled, the mdadm which assembles it
3735 * will take over the reshape.
3736 */
3737 struct mdinfo *content;
3738 int fd;
3739 struct mdstat_ent *mdstat;
3740 char *adev;
3741 dev_t devid;
3742
3743 sysfs_free(cc);
3744
3745 cc = st->ss->container_content(st, NULL);
3746
3747 for (content = cc; content ; content = content->next) {
3748 char *subarray;
3749 if (!content->reshape_active)
3750 continue;
3751
3752 subarray = strchr(content->text_version+1, '/')+1;
3753 mdstat = mdstat_by_subdev(subarray, container);
3754 if (!mdstat)
3755 continue;
3756 if (mdstat->active == 0) {
3757 pr_err("Skipping inactive array %s.\n",
3758 mdstat->devnm);
3759 free_mdstat(mdstat);
3760 mdstat = NULL;
3761 continue;
3762 }
3763 break;
3764 }
3765 if (!content)
3766 break;
3767
3768 devid = devnm2devid(mdstat->devnm);
3769 adev = map_dev(major(devid), minor(devid), 0);
3770 if (!adev)
3771 adev = content->text_version;
3772
3773 fd = open_dev(mdstat->devnm);
3774 if (fd < 0) {
3775 pr_err("Device %s cannot be opened for reshape.\n", adev);
3776 break;
3777 }
3778
3779 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3780 /* Do not allow for multiple reshape_array() calls for
3781 * the same array.
3782 * It can happen when reshape_array() returns without
3783 * error, when reshape is not finished (wrong reshape
3784 * starting/continuation conditions). Mdmon doesn't
3785 * switch to next array in container and reentry
3786 * conditions for the same array occur.
3787 * This is possibly interim until the behaviour of
3788 * reshape_array is resolved().
3789 */
3790 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3791 close(fd);
3792 break;
3793 }
3794 strcpy(last_devnm, mdstat->devnm);
3795
3796 if (sysfs_init(content, fd, mdstat->devnm)) {
3797 pr_err("Unable to initialize sysfs for %s\n",
3798 mdstat->devnm);
3799 rv = 1;
3800 break;
3801 }
3802
3803 if (mdmon_running(container))
3804 flush_mdmon(container);
3805
3806 rv = reshape_array(container, fd, adev, st,
3807 content, force, NULL, INVALID_SECTORS,
3808 backup_file, verbose, 1, restart,
3809 freeze_reshape);
3810 close(fd);
3811
3812 if (freeze_reshape) {
3813 sysfs_free(cc);
3814 exit(0);
3815 }
3816
3817 restart = 0;
3818 if (rv)
3819 break;
3820
3821 if (mdmon_running(container))
3822 flush_mdmon(container);
3823 }
3824 if (!rv)
3825 unfreeze(st);
3826 sysfs_free(cc);
3827 exit(0);
3828 }
3829
3830 /*
3831 * We run a child process in the background which performs the following
3832 * steps:
3833 * - wait for resync to reach a certain point
3834 * - suspend io to the following section
3835 * - backup that section
3836 * - allow resync to proceed further
3837 * - resume io
3838 * - discard the backup.
3839 *
3840 * When are combined in slightly different ways in the three cases.
3841 * Grow:
3842 * - suspend/backup/allow/wait/resume/discard
3843 * Shrink:
3844 * - allow/wait/suspend/backup/allow/wait/resume/discard
3845 * same-size:
3846 * - wait/resume/discard/suspend/backup/allow
3847 *
3848 * suspend/backup/allow always come together
3849 * wait/resume/discard do too.
3850 * For the same-size case we have two backups to improve flow.
3851 *
3852 */
3853
3854 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3855 unsigned long long backup_point,
3856 unsigned long long wait_point,
3857 unsigned long long *suspend_point,
3858 unsigned long long *reshape_completed, int *frozen)
3859 {
3860 /* This function is called repeatedly by the reshape manager.
3861 * It determines how much progress can safely be made and allows
3862 * that progress.
3863 * - 'info' identifies the array and particularly records in
3864 * ->reshape_progress the metadata's knowledge of progress
3865 * This is a sector offset from the start of the array
3866 * of the next array block to be relocated. This number
3867 * may increase from 0 or decrease from array_size, depending
3868 * on the type of reshape that is happening.
3869 * Note that in contrast, 'sync_completed' is a block count of the
3870 * reshape so far. It gives the distance between the start point
3871 * (head or tail of device) and the next place that data will be
3872 * written. It always increases.
3873 * - 'reshape' is the structure created by analyse_change
3874 * - 'backup_point' shows how much the metadata manager has backed-up
3875 * data. For reshapes with increasing progress, it is the next address
3876 * to be backed up, previous addresses have been backed-up. For
3877 * decreasing progress, it is the earliest address that has been
3878 * backed up - later address are also backed up.
3879 * So addresses between reshape_progress and backup_point are
3880 * backed up providing those are in the 'correct' order.
3881 * - 'wait_point' is an array address. When reshape_completed
3882 * passes this point, progress_reshape should return. It might
3883 * return earlier if it determines that ->reshape_progress needs
3884 * to be updated or further backup is needed.
3885 * - suspend_point is maintained by progress_reshape and the caller
3886 * should not touch it except to initialise to zero.
3887 * It is an array address and it only increases in 2.6.37 and earlier.
3888 * This makes it difficult to handle reducing reshapes with
3889 * external metadata.
3890 * However: it is similar to backup_point in that it records the
3891 * other end of a suspended region from reshape_progress.
3892 * it is moved to extend the region that is safe to backup and/or
3893 * reshape
3894 * - reshape_completed is read from sysfs and returned. The caller
3895 * should copy this into ->reshape_progress when it has reason to
3896 * believe that the metadata knows this, and any backup outside this
3897 * has been erased.
3898 *
3899 * Return value is:
3900 * 1 if more data from backup_point - but only as far as suspend_point,
3901 * should be backed up
3902 * 0 if things are progressing smoothly
3903 * -1 if the reshape is finished because it is all done,
3904 * -2 if the reshape is finished due to an error.
3905 */
3906
3907 int advancing = (reshape->after.data_disks
3908 >= reshape->before.data_disks);
3909 unsigned long long need_backup; /* All data between start of array and
3910 * here will at some point need to
3911 * be backed up.
3912 */
3913 unsigned long long read_offset, write_offset;
3914 unsigned long long write_range;
3915 unsigned long long max_progress, target, completed;
3916 unsigned long long array_size = (info->component_size
3917 * reshape->before.data_disks);
3918 int fd;
3919 char buf[20];
3920
3921 /* First, we unsuspend any region that is now known to be safe.
3922 * If suspend_point is on the 'wrong' side of reshape_progress, then
3923 * we don't have or need suspension at the moment. This is true for
3924 * native metadata when we don't need to back-up.
3925 */
3926 if (advancing) {
3927 if (info->reshape_progress <= *suspend_point)
3928 sysfs_set_num(info, NULL, "suspend_lo",
3929 info->reshape_progress);
3930 } else {
3931 /* Note: this won't work in 2.6.37 and before.
3932 * Something somewhere should make sure we don't need it!
3933 */
3934 if (info->reshape_progress >= *suspend_point)
3935 sysfs_set_num(info, NULL, "suspend_hi",
3936 info->reshape_progress);
3937 }
3938
3939 /* Now work out how far it is safe to progress.
3940 * If the read_offset for ->reshape_progress is less than
3941 * 'blocks' beyond the write_offset, we can only progress as far
3942 * as a backup.
3943 * Otherwise we can progress until the write_offset for the new location
3944 * reaches (within 'blocks' of) the read_offset at the current location.
3945 * However that region must be suspended unless we are using native
3946 * metadata.
3947 * If we need to suspend more, we limit it to 128M per device, which is
3948 * rather arbitrary and should be some time-based calculation.
3949 */
3950 read_offset = info->reshape_progress / reshape->before.data_disks;
3951 write_offset = info->reshape_progress / reshape->after.data_disks;
3952 write_range = info->new_chunk/512;
3953 if (reshape->before.data_disks == reshape->after.data_disks)
3954 need_backup = array_size;
3955 else
3956 need_backup = reshape->backup_blocks;
3957 if (advancing) {
3958 if (read_offset < write_offset + write_range)
3959 max_progress = backup_point;
3960 else
3961 max_progress =
3962 read_offset *
3963 reshape->after.data_disks;
3964 } else {
3965 if (read_offset > write_offset - write_range)
3966 /* Can only progress as far as has been backed up,
3967 * which must be suspended */
3968 max_progress = backup_point;
3969 else if (info->reshape_progress <= need_backup)
3970 max_progress = backup_point;
3971 else {
3972 if (info->array.major_version >= 0)
3973 /* Can progress until backup is needed */
3974 max_progress = need_backup;
3975 else {
3976 /* Can progress until metadata update is required */
3977 max_progress =
3978 read_offset *
3979 reshape->after.data_disks;
3980 /* but data must be suspended */
3981 if (max_progress < *suspend_point)
3982 max_progress = *suspend_point;
3983 }
3984 }
3985 }
3986
3987 /* We know it is safe to progress to 'max_progress' providing
3988 * it is suspended or we are using native metadata.
3989 * Consider extending suspend_point 128M per device if it
3990 * is less than 64M per device beyond reshape_progress.
3991 * But always do a multiple of 'blocks'
3992 * FIXME this is too big - it takes to long to complete
3993 * this much.
3994 */
3995 target = 64*1024*2 * min(reshape->before.data_disks,
3996 reshape->after.data_disks);
3997 target /= reshape->backup_blocks;
3998 if (target < 2)
3999 target = 2;
4000 target *= reshape->backup_blocks;
4001
4002 /* For externally managed metadata we always need to suspend IO to
4003 * the area being reshaped so we regularly push suspend_point forward.
4004 * For native metadata we only need the suspend if we are going to do
4005 * a backup.
4006 */
4007 if (advancing) {
4008 if ((need_backup > info->reshape_progress
4009 || info->array.major_version < 0) &&
4010 *suspend_point < info->reshape_progress + target) {
4011 if (need_backup < *suspend_point + 2 * target)
4012 *suspend_point = need_backup;
4013 else if (*suspend_point + 2 * target < array_size)
4014 *suspend_point += 2 * target;
4015 else
4016 *suspend_point = array_size;
4017 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
4018 if (max_progress > *suspend_point)
4019 max_progress = *suspend_point;
4020 }
4021 } else {
4022 if (info->array.major_version >= 0) {
4023 /* Only need to suspend when about to backup */
4024 if (info->reshape_progress < need_backup * 2 &&
4025 *suspend_point > 0) {
4026 *suspend_point = 0;
4027 sysfs_set_num(info, NULL, "suspend_lo", 0);
4028 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
4029 }
4030 } else {
4031 /* Need to suspend continually */
4032 if (info->reshape_progress < *suspend_point)
4033 *suspend_point = info->reshape_progress;
4034 if (*suspend_point + target < info->reshape_progress)
4035 /* No need to move suspend region yet */;
4036 else {
4037 if (*suspend_point >= 2 * target)
4038 *suspend_point -= 2 * target;
4039 else
4040 *suspend_point = 0;
4041 sysfs_set_num(info, NULL, "suspend_lo",
4042 *suspend_point);
4043 }
4044 if (max_progress < *suspend_point)
4045 max_progress = *suspend_point;
4046 }
4047 }
4048
4049 /* now set sync_max to allow that progress. sync_max, like
4050 * sync_completed is a count of sectors written per device, so
4051 * we find the difference between max_progress and the start point,
4052 * and divide that by after.data_disks to get a sync_max
4053 * number.
4054 * At the same time we convert wait_point to a similar number
4055 * for comparing against sync_completed.
4056 */
4057 /* scale down max_progress to per_disk */
4058 max_progress /= reshape->after.data_disks;
4059 /* Round to chunk size as some kernels give an erroneously high number */
4060 max_progress /= info->new_chunk/512;
4061 max_progress *= info->new_chunk/512;
4062 /* And round to old chunk size as the kernel wants that */
4063 max_progress /= info->array.chunk_size/512;
4064 max_progress *= info->array.chunk_size/512;
4065 /* Limit progress to the whole device */
4066 if (max_progress > info->component_size)
4067 max_progress = info->component_size;
4068 wait_point /= reshape->after.data_disks;
4069 if (!advancing) {
4070 /* switch from 'device offset' to 'processed block count' */
4071 max_progress = info->component_size - max_progress;
4072 wait_point = info->component_size - wait_point;
4073 }
4074
4075 if (!*frozen)
4076 sysfs_set_num(info, NULL, "sync_max", max_progress);
4077
4078 /* Now wait. If we have already reached the point that we were
4079 * asked to wait to, don't wait at all, else wait for any change.
4080 * We need to select on 'sync_completed' as that is the place that
4081 * notifications happen, but we are really interested in
4082 * 'reshape_position'
4083 */
4084 fd = sysfs_get_fd(info, NULL, "sync_completed");
4085 if (fd < 0)
4086 goto check_progress;
4087
4088 if (sysfs_fd_get_ll(fd, &completed) < 0)
4089 goto check_progress;
4090
4091 while (completed < max_progress && completed < wait_point) {
4092 /* Check that sync_action is still 'reshape' to avoid
4093 * waiting forever on a dead array
4094 */
4095 char action[20];
4096 if (sysfs_get_str(info, NULL, "sync_action",
4097 action, 20) <= 0 ||
4098 strncmp(action, "reshape", 7) != 0)
4099 break;
4100 /* Some kernels reset 'sync_completed' to zero
4101 * before setting 'sync_action' to 'idle'.
4102 * So we need these extra tests.
4103 */
4104 if (completed == 0 && advancing
4105 && strncmp(action, "idle", 4) == 0
4106 && info->reshape_progress > 0)
4107 break;
4108 if (completed == 0 && !advancing
4109 && strncmp(action, "idle", 4) == 0
4110 && info->reshape_progress < (info->component_size
4111 * reshape->after.data_disks))
4112 break;
4113 sysfs_wait(fd, NULL);
4114 if (sysfs_fd_get_ll(fd, &completed) < 0)
4115 goto check_progress;
4116 }
4117 /* Some kernels reset 'sync_completed' to zero,
4118 * we need to have real point we are in md.
4119 * So in that case, read 'reshape_position' from sysfs.
4120 */
4121 if (completed == 0) {
4122 unsigned long long reshapep;
4123 char action[20];
4124 if (sysfs_get_str(info, NULL, "sync_action",
4125 action, 20) > 0 &&
4126 strncmp(action, "idle", 4) == 0 &&
4127 sysfs_get_ll(info, NULL,
4128 "reshape_position", &reshapep) == 0)
4129 *reshape_completed = reshapep;
4130 } else {
4131 /* some kernels can give an incorrectly high
4132 * 'completed' number, so round down */
4133 completed /= (info->new_chunk/512);
4134 completed *= (info->new_chunk/512);
4135 /* Convert 'completed' back in to a 'progress' number */
4136 completed *= reshape->after.data_disks;
4137 if (!advancing)
4138 completed = (info->component_size
4139 * reshape->after.data_disks
4140 - completed);
4141 *reshape_completed = completed;
4142 }
4143
4144 close(fd);
4145
4146 /* We return the need_backup flag. Caller will decide
4147 * how much - a multiple of ->backup_blocks up to *suspend_point
4148 */
4149 if (advancing)
4150 return need_backup > info->reshape_progress;
4151 else
4152 return need_backup >= info->reshape_progress;
4153
4154 check_progress:
4155 /* if we couldn't read a number from sync_completed, then
4156 * either the reshape did complete, or it aborted.
4157 * We can tell which by checking for 'none' in reshape_position.
4158 * If it did abort, then it might immediately restart if it
4159 * it was just a device failure that leaves us degraded but
4160 * functioning.
4161 */
4162 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
4163 || strncmp(buf, "none", 4) != 0) {
4164 /* The abort might only be temporary. Wait up to 10
4165 * seconds for fd to contain a valid number again.
4166 */
4167 int wait = 10000;
4168 int rv = -2;
4169 unsigned long long new_sync_max;
4170 while (fd >= 0 && rv < 0 && wait > 0) {
4171 if (sysfs_wait(fd, &wait) != 1)
4172 break;
4173 switch (sysfs_fd_get_ll(fd, &completed)) {
4174 case 0:
4175 /* all good again */
4176 rv = 1;
4177 /* If "sync_max" is no longer max_progress
4178 * we need to freeze things
4179 */
4180 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
4181 *frozen = (new_sync_max != max_progress);
4182 break;
4183 case -2: /* read error - abort */
4184 wait = 0;
4185 break;
4186 }
4187 }
4188 if (fd >= 0)
4189 close(fd);
4190 return rv; /* abort */
4191 } else {
4192 /* Maybe racing with array shutdown - check state */
4193 if (fd >= 0)
4194 close(fd);
4195 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
4196 || strncmp(buf, "inactive", 8) == 0
4197 || strncmp(buf, "clear",5) == 0)
4198 return -2; /* abort */
4199 return -1; /* complete */
4200 }
4201 }
4202
4203 /* FIXME return status is never checked */
4204 static int grow_backup(struct mdinfo *sra,
4205 unsigned long long offset, /* per device */
4206 unsigned long stripes, /* per device, in old chunks */
4207 int *sources, unsigned long long *offsets,
4208 int disks, int chunk, int level, int layout,
4209 int dests, int *destfd, unsigned long long *destoffsets,
4210 int part, int *degraded,
4211 char *buf)
4212 {
4213 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4214 * to storage 'destfd' (offset 'destoffsets'), after first
4215 * suspending IO. Then allow resync to continue
4216 * over the suspended section.
4217 * Use part 'part' of the backup-super-block.
4218 */
4219 int odata = disks;
4220 int rv = 0;
4221 int i;
4222 unsigned long long ll;
4223 int new_degraded;
4224 //printf("offset %llu\n", offset);
4225 if (level >= 4)
4226 odata--;
4227 if (level == 6)
4228 odata--;
4229
4230 /* Check that array hasn't become degraded, else we might backup the wrong data */
4231 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4232 return -1; /* FIXME this error is ignored */
4233 new_degraded = (int)ll;
4234 if (new_degraded != *degraded) {
4235 /* check each device to ensure it is still working */
4236 struct mdinfo *sd;
4237 for (sd = sra->devs ; sd ; sd = sd->next) {
4238 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4239 continue;
4240 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4241 char sbuf[100];
4242
4243 if (sysfs_get_str(sra, sd, "state",
4244 sbuf, sizeof(sbuf)) < 0 ||
4245 strstr(sbuf, "faulty") ||
4246 strstr(sbuf, "in_sync") == NULL) {
4247 /* this device is dead */
4248 sd->disk.state = (1<<MD_DISK_FAULTY);
4249 if (sd->disk.raid_disk >= 0 &&
4250 sources[sd->disk.raid_disk] >= 0) {
4251 close(sources[sd->disk.raid_disk]);
4252 sources[sd->disk.raid_disk] = -1;
4253 }
4254 }
4255 }
4256 }
4257 *degraded = new_degraded;
4258 }
4259 if (part) {
4260 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4261 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4262 } else {
4263 bsb.arraystart = __cpu_to_le64(offset * odata);
4264 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4265 }
4266 if (part)
4267 bsb.magic[15] = '2';
4268 for (i = 0; i < dests; i++)
4269 if (part)
4270 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4271 else
4272 lseek64(destfd[i], destoffsets[i], 0);
4273
4274 rv = save_stripes(sources, offsets,
4275 disks, chunk, level, layout,
4276 dests, destfd,
4277 offset*512*odata, stripes * chunk * odata,
4278 buf);
4279
4280 if (rv)
4281 return rv;
4282 bsb.mtime = __cpu_to_le64(time(0));
4283 for (i = 0; i < dests; i++) {
4284 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4285
4286 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4287 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4288 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4289 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4290
4291 rv = -1;
4292 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4293 != destoffsets[i] - 4096)
4294 break;
4295 if (write(destfd[i], &bsb, 512) != 512)
4296 break;
4297 if (destoffsets[i] > 4096) {
4298 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4299 destoffsets[i]+stripes*chunk*odata)
4300 break;
4301 if (write(destfd[i], &bsb, 512) != 512)
4302 break;
4303 }
4304 fsync(destfd[i]);
4305 rv = 0;
4306 }
4307
4308 return rv;
4309 }
4310
4311 /* in 2.6.30, the value reported by sync_completed can be
4312 * less that it should be by one stripe.
4313 * This only happens when reshape hits sync_max and pauses.
4314 * So allow wait_backup to either extent sync_max further
4315 * than strictly necessary, or return before the
4316 * sync has got quite as far as we would really like.
4317 * This is what 'blocks2' is for.
4318 * The various caller give appropriate values so that
4319 * every works.
4320 */
4321 /* FIXME return value is often ignored */
4322 static int forget_backup(int dests, int *destfd,
4323 unsigned long long *destoffsets,
4324 int part)
4325 {
4326 /*
4327 * Erase backup 'part' (which is 0 or 1)
4328 */
4329 int i;
4330 int rv;
4331
4332 if (part) {
4333 bsb.arraystart2 = __cpu_to_le64(0);
4334 bsb.length2 = __cpu_to_le64(0);
4335 } else {
4336 bsb.arraystart = __cpu_to_le64(0);
4337 bsb.length = __cpu_to_le64(0);
4338 }
4339 bsb.mtime = __cpu_to_le64(time(0));
4340 rv = 0;
4341 for (i = 0; i < dests; i++) {
4342 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4343 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4344 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4345 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4346 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4347 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4348 destoffsets[i]-4096)
4349 rv = -1;
4350 if (rv == 0 &&
4351 write(destfd[i], &bsb, 512) != 512)
4352 rv = -1;
4353 fsync(destfd[i]);
4354 }
4355 return rv;
4356 }
4357
4358 static void fail(char *msg)
4359 {
4360 int rv;
4361 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4362 rv |= (write(2, "\n", 1) != 1);
4363 exit(rv ? 1 : 2);
4364 }
4365
4366 static char *abuf, *bbuf;
4367 static unsigned long long abuflen;
4368 static void validate(int afd, int bfd, unsigned long long offset)
4369 {
4370 /* check that the data in the backup against the array.
4371 * This is only used for regression testing and should not
4372 * be used while the array is active
4373 */
4374 if (afd < 0)
4375 return;
4376 lseek64(bfd, offset - 4096, 0);
4377 if (read(bfd, &bsb2, 512) != 512)
4378 fail("cannot read bsb");
4379 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4380 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4381 fail("first csum bad");
4382 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4383 fail("magic is bad");
4384 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4385 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4386 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4387 fail("second csum bad");
4388
4389 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4390 fail("devstart is wrong");
4391
4392 if (bsb2.length) {
4393 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4394
4395 if (abuflen < len) {
4396 free(abuf);
4397 free(bbuf);
4398 abuflen = len;
4399 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4400 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4401 abuflen = 0;
4402 /* just stop validating on mem-alloc failure */
4403 return;
4404 }
4405 }
4406
4407 lseek64(bfd, offset, 0);
4408 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4409 //printf("len %llu\n", len);
4410 fail("read first backup failed");
4411 }
4412 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4413 if ((unsigned long long)read(afd, abuf, len) != len)
4414 fail("read first from array failed");
4415 if (memcmp(bbuf, abuf, len) != 0) {
4416 #if 0
4417 int i;
4418 printf("offset=%llu len=%llu\n",
4419 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4420 for (i=0; i<len; i++)
4421 if (bbuf[i] != abuf[i]) {
4422 printf("first diff byte %d\n", i);
4423 break;
4424 }
4425 #endif
4426 fail("data1 compare failed");
4427 }
4428 }
4429 if (bsb2.length2) {
4430 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4431
4432 if (abuflen < len) {
4433 free(abuf);
4434 free(bbuf);
4435 abuflen = len;
4436 abuf = xmalloc(abuflen);
4437 bbuf = xmalloc(abuflen);
4438 }
4439
4440 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4441 if ((unsigned long long)read(bfd, bbuf, len) != len)
4442 fail("read second backup failed");
4443 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4444 if ((unsigned long long)read(afd, abuf, len) != len)
4445 fail("read second from array failed");
4446 if (memcmp(bbuf, abuf, len) != 0)
4447 fail("data2 compare failed");
4448 }
4449 }
4450
4451 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4452 struct supertype *st, unsigned long blocks,
4453 int *fds, unsigned long long *offsets,
4454 int dests, int *destfd, unsigned long long *destoffsets)
4455 {
4456 /* Monitor a reshape where backup is being performed using
4457 * 'native' mechanism - either to a backup file, or
4458 * to some space in a spare.
4459 */
4460 char *buf;
4461 int degraded = -1;
4462 unsigned long long speed;
4463 unsigned long long suspend_point, array_size;
4464 unsigned long long backup_point, wait_point;
4465 unsigned long long reshape_completed;
4466 int done = 0;
4467 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4468 int part = 0; /* The next part of the backup area to fill. It may already
4469 * be full, so we need to check */
4470 int level = reshape->level;
4471 int layout = reshape->before.layout;
4472 int data = reshape->before.data_disks;
4473 int disks = reshape->before.data_disks + reshape->parity;
4474 int chunk = sra->array.chunk_size;
4475 struct mdinfo *sd;
4476 unsigned long stripes;
4477 int uuid[4];
4478 int frozen = 0;
4479
4480 /* set up the backup-super-block. This requires the
4481 * uuid from the array.
4482 */
4483 /* Find a superblock */
4484 for (sd = sra->devs; sd; sd = sd->next) {
4485 char *dn;
4486 int devfd;
4487 int ok;
4488 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4489 continue;
4490 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4491 devfd = dev_open(dn, O_RDONLY);
4492 if (devfd < 0)
4493 continue;
4494 ok = st->ss->load_super(st, devfd, NULL);
4495 close(devfd);
4496 if (ok == 0)
4497 break;
4498 }
4499 if (!sd) {
4500 pr_err("Cannot find a superblock\n");
4501 return 0;
4502 }
4503
4504 memset(&bsb, 0, 512);
4505 memcpy(bsb.magic, "md_backup_data-1", 16);
4506 st->ss->uuid_from_super(st, uuid);
4507 memcpy(bsb.set_uuid, uuid, 16);
4508 bsb.mtime = __cpu_to_le64(time(0));
4509 bsb.devstart2 = blocks;
4510
4511 stripes = blocks / (sra->array.chunk_size/512) /
4512 reshape->before.data_disks;
4513
4514 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4515 /* Don't start the 'reshape' */
4516 return 0;
4517 if (reshape->before.data_disks == reshape->after.data_disks) {
4518 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4519 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4520 }
4521
4522 if (increasing) {
4523 array_size = sra->component_size * reshape->after.data_disks;
4524 backup_point = sra->reshape_progress;
4525 suspend_point = 0;
4526 } else {
4527 array_size = sra->component_size * reshape->before.data_disks;
4528 backup_point = reshape->backup_blocks;
4529 suspend_point = array_size;
4530 }
4531
4532 while (!done) {
4533 int rv;
4534
4535 /* Want to return as soon the oldest backup slot can
4536 * be released as that allows us to start backing up
4537 * some more, providing suspend_point has been
4538 * advanced, which it should have.
4539 */
4540 if (increasing) {
4541 wait_point = array_size;
4542 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4543 wait_point = (__le64_to_cpu(bsb.arraystart) +
4544 __le64_to_cpu(bsb.length));
4545 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4546 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4547 __le64_to_cpu(bsb.length2));
4548 } else {
4549 wait_point = 0;
4550 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4551 wait_point = __le64_to_cpu(bsb.arraystart);
4552 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4553 wait_point = __le64_to_cpu(bsb.arraystart2);
4554 }
4555
4556 reshape_completed = sra->reshape_progress;
4557 rv = progress_reshape(sra, reshape,
4558 backup_point, wait_point,
4559 &suspend_point, &reshape_completed,
4560 &frozen);
4561 /* external metadata would need to ping_monitor here */
4562 sra->reshape_progress = reshape_completed;
4563
4564 /* Clear any backup region that is before 'here' */
4565 if (increasing) {
4566 if (__le64_to_cpu(bsb.length) > 0 &&
4567 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4568 __le64_to_cpu(bsb.length)))
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 __le64_to_cpu(bsb.length2)))
4574 forget_backup(dests, destfd,
4575 destoffsets, 1);
4576 } else {
4577 if (__le64_to_cpu(bsb.length) > 0 &&
4578 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4579 forget_backup(dests, destfd,
4580 destoffsets, 0);
4581 if (__le64_to_cpu(bsb.length2) > 0 &&
4582 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4583 forget_backup(dests, destfd,
4584 destoffsets, 1);
4585 }
4586 if (sigterm)
4587 rv = -2;
4588 if (rv < 0) {
4589 if (rv == -1)
4590 done = 1;
4591 break;
4592 }
4593 if (rv == 0 && increasing && !st->ss->external) {
4594 /* No longer need to monitor this reshape */
4595 sysfs_set_str(sra, NULL, "sync_max", "max");
4596 done = 1;
4597 break;
4598 }
4599
4600 while (rv) {
4601 unsigned long long offset;
4602 unsigned long actual_stripes;
4603 /* Need to backup some data.
4604 * If 'part' is not used and the desired
4605 * backup size is suspended, do a backup,
4606 * then consider the next part.
4607 */
4608 /* Check that 'part' is unused */
4609 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4610 break;
4611 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4612 break;
4613
4614 offset = backup_point / data;
4615 actual_stripes = stripes;
4616 if (increasing) {
4617 if (offset + actual_stripes * (chunk/512) >
4618 sra->component_size)
4619 actual_stripes = ((sra->component_size - offset)
4620 / (chunk/512));
4621 if (offset + actual_stripes * (chunk/512) >
4622 suspend_point/data)
4623 break;
4624 } else {
4625 if (offset < actual_stripes * (chunk/512))
4626 actual_stripes = offset / (chunk/512);
4627 offset -= actual_stripes * (chunk/512);
4628 if (offset < suspend_point/data)
4629 break;
4630 }
4631 if (actual_stripes == 0)
4632 break;
4633 grow_backup(sra, offset, actual_stripes,
4634 fds, offsets,
4635 disks, chunk, level, layout,
4636 dests, destfd, destoffsets,
4637 part, &degraded, buf);
4638 validate(afd, destfd[0], destoffsets[0]);
4639 /* record where 'part' is up to */
4640 part = !part;
4641 if (increasing)
4642 backup_point += actual_stripes * (chunk/512) * data;
4643 else
4644 backup_point -= actual_stripes * (chunk/512) * data;
4645 }
4646 }
4647
4648 /* FIXME maybe call progress_reshape one more time instead */
4649 /* remove any remaining suspension */
4650 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4651 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4652 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4653 sysfs_set_num(sra, NULL, "sync_min", 0);
4654
4655 if (reshape->before.data_disks == reshape->after.data_disks)
4656 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4657 free(buf);
4658 return done;
4659 }
4660
4661 /*
4662 * If any spare contains md_back_data-1 which is recent wrt mtime,
4663 * write that data into the array and update the super blocks with
4664 * the new reshape_progress
4665 */
4666 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4667 char *backup_file, int verbose)
4668 {
4669 int i, j;
4670 int old_disks;
4671 unsigned long long *offsets;
4672 unsigned long long nstripe, ostripe;
4673 int ndata, odata;
4674
4675 odata = info->array.raid_disks - info->delta_disks - 1;
4676 if (info->array.level == 6) odata--; /* number of data disks */
4677 ndata = info->array.raid_disks - 1;
4678 if (info->new_level == 6) ndata--;
4679
4680 old_disks = info->array.raid_disks - info->delta_disks;
4681
4682 if (info->delta_disks <= 0)
4683 /* Didn't grow, so the backup file must have
4684 * been used
4685 */
4686 old_disks = cnt;
4687 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4688 struct mdinfo dinfo;
4689 int fd;
4690 int bsbsize;
4691 char *devname, namebuf[20];
4692 unsigned long long lo, hi;
4693
4694 /* This was a spare and may have some saved data on it.
4695 * Load the superblock, find and load the
4696 * backup_super_block.
4697 * If either fail, go on to next device.
4698 * If the backup contains no new info, just return
4699 * else restore data and update all superblocks
4700 */
4701 if (i == old_disks-1) {
4702 fd = open(backup_file, O_RDONLY);
4703 if (fd<0) {
4704 pr_err("backup file %s inaccessible: %s\n",
4705 backup_file, strerror(errno));
4706 continue;
4707 }
4708 devname = backup_file;
4709 } else {
4710 fd = fdlist[i];
4711 if (fd < 0)
4712 continue;
4713 if (st->ss->load_super(st, fd, NULL))
4714 continue;
4715
4716 st->ss->getinfo_super(st, &dinfo, NULL);
4717 st->ss->free_super(st);
4718
4719 if (lseek64(fd,
4720 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4721 0) < 0) {
4722 pr_err("Cannot seek on device %d\n", i);
4723 continue; /* Cannot seek */
4724 }
4725 sprintf(namebuf, "device-%d", i);
4726 devname = namebuf;
4727 }
4728 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4729 if (verbose)
4730 pr_err("Cannot read from %s\n", devname);
4731 continue; /* Cannot read */
4732 }
4733 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4734 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4735 if (verbose)
4736 pr_err("No backup metadata on %s\n", devname);
4737 continue;
4738 }
4739 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4740 if (verbose)
4741 pr_err("Bad backup-metadata checksum on %s\n", devname);
4742 continue; /* bad checksum */
4743 }
4744 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4745 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4746 if (verbose)
4747 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4748 continue; /* Bad second checksum */
4749 }
4750 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4751 if (verbose)
4752 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4753 continue; /* Wrong uuid */
4754 }
4755
4756 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4757 * sometimes they aren't... So allow considerable flexability in matching, and allow
4758 * this test to be overridden by an environment variable.
4759 */
4760 if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) ||
4761 time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) {
4762 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4763 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4764 (unsigned long)__le64_to_cpu(bsb.mtime),
4765 (unsigned long)info->array.utime);
4766 } else {
4767 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4768 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4769 continue; /* time stamp is too bad */
4770 }
4771 }
4772
4773 if (bsb.magic[15] == '1') {
4774 if (bsb.length == 0)
4775 continue;
4776 if (info->delta_disks >= 0) {
4777 /* reshape_progress is increasing */
4778 if (__le64_to_cpu(bsb.arraystart)
4779 + __le64_to_cpu(bsb.length)
4780 < info->reshape_progress) {
4781 nonew:
4782 if (verbose)
4783 pr_err("backup-metadata found on %s but is not needed\n", devname);
4784 continue; /* No new data here */
4785 }
4786 } else {
4787 /* reshape_progress is decreasing */
4788 if (__le64_to_cpu(bsb.arraystart) >=
4789 info->reshape_progress)
4790 goto nonew; /* No new data here */
4791 }
4792 } else {
4793 if (bsb.length == 0 && bsb.length2 == 0)
4794 continue;
4795 if (info->delta_disks >= 0) {
4796 /* reshape_progress is increasing */
4797 if ((__le64_to_cpu(bsb.arraystart)
4798 + __le64_to_cpu(bsb.length)
4799 < info->reshape_progress)
4800 &&
4801 (__le64_to_cpu(bsb.arraystart2)
4802 + __le64_to_cpu(bsb.length2)
4803 < info->reshape_progress))
4804 goto nonew; /* No new data here */
4805 } else {
4806 /* reshape_progress is decreasing */
4807 if (__le64_to_cpu(bsb.arraystart) >=
4808 info->reshape_progress &&
4809 __le64_to_cpu(bsb.arraystart2) >=
4810 info->reshape_progress)
4811 goto nonew; /* No new data here */
4812 }
4813 }
4814 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4815 second_fail:
4816 if (verbose)
4817 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4818 devname);
4819 continue; /* Cannot seek */
4820 }
4821 /* There should be a duplicate backup superblock 4k before here */
4822 if (lseek64(fd, -4096, 1) < 0 ||
4823 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4824 goto second_fail; /* Cannot find leading superblock */
4825 if (bsb.magic[15] == '1')
4826 bsbsize = offsetof(struct mdp_backup_super, pad1);
4827 else
4828 bsbsize = offsetof(struct mdp_backup_super, pad);
4829 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4830 goto second_fail; /* Cannot find leading superblock */
4831
4832 /* Now need the data offsets for all devices. */
4833 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4834 for(j=0; j<info->array.raid_disks; j++) {
4835 if (fdlist[j] < 0)
4836 continue;
4837 if (st->ss->load_super(st, fdlist[j], NULL))
4838 /* FIXME should be this be an error */
4839 continue;
4840 st->ss->getinfo_super(st, &dinfo, NULL);
4841 st->ss->free_super(st);
4842 offsets[j] = dinfo.data_offset * 512;
4843 }
4844 printf("%s: restoring critical section\n", Name);
4845
4846 if (restore_stripes(fdlist, offsets,
4847 info->array.raid_disks,
4848 info->new_chunk,
4849 info->new_level,
4850 info->new_layout,
4851 fd, __le64_to_cpu(bsb.devstart)*512,
4852 __le64_to_cpu(bsb.arraystart)*512,
4853 __le64_to_cpu(bsb.length)*512, NULL)) {
4854 /* didn't succeed, so giveup */
4855 if (verbose)
4856 pr_err("Error restoring backup from %s\n",
4857 devname);
4858 free(offsets);
4859 return 1;
4860 }
4861
4862 if (bsb.magic[15] == '2' &&
4863 restore_stripes(fdlist, offsets,
4864 info->array.raid_disks,
4865 info->new_chunk,
4866 info->new_level,
4867 info->new_layout,
4868 fd, __le64_to_cpu(bsb.devstart)*512 +
4869 __le64_to_cpu(bsb.devstart2)*512,
4870 __le64_to_cpu(bsb.arraystart2)*512,
4871 __le64_to_cpu(bsb.length2)*512, NULL)) {
4872 /* didn't succeed, so giveup */
4873 if (verbose)
4874 pr_err("Error restoring second backup from %s\n",
4875 devname);
4876 free(offsets);
4877 return 1;
4878 }
4879
4880 free(offsets);
4881
4882 /* Ok, so the data is restored. Let's update those superblocks. */
4883
4884 lo = hi = 0;
4885 if (bsb.length) {
4886 lo = __le64_to_cpu(bsb.arraystart);
4887 hi = lo + __le64_to_cpu(bsb.length);
4888 }
4889 if (bsb.magic[15] == '2' && bsb.length2) {
4890 unsigned long long lo1, hi1;
4891 lo1 = __le64_to_cpu(bsb.arraystart2);
4892 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4893 if (lo == hi) {
4894 lo = lo1;
4895 hi = hi1;
4896 } else if (lo < lo1)
4897 hi = hi1;
4898 else
4899 lo = lo1;
4900 }
4901 if (lo < hi &&
4902 (info->reshape_progress < lo ||
4903 info->reshape_progress > hi))
4904 /* backup does not affect reshape_progress*/ ;
4905 else if (info->delta_disks >= 0) {
4906 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4907 __le64_to_cpu(bsb.length);
4908 if (bsb.magic[15] == '2') {
4909 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4910 __le64_to_cpu(bsb.length2);
4911 if (p2 > info->reshape_progress)
4912 info->reshape_progress = p2;
4913 }
4914 } else {
4915 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4916 if (bsb.magic[15] == '2') {
4917 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4918 if (p2 < info->reshape_progress)
4919 info->reshape_progress = p2;
4920 }
4921 }
4922 for (j=0; j<info->array.raid_disks; j++) {
4923 if (fdlist[j] < 0)
4924 continue;
4925 if (st->ss->load_super(st, fdlist[j], NULL))
4926 continue;
4927 st->ss->getinfo_super(st, &dinfo, NULL);
4928 dinfo.reshape_progress = info->reshape_progress;
4929 st->ss->update_super(st, &dinfo,
4930 "_reshape_progress",
4931 NULL,0, 0, NULL);
4932 st->ss->store_super(st, fdlist[j]);
4933 st->ss->free_super(st);
4934 }
4935 return 0;
4936 }
4937 /* Didn't find any backup data, try to see if any
4938 * was needed.
4939 */
4940 if (info->delta_disks < 0) {
4941 /* When shrinking, the critical section is at the end.
4942 * So see if we are before the critical section.
4943 */
4944 unsigned long long first_block;
4945 nstripe = ostripe = 0;
4946 first_block = 0;
4947 while (ostripe >= nstripe) {
4948 ostripe += info->array.chunk_size / 512;
4949 first_block = ostripe * odata;
4950 nstripe = first_block / ndata / (info->new_chunk/512) *
4951 (info->new_chunk/512);
4952 }
4953
4954 if (info->reshape_progress >= first_block)
4955 return 0;
4956 }
4957 if (info->delta_disks > 0) {
4958 /* See if we are beyond the critical section. */
4959 unsigned long long last_block;
4960 nstripe = ostripe = 0;
4961 last_block = 0;
4962 while (nstripe >= ostripe) {
4963 nstripe += info->new_chunk / 512;
4964 last_block = nstripe * ndata;
4965 ostripe = last_block / odata / (info->array.chunk_size/512) *
4966 (info->array.chunk_size/512);
4967 }
4968
4969 if (info->reshape_progress >= last_block)
4970 return 0;
4971 }
4972 /* needed to recover critical section! */
4973 if (verbose)
4974 pr_err("Failed to find backup of critical section\n");
4975 return 1;
4976 }
4977
4978 int Grow_continue_command(char *devname, int fd,
4979 char *backup_file, int verbose)
4980 {
4981 int ret_val = 0;
4982 struct supertype *st = NULL;
4983 struct mdinfo *content = NULL;
4984 struct mdinfo array;
4985 char *subarray = NULL;
4986 struct mdinfo *cc = NULL;
4987 struct mdstat_ent *mdstat = NULL;
4988 int cfd = -1;
4989 int fd2;
4990
4991 dprintf("Grow continue from command line called for %s\n",
4992 devname);
4993
4994 st = super_by_fd(fd, &subarray);
4995 if (!st || !st->ss) {
4996 pr_err("Unable to determine metadata format for %s\n",
4997 devname);
4998 return 1;
4999 }
5000 dprintf("Grow continue is run for ");
5001 if (st->ss->external == 0) {
5002 int d;
5003 int cnt = 5;
5004 dprintf_cont("native array (%s)\n", devname);
5005 if (md_get_array_info(fd, &array.array) < 0) {
5006 pr_err("%s is not an active md array - aborting\n",
5007 devname);
5008 ret_val = 1;
5009 goto Grow_continue_command_exit;
5010 }
5011 content = &array;
5012 /* Need to load a superblock.
5013 * FIXME we should really get what we need from
5014 * sysfs
5015 */
5016 do {
5017 for (d = 0; d < MAX_DISKS; d++) {
5018 mdu_disk_info_t disk;
5019 char *dv;
5020 int err;
5021 disk.number = d;
5022 if (md_get_disk_info(fd, &disk) < 0)
5023 continue;
5024 if (disk.major == 0 && disk.minor == 0)
5025 continue;
5026 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
5027 continue;
5028 dv = map_dev(disk.major, disk.minor, 1);
5029 if (!dv)
5030 continue;
5031 fd2 = dev_open(dv, O_RDONLY);
5032 if (fd2 < 0)
5033 continue;
5034 err = st->ss->load_super(st, fd2, NULL);
5035 close(fd2);
5036 if (err)
5037 continue;
5038 break;
5039 }
5040 if (d == MAX_DISKS) {
5041 pr_err("Unable to load metadata for %s\n",
5042 devname);
5043 ret_val = 1;
5044 goto Grow_continue_command_exit;
5045 }
5046 st->ss->getinfo_super(st, content, NULL);
5047 if (!content->reshape_active)
5048 sleep(3);
5049 else
5050 break;
5051 } while (cnt-- > 0);
5052 } else {
5053 char *container;
5054
5055 if (subarray) {
5056 dprintf_cont("subarray (%s)\n", subarray);
5057 container = st->container_devnm;
5058 cfd = open_dev_excl(st->container_devnm);
5059 } else {
5060 container = st->devnm;
5061 close(fd);
5062 cfd = open_dev_excl(st->devnm);
5063 dprintf_cont("container (%s)\n", container);
5064 fd = cfd;
5065 }
5066 if (cfd < 0) {
5067 pr_err("Unable to open container for %s\n", devname);
5068 ret_val = 1;
5069 goto Grow_continue_command_exit;
5070 }
5071
5072 /* find in container array under reshape
5073 */
5074 ret_val = st->ss->load_container(st, cfd, NULL);
5075 if (ret_val) {
5076 pr_err("Cannot read superblock for %s\n",
5077 devname);
5078 ret_val = 1;
5079 goto Grow_continue_command_exit;
5080 }
5081
5082 cc = st->ss->container_content(st, subarray);
5083 for (content = cc; content ; content = content->next) {
5084 char *array;
5085 int allow_reshape = 1;
5086
5087 if (content->reshape_active == 0)
5088 continue;
5089 /* The decision about array or container wide
5090 * reshape is taken in Grow_continue based
5091 * content->reshape_active state, therefore we
5092 * need to check_reshape based on
5093 * reshape_active and subarray name
5094 */
5095 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
5096 allow_reshape = 0;
5097 if (content->reshape_active == CONTAINER_RESHAPE &&
5098 (content->array.state
5099 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
5100 allow_reshape = 0;
5101
5102 if (!allow_reshape) {
5103 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
5104 devname, container);
5105 ret_val = 1;
5106 goto Grow_continue_command_exit;
5107 }
5108
5109 array = strchr(content->text_version+1, '/')+1;
5110 mdstat = mdstat_by_subdev(array, container);
5111 if (!mdstat)
5112 continue;
5113 if (mdstat->active == 0) {
5114 pr_err("Skipping inactive array %s.\n",
5115 mdstat->devnm);
5116 free_mdstat(mdstat);
5117 mdstat = NULL;
5118 continue;
5119 }
5120 break;
5121 }
5122 if (!content) {
5123 pr_err("Unable to determine reshaped array for %s\n", devname);
5124 ret_val = 1;
5125 goto Grow_continue_command_exit;
5126 }
5127 fd2 = open_dev(mdstat->devnm);
5128 if (fd2 < 0) {
5129 pr_err("cannot open (%s)\n", mdstat->devnm);
5130 ret_val = 1;
5131 goto Grow_continue_command_exit;
5132 }
5133
5134 if (sysfs_init(content, fd2, mdstat->devnm)) {
5135 pr_err("Unable to initialize sysfs for %s, Grow cannot continue",
5136 mdstat->devnm);
5137 ret_val = 1;
5138 close(fd2);
5139 goto Grow_continue_command_exit;
5140 }
5141
5142 close(fd2);
5143
5144 /* start mdmon in case it is not running
5145 */
5146 if (!mdmon_running(container))
5147 start_mdmon(container);
5148 ping_monitor(container);
5149
5150 if (mdmon_running(container))
5151 st->update_tail = &st->updates;
5152 else {
5153 pr_err("No mdmon found. Grow cannot continue.\n");
5154 ret_val = 1;
5155 goto Grow_continue_command_exit;
5156 }
5157 }
5158
5159 /* verify that array under reshape is started from
5160 * correct position
5161 */
5162 if (verify_reshape_position(content, content->array.level) < 0) {
5163 ret_val = 1;
5164 goto Grow_continue_command_exit;
5165 }
5166
5167 /* continue reshape
5168 */
5169 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
5170
5171 Grow_continue_command_exit:
5172 if (cfd > -1)
5173 close(cfd);
5174 st->ss->free_super(st);
5175 free_mdstat(mdstat);
5176 sysfs_free(cc);
5177 free(subarray);
5178
5179 return ret_val;
5180 }
5181
5182 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
5183 char *backup_file, int forked, int freeze_reshape)
5184 {
5185 int ret_val = 2;
5186
5187 if (!info->reshape_active)
5188 return ret_val;
5189
5190 if (st->ss->external) {
5191 int cfd = open_dev(st->container_devnm);
5192
5193 if (cfd < 0)
5194 return 1;
5195
5196 st->ss->load_container(st, cfd, st->container_devnm);
5197 close(cfd);
5198 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
5199 st, info, 0, backup_file,
5200 0, forked,
5201 1 | info->reshape_active,
5202 freeze_reshape);
5203 } else
5204 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
5205 NULL, INVALID_SECTORS,
5206 backup_file, 0, forked,
5207 1 | info->reshape_active,
5208 freeze_reshape);
5209
5210 return ret_val;
5211 }
5212
5213 char *make_backup(char *name)
5214 {
5215 char *base = "backup_file-";
5216 int len;
5217 char *fname;
5218
5219 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
5220 fname = xmalloc(len);
5221 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
5222 return fname;
5223 }
5224
5225 char *locate_backup(char *name)
5226 {
5227 char *fl = make_backup(name);
5228 struct stat stb;
5229
5230 if (stat(fl, &stb) == 0 &&
5231 S_ISREG(stb.st_mode))
5232 return fl;
5233
5234 free(fl);
5235 return NULL;
5236 }
Unnamed repository; edit this file 'description' to name the repository.