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