]> 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
FIX: Detect error and rollback metadata
[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(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 rv = 1;
1672 break;
1673 }
1674 if (array.not_persistent == 0 &&
1675 array.major_version == 0 &&
1676 get_linux_version() < 3001000) {
1677 /* Dangerous to allow size to exceed 2TB */
1678 unsigned long long csize;
1679 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1680 if (csize >= 2ULL*1024*1024*1024)
1681 csize = 2ULL*1024*1024*1024;
1682 if ((min_csize == 0 || (min_csize
1683 > (long long)csize)))
1684 min_csize = csize;
1685 }
1686 }
1687 }
1688 if (rv) {
1689 fprintf(stderr, Name ": Cannot set size on "
1690 "array members.\n");
1691 goto size_change_error;
1692 }
1693 if (min_csize && size > min_csize) {
1694 fprintf(stderr, Name ": Cannot safely make this array "
1695 "use more than 2TB per device on this kernel.\n");
1696 rv = 1;
1697 goto size_change_error;
1698 }
1699 if (min_csize && size == 0) {
1700 /* Don't let the kernel choose a size - it will get
1701 * it wrong
1702 */
1703 fprintf(stderr, Name ": Limited v0.90 array to "
1704 "2TB per device\n");
1705 size = min_csize;
1706 }
1707 if (st->ss->external) {
1708 if (sra->array.level == 0) {
1709 rv = sysfs_set_str(sra, NULL, "level",
1710 "raid5");
1711 if (!rv) {
1712 raid0_takeover = 1;
1713 /* get array parametes after takeover
1714 * to chane one parameter at time only
1715 */
1716 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1717 }
1718 }
1719 /* make sure mdmon is
1720 * aware of the new level */
1721 if (!mdmon_running(st->container_dev))
1722 start_mdmon(st->container_dev);
1723 ping_monitor(container);
1724 if (mdmon_running(st->container_dev) &&
1725 st->update_tail == NULL)
1726 st->update_tail = &st->updates;
1727 }
1728
1729 array.size = size;
1730 if (array.size != size) {
1731 /* got truncated to 32bit, write to
1732 * component_size instead
1733 */
1734 if (sra)
1735 rv = sysfs_set_num(sra, NULL,
1736 "component_size", size);
1737 else
1738 rv = -1;
1739 } else {
1740 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1741
1742 /* manage array size when it is managed externally
1743 */
1744 if ((rv == 0) && st->ss->external)
1745 rv = set_array_size(st, sra, sra->text_version);
1746 }
1747
1748 if (raid0_takeover) {
1749 /* do not recync non-existing parity,
1750 * we will drop it anyway
1751 */
1752 sysfs_set_str(sra, NULL, "sync_action", "idle");
1753 /* go back to raid0, drop parity disk
1754 */
1755 sysfs_set_str(sra, NULL, "level", "raid0");
1756 ioctl(fd, GET_ARRAY_INFO, &array);
1757 }
1758
1759 size_change_error:
1760 if (rv != 0) {
1761 int err = errno;
1762
1763 /* restore metadata */
1764 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1765 UnSet, NULL, devname,
1766 ROLLBACK_METADATA_CHANGES,
1767 !quiet) == 0)
1768 sync_metadata(st);
1769 fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
1770 devname, strerror(err));
1771 if (err == EBUSY &&
1772 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1773 fprintf(stderr, " Bitmap must be removed before size can be changed\n");
1774 rv = 1;
1775 goto release;
1776 }
1777 if (assume_clean) {
1778 /* This will fail on kernels newer than 3.0 unless
1779 * a backport has been arranged.
1780 */
1781 if (sra == NULL ||
1782 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1783 fprintf(stderr, Name ": --assume-clean not support with --grow on this kernel\n");
1784 }
1785 ioctl(fd, GET_ARRAY_INFO, &array);
1786 size = get_component_size(fd)/2;
1787 if (size == 0)
1788 size = array.size;
1789 if (!quiet) {
1790 if (size == orig_size)
1791 fprintf(stderr, Name ": component size of %s "
1792 "unchanged at %lluK\n",
1793 devname, size);
1794 else
1795 fprintf(stderr, Name ": component size of %s "
1796 "has been set to %lluK\n",
1797 devname, size);
1798 }
1799 changed = 1;
1800 } else if (array.level != LEVEL_CONTAINER) {
1801 size = get_component_size(fd)/2;
1802 if (size == 0)
1803 size = array.size;
1804 }
1805
1806 /* See if there is anything else to do */
1807 if ((level == UnSet || level == array.level) &&
1808 (layout_str == NULL) &&
1809 (chunksize == 0 || chunksize == array.chunk_size) &&
1810 (raid_disks == 0 || raid_disks == array.raid_disks)) {
1811 /* Nothing more to do */
1812 if (!changed && !quiet)
1813 fprintf(stderr, Name ": %s: no change requested\n",
1814 devname);
1815 goto release;
1816 }
1817
1818 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1819 * current implementation assumes that following conditions must be met:
1820 * - RAID10:
1821 * - far_copies == 1
1822 * - near_copies == 2
1823 */
1824 if ((level == 0 && array.level == 10 && sra &&
1825 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1826 (level == 0 && array.level == 1 && sra)) {
1827 int err;
1828 err = remove_disks_for_takeover(st, sra, array.layout);
1829 if (err) {
1830 dprintf(Name": Array cannot be reshaped\n");
1831 if (cfd > -1)
1832 close(cfd);
1833 rv = 1;
1834 goto release;
1835 }
1836 /* Make sure mdmon has seen the device removal
1837 * and updated metadata before we continue with
1838 * level change
1839 */
1840 if (container)
1841 ping_monitor(container);
1842 }
1843
1844 memset(&info, 0, sizeof(info));
1845 info.array = array;
1846 sysfs_init(&info, fd, NoMdDev);
1847 strcpy(info.text_version, sra->text_version);
1848 info.component_size = size*2;
1849 info.new_level = level;
1850 info.new_chunk = chunksize * 1024;
1851 if (info.array.level == LEVEL_CONTAINER) {
1852 info.delta_disks = UnSet;
1853 info.array.raid_disks = raid_disks;
1854 } else if (raid_disks)
1855 info.delta_disks = raid_disks - info.array.raid_disks;
1856 else
1857 info.delta_disks = UnSet;
1858 if (layout_str == NULL) {
1859 info.new_layout = UnSet;
1860 if (info.array.level == 6 &&
1861 (info.new_level == 6 || info.new_level == UnSet) &&
1862 info.array.layout >= 16) {
1863 fprintf(stderr, Name
1864 ": %s has a non-standard layout. If you"
1865 " wish to preserve this\n"
1866 " during the reshape, please specify"
1867 " --layout=preserve\n"
1868 " If you want to change it, specify a"
1869 " layout or use --layout=normalise\n",
1870 devname);
1871 rv = 1;
1872 goto release;
1873 }
1874 } else if (strcmp(layout_str, "normalise") == 0 ||
1875 strcmp(layout_str, "normalize") == 0) {
1876 /* If we have a -6 RAID6 layout, remove the '-6'. */
1877 info.new_layout = UnSet;
1878 if (info.array.level == 6 && info.new_level == UnSet) {
1879 char l[40], *h;
1880 strcpy(l, map_num(r6layout, info.array.layout));
1881 h = strrchr(l, '-');
1882 if (h && strcmp(h, "-6") == 0) {
1883 *h = 0;
1884 info.new_layout = map_name(r6layout, l);
1885 }
1886 }
1887 } else if (strcmp(layout_str, "preserve") == 0) {
1888 info.new_layout = UnSet;
1889 } else {
1890 int l = info.new_level;
1891 if (l == UnSet)
1892 l = info.array.level;
1893 switch (l) {
1894 case 5:
1895 info.new_layout = map_name(r5layout, layout_str);
1896 break;
1897 case 6:
1898 info.new_layout = map_name(r6layout, layout_str);
1899 break;
1900 case 10:
1901 info.new_layout = parse_layout_10(layout_str);
1902 break;
1903 case LEVEL_FAULTY:
1904 info.new_layout = parse_layout_faulty(layout_str);
1905 break;
1906 default:
1907 fprintf(stderr, Name ": layout not meaningful"
1908 " with this level\n");
1909 rv = 1;
1910 goto release;
1911 }
1912 if (info.new_layout == UnSet) {
1913 fprintf(stderr, Name ": layout %s not understood"
1914 " for this level\n",
1915 layout_str);
1916 rv = 1;
1917 goto release;
1918 }
1919 }
1920
1921 if (array.level == LEVEL_FAULTY) {
1922 if (level != UnSet && level != array.level) {
1923 fprintf(stderr, Name ": cannot change level of Faulty device\n");
1924 rv =1 ;
1925 }
1926 if (chunksize) {
1927 fprintf(stderr, Name ": cannot set chunksize of Faulty device\n");
1928 rv =1 ;
1929 }
1930 if (raid_disks && raid_disks != 1) {
1931 fprintf(stderr, Name ": cannot set raid_disks of Faulty device\n");
1932 rv =1 ;
1933 }
1934 if (layout_str) {
1935 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
1936 dprintf("Cannot get array information.\n");
1937 goto release;
1938 }
1939 array.layout = info.new_layout;
1940 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1941 fprintf(stderr, Name ": failed to set new layout\n");
1942 rv = 1;
1943 } else if (!quiet)
1944 printf("layout for %s set to %d\n",
1945 devname, array.layout);
1946 }
1947 } else if (array.level == LEVEL_CONTAINER) {
1948 /* This change is to be applied to every array in the
1949 * container. This is only needed when the metadata imposes
1950 * restraints of the various arrays in the container.
1951 * Currently we only know that IMSM requires all arrays
1952 * to have the same number of devices so changing the
1953 * number of devices (On-Line Capacity Expansion) must be
1954 * performed at the level of the container
1955 */
1956 rv = reshape_container(container, devname, -1, st, &info,
1957 force, backup_file, quiet, 0, 0);
1958 frozen = 0;
1959 } else {
1960 /* get spare devices from external metadata
1961 */
1962 if (st->ss->external) {
1963 struct mdinfo *info2;
1964
1965 info2 = st->ss->container_content(st, subarray);
1966 if (info2) {
1967 info.array.spare_disks =
1968 info2->array.spare_disks;
1969 sysfs_free(info2);
1970 }
1971 }
1972
1973 /* Impose these changes on a single array. First
1974 * check that the metadata is OK with the change. */
1975
1976 if (reshape_super(st, info.component_size, info.new_level,
1977 info.new_layout, info.new_chunk,
1978 info.array.raid_disks, info.delta_disks,
1979 backup_file, devname, APPLY_METADATA_CHANGES,
1980 quiet)) {
1981 rv = 1;
1982 goto release;
1983 }
1984 sync_metadata(st);
1985 rv = reshape_array(container, fd, devname, st, &info, force,
1986 devlist, backup_file, quiet, 0, 0, 0);
1987 frozen = 0;
1988 }
1989 release:
1990 sysfs_free(sra);
1991 if (frozen > 0)
1992 unfreeze(st);
1993 return rv;
1994 }
1995
1996 /* verify_reshape_position()
1997 * Function checks if reshape position in metadata is not farther
1998 * than position in md.
1999 * Return value:
2000 * 0 : not valid sysfs entry
2001 * it can be caused by not started reshape, it should be started
2002 * by reshape array or raid0 array is before takeover
2003 * -1 : error, reshape position is obviously wrong
2004 * 1 : success, reshape progress correct or updated
2005 */
2006 static int verify_reshape_position(struct mdinfo *info, int level)
2007 {
2008 int ret_val = 0;
2009 char buf[40];
2010 int rv;
2011
2012 /* read sync_max, failure can mean raid0 array */
2013 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2014
2015 if (rv > 0) {
2016 char *ep;
2017 unsigned long long position = strtoull(buf, &ep, 0);
2018
2019 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2020 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2021 position *= get_data_disks(level,
2022 info->new_layout,
2023 info->array.raid_disks);
2024 if (info->reshape_progress < position) {
2025 dprintf("Corrected reshape progress (%llu) to "
2026 "md position (%llu)\n",
2027 info->reshape_progress, position);
2028 info->reshape_progress = position;
2029 ret_val = 1;
2030 } else if (info->reshape_progress > position) {
2031 fprintf(stderr, Name ": Fatal error: array "
2032 "reshape was not properly frozen "
2033 "(expected reshape position is %llu, "
2034 "but reshape progress is %llu.\n",
2035 position, info->reshape_progress);
2036 ret_val = -1;
2037 } else {
2038 dprintf("Reshape position in md and metadata "
2039 "are the same;");
2040 ret_val = 1;
2041 }
2042 }
2043 } else if (rv == 0) {
2044 /* for valid sysfs entry, 0-length content
2045 * should be indicated as error
2046 */
2047 ret_val = -1;
2048 }
2049
2050 return ret_val;
2051 }
2052
2053 static int reshape_array(char *container, int fd, char *devname,
2054 struct supertype *st, struct mdinfo *info,
2055 int force, struct mddev_dev *devlist,
2056 char *backup_file, int quiet, int forked,
2057 int restart, int freeze_reshape)
2058 {
2059 struct reshape reshape;
2060 int spares_needed;
2061 char *msg;
2062 int orig_level = UnSet;
2063 int disks, odisks;
2064
2065 struct mdu_array_info_s array;
2066 char *c;
2067
2068 struct mddev_dev *dv;
2069 int added_disks;
2070
2071 int *fdlist = NULL;
2072 unsigned long long *offsets = NULL;
2073 int d;
2074 int nrdisks;
2075 int err;
2076 unsigned long blocks;
2077 unsigned long cache;
2078 unsigned long long array_size;
2079 int done;
2080 struct mdinfo *sra = NULL;
2081
2082 /* when reshaping a RAID0, the component_size might be zero.
2083 * So try to fix that up.
2084 */
2085 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2086 dprintf("Cannot get array information.\n");
2087 goto release;
2088 }
2089 if (array.level == 0 && info->component_size == 0) {
2090 get_dev_size(fd, NULL, &array_size);
2091 info->component_size = array_size / array.raid_disks;
2092 }
2093
2094 if (info->reshape_active) {
2095 int new_level = info->new_level;
2096 info->new_level = UnSet;
2097 if (info->delta_disks > 0)
2098 info->array.raid_disks -= info->delta_disks;
2099 msg = analyse_change(info, &reshape);
2100 info->new_level = new_level;
2101 if (info->delta_disks > 0)
2102 info->array.raid_disks += info->delta_disks;
2103 if (!restart)
2104 /* Make sure the array isn't read-only */
2105 ioctl(fd, RESTART_ARRAY_RW, 0);
2106 } else
2107 msg = analyse_change(info, &reshape);
2108 if (msg) {
2109 fprintf(stderr, Name ": %s\n", msg);
2110 goto release;
2111 }
2112 if (restart &&
2113 (reshape.level != info->array.level ||
2114 reshape.before.layout != info->array.layout ||
2115 reshape.before.data_disks + reshape.parity
2116 != info->array.raid_disks - max(0, info->delta_disks))) {
2117 fprintf(stderr, Name ": reshape info is not in native format -"
2118 " cannot continue.\n");
2119 goto release;
2120 }
2121
2122 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2123 /* When reshape is restarted from '0', very begin of array
2124 * it is possible that for external metadata reshape and array
2125 * configuration doesn't happen.
2126 * Check if md has the same opinion, and reshape is restarted
2127 * from 0. If so, this is regular reshape start after reshape
2128 * switch in metadata to next array only.
2129 */
2130 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2131 (info->reshape_progress == 0))
2132 restart = 0;
2133 }
2134 if (restart) {
2135 /* reshape already started. just skip to monitoring the reshape */
2136 if (reshape.backup_blocks == 0)
2137 return 0;
2138 goto started;
2139 }
2140 /* The container is frozen but the array may not be.
2141 * So freeze the array so spares don't get put to the wrong use
2142 * FIXME there should probably be a cleaner separation between
2143 * freeze_array and freeze_container.
2144 */
2145 sysfs_freeze_array(info);
2146 /* Check we have enough spares to not be degraded */
2147 added_disks = 0;
2148 for (dv = devlist; dv ; dv=dv->next)
2149 added_disks++;
2150 spares_needed = max(reshape.before.data_disks,
2151 reshape.after.data_disks)
2152 + reshape.parity - array.raid_disks;
2153
2154 if (!force &&
2155 info->new_level > 1 && info->array.level > 1 &&
2156 spares_needed > info->array.spare_disks + added_disks) {
2157 fprintf(stderr,
2158 Name ": Need %d spare%s to avoid degraded array,"
2159 " and only have %d.\n"
2160 " Use --force to over-ride this check.\n",
2161 spares_needed,
2162 spares_needed == 1 ? "" : "s",
2163 info->array.spare_disks + added_disks);
2164 goto release;
2165 }
2166 /* Check we have enough spares to not fail */
2167 spares_needed = max(reshape.before.data_disks,
2168 reshape.after.data_disks)
2169 - array.raid_disks;
2170 if ((info->new_level > 1 || info->new_level == 0) &&
2171 spares_needed > info->array.spare_disks +added_disks) {
2172 fprintf(stderr,
2173 Name ": Need %d spare%s to create working array,"
2174 " and only have %d.\n",
2175 spares_needed,
2176 spares_needed == 1 ? "" : "s",
2177 info->array.spare_disks + added_disks);
2178 goto release;
2179 }
2180
2181 if (reshape.level != array.level) {
2182 char *c = map_num(pers, reshape.level);
2183 int err;
2184 if (c == NULL)
2185 goto release;
2186
2187 err = sysfs_set_str(info, NULL, "level", c);
2188 if (err) {
2189 err = errno;
2190 fprintf(stderr, Name ": %s: could not set level to %s\n",
2191 devname, c);
2192 if (err == EBUSY &&
2193 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2194 fprintf(stderr, " Bitmap must be removed"
2195 " before level can be changed\n");
2196 goto release;
2197 }
2198 if (!quiet)
2199 fprintf(stderr, Name ": level of %s changed to %s\n",
2200 devname, c);
2201 orig_level = array.level;
2202 sysfs_freeze_array(info);
2203
2204 if (reshape.level > 0 && st->ss->external) {
2205 /* make sure mdmon is aware of the new level */
2206 if (mdmon_running(st->container_dev))
2207 flush_mdmon(container);
2208
2209 if (!mdmon_running(st->container_dev))
2210 start_mdmon(st->container_dev);
2211 ping_monitor(container);
2212 if (mdmon_running(st->container_dev) &&
2213 st->update_tail == NULL)
2214 st->update_tail = &st->updates;
2215 }
2216 }
2217 /* ->reshape_super might have chosen some spares from the
2218 * container that it wants to be part of the new array.
2219 * We can collect them with ->container_content and give
2220 * them to the kernel.
2221 */
2222 if (st->ss->reshape_super && st->ss->container_content) {
2223 char *subarray = strchr(info->text_version+1, '/')+1;
2224 struct mdinfo *info2 =
2225 st->ss->container_content(st, subarray);
2226 struct mdinfo *d;
2227
2228 if (info2) {
2229 sysfs_init(info2, fd, st->devnum);
2230 /* When increasing number of devices, we need to set
2231 * new raid_disks before adding these, or they might
2232 * be rejected.
2233 */
2234 if (reshape.backup_blocks &&
2235 reshape.after.data_disks > reshape.before.data_disks)
2236 subarray_set_num(container, info2, "raid_disks",
2237 reshape.after.data_disks +
2238 reshape.parity);
2239 for (d = info2->devs; d; d = d->next) {
2240 if (d->disk.state == 0 &&
2241 d->disk.raid_disk >= 0) {
2242 /* This is a spare that wants to
2243 * be part of the array.
2244 */
2245 add_disk(fd, st, info2, d);
2246 }
2247 }
2248 sysfs_free(info2);
2249 }
2250 }
2251 /* We might have been given some devices to add to the
2252 * array. Now that the array has been changed to the right
2253 * level and frozen, we can safely add them.
2254 */
2255 if (devlist)
2256 Manage_subdevs(devname, fd, devlist, !quiet,
2257 0,NULL, 0);
2258
2259 if (reshape.backup_blocks == 0) {
2260 /* No restriping needed, but we might need to impose
2261 * some more changes: layout, raid_disks, chunk_size
2262 */
2263 /* read current array info */
2264 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2265 dprintf("Cannot get array information.\n");
2266 goto release;
2267 }
2268 /* compare current array info with new values and if
2269 * it is different update them to new */
2270 if (info->new_layout != UnSet &&
2271 info->new_layout != array.layout) {
2272 array.layout = info->new_layout;
2273 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2274 fprintf(stderr, Name ": failed to set new layout\n");
2275 goto release;
2276 } else if (!quiet)
2277 printf("layout for %s set to %d\n",
2278 devname, array.layout);
2279 }
2280 if (info->delta_disks != UnSet &&
2281 info->delta_disks != 0 &&
2282 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
2283 array.raid_disks += info->delta_disks;
2284 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2285 fprintf(stderr, Name ": failed to set raid disks\n");
2286 goto release;
2287 } else if (!quiet) {
2288 printf("raid_disks for %s set to %d\n",
2289 devname, array.raid_disks);
2290 }
2291 }
2292 if (info->new_chunk != 0 &&
2293 info->new_chunk != array.chunk_size) {
2294 if (sysfs_set_num(info, NULL,
2295 "chunk_size", info->new_chunk) != 0) {
2296 fprintf(stderr, Name ": failed to set chunk size\n");
2297 goto release;
2298 } else if (!quiet)
2299 printf("chunk size for %s set to %d\n",
2300 devname, array.chunk_size);
2301 }
2302 unfreeze(st);
2303 return 0;
2304 }
2305
2306 /*
2307 * There are three possibilities.
2308 * 1/ The array will shrink.
2309 * We need to ensure the reshape will pause before reaching
2310 * the 'critical section'. We also need to fork and wait for
2311 * that to happen. When it does we
2312 * suspend/backup/complete/unfreeze
2313 *
2314 * 2/ The array will not change size.
2315 * This requires that we keep a backup of a sliding window
2316 * so that we can restore data after a crash. So we need
2317 * to fork and monitor progress.
2318 * In future we will allow the data_offset to change, so
2319 * a sliding backup becomes unnecessary.
2320 *
2321 * 3/ The array will grow. This is relatively easy.
2322 * However the kernel's restripe routines will cheerfully
2323 * overwrite some early data before it is safe. So we
2324 * need to make a backup of the early parts of the array
2325 * and be ready to restore it if rebuild aborts very early.
2326 * For externally managed metadata, we still need a forked
2327 * child to monitor the reshape and suspend IO over the region
2328 * that is being reshaped.
2329 *
2330 * We backup data by writing it to one spare, or to a
2331 * file which was given on command line.
2332 *
2333 * In each case, we first make sure that storage is available
2334 * for the required backup.
2335 * Then we:
2336 * - request the shape change.
2337 * - fork to handle backup etc.
2338 */
2339 started:
2340 /* Check that we can hold all the data */
2341 get_dev_size(fd, NULL, &array_size);
2342 if (reshape.new_size < (array_size/512)) {
2343 fprintf(stderr,
2344 Name ": this change will reduce the size of the array.\n"
2345 " use --grow --array-size first to truncate array.\n"
2346 " e.g. mdadm --grow %s --array-size %llu\n",
2347 devname, reshape.new_size/2);
2348 goto release;
2349 }
2350
2351 sra = sysfs_read(fd, 0,
2352 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2353 GET_CACHE);
2354 if (!sra) {
2355 fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
2356 devname);
2357 goto release;
2358 }
2359
2360 /* Decide how many blocks (sectors) for a reshape
2361 * unit. The number we have so far is just a minimum
2362 */
2363 blocks = reshape.backup_blocks;
2364 if (reshape.before.data_disks ==
2365 reshape.after.data_disks) {
2366 /* Make 'blocks' bigger for better throughput, but
2367 * not so big that we reject it below.
2368 * Try for 16 megabytes
2369 */
2370 while (blocks * 32 < sra->component_size &&
2371 blocks < 16*1024*2)
2372 blocks *= 2;
2373 } else
2374 fprintf(stderr, Name ": Need to backup %luK of critical "
2375 "section..\n", blocks/2);
2376
2377 if (blocks >= sra->component_size/2) {
2378 fprintf(stderr, Name ": %s: Something wrong"
2379 " - reshape aborted\n",
2380 devname);
2381 goto release;
2382 }
2383
2384 /* Now we need to open all these devices so we can read/write.
2385 */
2386 nrdisks = max(reshape.before.data_disks,
2387 reshape.after.data_disks) + reshape.parity
2388 + sra->array.spare_disks;
2389 fdlist = malloc((1+nrdisks) * sizeof(int));
2390 offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
2391 if (!fdlist || !offsets) {
2392 fprintf(stderr, Name ": malloc failed: grow aborted\n");
2393 goto release;
2394 }
2395
2396 odisks = reshape.before.data_disks + reshape.parity;
2397 d = reshape_prepare_fdlist(devname, sra, odisks,
2398 nrdisks, blocks, backup_file,
2399 fdlist, offsets);
2400 if (d < 0) {
2401 goto release;
2402 }
2403 if ((st->ss->manage_reshape == NULL) ||
2404 (st->ss->recover_backup == NULL)) {
2405 if (backup_file == NULL) {
2406 if (reshape.after.data_disks <=
2407 reshape.before.data_disks) {
2408 fprintf(stderr, Name ": %s: Cannot grow - "
2409 "need backup-file\n", devname);
2410 goto release;
2411 } else if (sra->array.spare_disks == 0) {
2412 fprintf(stderr, Name ": %s: Cannot grow - "
2413 "need a spare or backup-file to backup "
2414 "critical section\n", devname);
2415 goto release;
2416 }
2417 } else {
2418 if (!reshape_open_backup_file(backup_file, fd, devname,
2419 (signed)blocks,
2420 fdlist+d, offsets+d,
2421 restart)) {
2422 goto release;
2423 }
2424 d++;
2425 }
2426 }
2427
2428 /* lastly, check that the internal stripe cache is
2429 * large enough, or it won't work.
2430 * It must hold at least 4 stripes of the larger
2431 * chunk size
2432 */
2433 cache = max(info->array.chunk_size, info->new_chunk);
2434 cache *= 4; /* 4 stripes minimum */
2435 cache /= 512; /* convert to sectors */
2436 disks = min(reshape.before.data_disks, reshape.after.data_disks);
2437 /* make sure there is room for 'blocks' with a bit to spare */
2438 if (cache < 16 + blocks / disks)
2439 cache = 16 + blocks / disks;
2440 cache /= (4096/512); /* Covert from sectors to pages */
2441
2442 if (sra->cache_size < cache)
2443 subarray_set_num(container, sra, "stripe_cache_size",
2444 cache+1);
2445
2446 /* Right, everything seems fine. Let's kick things off.
2447 * If only changing raid_disks, use ioctl, else use
2448 * sysfs.
2449 */
2450 sync_metadata(st);
2451
2452 sra->new_chunk = info->new_chunk;
2453
2454 if (restart) {
2455 /* for external metadata checkpoint saved by mdmon can be lost
2456 * or missed /due to e.g. crash/. Check if md is not during
2457 * restart farther than metadata points to.
2458 * If so, this means metadata information is obsolete.
2459 */
2460 if (st->ss->external)
2461 verify_reshape_position(info, reshape.level);
2462 sra->reshape_progress = info->reshape_progress;
2463 } else {
2464 sra->reshape_progress = 0;
2465 if (reshape.after.data_disks < reshape.before.data_disks)
2466 /* start from the end of the new array */
2467 sra->reshape_progress = (sra->component_size
2468 * reshape.after.data_disks);
2469 }
2470
2471 if (info->array.chunk_size == info->new_chunk &&
2472 reshape.before.layout == reshape.after.layout &&
2473 st->ss->external == 0) {
2474 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2475 ioctl(fd, GET_ARRAY_INFO, &array);
2476 array.raid_disks = reshape.after.data_disks + reshape.parity;
2477 if (!restart &&
2478 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2479 int err = errno;
2480
2481 fprintf(stderr,
2482 Name ": Cannot set device shape for %s: %s\n",
2483 devname, strerror(errno));
2484
2485 if (err == EBUSY &&
2486 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2487 fprintf(stderr,
2488 " Bitmap must be removed before"
2489 " shape can be changed\n");
2490
2491 goto release;
2492 }
2493 } else if (!restart) {
2494 /* set them all just in case some old 'new_*' value
2495 * persists from some earlier problem.
2496 */
2497 int err = 0;
2498 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2499 err = errno;
2500 if (!err && sysfs_set_num(sra, NULL, "layout",
2501 reshape.after.layout) < 0)
2502 err = errno;
2503 if (!err && subarray_set_num(container, sra, "raid_disks",
2504 reshape.after.data_disks +
2505 reshape.parity) < 0)
2506 err = errno;
2507 if (err) {
2508 fprintf(stderr, Name ": Cannot set device shape for %s\n",
2509 devname);
2510
2511 if (err == EBUSY &&
2512 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2513 fprintf(stderr,
2514 " Bitmap must be removed before"
2515 " shape can be changed\n");
2516 goto release;
2517 }
2518 }
2519
2520 err = start_reshape(sra, restart, reshape.before.data_disks,
2521 reshape.after.data_disks);
2522 if (err) {
2523 fprintf(stderr,
2524 Name ": Cannot %s reshape for %s\n",
2525 restart ? "continue" : "start",
2526 devname);
2527 goto release;
2528 }
2529 if (restart)
2530 sysfs_set_str(sra, NULL, "array_state", "active");
2531 if (freeze_reshape) {
2532 free(fdlist);
2533 free(offsets);
2534 sysfs_free(sra);
2535 fprintf(stderr, Name ": Reshape has to be continued from"
2536 " location %llu when root filesystem has been mounted.\n",
2537 sra->reshape_progress);
2538 return 1;
2539 }
2540
2541 /* Now we just need to kick off the reshape and watch, while
2542 * handling backups of the data...
2543 * This is all done by a forked background process.
2544 */
2545 switch(forked ? 0 : fork()) {
2546 case -1:
2547 fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n",
2548 strerror(errno));
2549 abort_reshape(sra);
2550 goto release;
2551 default:
2552 free(fdlist);
2553 free(offsets);
2554 sysfs_free(sra);
2555 return 0;
2556 case 0:
2557 map_fork();
2558 break;
2559 }
2560
2561 close(fd);
2562 if (check_env("MDADM_GROW_VERIFY"))
2563 fd = open(devname, O_RDONLY | O_DIRECT);
2564 else
2565 fd = -1;
2566 mlockall(MCL_FUTURE);
2567
2568 if (st->ss->external) {
2569 /* metadata handler takes it from here */
2570 done = st->ss->manage_reshape(
2571 fd, sra, &reshape, st, blocks,
2572 fdlist, offsets,
2573 d - odisks, fdlist+odisks,
2574 offsets+odisks);
2575 } else
2576 done = child_monitor(
2577 fd, sra, &reshape, st, blocks,
2578 fdlist, offsets,
2579 d - odisks, fdlist+odisks,
2580 offsets+odisks);
2581
2582 free(fdlist);
2583 free(offsets);
2584
2585 if (backup_file && done)
2586 unlink(backup_file);
2587 if (!done) {
2588 abort_reshape(sra);
2589 goto out;
2590 }
2591
2592 if (!st->ss->external &&
2593 !(reshape.before.data_disks != reshape.after.data_disks
2594 && info->custom_array_size) &&
2595 info->new_level == reshape.level &&
2596 !forked) {
2597 /* no need to wait for the reshape to finish as
2598 * there is nothing more to do.
2599 */
2600 sysfs_free(sra);
2601 exit(0);
2602 }
2603 wait_reshape(sra);
2604
2605 if (st->ss->external) {
2606 /* Re-load the metadata as much could have changed */
2607 int cfd = open_dev(st->container_dev);
2608 if (cfd >= 0) {
2609 flush_mdmon(container);
2610 st->ss->free_super(st);
2611 st->ss->load_container(st, cfd, container);
2612 close(cfd);
2613 }
2614 }
2615
2616 /* set new array size if required customer_array_size is used
2617 * by this metadata.
2618 */
2619 if (reshape.before.data_disks !=
2620 reshape.after.data_disks &&
2621 info->custom_array_size)
2622 set_array_size(st, info, info->text_version);
2623
2624 if (info->new_level != reshape.level) {
2625
2626 c = map_num(pers, info->new_level);
2627 if (c) {
2628 err = sysfs_set_str(sra, NULL, "level", c);
2629 if (err)
2630 fprintf(stderr, Name\
2631 ": %s: could not set level "
2632 "to %s\n", devname, c);
2633 }
2634 if (info->new_level == 0)
2635 st->update_tail = NULL;
2636 }
2637 out:
2638 sysfs_free(sra);
2639 if (forked)
2640 return 0;
2641 unfreeze(st);
2642 exit(0);
2643
2644 release:
2645 free(fdlist);
2646 free(offsets);
2647 if (orig_level != UnSet && sra) {
2648 c = map_num(pers, orig_level);
2649 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
2650 fprintf(stderr, Name ": aborting level change\n");
2651 }
2652 sysfs_free(sra);
2653 if (!forked)
2654 unfreeze(st);
2655 return 1;
2656 }
2657
2658 /* mdfd handle is passed to be closed in child process (after fork).
2659 */
2660 int reshape_container(char *container, char *devname,
2661 int mdfd,
2662 struct supertype *st,
2663 struct mdinfo *info,
2664 int force,
2665 char *backup_file,
2666 int quiet, int restart, int freeze_reshape)
2667 {
2668 struct mdinfo *cc = NULL;
2669 int rv = restart;
2670 int last_devnum = -1;
2671
2672 /* component_size is not meaningful for a container,
2673 * so pass '-1' meaning 'no change'
2674 */
2675 if (!restart &&
2676 reshape_super(st, -1, info->new_level,
2677 info->new_layout, info->new_chunk,
2678 info->array.raid_disks, info->delta_disks,
2679 backup_file, devname, APPLY_METADATA_CHANGES,
2680 quiet)) {
2681 unfreeze(st);
2682 return 1;
2683 }
2684
2685 sync_metadata(st);
2686
2687 /* ping monitor to be sure that update is on disk
2688 */
2689 ping_monitor(container);
2690
2691 switch (fork()) {
2692 case -1: /* error */
2693 perror("Cannot fork to complete reshape\n");
2694 unfreeze(st);
2695 return 1;
2696 default: /* parent */
2697 if (!freeze_reshape)
2698 printf(Name ": multi-array reshape continues"
2699 " in background\n");
2700 return 0;
2701 case 0: /* child */
2702 map_fork();
2703 break;
2704 }
2705
2706 /* close unused handle in child process
2707 */
2708 if (mdfd > -1)
2709 close(mdfd);
2710
2711 while(1) {
2712 /* For each member array with reshape_active,
2713 * we need to perform the reshape.
2714 * We pick the first array that needs reshaping and
2715 * reshape it. reshape_array() will re-read the metadata
2716 * so the next time through a different array should be
2717 * ready for reshape.
2718 * It is possible that the 'different' array will not
2719 * be assembled yet. In that case we simple exit.
2720 * When it is assembled, the mdadm which assembles it
2721 * will take over the reshape.
2722 */
2723 struct mdinfo *content;
2724 int fd;
2725 struct mdstat_ent *mdstat;
2726 char *adev;
2727
2728 sysfs_free(cc);
2729
2730 cc = st->ss->container_content(st, NULL);
2731
2732 for (content = cc; content ; content = content->next) {
2733 char *subarray;
2734 if (!content->reshape_active)
2735 continue;
2736
2737 subarray = strchr(content->text_version+1, '/')+1;
2738 mdstat = mdstat_by_subdev(subarray,
2739 devname2devnum(container));
2740 if (!mdstat)
2741 continue;
2742 if (mdstat->active == 0) {
2743 fprintf(stderr, Name ": Skipping inactive "
2744 "array md%i.\n", mdstat->devnum);
2745 free_mdstat(mdstat);
2746 mdstat = NULL;
2747 continue;
2748 }
2749 break;
2750 }
2751 if (!content)
2752 break;
2753
2754 adev = map_dev(dev2major(mdstat->devnum),
2755 dev2minor(mdstat->devnum),
2756 0);
2757 if (!adev)
2758 adev = content->text_version;
2759
2760 fd = open_dev(mdstat->devnum);
2761 if (fd < 0) {
2762 printf(Name ": Device %s cannot be opened for reshape.",
2763 adev);
2764 break;
2765 }
2766
2767 if (last_devnum == mdstat->devnum) {
2768 /* Do not allow for multiple reshape_array() calls for
2769 * the same array.
2770 * It can happen when reshape_array() returns without
2771 * error, when reshape is not finished (wrong reshape
2772 * starting/continuation conditions). Mdmon doesn't
2773 * switch to next array in container and reentry
2774 * conditions for the same array occur.
2775 * This is possibly interim until the behaviour of
2776 * reshape_array is resolved().
2777 */
2778 printf(Name ": Multiple reshape execution detected for "
2779 "device %s.", adev);
2780 close(fd);
2781 break;
2782 }
2783 last_devnum = mdstat->devnum;
2784
2785 sysfs_init(content, fd, mdstat->devnum);
2786
2787 if (mdmon_running(devname2devnum(container)))
2788 flush_mdmon(container);
2789
2790 rv = reshape_array(container, fd, adev, st,
2791 content, force, NULL,
2792 backup_file, quiet, 1, restart,
2793 freeze_reshape);
2794 close(fd);
2795
2796 if (freeze_reshape) {
2797 sysfs_free(cc);
2798 exit(0);
2799 }
2800
2801 restart = 0;
2802 if (rv)
2803 break;
2804
2805 if (mdmon_running(devname2devnum(container)))
2806 flush_mdmon(container);
2807 }
2808 if (!rv)
2809 unfreeze(st);
2810 sysfs_free(cc);
2811 exit(0);
2812 }
2813
2814 /*
2815 * We run a child process in the background which performs the following
2816 * steps:
2817 * - wait for resync to reach a certain point
2818 * - suspend io to the following section
2819 * - backup that section
2820 * - allow resync to proceed further
2821 * - resume io
2822 * - discard the backup.
2823 *
2824 * When are combined in slightly different ways in the three cases.
2825 * Grow:
2826 * - suspend/backup/allow/wait/resume/discard
2827 * Shrink:
2828 * - allow/wait/suspend/backup/allow/wait/resume/discard
2829 * same-size:
2830 * - wait/resume/discard/suspend/backup/allow
2831 *
2832 * suspend/backup/allow always come together
2833 * wait/resume/discard do too.
2834 * For the same-size case we have two backups to improve flow.
2835 *
2836 */
2837
2838 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
2839 unsigned long long backup_point,
2840 unsigned long long wait_point,
2841 unsigned long long *suspend_point,
2842 unsigned long long *reshape_completed)
2843 {
2844 /* This function is called repeatedly by the reshape manager.
2845 * It determines how much progress can safely be made and allows
2846 * that progress.
2847 * - 'info' identifies the array and particularly records in
2848 * ->reshape_progress the metadata's knowledge of progress
2849 * This is a sector offset from the start of the array
2850 * of the next array block to be relocated. This number
2851 * may increase from 0 or decrease from array_size, depending
2852 * on the type of reshape that is happening.
2853 * Note that in contrast, 'sync_completed' is a block count of the
2854 * reshape so far. It gives the distance between the start point
2855 * (head or tail of device) and the next place that data will be
2856 * written. It always increases.
2857 * - 'reshape' is the structure created by analyse_change
2858 * - 'backup_point' shows how much the metadata manager has backed-up
2859 * data. For reshapes with increasing progress, it is the next address
2860 * to be backed up, previous addresses have been backed-up. For
2861 * decreasing progress, it is the earliest address that has been
2862 * backed up - later address are also backed up.
2863 * So addresses between reshape_progress and backup_point are
2864 * backed up providing those are in the 'correct' order.
2865 * - 'wait_point' is an array address. When reshape_completed
2866 * passes this point, progress_reshape should return. It might
2867 * return earlier if it determines that ->reshape_progress needs
2868 * to be updated or further backup is needed.
2869 * - suspend_point is maintained by progress_reshape and the caller
2870 * should not touch it except to initialise to zero.
2871 * It is an array address and it only increases in 2.6.37 and earlier.
2872 * This makes it difficult to handle reducing reshapes with
2873 * external metadata.
2874 * However: it is similar to backup_point in that it records the
2875 * other end of a suspended region from reshape_progress.
2876 * it is moved to extend the region that is safe to backup and/or
2877 * reshape
2878 * - reshape_completed is read from sysfs and returned. The caller
2879 * should copy this into ->reshape_progress when it has reason to
2880 * believe that the metadata knows this, and any backup outside this
2881 * has been erased.
2882 *
2883 * Return value is:
2884 * 1 if more data from backup_point - but only as far as suspend_point,
2885 * should be backed up
2886 * 0 if things are progressing smoothly
2887 * -1 if the reshape is finished because it is all done,
2888 * -2 if the reshape is finished due to an error.
2889 */
2890
2891 int advancing = (reshape->after.data_disks
2892 >= reshape->before.data_disks);
2893 unsigned long long need_backup; /* All data between start of array and
2894 * here will at some point need to
2895 * be backed up.
2896 */
2897 unsigned long long read_offset, write_offset;
2898 unsigned long long write_range;
2899 unsigned long long max_progress, target, completed;
2900 unsigned long long array_size = (info->component_size
2901 * reshape->before.data_disks);
2902 int fd;
2903 char buf[20];
2904
2905 /* First, we unsuspend any region that is now known to be safe.
2906 * If suspend_point is on the 'wrong' side of reshape_progress, then
2907 * we don't have or need suspension at the moment. This is true for
2908 * native metadata when we don't need to back-up.
2909 */
2910 if (advancing) {
2911 if (info->reshape_progress <= *suspend_point)
2912 sysfs_set_num(info, NULL, "suspend_lo",
2913 info->reshape_progress);
2914 } else {
2915 /* Note: this won't work in 2.6.37 and before.
2916 * Something somewhere should make sure we don't need it!
2917 */
2918 if (info->reshape_progress >= *suspend_point)
2919 sysfs_set_num(info, NULL, "suspend_hi",
2920 info->reshape_progress);
2921 }
2922
2923 /* Now work out how far it is safe to progress.
2924 * If the read_offset for ->reshape_progress is less than
2925 * 'blocks' beyond the write_offset, we can only progress as far
2926 * as a backup.
2927 * Otherwise we can progress until the write_offset for the new location
2928 * reaches (within 'blocks' of) the read_offset at the current location.
2929 * However that region must be suspended unless we are using native
2930 * metadata.
2931 * If we need to suspend more, we limit it to 128M per device, which is
2932 * rather arbitrary and should be some time-based calculation.
2933 */
2934 read_offset = info->reshape_progress / reshape->before.data_disks;
2935 write_offset = info->reshape_progress / reshape->after.data_disks;
2936 write_range = info->new_chunk/512;
2937 if (reshape->before.data_disks == reshape->after.data_disks)
2938 need_backup = array_size;
2939 else
2940 need_backup = reshape->backup_blocks;
2941 if (advancing) {
2942 if (read_offset < write_offset + write_range)
2943 max_progress = backup_point;
2944 else
2945 max_progress =
2946 read_offset *
2947 reshape->after.data_disks;
2948 } else {
2949 if (read_offset > write_offset - write_range)
2950 /* Can only progress as far as has been backed up,
2951 * which must be suspended */
2952 max_progress = backup_point;
2953 else if (info->reshape_progress <= need_backup)
2954 max_progress = backup_point;
2955 else {
2956 if (info->array.major_version >= 0)
2957 /* Can progress until backup is needed */
2958 max_progress = need_backup;
2959 else {
2960 /* Can progress until metadata update is required */
2961 max_progress =
2962 read_offset *
2963 reshape->after.data_disks;
2964 /* but data must be suspended */
2965 if (max_progress < *suspend_point)
2966 max_progress = *suspend_point;
2967 }
2968 }
2969 }
2970
2971 /* We know it is safe to progress to 'max_progress' providing
2972 * it is suspended or we are using native metadata.
2973 * Consider extending suspend_point 128M per device if it
2974 * is less than 64M per device beyond reshape_progress.
2975 * But always do a multiple of 'blocks'
2976 * FIXME this is too big - it takes to long to complete
2977 * this much.
2978 */
2979 target = 64*1024*2 * min(reshape->before.data_disks,
2980 reshape->after.data_disks);
2981 target /= reshape->backup_blocks;
2982 if (target < 2)
2983 target = 2;
2984 target *= reshape->backup_blocks;
2985
2986 /* For externally managed metadata we always need to suspend IO to
2987 * the area being reshaped so we regularly push suspend_point forward.
2988 * For native metadata we only need the suspend if we are going to do
2989 * a backup.
2990 */
2991 if (advancing) {
2992 if ((need_backup > info->reshape_progress
2993 || info->array.major_version < 0) &&
2994 *suspend_point < info->reshape_progress + target) {
2995 if (need_backup < *suspend_point + 2 * target)
2996 *suspend_point = need_backup;
2997 else if (*suspend_point + 2 * target < array_size)
2998 *suspend_point += 2 * target;
2999 else
3000 *suspend_point = array_size;
3001 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3002 if (max_progress > *suspend_point)
3003 max_progress = *suspend_point;
3004 }
3005 } else {
3006 if (info->array.major_version >= 0) {
3007 /* Only need to suspend when about to backup */
3008 if (info->reshape_progress < need_backup * 2 &&
3009 *suspend_point > 0) {
3010 *suspend_point = 0;
3011 sysfs_set_num(info, NULL, "suspend_lo", 0);
3012 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3013 }
3014 } else {
3015 /* Need to suspend continually */
3016 if (info->reshape_progress < *suspend_point)
3017 *suspend_point = info->reshape_progress;
3018 if (*suspend_point + target < info->reshape_progress)
3019 /* No need to move suspend region yet */;
3020 else {
3021 if (*suspend_point >= 2 * target)
3022 *suspend_point -= 2 * target;
3023 else
3024 *suspend_point = 0;
3025 sysfs_set_num(info, NULL, "suspend_lo",
3026 *suspend_point);
3027 }
3028 if (max_progress < *suspend_point)
3029 max_progress = *suspend_point;
3030 }
3031 }
3032
3033 /* now set sync_max to allow that progress. sync_max, like
3034 * sync_completed is a count of sectors written per device, so
3035 * we find the difference between max_progress and the start point,
3036 * and divide that by after.data_disks to get a sync_max
3037 * number.
3038 * At the same time we convert wait_point to a similar number
3039 * for comparing against sync_completed.
3040 */
3041 /* scale down max_progress to per_disk */
3042 max_progress /= reshape->after.data_disks;
3043 /* Round to chunk size as some kernels give an erroneously high number */
3044 max_progress /= info->new_chunk/512;
3045 max_progress *= info->new_chunk/512;
3046 /* And round to old chunk size as the kernel wants that */
3047 max_progress /= info->array.chunk_size/512;
3048 max_progress *= info->array.chunk_size/512;
3049 /* Limit progress to the whole device */
3050 if (max_progress > info->component_size)
3051 max_progress = info->component_size;
3052 wait_point /= reshape->after.data_disks;
3053 if (!advancing) {
3054 /* switch from 'device offset' to 'processed block count' */
3055 max_progress = info->component_size - max_progress;
3056 wait_point = info->component_size - wait_point;
3057 }
3058
3059 sysfs_set_num(info, NULL, "sync_max", max_progress);
3060
3061 /* Now wait. If we have already reached the point that we were
3062 * asked to wait to, don't wait at all, else wait for any change.
3063 * We need to select on 'sync_completed' as that is the place that
3064 * notifications happen, but we are really interested in
3065 * 'reshape_position'
3066 */
3067 fd = sysfs_get_fd(info, NULL, "sync_completed");
3068 if (fd < 0)
3069 goto check_progress;
3070
3071 if (sysfs_fd_get_ll(fd, &completed) < 0)
3072 goto check_progress;
3073
3074 while (completed < max_progress && completed < wait_point) {
3075 /* Check that sync_action is still 'reshape' to avoid
3076 * waiting forever on a dead array
3077 */
3078 char action[20];
3079 fd_set rfds;
3080 if (sysfs_get_str(info, NULL, "sync_action",
3081 action, 20) <= 0 ||
3082 strncmp(action, "reshape", 7) != 0)
3083 break;
3084 /* Some kernels reset 'sync_completed' to zero
3085 * before setting 'sync_action' to 'idle'.
3086 * So we need these extra tests.
3087 */
3088 if (completed == 0 && advancing
3089 && info->reshape_progress > 0)
3090 break;
3091 if (completed == 0 && !advancing
3092 && info->reshape_progress < (info->component_size
3093 * reshape->after.data_disks))
3094 break;
3095 FD_ZERO(&rfds);
3096 FD_SET(fd, &rfds);
3097 select(fd+1, NULL, NULL, &rfds, NULL);
3098 if (sysfs_fd_get_ll(fd, &completed) < 0)
3099 goto check_progress;
3100 }
3101 /* Some kernels reset 'sync_completed' to zero,
3102 * we need to have real point we are in md
3103 */
3104 if (completed == 0)
3105 completed = max_progress;
3106
3107 /* some kernels can give an incorrectly high 'completed' number */
3108 completed /= (info->new_chunk/512);
3109 completed *= (info->new_chunk/512);
3110 /* Convert 'completed' back in to a 'progress' number */
3111 completed *= reshape->after.data_disks;
3112 if (!advancing) {
3113 completed = info->component_size * reshape->after.data_disks
3114 - completed;
3115 }
3116 *reshape_completed = completed;
3117
3118 close(fd);
3119
3120 /* We return the need_backup flag. Caller will decide
3121 * how much - a multiple of ->backup_blocks up to *suspend_point
3122 */
3123 if (advancing)
3124 return need_backup > info->reshape_progress;
3125 else
3126 return need_backup >= info->reshape_progress;
3127
3128 check_progress:
3129 /* if we couldn't read a number from sync_completed, then
3130 * either the reshape did complete, or it aborted.
3131 * We can tell which by checking for 'none' in reshape_position.
3132 * If it did abort, then it might immediately restart if it
3133 * it was just a device failure that leaves us degraded but
3134 * functioning.
3135 */
3136 strcpy(buf, "hi");
3137 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3138 || strncmp(buf, "none", 4) != 0) {
3139 /* The abort might only be temporary. Wait up to 10
3140 * seconds for fd to contain a valid number again.
3141 */
3142 struct timeval tv;
3143 int rv = -2;
3144 tv.tv_sec = 10;
3145 tv.tv_usec = 0;
3146 while (fd >= 0 && rv < 0 && tv.tv_sec > 0) {
3147 fd_set rfds;
3148 FD_ZERO(&rfds);
3149 FD_SET(fd, &rfds);
3150 if (select(fd+1, NULL, NULL, &rfds, &tv) != 1)
3151 break;
3152 switch (sysfs_fd_get_ll(fd, &completed)) {
3153 case 0:
3154 /* all good again */
3155 rv = 1;
3156 break;
3157 case -2: /* read error - abort */
3158 tv.tv_sec = 0;
3159 break;
3160 }
3161 }
3162 if (fd >= 0)
3163 close(fd);
3164 return rv; /* abort */
3165 } else {
3166 /* Maybe racing with array shutdown - check state */
3167 if (fd >= 0)
3168 close(fd);
3169 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3170 || strncmp(buf, "inactive", 8) == 0
3171 || strncmp(buf, "clear",5) == 0)
3172 return -2; /* abort */
3173 return -1; /* complete */
3174 }
3175 }
3176
3177 /* FIXME return status is never checked */
3178 static int grow_backup(struct mdinfo *sra,
3179 unsigned long long offset, /* per device */
3180 unsigned long stripes, /* per device, in old chunks */
3181 int *sources, unsigned long long *offsets,
3182 int disks, int chunk, int level, int layout,
3183 int dests, int *destfd, unsigned long long *destoffsets,
3184 int part, int *degraded,
3185 char *buf)
3186 {
3187 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3188 * to storage 'destfd' (offset 'destoffsets'), after first
3189 * suspending IO. Then allow resync to continue
3190 * over the suspended section.
3191 * Use part 'part' of the backup-super-block.
3192 */
3193 int odata = disks;
3194 int rv = 0;
3195 int i;
3196 unsigned long long ll;
3197 int new_degraded;
3198 //printf("offset %llu\n", offset);
3199 if (level >= 4)
3200 odata--;
3201 if (level == 6)
3202 odata--;
3203
3204 /* Check that array hasn't become degraded, else we might backup the wrong data */
3205 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3206 return -1; /* FIXME this error is ignored */
3207 new_degraded = (int)ll;
3208 if (new_degraded != *degraded) {
3209 /* check each device to ensure it is still working */
3210 struct mdinfo *sd;
3211 for (sd = sra->devs ; sd ; sd = sd->next) {
3212 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3213 continue;
3214 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3215 char sbuf[20];
3216 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3217 strstr(sbuf, "faulty") ||
3218 strstr(sbuf, "in_sync") == NULL) {
3219 /* this device is dead */
3220 sd->disk.state = (1<<MD_DISK_FAULTY);
3221 if (sd->disk.raid_disk >= 0 &&
3222 sources[sd->disk.raid_disk] >= 0) {
3223 close(sources[sd->disk.raid_disk]);
3224 sources[sd->disk.raid_disk] = -1;
3225 }
3226 }
3227 }
3228 }
3229 *degraded = new_degraded;
3230 }
3231 if (part) {
3232 bsb.arraystart2 = __cpu_to_le64(offset * odata);
3233 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
3234 } else {
3235 bsb.arraystart = __cpu_to_le64(offset * odata);
3236 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
3237 }
3238 if (part)
3239 bsb.magic[15] = '2';
3240 for (i = 0; i < dests; i++)
3241 if (part)
3242 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
3243 else
3244 lseek64(destfd[i], destoffsets[i], 0);
3245
3246 rv = save_stripes(sources, offsets,
3247 disks, chunk, level, layout,
3248 dests, destfd,
3249 offset*512*odata, stripes * chunk * odata,
3250 buf);
3251
3252 if (rv)
3253 return rv;
3254 bsb.mtime = __cpu_to_le64(time(0));
3255 for (i = 0; i < dests; i++) {
3256 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3257
3258 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3259 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3260 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3261 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3262
3263 rv = -1;
3264 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
3265 != destoffsets[i] - 4096)
3266 break;
3267 if (write(destfd[i], &bsb, 512) != 512)
3268 break;
3269 if (destoffsets[i] > 4096) {
3270 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
3271 destoffsets[i]+stripes*chunk*odata)
3272 break;
3273 if (write(destfd[i], &bsb, 512) != 512)
3274 break;
3275 }
3276 fsync(destfd[i]);
3277 rv = 0;
3278 }
3279
3280 return rv;
3281 }
3282
3283 /* in 2.6.30, the value reported by sync_completed can be
3284 * less that it should be by one stripe.
3285 * This only happens when reshape hits sync_max and pauses.
3286 * So allow wait_backup to either extent sync_max further
3287 * than strictly necessary, or return before the
3288 * sync has got quite as far as we would really like.
3289 * This is what 'blocks2' is for.
3290 * The various caller give appropriate values so that
3291 * every works.
3292 */
3293 /* FIXME return value is often ignored */
3294 static int forget_backup(int dests, int *destfd,
3295 unsigned long long *destoffsets,
3296 int part)
3297 {
3298 /*
3299 * Erase backup 'part' (which is 0 or 1)
3300 */
3301 int i;
3302 int rv;
3303
3304 if (part) {
3305 bsb.arraystart2 = __cpu_to_le64(0);
3306 bsb.length2 = __cpu_to_le64(0);
3307 } else {
3308 bsb.arraystart = __cpu_to_le64(0);
3309 bsb.length = __cpu_to_le64(0);
3310 }
3311 bsb.mtime = __cpu_to_le64(time(0));
3312 rv = 0;
3313 for (i = 0; i < dests; i++) {
3314 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3315 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3316 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3317 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3318 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3319 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
3320 destoffsets[i]-4096)
3321 rv = -1;
3322 if (rv == 0 &&
3323 write(destfd[i], &bsb, 512) != 512)
3324 rv = -1;
3325 fsync(destfd[i]);
3326 }
3327 return rv;
3328 }
3329
3330 static void fail(char *msg)
3331 {
3332 int rv;
3333 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
3334 rv |= (write(2, "\n", 1) != 1);
3335 exit(rv ? 1 : 2);
3336 }
3337
3338 static char *abuf, *bbuf;
3339 static unsigned long long abuflen;
3340 static void validate(int afd, int bfd, unsigned long long offset)
3341 {
3342 /* check that the data in the backup against the array.
3343 * This is only used for regression testing and should not
3344 * be used while the array is active
3345 */
3346 if (afd < 0)
3347 return;
3348 lseek64(bfd, offset - 4096, 0);
3349 if (read(bfd, &bsb2, 512) != 512)
3350 fail("cannot read bsb");
3351 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
3352 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
3353 fail("first csum bad");
3354 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
3355 fail("magic is bad");
3356 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
3357 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
3358 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
3359 fail("second csum bad");
3360
3361 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
3362 fail("devstart is wrong");
3363
3364 if (bsb2.length) {
3365 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
3366
3367 if (abuflen < len) {
3368 free(abuf);
3369 free(bbuf);
3370 abuflen = len;
3371 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
3372 posix_memalign((void**)&bbuf, 4096, abuflen)) {
3373 abuflen = 0;
3374 /* just stop validating on mem-alloc failure */
3375 return;
3376 }
3377 }
3378
3379 lseek64(bfd, offset, 0);
3380 if ((unsigned long long)read(bfd, bbuf, len) != len) {
3381 //printf("len %llu\n", len);
3382 fail("read first backup failed");
3383 }
3384 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
3385 if ((unsigned long long)read(afd, abuf, len) != len)
3386 fail("read first from array failed");
3387 if (memcmp(bbuf, abuf, len) != 0) {
3388 #if 0
3389 int i;
3390 printf("offset=%llu len=%llu\n",
3391 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
3392 for (i=0; i<len; i++)
3393 if (bbuf[i] != abuf[i]) {
3394 printf("first diff byte %d\n", i);
3395 break;
3396 }
3397 #endif
3398 fail("data1 compare failed");
3399 }
3400 }
3401 if (bsb2.length2) {
3402 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
3403
3404 if (abuflen < len) {
3405 free(abuf);
3406 free(bbuf);
3407 abuflen = len;
3408 abuf = malloc(abuflen);
3409 bbuf = malloc(abuflen);
3410 }
3411
3412 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
3413 if ((unsigned long long)read(bfd, bbuf, len) != len)
3414 fail("read second backup failed");
3415 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
3416 if ((unsigned long long)read(afd, abuf, len) != len)
3417 fail("read second from array failed");
3418 if (memcmp(bbuf, abuf, len) != 0)
3419 fail("data2 compare failed");
3420 }
3421 }
3422
3423 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
3424 struct supertype *st, unsigned long blocks,
3425 int *fds, unsigned long long *offsets,
3426 int dests, int *destfd, unsigned long long *destoffsets)
3427 {
3428 /* Monitor a reshape where backup is being performed using
3429 * 'native' mechanism - either to a backup file, or
3430 * to some space in a spare.
3431 */
3432 char *buf;
3433 int degraded = -1;
3434 unsigned long long speed;
3435 unsigned long long suspend_point, array_size;
3436 unsigned long long backup_point, wait_point;
3437 unsigned long long reshape_completed;
3438 int done = 0;
3439 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
3440 int part = 0; /* The next part of the backup area to fill. It may already
3441 * be full, so we need to check */
3442 int level = reshape->level;
3443 int layout = reshape->before.layout;
3444 int data = reshape->before.data_disks;
3445 int disks = reshape->before.data_disks + reshape->parity;
3446 int chunk = sra->array.chunk_size;
3447 struct mdinfo *sd;
3448 unsigned long stripes;
3449 int uuid[4];
3450
3451 /* set up the backup-super-block. This requires the
3452 * uuid from the array.
3453 */
3454 /* Find a superblock */
3455 for (sd = sra->devs; sd; sd = sd->next) {
3456 char *dn;
3457 int devfd;
3458 int ok;
3459 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3460 continue;
3461 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
3462 devfd = dev_open(dn, O_RDONLY);
3463 if (devfd < 0)
3464 continue;
3465 ok = st->ss->load_super(st, devfd, NULL);
3466 close(devfd);
3467 if (ok == 0)
3468 break;
3469 }
3470 if (!sd) {
3471 fprintf(stderr, Name ": Cannot find a superblock\n");
3472 return 0;
3473 }
3474
3475 memset(&bsb, 0, 512);
3476 memcpy(bsb.magic, "md_backup_data-1", 16);
3477 st->ss->uuid_from_super(st, uuid);
3478 memcpy(bsb.set_uuid, uuid, 16);
3479 bsb.mtime = __cpu_to_le64(time(0));
3480 bsb.devstart2 = blocks;
3481
3482 stripes = blocks / (sra->array.chunk_size/512) /
3483 reshape->before.data_disks;
3484
3485 if (posix_memalign((void**)&buf, 4096, disks * chunk))
3486 /* Don't start the 'reshape' */
3487 return 0;
3488 if (reshape->before.data_disks == reshape->after.data_disks) {
3489 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
3490 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
3491 }
3492
3493 if (increasing) {
3494 array_size = sra->component_size * reshape->after.data_disks;
3495 backup_point = sra->reshape_progress;
3496 suspend_point = 0;
3497 } else {
3498 array_size = sra->component_size * reshape->before.data_disks;
3499 backup_point = reshape->backup_blocks;
3500 suspend_point = array_size;
3501 }
3502
3503 while (!done) {
3504 int rv;
3505
3506 /* Want to return as soon the oldest backup slot can
3507 * be released as that allows us to start backing up
3508 * some more, providing suspend_point has been
3509 * advanced, which it should have.
3510 */
3511 if (increasing) {
3512 wait_point = array_size;
3513 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
3514 wait_point = (__le64_to_cpu(bsb.arraystart) +
3515 __le64_to_cpu(bsb.length));
3516 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
3517 wait_point = (__le64_to_cpu(bsb.arraystart2) +
3518 __le64_to_cpu(bsb.length2));
3519 } else {
3520 wait_point = 0;
3521 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
3522 wait_point = __le64_to_cpu(bsb.arraystart);
3523 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
3524 wait_point = __le64_to_cpu(bsb.arraystart2);
3525 }
3526
3527 rv = progress_reshape(sra, reshape,
3528 backup_point, wait_point,
3529 &suspend_point, &reshape_completed);
3530 /* external metadata would need to ping_monitor here */
3531 sra->reshape_progress = reshape_completed;
3532
3533 /* Clear any backup region that is before 'here' */
3534 if (increasing) {
3535 if (__le64_to_cpu(bsb.length) > 0 &&
3536 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
3537 __le64_to_cpu(bsb.length)))
3538 forget_backup(dests, destfd,
3539 destoffsets, 0);
3540 if (__le64_to_cpu(bsb.length2) > 0 &&
3541 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
3542 __le64_to_cpu(bsb.length2)))
3543 forget_backup(dests, destfd,
3544 destoffsets, 1);
3545 } else {
3546 if (__le64_to_cpu(bsb.length) > 0 &&
3547 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
3548 forget_backup(dests, destfd,
3549 destoffsets, 0);
3550 if (__le64_to_cpu(bsb.length2) > 0 &&
3551 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
3552 forget_backup(dests, destfd,
3553 destoffsets, 1);
3554 }
3555
3556 if (rv < 0) {
3557 if (rv == -1)
3558 done = 1;
3559 break;
3560 }
3561 if (rv == 0 && increasing && !st->ss->external) {
3562 /* No longer need to monitor this reshape */
3563 done = 1;
3564 break;
3565 }
3566
3567 while (rv) {
3568 unsigned long long offset;
3569 unsigned long actual_stripes;
3570 /* Need to backup some data.
3571 * If 'part' is not used and the desired
3572 * backup size is suspended, do a backup,
3573 * then consider the next part.
3574 */
3575 /* Check that 'part' is unused */
3576 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
3577 break;
3578 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
3579 break;
3580
3581 offset = backup_point / data;
3582 actual_stripes = stripes;
3583 if (increasing) {
3584 if (offset + actual_stripes * (chunk/512) >
3585 sra->component_size)
3586 actual_stripes = ((sra->component_size - offset)
3587 / (chunk/512));
3588 if (offset + actual_stripes * (chunk/512) >
3589 suspend_point/data)
3590 break;
3591 } else {
3592 if (offset < actual_stripes * (chunk/512))
3593 actual_stripes = offset / (chunk/512);
3594 offset -= actual_stripes * (chunk/512);
3595 if (offset < suspend_point/data)
3596 break;
3597 }
3598 if (actual_stripes == 0)
3599 break;
3600 grow_backup(sra, offset, actual_stripes,
3601 fds, offsets,
3602 disks, chunk, level, layout,
3603 dests, destfd, destoffsets,
3604 part, &degraded, buf);
3605 validate(afd, destfd[0], destoffsets[0]);
3606 /* record where 'part' is up to */
3607 part = !part;
3608 if (increasing)
3609 backup_point += actual_stripes * (chunk/512) * data;
3610 else
3611 backup_point -= actual_stripes * (chunk/512) * data;
3612 }
3613 }
3614
3615 /* FIXME maybe call progress_reshape one more time instead */
3616 abort_reshape(sra); /* remove any remaining suspension */
3617 if (reshape->before.data_disks == reshape->after.data_disks)
3618 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
3619 free(buf);
3620 return done;
3621 }
3622
3623 /*
3624 * If any spare contains md_back_data-1 which is recent wrt mtime,
3625 * write that data into the array and update the super blocks with
3626 * the new reshape_progress
3627 */
3628 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
3629 char *backup_file, int verbose)
3630 {
3631 int i, j;
3632 int old_disks;
3633 unsigned long long *offsets;
3634 unsigned long long nstripe, ostripe;
3635 int ndata, odata;
3636
3637 odata = info->array.raid_disks - info->delta_disks - 1;
3638 if (info->array.level == 6) odata--; /* number of data disks */
3639 ndata = info->array.raid_disks - 1;
3640 if (info->new_level == 6) ndata--;
3641
3642 old_disks = info->array.raid_disks - info->delta_disks;
3643
3644 if (info->delta_disks <= 0)
3645 /* Didn't grow, so the backup file must have
3646 * been used
3647 */
3648 old_disks = cnt;
3649 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
3650 struct mdinfo dinfo;
3651 int fd;
3652 int bsbsize;
3653 char *devname, namebuf[20];
3654 unsigned long long lo, hi;
3655
3656 /* This was a spare and may have some saved data on it.
3657 * Load the superblock, find and load the
3658 * backup_super_block.
3659 * If either fail, go on to next device.
3660 * If the backup contains no new info, just return
3661 * else restore data and update all superblocks
3662 */
3663 if (i == old_disks-1) {
3664 fd = open(backup_file, O_RDONLY);
3665 if (fd<0) {
3666 fprintf(stderr, Name ": backup file %s inaccessible: %s\n",
3667 backup_file, strerror(errno));
3668 continue;
3669 }
3670 devname = backup_file;
3671 } else {
3672 fd = fdlist[i];
3673 if (fd < 0)
3674 continue;
3675 if (st->ss->load_super(st, fd, NULL))
3676 continue;
3677
3678 st->ss->getinfo_super(st, &dinfo, NULL);
3679 st->ss->free_super(st);
3680
3681 if (lseek64(fd,
3682 (dinfo.data_offset + dinfo.component_size - 8) <<9,
3683 0) < 0) {
3684 fprintf(stderr, Name ": Cannot seek on device %d\n", i);
3685 continue; /* Cannot seek */
3686 }
3687 sprintf(namebuf, "device-%d", i);
3688 devname = namebuf;
3689 }
3690 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
3691 if (verbose)
3692 fprintf(stderr, Name ": Cannot read from %s\n", devname);
3693 continue; /* Cannot read */
3694 }
3695 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
3696 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
3697 if (verbose)
3698 fprintf(stderr, Name ": No backup metadata on %s\n", devname);
3699 continue;
3700 }
3701 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
3702 if (verbose)
3703 fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname);
3704 continue; /* bad checksum */
3705 }
3706 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
3707 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
3708 if (verbose)
3709 fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname);
3710 continue; /* Bad second checksum */
3711 }
3712 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
3713 if (verbose)
3714 fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname);
3715 continue; /* Wrong uuid */
3716 }
3717
3718 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3719 * sometimes they aren't... So allow considerable flexability in matching, and allow
3720 * this test to be overridden by an environment variable.
3721 */
3722 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
3723 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
3724 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3725 fprintf(stderr, Name ": accepting backup with timestamp %lu "
3726 "for array with timestamp %lu\n",
3727 (unsigned long)__le64_to_cpu(bsb.mtime),
3728 (unsigned long)info->array.utime);
3729 } else {
3730 if (verbose)
3731 fprintf(stderr, Name ": too-old timestamp on "
3732 "backup-metadata on %s\n", devname);
3733 continue; /* time stamp is too bad */
3734 }
3735 }
3736
3737 if (bsb.magic[15] == '1') {
3738 if (bsb.length == 0)
3739 continue;
3740 if (info->delta_disks >= 0) {
3741 /* reshape_progress is increasing */
3742 if (__le64_to_cpu(bsb.arraystart)
3743 + __le64_to_cpu(bsb.length)
3744 < info->reshape_progress) {
3745 nonew:
3746 if (verbose)
3747 fprintf(stderr, Name
3748 ": backup-metadata found on %s but is not needed\n", devname);
3749 continue; /* No new data here */
3750 }
3751 } else {
3752 /* reshape_progress is decreasing */
3753 if (__le64_to_cpu(bsb.arraystart) >=
3754 info->reshape_progress)
3755 goto nonew; /* No new data here */
3756 }
3757 } else {
3758 if (bsb.length == 0 && bsb.length2 == 0)
3759 continue;
3760 if (info->delta_disks >= 0) {
3761 /* reshape_progress is increasing */
3762 if ((__le64_to_cpu(bsb.arraystart)
3763 + __le64_to_cpu(bsb.length)
3764 < info->reshape_progress)
3765 &&
3766 (__le64_to_cpu(bsb.arraystart2)
3767 + __le64_to_cpu(bsb.length2)
3768 < info->reshape_progress))
3769 goto nonew; /* No new data here */
3770 } else {
3771 /* reshape_progress is decreasing */
3772 if (__le64_to_cpu(bsb.arraystart) >=
3773 info->reshape_progress &&
3774 __le64_to_cpu(bsb.arraystart2) >=
3775 info->reshape_progress)
3776 goto nonew; /* No new data here */
3777 }
3778 }
3779 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
3780 second_fail:
3781 if (verbose)
3782 fprintf(stderr, Name
3783 ": Failed to verify secondary backup-metadata block on %s\n",
3784 devname);
3785 continue; /* Cannot seek */
3786 }
3787 /* There should be a duplicate backup superblock 4k before here */
3788 if (lseek64(fd, -4096, 1) < 0 ||
3789 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
3790 goto second_fail; /* Cannot find leading superblock */
3791 if (bsb.magic[15] == '1')
3792 bsbsize = offsetof(struct mdp_backup_super, pad1);
3793 else
3794 bsbsize = offsetof(struct mdp_backup_super, pad);
3795 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
3796 goto second_fail; /* Cannot find leading superblock */
3797
3798 /* Now need the data offsets for all devices. */
3799 offsets = malloc(sizeof(*offsets)*info->array.raid_disks);
3800 for(j=0; j<info->array.raid_disks; j++) {
3801 if (fdlist[j] < 0)
3802 continue;
3803 if (st->ss->load_super(st, fdlist[j], NULL))
3804 /* FIXME should be this be an error */
3805 continue;
3806 st->ss->getinfo_super(st, &dinfo, NULL);
3807 st->ss->free_super(st);
3808 offsets[j] = dinfo.data_offset * 512;
3809 }
3810 printf(Name ": restoring critical section\n");
3811
3812 if (restore_stripes(fdlist, offsets,
3813 info->array.raid_disks,
3814 info->new_chunk,
3815 info->new_level,
3816 info->new_layout,
3817 fd, __le64_to_cpu(bsb.devstart)*512,
3818 __le64_to_cpu(bsb.arraystart)*512,
3819 __le64_to_cpu(bsb.length)*512, NULL)) {
3820 /* didn't succeed, so giveup */
3821 if (verbose)
3822 fprintf(stderr, Name ": Error restoring backup from %s\n",
3823 devname);
3824 free(offsets);
3825 return 1;
3826 }
3827
3828 if (bsb.magic[15] == '2' &&
3829 restore_stripes(fdlist, offsets,
3830 info->array.raid_disks,
3831 info->new_chunk,
3832 info->new_level,
3833 info->new_layout,
3834 fd, __le64_to_cpu(bsb.devstart)*512 +
3835 __le64_to_cpu(bsb.devstart2)*512,
3836 __le64_to_cpu(bsb.arraystart2)*512,
3837 __le64_to_cpu(bsb.length2)*512, NULL)) {
3838 /* didn't succeed, so giveup */
3839 if (verbose)
3840 fprintf(stderr, Name ": Error restoring second backup from %s\n",
3841 devname);
3842 free(offsets);
3843 return 1;
3844 }
3845
3846 free(offsets);
3847
3848 /* Ok, so the data is restored. Let's update those superblocks. */
3849
3850 lo = hi = 0;
3851 if (bsb.length) {
3852 lo = __le64_to_cpu(bsb.arraystart);
3853 hi = lo + __le64_to_cpu(bsb.length);
3854 }
3855 if (bsb.magic[15] == '2' && bsb.length2) {
3856 unsigned long long lo1, hi1;
3857 lo1 = __le64_to_cpu(bsb.arraystart2);
3858 hi1 = lo1 + __le64_to_cpu(bsb.length2);
3859 if (lo == hi) {
3860 lo = lo1;
3861 hi = hi1;
3862 } else if (lo < lo1)
3863 hi = hi1;
3864 else
3865 lo = lo1;
3866 }
3867 if (lo < hi &&
3868 (info->reshape_progress < lo ||
3869 info->reshape_progress > hi))
3870 /* backup does not affect reshape_progress*/ ;
3871 else if (info->delta_disks >= 0) {
3872 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
3873 __le64_to_cpu(bsb.length);
3874 if (bsb.magic[15] == '2') {
3875 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
3876 __le64_to_cpu(bsb.length2);
3877 if (p2 > info->reshape_progress)
3878 info->reshape_progress = p2;
3879 }
3880 } else {
3881 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
3882 if (bsb.magic[15] == '2') {
3883 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
3884 if (p2 < info->reshape_progress)
3885 info->reshape_progress = p2;
3886 }
3887 }
3888 for (j=0; j<info->array.raid_disks; j++) {
3889 if (fdlist[j] < 0) continue;
3890 if (st->ss->load_super(st, fdlist[j], NULL))
3891 continue;
3892 st->ss->getinfo_super(st, &dinfo, NULL);
3893 dinfo.reshape_progress = info->reshape_progress;
3894 st->ss->update_super(st, &dinfo,
3895 "_reshape_progress",
3896 NULL,0, 0, NULL);
3897 st->ss->store_super(st, fdlist[j]);
3898 st->ss->free_super(st);
3899 }
3900 return 0;
3901 }
3902 /* Didn't find any backup data, try to see if any
3903 * was needed.
3904 */
3905 if (info->delta_disks < 0) {
3906 /* When shrinking, the critical section is at the end.
3907 * So see if we are before the critical section.
3908 */
3909 unsigned long long first_block;
3910 nstripe = ostripe = 0;
3911 first_block = 0;
3912 while (ostripe >= nstripe) {
3913 ostripe += info->array.chunk_size / 512;
3914 first_block = ostripe * odata;
3915 nstripe = first_block / ndata / (info->new_chunk/512) *
3916 (info->new_chunk/512);
3917 }
3918
3919 if (info->reshape_progress >= first_block)
3920 return 0;
3921 }
3922 if (info->delta_disks > 0) {
3923 /* See if we are beyond the critical section. */
3924 unsigned long long last_block;
3925 nstripe = ostripe = 0;
3926 last_block = 0;
3927 while (nstripe >= ostripe) {
3928 nstripe += info->new_chunk / 512;
3929 last_block = nstripe * ndata;
3930 ostripe = last_block / odata / (info->array.chunk_size/512) *
3931 (info->array.chunk_size/512);
3932 }
3933
3934 if (info->reshape_progress >= last_block)
3935 return 0;
3936 }
3937 /* needed to recover critical section! */
3938 if (verbose)
3939 fprintf(stderr, Name ": Failed to find backup of critical section\n");
3940 return 1;
3941 }
3942
3943 int Grow_continue_command(char *devname, int fd,
3944 char *backup_file, int verbose)
3945 {
3946 int ret_val = 0;
3947 struct supertype *st = NULL;
3948 struct mdinfo *content = NULL;
3949 struct mdinfo array;
3950 char *subarray = NULL;
3951 struct mdinfo *cc = NULL;
3952 struct mdstat_ent *mdstat = NULL;
3953 char buf[40];
3954 int cfd = -1;
3955 int fd2 = -1;
3956
3957 dprintf("Grow continue from command line called for %s\n",
3958 devname);
3959
3960 st = super_by_fd(fd, &subarray);
3961 if (!st || !st->ss) {
3962 fprintf(stderr,
3963 Name ": Unable to determine metadata format for %s\n",
3964 devname);
3965 return 1;
3966 }
3967 dprintf("Grow continue is run for ");
3968 if (st->ss->external == 0) {
3969 dprintf("native array (%s)\n", devname);
3970 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
3971 fprintf(stderr, Name ": %s is not an active md array -"
3972 " aborting\n", devname);
3973 ret_val = 1;
3974 goto Grow_continue_command_exit;
3975 }
3976 content = &array;
3977 sysfs_init(content, fd, st->devnum);
3978 } else {
3979 int container_dev;
3980
3981 if (subarray) {
3982 dprintf("subarray (%s)\n", subarray);
3983 container_dev = st->container_dev;
3984 cfd = open_dev_excl(st->container_dev);
3985 } else {
3986 container_dev = st->devnum;
3987 close(fd);
3988 cfd = open_dev_excl(st->devnum);
3989 dprintf("container (%i)\n", container_dev);
3990 fd = cfd;
3991 }
3992 if (cfd < 0) {
3993 fprintf(stderr, Name ": Unable to open container "
3994 "for %s\n", devname);
3995 ret_val = 1;
3996 goto Grow_continue_command_exit;
3997 }
3998 fmt_devname(buf, container_dev);
3999
4000 /* find in container array under reshape
4001 */
4002 ret_val = st->ss->load_container(st, cfd, NULL);
4003 if (ret_val) {
4004 fprintf(stderr,
4005 Name ": Cannot read superblock for %s\n",
4006 devname);
4007 ret_val = 1;
4008 goto Grow_continue_command_exit;
4009 }
4010
4011 cc = st->ss->container_content(st, subarray);
4012 for (content = cc; content ; content = content->next) {
4013 char *array;
4014 int allow_reshape = 1;
4015
4016 if (content->reshape_active == 0)
4017 continue;
4018 /* The decision about array or container wide
4019 * reshape is taken in Grow_continue based
4020 * content->reshape_active state, therefore we
4021 * need to check_reshape based on
4022 * reshape_active and subarray name
4023 */
4024 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4025 allow_reshape = 0;
4026 if (content->reshape_active == CONTAINER_RESHAPE &&
4027 (content->array.state
4028 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4029 allow_reshape = 0;
4030
4031 if (!allow_reshape) {
4032 fprintf(stderr, Name
4033 ": cannot continue reshape of an array"
4034 " in container with unsupported"
4035 " metadata: %s(%s)\n",
4036 devname, buf);
4037 ret_val = 1;
4038 goto Grow_continue_command_exit;
4039 }
4040
4041 array = strchr(content->text_version+1, '/')+1;
4042 mdstat = mdstat_by_subdev(array, container_dev);
4043 if (!mdstat)
4044 continue;
4045 if (mdstat->active == 0) {
4046 fprintf(stderr, Name ": Skipping inactive "
4047 "array md%i.\n", mdstat->devnum);
4048 free_mdstat(mdstat);
4049 mdstat = NULL;
4050 continue;
4051 }
4052 break;
4053 }
4054 if (!content) {
4055 fprintf(stderr,
4056 Name ": Unable to determine reshaped "
4057 "array for %s\n", devname);
4058 ret_val = 1;
4059 goto Grow_continue_command_exit;
4060 }
4061 fd2 = open_dev(mdstat->devnum);
4062 if (fd2 < 0) {
4063 fprintf(stderr, Name ": cannot open (md%i)\n",
4064 mdstat->devnum);
4065 ret_val = 1;
4066 goto Grow_continue_command_exit;
4067 }
4068
4069 sysfs_init(content, fd2, mdstat->devnum);
4070
4071 /* start mdmon in case it is not running
4072 */
4073 if (!mdmon_running(container_dev))
4074 start_mdmon(container_dev);
4075 ping_monitor(buf);
4076
4077 if (mdmon_running(container_dev))
4078 st->update_tail = &st->updates;
4079 else {
4080 fprintf(stderr, Name ": No mdmon found. "
4081 "Grow cannot continue.\n");
4082 ret_val = 1;
4083 goto Grow_continue_command_exit;
4084 }
4085 }
4086
4087 /* verify that array under reshape is started from
4088 * correct position
4089 */
4090 if (verify_reshape_position(content,
4091 map_name(pers, mdstat->level)) < 0) {
4092 ret_val = 1;
4093 goto Grow_continue_command_exit;
4094 }
4095
4096 /* continue reshape
4097 */
4098 ret_val = Grow_continue(fd, st, content, backup_file, 0);
4099
4100 Grow_continue_command_exit:
4101 if (fd2 > -1)
4102 close(fd2);
4103 if (cfd > -1)
4104 close(cfd);
4105 st->ss->free_super(st);
4106 free_mdstat(mdstat);
4107 sysfs_free(cc);
4108 free(subarray);
4109
4110 return ret_val;
4111 }
4112
4113 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4114 char *backup_file, int freeze_reshape)
4115 {
4116 int ret_val = 2;
4117
4118 if (!info->reshape_active)
4119 return ret_val;
4120
4121 if (st->ss->external) {
4122 char container[40];
4123 int cfd = open_dev(st->container_dev);
4124
4125 if (cfd < 0)
4126 return 1;
4127
4128 fmt_devname(container, st->container_dev);
4129 st->ss->load_container(st, cfd, container);
4130 close(cfd);
4131 ret_val = reshape_container(container, NULL, mdfd,
4132 st, info, 0, backup_file,
4133 0, 1, freeze_reshape);
4134 } else
4135 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4136 NULL, backup_file, 0, 0, 1,
4137 freeze_reshape);
4138
4139 return ret_val;
4140 }
Unnamed repository; edit this file 'description' to name the repository.