]> 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: make warning about old metadata more explicit.
[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 compute_backup_blocks(int nchunk, int ochunk,
942 unsigned int ndata, unsigned int odata)
943 {
944 unsigned long a, b, blocks;
945 /* So how much do we need to backup.
946 * We need an amount of data which is both a whole number of
947 * old stripes and a whole number of new stripes.
948 * So LCM for (chunksize*datadisks).
949 */
950 a = (ochunk/512) * odata;
951 b = (nchunk/512) * ndata;
952 /* Find GCD */
953 while (a != b) {
954 if (a < b)
955 b -= a;
956 if (b < a)
957 a -= b;
958 }
959 /* LCM == product / GCD */
960 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
961
962 return blocks;
963 }
964
965 char *analyse_change(struct mdinfo *info, struct reshape *re)
966 {
967 /* Based on the current array state in info->array and
968 * the changes in info->new_* etc, determine:
969 * - whether the change is possible
970 * - Intermediate level/raid_disks/layout
971 * - whether a restriping reshape is needed
972 * - number of sectors in minimum change unit. This
973 * will cover a whole number of stripes in 'before' and
974 * 'after'.
975 *
976 * Return message if the change should be rejected
977 * NULL if the change can be achieved
978 *
979 * This can be called as part of starting a reshape, or
980 * when assembling an array that is undergoing reshape.
981 */
982 int new_disks;
983 /* delta_parity records change in number of devices
984 * caused by level change
985 */
986 int delta_parity = 0;
987
988 /* If a new level not explicitly given, we assume no-change */
989 if (info->new_level == UnSet)
990 info->new_level = info->array.level;
991
992 if (info->new_chunk)
993 switch (info->new_level) {
994 case 0:
995 case 4:
996 case 5:
997 case 6:
998 case 10:
999 /* chunk size is meaningful, must divide component_size
1000 * evenly
1001 */
1002 if (info->component_size % (info->new_chunk/512))
1003 return "New chunk size does not"
1004 " divide component size";
1005 break;
1006 default:
1007 return "chunk size not meaningful for this level";
1008 }
1009 else
1010 info->new_chunk = info->array.chunk_size;
1011
1012 switch (info->array.level) {
1013 default:
1014 return "Cannot understand this RAID level";
1015 case 1:
1016 /* RAID1 can convert to RAID1 with different disks, or
1017 * raid5 with 2 disks, or
1018 * raid0 with 1 disk
1019 */
1020 if (info->new_level > 1 &&
1021 (info->component_size & 7))
1022 return "Cannot convert RAID1 of this size - "
1023 "reduce size to multiple of 4K first.";
1024 if (info->new_level == 0) {
1025 if (info->delta_disks != UnSet &&
1026 info->delta_disks != 0)
1027 return "Cannot change number of disks "
1028 "with RAID1->RAID0 conversion";
1029 re->level = 0;
1030 re->before.data_disks = 1;
1031 re->after.data_disks = 1;
1032 re->before.layout = 0;
1033 re->backup_blocks = 0;
1034 re->parity = 0;
1035 return NULL;
1036 }
1037 if (info->new_level == 1) {
1038 if (info->delta_disks == UnSet)
1039 /* Don't know what to do */
1040 return "no change requested for Growing RAID1";
1041 re->level = 1;
1042 re->backup_blocks = 0;
1043 re->parity = 0;
1044 return NULL;
1045 }
1046 if (info->array.raid_disks == 2 &&
1047 info->new_level == 5) {
1048
1049 re->level = 5;
1050 re->before.data_disks = 1;
1051 if (info->delta_disks != UnSet &&
1052 info->delta_disks != 0)
1053 re->after.data_disks = 1 + info->delta_disks;
1054 else
1055 re->after.data_disks = 1;
1056 if (re->after.data_disks < 1)
1057 return "Number of disks too small for RAID5";
1058
1059 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1060 info->array.chunk_size = 65536;
1061 break;
1062 }
1063 /* Could do some multi-stage conversions, but leave that to
1064 * later.
1065 */
1066 return "Impossibly level change request for RAID1";
1067
1068 case 10:
1069 /* RAID10 can only be converted from near mode to
1070 * RAID0 by removing some devices
1071 */
1072 if ((info->array.layout & ~0xff) != 0x100)
1073 return "Cannot Grow RAID10 with far/offset layout";
1074 /* number of devices must be multiple of number of copies */
1075 if (info->array.raid_disks % (info->array.layout & 0xff))
1076 return "RAID10 layout too complex for Grow operation";
1077
1078 if (info->new_level != 0)
1079 return "RAID10 can only be changed to RAID0";
1080 new_disks = (info->array.raid_disks
1081 / (info->array.layout & 0xff));
1082 if (info->delta_disks == UnSet)
1083 info->delta_disks = (new_disks
1084 - info->array.raid_disks);
1085
1086 if (info->delta_disks != new_disks - info->array.raid_disks)
1087 return "New number of raid-devices impossible for RAID10";
1088 if (info->new_chunk &&
1089 info->new_chunk != info->array.chunk_size)
1090 return "Cannot change chunk-size with RAID10 Grow";
1091
1092 /* looks good */
1093 re->level = 0;
1094 re->parity = 0;
1095 re->before.data_disks = new_disks;
1096 re->after.data_disks = re->before.data_disks;
1097 re->before.layout = 0;
1098 re->backup_blocks = 0;
1099 return NULL;
1100
1101 case 0:
1102 /* RAID0 can be converted to RAID10, or to RAID456 */
1103 if (info->new_level == 10) {
1104 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1105 /* Assume near=2 layout */
1106 info->new_layout = 0x102;
1107 info->delta_disks = info->array.raid_disks;
1108 }
1109 if (info->new_layout == UnSet) {
1110 int copies = 1 + (info->delta_disks
1111 / info->array.raid_disks);
1112 if (info->array.raid_disks * (copies-1)
1113 != info->delta_disks)
1114 return "Impossible number of devices"
1115 " for RAID0->RAID10";
1116 info->new_layout = 0x100 + copies;
1117 }
1118 if (info->delta_disks == UnSet) {
1119 int copies = info->new_layout & 0xff;
1120 if (info->new_layout != 0x100 + copies)
1121 return "New layout impossible"
1122 " for RAID0->RAID10";;
1123 info->delta_disks = (copies - 1) *
1124 info->array.raid_disks;
1125 }
1126 if (info->new_chunk &&
1127 info->new_chunk != info->array.chunk_size)
1128 return "Cannot change chunk-size with RAID0->RAID10";
1129 /* looks good */
1130 re->level = 10;
1131 re->parity = 0;
1132 re->before.data_disks = (info->array.raid_disks +
1133 info->delta_disks);
1134 re->after.data_disks = re->before.data_disks;
1135 re->before.layout = info->new_layout;
1136 re->backup_blocks = 0;
1137 return NULL;
1138 }
1139
1140 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1141 * a raid4 style layout of the final level.
1142 */
1143 switch (info->new_level) {
1144 case 4:
1145 delta_parity = 1;
1146 case 0:
1147 re->level = 4;
1148 re->before.layout = 0;
1149 break;
1150 case 5:
1151 delta_parity = 1;
1152 re->level = 5;
1153 re->before.layout = ALGORITHM_PARITY_N;
1154 break;
1155 case 6:
1156 delta_parity = 2;
1157 re->level = 6;
1158 re->before.layout = ALGORITHM_PARITY_N;
1159 break;
1160 default:
1161 return "Impossible level change requested";
1162 }
1163 re->before.data_disks = info->array.raid_disks;
1164 /* determining 'after' layout happens outside this 'switch' */
1165 break;
1166
1167 case 4:
1168 info->array.layout = ALGORITHM_PARITY_N;
1169 case 5:
1170 switch (info->new_level) {
1171 case 0:
1172 delta_parity = -1;
1173 case 4:
1174 re->level = info->array.level;
1175 re->before.data_disks = info->array.raid_disks - 1;
1176 re->before.layout = info->array.layout;
1177 break;
1178 case 5:
1179 re->level = 5;
1180 re->before.data_disks = info->array.raid_disks - 1;
1181 re->before.layout = info->array.layout;
1182 break;
1183 case 6:
1184 delta_parity = 1;
1185 re->level = 6;
1186 re->before.data_disks = info->array.raid_disks - 1;
1187 switch (info->array.layout) {
1188 case ALGORITHM_LEFT_ASYMMETRIC:
1189 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1190 break;
1191 case ALGORITHM_RIGHT_ASYMMETRIC:
1192 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1193 break;
1194 case ALGORITHM_LEFT_SYMMETRIC:
1195 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1196 break;
1197 case ALGORITHM_RIGHT_SYMMETRIC:
1198 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1199 break;
1200 case ALGORITHM_PARITY_0:
1201 re->before.layout = ALGORITHM_PARITY_0_6;
1202 break;
1203 case ALGORITHM_PARITY_N:
1204 re->before.layout = ALGORITHM_PARITY_N_6;
1205 break;
1206 default:
1207 return "Cannot convert an array with this layout";
1208 }
1209 break;
1210 case 1:
1211 if (info->array.raid_disks != 2)
1212 return "Can only convert a 2-device array to RAID1";
1213 if (info->delta_disks != UnSet &&
1214 info->delta_disks != 0)
1215 return "Cannot set raid_disk when "
1216 "converting RAID5->RAID1";
1217 re->level = 1;
1218 break;
1219 default:
1220 return "Impossible level change requested";
1221 }
1222 break;
1223 case 6:
1224 switch (info->new_level) {
1225 case 4:
1226 case 5:
1227 delta_parity = -1;
1228 case 6:
1229 re->level = 6;
1230 re->before.data_disks = info->array.raid_disks - 2;
1231 re->before.layout = info->array.layout;
1232 break;
1233 default:
1234 return "Impossible level change requested";
1235 }
1236 break;
1237 }
1238
1239 /* If we reached here then it looks like a re-stripe is
1240 * happening. We have determined the intermediate level
1241 * and initial raid_disks/layout and stored these in 're'.
1242 *
1243 * We need to deduce the final layout that can be atomically
1244 * converted to the end state.
1245 */
1246 switch (info->new_level) {
1247 case 0:
1248 /* We can only get to RAID0 from RAID4 or RAID5
1249 * with appropriate layout and one extra device
1250 */
1251 if (re->level != 4 && re->level != 5)
1252 return "Cannot covert to RAID0 from this level";
1253
1254 switch (re->level) {
1255 case 4:
1256 re->after.layout = 0;
1257 break;
1258 case 5:
1259 re->after.layout = ALGORITHM_PARITY_N;
1260 break;
1261 }
1262 break;
1263
1264 case 4:
1265 /* We can only get to RAID4 from RAID5 */
1266 if (re->level != 4 && re->level != 5)
1267 return "Cannot convert to RAID4 from this level";
1268
1269 switch (re->level) {
1270 case 4:
1271 re->after.layout = 0;
1272 break;
1273 case 5:
1274 re->after.layout = ALGORITHM_PARITY_N;
1275 break;
1276 }
1277 break;
1278
1279 case 5:
1280 /* We get to RAID5 from RAID5 or RAID6 */
1281 if (re->level != 5 && re->level != 6)
1282 return "Cannot convert to RAID5 from this level";
1283
1284 switch (re->level) {
1285 case 5:
1286 if (info->new_layout == UnSet)
1287 re->after.layout = re->before.layout;
1288 else
1289 re->after.layout = info->new_layout;
1290 break;
1291 case 6:
1292 if (info->new_layout == UnSet)
1293 info->new_layout = re->before.layout;
1294
1295 /* after.layout needs to be raid6 version of new_layout */
1296 if (info->new_layout == ALGORITHM_PARITY_N)
1297 re->after.layout = ALGORITHM_PARITY_N;
1298 else {
1299 char layout[40];
1300 char *ls = map_num(r5layout, info->new_layout);
1301 int l;
1302 if (ls) {
1303 /* Current RAID6 layout has a RAID5
1304 * equivalent - good
1305 */
1306 strcat(strcpy(layout, ls), "-6");
1307 l = map_name(r6layout, layout);
1308 if (l == UnSet)
1309 return "Cannot find RAID6 layout"
1310 " to convert to";
1311 } else {
1312 /* Current RAID6 has no equivalent.
1313 * If it is already a '-6' layout we
1314 * can leave it unchanged, else we must
1315 * fail
1316 */
1317 ls = map_num(r6layout, info->new_layout);
1318 if (!ls ||
1319 strcmp(ls+strlen(ls)-2, "-6") != 0)
1320 return "Please specify new layout";
1321 l = info->new_layout;
1322 }
1323 re->after.layout = l;
1324 }
1325 }
1326 break;
1327
1328 case 6:
1329 /* We must already be at level 6 */
1330 if (re->level != 6)
1331 return "Impossible level change";
1332 if (info->new_layout == UnSet)
1333 re->after.layout = info->array.layout;
1334 else
1335 re->after.layout = info->new_layout;
1336 break;
1337 default:
1338 return "Impossible level change requested";
1339 }
1340 if (info->delta_disks == UnSet)
1341 info->delta_disks = delta_parity;
1342
1343 re->after.data_disks = (re->before.data_disks
1344 + info->delta_disks
1345 - delta_parity);
1346 switch (re->level) {
1347 case 6: re->parity = 2;
1348 break;
1349 case 4:
1350 case 5: re->parity = 1;
1351 break;
1352 default: re->parity = 0;
1353 break;
1354 }
1355 /* So we have a restripe operation, we need to calculate the number
1356 * of blocks per reshape operation.
1357 */
1358 if (info->new_chunk == 0)
1359 info->new_chunk = info->array.chunk_size;
1360 if (re->after.data_disks == re->before.data_disks &&
1361 re->after.layout == re->before.layout &&
1362 info->new_chunk == info->array.chunk_size) {
1363 /* Nothing to change */
1364 re->backup_blocks = 0;
1365 return NULL;
1366 }
1367 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1368 /* chunk and layout changes make no difference */
1369 re->backup_blocks = 0;
1370 return NULL;
1371 }
1372
1373 if (re->after.data_disks == re->before.data_disks &&
1374 get_linux_version() < 2006032)
1375 return "in-place reshape is not safe before 2.6.32 - sorry.";
1376
1377 if (re->after.data_disks < re->before.data_disks &&
1378 get_linux_version() < 2006030)
1379 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1380
1381 re->backup_blocks = compute_backup_blocks(
1382 info->new_chunk, info->array.chunk_size,
1383 re->after.data_disks,
1384 re->before.data_disks);
1385
1386 re->new_size = info->component_size * re->after.data_disks;
1387 return NULL;
1388 }
1389
1390 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1391 char *text_version)
1392 {
1393 struct mdinfo *info;
1394 char *subarray;
1395 int ret_val = -1;
1396
1397 if ((st == NULL) || (sra == NULL))
1398 return ret_val;
1399
1400 if (text_version == NULL)
1401 text_version = sra->text_version;
1402 subarray = strchr(text_version+1, '/')+1;
1403 info = st->ss->container_content(st, subarray);
1404 if (info) {
1405 unsigned long long current_size = 0;
1406 unsigned long long new_size =
1407 info->custom_array_size/2;
1408
1409 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1410 new_size > current_size) {
1411 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1412 < 0)
1413 dprintf("Error: Cannot set array size");
1414 else {
1415 ret_val = 0;
1416 dprintf("Array size changed");
1417 }
1418 dprintf(" from %llu to %llu.\n",
1419 current_size, new_size);
1420 }
1421 sysfs_free(info);
1422 } else
1423 dprintf("Error: set_array_size(): info pointer in NULL\n");
1424
1425 return ret_val;
1426 }
1427
1428 static int reshape_array(char *container, int fd, char *devname,
1429 struct supertype *st, struct mdinfo *info,
1430 int force, struct mddev_dev *devlist,
1431 char *backup_file, int verbose, int forked,
1432 int restart, int freeze_reshape);
1433 static int reshape_container(char *container, char *devname,
1434 int mdfd,
1435 struct supertype *st,
1436 struct mdinfo *info,
1437 int force,
1438 char *backup_file,
1439 int verbose, int restart, int freeze_reshape);
1440
1441 int Grow_reshape(char *devname, int fd,
1442 struct mddev_dev *devlist,
1443 struct context *c, struct shape *s)
1444 {
1445 /* Make some changes in the shape of an array.
1446 * The kernel must support the change.
1447 *
1448 * There are three different changes. Each can trigger
1449 * a resync or recovery so we freeze that until we have
1450 * requested everything (if kernel supports freezing - 2.6.30).
1451 * The steps are:
1452 * - change size (i.e. component_size)
1453 * - change level
1454 * - change layout/chunksize/ndisks
1455 *
1456 * The last can require a reshape. It is different on different
1457 * levels so we need to check the level before actioning it.
1458 * Some times the level change needs to be requested after the
1459 * reshape (e.g. raid6->raid5, raid5->raid0)
1460 *
1461 */
1462 struct mdu_array_info_s array;
1463 int rv = 0;
1464 struct supertype *st;
1465 char *subarray = NULL;
1466
1467 int frozen;
1468 int changed = 0;
1469 char *container = NULL;
1470 char container_buf[20];
1471 int cfd = -1;
1472
1473 struct mddev_dev *dv;
1474 int added_disks;
1475
1476 struct mdinfo info;
1477 struct mdinfo *sra;
1478
1479 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1480 pr_err("%s is not an active md array - aborting\n",
1481 devname);
1482 return 1;
1483 }
1484
1485 if (s->size > 0 &&
1486 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1487 pr_err("cannot change component size at the same time "
1488 "as other changes.\n"
1489 " Change size first, then check data is intact before "
1490 "making other changes.\n");
1491 return 1;
1492 }
1493
1494 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1495 get_linux_version() < 2006032 &&
1496 !check_env("MDADM_FORCE_FEWER")) {
1497 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1498 " Please use a newer kernel\n");
1499 return 1;
1500 }
1501
1502 st = super_by_fd(fd, &subarray);
1503 if (!st) {
1504 pr_err("Unable to determine metadata format for %s\n", devname);
1505 return 1;
1506 }
1507 if (s->raiddisks > st->max_devs) {
1508 pr_err("Cannot increase raid-disks on this array"
1509 " beyond %d\n", st->max_devs);
1510 return 1;
1511 }
1512
1513 /* in the external case we need to check that the requested reshape is
1514 * supported, and perform an initial check that the container holds the
1515 * pre-requisite spare devices (mdmon owns final validation)
1516 */
1517 if (st->ss->external) {
1518 int container_dev;
1519 int rv;
1520
1521 if (subarray) {
1522 container_dev = st->container_dev;
1523 cfd = open_dev_excl(st->container_dev);
1524 } else {
1525 container_dev = st->devnum;
1526 close(fd);
1527 cfd = open_dev_excl(st->devnum);
1528 fd = cfd;
1529 }
1530 if (cfd < 0) {
1531 pr_err("Unable to open container for %s\n",
1532 devname);
1533 free(subarray);
1534 return 1;
1535 }
1536
1537 fmt_devname(container_buf, container_dev);
1538 container = container_buf;
1539
1540 rv = st->ss->load_container(st, cfd, NULL);
1541
1542 if (rv) {
1543 pr_err("Cannot read superblock for %s\n",
1544 devname);
1545 free(subarray);
1546 return 1;
1547 }
1548
1549 /* check if operation is supported for metadata handler */
1550 if (st->ss->container_content) {
1551 struct mdinfo *cc = NULL;
1552 struct mdinfo *content = NULL;
1553
1554 cc = st->ss->container_content(st, subarray);
1555 for (content = cc; content ; content = content->next) {
1556 int allow_reshape = 1;
1557
1558 /* check if reshape is allowed based on metadata
1559 * indications stored in content.array.status
1560 */
1561 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1562 allow_reshape = 0;
1563 if (content->array.state
1564 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1565 allow_reshape = 0;
1566 if (!allow_reshape) {
1567 pr_err("cannot reshape arrays in"
1568 " container with unsupported"
1569 " metadata: %s(%s)\n",
1570 devname, container_buf);
1571 sysfs_free(cc);
1572 free(subarray);
1573 return 1;
1574 }
1575 }
1576 sysfs_free(cc);
1577 }
1578 if (mdmon_running(container_dev))
1579 st->update_tail = &st->updates;
1580 }
1581
1582 added_disks = 0;
1583 for (dv = devlist; dv; dv = dv->next)
1584 added_disks++;
1585 if (s->raiddisks > array.raid_disks &&
1586 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1587 !c->force) {
1588 pr_err("Need %d spare%s to avoid degraded array,"
1589 " and only have %d.\n"
1590 " Use --force to over-ride this check.\n",
1591 s->raiddisks - array.raid_disks,
1592 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1593 array.spare_disks + added_disks);
1594 return 1;
1595 }
1596
1597 sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS
1598 | GET_STATE | GET_VERSION);
1599 if (sra) {
1600 if (st->ss->external && subarray == NULL) {
1601 array.level = LEVEL_CONTAINER;
1602 sra->array.level = LEVEL_CONTAINER;
1603 }
1604 } else {
1605 pr_err("failed to read sysfs parameters for %s\n",
1606 devname);
1607 return 1;
1608 }
1609 frozen = freeze(st);
1610 if (frozen < -1) {
1611 /* freeze() already spewed the reason */
1612 sysfs_free(sra);
1613 return 1;
1614 } else if (frozen < 0) {
1615 pr_err("%s is performing resync/recovery and cannot"
1616 " be reshaped\n", devname);
1617 sysfs_free(sra);
1618 return 1;
1619 }
1620
1621 /* ========= set size =============== */
1622 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1623 unsigned long long orig_size = get_component_size(fd)/2;
1624 unsigned long long min_csize;
1625 struct mdinfo *mdi;
1626 int raid0_takeover = 0;
1627
1628 if (orig_size == 0)
1629 orig_size = (unsigned) array.size;
1630
1631 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1632 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1633 rv = 1;
1634 goto release;
1635 }
1636 sync_metadata(st);
1637 if (st->ss->external) {
1638 /* metadata can have size limitation
1639 * update size value according to metadata information
1640 */
1641 struct mdinfo *sizeinfo =
1642 st->ss->container_content(st, subarray);
1643 if (sizeinfo) {
1644 unsigned long long new_size =
1645 sizeinfo->custom_array_size/2;
1646 int data_disks = get_data_disks(
1647 sizeinfo->array.level,
1648 sizeinfo->array.layout,
1649 sizeinfo->array.raid_disks);
1650 new_size /= data_disks;
1651 dprintf("Metadata size correction from %llu to "
1652 "%llu (%llu)\n", orig_size, new_size,
1653 new_size * data_disks);
1654 s->size = new_size;
1655 sysfs_free(sizeinfo);
1656 }
1657 }
1658
1659 /* Update the size of each member device in case
1660 * they have been resized. This will never reduce
1661 * below the current used-size. The "size" attribute
1662 * understands '0' to mean 'max'.
1663 */
1664 min_csize = 0;
1665 rv = 0;
1666 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1667 if (sysfs_set_num(sra, mdi, "size",
1668 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1669 /* Probably kernel refusing to let us
1670 * reduce the size - not an error.
1671 */
1672 break;
1673 }
1674 if (array.not_persistent == 0 &&
1675 array.major_version == 0 &&
1676 get_linux_version() < 3001000) {
1677 /* Dangerous to allow size to exceed 2TB */
1678 unsigned long long csize;
1679 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1680 if (csize >= 2ULL*1024*1024*1024)
1681 csize = 2ULL*1024*1024*1024;
1682 if ((min_csize == 0 || (min_csize
1683 > csize)))
1684 min_csize = csize;
1685 }
1686 }
1687 }
1688 if (rv) {
1689 pr_err("Cannot set size on "
1690 "array members.\n");
1691 goto size_change_error;
1692 }
1693 if (min_csize && s->size > min_csize) {
1694 pr_err("Cannot safely make this array "
1695 "use more than 2TB per device on this kernel.\n");
1696 rv = 1;
1697 goto size_change_error;
1698 }
1699 if (min_csize && s->size == MAX_SIZE) {
1700 /* Don't let the kernel choose a size - it will get
1701 * it wrong
1702 */
1703 pr_err("Limited v0.90 array to "
1704 "2TB per device\n");
1705 s->size = min_csize;
1706 }
1707 if (st->ss->external) {
1708 if (sra->array.level == 0) {
1709 rv = sysfs_set_str(sra, NULL, "level",
1710 "raid5");
1711 if (!rv) {
1712 raid0_takeover = 1;
1713 /* get array parametes after takeover
1714 * to chane one parameter at time only
1715 */
1716 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1717 }
1718 }
1719 /* make sure mdmon is
1720 * aware of the new level */
1721 if (!mdmon_running(st->container_dev))
1722 start_mdmon(st->container_dev);
1723 ping_monitor(container);
1724 if (mdmon_running(st->container_dev) &&
1725 st->update_tail == NULL)
1726 st->update_tail = &st->updates;
1727 }
1728
1729 if (s->size == MAX_SIZE)
1730 s->size = 0;
1731 array.size = s->size;
1732 if ((unsigned)array.size != s->size) {
1733 /* got truncated to 32bit, write to
1734 * component_size instead
1735 */
1736 if (sra)
1737 rv = sysfs_set_num(sra, NULL,
1738 "component_size", s->size);
1739 else
1740 rv = -1;
1741 } else {
1742 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1743
1744 /* manage array size when it is managed externally
1745 */
1746 if ((rv == 0) && st->ss->external)
1747 rv = set_array_size(st, sra, sra->text_version);
1748 }
1749
1750 if (raid0_takeover) {
1751 /* do not recync non-existing parity,
1752 * we will drop it anyway
1753 */
1754 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1755 /* go back to raid0, drop parity disk
1756 */
1757 sysfs_set_str(sra, NULL, "level", "raid0");
1758 ioctl(fd, GET_ARRAY_INFO, &array);
1759 }
1760
1761 size_change_error:
1762 if (rv != 0) {
1763 int err = errno;
1764
1765 /* restore metadata */
1766 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1767 UnSet, NULL, devname,
1768 ROLLBACK_METADATA_CHANGES,
1769 c->verbose) == 0)
1770 sync_metadata(st);
1771 pr_err("Cannot set device size for %s: %s\n",
1772 devname, strerror(err));
1773 if (err == EBUSY &&
1774 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1775 cont_err("Bitmap must be removed before size can be changed\n");
1776 rv = 1;
1777 goto release;
1778 }
1779 if (s->assume_clean) {
1780 /* This will fail on kernels newer than 3.0 unless
1781 * a backport has been arranged.
1782 */
1783 if (sra == NULL ||
1784 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1785 pr_err("--assume-clean not support with --grow on this kernel\n");
1786 }
1787 ioctl(fd, GET_ARRAY_INFO, &array);
1788 s->size = get_component_size(fd)/2;
1789 if (s->size == 0)
1790 s->size = array.size;
1791 if (c->verbose >= 0) {
1792 if (s->size == orig_size)
1793 pr_err("component size of %s "
1794 "unchanged at %lluK\n",
1795 devname, s->size);
1796 else
1797 pr_err("component size of %s "
1798 "has been set to %lluK\n",
1799 devname, s->size);
1800 }
1801 changed = 1;
1802 } else if (array.level != LEVEL_CONTAINER) {
1803 s->size = get_component_size(fd)/2;
1804 if (s->size == 0)
1805 s->size = array.size;
1806 }
1807
1808 /* See if there is anything else to do */
1809 if ((s->level == UnSet || s->level == array.level) &&
1810 (s->layout_str == NULL) &&
1811 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1812 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1813 /* Nothing more to do */
1814 if (!changed && c->verbose >= 0)
1815 pr_err("%s: no change requested\n",
1816 devname);
1817 goto release;
1818 }
1819
1820 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1821 * current implementation assumes that following conditions must be met:
1822 * - RAID10:
1823 * - far_copies == 1
1824 * - near_copies == 2
1825 */
1826 if ((s->level == 0 && array.level == 10 && sra &&
1827 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1828 (s->level == 0 && array.level == 1 && sra)) {
1829 int err;
1830 err = remove_disks_for_takeover(st, sra, array.layout);
1831 if (err) {
1832 dprintf(Name": Array cannot be reshaped\n");
1833 if (cfd > -1)
1834 close(cfd);
1835 rv = 1;
1836 goto release;
1837 }
1838 /* Make sure mdmon has seen the device removal
1839 * and updated metadata before we continue with
1840 * level change
1841 */
1842 if (container)
1843 ping_monitor(container);
1844 }
1845
1846 memset(&info, 0, sizeof(info));
1847 info.array = array;
1848 sysfs_init(&info, fd, NoMdDev);
1849 strcpy(info.text_version, sra->text_version);
1850 info.component_size = s->size*2;
1851 info.new_level = s->level;
1852 info.new_chunk = s->chunk * 1024;
1853 if (info.array.level == LEVEL_CONTAINER) {
1854 info.delta_disks = UnSet;
1855 info.array.raid_disks = s->raiddisks;
1856 } else if (s->raiddisks)
1857 info.delta_disks = s->raiddisks - info.array.raid_disks;
1858 else
1859 info.delta_disks = UnSet;
1860 if (s->layout_str == NULL) {
1861 info.new_layout = UnSet;
1862 if (info.array.level == 6 &&
1863 (info.new_level == 6 || info.new_level == UnSet) &&
1864 info.array.layout >= 16) {
1865 pr_err("%s has a non-standard layout. If you"
1866 " wish to preserve this\n", devname);
1867 cont_err("during the reshape, please specify"
1868 " --layout=preserve\n");
1869 cont_err("If you want to change it, specify a"
1870 " layout or use --layout=normalise\n");
1871 rv = 1;
1872 goto release;
1873 }
1874 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1875 strcmp(s->layout_str, "normalize") == 0) {
1876 /* If we have a -6 RAID6 layout, remove the '-6'. */
1877 info.new_layout = UnSet;
1878 if (info.array.level == 6 && info.new_level == UnSet) {
1879 char l[40], *h;
1880 strcpy(l, map_num(r6layout, info.array.layout));
1881 h = strrchr(l, '-');
1882 if (h && strcmp(h, "-6") == 0) {
1883 *h = 0;
1884 info.new_layout = map_name(r6layout, l);
1885 }
1886 } else {
1887 pr_err("%s is only meaningful when reshaping"
1888 " a RAID6 array.\n", s->layout_str);
1889 rv = 1;
1890 goto release;
1891 }
1892 } else if (strcmp(s->layout_str, "preserve") == 0) {
1893 /* This means that a non-standard RAID6 layout
1894 * is OK.
1895 * In particular:
1896 * - When reshape a RAID6 (e.g. adding a device)
1897 * which is in a non-standard layout, it is OK
1898 * to preserve that layout.
1899 * - When converting a RAID5 to RAID6, leave it in
1900 * the XXX-6 layout, don't re-layout.
1901 */
1902 if (info.array.level == 6 && info.new_level == UnSet)
1903 info.new_layout = info.array.layout;
1904 else if (info.array.level == 5 && info.new_level == 6) {
1905 char l[40];
1906 strcpy(l, map_num(r5layout, info.array.layout));
1907 strcat(l, "-6");
1908 info.new_layout = map_name(r6layout, l);
1909 } else {
1910 pr_err("%s in only meaningful when reshaping"
1911 " to RAID6\n", s->layout_str);
1912 rv = 1;
1913 goto release;
1914 }
1915 } else {
1916 int l = info.new_level;
1917 if (l == UnSet)
1918 l = info.array.level;
1919 switch (l) {
1920 case 5:
1921 info.new_layout = map_name(r5layout, s->layout_str);
1922 break;
1923 case 6:
1924 info.new_layout = map_name(r6layout, s->layout_str);
1925 break;
1926 case 10:
1927 info.new_layout = parse_layout_10(s->layout_str);
1928 break;
1929 case LEVEL_FAULTY:
1930 info.new_layout = parse_layout_faulty(s->layout_str);
1931 break;
1932 default:
1933 pr_err("layout not meaningful"
1934 " with this level\n");
1935 rv = 1;
1936 goto release;
1937 }
1938 if (info.new_layout == UnSet) {
1939 pr_err("layout %s not understood"
1940 " for this level\n",
1941 s->layout_str);
1942 rv = 1;
1943 goto release;
1944 }
1945 }
1946
1947 if (array.level == LEVEL_FAULTY) {
1948 if (s->level != UnSet && s->level != array.level) {
1949 pr_err("cannot change level of Faulty device\n");
1950 rv =1 ;
1951 }
1952 if (s->chunk) {
1953 pr_err("cannot set chunksize of Faulty device\n");
1954 rv =1 ;
1955 }
1956 if (s->raiddisks && s->raiddisks != 1) {
1957 pr_err("cannot set raid_disks of Faulty device\n");
1958 rv =1 ;
1959 }
1960 if (s->layout_str) {
1961 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
1962 dprintf("Cannot get array information.\n");
1963 goto release;
1964 }
1965 array.layout = info.new_layout;
1966 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1967 pr_err("failed to set new layout\n");
1968 rv = 1;
1969 } else if (c->verbose >= 0)
1970 printf("layout for %s set to %d\n",
1971 devname, array.layout);
1972 }
1973 } else if (array.level == LEVEL_CONTAINER) {
1974 /* This change is to be applied to every array in the
1975 * container. This is only needed when the metadata imposes
1976 * restraints of the various arrays in the container.
1977 * Currently we only know that IMSM requires all arrays
1978 * to have the same number of devices so changing the
1979 * number of devices (On-Line Capacity Expansion) must be
1980 * performed at the level of the container
1981 */
1982 rv = reshape_container(container, devname, -1, st, &info,
1983 c->force, c->backup_file, c->verbose, 0, 0);
1984 frozen = 0;
1985 } else {
1986 /* get spare devices from external metadata
1987 */
1988 if (st->ss->external) {
1989 struct mdinfo *info2;
1990
1991 info2 = st->ss->container_content(st, subarray);
1992 if (info2) {
1993 info.array.spare_disks =
1994 info2->array.spare_disks;
1995 sysfs_free(info2);
1996 }
1997 }
1998
1999 /* Impose these changes on a single array. First
2000 * check that the metadata is OK with the change. */
2001
2002 if (reshape_super(st, 0, info.new_level,
2003 info.new_layout, info.new_chunk,
2004 info.array.raid_disks, info.delta_disks,
2005 c->backup_file, devname, APPLY_METADATA_CHANGES,
2006 c->verbose)) {
2007 rv = 1;
2008 goto release;
2009 }
2010 sync_metadata(st);
2011 rv = reshape_array(container, fd, devname, st, &info, c->force,
2012 devlist, c->backup_file, c->verbose, 0, 0, 0);
2013 frozen = 0;
2014 }
2015 release:
2016 sysfs_free(sra);
2017 if (frozen > 0)
2018 unfreeze(st);
2019 return rv;
2020 }
2021
2022 /* verify_reshape_position()
2023 * Function checks if reshape position in metadata is not farther
2024 * than position in md.
2025 * Return value:
2026 * 0 : not valid sysfs entry
2027 * it can be caused by not started reshape, it should be started
2028 * by reshape array or raid0 array is before takeover
2029 * -1 : error, reshape position is obviously wrong
2030 * 1 : success, reshape progress correct or updated
2031 */
2032 static int verify_reshape_position(struct mdinfo *info, int level)
2033 {
2034 int ret_val = 0;
2035 char buf[40];
2036 int rv;
2037
2038 /* read sync_max, failure can mean raid0 array */
2039 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2040
2041 if (rv > 0) {
2042 char *ep;
2043 unsigned long long position = strtoull(buf, &ep, 0);
2044
2045 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2046 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2047 position *= get_data_disks(level,
2048 info->new_layout,
2049 info->array.raid_disks);
2050 if (info->reshape_progress < position) {
2051 dprintf("Corrected reshape progress (%llu) to "
2052 "md position (%llu)\n",
2053 info->reshape_progress, position);
2054 info->reshape_progress = position;
2055 ret_val = 1;
2056 } else if (info->reshape_progress > position) {
2057 pr_err("Fatal error: array "
2058 "reshape was not properly frozen "
2059 "(expected reshape position is %llu, "
2060 "but reshape progress is %llu.\n",
2061 position, info->reshape_progress);
2062 ret_val = -1;
2063 } else {
2064 dprintf("Reshape position in md and metadata "
2065 "are the same;");
2066 ret_val = 1;
2067 }
2068 }
2069 } else if (rv == 0) {
2070 /* for valid sysfs entry, 0-length content
2071 * should be indicated as error
2072 */
2073 ret_val = -1;
2074 }
2075
2076 return ret_val;
2077 }
2078
2079 static int reshape_array(char *container, int fd, char *devname,
2080 struct supertype *st, struct mdinfo *info,
2081 int force, struct mddev_dev *devlist,
2082 char *backup_file, int verbose, int forked,
2083 int restart, int freeze_reshape)
2084 {
2085 struct reshape reshape;
2086 int spares_needed;
2087 char *msg;
2088 int orig_level = UnSet;
2089 int disks, odisks;
2090 int delayed;
2091
2092 struct mdu_array_info_s array;
2093 char *c;
2094
2095 struct mddev_dev *dv;
2096 int added_disks;
2097
2098 int *fdlist = NULL;
2099 unsigned long long *offsets = NULL;
2100 int d;
2101 int nrdisks;
2102 int err;
2103 unsigned long blocks;
2104 unsigned long cache;
2105 unsigned long long array_size;
2106 int done;
2107 struct mdinfo *sra = NULL;
2108
2109 /* when reshaping a RAID0, the component_size might be zero.
2110 * So try to fix that up.
2111 */
2112 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2113 dprintf("Cannot get array information.\n");
2114 goto release;
2115 }
2116 if (array.level == 0 && info->component_size == 0) {
2117 get_dev_size(fd, NULL, &array_size);
2118 info->component_size = array_size / array.raid_disks;
2119 }
2120
2121 if (info->reshape_active) {
2122 int new_level = info->new_level;
2123 info->new_level = UnSet;
2124 if (info->delta_disks > 0)
2125 info->array.raid_disks -= info->delta_disks;
2126 msg = analyse_change(info, &reshape);
2127 info->new_level = new_level;
2128 if (info->delta_disks > 0)
2129 info->array.raid_disks += info->delta_disks;
2130 if (!restart)
2131 /* Make sure the array isn't read-only */
2132 ioctl(fd, RESTART_ARRAY_RW, 0);
2133 } else
2134 msg = analyse_change(info, &reshape);
2135 if (msg) {
2136 pr_err("%s\n", msg);
2137 goto release;
2138 }
2139 if (restart &&
2140 (reshape.level != info->array.level ||
2141 reshape.before.layout != info->array.layout ||
2142 reshape.before.data_disks + reshape.parity
2143 != info->array.raid_disks - max(0, info->delta_disks))) {
2144 pr_err("reshape info is not in native format -"
2145 " cannot continue.\n");
2146 goto release;
2147 }
2148
2149 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2150 /* When reshape is restarted from '0', very begin of array
2151 * it is possible that for external metadata reshape and array
2152 * configuration doesn't happen.
2153 * Check if md has the same opinion, and reshape is restarted
2154 * from 0. If so, this is regular reshape start after reshape
2155 * switch in metadata to next array only.
2156 */
2157 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2158 (info->reshape_progress == 0))
2159 restart = 0;
2160 }
2161 if (restart) {
2162 /* reshape already started. just skip to monitoring the reshape */
2163 if (reshape.backup_blocks == 0)
2164 return 0;
2165 goto started;
2166 }
2167 /* The container is frozen but the array may not be.
2168 * So freeze the array so spares don't get put to the wrong use
2169 * FIXME there should probably be a cleaner separation between
2170 * freeze_array and freeze_container.
2171 */
2172 sysfs_freeze_array(info);
2173 /* Check we have enough spares to not be degraded */
2174 added_disks = 0;
2175 for (dv = devlist; dv ; dv=dv->next)
2176 added_disks++;
2177 spares_needed = max(reshape.before.data_disks,
2178 reshape.after.data_disks)
2179 + reshape.parity - array.raid_disks;
2180
2181 if (!force &&
2182 info->new_level > 1 && info->array.level > 1 &&
2183 spares_needed > info->array.spare_disks + added_disks) {
2184 pr_err("Need %d spare%s to avoid degraded array,"
2185 " and only have %d.\n"
2186 " Use --force to over-ride this check.\n",
2187 spares_needed,
2188 spares_needed == 1 ? "" : "s",
2189 info->array.spare_disks + added_disks);
2190 goto release;
2191 }
2192 /* Check we have enough spares to not fail */
2193 spares_needed = max(reshape.before.data_disks,
2194 reshape.after.data_disks)
2195 - array.raid_disks;
2196 if ((info->new_level > 1 || info->new_level == 0) &&
2197 spares_needed > info->array.spare_disks +added_disks) {
2198 pr_err("Need %d spare%s to create working array,"
2199 " and only have %d.\n",
2200 spares_needed,
2201 spares_needed == 1 ? "" : "s",
2202 info->array.spare_disks + added_disks);
2203 goto release;
2204 }
2205
2206 if (reshape.level != array.level) {
2207 char *c = map_num(pers, reshape.level);
2208 int err;
2209 if (c == NULL)
2210 goto release;
2211
2212 err = sysfs_set_str(info, NULL, "level", c);
2213 if (err) {
2214 err = errno;
2215 pr_err("%s: could not set level to %s\n",
2216 devname, c);
2217 if (err == EBUSY &&
2218 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2219 cont_err("Bitmap must be removed"
2220 " before level can be changed\n");
2221 goto release;
2222 }
2223 if (verbose >= 0)
2224 pr_err("level of %s changed to %s\n",
2225 devname, c);
2226 orig_level = array.level;
2227 sysfs_freeze_array(info);
2228
2229 if (reshape.level > 0 && st->ss->external) {
2230 /* make sure mdmon is aware of the new level */
2231 if (mdmon_running(st->container_dev))
2232 flush_mdmon(container);
2233
2234 if (!mdmon_running(st->container_dev))
2235 start_mdmon(st->container_dev);
2236 ping_monitor(container);
2237 if (mdmon_running(st->container_dev) &&
2238 st->update_tail == NULL)
2239 st->update_tail = &st->updates;
2240 }
2241 }
2242 /* ->reshape_super might have chosen some spares from the
2243 * container that it wants to be part of the new array.
2244 * We can collect them with ->container_content and give
2245 * them to the kernel.
2246 */
2247 if (st->ss->reshape_super && st->ss->container_content) {
2248 char *subarray = strchr(info->text_version+1, '/')+1;
2249 struct mdinfo *info2 =
2250 st->ss->container_content(st, subarray);
2251 struct mdinfo *d;
2252
2253 if (info2) {
2254 sysfs_init(info2, fd, st->devnum);
2255 /* When increasing number of devices, we need to set
2256 * new raid_disks before adding these, or they might
2257 * be rejected.
2258 */
2259 if (reshape.backup_blocks &&
2260 reshape.after.data_disks > reshape.before.data_disks)
2261 subarray_set_num(container, info2, "raid_disks",
2262 reshape.after.data_disks +
2263 reshape.parity);
2264 for (d = info2->devs; d; d = d->next) {
2265 if (d->disk.state == 0 &&
2266 d->disk.raid_disk >= 0) {
2267 /* This is a spare that wants to
2268 * be part of the array.
2269 */
2270 add_disk(fd, st, info2, d);
2271 }
2272 }
2273 sysfs_free(info2);
2274 }
2275 }
2276 /* We might have been given some devices to add to the
2277 * array. Now that the array has been changed to the right
2278 * level and frozen, we can safely add them.
2279 */
2280 if (devlist)
2281 Manage_subdevs(devname, fd, devlist, verbose,
2282 0,NULL, 0);
2283
2284 if (reshape.backup_blocks == 0) {
2285 /* No restriping needed, but we might need to impose
2286 * some more changes: layout, raid_disks, chunk_size
2287 */
2288 /* read current array info */
2289 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2290 dprintf("Cannot get array information.\n");
2291 goto release;
2292 }
2293 /* compare current array info with new values and if
2294 * it is different update them to new */
2295 if (info->new_layout != UnSet &&
2296 info->new_layout != array.layout) {
2297 array.layout = info->new_layout;
2298 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2299 pr_err("failed to set new layout\n");
2300 goto release;
2301 } else if (verbose >= 0)
2302 printf("layout for %s set to %d\n",
2303 devname, array.layout);
2304 }
2305 if (info->delta_disks != UnSet &&
2306 info->delta_disks != 0 &&
2307 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
2308 array.raid_disks += info->delta_disks;
2309 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2310 pr_err("failed to set raid disks\n");
2311 goto release;
2312 } else if (verbose >= 0) {
2313 printf("raid_disks for %s set to %d\n",
2314 devname, array.raid_disks);
2315 }
2316 }
2317 if (info->new_chunk != 0 &&
2318 info->new_chunk != array.chunk_size) {
2319 if (sysfs_set_num(info, NULL,
2320 "chunk_size", info->new_chunk) != 0) {
2321 pr_err("failed to set chunk size\n");
2322 goto release;
2323 } else if (verbose >= 0)
2324 printf("chunk size for %s set to %d\n",
2325 devname, array.chunk_size);
2326 }
2327 unfreeze(st);
2328 return 0;
2329 }
2330
2331 /*
2332 * There are three possibilities.
2333 * 1/ The array will shrink.
2334 * We need to ensure the reshape will pause before reaching
2335 * the 'critical section'. We also need to fork and wait for
2336 * that to happen. When it does we
2337 * suspend/backup/complete/unfreeze
2338 *
2339 * 2/ The array will not change size.
2340 * This requires that we keep a backup of a sliding window
2341 * so that we can restore data after a crash. So we need
2342 * to fork and monitor progress.
2343 * In future we will allow the data_offset to change, so
2344 * a sliding backup becomes unnecessary.
2345 *
2346 * 3/ The array will grow. This is relatively easy.
2347 * However the kernel's restripe routines will cheerfully
2348 * overwrite some early data before it is safe. So we
2349 * need to make a backup of the early parts of the array
2350 * and be ready to restore it if rebuild aborts very early.
2351 * For externally managed metadata, we still need a forked
2352 * child to monitor the reshape and suspend IO over the region
2353 * that is being reshaped.
2354 *
2355 * We backup data by writing it to one spare, or to a
2356 * file which was given on command line.
2357 *
2358 * In each case, we first make sure that storage is available
2359 * for the required backup.
2360 * Then we:
2361 * - request the shape change.
2362 * - fork to handle backup etc.
2363 */
2364 started:
2365 /* Check that we can hold all the data */
2366 get_dev_size(fd, NULL, &array_size);
2367 if (reshape.new_size < (array_size/512)) {
2368 pr_err("this change will reduce the size of the array.\n"
2369 " use --grow --array-size first to truncate array.\n"
2370 " e.g. mdadm --grow %s --array-size %llu\n",
2371 devname, reshape.new_size/2);
2372 goto release;
2373 }
2374
2375 sra = sysfs_read(fd, 0,
2376 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2377 GET_CACHE);
2378 if (!sra) {
2379 pr_err("%s: Cannot get array details from sysfs\n",
2380 devname);
2381 goto release;
2382 }
2383
2384 /* Decide how many blocks (sectors) for a reshape
2385 * unit. The number we have so far is just a minimum
2386 */
2387 blocks = reshape.backup_blocks;
2388 if (reshape.before.data_disks ==
2389 reshape.after.data_disks) {
2390 /* Make 'blocks' bigger for better throughput, but
2391 * not so big that we reject it below.
2392 * Try for 16 megabytes
2393 */
2394 while (blocks * 32 < sra->component_size &&
2395 blocks < 16*1024*2)
2396 blocks *= 2;
2397 } else
2398 pr_err("Need to backup %luK of critical "
2399 "section..\n", blocks/2);
2400
2401 if (blocks >= sra->component_size/2) {
2402 pr_err("%s: Something wrong"
2403 " - reshape aborted\n",
2404 devname);
2405 goto release;
2406 }
2407
2408 /* Now we need to open all these devices so we can read/write.
2409 */
2410 nrdisks = max(reshape.before.data_disks,
2411 reshape.after.data_disks) + reshape.parity
2412 + sra->array.spare_disks;
2413 fdlist = xcalloc((1+nrdisks), sizeof(int));
2414 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
2415
2416 odisks = reshape.before.data_disks + reshape.parity;
2417 d = reshape_prepare_fdlist(devname, sra, odisks,
2418 nrdisks, blocks, backup_file,
2419 fdlist, offsets);
2420 if (d < 0) {
2421 goto release;
2422 }
2423 if ((st->ss->manage_reshape == NULL) ||
2424 (st->ss->recover_backup == NULL)) {
2425 if (backup_file == NULL) {
2426 if (reshape.after.data_disks <=
2427 reshape.before.data_disks) {
2428 pr_err("%s: Cannot grow - "
2429 "need backup-file\n", devname);
2430 goto release;
2431 } else if (sra->array.spare_disks == 0) {
2432 pr_err("%s: Cannot grow - "
2433 "need a spare or backup-file to backup "
2434 "critical section\n", devname);
2435 goto release;
2436 }
2437 } else {
2438 if (!reshape_open_backup_file(backup_file, fd, devname,
2439 (signed)blocks,
2440 fdlist+d, offsets+d,
2441 restart)) {
2442 goto release;
2443 }
2444 d++;
2445 }
2446 }
2447
2448 /* lastly, check that the internal stripe cache is
2449 * large enough, or it won't work.
2450 * It must hold at least 4 stripes of the larger
2451 * chunk size
2452 */
2453 cache = max(info->array.chunk_size, info->new_chunk);
2454 cache *= 4; /* 4 stripes minimum */
2455 cache /= 512; /* convert to sectors */
2456 disks = min(reshape.before.data_disks, reshape.after.data_disks);
2457 /* make sure there is room for 'blocks' with a bit to spare */
2458 if (cache < 16 + blocks / disks)
2459 cache = 16 + blocks / disks;
2460 cache /= (4096/512); /* Covert from sectors to pages */
2461
2462 if (sra->cache_size < cache)
2463 subarray_set_num(container, sra, "stripe_cache_size",
2464 cache+1);
2465
2466 /* Right, everything seems fine. Let's kick things off.
2467 * If only changing raid_disks, use ioctl, else use
2468 * sysfs.
2469 */
2470 sync_metadata(st);
2471
2472 sra->new_chunk = info->new_chunk;
2473
2474 if (restart) {
2475 /* for external metadata checkpoint saved by mdmon can be lost
2476 * or missed /due to e.g. crash/. Check if md is not during
2477 * restart farther than metadata points to.
2478 * If so, this means metadata information is obsolete.
2479 */
2480 if (st->ss->external)
2481 verify_reshape_position(info, reshape.level);
2482 sra->reshape_progress = info->reshape_progress;
2483 } else {
2484 sra->reshape_progress = 0;
2485 if (reshape.after.data_disks < reshape.before.data_disks)
2486 /* start from the end of the new array */
2487 sra->reshape_progress = (sra->component_size
2488 * reshape.after.data_disks);
2489 }
2490
2491 if (info->array.chunk_size == info->new_chunk &&
2492 reshape.before.layout == reshape.after.layout &&
2493 st->ss->external == 0) {
2494 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2495 ioctl(fd, GET_ARRAY_INFO, &array);
2496 array.raid_disks = reshape.after.data_disks + reshape.parity;
2497 if (!restart &&
2498 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2499 int err = errno;
2500
2501 pr_err("Cannot set device shape for %s: %s\n",
2502 devname, strerror(errno));
2503
2504 if (err == EBUSY &&
2505 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2506 cont_err("Bitmap must be removed before"
2507 " shape can be changed\n");
2508
2509 goto release;
2510 }
2511 } else if (!restart) {
2512 /* set them all just in case some old 'new_*' value
2513 * persists from some earlier problem.
2514 */
2515 int err = 0;
2516 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2517 err = errno;
2518 if (!err && sysfs_set_num(sra, NULL, "layout",
2519 reshape.after.layout) < 0)
2520 err = errno;
2521 if (!err && subarray_set_num(container, sra, "raid_disks",
2522 reshape.after.data_disks +
2523 reshape.parity) < 0)
2524 err = errno;
2525 if (err) {
2526 pr_err("Cannot set device shape for %s\n",
2527 devname);
2528
2529 if (err == EBUSY &&
2530 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2531 cont_err("Bitmap must be removed before"
2532 " shape can be changed\n");
2533 goto release;
2534 }
2535 }
2536
2537 err = start_reshape(sra, restart, reshape.before.data_disks,
2538 reshape.after.data_disks);
2539 if (err) {
2540 pr_err("Cannot %s reshape for %s\n",
2541 restart ? "continue" : "start",
2542 devname);
2543 goto release;
2544 }
2545 if (restart)
2546 sysfs_set_str(sra, NULL, "array_state", "active");
2547 if (freeze_reshape) {
2548 free(fdlist);
2549 free(offsets);
2550 sysfs_free(sra);
2551 pr_err("Reshape has to be continued from"
2552 " location %llu when root filesystem has been mounted.\n",
2553 sra->reshape_progress);
2554 return 1;
2555 }
2556
2557 /* Now we just need to kick off the reshape and watch, while
2558 * handling backups of the data...
2559 * This is all done by a forked background process.
2560 */
2561 switch(forked ? 0 : fork()) {
2562 case -1:
2563 pr_err("Cannot run child to monitor reshape: %s\n",
2564 strerror(errno));
2565 abort_reshape(sra);
2566 goto release;
2567 default:
2568 free(fdlist);
2569 free(offsets);
2570 sysfs_free(sra);
2571 return 0;
2572 case 0:
2573 map_fork();
2574 break;
2575 }
2576
2577 /* If another array on the same devices is busy, the
2578 * reshape will wait for them. This would mean that
2579 * the first section that we suspend will stay suspended
2580 * for a long time. So check on that possibility
2581 * by looking for "DELAYED" in /proc/mdstat, and if found,
2582 * wait a while
2583 */
2584 do {
2585 struct mdstat_ent *mds, *m;
2586 delayed = 0;
2587 mds = mdstat_read(0, 0);
2588 for (m = mds; m; m = mds->next)
2589 if (m->devnum == devname2devnum(sra->sys_name)) {
2590 if (m->resync &&
2591 m->percent == RESYNC_DELAYED)
2592 delayed = 1;
2593 if (m->resync == 0)
2594 /* Haven't started the reshape thread
2595 * yet, wait a bit
2596 */
2597 delayed = 2;
2598 break;
2599 }
2600 free_mdstat(mds);
2601 if (delayed == 1 && get_linux_version() < 3007000) {
2602 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
2603 " You might experience problems until other reshapes complete.\n");
2604 delayed = 0;
2605 }
2606 if (delayed)
2607 sleep(30 - (delayed-1) * 25);
2608 } while (delayed);
2609
2610 close(fd);
2611 if (check_env("MDADM_GROW_VERIFY"))
2612 fd = open(devname, O_RDONLY | O_DIRECT);
2613 else
2614 fd = -1;
2615 mlockall(MCL_FUTURE);
2616
2617 if (st->ss->external) {
2618 /* metadata handler takes it from here */
2619 done = st->ss->manage_reshape(
2620 fd, sra, &reshape, st, blocks,
2621 fdlist, offsets,
2622 d - odisks, fdlist+odisks,
2623 offsets+odisks);
2624 } else
2625 done = child_monitor(
2626 fd, sra, &reshape, st, blocks,
2627 fdlist, offsets,
2628 d - odisks, fdlist+odisks,
2629 offsets+odisks);
2630
2631 free(fdlist);
2632 free(offsets);
2633
2634 if (backup_file && done)
2635 unlink(backup_file);
2636 if (!done) {
2637 abort_reshape(sra);
2638 goto out;
2639 }
2640
2641 if (!st->ss->external &&
2642 !(reshape.before.data_disks != reshape.after.data_disks
2643 && info->custom_array_size) &&
2644 info->new_level == reshape.level &&
2645 !forked) {
2646 /* no need to wait for the reshape to finish as
2647 * there is nothing more to do.
2648 */
2649 sysfs_free(sra);
2650 exit(0);
2651 }
2652 wait_reshape(sra);
2653
2654 if (st->ss->external) {
2655 /* Re-load the metadata as much could have changed */
2656 int cfd = open_dev(st->container_dev);
2657 if (cfd >= 0) {
2658 flush_mdmon(container);
2659 st->ss->free_super(st);
2660 st->ss->load_container(st, cfd, container);
2661 close(cfd);
2662 }
2663 }
2664
2665 /* set new array size if required customer_array_size is used
2666 * by this metadata.
2667 */
2668 if (reshape.before.data_disks !=
2669 reshape.after.data_disks &&
2670 info->custom_array_size)
2671 set_array_size(st, info, info->text_version);
2672
2673 if (info->new_level != reshape.level) {
2674
2675 c = map_num(pers, info->new_level);
2676 if (c) {
2677 err = sysfs_set_str(sra, NULL, "level", c);
2678 if (err)
2679 pr_err("%s: could not set level "
2680 "to %s\n", devname, c);
2681 }
2682 if (info->new_level == 0)
2683 st->update_tail = NULL;
2684 }
2685 out:
2686 sysfs_free(sra);
2687 if (forked)
2688 return 0;
2689 unfreeze(st);
2690 exit(0);
2691
2692 release:
2693 free(fdlist);
2694 free(offsets);
2695 if (orig_level != UnSet && sra) {
2696 c = map_num(pers, orig_level);
2697 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
2698 pr_err("aborting level change\n");
2699 }
2700 sysfs_free(sra);
2701 if (!forked)
2702 unfreeze(st);
2703 return 1;
2704 }
2705
2706 /* mdfd handle is passed to be closed in child process (after fork).
2707 */
2708 int reshape_container(char *container, char *devname,
2709 int mdfd,
2710 struct supertype *st,
2711 struct mdinfo *info,
2712 int force,
2713 char *backup_file,
2714 int verbose, int restart, int freeze_reshape)
2715 {
2716 struct mdinfo *cc = NULL;
2717 int rv = restart;
2718 int last_devnum = -1;
2719
2720 /* component_size is not meaningful for a container,
2721 * so pass '0' meaning 'no change'
2722 */
2723 if (!restart &&
2724 reshape_super(st, 0, info->new_level,
2725 info->new_layout, info->new_chunk,
2726 info->array.raid_disks, info->delta_disks,
2727 backup_file, devname, APPLY_METADATA_CHANGES,
2728 verbose)) {
2729 unfreeze(st);
2730 return 1;
2731 }
2732
2733 sync_metadata(st);
2734
2735 /* ping monitor to be sure that update is on disk
2736 */
2737 ping_monitor(container);
2738
2739 switch (fork()) {
2740 case -1: /* error */
2741 perror("Cannot fork to complete reshape\n");
2742 unfreeze(st);
2743 return 1;
2744 default: /* parent */
2745 if (!freeze_reshape)
2746 printf(Name ": multi-array reshape continues"
2747 " in background\n");
2748 return 0;
2749 case 0: /* child */
2750 map_fork();
2751 break;
2752 }
2753
2754 /* close unused handle in child process
2755 */
2756 if (mdfd > -1)
2757 close(mdfd);
2758
2759 while(1) {
2760 /* For each member array with reshape_active,
2761 * we need to perform the reshape.
2762 * We pick the first array that needs reshaping and
2763 * reshape it. reshape_array() will re-read the metadata
2764 * so the next time through a different array should be
2765 * ready for reshape.
2766 * It is possible that the 'different' array will not
2767 * be assembled yet. In that case we simple exit.
2768 * When it is assembled, the mdadm which assembles it
2769 * will take over the reshape.
2770 */
2771 struct mdinfo *content;
2772 int fd;
2773 struct mdstat_ent *mdstat;
2774 char *adev;
2775
2776 sysfs_free(cc);
2777
2778 cc = st->ss->container_content(st, NULL);
2779
2780 for (content = cc; content ; content = content->next) {
2781 char *subarray;
2782 if (!content->reshape_active)
2783 continue;
2784
2785 subarray = strchr(content->text_version+1, '/')+1;
2786 mdstat = mdstat_by_subdev(subarray,
2787 devname2devnum(container));
2788 if (!mdstat)
2789 continue;
2790 if (mdstat->active == 0) {
2791 pr_err("Skipping inactive "
2792 "array md%i.\n", mdstat->devnum);
2793 free_mdstat(mdstat);
2794 mdstat = NULL;
2795 continue;
2796 }
2797 break;
2798 }
2799 if (!content)
2800 break;
2801
2802 adev = map_dev(dev2major(mdstat->devnum),
2803 dev2minor(mdstat->devnum),
2804 0);
2805 if (!adev)
2806 adev = content->text_version;
2807
2808 fd = open_dev(mdstat->devnum);
2809 if (fd < 0) {
2810 printf(Name ": Device %s cannot be opened for reshape.",
2811 adev);
2812 break;
2813 }
2814
2815 if (last_devnum == mdstat->devnum) {
2816 /* Do not allow for multiple reshape_array() calls for
2817 * the same array.
2818 * It can happen when reshape_array() returns without
2819 * error, when reshape is not finished (wrong reshape
2820 * starting/continuation conditions). Mdmon doesn't
2821 * switch to next array in container and reentry
2822 * conditions for the same array occur.
2823 * This is possibly interim until the behaviour of
2824 * reshape_array is resolved().
2825 */
2826 printf(Name ": Multiple reshape execution detected for "
2827 "device %s.", adev);
2828 close(fd);
2829 break;
2830 }
2831 last_devnum = mdstat->devnum;
2832
2833 sysfs_init(content, fd, mdstat->devnum);
2834
2835 if (mdmon_running(devname2devnum(container)))
2836 flush_mdmon(container);
2837
2838 rv = reshape_array(container, fd, adev, st,
2839 content, force, NULL,
2840 backup_file, verbose, 1, restart,
2841 freeze_reshape);
2842 close(fd);
2843
2844 if (freeze_reshape) {
2845 sysfs_free(cc);
2846 exit(0);
2847 }
2848
2849 restart = 0;
2850 if (rv)
2851 break;
2852
2853 if (mdmon_running(devname2devnum(container)))
2854 flush_mdmon(container);
2855 }
2856 if (!rv)
2857 unfreeze(st);
2858 sysfs_free(cc);
2859 exit(0);
2860 }
2861
2862 /*
2863 * We run a child process in the background which performs the following
2864 * steps:
2865 * - wait for resync to reach a certain point
2866 * - suspend io to the following section
2867 * - backup that section
2868 * - allow resync to proceed further
2869 * - resume io
2870 * - discard the backup.
2871 *
2872 * When are combined in slightly different ways in the three cases.
2873 * Grow:
2874 * - suspend/backup/allow/wait/resume/discard
2875 * Shrink:
2876 * - allow/wait/suspend/backup/allow/wait/resume/discard
2877 * same-size:
2878 * - wait/resume/discard/suspend/backup/allow
2879 *
2880 * suspend/backup/allow always come together
2881 * wait/resume/discard do too.
2882 * For the same-size case we have two backups to improve flow.
2883 *
2884 */
2885
2886 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
2887 unsigned long long backup_point,
2888 unsigned long long wait_point,
2889 unsigned long long *suspend_point,
2890 unsigned long long *reshape_completed)
2891 {
2892 /* This function is called repeatedly by the reshape manager.
2893 * It determines how much progress can safely be made and allows
2894 * that progress.
2895 * - 'info' identifies the array and particularly records in
2896 * ->reshape_progress the metadata's knowledge of progress
2897 * This is a sector offset from the start of the array
2898 * of the next array block to be relocated. This number
2899 * may increase from 0 or decrease from array_size, depending
2900 * on the type of reshape that is happening.
2901 * Note that in contrast, 'sync_completed' is a block count of the
2902 * reshape so far. It gives the distance between the start point
2903 * (head or tail of device) and the next place that data will be
2904 * written. It always increases.
2905 * - 'reshape' is the structure created by analyse_change
2906 * - 'backup_point' shows how much the metadata manager has backed-up
2907 * data. For reshapes with increasing progress, it is the next address
2908 * to be backed up, previous addresses have been backed-up. For
2909 * decreasing progress, it is the earliest address that has been
2910 * backed up - later address are also backed up.
2911 * So addresses between reshape_progress and backup_point are
2912 * backed up providing those are in the 'correct' order.
2913 * - 'wait_point' is an array address. When reshape_completed
2914 * passes this point, progress_reshape should return. It might
2915 * return earlier if it determines that ->reshape_progress needs
2916 * to be updated or further backup is needed.
2917 * - suspend_point is maintained by progress_reshape and the caller
2918 * should not touch it except to initialise to zero.
2919 * It is an array address and it only increases in 2.6.37 and earlier.
2920 * This makes it difficult to handle reducing reshapes with
2921 * external metadata.
2922 * However: it is similar to backup_point in that it records the
2923 * other end of a suspended region from reshape_progress.
2924 * it is moved to extend the region that is safe to backup and/or
2925 * reshape
2926 * - reshape_completed is read from sysfs and returned. The caller
2927 * should copy this into ->reshape_progress when it has reason to
2928 * believe that the metadata knows this, and any backup outside this
2929 * has been erased.
2930 *
2931 * Return value is:
2932 * 1 if more data from backup_point - but only as far as suspend_point,
2933 * should be backed up
2934 * 0 if things are progressing smoothly
2935 * -1 if the reshape is finished because it is all done,
2936 * -2 if the reshape is finished due to an error.
2937 */
2938
2939 int advancing = (reshape->after.data_disks
2940 >= reshape->before.data_disks);
2941 unsigned long long need_backup; /* All data between start of array and
2942 * here will at some point need to
2943 * be backed up.
2944 */
2945 unsigned long long read_offset, write_offset;
2946 unsigned long long write_range;
2947 unsigned long long max_progress, target, completed;
2948 unsigned long long array_size = (info->component_size
2949 * reshape->before.data_disks);
2950 int fd;
2951 char buf[20];
2952
2953 /* First, we unsuspend any region that is now known to be safe.
2954 * If suspend_point is on the 'wrong' side of reshape_progress, then
2955 * we don't have or need suspension at the moment. This is true for
2956 * native metadata when we don't need to back-up.
2957 */
2958 if (advancing) {
2959 if (info->reshape_progress <= *suspend_point)
2960 sysfs_set_num(info, NULL, "suspend_lo",
2961 info->reshape_progress);
2962 } else {
2963 /* Note: this won't work in 2.6.37 and before.
2964 * Something somewhere should make sure we don't need it!
2965 */
2966 if (info->reshape_progress >= *suspend_point)
2967 sysfs_set_num(info, NULL, "suspend_hi",
2968 info->reshape_progress);
2969 }
2970
2971 /* Now work out how far it is safe to progress.
2972 * If the read_offset for ->reshape_progress is less than
2973 * 'blocks' beyond the write_offset, we can only progress as far
2974 * as a backup.
2975 * Otherwise we can progress until the write_offset for the new location
2976 * reaches (within 'blocks' of) the read_offset at the current location.
2977 * However that region must be suspended unless we are using native
2978 * metadata.
2979 * If we need to suspend more, we limit it to 128M per device, which is
2980 * rather arbitrary and should be some time-based calculation.
2981 */
2982 read_offset = info->reshape_progress / reshape->before.data_disks;
2983 write_offset = info->reshape_progress / reshape->after.data_disks;
2984 write_range = info->new_chunk/512;
2985 if (reshape->before.data_disks == reshape->after.data_disks)
2986 need_backup = array_size;
2987 else
2988 need_backup = reshape->backup_blocks;
2989 if (advancing) {
2990 if (read_offset < write_offset + write_range)
2991 max_progress = backup_point;
2992 else
2993 max_progress =
2994 read_offset *
2995 reshape->after.data_disks;
2996 } else {
2997 if (read_offset > write_offset - write_range)
2998 /* Can only progress as far as has been backed up,
2999 * which must be suspended */
3000 max_progress = backup_point;
3001 else if (info->reshape_progress <= need_backup)
3002 max_progress = backup_point;
3003 else {
3004 if (info->array.major_version >= 0)
3005 /* Can progress until backup is needed */
3006 max_progress = need_backup;
3007 else {
3008 /* Can progress until metadata update is required */
3009 max_progress =
3010 read_offset *
3011 reshape->after.data_disks;
3012 /* but data must be suspended */
3013 if (max_progress < *suspend_point)
3014 max_progress = *suspend_point;
3015 }
3016 }
3017 }
3018
3019 /* We know it is safe to progress to 'max_progress' providing
3020 * it is suspended or we are using native metadata.
3021 * Consider extending suspend_point 128M per device if it
3022 * is less than 64M per device beyond reshape_progress.
3023 * But always do a multiple of 'blocks'
3024 * FIXME this is too big - it takes to long to complete
3025 * this much.
3026 */
3027 target = 64*1024*2 * min(reshape->before.data_disks,
3028 reshape->after.data_disks);
3029 target /= reshape->backup_blocks;
3030 if (target < 2)
3031 target = 2;
3032 target *= reshape->backup_blocks;
3033
3034 /* For externally managed metadata we always need to suspend IO to
3035 * the area being reshaped so we regularly push suspend_point forward.
3036 * For native metadata we only need the suspend if we are going to do
3037 * a backup.
3038 */
3039 if (advancing) {
3040 if ((need_backup > info->reshape_progress
3041 || info->array.major_version < 0) &&
3042 *suspend_point < info->reshape_progress + target) {
3043 if (need_backup < *suspend_point + 2 * target)
3044 *suspend_point = need_backup;
3045 else if (*suspend_point + 2 * target < array_size)
3046 *suspend_point += 2 * target;
3047 else
3048 *suspend_point = array_size;
3049 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3050 if (max_progress > *suspend_point)
3051 max_progress = *suspend_point;
3052 }
3053 } else {
3054 if (info->array.major_version >= 0) {
3055 /* Only need to suspend when about to backup */
3056 if (info->reshape_progress < need_backup * 2 &&
3057 *suspend_point > 0) {
3058 *suspend_point = 0;
3059 sysfs_set_num(info, NULL, "suspend_lo", 0);
3060 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3061 }
3062 } else {
3063 /* Need to suspend continually */
3064 if (info->reshape_progress < *suspend_point)
3065 *suspend_point = info->reshape_progress;
3066 if (*suspend_point + target < info->reshape_progress)
3067 /* No need to move suspend region yet */;
3068 else {
3069 if (*suspend_point >= 2 * target)
3070 *suspend_point -= 2 * target;
3071 else
3072 *suspend_point = 0;
3073 sysfs_set_num(info, NULL, "suspend_lo",
3074 *suspend_point);
3075 }
3076 if (max_progress < *suspend_point)
3077 max_progress = *suspend_point;
3078 }
3079 }
3080
3081 /* now set sync_max to allow that progress. sync_max, like
3082 * sync_completed is a count of sectors written per device, so
3083 * we find the difference between max_progress and the start point,
3084 * and divide that by after.data_disks to get a sync_max
3085 * number.
3086 * At the same time we convert wait_point to a similar number
3087 * for comparing against sync_completed.
3088 */
3089 /* scale down max_progress to per_disk */
3090 max_progress /= reshape->after.data_disks;
3091 /* Round to chunk size as some kernels give an erroneously high number */
3092 max_progress /= info->new_chunk/512;
3093 max_progress *= info->new_chunk/512;
3094 /* And round to old chunk size as the kernel wants that */
3095 max_progress /= info->array.chunk_size/512;
3096 max_progress *= info->array.chunk_size/512;
3097 /* Limit progress to the whole device */
3098 if (max_progress > info->component_size)
3099 max_progress = info->component_size;
3100 wait_point /= reshape->after.data_disks;
3101 if (!advancing) {
3102 /* switch from 'device offset' to 'processed block count' */
3103 max_progress = info->component_size - max_progress;
3104 wait_point = info->component_size - wait_point;
3105 }
3106
3107 sysfs_set_num(info, NULL, "sync_max", max_progress);
3108
3109 /* Now wait. If we have already reached the point that we were
3110 * asked to wait to, don't wait at all, else wait for any change.
3111 * We need to select on 'sync_completed' as that is the place that
3112 * notifications happen, but we are really interested in
3113 * 'reshape_position'
3114 */
3115 fd = sysfs_get_fd(info, NULL, "sync_completed");
3116 if (fd < 0)
3117 goto check_progress;
3118
3119 if (sysfs_fd_get_ll(fd, &completed) < 0)
3120 goto check_progress;
3121
3122 while (completed < max_progress && completed < wait_point) {
3123 /* Check that sync_action is still 'reshape' to avoid
3124 * waiting forever on a dead array
3125 */
3126 char action[20];
3127 fd_set rfds;
3128 if (sysfs_get_str(info, NULL, "sync_action",
3129 action, 20) <= 0 ||
3130 strncmp(action, "reshape", 7) != 0)
3131 break;
3132 /* Some kernels reset 'sync_completed' to zero
3133 * before setting 'sync_action' to 'idle'.
3134 * So we need these extra tests.
3135 */
3136 if (completed == 0 && advancing
3137 && info->reshape_progress > 0)
3138 break;
3139 if (completed == 0 && !advancing
3140 && info->reshape_progress < (info->component_size
3141 * reshape->after.data_disks))
3142 break;
3143 FD_ZERO(&rfds);
3144 FD_SET(fd, &rfds);
3145 select(fd+1, NULL, NULL, &rfds, NULL);
3146 if (sysfs_fd_get_ll(fd, &completed) < 0)
3147 goto check_progress;
3148 }
3149 /* Some kernels reset 'sync_completed' to zero,
3150 * we need to have real point we are in md
3151 */
3152 if (completed == 0)
3153 completed = max_progress;
3154
3155 /* some kernels can give an incorrectly high 'completed' number */
3156 completed /= (info->new_chunk/512);
3157 completed *= (info->new_chunk/512);
3158 /* Convert 'completed' back in to a 'progress' number */
3159 completed *= reshape->after.data_disks;
3160 if (!advancing) {
3161 completed = info->component_size * reshape->after.data_disks
3162 - completed;
3163 }
3164 *reshape_completed = completed;
3165
3166 close(fd);
3167
3168 /* We return the need_backup flag. Caller will decide
3169 * how much - a multiple of ->backup_blocks up to *suspend_point
3170 */
3171 if (advancing)
3172 return need_backup > info->reshape_progress;
3173 else
3174 return need_backup >= info->reshape_progress;
3175
3176 check_progress:
3177 /* if we couldn't read a number from sync_completed, then
3178 * either the reshape did complete, or it aborted.
3179 * We can tell which by checking for 'none' in reshape_position.
3180 * If it did abort, then it might immediately restart if it
3181 * it was just a device failure that leaves us degraded but
3182 * functioning.
3183 */
3184 strcpy(buf, "hi");
3185 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3186 || strncmp(buf, "none", 4) != 0) {
3187 /* The abort might only be temporary. Wait up to 10
3188 * seconds for fd to contain a valid number again.
3189 */
3190 struct timeval tv;
3191 int rv = -2;
3192 tv.tv_sec = 10;
3193 tv.tv_usec = 0;
3194 while (fd >= 0 && rv < 0 && tv.tv_sec > 0) {
3195 fd_set rfds;
3196 FD_ZERO(&rfds);
3197 FD_SET(fd, &rfds);
3198 if (select(fd+1, NULL, NULL, &rfds, &tv) != 1)
3199 break;
3200 switch (sysfs_fd_get_ll(fd, &completed)) {
3201 case 0:
3202 /* all good again */
3203 rv = 1;
3204 break;
3205 case -2: /* read error - abort */
3206 tv.tv_sec = 0;
3207 break;
3208 }
3209 }
3210 if (fd >= 0)
3211 close(fd);
3212 return rv; /* abort */
3213 } else {
3214 /* Maybe racing with array shutdown - check state */
3215 if (fd >= 0)
3216 close(fd);
3217 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3218 || strncmp(buf, "inactive", 8) == 0
3219 || strncmp(buf, "clear",5) == 0)
3220 return -2; /* abort */
3221 return -1; /* complete */
3222 }
3223 }
3224
3225 /* FIXME return status is never checked */
3226 static int grow_backup(struct mdinfo *sra,
3227 unsigned long long offset, /* per device */
3228 unsigned long stripes, /* per device, in old chunks */
3229 int *sources, unsigned long long *offsets,
3230 int disks, int chunk, int level, int layout,
3231 int dests, int *destfd, unsigned long long *destoffsets,
3232 int part, int *degraded,
3233 char *buf)
3234 {
3235 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3236 * to storage 'destfd' (offset 'destoffsets'), after first
3237 * suspending IO. Then allow resync to continue
3238 * over the suspended section.
3239 * Use part 'part' of the backup-super-block.
3240 */
3241 int odata = disks;
3242 int rv = 0;
3243 int i;
3244 unsigned long long ll;
3245 int new_degraded;
3246 //printf("offset %llu\n", offset);
3247 if (level >= 4)
3248 odata--;
3249 if (level == 6)
3250 odata--;
3251
3252 /* Check that array hasn't become degraded, else we might backup the wrong data */
3253 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3254 return -1; /* FIXME this error is ignored */
3255 new_degraded = (int)ll;
3256 if (new_degraded != *degraded) {
3257 /* check each device to ensure it is still working */
3258 struct mdinfo *sd;
3259 for (sd = sra->devs ; sd ; sd = sd->next) {
3260 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3261 continue;
3262 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3263 char sbuf[20];
3264 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3265 strstr(sbuf, "faulty") ||
3266 strstr(sbuf, "in_sync") == NULL) {
3267 /* this device is dead */
3268 sd->disk.state = (1<<MD_DISK_FAULTY);
3269 if (sd->disk.raid_disk >= 0 &&
3270 sources[sd->disk.raid_disk] >= 0) {
3271 close(sources[sd->disk.raid_disk]);
3272 sources[sd->disk.raid_disk] = -1;
3273 }
3274 }
3275 }
3276 }
3277 *degraded = new_degraded;
3278 }
3279 if (part) {
3280 bsb.arraystart2 = __cpu_to_le64(offset * odata);
3281 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
3282 } else {
3283 bsb.arraystart = __cpu_to_le64(offset * odata);
3284 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
3285 }
3286 if (part)
3287 bsb.magic[15] = '2';
3288 for (i = 0; i < dests; i++)
3289 if (part)
3290 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
3291 else
3292 lseek64(destfd[i], destoffsets[i], 0);
3293
3294 rv = save_stripes(sources, offsets,
3295 disks, chunk, level, layout,
3296 dests, destfd,
3297 offset*512*odata, stripes * chunk * odata,
3298 buf);
3299
3300 if (rv)
3301 return rv;
3302 bsb.mtime = __cpu_to_le64(time(0));
3303 for (i = 0; i < dests; i++) {
3304 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3305
3306 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3307 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3308 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3309 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3310
3311 rv = -1;
3312 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
3313 != destoffsets[i] - 4096)
3314 break;
3315 if (write(destfd[i], &bsb, 512) != 512)
3316 break;
3317 if (destoffsets[i] > 4096) {
3318 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
3319 destoffsets[i]+stripes*chunk*odata)
3320 break;
3321 if (write(destfd[i], &bsb, 512) != 512)
3322 break;
3323 }
3324 fsync(destfd[i]);
3325 rv = 0;
3326 }
3327
3328 return rv;
3329 }
3330
3331 /* in 2.6.30, the value reported by sync_completed can be
3332 * less that it should be by one stripe.
3333 * This only happens when reshape hits sync_max and pauses.
3334 * So allow wait_backup to either extent sync_max further
3335 * than strictly necessary, or return before the
3336 * sync has got quite as far as we would really like.
3337 * This is what 'blocks2' is for.
3338 * The various caller give appropriate values so that
3339 * every works.
3340 */
3341 /* FIXME return value is often ignored */
3342 static int forget_backup(int dests, int *destfd,
3343 unsigned long long *destoffsets,
3344 int part)
3345 {
3346 /*
3347 * Erase backup 'part' (which is 0 or 1)
3348 */
3349 int i;
3350 int rv;
3351
3352 if (part) {
3353 bsb.arraystart2 = __cpu_to_le64(0);
3354 bsb.length2 = __cpu_to_le64(0);
3355 } else {
3356 bsb.arraystart = __cpu_to_le64(0);
3357 bsb.length = __cpu_to_le64(0);
3358 }
3359 bsb.mtime = __cpu_to_le64(time(0));
3360 rv = 0;
3361 for (i = 0; i < dests; i++) {
3362 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3363 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3364 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3365 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3366 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3367 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
3368 destoffsets[i]-4096)
3369 rv = -1;
3370 if (rv == 0 &&
3371 write(destfd[i], &bsb, 512) != 512)
3372 rv = -1;
3373 fsync(destfd[i]);
3374 }
3375 return rv;
3376 }
3377
3378 static void fail(char *msg)
3379 {
3380 int rv;
3381 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
3382 rv |= (write(2, "\n", 1) != 1);
3383 exit(rv ? 1 : 2);
3384 }
3385
3386 static char *abuf, *bbuf;
3387 static unsigned long long abuflen;
3388 static void validate(int afd, int bfd, unsigned long long offset)
3389 {
3390 /* check that the data in the backup against the array.
3391 * This is only used for regression testing and should not
3392 * be used while the array is active
3393 */
3394 if (afd < 0)
3395 return;
3396 lseek64(bfd, offset - 4096, 0);
3397 if (read(bfd, &bsb2, 512) != 512)
3398 fail("cannot read bsb");
3399 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
3400 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
3401 fail("first csum bad");
3402 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
3403 fail("magic is bad");
3404 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
3405 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
3406 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
3407 fail("second csum bad");
3408
3409 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
3410 fail("devstart is wrong");
3411
3412 if (bsb2.length) {
3413 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
3414
3415 if (abuflen < len) {
3416 free(abuf);
3417 free(bbuf);
3418 abuflen = len;
3419 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
3420 posix_memalign((void**)&bbuf, 4096, abuflen)) {
3421 abuflen = 0;
3422 /* just stop validating on mem-alloc failure */
3423 return;
3424 }
3425 }
3426
3427 lseek64(bfd, offset, 0);
3428 if ((unsigned long long)read(bfd, bbuf, len) != len) {
3429 //printf("len %llu\n", len);
3430 fail("read first backup failed");
3431 }
3432 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
3433 if ((unsigned long long)read(afd, abuf, len) != len)
3434 fail("read first from array failed");
3435 if (memcmp(bbuf, abuf, len) != 0) {
3436 #if 0
3437 int i;
3438 printf("offset=%llu len=%llu\n",
3439 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
3440 for (i=0; i<len; i++)
3441 if (bbuf[i] != abuf[i]) {
3442 printf("first diff byte %d\n", i);
3443 break;
3444 }
3445 #endif
3446 fail("data1 compare failed");
3447 }
3448 }
3449 if (bsb2.length2) {
3450 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
3451
3452 if (abuflen < len) {
3453 free(abuf);
3454 free(bbuf);
3455 abuflen = len;
3456 abuf = xmalloc(abuflen);
3457 bbuf = xmalloc(abuflen);
3458 }
3459
3460 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
3461 if ((unsigned long long)read(bfd, bbuf, len) != len)
3462 fail("read second backup failed");
3463 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
3464 if ((unsigned long long)read(afd, abuf, len) != len)
3465 fail("read second from array failed");
3466 if (memcmp(bbuf, abuf, len) != 0)
3467 fail("data2 compare failed");
3468 }
3469 }
3470
3471 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
3472 struct supertype *st, unsigned long blocks,
3473 int *fds, unsigned long long *offsets,
3474 int dests, int *destfd, unsigned long long *destoffsets)
3475 {
3476 /* Monitor a reshape where backup is being performed using
3477 * 'native' mechanism - either to a backup file, or
3478 * to some space in a spare.
3479 */
3480 char *buf;
3481 int degraded = -1;
3482 unsigned long long speed;
3483 unsigned long long suspend_point, array_size;
3484 unsigned long long backup_point, wait_point;
3485 unsigned long long reshape_completed;
3486 int done = 0;
3487 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
3488 int part = 0; /* The next part of the backup area to fill. It may already
3489 * be full, so we need to check */
3490 int level = reshape->level;
3491 int layout = reshape->before.layout;
3492 int data = reshape->before.data_disks;
3493 int disks = reshape->before.data_disks + reshape->parity;
3494 int chunk = sra->array.chunk_size;
3495 struct mdinfo *sd;
3496 unsigned long stripes;
3497 int uuid[4];
3498
3499 /* set up the backup-super-block. This requires the
3500 * uuid from the array.
3501 */
3502 /* Find a superblock */
3503 for (sd = sra->devs; sd; sd = sd->next) {
3504 char *dn;
3505 int devfd;
3506 int ok;
3507 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3508 continue;
3509 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
3510 devfd = dev_open(dn, O_RDONLY);
3511 if (devfd < 0)
3512 continue;
3513 ok = st->ss->load_super(st, devfd, NULL);
3514 close(devfd);
3515 if (ok == 0)
3516 break;
3517 }
3518 if (!sd) {
3519 pr_err("Cannot find a superblock\n");
3520 return 0;
3521 }
3522
3523 memset(&bsb, 0, 512);
3524 memcpy(bsb.magic, "md_backup_data-1", 16);
3525 st->ss->uuid_from_super(st, uuid);
3526 memcpy(bsb.set_uuid, uuid, 16);
3527 bsb.mtime = __cpu_to_le64(time(0));
3528 bsb.devstart2 = blocks;
3529
3530 stripes = blocks / (sra->array.chunk_size/512) /
3531 reshape->before.data_disks;
3532
3533 if (posix_memalign((void**)&buf, 4096, disks * chunk))
3534 /* Don't start the 'reshape' */
3535 return 0;
3536 if (reshape->before.data_disks == reshape->after.data_disks) {
3537 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
3538 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
3539 }
3540
3541 if (increasing) {
3542 array_size = sra->component_size * reshape->after.data_disks;
3543 backup_point = sra->reshape_progress;
3544 suspend_point = 0;
3545 } else {
3546 array_size = sra->component_size * reshape->before.data_disks;
3547 backup_point = reshape->backup_blocks;
3548 suspend_point = array_size;
3549 }
3550
3551 while (!done) {
3552 int rv;
3553
3554 /* Want to return as soon the oldest backup slot can
3555 * be released as that allows us to start backing up
3556 * some more, providing suspend_point has been
3557 * advanced, which it should have.
3558 */
3559 if (increasing) {
3560 wait_point = array_size;
3561 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
3562 wait_point = (__le64_to_cpu(bsb.arraystart) +
3563 __le64_to_cpu(bsb.length));
3564 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
3565 wait_point = (__le64_to_cpu(bsb.arraystart2) +
3566 __le64_to_cpu(bsb.length2));
3567 } else {
3568 wait_point = 0;
3569 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
3570 wait_point = __le64_to_cpu(bsb.arraystart);
3571 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
3572 wait_point = __le64_to_cpu(bsb.arraystart2);
3573 }
3574
3575 rv = progress_reshape(sra, reshape,
3576 backup_point, wait_point,
3577 &suspend_point, &reshape_completed);
3578 /* external metadata would need to ping_monitor here */
3579 sra->reshape_progress = reshape_completed;
3580
3581 /* Clear any backup region that is before 'here' */
3582 if (increasing) {
3583 if (__le64_to_cpu(bsb.length) > 0 &&
3584 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
3585 __le64_to_cpu(bsb.length)))
3586 forget_backup(dests, destfd,
3587 destoffsets, 0);
3588 if (__le64_to_cpu(bsb.length2) > 0 &&
3589 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
3590 __le64_to_cpu(bsb.length2)))
3591 forget_backup(dests, destfd,
3592 destoffsets, 1);
3593 } else {
3594 if (__le64_to_cpu(bsb.length) > 0 &&
3595 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
3596 forget_backup(dests, destfd,
3597 destoffsets, 0);
3598 if (__le64_to_cpu(bsb.length2) > 0 &&
3599 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
3600 forget_backup(dests, destfd,
3601 destoffsets, 1);
3602 }
3603
3604 if (rv < 0) {
3605 if (rv == -1)
3606 done = 1;
3607 break;
3608 }
3609 if (rv == 0 && increasing && !st->ss->external) {
3610 /* No longer need to monitor this reshape */
3611 done = 1;
3612 break;
3613 }
3614
3615 while (rv) {
3616 unsigned long long offset;
3617 unsigned long actual_stripes;
3618 /* Need to backup some data.
3619 * If 'part' is not used and the desired
3620 * backup size is suspended, do a backup,
3621 * then consider the next part.
3622 */
3623 /* Check that 'part' is unused */
3624 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
3625 break;
3626 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
3627 break;
3628
3629 offset = backup_point / data;
3630 actual_stripes = stripes;
3631 if (increasing) {
3632 if (offset + actual_stripes * (chunk/512) >
3633 sra->component_size)
3634 actual_stripes = ((sra->component_size - offset)
3635 / (chunk/512));
3636 if (offset + actual_stripes * (chunk/512) >
3637 suspend_point/data)
3638 break;
3639 } else {
3640 if (offset < actual_stripes * (chunk/512))
3641 actual_stripes = offset / (chunk/512);
3642 offset -= actual_stripes * (chunk/512);
3643 if (offset < suspend_point/data)
3644 break;
3645 }
3646 if (actual_stripes == 0)
3647 break;
3648 grow_backup(sra, offset, actual_stripes,
3649 fds, offsets,
3650 disks, chunk, level, layout,
3651 dests, destfd, destoffsets,
3652 part, &degraded, buf);
3653 validate(afd, destfd[0], destoffsets[0]);
3654 /* record where 'part' is up to */
3655 part = !part;
3656 if (increasing)
3657 backup_point += actual_stripes * (chunk/512) * data;
3658 else
3659 backup_point -= actual_stripes * (chunk/512) * data;
3660 }
3661 }
3662
3663 /* FIXME maybe call progress_reshape one more time instead */
3664 abort_reshape(sra); /* remove any remaining suspension */
3665 if (reshape->before.data_disks == reshape->after.data_disks)
3666 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
3667 free(buf);
3668 return done;
3669 }
3670
3671 /*
3672 * If any spare contains md_back_data-1 which is recent wrt mtime,
3673 * write that data into the array and update the super blocks with
3674 * the new reshape_progress
3675 */
3676 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
3677 char *backup_file, int verbose)
3678 {
3679 int i, j;
3680 int old_disks;
3681 unsigned long long *offsets;
3682 unsigned long long nstripe, ostripe;
3683 int ndata, odata;
3684
3685 odata = info->array.raid_disks - info->delta_disks - 1;
3686 if (info->array.level == 6) odata--; /* number of data disks */
3687 ndata = info->array.raid_disks - 1;
3688 if (info->new_level == 6) ndata--;
3689
3690 old_disks = info->array.raid_disks - info->delta_disks;
3691
3692 if (info->delta_disks <= 0)
3693 /* Didn't grow, so the backup file must have
3694 * been used
3695 */
3696 old_disks = cnt;
3697 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
3698 struct mdinfo dinfo;
3699 int fd;
3700 int bsbsize;
3701 char *devname, namebuf[20];
3702 unsigned long long lo, hi;
3703
3704 /* This was a spare and may have some saved data on it.
3705 * Load the superblock, find and load the
3706 * backup_super_block.
3707 * If either fail, go on to next device.
3708 * If the backup contains no new info, just return
3709 * else restore data and update all superblocks
3710 */
3711 if (i == old_disks-1) {
3712 fd = open(backup_file, O_RDONLY);
3713 if (fd<0) {
3714 pr_err("backup file %s inaccessible: %s\n",
3715 backup_file, strerror(errno));
3716 continue;
3717 }
3718 devname = backup_file;
3719 } else {
3720 fd = fdlist[i];
3721 if (fd < 0)
3722 continue;
3723 if (st->ss->load_super(st, fd, NULL))
3724 continue;
3725
3726 st->ss->getinfo_super(st, &dinfo, NULL);
3727 st->ss->free_super(st);
3728
3729 if (lseek64(fd,
3730 (dinfo.data_offset + dinfo.component_size - 8) <<9,
3731 0) < 0) {
3732 pr_err("Cannot seek on device %d\n", i);
3733 continue; /* Cannot seek */
3734 }
3735 sprintf(namebuf, "device-%d", i);
3736 devname = namebuf;
3737 }
3738 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
3739 if (verbose)
3740 pr_err("Cannot read from %s\n", devname);
3741 continue; /* Cannot read */
3742 }
3743 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
3744 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
3745 if (verbose)
3746 pr_err("No backup metadata on %s\n", devname);
3747 continue;
3748 }
3749 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
3750 if (verbose)
3751 pr_err("Bad backup-metadata checksum on %s\n", devname);
3752 continue; /* bad checksum */
3753 }
3754 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
3755 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
3756 if (verbose)
3757 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
3758 continue; /* Bad second checksum */
3759 }
3760 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
3761 if (verbose)
3762 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
3763 continue; /* Wrong uuid */
3764 }
3765
3766 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3767 * sometimes they aren't... So allow considerable flexability in matching, and allow
3768 * this test to be overridden by an environment variable.
3769 */
3770 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
3771 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
3772 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3773 pr_err("accepting backup with timestamp %lu "
3774 "for array with timestamp %lu\n",
3775 (unsigned long)__le64_to_cpu(bsb.mtime),
3776 (unsigned long)info->array.utime);
3777 } else {
3778 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
3779 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
3780 continue; /* time stamp is too bad */
3781 }
3782 }
3783
3784 if (bsb.magic[15] == '1') {
3785 if (bsb.length == 0)
3786 continue;
3787 if (info->delta_disks >= 0) {
3788 /* reshape_progress is increasing */
3789 if (__le64_to_cpu(bsb.arraystart)
3790 + __le64_to_cpu(bsb.length)
3791 < info->reshape_progress) {
3792 nonew:
3793 if (verbose)
3794 pr_err("backup-metadata found on %s but is not needed\n", devname);
3795 continue; /* No new data here */
3796 }
3797 } else {
3798 /* reshape_progress is decreasing */
3799 if (__le64_to_cpu(bsb.arraystart) >=
3800 info->reshape_progress)
3801 goto nonew; /* No new data here */
3802 }
3803 } else {
3804 if (bsb.length == 0 && bsb.length2 == 0)
3805 continue;
3806 if (info->delta_disks >= 0) {
3807 /* reshape_progress is increasing */
3808 if ((__le64_to_cpu(bsb.arraystart)
3809 + __le64_to_cpu(bsb.length)
3810 < info->reshape_progress)
3811 &&
3812 (__le64_to_cpu(bsb.arraystart2)
3813 + __le64_to_cpu(bsb.length2)
3814 < info->reshape_progress))
3815 goto nonew; /* No new data here */
3816 } else {
3817 /* reshape_progress is decreasing */
3818 if (__le64_to_cpu(bsb.arraystart) >=
3819 info->reshape_progress &&
3820 __le64_to_cpu(bsb.arraystart2) >=
3821 info->reshape_progress)
3822 goto nonew; /* No new data here */
3823 }
3824 }
3825 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
3826 second_fail:
3827 if (verbose)
3828 pr_err("Failed to verify secondary backup-metadata block on %s\n",
3829 devname);
3830 continue; /* Cannot seek */
3831 }
3832 /* There should be a duplicate backup superblock 4k before here */
3833 if (lseek64(fd, -4096, 1) < 0 ||
3834 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
3835 goto second_fail; /* Cannot find leading superblock */
3836 if (bsb.magic[15] == '1')
3837 bsbsize = offsetof(struct mdp_backup_super, pad1);
3838 else
3839 bsbsize = offsetof(struct mdp_backup_super, pad);
3840 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
3841 goto second_fail; /* Cannot find leading superblock */
3842
3843 /* Now need the data offsets for all devices. */
3844 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
3845 for(j=0; j<info->array.raid_disks; j++) {
3846 if (fdlist[j] < 0)
3847 continue;
3848 if (st->ss->load_super(st, fdlist[j], NULL))
3849 /* FIXME should be this be an error */
3850 continue;
3851 st->ss->getinfo_super(st, &dinfo, NULL);
3852 st->ss->free_super(st);
3853 offsets[j] = dinfo.data_offset * 512;
3854 }
3855 printf(Name ": restoring critical section\n");
3856
3857 if (restore_stripes(fdlist, offsets,
3858 info->array.raid_disks,
3859 info->new_chunk,
3860 info->new_level,
3861 info->new_layout,
3862 fd, __le64_to_cpu(bsb.devstart)*512,
3863 __le64_to_cpu(bsb.arraystart)*512,
3864 __le64_to_cpu(bsb.length)*512, NULL)) {
3865 /* didn't succeed, so giveup */
3866 if (verbose)
3867 pr_err("Error restoring backup from %s\n",
3868 devname);
3869 free(offsets);
3870 return 1;
3871 }
3872
3873 if (bsb.magic[15] == '2' &&
3874 restore_stripes(fdlist, offsets,
3875 info->array.raid_disks,
3876 info->new_chunk,
3877 info->new_level,
3878 info->new_layout,
3879 fd, __le64_to_cpu(bsb.devstart)*512 +
3880 __le64_to_cpu(bsb.devstart2)*512,
3881 __le64_to_cpu(bsb.arraystart2)*512,
3882 __le64_to_cpu(bsb.length2)*512, NULL)) {
3883 /* didn't succeed, so giveup */
3884 if (verbose)
3885 pr_err("Error restoring second backup from %s\n",
3886 devname);
3887 free(offsets);
3888 return 1;
3889 }
3890
3891 free(offsets);
3892
3893 /* Ok, so the data is restored. Let's update those superblocks. */
3894
3895 lo = hi = 0;
3896 if (bsb.length) {
3897 lo = __le64_to_cpu(bsb.arraystart);
3898 hi = lo + __le64_to_cpu(bsb.length);
3899 }
3900 if (bsb.magic[15] == '2' && bsb.length2) {
3901 unsigned long long lo1, hi1;
3902 lo1 = __le64_to_cpu(bsb.arraystart2);
3903 hi1 = lo1 + __le64_to_cpu(bsb.length2);
3904 if (lo == hi) {
3905 lo = lo1;
3906 hi = hi1;
3907 } else if (lo < lo1)
3908 hi = hi1;
3909 else
3910 lo = lo1;
3911 }
3912 if (lo < hi &&
3913 (info->reshape_progress < lo ||
3914 info->reshape_progress > hi))
3915 /* backup does not affect reshape_progress*/ ;
3916 else if (info->delta_disks >= 0) {
3917 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
3918 __le64_to_cpu(bsb.length);
3919 if (bsb.magic[15] == '2') {
3920 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
3921 __le64_to_cpu(bsb.length2);
3922 if (p2 > info->reshape_progress)
3923 info->reshape_progress = p2;
3924 }
3925 } else {
3926 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
3927 if (bsb.magic[15] == '2') {
3928 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
3929 if (p2 < info->reshape_progress)
3930 info->reshape_progress = p2;
3931 }
3932 }
3933 for (j=0; j<info->array.raid_disks; j++) {
3934 if (fdlist[j] < 0)
3935 continue;
3936 if (st->ss->load_super(st, fdlist[j], NULL))
3937 continue;
3938 st->ss->getinfo_super(st, &dinfo, NULL);
3939 dinfo.reshape_progress = info->reshape_progress;
3940 st->ss->update_super(st, &dinfo,
3941 "_reshape_progress",
3942 NULL,0, 0, NULL);
3943 st->ss->store_super(st, fdlist[j]);
3944 st->ss->free_super(st);
3945 }
3946 return 0;
3947 }
3948 /* Didn't find any backup data, try to see if any
3949 * was needed.
3950 */
3951 if (info->delta_disks < 0) {
3952 /* When shrinking, the critical section is at the end.
3953 * So see if we are before the critical section.
3954 */
3955 unsigned long long first_block;
3956 nstripe = ostripe = 0;
3957 first_block = 0;
3958 while (ostripe >= nstripe) {
3959 ostripe += info->array.chunk_size / 512;
3960 first_block = ostripe * odata;
3961 nstripe = first_block / ndata / (info->new_chunk/512) *
3962 (info->new_chunk/512);
3963 }
3964
3965 if (info->reshape_progress >= first_block)
3966 return 0;
3967 }
3968 if (info->delta_disks > 0) {
3969 /* See if we are beyond the critical section. */
3970 unsigned long long last_block;
3971 nstripe = ostripe = 0;
3972 last_block = 0;
3973 while (nstripe >= ostripe) {
3974 nstripe += info->new_chunk / 512;
3975 last_block = nstripe * ndata;
3976 ostripe = last_block / odata / (info->array.chunk_size/512) *
3977 (info->array.chunk_size/512);
3978 }
3979
3980 if (info->reshape_progress >= last_block)
3981 return 0;
3982 }
3983 /* needed to recover critical section! */
3984 if (verbose)
3985 pr_err("Failed to find backup of critical section\n");
3986 return 1;
3987 }
3988
3989 int Grow_continue_command(char *devname, int fd,
3990 char *backup_file, int verbose)
3991 {
3992 int ret_val = 0;
3993 struct supertype *st = NULL;
3994 struct mdinfo *content = NULL;
3995 struct mdinfo array;
3996 char *subarray = NULL;
3997 struct mdinfo *cc = NULL;
3998 struct mdstat_ent *mdstat = NULL;
3999 char buf[40];
4000 int cfd = -1;
4001 int fd2 = -1;
4002
4003 dprintf("Grow continue from command line called for %s\n",
4004 devname);
4005
4006 st = super_by_fd(fd, &subarray);
4007 if (!st || !st->ss) {
4008 pr_err("Unable to determine metadata format for %s\n",
4009 devname);
4010 return 1;
4011 }
4012 dprintf("Grow continue is run for ");
4013 if (st->ss->external == 0) {
4014 dprintf("native array (%s)\n", devname);
4015 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
4016 pr_err("%s is not an active md array -"
4017 " aborting\n", devname);
4018 ret_val = 1;
4019 goto Grow_continue_command_exit;
4020 }
4021 content = &array;
4022 sysfs_init(content, fd, st->devnum);
4023 } else {
4024 int container_dev;
4025
4026 if (subarray) {
4027 dprintf("subarray (%s)\n", subarray);
4028 container_dev = st->container_dev;
4029 cfd = open_dev_excl(st->container_dev);
4030 } else {
4031 container_dev = st->devnum;
4032 close(fd);
4033 cfd = open_dev_excl(st->devnum);
4034 dprintf("container (%i)\n", container_dev);
4035 fd = cfd;
4036 }
4037 if (cfd < 0) {
4038 pr_err("Unable to open container "
4039 "for %s\n", devname);
4040 ret_val = 1;
4041 goto Grow_continue_command_exit;
4042 }
4043 fmt_devname(buf, container_dev);
4044
4045 /* find in container array under reshape
4046 */
4047 ret_val = st->ss->load_container(st, cfd, NULL);
4048 if (ret_val) {
4049 pr_err("Cannot read superblock for %s\n",
4050 devname);
4051 ret_val = 1;
4052 goto Grow_continue_command_exit;
4053 }
4054
4055 cc = st->ss->container_content(st, subarray);
4056 for (content = cc; content ; content = content->next) {
4057 char *array;
4058 int allow_reshape = 1;
4059
4060 if (content->reshape_active == 0)
4061 continue;
4062 /* The decision about array or container wide
4063 * reshape is taken in Grow_continue based
4064 * content->reshape_active state, therefore we
4065 * need to check_reshape based on
4066 * reshape_active and subarray name
4067 */
4068 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4069 allow_reshape = 0;
4070 if (content->reshape_active == CONTAINER_RESHAPE &&
4071 (content->array.state
4072 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4073 allow_reshape = 0;
4074
4075 if (!allow_reshape) {
4076 pr_err("cannot continue reshape of an array"
4077 " in container with unsupported"
4078 " metadata: %s(%s)\n",
4079 devname, buf);
4080 ret_val = 1;
4081 goto Grow_continue_command_exit;
4082 }
4083
4084 array = strchr(content->text_version+1, '/')+1;
4085 mdstat = mdstat_by_subdev(array, container_dev);
4086 if (!mdstat)
4087 continue;
4088 if (mdstat->active == 0) {
4089 pr_err("Skipping inactive "
4090 "array md%i.\n", mdstat->devnum);
4091 free_mdstat(mdstat);
4092 mdstat = NULL;
4093 continue;
4094 }
4095 break;
4096 }
4097 if (!content) {
4098 pr_err("Unable to determine reshaped "
4099 "array for %s\n", devname);
4100 ret_val = 1;
4101 goto Grow_continue_command_exit;
4102 }
4103 fd2 = open_dev(mdstat->devnum);
4104 if (fd2 < 0) {
4105 pr_err("cannot open (md%i)\n",
4106 mdstat->devnum);
4107 ret_val = 1;
4108 goto Grow_continue_command_exit;
4109 }
4110
4111 sysfs_init(content, fd2, mdstat->devnum);
4112
4113 /* start mdmon in case it is not running
4114 */
4115 if (!mdmon_running(container_dev))
4116 start_mdmon(container_dev);
4117 ping_monitor(buf);
4118
4119 if (mdmon_running(container_dev))
4120 st->update_tail = &st->updates;
4121 else {
4122 pr_err("No mdmon found. "
4123 "Grow cannot continue.\n");
4124 ret_val = 1;
4125 goto Grow_continue_command_exit;
4126 }
4127 }
4128
4129 /* verify that array under reshape is started from
4130 * correct position
4131 */
4132 if (verify_reshape_position(content,
4133 map_name(pers, mdstat->level)) < 0) {
4134 ret_val = 1;
4135 goto Grow_continue_command_exit;
4136 }
4137
4138 /* continue reshape
4139 */
4140 ret_val = Grow_continue(fd, st, content, backup_file, 0);
4141
4142 Grow_continue_command_exit:
4143 if (fd2 > -1)
4144 close(fd2);
4145 if (cfd > -1)
4146 close(cfd);
4147 st->ss->free_super(st);
4148 free_mdstat(mdstat);
4149 sysfs_free(cc);
4150 free(subarray);
4151
4152 return ret_val;
4153 }
4154
4155 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4156 char *backup_file, int freeze_reshape)
4157 {
4158 int ret_val = 2;
4159
4160 if (!info->reshape_active)
4161 return ret_val;
4162
4163 if (st->ss->external) {
4164 char container[40];
4165 int cfd = open_dev(st->container_dev);
4166
4167 if (cfd < 0)
4168 return 1;
4169
4170 fmt_devname(container, st->container_dev);
4171 st->ss->load_container(st, cfd, container);
4172 close(cfd);
4173 ret_val = reshape_container(container, NULL, mdfd,
4174 st, info, 0, backup_file,
4175 0, 1, freeze_reshape);
4176 } else
4177 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4178 NULL, backup_file, 0, 0, 1,
4179 freeze_reshape);
4180
4181 return ret_val;
4182 }
Unnamed repository; edit this file 'description' to name the repository.