]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Grow: fix reshape from RAID5 to RAID1.
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2012 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
28 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
29 #error no endian defined
30 #endif
31 #include "md_u.h"
32 #include "md_p.h"
33
34 #ifndef offsetof
35 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
36 #endif
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char *backup_file,
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
51 dprintf("Called restore_backup()\n");
52 fdlist = xmalloc(sizeof(int) * disk_count);
53
54 for (i = 0; i < next_spare; i++)
55 fdlist[i] = -1;
56 for (dev = content->devs; dev; dev = dev->next) {
57 char buf[22];
58 int fd;
59 sprintf(buf, "%d:%d",
60 dev->disk.major,
61 dev->disk.minor);
62 fd = dev_open(buf, O_RDWR);
63
64 if (dev->disk.raid_disk >= 0)
65 fdlist[dev->disk.raid_disk] = fd;
66 else
67 fdlist[next_spare++] = fd;
68 }
69
70 if (st->ss->external && st->ss->recover_backup)
71 err = st->ss->recover_backup(st, content);
72 else
73 err = Grow_restart(st, content, fdlist, next_spare,
74 backup_file, verbose > 0);
75
76 while (next_spare > 0) {
77 next_spare--;
78 if (fdlist[next_spare] >= 0)
79 close(fdlist[next_spare]);
80 }
81 free(fdlist);
82 if (err) {
83 pr_err("Failed to restore critical"
84 " section for reshape - sorry.\n");
85 if (!backup_file)
86 pr_err("Possibly you need"
87 " to specify a --backup-file\n");
88 return 1;
89 }
90
91 dprintf("restore_backup() returns status OK.\n");
92 return 0;
93 }
94
95 int Grow_Add_device(char *devname, int fd, char *newdev)
96 {
97 /* Add a device to an active array.
98 * Currently, just extend a linear array.
99 * This requires writing a new superblock on the
100 * new device, calling the kernel to add the device,
101 * and if that succeeds, update the superblock on
102 * all other devices.
103 * This means that we need to *find* all other devices.
104 */
105 struct mdinfo info;
106
107 struct stat stb;
108 int nfd, fd2;
109 int d, nd;
110 struct supertype *st = NULL;
111 char *subarray = NULL;
112
113 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
114 pr_err("cannot get array info for %s\n", devname);
115 return 1;
116 }
117
118 if (info.array.level != -1) {
119 pr_err("can only add devices to linear arrays\n");
120 return 1;
121 }
122
123 st = super_by_fd(fd, &subarray);
124 if (!st) {
125 pr_err("cannot handle arrays with superblock version %d\n",
126 info.array.major_version);
127 return 1;
128 }
129
130 if (subarray) {
131 pr_err("Cannot grow linear sub-arrays yet\n");
132 free(subarray);
133 free(st);
134 return 1;
135 }
136
137 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
138 if (nfd < 0) {
139 pr_err("cannot open %s\n", newdev);
140 free(st);
141 return 1;
142 }
143 fstat(nfd, &stb);
144 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
145 pr_err("%s is not a block device!\n", newdev);
146 close(nfd);
147 free(st);
148 return 1;
149 }
150 /* now check out all the devices and make sure we can read the
151 * superblock */
152 for (d=0 ; d < info.array.raid_disks ; d++) {
153 mdu_disk_info_t disk;
154 char *dv;
155
156 st->ss->free_super(st);
157
158 disk.number = d;
159 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
160 pr_err("cannot get device detail for device %d\n",
161 d);
162 close(nfd);
163 free(st);
164 return 1;
165 }
166 dv = map_dev(disk.major, disk.minor, 1);
167 if (!dv) {
168 pr_err("cannot find device file for device %d\n",
169 d);
170 close(nfd);
171 free(st);
172 return 1;
173 }
174 fd2 = dev_open(dv, O_RDWR);
175 if (fd2 < 0) {
176 pr_err("cannot open device file %s\n", dv);
177 close(nfd);
178 free(st);
179 return 1;
180 }
181
182 if (st->ss->load_super(st, fd2, NULL)) {
183 pr_err("cannot find super block on %s\n", dv);
184 close(nfd);
185 close(fd2);
186 free(st);
187 return 1;
188 }
189 close(fd2);
190 }
191 /* Ok, looks good. Lets update the superblock and write it out to
192 * newdev.
193 */
194
195 info.disk.number = d;
196 info.disk.major = major(stb.st_rdev);
197 info.disk.minor = minor(stb.st_rdev);
198 info.disk.raid_disk = d;
199 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
200 st->ss->update_super(st, &info, "linear-grow-new", newdev,
201 0, 0, NULL);
202
203 if (st->ss->store_super(st, nfd)) {
204 pr_err("Cannot store new superblock on %s\n",
205 newdev);
206 close(nfd);
207 return 1;
208 }
209 close(nfd);
210
211 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
212 pr_err("Cannot add new disk to this array\n");
213 return 1;
214 }
215 /* Well, that seems to have worked.
216 * Now go through and update all superblocks
217 */
218
219 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
220 pr_err("cannot get array info for %s\n", devname);
221 return 1;
222 }
223
224 nd = d;
225 for (d=0 ; d < info.array.raid_disks ; d++) {
226 mdu_disk_info_t disk;
227 char *dv;
228
229 disk.number = d;
230 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
231 pr_err("cannot get device detail for device %d\n",
232 d);
233 return 1;
234 }
235 dv = map_dev(disk.major, disk.minor, 1);
236 if (!dv) {
237 pr_err("cannot find device file for device %d\n",
238 d);
239 return 1;
240 }
241 fd2 = dev_open(dv, O_RDWR);
242 if (fd2 < 0) {
243 pr_err("cannot open device file %s\n", dv);
244 return 1;
245 }
246 if (st->ss->load_super(st, fd2, NULL)) {
247 pr_err("cannot find super block on %s\n", dv);
248 close(fd);
249 return 1;
250 }
251 info.array.raid_disks = nd+1;
252 info.array.nr_disks = nd+1;
253 info.array.active_disks = nd+1;
254 info.array.working_disks = nd+1;
255
256 st->ss->update_super(st, &info, "linear-grow-update", dv,
257 0, 0, NULL);
258
259 if (st->ss->store_super(st, fd2)) {
260 pr_err("Cannot store new superblock on %s\n", dv);
261 close(fd2);
262 return 1;
263 }
264 close(fd2);
265 }
266
267 return 0;
268 }
269
270 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
271 {
272 /*
273 * First check that array doesn't have a bitmap
274 * Then create the bitmap
275 * Then add it
276 *
277 * For internal bitmaps, we need to check the version,
278 * find all the active devices, and write the bitmap block
279 * to all devices
280 */
281 mdu_bitmap_file_t bmf;
282 mdu_array_info_t array;
283 struct supertype *st;
284 char *subarray = NULL;
285 int major = BITMAP_MAJOR_HI;
286 int vers = md_get_version(fd);
287 unsigned long long bitmapsize, array_size;
288
289 if (vers < 9003) {
290 major = BITMAP_MAJOR_HOSTENDIAN;
291 pr_err("Warning - bitmaps created on this kernel"
292 " are not portable\n"
293 " between different architectures. Consider upgrading"
294 " the Linux kernel.\n");
295 }
296
297 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
298 if (errno == ENOMEM)
299 pr_err("Memory allocation failure.\n");
300 else
301 pr_err("bitmaps not supported by this kernel.\n");
302 return 1;
303 }
304 if (bmf.pathname[0]) {
305 if (strcmp(s->bitmap_file,"none")==0) {
306 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
307 pr_err("failed to remove bitmap %s\n",
308 bmf.pathname);
309 return 1;
310 }
311 return 0;
312 }
313 pr_err("%s already has a bitmap (%s)\n",
314 devname, bmf.pathname);
315 return 1;
316 }
317 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
318 pr_err("cannot get array status for %s\n", devname);
319 return 1;
320 }
321 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
322 if (strcmp(s->bitmap_file, "none")==0) {
323 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
324 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
325 pr_err("failed to remove internal bitmap.\n");
326 return 1;
327 }
328 return 0;
329 }
330 pr_err("Internal bitmap already present on %s\n",
331 devname);
332 return 1;
333 }
334
335 if (strcmp(s->bitmap_file, "none") == 0) {
336 pr_err("no bitmap found on %s\n", devname);
337 return 1;
338 }
339 if (array.level <= 0) {
340 pr_err("Bitmaps not meaningful with level %s\n",
341 map_num(pers, array.level)?:"of this array");
342 return 1;
343 }
344 bitmapsize = array.size;
345 bitmapsize <<= 1;
346 if (get_dev_size(fd, NULL, &array_size) &&
347 array_size > (0x7fffffffULL<<9)) {
348 /* Array is big enough that we cannot trust array.size
349 * try other approaches
350 */
351 bitmapsize = get_component_size(fd);
352 }
353 if (bitmapsize == 0) {
354 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
355 return 1;
356 }
357
358 if (array.level == 10) {
359 int ncopies = (array.layout&255)*((array.layout>>8)&255);
360 bitmapsize = bitmapsize * array.raid_disks / ncopies;
361 }
362
363 st = super_by_fd(fd, &subarray);
364 if (!st) {
365 pr_err("Cannot understand version %d.%d\n",
366 array.major_version, array.minor_version);
367 return 1;
368 }
369 if (subarray) {
370 pr_err("Cannot add bitmaps to sub-arrays yet\n");
371 free(subarray);
372 free(st);
373 return 1;
374 }
375 if (strcmp(s->bitmap_file, "internal") == 0) {
376 int rv;
377 int d;
378 int offset_setable = 0;
379 struct mdinfo *mdi;
380 if (st->ss->add_internal_bitmap == NULL) {
381 pr_err("Internal bitmaps not supported "
382 "with %s metadata\n", st->ss->name);
383 return 1;
384 }
385 mdi = sysfs_read(fd, -1, GET_BITMAP_LOCATION);
386 if (mdi)
387 offset_setable = 1;
388 for (d=0; d< st->max_devs; d++) {
389 mdu_disk_info_t disk;
390 char *dv;
391 disk.number = d;
392 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
393 continue;
394 if (disk.major == 0 &&
395 disk.minor == 0)
396 continue;
397 if ((disk.state & (1<<MD_DISK_SYNC))==0)
398 continue;
399 dv = map_dev(disk.major, disk.minor, 1);
400 if (dv) {
401 int fd2 = dev_open(dv, O_RDWR);
402 if (fd2 < 0)
403 continue;
404 if (st->ss->load_super(st, fd2, NULL)==0) {
405 if (st->ss->add_internal_bitmap(
406 st,
407 &s->bitmap_chunk, c->delay, s->write_behind,
408 bitmapsize, offset_setable,
409 major)
410 )
411 st->ss->write_bitmap(st, fd2);
412 else {
413 pr_err("failed to create internal bitmap"
414 " - chunksize problem.\n");
415 close(fd2);
416 return 1;
417 }
418 }
419 close(fd2);
420 }
421 }
422 if (offset_setable) {
423 st->ss->getinfo_super(st, mdi, NULL);
424 sysfs_init(mdi, fd, -1);
425 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
426 mdi->bitmap_offset);
427 } else {
428 array.state |= (1<<MD_SB_BITMAP_PRESENT);
429 rv = ioctl(fd, SET_ARRAY_INFO, &array);
430 }
431 if (rv < 0) {
432 if (errno == EBUSY)
433 pr_err("Cannot add bitmap while array is"
434 " resyncing or reshaping etc.\n");
435 pr_err("failed to set internal bitmap.\n");
436 return 1;
437 }
438 } else {
439 int uuid[4];
440 int bitmap_fd;
441 int d;
442 int max_devs = st->max_devs;
443
444 /* try to load a superblock */
445 for (d = 0; d < max_devs; d++) {
446 mdu_disk_info_t disk;
447 char *dv;
448 int fd2;
449 disk.number = d;
450 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
451 continue;
452 if ((disk.major==0 && disk.minor==0) ||
453 (disk.state & (1<<MD_DISK_REMOVED)))
454 continue;
455 dv = map_dev(disk.major, disk.minor, 1);
456 if (!dv)
457 continue;
458 fd2 = dev_open(dv, O_RDONLY);
459 if (fd2 >= 0) {
460 if (st->ss->load_super(st, fd2, NULL) == 0) {
461 close(fd2);
462 st->ss->uuid_from_super(st, uuid);
463 break;
464 }
465 close(fd2);
466 }
467 }
468 if (d == max_devs) {
469 pr_err("cannot find UUID for array!\n");
470 return 1;
471 }
472 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
473 c->delay, s->write_behind, bitmapsize, major)) {
474 return 1;
475 }
476 bitmap_fd = open(s->bitmap_file, O_RDWR);
477 if (bitmap_fd < 0) {
478 pr_err("weird: %s cannot be opened\n",
479 s->bitmap_file);
480 return 1;
481 }
482 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
483 int err = errno;
484 if (errno == EBUSY)
485 pr_err("Cannot add bitmap while array is"
486 " resyncing or reshaping etc.\n");
487 pr_err("Cannot set bitmap file for %s: %s\n",
488 devname, strerror(err));
489 return 1;
490 }
491 }
492
493 return 0;
494 }
495
496 /*
497 * When reshaping an array we might need to backup some data.
498 * This is written to all spares with a 'super_block' describing it.
499 * The superblock goes 4K from the end of the used space on the
500 * device.
501 * It if written after the backup is complete.
502 * It has the following structure.
503 */
504
505 static struct mdp_backup_super {
506 char magic[16]; /* md_backup_data-1 or -2 */
507 __u8 set_uuid[16];
508 __u64 mtime;
509 /* start/sizes in 512byte sectors */
510 __u64 devstart; /* address on backup device/file of data */
511 __u64 arraystart;
512 __u64 length;
513 __u32 sb_csum; /* csum of preceeding bytes. */
514 __u32 pad1;
515 __u64 devstart2; /* offset in to data of second section */
516 __u64 arraystart2;
517 __u64 length2;
518 __u32 sb_csum2; /* csum of preceeding bytes. */
519 __u8 pad[512-68-32];
520 } __attribute__((aligned(512))) bsb, bsb2;
521
522 static __u32 bsb_csum(char *buf, int len)
523 {
524 int i;
525 int csum = 0;
526 for (i = 0; i < len; i++)
527 csum = (csum<<3) + buf[0];
528 return __cpu_to_le32(csum);
529 }
530
531 static int check_idle(struct supertype *st)
532 {
533 /* Check that all member arrays for this container, or the
534 * container of this array, are idle
535 */
536 int container_dev = (st->container_dev != NoMdDev
537 ? st->container_dev : st->devnum);
538 char container[40];
539 struct mdstat_ent *ent, *e;
540 int is_idle = 1;
541
542 fmt_devname(container, container_dev);
543 ent = mdstat_read(0, 0);
544 for (e = ent ; e; e = e->next) {
545 if (!is_container_member(e, container))
546 continue;
547 if (e->percent >= 0) {
548 is_idle = 0;
549 break;
550 }
551 }
552 free_mdstat(ent);
553 return is_idle;
554 }
555
556 static int freeze_container(struct supertype *st)
557 {
558 int container_dev = (st->container_dev != NoMdDev
559 ? st->container_dev : st->devnum);
560 char container[40];
561
562 if (!check_idle(st))
563 return -1;
564
565 fmt_devname(container, container_dev);
566
567 if (block_monitor(container, 1)) {
568 pr_err("failed to freeze container\n");
569 return -2;
570 }
571
572 return 1;
573 }
574
575 static void unfreeze_container(struct supertype *st)
576 {
577 int container_dev = (st->container_dev != NoMdDev
578 ? st->container_dev : st->devnum);
579 char container[40];
580
581 fmt_devname(container, container_dev);
582
583 unblock_monitor(container, 1);
584 }
585
586 static int freeze(struct supertype *st)
587 {
588 /* Try to freeze resync/rebuild on this array/container.
589 * Return -1 if the array is busy,
590 * return -2 container cannot be frozen,
591 * return 0 if this kernel doesn't support 'frozen'
592 * return 1 if it worked.
593 */
594 if (st->ss->external)
595 return freeze_container(st);
596 else {
597 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
598 int err;
599 char buf[20];
600
601 if (!sra)
602 return -1;
603 /* Need to clear any 'read-auto' status */
604 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
605 strncmp(buf, "read-auto", 9) == 0)
606 sysfs_set_str(sra, NULL, "array_state", "clean");
607
608 err = sysfs_freeze_array(sra);
609 sysfs_free(sra);
610 return err;
611 }
612 }
613
614 static void unfreeze(struct supertype *st)
615 {
616 if (st->ss->external)
617 return unfreeze_container(st);
618 else {
619 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
620
621 if (sra)
622 sysfs_set_str(sra, NULL, "sync_action", "idle");
623 sysfs_free(sra);
624 }
625 }
626
627 static void wait_reshape(struct mdinfo *sra)
628 {
629 int fd = sysfs_get_fd(sra, NULL, "sync_action");
630 char action[20];
631
632 if (fd < 0)
633 return;
634
635 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
636 strncmp(action, "reshape", 7) == 0) {
637 fd_set rfds;
638 FD_ZERO(&rfds);
639 FD_SET(fd, &rfds);
640 select(fd+1, NULL, NULL, &rfds, NULL);
641 }
642 close(fd);
643 }
644
645 static int reshape_super(struct supertype *st, unsigned long long size,
646 int level, int layout, int chunksize, int raid_disks,
647 int delta_disks, char *backup_file, char *dev,
648 int direction, int verbose)
649 {
650 /* nothing extra to check in the native case */
651 if (!st->ss->external)
652 return 0;
653 if (!st->ss->reshape_super ||
654 !st->ss->manage_reshape) {
655 pr_err("%s metadata does not support reshape\n",
656 st->ss->name);
657 return 1;
658 }
659
660 return st->ss->reshape_super(st, size, level, layout, chunksize,
661 raid_disks, delta_disks, backup_file, dev,
662 direction, verbose);
663 }
664
665 static void sync_metadata(struct supertype *st)
666 {
667 if (st->ss->external) {
668 if (st->update_tail) {
669 flush_metadata_updates(st);
670 st->update_tail = &st->updates;
671 } else
672 st->ss->sync_metadata(st);
673 }
674 }
675
676 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
677 {
678 /* when dealing with external metadata subarrays we need to be
679 * prepared to handle EAGAIN. The kernel may need to wait for
680 * mdmon to mark the array active so the kernel can handle
681 * allocations/writeback when preparing the reshape action
682 * (md_allow_write()). We temporarily disable safe_mode_delay
683 * to close a race with the array_state going clean before the
684 * next write to raid_disks / stripe_cache_size
685 */
686 char safe[50];
687 int rc;
688
689 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
690 if (!container ||
691 (strcmp(name, "raid_disks") != 0 &&
692 strcmp(name, "stripe_cache_size") != 0))
693 return sysfs_set_num(sra, NULL, name, n);
694
695 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
696 if (rc <= 0)
697 return -1;
698 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
699 rc = sysfs_set_num(sra, NULL, name, n);
700 if (rc < 0 && errno == EAGAIN) {
701 ping_monitor(container);
702 /* if we get EAGAIN here then the monitor is not active
703 * so stop trying
704 */
705 rc = sysfs_set_num(sra, NULL, name, n);
706 }
707 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
708 return rc;
709 }
710
711 int start_reshape(struct mdinfo *sra, int already_running,
712 int before_data_disks, int data_disks)
713 {
714 int err;
715 unsigned long long sync_max_to_set;
716
717 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
718 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
719 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
720 sra->reshape_progress);
721 if (before_data_disks <= data_disks)
722 sync_max_to_set = sra->reshape_progress / data_disks;
723 else
724 sync_max_to_set = (sra->component_size * data_disks
725 - sra->reshape_progress) / data_disks;
726 if (!already_running)
727 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
728 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
729 if (!already_running)
730 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
731
732 return err;
733 }
734
735 void abort_reshape(struct mdinfo *sra)
736 {
737 sysfs_set_str(sra, NULL, "sync_action", "idle");
738 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
739 sysfs_set_num(sra, NULL, "suspend_hi", 0);
740 sysfs_set_num(sra, NULL, "suspend_lo", 0);
741 sysfs_set_num(sra, NULL, "sync_min", 0);
742 sysfs_set_str(sra, NULL, "sync_max", "max");
743 }
744
745 int remove_disks_for_takeover(struct supertype *st,
746 struct mdinfo *sra,
747 int layout)
748 {
749 int nr_of_copies;
750 struct mdinfo *remaining;
751 int slot;
752
753 if (sra->array.level == 10)
754 nr_of_copies = layout & 0xff;
755 else if (sra->array.level == 1)
756 nr_of_copies = sra->array.raid_disks;
757 else
758 return 1;
759
760 remaining = sra->devs;
761 sra->devs = NULL;
762 /* for each 'copy', select one device and remove from the list. */
763 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
764 struct mdinfo **diskp;
765 int found = 0;
766
767 /* Find a working device to keep */
768 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
769 struct mdinfo *disk = *diskp;
770
771 if (disk->disk.raid_disk < slot)
772 continue;
773 if (disk->disk.raid_disk >= slot + nr_of_copies)
774 continue;
775 if (disk->disk.state & (1<<MD_DISK_REMOVED))
776 continue;
777 if (disk->disk.state & (1<<MD_DISK_FAULTY))
778 continue;
779 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
780 continue;
781
782 /* We have found a good disk to use! */
783 *diskp = disk->next;
784 disk->next = sra->devs;
785 sra->devs = disk;
786 found = 1;
787 break;
788 }
789 if (!found)
790 break;
791 }
792
793 if (slot < sra->array.raid_disks) {
794 /* didn't find all slots */
795 struct mdinfo **e;
796 e = &remaining;
797 while (*e)
798 e = &(*e)->next;
799 *e = sra->devs;
800 sra->devs = remaining;
801 return 1;
802 }
803
804 /* Remove all 'remaining' devices from the array */
805 while (remaining) {
806 struct mdinfo *sd = remaining;
807 remaining = sd->next;
808
809 sysfs_set_str(sra, sd, "state", "faulty");
810 sysfs_set_str(sra, sd, "slot", "none");
811 /* for external metadata disks should be removed in mdmon */
812 if (!st->ss->external)
813 sysfs_set_str(sra, sd, "state", "remove");
814 sd->disk.state |= (1<<MD_DISK_REMOVED);
815 sd->disk.state &= ~(1<<MD_DISK_SYNC);
816 sd->next = sra->devs;
817 sra->devs = sd;
818 }
819 return 0;
820 }
821
822 void reshape_free_fdlist(int *fdlist,
823 unsigned long long *offsets,
824 int size)
825 {
826 int i;
827
828 for (i = 0; i < size; i++)
829 if (fdlist[i] >= 0)
830 close(fdlist[i]);
831
832 free(fdlist);
833 free(offsets);
834 }
835
836 int reshape_prepare_fdlist(char *devname,
837 struct mdinfo *sra,
838 int raid_disks,
839 int nrdisks,
840 unsigned long blocks,
841 char *backup_file,
842 int *fdlist,
843 unsigned long long *offsets)
844 {
845 int d = 0;
846 struct mdinfo *sd;
847
848 for (d = 0; d <= nrdisks; d++)
849 fdlist[d] = -1;
850 d = raid_disks;
851 for (sd = sra->devs; sd; sd = sd->next) {
852 if (sd->disk.state & (1<<MD_DISK_FAULTY))
853 continue;
854 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
855 char *dn = map_dev(sd->disk.major,
856 sd->disk.minor, 1);
857 fdlist[sd->disk.raid_disk]
858 = dev_open(dn, O_RDONLY);
859 offsets[sd->disk.raid_disk] = sd->data_offset*512;
860 if (fdlist[sd->disk.raid_disk] < 0) {
861 pr_err("%s: cannot open component %s\n",
862 devname, dn ? dn : "-unknown-");
863 d = -1;
864 goto release;
865 }
866 } else if (backup_file == NULL) {
867 /* spare */
868 char *dn = map_dev(sd->disk.major,
869 sd->disk.minor, 1);
870 fdlist[d] = dev_open(dn, O_RDWR);
871 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
872 if (fdlist[d] < 0) {
873 pr_err("%s: cannot open component %s\n",
874 devname, dn ? dn : "-unknown-");
875 d = -1;
876 goto release;
877 }
878 d++;
879 }
880 }
881 release:
882 return d;
883 }
884
885 int reshape_open_backup_file(char *backup_file,
886 int fd,
887 char *devname,
888 long blocks,
889 int *fdlist,
890 unsigned long long *offsets,
891 int restart)
892 {
893 /* Return 1 on success, 0 on any form of failure */
894 /* need to check backup file is large enough */
895 char buf[512];
896 struct stat stb;
897 unsigned int dev;
898 int i;
899
900 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
901 S_IRUSR | S_IWUSR);
902 *offsets = 8 * 512;
903 if (*fdlist < 0) {
904 pr_err("%s: cannot create backup file %s: %s\n",
905 devname, backup_file, strerror(errno));
906 return 0;
907 }
908 /* Guard against backup file being on array device.
909 * If array is partitioned or if LVM etc is in the
910 * way this will not notice, but it is better than
911 * nothing.
912 */
913 fstat(*fdlist, &stb);
914 dev = stb.st_dev;
915 fstat(fd, &stb);
916 if (stb.st_rdev == dev) {
917 pr_err("backup file must NOT be"
918 " on the array being reshaped.\n");
919 close(*fdlist);
920 return 0;
921 }
922
923 memset(buf, 0, 512);
924 for (i=0; i < blocks + 8 ; i++) {
925 if (write(*fdlist, buf, 512) != 512) {
926 pr_err("%s: cannot create"
927 " backup file %s: %s\n",
928 devname, backup_file, strerror(errno));
929 return 0;
930 }
931 }
932 if (fsync(*fdlist) != 0) {
933 pr_err("%s: cannot create backup file %s: %s\n",
934 devname, backup_file, strerror(errno));
935 return 0;
936 }
937
938 return 1;
939 }
940
941 unsigned long GCD(unsigned long a, unsigned long b)
942 {
943 while (a != b) {
944 if (a < b)
945 b -= a;
946 if (b < a)
947 a -= b;
948 }
949 return a;
950 }
951
952 unsigned long compute_backup_blocks(int nchunk, int ochunk,
953 unsigned int ndata, unsigned int odata)
954 {
955 unsigned long a, b, blocks;
956 /* So how much do we need to backup.
957 * We need an amount of data which is both a whole number of
958 * old stripes and a whole number of new stripes.
959 * So LCM for (chunksize*datadisks).
960 */
961 a = (ochunk/512) * odata;
962 b = (nchunk/512) * ndata;
963 /* Find GCD */
964 a = GCD(a, b);
965 /* LCM == product / GCD */
966 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
967
968 return blocks;
969 }
970
971 char *analyse_change(struct mdinfo *info, struct reshape *re)
972 {
973 /* Based on the current array state in info->array and
974 * the changes in info->new_* etc, determine:
975 * - whether the change is possible
976 * - Intermediate level/raid_disks/layout
977 * - whether a restriping reshape is needed
978 * - number of sectors in minimum change unit. This
979 * will cover a whole number of stripes in 'before' and
980 * 'after'.
981 *
982 * Return message if the change should be rejected
983 * NULL if the change can be achieved
984 *
985 * This can be called as part of starting a reshape, or
986 * when assembling an array that is undergoing reshape.
987 */
988 int near, far, offset, copies;
989 int new_disks;
990 int old_chunk, new_chunk;
991 /* delta_parity records change in number of devices
992 * caused by level change
993 */
994 int delta_parity = 0;
995
996 /* If a new level not explicitly given, we assume no-change */
997 if (info->new_level == UnSet)
998 info->new_level = info->array.level;
999
1000 if (info->new_chunk)
1001 switch (info->new_level) {
1002 case 0:
1003 case 4:
1004 case 5:
1005 case 6:
1006 case 10:
1007 /* chunk size is meaningful, must divide component_size
1008 * evenly
1009 */
1010 if (info->component_size % (info->new_chunk/512))
1011 return "New chunk size does not"
1012 " divide component size";
1013 break;
1014 default:
1015 return "chunk size not meaningful for this level";
1016 }
1017 else
1018 info->new_chunk = info->array.chunk_size;
1019
1020 switch (info->array.level) {
1021 default:
1022 return "Cannot understand this RAID level";
1023 case 1:
1024 /* RAID1 can convert to RAID1 with different disks, or
1025 * raid5 with 2 disks, or
1026 * raid0 with 1 disk
1027 */
1028 if (info->new_level > 1 &&
1029 (info->component_size & 7))
1030 return "Cannot convert RAID1 of this size - "
1031 "reduce size to multiple of 4K first.";
1032 if (info->new_level == 0) {
1033 if (info->delta_disks != UnSet &&
1034 info->delta_disks != 0)
1035 return "Cannot change number of disks "
1036 "with RAID1->RAID0 conversion";
1037 re->level = 0;
1038 re->before.data_disks = 1;
1039 re->after.data_disks = 1;
1040 re->before.layout = 0;
1041 re->backup_blocks = 0;
1042 re->parity = 0;
1043 return NULL;
1044 }
1045 if (info->new_level == 1) {
1046 if (info->delta_disks == UnSet)
1047 /* Don't know what to do */
1048 return "no change requested for Growing RAID1";
1049 re->level = 1;
1050 re->backup_blocks = 0;
1051 re->parity = 0;
1052 return NULL;
1053 }
1054 if (info->array.raid_disks == 2 &&
1055 info->new_level == 5) {
1056
1057 re->level = 5;
1058 re->before.data_disks = 1;
1059 if (info->delta_disks != UnSet &&
1060 info->delta_disks != 0)
1061 re->after.data_disks = 1 + info->delta_disks;
1062 else
1063 re->after.data_disks = 1;
1064 if (re->after.data_disks < 1)
1065 return "Number of disks too small for RAID5";
1066
1067 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1068 info->array.chunk_size = 65536;
1069 break;
1070 }
1071 /* Could do some multi-stage conversions, but leave that to
1072 * later.
1073 */
1074 return "Impossibly level change request for RAID1";
1075
1076 case 10:
1077 /* RAID10 can be converted from near mode to
1078 * RAID0 by removing some devices.
1079 * It can also be reshaped if the kernel supports
1080 * new_data_offset.
1081 */
1082 switch (info->new_level) {
1083 case 0:
1084 if ((info->array.layout & ~0xff) != 0x100)
1085 return "Cannot Grow RAID10 with far/offset layout";
1086 /* number of devices must be multiple of number of copies */
1087 if (info->array.raid_disks % (info->array.layout & 0xff))
1088 return "RAID10 layout too complex for Grow operation";
1089
1090 new_disks = (info->array.raid_disks
1091 / (info->array.layout & 0xff));
1092 if (info->delta_disks == UnSet)
1093 info->delta_disks = (new_disks
1094 - info->array.raid_disks);
1095
1096 if (info->delta_disks != new_disks - info->array.raid_disks)
1097 return "New number of raid-devices impossible for RAID10";
1098 if (info->new_chunk &&
1099 info->new_chunk != info->array.chunk_size)
1100 return "Cannot change chunk-size with RAID10 Grow";
1101
1102 /* looks good */
1103 re->level = 0;
1104 re->parity = 0;
1105 re->before.data_disks = new_disks;
1106 re->after.data_disks = re->before.data_disks;
1107 re->before.layout = 0;
1108 re->backup_blocks = 0;
1109 return NULL;
1110
1111 case 10:
1112 near = info->array.layout & 0xff;
1113 far = (info->array.layout >> 8) & 0xff;
1114 offset = info->array.layout & 0x10000;
1115 if (far > 1 && !offset)
1116 return "Cannot reshape RAID10 in far-mode";
1117 copies = near * far;
1118
1119 old_chunk = info->array.chunk_size * far;
1120
1121 if (info->new_layout == UnSet)
1122 info->new_layout = info->array.layout;
1123 else {
1124 near = info->new_layout & 0xff;
1125 far = (info->new_layout >> 8) & 0xff;
1126 offset = info->new_layout & 0x10000;
1127 if (far > 1 && !offset)
1128 return "Cannot reshape RAID10 to far-mode";
1129 if (near * far != copies)
1130 return "Cannot change number of copies"
1131 " when reshaping RAID10";
1132 }
1133 if (info->delta_disks == UnSet)
1134 info->delta_disks = 0;
1135 new_disks = (info->array.raid_disks +
1136 info->delta_disks);
1137
1138 new_chunk = info->new_chunk * far;
1139
1140 re->level = 10;
1141 re->parity = 0;
1142 re->before.layout = info->array.layout;
1143 re->before.data_disks = info->array.raid_disks;
1144 re->after.layout = info->new_layout;
1145 re->after.data_disks = new_disks;
1146 /* For RAID10 we don't do backup, and there is
1147 * no need to synchronise stripes on both
1148 * 'old' and 'new'. So the important
1149 * number is the minimum data_offset difference
1150 * which is the larger of (offset copies * chunk).
1151 */
1152
1153 re->backup_blocks = max(old_chunk, new_chunk) / 512;
1154 if (new_disks < re->before.data_disks &&
1155 info->space_after < re->backup_blocks)
1156 /* Reduce component size by one chunk */
1157 re->new_size = (info->component_size -
1158 re->backup_blocks);
1159 else
1160 re->new_size = info->component_size;
1161 re->new_size = re->new_size * new_disks / copies;
1162 return NULL;
1163
1164 default:
1165 return "RAID10 can only be changed to RAID0";
1166 }
1167 case 0:
1168 /* RAID0 can be converted to RAID10, or to RAID456 */
1169 if (info->new_level == 10) {
1170 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1171 /* Assume near=2 layout */
1172 info->new_layout = 0x102;
1173 info->delta_disks = info->array.raid_disks;
1174 }
1175 if (info->new_layout == UnSet) {
1176 int copies = 1 + (info->delta_disks
1177 / info->array.raid_disks);
1178 if (info->array.raid_disks * (copies-1)
1179 != info->delta_disks)
1180 return "Impossible number of devices"
1181 " for RAID0->RAID10";
1182 info->new_layout = 0x100 + copies;
1183 }
1184 if (info->delta_disks == UnSet) {
1185 int copies = info->new_layout & 0xff;
1186 if (info->new_layout != 0x100 + copies)
1187 return "New layout impossible"
1188 " for RAID0->RAID10";;
1189 info->delta_disks = (copies - 1) *
1190 info->array.raid_disks;
1191 }
1192 if (info->new_chunk &&
1193 info->new_chunk != info->array.chunk_size)
1194 return "Cannot change chunk-size with RAID0->RAID10";
1195 /* looks good */
1196 re->level = 10;
1197 re->parity = 0;
1198 re->before.data_disks = (info->array.raid_disks +
1199 info->delta_disks);
1200 re->after.data_disks = re->before.data_disks;
1201 re->before.layout = info->new_layout;
1202 re->backup_blocks = 0;
1203 return NULL;
1204 }
1205
1206 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1207 * a raid4 style layout of the final level.
1208 */
1209 switch (info->new_level) {
1210 case 4:
1211 delta_parity = 1;
1212 case 0:
1213 re->level = 4;
1214 re->before.layout = 0;
1215 break;
1216 case 5:
1217 delta_parity = 1;
1218 re->level = 5;
1219 re->before.layout = ALGORITHM_PARITY_N;
1220 break;
1221 case 6:
1222 delta_parity = 2;
1223 re->level = 6;
1224 re->before.layout = ALGORITHM_PARITY_N;
1225 break;
1226 default:
1227 return "Impossible level change requested";
1228 }
1229 re->before.data_disks = info->array.raid_disks;
1230 /* determining 'after' layout happens outside this 'switch' */
1231 break;
1232
1233 case 4:
1234 info->array.layout = ALGORITHM_PARITY_N;
1235 case 5:
1236 switch (info->new_level) {
1237 case 0:
1238 delta_parity = -1;
1239 case 4:
1240 re->level = info->array.level;
1241 re->before.data_disks = info->array.raid_disks - 1;
1242 re->before.layout = info->array.layout;
1243 break;
1244 case 5:
1245 re->level = 5;
1246 re->before.data_disks = info->array.raid_disks - 1;
1247 re->before.layout = info->array.layout;
1248 break;
1249 case 6:
1250 delta_parity = 1;
1251 re->level = 6;
1252 re->before.data_disks = info->array.raid_disks - 1;
1253 switch (info->array.layout) {
1254 case ALGORITHM_LEFT_ASYMMETRIC:
1255 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1256 break;
1257 case ALGORITHM_RIGHT_ASYMMETRIC:
1258 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1259 break;
1260 case ALGORITHM_LEFT_SYMMETRIC:
1261 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1262 break;
1263 case ALGORITHM_RIGHT_SYMMETRIC:
1264 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1265 break;
1266 case ALGORITHM_PARITY_0:
1267 re->before.layout = ALGORITHM_PARITY_0_6;
1268 break;
1269 case ALGORITHM_PARITY_N:
1270 re->before.layout = ALGORITHM_PARITY_N_6;
1271 break;
1272 default:
1273 return "Cannot convert an array with this layout";
1274 }
1275 break;
1276 case 1:
1277 if (info->array.raid_disks != 2)
1278 return "Can only convert a 2-device array to RAID1";
1279 if (info->delta_disks != UnSet &&
1280 info->delta_disks != 0)
1281 return "Cannot set raid_disk when "
1282 "converting RAID5->RAID1";
1283 re->level = 1;
1284 re->backup_blocks = 0;
1285 info->new_chunk = 0;
1286 return NULL;
1287 default:
1288 return "Impossible level change requested";
1289 }
1290 break;
1291 case 6:
1292 switch (info->new_level) {
1293 case 4:
1294 case 5:
1295 delta_parity = -1;
1296 case 6:
1297 re->level = 6;
1298 re->before.data_disks = info->array.raid_disks - 2;
1299 re->before.layout = info->array.layout;
1300 break;
1301 default:
1302 return "Impossible level change requested";
1303 }
1304 break;
1305 }
1306
1307 /* If we reached here then it looks like a re-stripe is
1308 * happening. We have determined the intermediate level
1309 * and initial raid_disks/layout and stored these in 're'.
1310 *
1311 * We need to deduce the final layout that can be atomically
1312 * converted to the end state.
1313 */
1314 switch (info->new_level) {
1315 case 0:
1316 /* We can only get to RAID0 from RAID4 or RAID5
1317 * with appropriate layout and one extra device
1318 */
1319 if (re->level != 4 && re->level != 5)
1320 return "Cannot covert to RAID0 from this level";
1321
1322 switch (re->level) {
1323 case 4:
1324 re->after.layout = 0;
1325 break;
1326 case 5:
1327 re->after.layout = ALGORITHM_PARITY_N;
1328 break;
1329 }
1330 break;
1331
1332 case 4:
1333 /* We can only get to RAID4 from RAID5 */
1334 if (re->level != 4 && re->level != 5)
1335 return "Cannot convert to RAID4 from this level";
1336
1337 switch (re->level) {
1338 case 4:
1339 re->after.layout = 0;
1340 break;
1341 case 5:
1342 re->after.layout = ALGORITHM_PARITY_N;
1343 break;
1344 }
1345 break;
1346
1347 case 5:
1348 /* We get to RAID5 from RAID5 or RAID6 */
1349 if (re->level != 5 && re->level != 6)
1350 return "Cannot convert to RAID5 from this level";
1351
1352 switch (re->level) {
1353 case 5:
1354 if (info->new_layout == UnSet)
1355 re->after.layout = re->before.layout;
1356 else
1357 re->after.layout = info->new_layout;
1358 break;
1359 case 6:
1360 if (info->new_layout == UnSet)
1361 info->new_layout = re->before.layout;
1362
1363 /* after.layout needs to be raid6 version of new_layout */
1364 if (info->new_layout == ALGORITHM_PARITY_N)
1365 re->after.layout = ALGORITHM_PARITY_N;
1366 else {
1367 char layout[40];
1368 char *ls = map_num(r5layout, info->new_layout);
1369 int l;
1370 if (ls) {
1371 /* Current RAID6 layout has a RAID5
1372 * equivalent - good
1373 */
1374 strcat(strcpy(layout, ls), "-6");
1375 l = map_name(r6layout, layout);
1376 if (l == UnSet)
1377 return "Cannot find RAID6 layout"
1378 " to convert to";
1379 } else {
1380 /* Current RAID6 has no equivalent.
1381 * If it is already a '-6' layout we
1382 * can leave it unchanged, else we must
1383 * fail
1384 */
1385 ls = map_num(r6layout, info->new_layout);
1386 if (!ls ||
1387 strcmp(ls+strlen(ls)-2, "-6") != 0)
1388 return "Please specify new layout";
1389 l = info->new_layout;
1390 }
1391 re->after.layout = l;
1392 }
1393 }
1394 break;
1395
1396 case 6:
1397 /* We must already be at level 6 */
1398 if (re->level != 6)
1399 return "Impossible level change";
1400 if (info->new_layout == UnSet)
1401 re->after.layout = info->array.layout;
1402 else
1403 re->after.layout = info->new_layout;
1404 break;
1405 default:
1406 return "Impossible level change requested";
1407 }
1408 if (info->delta_disks == UnSet)
1409 info->delta_disks = delta_parity;
1410
1411 re->after.data_disks = (re->before.data_disks
1412 + info->delta_disks
1413 - delta_parity);
1414 switch (re->level) {
1415 case 6: re->parity = 2;
1416 break;
1417 case 4:
1418 case 5: re->parity = 1;
1419 break;
1420 default: re->parity = 0;
1421 break;
1422 }
1423 /* So we have a restripe operation, we need to calculate the number
1424 * of blocks per reshape operation.
1425 */
1426 if (info->new_chunk == 0)
1427 info->new_chunk = info->array.chunk_size;
1428 if (re->after.data_disks == re->before.data_disks &&
1429 re->after.layout == re->before.layout &&
1430 info->new_chunk == info->array.chunk_size) {
1431 /* Nothing to change */
1432 re->backup_blocks = 0;
1433 return NULL;
1434 }
1435 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1436 /* chunk and layout changes make no difference */
1437 re->backup_blocks = 0;
1438 return NULL;
1439 }
1440
1441 if (re->after.data_disks == re->before.data_disks &&
1442 get_linux_version() < 2006032)
1443 return "in-place reshape is not safe before 2.6.32 - sorry.";
1444
1445 if (re->after.data_disks < re->before.data_disks &&
1446 get_linux_version() < 2006030)
1447 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1448
1449 re->backup_blocks = compute_backup_blocks(
1450 info->new_chunk, info->array.chunk_size,
1451 re->after.data_disks,
1452 re->before.data_disks);
1453
1454 re->new_size = info->component_size * re->after.data_disks;
1455 return NULL;
1456 }
1457
1458 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1459 char *text_version)
1460 {
1461 struct mdinfo *info;
1462 char *subarray;
1463 int ret_val = -1;
1464
1465 if ((st == NULL) || (sra == NULL))
1466 return ret_val;
1467
1468 if (text_version == NULL)
1469 text_version = sra->text_version;
1470 subarray = strchr(text_version+1, '/')+1;
1471 info = st->ss->container_content(st, subarray);
1472 if (info) {
1473 unsigned long long current_size = 0;
1474 unsigned long long new_size =
1475 info->custom_array_size/2;
1476
1477 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1478 new_size > current_size) {
1479 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1480 < 0)
1481 dprintf("Error: Cannot set array size");
1482 else {
1483 ret_val = 0;
1484 dprintf("Array size changed");
1485 }
1486 dprintf(" from %llu to %llu.\n",
1487 current_size, new_size);
1488 }
1489 sysfs_free(info);
1490 } else
1491 dprintf("Error: set_array_size(): info pointer in NULL\n");
1492
1493 return ret_val;
1494 }
1495
1496 static int reshape_array(char *container, int fd, char *devname,
1497 struct supertype *st, struct mdinfo *info,
1498 int force, struct mddev_dev *devlist,
1499 unsigned long long data_offset,
1500 char *backup_file, int verbose, int forked,
1501 int restart, int freeze_reshape);
1502 static int reshape_container(char *container, char *devname,
1503 int mdfd,
1504 struct supertype *st,
1505 struct mdinfo *info,
1506 int force,
1507 char *backup_file,
1508 int verbose, int restart, int freeze_reshape);
1509
1510 int Grow_reshape(char *devname, int fd,
1511 struct mddev_dev *devlist,
1512 unsigned long long data_offset,
1513 struct context *c, struct shape *s)
1514 {
1515 /* Make some changes in the shape of an array.
1516 * The kernel must support the change.
1517 *
1518 * There are three different changes. Each can trigger
1519 * a resync or recovery so we freeze that until we have
1520 * requested everything (if kernel supports freezing - 2.6.30).
1521 * The steps are:
1522 * - change size (i.e. component_size)
1523 * - change level
1524 * - change layout/chunksize/ndisks
1525 *
1526 * The last can require a reshape. It is different on different
1527 * levels so we need to check the level before actioning it.
1528 * Some times the level change needs to be requested after the
1529 * reshape (e.g. raid6->raid5, raid5->raid0)
1530 *
1531 */
1532 struct mdu_array_info_s array;
1533 int rv = 0;
1534 struct supertype *st;
1535 char *subarray = NULL;
1536
1537 int frozen;
1538 int changed = 0;
1539 char *container = NULL;
1540 char container_buf[20];
1541 int cfd = -1;
1542
1543 struct mddev_dev *dv;
1544 int added_disks;
1545
1546 struct mdinfo info;
1547 struct mdinfo *sra;
1548
1549
1550 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1551 fprintf(stderr, Name ": %s is not an active md array - aborting\n",
1552 devname);
1553 return 1;
1554 }
1555 if (data_offset != INVALID_SECTORS && array.level != 10) {
1556 pr_err("--grow --data-offset not yet supported\n");
1557 return 1;
1558 }
1559
1560 if (s->size > 0 &&
1561 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1562 pr_err("cannot change component size at the same time "
1563 "as other changes.\n"
1564 " Change size first, then check data is intact before "
1565 "making other changes.\n");
1566 return 1;
1567 }
1568
1569 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1570 get_linux_version() < 2006032 &&
1571 !check_env("MDADM_FORCE_FEWER")) {
1572 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1573 " Please use a newer kernel\n");
1574 return 1;
1575 }
1576
1577 st = super_by_fd(fd, &subarray);
1578 if (!st) {
1579 pr_err("Unable to determine metadata format for %s\n", devname);
1580 return 1;
1581 }
1582 if (s->raiddisks > st->max_devs) {
1583 pr_err("Cannot increase raid-disks on this array"
1584 " beyond %d\n", st->max_devs);
1585 return 1;
1586 }
1587
1588 /* in the external case we need to check that the requested reshape is
1589 * supported, and perform an initial check that the container holds the
1590 * pre-requisite spare devices (mdmon owns final validation)
1591 */
1592 if (st->ss->external) {
1593 int container_dev;
1594 int rv;
1595
1596 if (subarray) {
1597 container_dev = st->container_dev;
1598 cfd = open_dev_excl(st->container_dev);
1599 } else {
1600 container_dev = st->devnum;
1601 close(fd);
1602 cfd = open_dev_excl(st->devnum);
1603 fd = cfd;
1604 }
1605 if (cfd < 0) {
1606 pr_err("Unable to open container for %s\n",
1607 devname);
1608 free(subarray);
1609 return 1;
1610 }
1611
1612 fmt_devname(container_buf, container_dev);
1613 container = container_buf;
1614
1615 rv = st->ss->load_container(st, cfd, NULL);
1616
1617 if (rv) {
1618 pr_err("Cannot read superblock for %s\n",
1619 devname);
1620 free(subarray);
1621 return 1;
1622 }
1623
1624 /* check if operation is supported for metadata handler */
1625 if (st->ss->container_content) {
1626 struct mdinfo *cc = NULL;
1627 struct mdinfo *content = NULL;
1628
1629 cc = st->ss->container_content(st, subarray);
1630 for (content = cc; content ; content = content->next) {
1631 int allow_reshape = 1;
1632
1633 /* check if reshape is allowed based on metadata
1634 * indications stored in content.array.status
1635 */
1636 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1637 allow_reshape = 0;
1638 if (content->array.state
1639 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1640 allow_reshape = 0;
1641 if (!allow_reshape) {
1642 pr_err("cannot reshape arrays in"
1643 " container with unsupported"
1644 " metadata: %s(%s)\n",
1645 devname, container_buf);
1646 sysfs_free(cc);
1647 free(subarray);
1648 return 1;
1649 }
1650 }
1651 sysfs_free(cc);
1652 }
1653 if (mdmon_running(container_dev))
1654 st->update_tail = &st->updates;
1655 }
1656
1657 added_disks = 0;
1658 for (dv = devlist; dv; dv = dv->next)
1659 added_disks++;
1660 if (s->raiddisks > array.raid_disks &&
1661 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1662 !c->force) {
1663 pr_err("Need %d spare%s to avoid degraded array,"
1664 " and only have %d.\n"
1665 " Use --force to over-ride this check.\n",
1666 s->raiddisks - array.raid_disks,
1667 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1668 array.spare_disks + added_disks);
1669 return 1;
1670 }
1671
1672 sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS
1673 | GET_STATE | GET_VERSION);
1674 if (sra) {
1675 if (st->ss->external && subarray == NULL) {
1676 array.level = LEVEL_CONTAINER;
1677 sra->array.level = LEVEL_CONTAINER;
1678 }
1679 } else {
1680 pr_err("failed to read sysfs parameters for %s\n",
1681 devname);
1682 return 1;
1683 }
1684 frozen = freeze(st);
1685 if (frozen < -1) {
1686 /* freeze() already spewed the reason */
1687 sysfs_free(sra);
1688 return 1;
1689 } else if (frozen < 0) {
1690 pr_err("%s is performing resync/recovery and cannot"
1691 " be reshaped\n", devname);
1692 sysfs_free(sra);
1693 return 1;
1694 }
1695
1696 /* ========= set size =============== */
1697 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1698 unsigned long long orig_size = get_component_size(fd)/2;
1699 unsigned long long min_csize;
1700 struct mdinfo *mdi;
1701 int raid0_takeover = 0;
1702
1703 if (orig_size == 0)
1704 orig_size = (unsigned) array.size;
1705
1706 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1707 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1708 rv = 1;
1709 goto release;
1710 }
1711 sync_metadata(st);
1712 if (st->ss->external) {
1713 /* metadata can have size limitation
1714 * update size value according to metadata information
1715 */
1716 struct mdinfo *sizeinfo =
1717 st->ss->container_content(st, subarray);
1718 if (sizeinfo) {
1719 unsigned long long new_size =
1720 sizeinfo->custom_array_size/2;
1721 int data_disks = get_data_disks(
1722 sizeinfo->array.level,
1723 sizeinfo->array.layout,
1724 sizeinfo->array.raid_disks);
1725 new_size /= data_disks;
1726 dprintf("Metadata size correction from %llu to "
1727 "%llu (%llu)\n", orig_size, new_size,
1728 new_size * data_disks);
1729 s->size = new_size;
1730 sysfs_free(sizeinfo);
1731 }
1732 }
1733
1734 /* Update the size of each member device in case
1735 * they have been resized. This will never reduce
1736 * below the current used-size. The "size" attribute
1737 * understands '0' to mean 'max'.
1738 */
1739 min_csize = 0;
1740 rv = 0;
1741 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1742 if (sysfs_set_num(sra, mdi, "size",
1743 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1744 /* Probably kernel refusing to let us
1745 * reduce the size - not an error.
1746 */
1747 break;
1748 }
1749 if (array.not_persistent == 0 &&
1750 array.major_version == 0 &&
1751 get_linux_version() < 3001000) {
1752 /* Dangerous to allow size to exceed 2TB */
1753 unsigned long long csize;
1754 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1755 if (csize >= 2ULL*1024*1024*1024)
1756 csize = 2ULL*1024*1024*1024;
1757 if ((min_csize == 0 || (min_csize
1758 > csize)))
1759 min_csize = csize;
1760 }
1761 }
1762 }
1763 if (rv) {
1764 pr_err("Cannot set size on "
1765 "array members.\n");
1766 goto size_change_error;
1767 }
1768 if (min_csize && s->size > min_csize) {
1769 pr_err("Cannot safely make this array "
1770 "use more than 2TB per device on this kernel.\n");
1771 rv = 1;
1772 goto size_change_error;
1773 }
1774 if (min_csize && s->size == MAX_SIZE) {
1775 /* Don't let the kernel choose a size - it will get
1776 * it wrong
1777 */
1778 pr_err("Limited v0.90 array to "
1779 "2TB per device\n");
1780 s->size = min_csize;
1781 }
1782 if (st->ss->external) {
1783 if (sra->array.level == 0) {
1784 rv = sysfs_set_str(sra, NULL, "level",
1785 "raid5");
1786 if (!rv) {
1787 raid0_takeover = 1;
1788 /* get array parametes after takeover
1789 * to chane one parameter at time only
1790 */
1791 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1792 }
1793 }
1794 /* make sure mdmon is
1795 * aware of the new level */
1796 if (!mdmon_running(st->container_dev))
1797 start_mdmon(st->container_dev);
1798 ping_monitor(container);
1799 if (mdmon_running(st->container_dev) &&
1800 st->update_tail == NULL)
1801 st->update_tail = &st->updates;
1802 }
1803
1804 if (s->size == MAX_SIZE)
1805 s->size = 0;
1806 array.size = s->size;
1807 if ((unsigned)array.size != s->size) {
1808 /* got truncated to 32bit, write to
1809 * component_size instead
1810 */
1811 if (sra)
1812 rv = sysfs_set_num(sra, NULL,
1813 "component_size", s->size);
1814 else
1815 rv = -1;
1816 } else {
1817 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1818
1819 /* manage array size when it is managed externally
1820 */
1821 if ((rv == 0) && st->ss->external)
1822 rv = set_array_size(st, sra, sra->text_version);
1823 }
1824
1825 if (raid0_takeover) {
1826 /* do not recync non-existing parity,
1827 * we will drop it anyway
1828 */
1829 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1830 /* go back to raid0, drop parity disk
1831 */
1832 sysfs_set_str(sra, NULL, "level", "raid0");
1833 ioctl(fd, GET_ARRAY_INFO, &array);
1834 }
1835
1836 size_change_error:
1837 if (rv != 0) {
1838 int err = errno;
1839
1840 /* restore metadata */
1841 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1842 UnSet, NULL, devname,
1843 ROLLBACK_METADATA_CHANGES,
1844 c->verbose) == 0)
1845 sync_metadata(st);
1846 pr_err("Cannot set device size for %s: %s\n",
1847 devname, strerror(err));
1848 if (err == EBUSY &&
1849 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1850 cont_err("Bitmap must be removed before size can be changed\n");
1851 rv = 1;
1852 goto release;
1853 }
1854 if (s->assume_clean) {
1855 /* This will fail on kernels older than 3.0 unless
1856 * a backport has been arranged.
1857 */
1858 if (sra == NULL ||
1859 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1860 pr_err("--assume-clean not supported with --grow on this kernel\n");
1861 }
1862 ioctl(fd, GET_ARRAY_INFO, &array);
1863 s->size = get_component_size(fd)/2;
1864 if (s->size == 0)
1865 s->size = array.size;
1866 if (c->verbose >= 0) {
1867 if (s->size == orig_size)
1868 pr_err("component size of %s "
1869 "unchanged at %lluK\n",
1870 devname, s->size);
1871 else
1872 pr_err("component size of %s "
1873 "has been set to %lluK\n",
1874 devname, s->size);
1875 }
1876 changed = 1;
1877 } else if (array.level != LEVEL_CONTAINER) {
1878 s->size = get_component_size(fd)/2;
1879 if (s->size == 0)
1880 s->size = array.size;
1881 }
1882
1883 /* See if there is anything else to do */
1884 if ((s->level == UnSet || s->level == array.level) &&
1885 (s->layout_str == NULL) &&
1886 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1887 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1888 /* Nothing more to do */
1889 if (!changed && c->verbose >= 0)
1890 pr_err("%s: no change requested\n",
1891 devname);
1892 goto release;
1893 }
1894
1895 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1896 * current implementation assumes that following conditions must be met:
1897 * - RAID10:
1898 * - far_copies == 1
1899 * - near_copies == 2
1900 */
1901 if ((s->level == 0 && array.level == 10 && sra &&
1902 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1903 (s->level == 0 && array.level == 1 && sra)) {
1904 int err;
1905 err = remove_disks_for_takeover(st, sra, array.layout);
1906 if (err) {
1907 dprintf(Name": Array cannot be reshaped\n");
1908 if (cfd > -1)
1909 close(cfd);
1910 rv = 1;
1911 goto release;
1912 }
1913 /* Make sure mdmon has seen the device removal
1914 * and updated metadata before we continue with
1915 * level change
1916 */
1917 if (container)
1918 ping_monitor(container);
1919 }
1920
1921 memset(&info, 0, sizeof(info));
1922 info.array = array;
1923 sysfs_init(&info, fd, NoMdDev);
1924 strcpy(info.text_version, sra->text_version);
1925 info.component_size = s->size*2;
1926 info.new_level = s->level;
1927 info.new_chunk = s->chunk * 1024;
1928 if (info.array.level == LEVEL_CONTAINER) {
1929 info.delta_disks = UnSet;
1930 info.array.raid_disks = s->raiddisks;
1931 } else if (s->raiddisks)
1932 info.delta_disks = s->raiddisks - info.array.raid_disks;
1933 else
1934 info.delta_disks = UnSet;
1935 if (s->layout_str == NULL) {
1936 info.new_layout = UnSet;
1937 if (info.array.level == 6 &&
1938 (info.new_level == 6 || info.new_level == UnSet) &&
1939 info.array.layout >= 16) {
1940 pr_err("%s has a non-standard layout. If you"
1941 " wish to preserve this\n", devname);
1942 cont_err("during the reshape, please specify"
1943 " --layout=preserve\n");
1944 cont_err("If you want to change it, specify a"
1945 " layout or use --layout=normalise\n");
1946 rv = 1;
1947 goto release;
1948 }
1949 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1950 strcmp(s->layout_str, "normalize") == 0) {
1951 /* If we have a -6 RAID6 layout, remove the '-6'. */
1952 info.new_layout = UnSet;
1953 if (info.array.level == 6 && info.new_level == UnSet) {
1954 char l[40], *h;
1955 strcpy(l, map_num(r6layout, info.array.layout));
1956 h = strrchr(l, '-');
1957 if (h && strcmp(h, "-6") == 0) {
1958 *h = 0;
1959 info.new_layout = map_name(r6layout, l);
1960 }
1961 } else {
1962 pr_err("%s is only meaningful when reshaping"
1963 " a RAID6 array.\n", s->layout_str);
1964 rv = 1;
1965 goto release;
1966 }
1967 } else if (strcmp(s->layout_str, "preserve") == 0) {
1968 /* This means that a non-standard RAID6 layout
1969 * is OK.
1970 * In particular:
1971 * - When reshape a RAID6 (e.g. adding a device)
1972 * which is in a non-standard layout, it is OK
1973 * to preserve that layout.
1974 * - When converting a RAID5 to RAID6, leave it in
1975 * the XXX-6 layout, don't re-layout.
1976 */
1977 if (info.array.level == 6 && info.new_level == UnSet)
1978 info.new_layout = info.array.layout;
1979 else if (info.array.level == 5 && info.new_level == 6) {
1980 char l[40];
1981 strcpy(l, map_num(r5layout, info.array.layout));
1982 strcat(l, "-6");
1983 info.new_layout = map_name(r6layout, l);
1984 } else {
1985 pr_err("%s in only meaningful when reshaping"
1986 " to RAID6\n", s->layout_str);
1987 rv = 1;
1988 goto release;
1989 }
1990 } else {
1991 int l = info.new_level;
1992 if (l == UnSet)
1993 l = info.array.level;
1994 switch (l) {
1995 case 5:
1996 info.new_layout = map_name(r5layout, s->layout_str);
1997 break;
1998 case 6:
1999 info.new_layout = map_name(r6layout, s->layout_str);
2000 break;
2001 case 10:
2002 info.new_layout = parse_layout_10(s->layout_str);
2003 break;
2004 case LEVEL_FAULTY:
2005 info.new_layout = parse_layout_faulty(s->layout_str);
2006 break;
2007 default:
2008 pr_err("layout not meaningful"
2009 " with this level\n");
2010 rv = 1;
2011 goto release;
2012 }
2013 if (info.new_layout == UnSet) {
2014 pr_err("layout %s not understood"
2015 " for this level\n",
2016 s->layout_str);
2017 rv = 1;
2018 goto release;
2019 }
2020 }
2021
2022 if (array.level == LEVEL_FAULTY) {
2023 if (s->level != UnSet && s->level != array.level) {
2024 pr_err("cannot change level of Faulty device\n");
2025 rv =1 ;
2026 }
2027 if (s->chunk) {
2028 pr_err("cannot set chunksize of Faulty device\n");
2029 rv =1 ;
2030 }
2031 if (s->raiddisks && s->raiddisks != 1) {
2032 pr_err("cannot set raid_disks of Faulty device\n");
2033 rv =1 ;
2034 }
2035 if (s->layout_str) {
2036 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2037 dprintf("Cannot get array information.\n");
2038 goto release;
2039 }
2040 array.layout = info.new_layout;
2041 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2042 pr_err("failed to set new layout\n");
2043 rv = 1;
2044 } else if (c->verbose >= 0)
2045 printf("layout for %s set to %d\n",
2046 devname, array.layout);
2047 }
2048 } else if (array.level == LEVEL_CONTAINER) {
2049 /* This change is to be applied to every array in the
2050 * container. This is only needed when the metadata imposes
2051 * restraints of the various arrays in the container.
2052 * Currently we only know that IMSM requires all arrays
2053 * to have the same number of devices so changing the
2054 * number of devices (On-Line Capacity Expansion) must be
2055 * performed at the level of the container
2056 */
2057 rv = reshape_container(container, devname, -1, st, &info,
2058 c->force, c->backup_file, c->verbose, 0, 0);
2059 frozen = 0;
2060 } else {
2061 /* get spare devices from external metadata
2062 */
2063 if (st->ss->external) {
2064 struct mdinfo *info2;
2065
2066 info2 = st->ss->container_content(st, subarray);
2067 if (info2) {
2068 info.array.spare_disks =
2069 info2->array.spare_disks;
2070 sysfs_free(info2);
2071 }
2072 }
2073
2074 /* Impose these changes on a single array. First
2075 * check that the metadata is OK with the change. */
2076
2077 if (reshape_super(st, 0, info.new_level,
2078 info.new_layout, info.new_chunk,
2079 info.array.raid_disks, info.delta_disks,
2080 c->backup_file, devname, APPLY_METADATA_CHANGES,
2081 c->verbose)) {
2082 rv = 1;
2083 goto release;
2084 }
2085 sync_metadata(st);
2086 rv = reshape_array(container, fd, devname, st, &info, c->force,
2087 devlist, data_offset, c->backup_file, c->verbose,
2088 0, 0, 0);
2089 frozen = 0;
2090 }
2091 release:
2092 sysfs_free(sra);
2093 if (frozen > 0)
2094 unfreeze(st);
2095 return rv;
2096 }
2097
2098 /* verify_reshape_position()
2099 * Function checks if reshape position in metadata is not farther
2100 * than position in md.
2101 * Return value:
2102 * 0 : not valid sysfs entry
2103 * it can be caused by not started reshape, it should be started
2104 * by reshape array or raid0 array is before takeover
2105 * -1 : error, reshape position is obviously wrong
2106 * 1 : success, reshape progress correct or updated
2107 */
2108 static int verify_reshape_position(struct mdinfo *info, int level)
2109 {
2110 int ret_val = 0;
2111 char buf[40];
2112 int rv;
2113
2114 /* read sync_max, failure can mean raid0 array */
2115 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2116
2117 if (rv > 0) {
2118 char *ep;
2119 unsigned long long position = strtoull(buf, &ep, 0);
2120
2121 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2122 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2123 position *= get_data_disks(level,
2124 info->new_layout,
2125 info->array.raid_disks);
2126 if (info->reshape_progress < position) {
2127 dprintf("Corrected reshape progress (%llu) to "
2128 "md position (%llu)\n",
2129 info->reshape_progress, position);
2130 info->reshape_progress = position;
2131 ret_val = 1;
2132 } else if (info->reshape_progress > position) {
2133 pr_err("Fatal error: array "
2134 "reshape was not properly frozen "
2135 "(expected reshape position is %llu, "
2136 "but reshape progress is %llu.\n",
2137 position, info->reshape_progress);
2138 ret_val = -1;
2139 } else {
2140 dprintf("Reshape position in md and metadata "
2141 "are the same;");
2142 ret_val = 1;
2143 }
2144 }
2145 } else if (rv == 0) {
2146 /* for valid sysfs entry, 0-length content
2147 * should be indicated as error
2148 */
2149 ret_val = -1;
2150 }
2151
2152 return ret_val;
2153 }
2154
2155 static int raid10_reshape(char *container, int fd, char *devname,
2156 struct supertype *st, struct mdinfo *info,
2157 struct reshape *reshape,
2158 unsigned long long data_offset,
2159 int force, int verbose)
2160 {
2161 /* Changing raid_disks, layout, chunksize or possibly
2162 * just data_offset for a RAID10.
2163 * We must always change data_offset. We change by at least
2164 * ->backup_blocks which is the largest of the old and new
2165 * chunk sizes.
2166 * If raid_disks is increasing, then data_offset must decrease
2167 * by at least this copy size.
2168 * If raid_disks is unchanged, data_offset must increase or
2169 * decrease by at least backup_blocks but preferably by much more.
2170 * We choose half of the available space.
2171 * If raid_disks is decreasing, data_offset must increase by
2172 * at least backup_blocks. To allow of this, component_size
2173 * must be decreased by the same amount.
2174 *
2175 * So we calculate the required minimum and direction, possibly
2176 * reduce the component_size, then iterate through the devices
2177 * and set the new_data_offset.
2178 * If that all works, we set chunk_size, layout, raid_disks, and start
2179 * 'reshape'
2180 */
2181 struct mdinfo *sra, *sd;
2182 unsigned long long min;
2183 int dir = 0;
2184 int err = 0;
2185
2186 sra = sysfs_read(fd, 0,
2187 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2188 );
2189 if (!sra) {
2190 fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
2191 devname);
2192 goto release;
2193 }
2194 min = reshape->backup_blocks;
2195
2196 if (info->delta_disks)
2197 sysfs_set_str(sra, NULL, "reshape_direction",
2198 info->delta_disks < 0 ? "backwards" : "forwards");
2199 if (info->delta_disks < 0 &&
2200 info->space_after < reshape->backup_blocks) {
2201 int rv = sysfs_set_num(sra, NULL, "component_size",
2202 (sra->component_size -
2203 reshape->backup_blocks)/2);
2204 if (rv) {
2205 fprintf(stderr, Name ": cannot reduce component size\n");
2206 goto release;
2207 }
2208 }
2209 for (sd = sra->devs; sd; sd = sd->next) {
2210 char *dn;
2211 int dfd;
2212 int rv;
2213 struct supertype *st2;
2214 struct mdinfo info2;
2215
2216 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2217 continue;
2218 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2219 dfd = dev_open(dn, O_RDONLY);
2220 if (dfd < 0) {
2221 fprintf(stderr,
2222 Name ": %s: cannot open component %s\n",
2223 devname, dn ? dn : "-unknown-");
2224 rv = -1;
2225 goto release;
2226 }
2227 st2 = dup_super(st);
2228 rv = st2->ss->load_super(st2,dfd, NULL);
2229 close(dfd);
2230 if (rv) {
2231 free(st2);
2232 fprintf(stderr, ": %s: cannot get superblock from %s\n",
2233 devname, dn);
2234 goto release;
2235 }
2236 st2->ss->getinfo_super(st2, &info2, NULL);
2237 st2->ss->free_super(st2);
2238 free(st2);
2239 if (info->delta_disks < 0) {
2240 /* Don't need any space as array is shrinking
2241 * just move data_offset up by min
2242 */
2243 if (data_offset == 1)
2244 info2.new_data_offset = info2.data_offset + min;
2245 else {
2246 if ((unsigned long long)data_offset
2247 < info2.data_offset + min) {
2248 fprintf(stderr, Name ": --data-offset too small for %s\n",
2249 dn);
2250 goto release;
2251 }
2252 info2.new_data_offset = data_offset;
2253 }
2254 } else if (info->delta_disks > 0) {
2255 /* need space before */
2256 if (info2.space_before < min) {
2257 fprintf(stderr, Name ": Insufficient head-space for reshape on %s\n",
2258 dn);
2259 goto release;
2260 }
2261 if (data_offset == 1)
2262 info2.new_data_offset = info2.data_offset - min;
2263 else {
2264 if ((unsigned long long)data_offset
2265 > info2.data_offset - min) {
2266 fprintf(stderr, Name ": --data-offset too large for %s\n",
2267 dn);
2268 goto release;
2269 }
2270 info2.new_data_offset = data_offset;
2271 }
2272 } else {
2273 if (dir == 0) {
2274 /* can move up or down. 'data_offset'
2275 * might guide us, otherwise choose
2276 * direction with most space
2277 */
2278 if (data_offset == 1) {
2279 if (info2.space_before > info2.space_after)
2280 dir = -1;
2281 else
2282 dir = 1;
2283 } else if (data_offset < info2.data_offset)
2284 dir = -1;
2285 else
2286 dir = 1;
2287 sysfs_set_str(sra, NULL, "reshape_direction",
2288 dir == 1 ? "backwards" : "forwards");
2289 }
2290 switch (dir) {
2291 case 1: /* Increase data offset */
2292 if (info2.space_after < min) {
2293 fprintf(stderr, Name ": Insufficient tail-space for reshape on %s\n",
2294 dn);
2295 goto release;
2296 }
2297 if (data_offset != 1 &&
2298 data_offset < info2.data_offset + min) {
2299 fprintf(stderr, Name ": --data-offset too small on %s\n",
2300 dn);
2301 goto release;
2302 }
2303 if (data_offset != 1)
2304 info2.new_data_offset = data_offset;
2305 else {
2306 unsigned long long off =
2307 info2.space_after / 2;
2308 off &= ~7ULL;
2309 if (off < min)
2310 off = min;
2311 info2.new_data_offset =
2312 info2.data_offset + off;
2313 }
2314 break;
2315 case -1: /* Decrease data offset */
2316 if (info2.space_before < min) {
2317 fprintf(stderr, Name ": insufficient head-room on %s\n",
2318 dn);
2319 goto release;
2320 }
2321 if (data_offset != 1 &&
2322 data_offset < info2.data_offset - min) {
2323 fprintf(stderr, Name ": --data-offset too small on %s\n",
2324 dn);
2325 goto release;
2326 }
2327 if (data_offset != 1)
2328 info2.new_data_offset = data_offset;
2329 else {
2330 unsigned long long off =
2331 info2.space_before / 2;
2332 off &= ~7ULL;
2333 if (off < min)
2334 off = min;
2335 info2.new_data_offset =
2336 info2.data_offset - off;
2337 }
2338 break;
2339 }
2340 }
2341 if (sysfs_set_num(sra, sd, "new_offset",
2342 info2.new_data_offset) < 0) {
2343 err = errno;
2344 fprintf(stderr, Name ": Cannot set new_offset for %s\n",
2345 dn);
2346 break;
2347 }
2348 }
2349 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2350 err = errno;
2351 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2352 err = errno;
2353 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2354 info->array.raid_disks + info->delta_disks) < 0)
2355 err = errno;
2356 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2357 err = errno;
2358 if (err) {
2359 fprintf(stderr, Name ": Cannot set array shape for %s\n",
2360 devname);
2361 if (err == EBUSY &&
2362 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2363 fprintf(stderr,
2364 " Bitmap must be removed before"
2365 " shape can be changed\n");
2366 goto release;
2367 }
2368 sysfs_free(sra);
2369 return 0;
2370 release:
2371 sysfs_free(sra);
2372 return 1;
2373 }
2374
2375 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2376 {
2377 struct mdinfo *sra, *sd;
2378 unsigned long long min_space_before, min_space_after;
2379 int first = 1;
2380
2381 sra = sysfs_read(fd, 0, GET_DEVS);
2382 if (!sra)
2383 return;
2384 for (sd = sra->devs; sd; sd = sd->next) {
2385 char *dn;
2386 int dfd;
2387 struct supertype *st2;
2388 struct mdinfo info2;
2389
2390 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2391 continue;
2392 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2393 dfd = dev_open(dn, O_RDONLY);
2394 if (dfd < 0)
2395 break;
2396 st2 = dup_super(st);
2397 if (st2->ss->load_super(st2,dfd, NULL)) {
2398 close(dfd);
2399 free(st2);
2400 break;
2401 }
2402 close(dfd);
2403 st2->ss->getinfo_super(st2, &info2, NULL);
2404 st2->ss->free_super(st2);
2405 free(st2);
2406 if (first ||
2407 min_space_before > info2.space_before)
2408 min_space_before = info2.space_before;
2409 if (first ||
2410 min_space_after > info2.space_after)
2411 min_space_after = info2.space_after;
2412 first = 0;
2413 }
2414 if (sd == NULL && !first) {
2415 info->space_after = min_space_after;
2416 info->space_before = min_space_before;
2417 }
2418 sysfs_free(sra);
2419 }
2420
2421 static int reshape_array(char *container, int fd, char *devname,
2422 struct supertype *st, struct mdinfo *info,
2423 int force, struct mddev_dev *devlist,
2424 unsigned long long data_offset,
2425 char *backup_file, int verbose, int forked,
2426 int restart, int freeze_reshape)
2427 {
2428 struct reshape reshape;
2429 int spares_needed;
2430 char *msg;
2431 int orig_level = UnSet;
2432 int disks, odisks;
2433 int delayed;
2434
2435 struct mdu_array_info_s array;
2436 char *c;
2437
2438 struct mddev_dev *dv;
2439 int added_disks;
2440
2441 int *fdlist = NULL;
2442 unsigned long long *offsets = NULL;
2443 int d;
2444 int nrdisks;
2445 int err;
2446 unsigned long blocks;
2447 unsigned long cache;
2448 unsigned long long array_size;
2449 int done;
2450 struct mdinfo *sra = NULL;
2451
2452 /* when reshaping a RAID0, the component_size might be zero.
2453 * So try to fix that up.
2454 */
2455 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2456 dprintf("Cannot get array information.\n");
2457 goto release;
2458 }
2459 if (array.level == 0 && info->component_size == 0) {
2460 get_dev_size(fd, NULL, &array_size);
2461 info->component_size = array_size / array.raid_disks;
2462 }
2463
2464 if (array.level == 10)
2465 /* Need space_after info */
2466 get_space_after(fd, st, info);
2467
2468 if (info->reshape_active) {
2469 int new_level = info->new_level;
2470 info->new_level = UnSet;
2471 if (info->delta_disks > 0)
2472 info->array.raid_disks -= info->delta_disks;
2473 msg = analyse_change(info, &reshape);
2474 info->new_level = new_level;
2475 if (info->delta_disks > 0)
2476 info->array.raid_disks += info->delta_disks;
2477 if (!restart)
2478 /* Make sure the array isn't read-only */
2479 ioctl(fd, RESTART_ARRAY_RW, 0);
2480 } else
2481 msg = analyse_change(info, &reshape);
2482 if (msg) {
2483 pr_err("%s\n", msg);
2484 goto release;
2485 }
2486 if (restart &&
2487 (reshape.level != info->array.level ||
2488 reshape.before.layout != info->array.layout ||
2489 reshape.before.data_disks + reshape.parity
2490 != info->array.raid_disks - max(0, info->delta_disks))) {
2491 pr_err("reshape info is not in native format -"
2492 " cannot continue.\n");
2493 goto release;
2494 }
2495
2496 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2497 /* When reshape is restarted from '0', very begin of array
2498 * it is possible that for external metadata reshape and array
2499 * configuration doesn't happen.
2500 * Check if md has the same opinion, and reshape is restarted
2501 * from 0. If so, this is regular reshape start after reshape
2502 * switch in metadata to next array only.
2503 */
2504 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2505 (info->reshape_progress == 0))
2506 restart = 0;
2507 }
2508 if (restart) {
2509 /* reshape already started. just skip to monitoring the reshape */
2510 if (reshape.backup_blocks == 0)
2511 return 0;
2512 goto started;
2513 }
2514 /* The container is frozen but the array may not be.
2515 * So freeze the array so spares don't get put to the wrong use
2516 * FIXME there should probably be a cleaner separation between
2517 * freeze_array and freeze_container.
2518 */
2519 sysfs_freeze_array(info);
2520 /* Check we have enough spares to not be degraded */
2521 added_disks = 0;
2522 for (dv = devlist; dv ; dv=dv->next)
2523 added_disks++;
2524 spares_needed = max(reshape.before.data_disks,
2525 reshape.after.data_disks)
2526 + reshape.parity - array.raid_disks;
2527
2528 if (!force &&
2529 info->new_level > 1 && info->array.level > 1 &&
2530 spares_needed > info->array.spare_disks + added_disks) {
2531 pr_err("Need %d spare%s to avoid degraded array,"
2532 " and only have %d.\n"
2533 " Use --force to over-ride this check.\n",
2534 spares_needed,
2535 spares_needed == 1 ? "" : "s",
2536 info->array.spare_disks + added_disks);
2537 goto release;
2538 }
2539 /* Check we have enough spares to not fail */
2540 spares_needed = max(reshape.before.data_disks,
2541 reshape.after.data_disks)
2542 - array.raid_disks;
2543 if ((info->new_level > 1 || info->new_level == 0) &&
2544 spares_needed > info->array.spare_disks +added_disks) {
2545 pr_err("Need %d spare%s to create working array,"
2546 " and only have %d.\n",
2547 spares_needed,
2548 spares_needed == 1 ? "" : "s",
2549 info->array.spare_disks + added_disks);
2550 goto release;
2551 }
2552
2553 if (reshape.level != array.level) {
2554 char *c = map_num(pers, reshape.level);
2555 int err;
2556 if (c == NULL)
2557 goto release;
2558
2559 err = sysfs_set_str(info, NULL, "level", c);
2560 if (err) {
2561 err = errno;
2562 pr_err("%s: could not set level to %s\n",
2563 devname, c);
2564 if (err == EBUSY &&
2565 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2566 cont_err("Bitmap must be removed"
2567 " before level can be changed\n");
2568 goto release;
2569 }
2570 if (verbose >= 0)
2571 pr_err("level of %s changed to %s\n",
2572 devname, c);
2573 orig_level = array.level;
2574 sysfs_freeze_array(info);
2575
2576 if (reshape.level > 0 && st->ss->external) {
2577 /* make sure mdmon is aware of the new level */
2578 if (mdmon_running(st->container_dev))
2579 flush_mdmon(container);
2580
2581 if (!mdmon_running(st->container_dev))
2582 start_mdmon(st->container_dev);
2583 ping_monitor(container);
2584 if (mdmon_running(st->container_dev) &&
2585 st->update_tail == NULL)
2586 st->update_tail = &st->updates;
2587 }
2588 }
2589 /* ->reshape_super might have chosen some spares from the
2590 * container that it wants to be part of the new array.
2591 * We can collect them with ->container_content and give
2592 * them to the kernel.
2593 */
2594 if (st->ss->reshape_super && st->ss->container_content) {
2595 char *subarray = strchr(info->text_version+1, '/')+1;
2596 struct mdinfo *info2 =
2597 st->ss->container_content(st, subarray);
2598 struct mdinfo *d;
2599
2600 if (info2) {
2601 sysfs_init(info2, fd, st->devnum);
2602 /* When increasing number of devices, we need to set
2603 * new raid_disks before adding these, or they might
2604 * be rejected.
2605 */
2606 if (reshape.backup_blocks &&
2607 reshape.after.data_disks > reshape.before.data_disks)
2608 subarray_set_num(container, info2, "raid_disks",
2609 reshape.after.data_disks +
2610 reshape.parity);
2611 for (d = info2->devs; d; d = d->next) {
2612 if (d->disk.state == 0 &&
2613 d->disk.raid_disk >= 0) {
2614 /* This is a spare that wants to
2615 * be part of the array.
2616 */
2617 add_disk(fd, st, info2, d);
2618 }
2619 }
2620 sysfs_free(info2);
2621 }
2622 }
2623 /* We might have been given some devices to add to the
2624 * array. Now that the array has been changed to the right
2625 * level and frozen, we can safely add them.
2626 */
2627 if (devlist)
2628 Manage_subdevs(devname, fd, devlist, verbose,
2629 0,NULL, 0);
2630
2631 if (reshape.backup_blocks == 0) {
2632 /* No restriping needed, but we might need to impose
2633 * some more changes: layout, raid_disks, chunk_size
2634 */
2635 /* read current array info */
2636 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2637 dprintf("Cannot get array information.\n");
2638 goto release;
2639 }
2640 /* compare current array info with new values and if
2641 * it is different update them to new */
2642 if (info->new_layout != UnSet &&
2643 info->new_layout != array.layout) {
2644 array.layout = info->new_layout;
2645 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2646 pr_err("failed to set new layout\n");
2647 goto release;
2648 } else if (verbose >= 0)
2649 printf("layout for %s set to %d\n",
2650 devname, array.layout);
2651 }
2652 if (info->delta_disks != UnSet &&
2653 info->delta_disks != 0 &&
2654 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
2655 array.raid_disks += info->delta_disks;
2656 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2657 pr_err("failed to set raid disks\n");
2658 goto release;
2659 } else if (verbose >= 0) {
2660 printf("raid_disks for %s set to %d\n",
2661 devname, array.raid_disks);
2662 }
2663 }
2664 if (info->new_chunk != 0 &&
2665 info->new_chunk != array.chunk_size) {
2666 if (sysfs_set_num(info, NULL,
2667 "chunk_size", info->new_chunk) != 0) {
2668 pr_err("failed to set chunk size\n");
2669 goto release;
2670 } else if (verbose >= 0)
2671 printf("chunk size for %s set to %d\n",
2672 devname, array.chunk_size);
2673 }
2674 unfreeze(st);
2675 return 0;
2676 }
2677
2678 /*
2679 * There are three possibilities.
2680 * 1/ The array will shrink.
2681 * We need to ensure the reshape will pause before reaching
2682 * the 'critical section'. We also need to fork and wait for
2683 * that to happen. When it does we
2684 * suspend/backup/complete/unfreeze
2685 *
2686 * 2/ The array will not change size.
2687 * This requires that we keep a backup of a sliding window
2688 * so that we can restore data after a crash. So we need
2689 * to fork and monitor progress.
2690 * In future we will allow the data_offset to change, so
2691 * a sliding backup becomes unnecessary.
2692 *
2693 * 3/ The array will grow. This is relatively easy.
2694 * However the kernel's restripe routines will cheerfully
2695 * overwrite some early data before it is safe. So we
2696 * need to make a backup of the early parts of the array
2697 * and be ready to restore it if rebuild aborts very early.
2698 * For externally managed metadata, we still need a forked
2699 * child to monitor the reshape and suspend IO over the region
2700 * that is being reshaped.
2701 *
2702 * We backup data by writing it to one spare, or to a
2703 * file which was given on command line.
2704 *
2705 * In each case, we first make sure that storage is available
2706 * for the required backup.
2707 * Then we:
2708 * - request the shape change.
2709 * - fork to handle backup etc.
2710 */
2711 /* Check that we can hold all the data */
2712 get_dev_size(fd, NULL, &array_size);
2713 if (reshape.new_size < (array_size/512)) {
2714 pr_err("this change will reduce the size of the array.\n"
2715 " use --grow --array-size first to truncate array.\n"
2716 " e.g. mdadm --grow %s --array-size %llu\n",
2717 devname, reshape.new_size/2);
2718 goto release;
2719 }
2720
2721 started:
2722
2723 if (array.level == 10) {
2724 /* Reshaping RAID10 does not require and data backup by
2725 * user-space. Instead it requires that the data_offset
2726 * is changed to avoid the need for backup.
2727 * So this is handled very separately
2728 */
2729 if (restart)
2730 /* Nothing to do. */
2731 return 0;
2732 return raid10_reshape(container, fd, devname, st, info,
2733 &reshape, data_offset,
2734 force, verbose);
2735 }
2736 sra = sysfs_read(fd, 0,
2737 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2738 GET_CACHE);
2739 if (!sra) {
2740 pr_err("%s: Cannot get array details from sysfs\n",
2741 devname);
2742 goto release;
2743 }
2744
2745 /* Decide how many blocks (sectors) for a reshape
2746 * unit. The number we have so far is just a minimum
2747 */
2748 blocks = reshape.backup_blocks;
2749 if (reshape.before.data_disks ==
2750 reshape.after.data_disks) {
2751 /* Make 'blocks' bigger for better throughput, but
2752 * not so big that we reject it below.
2753 * Try for 16 megabytes
2754 */
2755 while (blocks * 32 < sra->component_size &&
2756 blocks < 16*1024*2)
2757 blocks *= 2;
2758 } else
2759 pr_err("Need to backup %luK of critical "
2760 "section..\n", blocks/2);
2761
2762 if (blocks >= sra->component_size/2) {
2763 pr_err("%s: Something wrong"
2764 " - reshape aborted\n",
2765 devname);
2766 goto release;
2767 }
2768
2769 /* Now we need to open all these devices so we can read/write.
2770 */
2771 nrdisks = max(reshape.before.data_disks,
2772 reshape.after.data_disks) + reshape.parity
2773 + sra->array.spare_disks;
2774 fdlist = xcalloc((1+nrdisks), sizeof(int));
2775 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
2776
2777 odisks = reshape.before.data_disks + reshape.parity;
2778 d = reshape_prepare_fdlist(devname, sra, odisks,
2779 nrdisks, blocks, backup_file,
2780 fdlist, offsets);
2781 if (d < 0) {
2782 goto release;
2783 }
2784 if ((st->ss->manage_reshape == NULL) ||
2785 (st->ss->recover_backup == NULL)) {
2786 if (backup_file == NULL) {
2787 if (reshape.after.data_disks <=
2788 reshape.before.data_disks) {
2789 pr_err("%s: Cannot grow - "
2790 "need backup-file\n", devname);
2791 goto release;
2792 } else if (sra->array.spare_disks == 0) {
2793 pr_err("%s: Cannot grow - "
2794 "need a spare or backup-file to backup "
2795 "critical section\n", devname);
2796 goto release;
2797 }
2798 } else {
2799 if (!reshape_open_backup_file(backup_file, fd, devname,
2800 (signed)blocks,
2801 fdlist+d, offsets+d,
2802 restart)) {
2803 goto release;
2804 }
2805 d++;
2806 }
2807 }
2808
2809 /* lastly, check that the internal stripe cache is
2810 * large enough, or it won't work.
2811 * It must hold at least 4 stripes of the larger
2812 * chunk size
2813 */
2814 cache = max(info->array.chunk_size, info->new_chunk);
2815 cache *= 4; /* 4 stripes minimum */
2816 cache /= 512; /* convert to sectors */
2817 disks = min(reshape.before.data_disks, reshape.after.data_disks);
2818 /* make sure there is room for 'blocks' with a bit to spare */
2819 if (cache < 16 + blocks / disks)
2820 cache = 16 + blocks / disks;
2821 cache /= (4096/512); /* Covert from sectors to pages */
2822
2823 if (sra->cache_size < cache)
2824 subarray_set_num(container, sra, "stripe_cache_size",
2825 cache+1);
2826
2827 /* Right, everything seems fine. Let's kick things off.
2828 * If only changing raid_disks, use ioctl, else use
2829 * sysfs.
2830 */
2831 sync_metadata(st);
2832
2833 sra->new_chunk = info->new_chunk;
2834
2835 if (restart) {
2836 /* for external metadata checkpoint saved by mdmon can be lost
2837 * or missed /due to e.g. crash/. Check if md is not during
2838 * restart farther than metadata points to.
2839 * If so, this means metadata information is obsolete.
2840 */
2841 if (st->ss->external)
2842 verify_reshape_position(info, reshape.level);
2843 sra->reshape_progress = info->reshape_progress;
2844 } else {
2845 sra->reshape_progress = 0;
2846 if (reshape.after.data_disks < reshape.before.data_disks)
2847 /* start from the end of the new array */
2848 sra->reshape_progress = (sra->component_size
2849 * reshape.after.data_disks);
2850 }
2851
2852 if (info->array.chunk_size == info->new_chunk &&
2853 reshape.before.layout == reshape.after.layout &&
2854 st->ss->external == 0) {
2855 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2856 ioctl(fd, GET_ARRAY_INFO, &array);
2857 array.raid_disks = reshape.after.data_disks + reshape.parity;
2858 if (!restart &&
2859 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2860 int err = errno;
2861
2862 pr_err("Cannot set device shape for %s: %s\n",
2863 devname, strerror(errno));
2864
2865 if (err == EBUSY &&
2866 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2867 cont_err("Bitmap must be removed before"
2868 " shape can be changed\n");
2869
2870 goto release;
2871 }
2872 } else if (!restart) {
2873 /* set them all just in case some old 'new_*' value
2874 * persists from some earlier problem.
2875 */
2876 int err = 0;
2877 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2878 err = errno;
2879 if (!err && sysfs_set_num(sra, NULL, "layout",
2880 reshape.after.layout) < 0)
2881 err = errno;
2882 if (!err && subarray_set_num(container, sra, "raid_disks",
2883 reshape.after.data_disks +
2884 reshape.parity) < 0)
2885 err = errno;
2886 if (err) {
2887 pr_err("Cannot set device shape for %s\n",
2888 devname);
2889
2890 if (err == EBUSY &&
2891 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2892 cont_err("Bitmap must be removed before"
2893 " shape can be changed\n");
2894 goto release;
2895 }
2896 }
2897
2898 err = start_reshape(sra, restart, reshape.before.data_disks,
2899 reshape.after.data_disks);
2900 if (err) {
2901 pr_err("Cannot %s reshape for %s\n",
2902 restart ? "continue" : "start",
2903 devname);
2904 goto release;
2905 }
2906 if (restart)
2907 sysfs_set_str(sra, NULL, "array_state", "active");
2908 if (freeze_reshape) {
2909 free(fdlist);
2910 free(offsets);
2911 sysfs_free(sra);
2912 pr_err("Reshape has to be continued from"
2913 " location %llu when root filesystem has been mounted.\n",
2914 sra->reshape_progress);
2915 return 1;
2916 }
2917
2918 /* Now we just need to kick off the reshape and watch, while
2919 * handling backups of the data...
2920 * This is all done by a forked background process.
2921 */
2922 switch(forked ? 0 : fork()) {
2923 case -1:
2924 pr_err("Cannot run child to monitor reshape: %s\n",
2925 strerror(errno));
2926 abort_reshape(sra);
2927 goto release;
2928 default:
2929 free(fdlist);
2930 free(offsets);
2931 sysfs_free(sra);
2932 return 0;
2933 case 0:
2934 map_fork();
2935 break;
2936 }
2937
2938 /* If another array on the same devices is busy, the
2939 * reshape will wait for them. This would mean that
2940 * the first section that we suspend will stay suspended
2941 * for a long time. So check on that possibility
2942 * by looking for "DELAYED" in /proc/mdstat, and if found,
2943 * wait a while
2944 */
2945 do {
2946 struct mdstat_ent *mds, *m;
2947 delayed = 0;
2948 mds = mdstat_read(0, 0);
2949 for (m = mds; m; m = mds->next)
2950 if (m->devnum == devname2devnum(sra->sys_name)) {
2951 if (m->resync &&
2952 m->percent == RESYNC_DELAYED)
2953 delayed = 1;
2954 if (m->resync == 0)
2955 /* Haven't started the reshape thread
2956 * yet, wait a bit
2957 */
2958 delayed = 2;
2959 break;
2960 }
2961 free_mdstat(mds);
2962 if (delayed == 1 && get_linux_version() < 3007000) {
2963 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
2964 " You might experience problems until other reshapes complete.\n");
2965 delayed = 0;
2966 }
2967 if (delayed)
2968 sleep(30 - (delayed-1) * 25);
2969 } while (delayed);
2970
2971 close(fd);
2972 if (check_env("MDADM_GROW_VERIFY"))
2973 fd = open(devname, O_RDONLY | O_DIRECT);
2974 else
2975 fd = -1;
2976 mlockall(MCL_FUTURE);
2977
2978 if (st->ss->external) {
2979 /* metadata handler takes it from here */
2980 done = st->ss->manage_reshape(
2981 fd, sra, &reshape, st, blocks,
2982 fdlist, offsets,
2983 d - odisks, fdlist+odisks,
2984 offsets+odisks);
2985 } else
2986 done = child_monitor(
2987 fd, sra, &reshape, st, blocks,
2988 fdlist, offsets,
2989 d - odisks, fdlist+odisks,
2990 offsets+odisks);
2991
2992 free(fdlist);
2993 free(offsets);
2994
2995 if (backup_file && done)
2996 unlink(backup_file);
2997 if (!done) {
2998 abort_reshape(sra);
2999 goto out;
3000 }
3001
3002 if (!st->ss->external &&
3003 !(reshape.before.data_disks != reshape.after.data_disks
3004 && info->custom_array_size) &&
3005 info->new_level == reshape.level &&
3006 !forked) {
3007 /* no need to wait for the reshape to finish as
3008 * there is nothing more to do.
3009 */
3010 sysfs_free(sra);
3011 exit(0);
3012 }
3013 wait_reshape(sra);
3014
3015 if (st->ss->external) {
3016 /* Re-load the metadata as much could have changed */
3017 int cfd = open_dev(st->container_dev);
3018 if (cfd >= 0) {
3019 flush_mdmon(container);
3020 st->ss->free_super(st);
3021 st->ss->load_container(st, cfd, container);
3022 close(cfd);
3023 }
3024 }
3025
3026 /* set new array size if required customer_array_size is used
3027 * by this metadata.
3028 */
3029 if (reshape.before.data_disks !=
3030 reshape.after.data_disks &&
3031 info->custom_array_size)
3032 set_array_size(st, info, info->text_version);
3033
3034 if (info->new_level != reshape.level) {
3035
3036 c = map_num(pers, info->new_level);
3037 if (c) {
3038 err = sysfs_set_str(sra, NULL, "level", c);
3039 if (err)
3040 pr_err("%s: could not set level "
3041 "to %s\n", devname, c);
3042 }
3043 if (info->new_level == 0)
3044 st->update_tail = NULL;
3045 }
3046 out:
3047 sysfs_free(sra);
3048 if (forked)
3049 return 0;
3050 unfreeze(st);
3051 exit(0);
3052
3053 release:
3054 free(fdlist);
3055 free(offsets);
3056 if (orig_level != UnSet && sra) {
3057 c = map_num(pers, orig_level);
3058 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3059 pr_err("aborting level change\n");
3060 }
3061 sysfs_free(sra);
3062 if (!forked)
3063 unfreeze(st);
3064 return 1;
3065 }
3066
3067 /* mdfd handle is passed to be closed in child process (after fork).
3068 */
3069 int reshape_container(char *container, char *devname,
3070 int mdfd,
3071 struct supertype *st,
3072 struct mdinfo *info,
3073 int force,
3074 char *backup_file,
3075 int verbose, int restart, int freeze_reshape)
3076 {
3077 struct mdinfo *cc = NULL;
3078 int rv = restart;
3079 int last_devnum = -1;
3080
3081 /* component_size is not meaningful for a container,
3082 * so pass '0' meaning 'no change'
3083 */
3084 if (!restart &&
3085 reshape_super(st, 0, info->new_level,
3086 info->new_layout, info->new_chunk,
3087 info->array.raid_disks, info->delta_disks,
3088 backup_file, devname, APPLY_METADATA_CHANGES,
3089 verbose)) {
3090 unfreeze(st);
3091 return 1;
3092 }
3093
3094 sync_metadata(st);
3095
3096 /* ping monitor to be sure that update is on disk
3097 */
3098 ping_monitor(container);
3099
3100 switch (fork()) {
3101 case -1: /* error */
3102 perror("Cannot fork to complete reshape\n");
3103 unfreeze(st);
3104 return 1;
3105 default: /* parent */
3106 if (!freeze_reshape)
3107 printf(Name ": multi-array reshape continues"
3108 " in background\n");
3109 return 0;
3110 case 0: /* child */
3111 map_fork();
3112 break;
3113 }
3114
3115 /* close unused handle in child process
3116 */
3117 if (mdfd > -1)
3118 close(mdfd);
3119
3120 while(1) {
3121 /* For each member array with reshape_active,
3122 * we need to perform the reshape.
3123 * We pick the first array that needs reshaping and
3124 * reshape it. reshape_array() will re-read the metadata
3125 * so the next time through a different array should be
3126 * ready for reshape.
3127 * It is possible that the 'different' array will not
3128 * be assembled yet. In that case we simple exit.
3129 * When it is assembled, the mdadm which assembles it
3130 * will take over the reshape.
3131 */
3132 struct mdinfo *content;
3133 int fd;
3134 struct mdstat_ent *mdstat;
3135 char *adev;
3136
3137 sysfs_free(cc);
3138
3139 cc = st->ss->container_content(st, NULL);
3140
3141 for (content = cc; content ; content = content->next) {
3142 char *subarray;
3143 if (!content->reshape_active)
3144 continue;
3145
3146 subarray = strchr(content->text_version+1, '/')+1;
3147 mdstat = mdstat_by_subdev(subarray,
3148 devname2devnum(container));
3149 if (!mdstat)
3150 continue;
3151 if (mdstat->active == 0) {
3152 pr_err("Skipping inactive "
3153 "array md%i.\n", mdstat->devnum);
3154 free_mdstat(mdstat);
3155 mdstat = NULL;
3156 continue;
3157 }
3158 break;
3159 }
3160 if (!content)
3161 break;
3162
3163 adev = map_dev(dev2major(mdstat->devnum),
3164 dev2minor(mdstat->devnum),
3165 0);
3166 if (!adev)
3167 adev = content->text_version;
3168
3169 fd = open_dev(mdstat->devnum);
3170 if (fd < 0) {
3171 printf(Name ": Device %s cannot be opened for reshape.",
3172 adev);
3173 break;
3174 }
3175
3176 if (last_devnum == mdstat->devnum) {
3177 /* Do not allow for multiple reshape_array() calls for
3178 * the same array.
3179 * It can happen when reshape_array() returns without
3180 * error, when reshape is not finished (wrong reshape
3181 * starting/continuation conditions). Mdmon doesn't
3182 * switch to next array in container and reentry
3183 * conditions for the same array occur.
3184 * This is possibly interim until the behaviour of
3185 * reshape_array is resolved().
3186 */
3187 printf(Name ": Multiple reshape execution detected for "
3188 "device %s.", adev);
3189 close(fd);
3190 break;
3191 }
3192 last_devnum = mdstat->devnum;
3193
3194 sysfs_init(content, fd, mdstat->devnum);
3195
3196 if (mdmon_running(devname2devnum(container)))
3197 flush_mdmon(container);
3198
3199 rv = reshape_array(container, fd, adev, st,
3200 content, force, NULL, 0ULL,
3201 backup_file, verbose, 1, restart,
3202 freeze_reshape);
3203 close(fd);
3204
3205 if (freeze_reshape) {
3206 sysfs_free(cc);
3207 exit(0);
3208 }
3209
3210 restart = 0;
3211 if (rv)
3212 break;
3213
3214 if (mdmon_running(devname2devnum(container)))
3215 flush_mdmon(container);
3216 }
3217 if (!rv)
3218 unfreeze(st);
3219 sysfs_free(cc);
3220 exit(0);
3221 }
3222
3223 /*
3224 * We run a child process in the background which performs the following
3225 * steps:
3226 * - wait for resync to reach a certain point
3227 * - suspend io to the following section
3228 * - backup that section
3229 * - allow resync to proceed further
3230 * - resume io
3231 * - discard the backup.
3232 *
3233 * When are combined in slightly different ways in the three cases.
3234 * Grow:
3235 * - suspend/backup/allow/wait/resume/discard
3236 * Shrink:
3237 * - allow/wait/suspend/backup/allow/wait/resume/discard
3238 * same-size:
3239 * - wait/resume/discard/suspend/backup/allow
3240 *
3241 * suspend/backup/allow always come together
3242 * wait/resume/discard do too.
3243 * For the same-size case we have two backups to improve flow.
3244 *
3245 */
3246
3247 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3248 unsigned long long backup_point,
3249 unsigned long long wait_point,
3250 unsigned long long *suspend_point,
3251 unsigned long long *reshape_completed)
3252 {
3253 /* This function is called repeatedly by the reshape manager.
3254 * It determines how much progress can safely be made and allows
3255 * that progress.
3256 * - 'info' identifies the array and particularly records in
3257 * ->reshape_progress the metadata's knowledge of progress
3258 * This is a sector offset from the start of the array
3259 * of the next array block to be relocated. This number
3260 * may increase from 0 or decrease from array_size, depending
3261 * on the type of reshape that is happening.
3262 * Note that in contrast, 'sync_completed' is a block count of the
3263 * reshape so far. It gives the distance between the start point
3264 * (head or tail of device) and the next place that data will be
3265 * written. It always increases.
3266 * - 'reshape' is the structure created by analyse_change
3267 * - 'backup_point' shows how much the metadata manager has backed-up
3268 * data. For reshapes with increasing progress, it is the next address
3269 * to be backed up, previous addresses have been backed-up. For
3270 * decreasing progress, it is the earliest address that has been
3271 * backed up - later address are also backed up.
3272 * So addresses between reshape_progress and backup_point are
3273 * backed up providing those are in the 'correct' order.
3274 * - 'wait_point' is an array address. When reshape_completed
3275 * passes this point, progress_reshape should return. It might
3276 * return earlier if it determines that ->reshape_progress needs
3277 * to be updated or further backup is needed.
3278 * - suspend_point is maintained by progress_reshape and the caller
3279 * should not touch it except to initialise to zero.
3280 * It is an array address and it only increases in 2.6.37 and earlier.
3281 * This makes it difficult to handle reducing reshapes with
3282 * external metadata.
3283 * However: it is similar to backup_point in that it records the
3284 * other end of a suspended region from reshape_progress.
3285 * it is moved to extend the region that is safe to backup and/or
3286 * reshape
3287 * - reshape_completed is read from sysfs and returned. The caller
3288 * should copy this into ->reshape_progress when it has reason to
3289 * believe that the metadata knows this, and any backup outside this
3290 * has been erased.
3291 *
3292 * Return value is:
3293 * 1 if more data from backup_point - but only as far as suspend_point,
3294 * should be backed up
3295 * 0 if things are progressing smoothly
3296 * -1 if the reshape is finished because it is all done,
3297 * -2 if the reshape is finished due to an error.
3298 */
3299
3300 int advancing = (reshape->after.data_disks
3301 >= reshape->before.data_disks);
3302 unsigned long long need_backup; /* All data between start of array and
3303 * here will at some point need to
3304 * be backed up.
3305 */
3306 unsigned long long read_offset, write_offset;
3307 unsigned long long write_range;
3308 unsigned long long max_progress, target, completed;
3309 unsigned long long array_size = (info->component_size
3310 * reshape->before.data_disks);
3311 int fd;
3312 char buf[20];
3313
3314 /* First, we unsuspend any region that is now known to be safe.
3315 * If suspend_point is on the 'wrong' side of reshape_progress, then
3316 * we don't have or need suspension at the moment. This is true for
3317 * native metadata when we don't need to back-up.
3318 */
3319 if (advancing) {
3320 if (info->reshape_progress <= *suspend_point)
3321 sysfs_set_num(info, NULL, "suspend_lo",
3322 info->reshape_progress);
3323 } else {
3324 /* Note: this won't work in 2.6.37 and before.
3325 * Something somewhere should make sure we don't need it!
3326 */
3327 if (info->reshape_progress >= *suspend_point)
3328 sysfs_set_num(info, NULL, "suspend_hi",
3329 info->reshape_progress);
3330 }
3331
3332 /* Now work out how far it is safe to progress.
3333 * If the read_offset for ->reshape_progress is less than
3334 * 'blocks' beyond the write_offset, we can only progress as far
3335 * as a backup.
3336 * Otherwise we can progress until the write_offset for the new location
3337 * reaches (within 'blocks' of) the read_offset at the current location.
3338 * However that region must be suspended unless we are using native
3339 * metadata.
3340 * If we need to suspend more, we limit it to 128M per device, which is
3341 * rather arbitrary and should be some time-based calculation.
3342 */
3343 read_offset = info->reshape_progress / reshape->before.data_disks;
3344 write_offset = info->reshape_progress / reshape->after.data_disks;
3345 write_range = info->new_chunk/512;
3346 if (reshape->before.data_disks == reshape->after.data_disks)
3347 need_backup = array_size;
3348 else
3349 need_backup = reshape->backup_blocks;
3350 if (advancing) {
3351 if (read_offset < write_offset + write_range)
3352 max_progress = backup_point;
3353 else
3354 max_progress =
3355 read_offset *
3356 reshape->after.data_disks;
3357 } else {
3358 if (read_offset > write_offset - write_range)
3359 /* Can only progress as far as has been backed up,
3360 * which must be suspended */
3361 max_progress = backup_point;
3362 else if (info->reshape_progress <= need_backup)
3363 max_progress = backup_point;
3364 else {
3365 if (info->array.major_version >= 0)
3366 /* Can progress until backup is needed */
3367 max_progress = need_backup;
3368 else {
3369 /* Can progress until metadata update is required */
3370 max_progress =
3371 read_offset *
3372 reshape->after.data_disks;
3373 /* but data must be suspended */
3374 if (max_progress < *suspend_point)
3375 max_progress = *suspend_point;
3376 }
3377 }
3378 }
3379
3380 /* We know it is safe to progress to 'max_progress' providing
3381 * it is suspended or we are using native metadata.
3382 * Consider extending suspend_point 128M per device if it
3383 * is less than 64M per device beyond reshape_progress.
3384 * But always do a multiple of 'blocks'
3385 * FIXME this is too big - it takes to long to complete
3386 * this much.
3387 */
3388 target = 64*1024*2 * min(reshape->before.data_disks,
3389 reshape->after.data_disks);
3390 target /= reshape->backup_blocks;
3391 if (target < 2)
3392 target = 2;
3393 target *= reshape->backup_blocks;
3394
3395 /* For externally managed metadata we always need to suspend IO to
3396 * the area being reshaped so we regularly push suspend_point forward.
3397 * For native metadata we only need the suspend if we are going to do
3398 * a backup.
3399 */
3400 if (advancing) {
3401 if ((need_backup > info->reshape_progress
3402 || info->array.major_version < 0) &&
3403 *suspend_point < info->reshape_progress + target) {
3404 if (need_backup < *suspend_point + 2 * target)
3405 *suspend_point = need_backup;
3406 else if (*suspend_point + 2 * target < array_size)
3407 *suspend_point += 2 * target;
3408 else
3409 *suspend_point = array_size;
3410 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3411 if (max_progress > *suspend_point)
3412 max_progress = *suspend_point;
3413 }
3414 } else {
3415 if (info->array.major_version >= 0) {
3416 /* Only need to suspend when about to backup */
3417 if (info->reshape_progress < need_backup * 2 &&
3418 *suspend_point > 0) {
3419 *suspend_point = 0;
3420 sysfs_set_num(info, NULL, "suspend_lo", 0);
3421 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3422 }
3423 } else {
3424 /* Need to suspend continually */
3425 if (info->reshape_progress < *suspend_point)
3426 *suspend_point = info->reshape_progress;
3427 if (*suspend_point + target < info->reshape_progress)
3428 /* No need to move suspend region yet */;
3429 else {
3430 if (*suspend_point >= 2 * target)
3431 *suspend_point -= 2 * target;
3432 else
3433 *suspend_point = 0;
3434 sysfs_set_num(info, NULL, "suspend_lo",
3435 *suspend_point);
3436 }
3437 if (max_progress < *suspend_point)
3438 max_progress = *suspend_point;
3439 }
3440 }
3441
3442 /* now set sync_max to allow that progress. sync_max, like
3443 * sync_completed is a count of sectors written per device, so
3444 * we find the difference between max_progress and the start point,
3445 * and divide that by after.data_disks to get a sync_max
3446 * number.
3447 * At the same time we convert wait_point to a similar number
3448 * for comparing against sync_completed.
3449 */
3450 /* scale down max_progress to per_disk */
3451 max_progress /= reshape->after.data_disks;
3452 /* Round to chunk size as some kernels give an erroneously high number */
3453 max_progress /= info->new_chunk/512;
3454 max_progress *= info->new_chunk/512;
3455 /* And round to old chunk size as the kernel wants that */
3456 max_progress /= info->array.chunk_size/512;
3457 max_progress *= info->array.chunk_size/512;
3458 /* Limit progress to the whole device */
3459 if (max_progress > info->component_size)
3460 max_progress = info->component_size;
3461 wait_point /= reshape->after.data_disks;
3462 if (!advancing) {
3463 /* switch from 'device offset' to 'processed block count' */
3464 max_progress = info->component_size - max_progress;
3465 wait_point = info->component_size - wait_point;
3466 }
3467
3468 sysfs_set_num(info, NULL, "sync_max", max_progress);
3469
3470 /* Now wait. If we have already reached the point that we were
3471 * asked to wait to, don't wait at all, else wait for any change.
3472 * We need to select on 'sync_completed' as that is the place that
3473 * notifications happen, but we are really interested in
3474 * 'reshape_position'
3475 */
3476 fd = sysfs_get_fd(info, NULL, "sync_completed");
3477 if (fd < 0)
3478 goto check_progress;
3479
3480 if (sysfs_fd_get_ll(fd, &completed) < 0)
3481 goto check_progress;
3482
3483 while (completed < max_progress && completed < wait_point) {
3484 /* Check that sync_action is still 'reshape' to avoid
3485 * waiting forever on a dead array
3486 */
3487 char action[20];
3488 fd_set rfds;
3489 if (sysfs_get_str(info, NULL, "sync_action",
3490 action, 20) <= 0 ||
3491 strncmp(action, "reshape", 7) != 0)
3492 break;
3493 /* Some kernels reset 'sync_completed' to zero
3494 * before setting 'sync_action' to 'idle'.
3495 * So we need these extra tests.
3496 */
3497 if (completed == 0 && advancing
3498 && info->reshape_progress > 0)
3499 break;
3500 if (completed == 0 && !advancing
3501 && info->reshape_progress < (info->component_size
3502 * reshape->after.data_disks))
3503 break;
3504 FD_ZERO(&rfds);
3505 FD_SET(fd, &rfds);
3506 select(fd+1, NULL, NULL, &rfds, NULL);
3507 if (sysfs_fd_get_ll(fd, &completed) < 0)
3508 goto check_progress;
3509 }
3510 /* Some kernels reset 'sync_completed' to zero,
3511 * we need to have real point we are in md
3512 */
3513 if (completed == 0)
3514 completed = max_progress;
3515
3516 /* some kernels can give an incorrectly high 'completed' number */
3517 completed /= (info->new_chunk/512);
3518 completed *= (info->new_chunk/512);
3519 /* Convert 'completed' back in to a 'progress' number */
3520 completed *= reshape->after.data_disks;
3521 if (!advancing) {
3522 completed = info->component_size * reshape->after.data_disks
3523 - completed;
3524 }
3525 *reshape_completed = completed;
3526
3527 close(fd);
3528
3529 /* We return the need_backup flag. Caller will decide
3530 * how much - a multiple of ->backup_blocks up to *suspend_point
3531 */
3532 if (advancing)
3533 return need_backup > info->reshape_progress;
3534 else
3535 return need_backup >= info->reshape_progress;
3536
3537 check_progress:
3538 /* if we couldn't read a number from sync_completed, then
3539 * either the reshape did complete, or it aborted.
3540 * We can tell which by checking for 'none' in reshape_position.
3541 * If it did abort, then it might immediately restart if it
3542 * it was just a device failure that leaves us degraded but
3543 * functioning.
3544 */
3545 strcpy(buf, "hi");
3546 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3547 || strncmp(buf, "none", 4) != 0) {
3548 /* The abort might only be temporary. Wait up to 10
3549 * seconds for fd to contain a valid number again.
3550 */
3551 struct timeval tv;
3552 int rv = -2;
3553 tv.tv_sec = 10;
3554 tv.tv_usec = 0;
3555 while (fd >= 0 && rv < 0 && tv.tv_sec > 0) {
3556 fd_set rfds;
3557 FD_ZERO(&rfds);
3558 FD_SET(fd, &rfds);
3559 if (select(fd+1, NULL, NULL, &rfds, &tv) != 1)
3560 break;
3561 switch (sysfs_fd_get_ll(fd, &completed)) {
3562 case 0:
3563 /* all good again */
3564 rv = 1;
3565 break;
3566 case -2: /* read error - abort */
3567 tv.tv_sec = 0;
3568 break;
3569 }
3570 }
3571 if (fd >= 0)
3572 close(fd);
3573 return rv; /* abort */
3574 } else {
3575 /* Maybe racing with array shutdown - check state */
3576 if (fd >= 0)
3577 close(fd);
3578 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3579 || strncmp(buf, "inactive", 8) == 0
3580 || strncmp(buf, "clear",5) == 0)
3581 return -2; /* abort */
3582 return -1; /* complete */
3583 }
3584 }
3585
3586 /* FIXME return status is never checked */
3587 static int grow_backup(struct mdinfo *sra,
3588 unsigned long long offset, /* per device */
3589 unsigned long stripes, /* per device, in old chunks */
3590 int *sources, unsigned long long *offsets,
3591 int disks, int chunk, int level, int layout,
3592 int dests, int *destfd, unsigned long long *destoffsets,
3593 int part, int *degraded,
3594 char *buf)
3595 {
3596 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3597 * to storage 'destfd' (offset 'destoffsets'), after first
3598 * suspending IO. Then allow resync to continue
3599 * over the suspended section.
3600 * Use part 'part' of the backup-super-block.
3601 */
3602 int odata = disks;
3603 int rv = 0;
3604 int i;
3605 unsigned long long ll;
3606 int new_degraded;
3607 //printf("offset %llu\n", offset);
3608 if (level >= 4)
3609 odata--;
3610 if (level == 6)
3611 odata--;
3612
3613 /* Check that array hasn't become degraded, else we might backup the wrong data */
3614 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3615 return -1; /* FIXME this error is ignored */
3616 new_degraded = (int)ll;
3617 if (new_degraded != *degraded) {
3618 /* check each device to ensure it is still working */
3619 struct mdinfo *sd;
3620 for (sd = sra->devs ; sd ; sd = sd->next) {
3621 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3622 continue;
3623 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3624 char sbuf[20];
3625 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3626 strstr(sbuf, "faulty") ||
3627 strstr(sbuf, "in_sync") == NULL) {
3628 /* this device is dead */
3629 sd->disk.state = (1<<MD_DISK_FAULTY);
3630 if (sd->disk.raid_disk >= 0 &&
3631 sources[sd->disk.raid_disk] >= 0) {
3632 close(sources[sd->disk.raid_disk]);
3633 sources[sd->disk.raid_disk] = -1;
3634 }
3635 }
3636 }
3637 }
3638 *degraded = new_degraded;
3639 }
3640 if (part) {
3641 bsb.arraystart2 = __cpu_to_le64(offset * odata);
3642 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
3643 } else {
3644 bsb.arraystart = __cpu_to_le64(offset * odata);
3645 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
3646 }
3647 if (part)
3648 bsb.magic[15] = '2';
3649 for (i = 0; i < dests; i++)
3650 if (part)
3651 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
3652 else
3653 lseek64(destfd[i], destoffsets[i], 0);
3654
3655 rv = save_stripes(sources, offsets,
3656 disks, chunk, level, layout,
3657 dests, destfd,
3658 offset*512*odata, stripes * chunk * odata,
3659 buf);
3660
3661 if (rv)
3662 return rv;
3663 bsb.mtime = __cpu_to_le64(time(0));
3664 for (i = 0; i < dests; i++) {
3665 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3666
3667 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3668 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3669 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3670 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3671
3672 rv = -1;
3673 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
3674 != destoffsets[i] - 4096)
3675 break;
3676 if (write(destfd[i], &bsb, 512) != 512)
3677 break;
3678 if (destoffsets[i] > 4096) {
3679 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
3680 destoffsets[i]+stripes*chunk*odata)
3681 break;
3682 if (write(destfd[i], &bsb, 512) != 512)
3683 break;
3684 }
3685 fsync(destfd[i]);
3686 rv = 0;
3687 }
3688
3689 return rv;
3690 }
3691
3692 /* in 2.6.30, the value reported by sync_completed can be
3693 * less that it should be by one stripe.
3694 * This only happens when reshape hits sync_max and pauses.
3695 * So allow wait_backup to either extent sync_max further
3696 * than strictly necessary, or return before the
3697 * sync has got quite as far as we would really like.
3698 * This is what 'blocks2' is for.
3699 * The various caller give appropriate values so that
3700 * every works.
3701 */
3702 /* FIXME return value is often ignored */
3703 static int forget_backup(int dests, int *destfd,
3704 unsigned long long *destoffsets,
3705 int part)
3706 {
3707 /*
3708 * Erase backup 'part' (which is 0 or 1)
3709 */
3710 int i;
3711 int rv;
3712
3713 if (part) {
3714 bsb.arraystart2 = __cpu_to_le64(0);
3715 bsb.length2 = __cpu_to_le64(0);
3716 } else {
3717 bsb.arraystart = __cpu_to_le64(0);
3718 bsb.length = __cpu_to_le64(0);
3719 }
3720 bsb.mtime = __cpu_to_le64(time(0));
3721 rv = 0;
3722 for (i = 0; i < dests; i++) {
3723 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3724 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3725 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3726 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3727 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3728 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
3729 destoffsets[i]-4096)
3730 rv = -1;
3731 if (rv == 0 &&
3732 write(destfd[i], &bsb, 512) != 512)
3733 rv = -1;
3734 fsync(destfd[i]);
3735 }
3736 return rv;
3737 }
3738
3739 static void fail(char *msg)
3740 {
3741 int rv;
3742 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
3743 rv |= (write(2, "\n", 1) != 1);
3744 exit(rv ? 1 : 2);
3745 }
3746
3747 static char *abuf, *bbuf;
3748 static unsigned long long abuflen;
3749 static void validate(int afd, int bfd, unsigned long long offset)
3750 {
3751 /* check that the data in the backup against the array.
3752 * This is only used for regression testing and should not
3753 * be used while the array is active
3754 */
3755 if (afd < 0)
3756 return;
3757 lseek64(bfd, offset - 4096, 0);
3758 if (read(bfd, &bsb2, 512) != 512)
3759 fail("cannot read bsb");
3760 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
3761 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
3762 fail("first csum bad");
3763 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
3764 fail("magic is bad");
3765 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
3766 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
3767 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
3768 fail("second csum bad");
3769
3770 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
3771 fail("devstart is wrong");
3772
3773 if (bsb2.length) {
3774 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
3775
3776 if (abuflen < len) {
3777 free(abuf);
3778 free(bbuf);
3779 abuflen = len;
3780 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
3781 posix_memalign((void**)&bbuf, 4096, abuflen)) {
3782 abuflen = 0;
3783 /* just stop validating on mem-alloc failure */
3784 return;
3785 }
3786 }
3787
3788 lseek64(bfd, offset, 0);
3789 if ((unsigned long long)read(bfd, bbuf, len) != len) {
3790 //printf("len %llu\n", len);
3791 fail("read first backup failed");
3792 }
3793 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
3794 if ((unsigned long long)read(afd, abuf, len) != len)
3795 fail("read first from array failed");
3796 if (memcmp(bbuf, abuf, len) != 0) {
3797 #if 0
3798 int i;
3799 printf("offset=%llu len=%llu\n",
3800 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
3801 for (i=0; i<len; i++)
3802 if (bbuf[i] != abuf[i]) {
3803 printf("first diff byte %d\n", i);
3804 break;
3805 }
3806 #endif
3807 fail("data1 compare failed");
3808 }
3809 }
3810 if (bsb2.length2) {
3811 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
3812
3813 if (abuflen < len) {
3814 free(abuf);
3815 free(bbuf);
3816 abuflen = len;
3817 abuf = xmalloc(abuflen);
3818 bbuf = xmalloc(abuflen);
3819 }
3820
3821 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
3822 if ((unsigned long long)read(bfd, bbuf, len) != len)
3823 fail("read second backup failed");
3824 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
3825 if ((unsigned long long)read(afd, abuf, len) != len)
3826 fail("read second from array failed");
3827 if (memcmp(bbuf, abuf, len) != 0)
3828 fail("data2 compare failed");
3829 }
3830 }
3831
3832 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
3833 struct supertype *st, unsigned long blocks,
3834 int *fds, unsigned long long *offsets,
3835 int dests, int *destfd, unsigned long long *destoffsets)
3836 {
3837 /* Monitor a reshape where backup is being performed using
3838 * 'native' mechanism - either to a backup file, or
3839 * to some space in a spare.
3840 */
3841 char *buf;
3842 int degraded = -1;
3843 unsigned long long speed;
3844 unsigned long long suspend_point, array_size;
3845 unsigned long long backup_point, wait_point;
3846 unsigned long long reshape_completed;
3847 int done = 0;
3848 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
3849 int part = 0; /* The next part of the backup area to fill. It may already
3850 * be full, so we need to check */
3851 int level = reshape->level;
3852 int layout = reshape->before.layout;
3853 int data = reshape->before.data_disks;
3854 int disks = reshape->before.data_disks + reshape->parity;
3855 int chunk = sra->array.chunk_size;
3856 struct mdinfo *sd;
3857 unsigned long stripes;
3858 int uuid[4];
3859
3860 /* set up the backup-super-block. This requires the
3861 * uuid from the array.
3862 */
3863 /* Find a superblock */
3864 for (sd = sra->devs; sd; sd = sd->next) {
3865 char *dn;
3866 int devfd;
3867 int ok;
3868 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3869 continue;
3870 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
3871 devfd = dev_open(dn, O_RDONLY);
3872 if (devfd < 0)
3873 continue;
3874 ok = st->ss->load_super(st, devfd, NULL);
3875 close(devfd);
3876 if (ok == 0)
3877 break;
3878 }
3879 if (!sd) {
3880 pr_err("Cannot find a superblock\n");
3881 return 0;
3882 }
3883
3884 memset(&bsb, 0, 512);
3885 memcpy(bsb.magic, "md_backup_data-1", 16);
3886 st->ss->uuid_from_super(st, uuid);
3887 memcpy(bsb.set_uuid, uuid, 16);
3888 bsb.mtime = __cpu_to_le64(time(0));
3889 bsb.devstart2 = blocks;
3890
3891 stripes = blocks / (sra->array.chunk_size/512) /
3892 reshape->before.data_disks;
3893
3894 if (posix_memalign((void**)&buf, 4096, disks * chunk))
3895 /* Don't start the 'reshape' */
3896 return 0;
3897 if (reshape->before.data_disks == reshape->after.data_disks) {
3898 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
3899 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
3900 }
3901
3902 if (increasing) {
3903 array_size = sra->component_size * reshape->after.data_disks;
3904 backup_point = sra->reshape_progress;
3905 suspend_point = 0;
3906 } else {
3907 array_size = sra->component_size * reshape->before.data_disks;
3908 backup_point = reshape->backup_blocks;
3909 suspend_point = array_size;
3910 }
3911
3912 while (!done) {
3913 int rv;
3914
3915 /* Want to return as soon the oldest backup slot can
3916 * be released as that allows us to start backing up
3917 * some more, providing suspend_point has been
3918 * advanced, which it should have.
3919 */
3920 if (increasing) {
3921 wait_point = array_size;
3922 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
3923 wait_point = (__le64_to_cpu(bsb.arraystart) +
3924 __le64_to_cpu(bsb.length));
3925 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
3926 wait_point = (__le64_to_cpu(bsb.arraystart2) +
3927 __le64_to_cpu(bsb.length2));
3928 } else {
3929 wait_point = 0;
3930 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
3931 wait_point = __le64_to_cpu(bsb.arraystart);
3932 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
3933 wait_point = __le64_to_cpu(bsb.arraystart2);
3934 }
3935
3936 rv = progress_reshape(sra, reshape,
3937 backup_point, wait_point,
3938 &suspend_point, &reshape_completed);
3939 /* external metadata would need to ping_monitor here */
3940 sra->reshape_progress = reshape_completed;
3941
3942 /* Clear any backup region that is before 'here' */
3943 if (increasing) {
3944 if (__le64_to_cpu(bsb.length) > 0 &&
3945 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
3946 __le64_to_cpu(bsb.length)))
3947 forget_backup(dests, destfd,
3948 destoffsets, 0);
3949 if (__le64_to_cpu(bsb.length2) > 0 &&
3950 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
3951 __le64_to_cpu(bsb.length2)))
3952 forget_backup(dests, destfd,
3953 destoffsets, 1);
3954 } else {
3955 if (__le64_to_cpu(bsb.length) > 0 &&
3956 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
3957 forget_backup(dests, destfd,
3958 destoffsets, 0);
3959 if (__le64_to_cpu(bsb.length2) > 0 &&
3960 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
3961 forget_backup(dests, destfd,
3962 destoffsets, 1);
3963 }
3964
3965 if (rv < 0) {
3966 if (rv == -1)
3967 done = 1;
3968 break;
3969 }
3970 if (rv == 0 && increasing && !st->ss->external) {
3971 /* No longer need to monitor this reshape */
3972 done = 1;
3973 break;
3974 }
3975
3976 while (rv) {
3977 unsigned long long offset;
3978 unsigned long actual_stripes;
3979 /* Need to backup some data.
3980 * If 'part' is not used and the desired
3981 * backup size is suspended, do a backup,
3982 * then consider the next part.
3983 */
3984 /* Check that 'part' is unused */
3985 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
3986 break;
3987 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
3988 break;
3989
3990 offset = backup_point / data;
3991 actual_stripes = stripes;
3992 if (increasing) {
3993 if (offset + actual_stripes * (chunk/512) >
3994 sra->component_size)
3995 actual_stripes = ((sra->component_size - offset)
3996 / (chunk/512));
3997 if (offset + actual_stripes * (chunk/512) >
3998 suspend_point/data)
3999 break;
4000 } else {
4001 if (offset < actual_stripes * (chunk/512))
4002 actual_stripes = offset / (chunk/512);
4003 offset -= actual_stripes * (chunk/512);
4004 if (offset < suspend_point/data)
4005 break;
4006 }
4007 if (actual_stripes == 0)
4008 break;
4009 grow_backup(sra, offset, actual_stripes,
4010 fds, offsets,
4011 disks, chunk, level, layout,
4012 dests, destfd, destoffsets,
4013 part, &degraded, buf);
4014 validate(afd, destfd[0], destoffsets[0]);
4015 /* record where 'part' is up to */
4016 part = !part;
4017 if (increasing)
4018 backup_point += actual_stripes * (chunk/512) * data;
4019 else
4020 backup_point -= actual_stripes * (chunk/512) * data;
4021 }
4022 }
4023
4024 /* FIXME maybe call progress_reshape one more time instead */
4025 abort_reshape(sra); /* remove any remaining suspension */
4026 if (reshape->before.data_disks == reshape->after.data_disks)
4027 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4028 free(buf);
4029 return done;
4030 }
4031
4032 /*
4033 * If any spare contains md_back_data-1 which is recent wrt mtime,
4034 * write that data into the array and update the super blocks with
4035 * the new reshape_progress
4036 */
4037 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4038 char *backup_file, int verbose)
4039 {
4040 int i, j;
4041 int old_disks;
4042 unsigned long long *offsets;
4043 unsigned long long nstripe, ostripe;
4044 int ndata, odata;
4045
4046 odata = info->array.raid_disks - info->delta_disks - 1;
4047 if (info->array.level == 6) odata--; /* number of data disks */
4048 ndata = info->array.raid_disks - 1;
4049 if (info->new_level == 6) ndata--;
4050
4051 old_disks = info->array.raid_disks - info->delta_disks;
4052
4053 if (info->delta_disks <= 0)
4054 /* Didn't grow, so the backup file must have
4055 * been used
4056 */
4057 old_disks = cnt;
4058 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4059 struct mdinfo dinfo;
4060 int fd;
4061 int bsbsize;
4062 char *devname, namebuf[20];
4063 unsigned long long lo, hi;
4064
4065 /* This was a spare and may have some saved data on it.
4066 * Load the superblock, find and load the
4067 * backup_super_block.
4068 * If either fail, go on to next device.
4069 * If the backup contains no new info, just return
4070 * else restore data and update all superblocks
4071 */
4072 if (i == old_disks-1) {
4073 fd = open(backup_file, O_RDONLY);
4074 if (fd<0) {
4075 pr_err("backup file %s inaccessible: %s\n",
4076 backup_file, strerror(errno));
4077 continue;
4078 }
4079 devname = backup_file;
4080 } else {
4081 fd = fdlist[i];
4082 if (fd < 0)
4083 continue;
4084 if (st->ss->load_super(st, fd, NULL))
4085 continue;
4086
4087 st->ss->getinfo_super(st, &dinfo, NULL);
4088 st->ss->free_super(st);
4089
4090 if (lseek64(fd,
4091 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4092 0) < 0) {
4093 pr_err("Cannot seek on device %d\n", i);
4094 continue; /* Cannot seek */
4095 }
4096 sprintf(namebuf, "device-%d", i);
4097 devname = namebuf;
4098 }
4099 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4100 if (verbose)
4101 pr_err("Cannot read from %s\n", devname);
4102 continue; /* Cannot read */
4103 }
4104 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4105 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4106 if (verbose)
4107 pr_err("No backup metadata on %s\n", devname);
4108 continue;
4109 }
4110 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4111 if (verbose)
4112 pr_err("Bad backup-metadata checksum on %s\n", devname);
4113 continue; /* bad checksum */
4114 }
4115 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4116 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4117 if (verbose)
4118 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4119 continue; /* Bad second checksum */
4120 }
4121 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4122 if (verbose)
4123 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4124 continue; /* Wrong uuid */
4125 }
4126
4127 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4128 * sometimes they aren't... So allow considerable flexability in matching, and allow
4129 * this test to be overridden by an environment variable.
4130 */
4131 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4132 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4133 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4134 pr_err("accepting backup with timestamp %lu "
4135 "for array with timestamp %lu\n",
4136 (unsigned long)__le64_to_cpu(bsb.mtime),
4137 (unsigned long)info->array.utime);
4138 } else {
4139 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4140 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4141 continue; /* time stamp is too bad */
4142 }
4143 }
4144
4145 if (bsb.magic[15] == '1') {
4146 if (bsb.length == 0)
4147 continue;
4148 if (info->delta_disks >= 0) {
4149 /* reshape_progress is increasing */
4150 if (__le64_to_cpu(bsb.arraystart)
4151 + __le64_to_cpu(bsb.length)
4152 < info->reshape_progress) {
4153 nonew:
4154 if (verbose)
4155 pr_err("backup-metadata found on %s but is not needed\n", devname);
4156 continue; /* No new data here */
4157 }
4158 } else {
4159 /* reshape_progress is decreasing */
4160 if (__le64_to_cpu(bsb.arraystart) >=
4161 info->reshape_progress)
4162 goto nonew; /* No new data here */
4163 }
4164 } else {
4165 if (bsb.length == 0 && bsb.length2 == 0)
4166 continue;
4167 if (info->delta_disks >= 0) {
4168 /* reshape_progress is increasing */
4169 if ((__le64_to_cpu(bsb.arraystart)
4170 + __le64_to_cpu(bsb.length)
4171 < info->reshape_progress)
4172 &&
4173 (__le64_to_cpu(bsb.arraystart2)
4174 + __le64_to_cpu(bsb.length2)
4175 < info->reshape_progress))
4176 goto nonew; /* No new data here */
4177 } else {
4178 /* reshape_progress is decreasing */
4179 if (__le64_to_cpu(bsb.arraystart) >=
4180 info->reshape_progress &&
4181 __le64_to_cpu(bsb.arraystart2) >=
4182 info->reshape_progress)
4183 goto nonew; /* No new data here */
4184 }
4185 }
4186 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4187 second_fail:
4188 if (verbose)
4189 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4190 devname);
4191 continue; /* Cannot seek */
4192 }
4193 /* There should be a duplicate backup superblock 4k before here */
4194 if (lseek64(fd, -4096, 1) < 0 ||
4195 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4196 goto second_fail; /* Cannot find leading superblock */
4197 if (bsb.magic[15] == '1')
4198 bsbsize = offsetof(struct mdp_backup_super, pad1);
4199 else
4200 bsbsize = offsetof(struct mdp_backup_super, pad);
4201 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4202 goto second_fail; /* Cannot find leading superblock */
4203
4204 /* Now need the data offsets for all devices. */
4205 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4206 for(j=0; j<info->array.raid_disks; j++) {
4207 if (fdlist[j] < 0)
4208 continue;
4209 if (st->ss->load_super(st, fdlist[j], NULL))
4210 /* FIXME should be this be an error */
4211 continue;
4212 st->ss->getinfo_super(st, &dinfo, NULL);
4213 st->ss->free_super(st);
4214 offsets[j] = dinfo.data_offset * 512;
4215 }
4216 printf(Name ": restoring critical section\n");
4217
4218 if (restore_stripes(fdlist, offsets,
4219 info->array.raid_disks,
4220 info->new_chunk,
4221 info->new_level,
4222 info->new_layout,
4223 fd, __le64_to_cpu(bsb.devstart)*512,
4224 __le64_to_cpu(bsb.arraystart)*512,
4225 __le64_to_cpu(bsb.length)*512, NULL)) {
4226 /* didn't succeed, so giveup */
4227 if (verbose)
4228 pr_err("Error restoring backup from %s\n",
4229 devname);
4230 free(offsets);
4231 return 1;
4232 }
4233
4234 if (bsb.magic[15] == '2' &&
4235 restore_stripes(fdlist, offsets,
4236 info->array.raid_disks,
4237 info->new_chunk,
4238 info->new_level,
4239 info->new_layout,
4240 fd, __le64_to_cpu(bsb.devstart)*512 +
4241 __le64_to_cpu(bsb.devstart2)*512,
4242 __le64_to_cpu(bsb.arraystart2)*512,
4243 __le64_to_cpu(bsb.length2)*512, NULL)) {
4244 /* didn't succeed, so giveup */
4245 if (verbose)
4246 pr_err("Error restoring second backup from %s\n",
4247 devname);
4248 free(offsets);
4249 return 1;
4250 }
4251
4252 free(offsets);
4253
4254 /* Ok, so the data is restored. Let's update those superblocks. */
4255
4256 lo = hi = 0;
4257 if (bsb.length) {
4258 lo = __le64_to_cpu(bsb.arraystart);
4259 hi = lo + __le64_to_cpu(bsb.length);
4260 }
4261 if (bsb.magic[15] == '2' && bsb.length2) {
4262 unsigned long long lo1, hi1;
4263 lo1 = __le64_to_cpu(bsb.arraystart2);
4264 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4265 if (lo == hi) {
4266 lo = lo1;
4267 hi = hi1;
4268 } else if (lo < lo1)
4269 hi = hi1;
4270 else
4271 lo = lo1;
4272 }
4273 if (lo < hi &&
4274 (info->reshape_progress < lo ||
4275 info->reshape_progress > hi))
4276 /* backup does not affect reshape_progress*/ ;
4277 else if (info->delta_disks >= 0) {
4278 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4279 __le64_to_cpu(bsb.length);
4280 if (bsb.magic[15] == '2') {
4281 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4282 __le64_to_cpu(bsb.length2);
4283 if (p2 > info->reshape_progress)
4284 info->reshape_progress = p2;
4285 }
4286 } else {
4287 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4288 if (bsb.magic[15] == '2') {
4289 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4290 if (p2 < info->reshape_progress)
4291 info->reshape_progress = p2;
4292 }
4293 }
4294 for (j=0; j<info->array.raid_disks; j++) {
4295 if (fdlist[j] < 0)
4296 continue;
4297 if (st->ss->load_super(st, fdlist[j], NULL))
4298 continue;
4299 st->ss->getinfo_super(st, &dinfo, NULL);
4300 dinfo.reshape_progress = info->reshape_progress;
4301 st->ss->update_super(st, &dinfo,
4302 "_reshape_progress",
4303 NULL,0, 0, NULL);
4304 st->ss->store_super(st, fdlist[j]);
4305 st->ss->free_super(st);
4306 }
4307 return 0;
4308 }
4309 /* Didn't find any backup data, try to see if any
4310 * was needed.
4311 */
4312 if (info->delta_disks < 0) {
4313 /* When shrinking, the critical section is at the end.
4314 * So see if we are before the critical section.
4315 */
4316 unsigned long long first_block;
4317 nstripe = ostripe = 0;
4318 first_block = 0;
4319 while (ostripe >= nstripe) {
4320 ostripe += info->array.chunk_size / 512;
4321 first_block = ostripe * odata;
4322 nstripe = first_block / ndata / (info->new_chunk/512) *
4323 (info->new_chunk/512);
4324 }
4325
4326 if (info->reshape_progress >= first_block)
4327 return 0;
4328 }
4329 if (info->delta_disks > 0) {
4330 /* See if we are beyond the critical section. */
4331 unsigned long long last_block;
4332 nstripe = ostripe = 0;
4333 last_block = 0;
4334 while (nstripe >= ostripe) {
4335 nstripe += info->new_chunk / 512;
4336 last_block = nstripe * ndata;
4337 ostripe = last_block / odata / (info->array.chunk_size/512) *
4338 (info->array.chunk_size/512);
4339 }
4340
4341 if (info->reshape_progress >= last_block)
4342 return 0;
4343 }
4344 /* needed to recover critical section! */
4345 if (verbose)
4346 pr_err("Failed to find backup of critical section\n");
4347 return 1;
4348 }
4349
4350 int Grow_continue_command(char *devname, int fd,
4351 char *backup_file, int verbose)
4352 {
4353 int ret_val = 0;
4354 struct supertype *st = NULL;
4355 struct mdinfo *content = NULL;
4356 struct mdinfo array;
4357 char *subarray = NULL;
4358 struct mdinfo *cc = NULL;
4359 struct mdstat_ent *mdstat = NULL;
4360 char buf[40];
4361 int cfd = -1;
4362 int fd2 = -1;
4363
4364 dprintf("Grow continue from command line called for %s\n",
4365 devname);
4366
4367 st = super_by_fd(fd, &subarray);
4368 if (!st || !st->ss) {
4369 pr_err("Unable to determine metadata format for %s\n",
4370 devname);
4371 return 1;
4372 }
4373 dprintf("Grow continue is run for ");
4374 if (st->ss->external == 0) {
4375 int d;
4376 dprintf("native array (%s)\n", devname);
4377 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4378 pr_err("%s is not an active md array -"
4379 " aborting\n", devname);
4380 ret_val = 1;
4381 goto Grow_continue_command_exit;
4382 }
4383 content = &array;
4384 /* Need to load a superblock.
4385 * FIXME we should really get what we need from
4386 * sysfs
4387 */
4388 for (d = 0; d < MAX_DISKS; d++) {
4389 mdu_disk_info_t disk;
4390 char *dv;
4391 int err;
4392 disk.number = d;
4393 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4394 continue;
4395 if (disk.major == 0 && disk.minor == 0)
4396 continue;
4397 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4398 continue;
4399 dv = map_dev(disk.major, disk.minor, 1);
4400 if (!dv)
4401 continue;
4402 fd2 = dev_open(dv, O_RDONLY);
4403 if (fd2 < 0)
4404 continue;
4405 err = st->ss->load_super(st, fd2, NULL);
4406 close(fd2);
4407 if (err)
4408 continue;
4409 break;
4410 }
4411 if (d == MAX_DISKS) {
4412 pr_err("Unable to load metadata for %s\n",
4413 devname);
4414 ret_val = 1;
4415 goto Grow_continue_command_exit;
4416 }
4417 st->ss->getinfo_super(st, content, NULL);
4418 } else {
4419 int container_dev;
4420
4421 if (subarray) {
4422 dprintf("subarray (%s)\n", subarray);
4423 container_dev = st->container_dev;
4424 cfd = open_dev_excl(st->container_dev);
4425 } else {
4426 container_dev = st->devnum;
4427 close(fd);
4428 cfd = open_dev_excl(st->devnum);
4429 dprintf("container (%i)\n", container_dev);
4430 fd = cfd;
4431 }
4432 if (cfd < 0) {
4433 pr_err("Unable to open container "
4434 "for %s\n", devname);
4435 ret_val = 1;
4436 goto Grow_continue_command_exit;
4437 }
4438 fmt_devname(buf, container_dev);
4439
4440 /* find in container array under reshape
4441 */
4442 ret_val = st->ss->load_container(st, cfd, NULL);
4443 if (ret_val) {
4444 pr_err("Cannot read superblock for %s\n",
4445 devname);
4446 ret_val = 1;
4447 goto Grow_continue_command_exit;
4448 }
4449
4450 cc = st->ss->container_content(st, subarray);
4451 for (content = cc; content ; content = content->next) {
4452 char *array;
4453 int allow_reshape = 1;
4454
4455 if (content->reshape_active == 0)
4456 continue;
4457 /* The decision about array or container wide
4458 * reshape is taken in Grow_continue based
4459 * content->reshape_active state, therefore we
4460 * need to check_reshape based on
4461 * reshape_active and subarray name
4462 */
4463 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4464 allow_reshape = 0;
4465 if (content->reshape_active == CONTAINER_RESHAPE &&
4466 (content->array.state
4467 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4468 allow_reshape = 0;
4469
4470 if (!allow_reshape) {
4471 pr_err("cannot continue reshape of an array"
4472 " in container with unsupported"
4473 " metadata: %s(%s)\n",
4474 devname, buf);
4475 ret_val = 1;
4476 goto Grow_continue_command_exit;
4477 }
4478
4479 array = strchr(content->text_version+1, '/')+1;
4480 mdstat = mdstat_by_subdev(array, container_dev);
4481 if (!mdstat)
4482 continue;
4483 if (mdstat->active == 0) {
4484 pr_err("Skipping inactive "
4485 "array md%i.\n", mdstat->devnum);
4486 free_mdstat(mdstat);
4487 mdstat = NULL;
4488 continue;
4489 }
4490 break;
4491 }
4492 if (!content) {
4493 pr_err("Unable to determine reshaped "
4494 "array for %s\n", devname);
4495 ret_val = 1;
4496 goto Grow_continue_command_exit;
4497 }
4498 fd2 = open_dev(mdstat->devnum);
4499 if (fd2 < 0) {
4500 pr_err("cannot open (md%i)\n",
4501 mdstat->devnum);
4502 ret_val = 1;
4503 goto Grow_continue_command_exit;
4504 }
4505
4506 sysfs_init(content, fd2, mdstat->devnum);
4507
4508 /* start mdmon in case it is not running
4509 */
4510 if (!mdmon_running(container_dev))
4511 start_mdmon(container_dev);
4512 ping_monitor(buf);
4513
4514 if (mdmon_running(container_dev))
4515 st->update_tail = &st->updates;
4516 else {
4517 pr_err("No mdmon found. "
4518 "Grow cannot continue.\n");
4519 ret_val = 1;
4520 goto Grow_continue_command_exit;
4521 }
4522 }
4523
4524 /* verify that array under reshape is started from
4525 * correct position
4526 */
4527 if (verify_reshape_position(content, content->array.level) < 0) {
4528 ret_val = 1;
4529 goto Grow_continue_command_exit;
4530 }
4531
4532 /* continue reshape
4533 */
4534 ret_val = Grow_continue(fd, st, content, backup_file, 0);
4535
4536 Grow_continue_command_exit:
4537 if (fd2 > -1)
4538 close(fd2);
4539 if (cfd > -1)
4540 close(cfd);
4541 st->ss->free_super(st);
4542 free_mdstat(mdstat);
4543 sysfs_free(cc);
4544 free(subarray);
4545
4546 return ret_val;
4547 }
4548
4549 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4550 char *backup_file, int freeze_reshape)
4551 {
4552 int ret_val = 2;
4553
4554 if (!info->reshape_active)
4555 return ret_val;
4556
4557 if (st->ss->external) {
4558 char container[40];
4559 int cfd = open_dev(st->container_dev);
4560
4561 if (cfd < 0)
4562 return 1;
4563
4564 fmt_devname(container, st->container_dev);
4565 st->ss->load_container(st, cfd, container);
4566 close(cfd);
4567 ret_val = reshape_container(container, NULL, mdfd,
4568 st, info, 0, backup_file,
4569 0, 1, freeze_reshape);
4570 } else
4571 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4572 NULL, 0ULL, backup_file, 0, 0, 1,
4573 freeze_reshape);
4574
4575 return ret_val;
4576 }
Unnamed repository; edit this file 'description' to name the repository.