]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
Remove lots of unnecessary white space.
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2012 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27 #include <stdint.h>
28
29 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
30 #error no endian defined
31 #endif
32 #include "md_u.h"
33 #include "md_p.h"
34
35 #ifndef offsetof
36 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
37 #endif
38
39 int restore_backup(struct supertype *st,
40 struct mdinfo *content,
41 int working_disks,
42 int next_spare,
43 char *backup_file,
44 int verbose)
45 {
46 int i;
47 int *fdlist;
48 struct mdinfo *dev;
49 int err;
50 int disk_count = next_spare + working_disks;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61 sprintf(buf, "%d:%d",
62 dev->disk.major,
63 dev->disk.minor);
64 fd = dev_open(buf, O_RDWR);
65
66 if (dev->disk.raid_disk >= 0)
67 fdlist[dev->disk.raid_disk] = fd;
68 else
69 fdlist[next_spare++] = fd;
70 }
71
72 if (st->ss->external && st->ss->recover_backup)
73 err = st->ss->recover_backup(st, content);
74 else
75 err = Grow_restart(st, content, fdlist, next_spare,
76 backup_file, verbose > 0);
77
78 while (next_spare > 0) {
79 next_spare--;
80 if (fdlist[next_spare] >= 0)
81 close(fdlist[next_spare]);
82 }
83 free(fdlist);
84 if (err) {
85 pr_err("Failed to restore critical"
86 " section for reshape - sorry.\n");
87 if (!backup_file)
88 pr_err("Possibly you need"
89 " to specify a --backup-file\n");
90 return 1;
91 }
92
93 dprintf("restore_backup() returns status OK.\n");
94 return 0;
95 }
96
97 int Grow_Add_device(char *devname, int fd, char *newdev)
98 {
99 /* Add a device to an active array.
100 * Currently, just extend a linear array.
101 * This requires writing a new superblock on the
102 * new device, calling the kernel to add the device,
103 * and if that succeeds, update the superblock on
104 * all other devices.
105 * This means that we need to *find* all other devices.
106 */
107 struct mdinfo info;
108
109 struct stat stb;
110 int nfd, fd2;
111 int d, nd;
112 struct supertype *st = NULL;
113 char *subarray = NULL;
114
115 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
116 pr_err("cannot get array info for %s\n", devname);
117 return 1;
118 }
119
120 if (info.array.level != -1) {
121 pr_err("can only add devices to linear arrays\n");
122 return 1;
123 }
124
125 st = super_by_fd(fd, &subarray);
126 if (!st) {
127 pr_err("cannot handle arrays with superblock version %d\n",
128 info.array.major_version);
129 return 1;
130 }
131
132 if (subarray) {
133 pr_err("Cannot grow linear sub-arrays yet\n");
134 free(subarray);
135 free(st);
136 return 1;
137 }
138
139 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
140 if (nfd < 0) {
141 pr_err("cannot open %s\n", newdev);
142 free(st);
143 return 1;
144 }
145 fstat(nfd, &stb);
146 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
147 pr_err("%s is not a block device!\n", newdev);
148 close(nfd);
149 free(st);
150 return 1;
151 }
152 /* now check out all the devices and make sure we can read the
153 * 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 pr_err("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 pr_err("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 pr_err("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 pr_err("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 pr_err("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 pr_err("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 pr_err("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 pr_err("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 pr_err("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 pr_err("cannot open device file %s\n", dv);
246 return 1;
247 }
248 if (st->ss->load_super(st, fd2, NULL)) {
249 pr_err("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 pr_err("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, struct context *c, struct shape *s)
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 pr_err("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 pr_err("Memory allocation failure.\n");
302 else
303 pr_err("bitmaps not supported by this kernel.\n");
304 return 1;
305 }
306 if (bmf.pathname[0]) {
307 if (strcmp(s->bitmap_file,"none")==0) {
308 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
309 pr_err("failed to remove bitmap %s\n",
310 bmf.pathname);
311 return 1;
312 }
313 return 0;
314 }
315 pr_err("%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 pr_err("cannot get array status for %s\n", devname);
321 return 1;
322 }
323 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
324 if (strcmp(s->bitmap_file, "none")==0) {
325 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
326 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
327 pr_err("failed to remove internal bitmap.\n");
328 return 1;
329 }
330 return 0;
331 }
332 pr_err("Internal bitmap already present on %s\n",
333 devname);
334 return 1;
335 }
336
337 if (strcmp(s->bitmap_file, "none") == 0) {
338 pr_err("no bitmap found on %s\n", devname);
339 return 1;
340 }
341 if (array.level <= 0) {
342 pr_err("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 pr_err("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 pr_err("Cannot understand version %d.%d\n",
368 array.major_version, array.minor_version);
369 return 1;
370 }
371 if (subarray) {
372 pr_err("Cannot add bitmaps to sub-arrays yet\n");
373 free(subarray);
374 free(st);
375 return 1;
376 }
377 if (strcmp(s->bitmap_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 pr_err("Internal bitmaps not supported "
384 "with %s metadata\n", st->ss->name);
385 return 1;
386 }
387 mdi = sysfs_read(fd, NULL, 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 &s->bitmap_chunk, c->delay, s->write_behind,
410 bitmapsize, offset_setable,
411 major)
412 )
413 st->ss->write_bitmap(st, fd2);
414 else {
415 pr_err("failed to create internal bitmap"
416 " - 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, NULL);
427 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
428 mdi->bitmap_offset);
429 } else {
430 array.state |= (1<<MD_SB_BITMAP_PRESENT);
431 rv = ioctl(fd, SET_ARRAY_INFO, &array);
432 }
433 if (rv < 0) {
434 if (errno == EBUSY)
435 pr_err("Cannot add bitmap while array is"
436 " resyncing or reshaping etc.\n");
437 pr_err("failed to set internal bitmap.\n");
438 return 1;
439 }
440 } else {
441 int uuid[4];
442 int bitmap_fd;
443 int d;
444 int max_devs = st->max_devs;
445
446 /* try to load a superblock */
447 for (d = 0; d < max_devs; d++) {
448 mdu_disk_info_t disk;
449 char *dv;
450 int fd2;
451 disk.number = d;
452 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
453 continue;
454 if ((disk.major==0 && disk.minor==0) ||
455 (disk.state & (1<<MD_DISK_REMOVED)))
456 continue;
457 dv = map_dev(disk.major, disk.minor, 1);
458 if (!dv)
459 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 pr_err("cannot find UUID for array!\n");
472 return 1;
473 }
474 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
475 c->delay, s->write_behind, bitmapsize, major)) {
476 return 1;
477 }
478 bitmap_fd = open(s->bitmap_file, O_RDWR);
479 if (bitmap_fd < 0) {
480 pr_err("weird: %s cannot be opened\n",
481 s->bitmap_file);
482 return 1;
483 }
484 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
485 int err = errno;
486 if (errno == EBUSY)
487 pr_err("Cannot add bitmap while array is"
488 " resyncing or reshaping etc.\n");
489 pr_err("Cannot set bitmap file for %s: %s\n",
490 devname, strerror(err));
491 return 1;
492 }
493 }
494
495 return 0;
496 }
497
498 /*
499 * When reshaping an array we might need to backup some data.
500 * This is written to all spares with a 'super_block' describing it.
501 * The superblock goes 4K from the end of the used space on the
502 * device.
503 * It if written after the backup is complete.
504 * It has the following structure.
505 */
506
507 static struct mdp_backup_super {
508 char magic[16]; /* md_backup_data-1 or -2 */
509 __u8 set_uuid[16];
510 __u64 mtime;
511 /* start/sizes in 512byte sectors */
512 __u64 devstart; /* address on backup device/file of data */
513 __u64 arraystart;
514 __u64 length;
515 __u32 sb_csum; /* csum of preceeding bytes. */
516 __u32 pad1;
517 __u64 devstart2; /* offset in to data of second section */
518 __u64 arraystart2;
519 __u64 length2;
520 __u32 sb_csum2; /* csum of preceeding bytes. */
521 __u8 pad[512-68-32];
522 } __attribute__((aligned(512))) bsb, bsb2;
523
524 static __u32 bsb_csum(char *buf, int len)
525 {
526 int i;
527 int csum = 0;
528 for (i = 0; i < len; i++)
529 csum = (csum<<3) + buf[0];
530 return __cpu_to_le32(csum);
531 }
532
533 static int check_idle(struct supertype *st)
534 {
535 /* Check that all member arrays for this container, or the
536 * container of this array, are idle
537 */
538 char *container = (st->container_devnm[0]
539 ? st->container_devnm : st->devnm);
540 struct mdstat_ent *ent, *e;
541 int is_idle = 1;
542
543 ent = mdstat_read(0, 0);
544 for (e = ent ; e; e = e->next) {
545 if (!is_container_member(e, container))
546 continue;
547 if (e->percent >= 0) {
548 is_idle = 0;
549 break;
550 }
551 }
552 free_mdstat(ent);
553 return is_idle;
554 }
555
556 static int freeze_container(struct supertype *st)
557 {
558 char *container = (st->container_devnm[0]
559 ? st->container_devnm : st->devnm);
560
561 if (!check_idle(st))
562 return -1;
563
564 if (block_monitor(container, 1)) {
565 pr_err("failed to freeze container\n");
566 return -2;
567 }
568
569 return 1;
570 }
571
572 static void unfreeze_container(struct supertype *st)
573 {
574 char *container = (st->container_devnm[0]
575 ? st->container_devnm : st->devnm);
576
577 unblock_monitor(container, 1);
578 }
579
580 static int freeze(struct supertype *st)
581 {
582 /* Try to freeze resync/rebuild on this array/container.
583 * Return -1 if the array is busy,
584 * return -2 container cannot be frozen,
585 * return 0 if this kernel doesn't support 'frozen'
586 * return 1 if it worked.
587 */
588 if (st->ss->external)
589 return freeze_container(st);
590 else {
591 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
592 int err;
593 char buf[20];
594
595 if (!sra)
596 return -1;
597 /* Need to clear any 'read-auto' status */
598 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
599 strncmp(buf, "read-auto", 9) == 0)
600 sysfs_set_str(sra, NULL, "array_state", "clean");
601
602 err = sysfs_freeze_array(sra);
603 sysfs_free(sra);
604 return err;
605 }
606 }
607
608 static void unfreeze(struct supertype *st)
609 {
610 if (st->ss->external)
611 return unfreeze_container(st);
612 else {
613 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
614
615 if (sra)
616 sysfs_set_str(sra, NULL, "sync_action", "idle");
617 sysfs_free(sra);
618 }
619 }
620
621 static void wait_reshape(struct mdinfo *sra)
622 {
623 int fd = sysfs_get_fd(sra, NULL, "sync_action");
624 char action[20];
625
626 if (fd < 0)
627 return;
628
629 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
630 strncmp(action, "reshape", 7) == 0) {
631 fd_set rfds;
632 FD_ZERO(&rfds);
633 FD_SET(fd, &rfds);
634 select(fd+1, NULL, NULL, &rfds, NULL);
635 }
636 close(fd);
637 }
638
639 static int reshape_super(struct supertype *st, unsigned long long size,
640 int level, int layout, int chunksize, int raid_disks,
641 int delta_disks, char *backup_file, char *dev,
642 int direction, int verbose)
643 {
644 /* nothing extra to check in the native case */
645 if (!st->ss->external)
646 return 0;
647 if (!st->ss->reshape_super ||
648 !st->ss->manage_reshape) {
649 pr_err("%s metadata does not support reshape\n",
650 st->ss->name);
651 return 1;
652 }
653
654 return st->ss->reshape_super(st, size, level, layout, chunksize,
655 raid_disks, delta_disks, backup_file, dev,
656 direction, verbose);
657 }
658
659 static void sync_metadata(struct supertype *st)
660 {
661 if (st->ss->external) {
662 if (st->update_tail) {
663 flush_metadata_updates(st);
664 st->update_tail = &st->updates;
665 } else
666 st->ss->sync_metadata(st);
667 }
668 }
669
670 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
671 {
672 /* when dealing with external metadata subarrays we need to be
673 * prepared to handle EAGAIN. The kernel may need to wait for
674 * mdmon to mark the array active so the kernel can handle
675 * allocations/writeback when preparing the reshape action
676 * (md_allow_write()). We temporarily disable safe_mode_delay
677 * to close a race with the array_state going clean before the
678 * next write to raid_disks / stripe_cache_size
679 */
680 char safe[50];
681 int rc;
682
683 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
684 if (!container ||
685 (strcmp(name, "raid_disks") != 0 &&
686 strcmp(name, "stripe_cache_size") != 0))
687 return sysfs_set_num(sra, NULL, name, n);
688
689 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
690 if (rc <= 0)
691 return -1;
692 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
693 rc = sysfs_set_num(sra, NULL, name, n);
694 if (rc < 0 && errno == EAGAIN) {
695 ping_monitor(container);
696 /* if we get EAGAIN here then the monitor is not active
697 * so stop trying
698 */
699 rc = sysfs_set_num(sra, NULL, name, n);
700 }
701 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
702 return rc;
703 }
704
705 int start_reshape(struct mdinfo *sra, int already_running,
706 int before_data_disks, int data_disks)
707 {
708 int err;
709 unsigned long long sync_max_to_set;
710
711 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
712 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
713 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
714 sra->reshape_progress);
715 if (before_data_disks <= data_disks)
716 sync_max_to_set = sra->reshape_progress / data_disks;
717 else
718 sync_max_to_set = (sra->component_size * data_disks
719 - sra->reshape_progress) / data_disks;
720 if (!already_running)
721 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
722 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
723 if (!already_running)
724 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
725
726 return err;
727 }
728
729 void abort_reshape(struct mdinfo *sra)
730 {
731 sysfs_set_str(sra, NULL, "sync_action", "idle");
732 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
733 sysfs_set_num(sra, NULL, "suspend_hi", 0);
734 sysfs_set_num(sra, NULL, "suspend_lo", 0);
735 sysfs_set_num(sra, NULL, "sync_min", 0);
736 sysfs_set_str(sra, NULL, "sync_max", "max");
737 }
738
739 int remove_disks_for_takeover(struct supertype *st,
740 struct mdinfo *sra,
741 int layout)
742 {
743 int nr_of_copies;
744 struct mdinfo *remaining;
745 int slot;
746
747 if (sra->array.level == 10)
748 nr_of_copies = layout & 0xff;
749 else if (sra->array.level == 1)
750 nr_of_copies = sra->array.raid_disks;
751 else
752 return 1;
753
754 remaining = sra->devs;
755 sra->devs = NULL;
756 /* for each 'copy', select one device and remove from the list. */
757 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
758 struct mdinfo **diskp;
759 int found = 0;
760
761 /* Find a working device to keep */
762 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
763 struct mdinfo *disk = *diskp;
764
765 if (disk->disk.raid_disk < slot)
766 continue;
767 if (disk->disk.raid_disk >= slot + nr_of_copies)
768 continue;
769 if (disk->disk.state & (1<<MD_DISK_REMOVED))
770 continue;
771 if (disk->disk.state & (1<<MD_DISK_FAULTY))
772 continue;
773 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
774 continue;
775
776 /* We have found a good disk to use! */
777 *diskp = disk->next;
778 disk->next = sra->devs;
779 sra->devs = disk;
780 found = 1;
781 break;
782 }
783 if (!found)
784 break;
785 }
786
787 if (slot < sra->array.raid_disks) {
788 /* didn't find all slots */
789 struct mdinfo **e;
790 e = &remaining;
791 while (*e)
792 e = &(*e)->next;
793 *e = sra->devs;
794 sra->devs = remaining;
795 return 1;
796 }
797
798 /* Remove all 'remaining' devices from the array */
799 while (remaining) {
800 struct mdinfo *sd = remaining;
801 remaining = sd->next;
802
803 sysfs_set_str(sra, sd, "state", "faulty");
804 sysfs_set_str(sra, sd, "slot", "none");
805 /* for external metadata disks should be removed in mdmon */
806 if (!st->ss->external)
807 sysfs_set_str(sra, sd, "state", "remove");
808 sd->disk.state |= (1<<MD_DISK_REMOVED);
809 sd->disk.state &= ~(1<<MD_DISK_SYNC);
810 sd->next = sra->devs;
811 sra->devs = sd;
812 }
813 return 0;
814 }
815
816 void reshape_free_fdlist(int *fdlist,
817 unsigned long long *offsets,
818 int size)
819 {
820 int i;
821
822 for (i = 0; i < size; i++)
823 if (fdlist[i] >= 0)
824 close(fdlist[i]);
825
826 free(fdlist);
827 free(offsets);
828 }
829
830 int reshape_prepare_fdlist(char *devname,
831 struct mdinfo *sra,
832 int raid_disks,
833 int nrdisks,
834 unsigned long blocks,
835 char *backup_file,
836 int *fdlist,
837 unsigned long long *offsets)
838 {
839 int d = 0;
840 struct mdinfo *sd;
841
842 enable_fds(nrdisks);
843 for (d = 0; d <= nrdisks; d++)
844 fdlist[d] = -1;
845 d = raid_disks;
846 for (sd = sra->devs; sd; sd = sd->next) {
847 if (sd->disk.state & (1<<MD_DISK_FAULTY))
848 continue;
849 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
850 char *dn = map_dev(sd->disk.major,
851 sd->disk.minor, 1);
852 fdlist[sd->disk.raid_disk]
853 = dev_open(dn, O_RDONLY);
854 offsets[sd->disk.raid_disk] = sd->data_offset*512;
855 if (fdlist[sd->disk.raid_disk] < 0) {
856 pr_err("%s: cannot open component %s\n",
857 devname, dn ? dn : "-unknown-");
858 d = -1;
859 goto release;
860 }
861 } else if (backup_file == NULL) {
862 /* spare */
863 char *dn = map_dev(sd->disk.major,
864 sd->disk.minor, 1);
865 fdlist[d] = dev_open(dn, O_RDWR);
866 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
867 if (fdlist[d] < 0) {
868 pr_err("%s: cannot open component %s\n",
869 devname, dn ? dn : "-unknown-");
870 d = -1;
871 goto release;
872 }
873 d++;
874 }
875 }
876 release:
877 return d;
878 }
879
880 int reshape_open_backup_file(char *backup_file,
881 int fd,
882 char *devname,
883 long blocks,
884 int *fdlist,
885 unsigned long long *offsets,
886 int restart)
887 {
888 /* Return 1 on success, 0 on any form of failure */
889 /* need to check backup file is large enough */
890 char buf[512];
891 struct stat stb;
892 unsigned int dev;
893 int i;
894
895 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
896 S_IRUSR | S_IWUSR);
897 *offsets = 8 * 512;
898 if (*fdlist < 0) {
899 pr_err("%s: cannot create backup file %s: %s\n",
900 devname, backup_file, strerror(errno));
901 return 0;
902 }
903 /* Guard against backup file being on array device.
904 * If array is partitioned or if LVM etc is in the
905 * way this will not notice, but it is better than
906 * nothing.
907 */
908 fstat(*fdlist, &stb);
909 dev = stb.st_dev;
910 fstat(fd, &stb);
911 if (stb.st_rdev == dev) {
912 pr_err("backup file must NOT be"
913 " on the array being reshaped.\n");
914 close(*fdlist);
915 return 0;
916 }
917
918 memset(buf, 0, 512);
919 for (i=0; i < blocks + 8 ; i++) {
920 if (write(*fdlist, buf, 512) != 512) {
921 pr_err("%s: cannot create"
922 " backup file %s: %s\n",
923 devname, backup_file, strerror(errno));
924 return 0;
925 }
926 }
927 if (fsync(*fdlist) != 0) {
928 pr_err("%s: cannot create backup file %s: %s\n",
929 devname, backup_file, strerror(errno));
930 return 0;
931 }
932
933 return 1;
934 }
935
936 unsigned long GCD(unsigned long a, unsigned long b)
937 {
938 while (a != b) {
939 if (a < b)
940 b -= a;
941 if (b < a)
942 a -= b;
943 }
944 return a;
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 a = GCD(a, b);
960 /* LCM == product / GCD */
961 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
962
963 return blocks;
964 }
965
966 char *analyse_change(struct mdinfo *info, struct reshape *re)
967 {
968 /* Based on the current array state in info->array and
969 * the changes in info->new_* etc, determine:
970 * - whether the change is possible
971 * - Intermediate level/raid_disks/layout
972 * - whether a restriping reshape is needed
973 * - number of sectors in minimum change unit. This
974 * will cover a whole number of stripes in 'before' and
975 * 'after'.
976 *
977 * Return message if the change should be rejected
978 * NULL if the change can be achieved
979 *
980 * This can be called as part of starting a reshape, or
981 * when assembling an array that is undergoing reshape.
982 */
983 int near, far, offset, copies;
984 int new_disks;
985 int old_chunk, new_chunk;
986 /* delta_parity records change in number of devices
987 * caused by level change
988 */
989 int delta_parity = 0;
990
991 memset(re, 0, sizeof(*re));
992
993 /* If a new level not explicitly given, we assume no-change */
994 if (info->new_level == UnSet)
995 info->new_level = info->array.level;
996
997 if (info->new_chunk)
998 switch (info->new_level) {
999 case 0:
1000 case 4:
1001 case 5:
1002 case 6:
1003 case 10:
1004 /* chunk size is meaningful, must divide component_size
1005 * evenly
1006 */
1007 if (info->component_size % (info->new_chunk/512))
1008 return "New chunk size does not"
1009 " divide component size";
1010 break;
1011 default:
1012 return "chunk size not meaningful for this level";
1013 }
1014 else
1015 info->new_chunk = info->array.chunk_size;
1016
1017 switch (info->array.level) {
1018 default:
1019 return "Cannot understand this RAID level";
1020 case 1:
1021 /* RAID1 can convert to RAID1 with different disks, or
1022 * raid5 with 2 disks, or
1023 * raid0 with 1 disk
1024 */
1025 if (info->new_level > 1 &&
1026 (info->component_size & 7))
1027 return "Cannot convert RAID1 of this size - "
1028 "reduce size to multiple of 4K first.";
1029 if (info->new_level == 0) {
1030 if (info->delta_disks != UnSet &&
1031 info->delta_disks != 0)
1032 return "Cannot change number of disks "
1033 "with RAID1->RAID0 conversion";
1034 re->level = 0;
1035 re->before.data_disks = 1;
1036 re->after.data_disks = 1;
1037 return NULL;
1038 }
1039 if (info->new_level == 1) {
1040 if (info->delta_disks == UnSet)
1041 /* Don't know what to do */
1042 return "no change requested for Growing RAID1";
1043 re->level = 1;
1044 return NULL;
1045 }
1046 if (info->array.raid_disks == 2 &&
1047 info->new_level == 5) {
1048
1049 re->level = 5;
1050 re->before.data_disks = 1;
1051 if (info->delta_disks != UnSet &&
1052 info->delta_disks != 0)
1053 re->after.data_disks = 1 + info->delta_disks;
1054 else
1055 re->after.data_disks = 1;
1056 if (re->after.data_disks < 1)
1057 return "Number of disks too small for RAID5";
1058
1059 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1060 info->array.chunk_size = 65536;
1061 break;
1062 }
1063 /* Could do some multi-stage conversions, but leave that to
1064 * later.
1065 */
1066 return "Impossibly level change request for RAID1";
1067
1068 case 10:
1069 /* RAID10 can be converted from near mode to
1070 * RAID0 by removing some devices.
1071 * It can also be reshaped if the kernel supports
1072 * new_data_offset.
1073 */
1074 switch (info->new_level) {
1075 case 0:
1076 if ((info->array.layout & ~0xff) != 0x100)
1077 return "Cannot Grow RAID10 with far/offset layout";
1078 /* number of devices must be multiple of number of copies */
1079 if (info->array.raid_disks % (info->array.layout & 0xff))
1080 return "RAID10 layout too complex for Grow operation";
1081
1082 new_disks = (info->array.raid_disks
1083 / (info->array.layout & 0xff));
1084 if (info->delta_disks == UnSet)
1085 info->delta_disks = (new_disks
1086 - info->array.raid_disks);
1087
1088 if (info->delta_disks != new_disks - info->array.raid_disks)
1089 return "New number of raid-devices impossible for RAID10";
1090 if (info->new_chunk &&
1091 info->new_chunk != info->array.chunk_size)
1092 return "Cannot change chunk-size with RAID10 Grow";
1093
1094 /* looks good */
1095 re->level = 0;
1096 re->before.data_disks = new_disks;
1097 re->after.data_disks = re->before.data_disks;
1098 return NULL;
1099
1100 case 10:
1101 near = info->array.layout & 0xff;
1102 far = (info->array.layout >> 8) & 0xff;
1103 offset = info->array.layout & 0x10000;
1104 if (far > 1 && !offset)
1105 return "Cannot reshape RAID10 in far-mode";
1106 copies = near * far;
1107
1108 old_chunk = info->array.chunk_size * far;
1109
1110 if (info->new_layout == UnSet)
1111 info->new_layout = info->array.layout;
1112 else {
1113 near = info->new_layout & 0xff;
1114 far = (info->new_layout >> 8) & 0xff;
1115 offset = info->new_layout & 0x10000;
1116 if (far > 1 && !offset)
1117 return "Cannot reshape RAID10 to far-mode";
1118 if (near * far != copies)
1119 return "Cannot change number of copies"
1120 " when reshaping RAID10";
1121 }
1122 if (info->delta_disks == UnSet)
1123 info->delta_disks = 0;
1124 new_disks = (info->array.raid_disks +
1125 info->delta_disks);
1126
1127 new_chunk = info->new_chunk * far;
1128
1129 re->level = 10;
1130 re->before.layout = info->array.layout;
1131 re->before.data_disks = info->array.raid_disks;
1132 re->after.layout = info->new_layout;
1133 re->after.data_disks = new_disks;
1134 /* For RAID10 we don't do backup but do allow reshape,
1135 * so set backup_blocks to INVALID_SECTORS rather than
1136 * zero.
1137 * And there is no need to synchronise stripes on both
1138 * 'old' and 'new'. So the important
1139 * number is the minimum data_offset difference
1140 * which is the larger of (offset copies * chunk).
1141 */
1142 re->backup_blocks = INVALID_SECTORS;
1143 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1144 if (new_disks < re->before.data_disks &&
1145 info->space_after < re->min_offset_change)
1146 /* Reduce component size by one chunk */
1147 re->new_size = (info->component_size -
1148 re->min_offset_change);
1149 else
1150 re->new_size = info->component_size;
1151 re->new_size = re->new_size * new_disks / copies;
1152 return NULL;
1153
1154 default:
1155 return "RAID10 can only be changed to RAID0";
1156 }
1157 case 0:
1158 /* RAID0 can be converted to RAID10, or to RAID456 */
1159 if (info->new_level == 10) {
1160 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1161 /* Assume near=2 layout */
1162 info->new_layout = 0x102;
1163 info->delta_disks = info->array.raid_disks;
1164 }
1165 if (info->new_layout == UnSet) {
1166 int copies = 1 + (info->delta_disks
1167 / info->array.raid_disks);
1168 if (info->array.raid_disks * (copies-1)
1169 != info->delta_disks)
1170 return "Impossible number of devices"
1171 " for RAID0->RAID10";
1172 info->new_layout = 0x100 + copies;
1173 }
1174 if (info->delta_disks == UnSet) {
1175 int copies = info->new_layout & 0xff;
1176 if (info->new_layout != 0x100 + copies)
1177 return "New layout impossible"
1178 " for RAID0->RAID10";;
1179 info->delta_disks = (copies - 1) *
1180 info->array.raid_disks;
1181 }
1182 if (info->new_chunk &&
1183 info->new_chunk != info->array.chunk_size)
1184 return "Cannot change chunk-size with RAID0->RAID10";
1185 /* looks good */
1186 re->level = 10;
1187 re->before.data_disks = (info->array.raid_disks +
1188 info->delta_disks);
1189 re->after.data_disks = re->before.data_disks;
1190 re->before.layout = info->new_layout;
1191 return NULL;
1192 }
1193
1194 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1195 * a raid4 style layout of the final level.
1196 */
1197 switch (info->new_level) {
1198 case 4:
1199 delta_parity = 1;
1200 case 0:
1201 re->level = 4;
1202 re->before.layout = 0;
1203 break;
1204 case 5:
1205 delta_parity = 1;
1206 re->level = 5;
1207 re->before.layout = ALGORITHM_PARITY_N;
1208 break;
1209 case 6:
1210 delta_parity = 2;
1211 re->level = 6;
1212 re->before.layout = ALGORITHM_PARITY_N;
1213 break;
1214 default:
1215 return "Impossible level change requested";
1216 }
1217 re->before.data_disks = info->array.raid_disks;
1218 /* determining 'after' layout happens outside this 'switch' */
1219 break;
1220
1221 case 4:
1222 info->array.layout = ALGORITHM_PARITY_N;
1223 case 5:
1224 switch (info->new_level) {
1225 case 0:
1226 delta_parity = -1;
1227 case 4:
1228 re->level = info->array.level;
1229 re->before.data_disks = info->array.raid_disks - 1;
1230 re->before.layout = info->array.layout;
1231 break;
1232 case 5:
1233 re->level = 5;
1234 re->before.data_disks = info->array.raid_disks - 1;
1235 re->before.layout = info->array.layout;
1236 break;
1237 case 6:
1238 delta_parity = 1;
1239 re->level = 6;
1240 re->before.data_disks = info->array.raid_disks - 1;
1241 switch (info->array.layout) {
1242 case ALGORITHM_LEFT_ASYMMETRIC:
1243 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1244 break;
1245 case ALGORITHM_RIGHT_ASYMMETRIC:
1246 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1247 break;
1248 case ALGORITHM_LEFT_SYMMETRIC:
1249 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1250 break;
1251 case ALGORITHM_RIGHT_SYMMETRIC:
1252 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1253 break;
1254 case ALGORITHM_PARITY_0:
1255 re->before.layout = ALGORITHM_PARITY_0_6;
1256 break;
1257 case ALGORITHM_PARITY_N:
1258 re->before.layout = ALGORITHM_PARITY_N_6;
1259 break;
1260 default:
1261 return "Cannot convert an array with this layout";
1262 }
1263 break;
1264 case 1:
1265 if (info->array.raid_disks != 2)
1266 return "Can only convert a 2-device array to RAID1";
1267 if (info->delta_disks != UnSet &&
1268 info->delta_disks != 0)
1269 return "Cannot set raid_disk when "
1270 "converting RAID5->RAID1";
1271 re->level = 1;
1272 info->new_chunk = 0;
1273 return NULL;
1274 default:
1275 return "Impossible level change requested";
1276 }
1277 break;
1278 case 6:
1279 switch (info->new_level) {
1280 case 4:
1281 case 5:
1282 delta_parity = -1;
1283 case 6:
1284 re->level = 6;
1285 re->before.data_disks = info->array.raid_disks - 2;
1286 re->before.layout = info->array.layout;
1287 break;
1288 default:
1289 return "Impossible level change requested";
1290 }
1291 break;
1292 }
1293
1294 /* If we reached here then it looks like a re-stripe is
1295 * happening. We have determined the intermediate level
1296 * and initial raid_disks/layout and stored these in 're'.
1297 *
1298 * We need to deduce the final layout that can be atomically
1299 * converted to the end state.
1300 */
1301 switch (info->new_level) {
1302 case 0:
1303 /* We can only get to RAID0 from RAID4 or RAID5
1304 * with appropriate layout and one extra device
1305 */
1306 if (re->level != 4 && re->level != 5)
1307 return "Cannot covert to RAID0 from this level";
1308
1309 switch (re->level) {
1310 case 4:
1311 re->before.layout = 0;
1312 re->after.layout = 0;
1313 break;
1314 case 5:
1315 re->after.layout = ALGORITHM_PARITY_N;
1316 break;
1317 }
1318 break;
1319
1320 case 4:
1321 /* We can only get to RAID4 from RAID5 */
1322 if (re->level != 4 && re->level != 5)
1323 return "Cannot convert to RAID4 from this level";
1324
1325 switch (re->level) {
1326 case 4:
1327 re->before.layout = 0;
1328 re->after.layout = 0;
1329 break;
1330 case 5:
1331 re->after.layout = ALGORITHM_PARITY_N;
1332 break;
1333 }
1334 break;
1335
1336 case 5:
1337 /* We get to RAID5 from RAID5 or RAID6 */
1338 if (re->level != 5 && re->level != 6)
1339 return "Cannot convert to RAID5 from this level";
1340
1341 switch (re->level) {
1342 case 5:
1343 if (info->new_layout == UnSet)
1344 re->after.layout = re->before.layout;
1345 else
1346 re->after.layout = info->new_layout;
1347 break;
1348 case 6:
1349 if (info->new_layout == UnSet)
1350 info->new_layout = re->before.layout;
1351
1352 /* after.layout needs to be raid6 version of new_layout */
1353 if (info->new_layout == ALGORITHM_PARITY_N)
1354 re->after.layout = ALGORITHM_PARITY_N;
1355 else {
1356 char layout[40];
1357 char *ls = map_num(r5layout, info->new_layout);
1358 int l;
1359 if (ls) {
1360 /* Current RAID6 layout has a RAID5
1361 * equivalent - good
1362 */
1363 strcat(strcpy(layout, ls), "-6");
1364 l = map_name(r6layout, layout);
1365 if (l == UnSet)
1366 return "Cannot find RAID6 layout"
1367 " to convert to";
1368 } else {
1369 /* Current RAID6 has no equivalent.
1370 * If it is already a '-6' layout we
1371 * can leave it unchanged, else we must
1372 * fail
1373 */
1374 ls = map_num(r6layout, info->new_layout);
1375 if (!ls ||
1376 strcmp(ls+strlen(ls)-2, "-6") != 0)
1377 return "Please specify new layout";
1378 l = info->new_layout;
1379 }
1380 re->after.layout = l;
1381 }
1382 }
1383 break;
1384
1385 case 6:
1386 /* We must already be at level 6 */
1387 if (re->level != 6)
1388 return "Impossible level change";
1389 if (info->new_layout == UnSet)
1390 re->after.layout = info->array.layout;
1391 else
1392 re->after.layout = info->new_layout;
1393 break;
1394 default:
1395 return "Impossible level change requested";
1396 }
1397 if (info->delta_disks == UnSet)
1398 info->delta_disks = delta_parity;
1399
1400 re->after.data_disks = (re->before.data_disks
1401 + info->delta_disks
1402 - delta_parity);
1403 switch (re->level) {
1404 case 6: re->parity = 2;
1405 break;
1406 case 4:
1407 case 5: re->parity = 1;
1408 break;
1409 default: re->parity = 0;
1410 break;
1411 }
1412 /* So we have a restripe operation, we need to calculate the number
1413 * of blocks per reshape operation.
1414 */
1415 if (info->new_chunk == 0)
1416 info->new_chunk = info->array.chunk_size;
1417 if (re->after.data_disks == re->before.data_disks &&
1418 re->after.layout == re->before.layout &&
1419 info->new_chunk == info->array.chunk_size) {
1420 /* Nothing to change, can change level immediately. */
1421 re->level = info->new_level;
1422 re->backup_blocks = 0;
1423 return NULL;
1424 }
1425 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1426 /* chunk and layout changes make no difference */
1427 re->level = info->new_level;
1428 re->backup_blocks = 0;
1429 return NULL;
1430 }
1431
1432 if (re->after.data_disks == re->before.data_disks &&
1433 get_linux_version() < 2006032)
1434 return "in-place reshape is not safe before 2.6.32 - sorry.";
1435
1436 if (re->after.data_disks < re->before.data_disks &&
1437 get_linux_version() < 2006030)
1438 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1439
1440 re->backup_blocks = compute_backup_blocks(
1441 info->new_chunk, info->array.chunk_size,
1442 re->after.data_disks,
1443 re->before.data_disks);
1444 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1445
1446 re->new_size = info->component_size * re->after.data_disks;
1447 return NULL;
1448 }
1449
1450 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1451 char *text_version)
1452 {
1453 struct mdinfo *info;
1454 char *subarray;
1455 int ret_val = -1;
1456
1457 if ((st == NULL) || (sra == NULL))
1458 return ret_val;
1459
1460 if (text_version == NULL)
1461 text_version = sra->text_version;
1462 subarray = strchr(text_version+1, '/')+1;
1463 info = st->ss->container_content(st, subarray);
1464 if (info) {
1465 unsigned long long current_size = 0;
1466 unsigned long long new_size =
1467 info->custom_array_size/2;
1468
1469 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1470 new_size > current_size) {
1471 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1472 < 0)
1473 dprintf("Error: Cannot set array size");
1474 else {
1475 ret_val = 0;
1476 dprintf("Array size changed");
1477 }
1478 dprintf(" from %llu to %llu.\n",
1479 current_size, new_size);
1480 }
1481 sysfs_free(info);
1482 } else
1483 dprintf("Error: set_array_size(): info pointer in NULL\n");
1484
1485 return ret_val;
1486 }
1487
1488 static int reshape_array(char *container, int fd, char *devname,
1489 struct supertype *st, struct mdinfo *info,
1490 int force, struct mddev_dev *devlist,
1491 unsigned long long data_offset,
1492 char *backup_file, int verbose, int forked,
1493 int restart, int freeze_reshape);
1494 static int reshape_container(char *container, char *devname,
1495 int mdfd,
1496 struct supertype *st,
1497 struct mdinfo *info,
1498 int force,
1499 char *backup_file,
1500 int verbose, int restart, int freeze_reshape);
1501
1502 int Grow_reshape(char *devname, int fd,
1503 struct mddev_dev *devlist,
1504 unsigned long long data_offset,
1505 struct context *c, struct shape *s)
1506 {
1507 /* Make some changes in the shape of an array.
1508 * The kernel must support the change.
1509 *
1510 * There are three different changes. Each can trigger
1511 * a resync or recovery so we freeze that until we have
1512 * requested everything (if kernel supports freezing - 2.6.30).
1513 * The steps are:
1514 * - change size (i.e. component_size)
1515 * - change level
1516 * - change layout/chunksize/ndisks
1517 *
1518 * The last can require a reshape. It is different on different
1519 * levels so we need to check the level before actioning it.
1520 * Some times the level change needs to be requested after the
1521 * reshape (e.g. raid6->raid5, raid5->raid0)
1522 *
1523 */
1524 struct mdu_array_info_s array;
1525 int rv = 0;
1526 struct supertype *st;
1527 char *subarray = NULL;
1528
1529 int frozen;
1530 int changed = 0;
1531 char *container = NULL;
1532 int cfd = -1;
1533
1534 struct mddev_dev *dv;
1535 int added_disks;
1536
1537 struct mdinfo info;
1538 struct mdinfo *sra;
1539
1540 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1541 pr_err("%s is not an active md array - aborting\n",
1542 devname);
1543 return 1;
1544 }
1545 if (data_offset != INVALID_SECTORS && array.level != 10
1546 && (array.level < 4 || array.level > 6)) {
1547 pr_err("--grow --data-offset not yet supported\n");
1548 return 1;
1549 }
1550
1551 if (s->size > 0 &&
1552 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1553 pr_err("cannot change component size at the same time "
1554 "as other changes.\n"
1555 " Change size first, then check data is intact before "
1556 "making other changes.\n");
1557 return 1;
1558 }
1559
1560 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1561 get_linux_version() < 2006032 &&
1562 !check_env("MDADM_FORCE_FEWER")) {
1563 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1564 " Please use a newer kernel\n");
1565 return 1;
1566 }
1567
1568 st = super_by_fd(fd, &subarray);
1569 if (!st) {
1570 pr_err("Unable to determine metadata format for %s\n", devname);
1571 return 1;
1572 }
1573 if (s->raiddisks > st->max_devs) {
1574 pr_err("Cannot increase raid-disks on this array"
1575 " beyond %d\n", st->max_devs);
1576 return 1;
1577 }
1578
1579 /* in the external case we need to check that the requested reshape is
1580 * supported, and perform an initial check that the container holds the
1581 * pre-requisite spare devices (mdmon owns final validation)
1582 */
1583 if (st->ss->external) {
1584 int rv;
1585
1586 if (subarray) {
1587 container = st->container_devnm;
1588 cfd = open_dev_excl(st->container_devnm);
1589 } else {
1590 container = st->devnm;
1591 close(fd);
1592 cfd = open_dev_excl(st->devnm);
1593 fd = cfd;
1594 }
1595 if (cfd < 0) {
1596 pr_err("Unable to open container for %s\n",
1597 devname);
1598 free(subarray);
1599 return 1;
1600 }
1601
1602 rv = st->ss->load_container(st, cfd, NULL);
1603
1604 if (rv) {
1605 pr_err("Cannot read superblock for %s\n",
1606 devname);
1607 free(subarray);
1608 return 1;
1609 }
1610
1611 /* check if operation is supported for metadata handler */
1612 if (st->ss->container_content) {
1613 struct mdinfo *cc = NULL;
1614 struct mdinfo *content = NULL;
1615
1616 cc = st->ss->container_content(st, subarray);
1617 for (content = cc; content ; content = content->next) {
1618 int allow_reshape = 1;
1619
1620 /* check if reshape is allowed based on metadata
1621 * indications stored in content.array.status
1622 */
1623 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1624 allow_reshape = 0;
1625 if (content->array.state
1626 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1627 allow_reshape = 0;
1628 if (!allow_reshape) {
1629 pr_err("cannot reshape arrays in"
1630 " container with unsupported"
1631 " metadata: %s(%s)\n",
1632 devname, container);
1633 sysfs_free(cc);
1634 free(subarray);
1635 return 1;
1636 }
1637 }
1638 sysfs_free(cc);
1639 }
1640 if (mdmon_running(container))
1641 st->update_tail = &st->updates;
1642 }
1643
1644 added_disks = 0;
1645 for (dv = devlist; dv; dv = dv->next)
1646 added_disks++;
1647 if (s->raiddisks > array.raid_disks &&
1648 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1649 !c->force) {
1650 pr_err("Need %d spare%s to avoid degraded array,"
1651 " and only have %d.\n"
1652 " Use --force to over-ride this check.\n",
1653 s->raiddisks - array.raid_disks,
1654 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1655 array.spare_disks + added_disks);
1656 return 1;
1657 }
1658
1659 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1660 | GET_STATE | GET_VERSION);
1661 if (sra) {
1662 if (st->ss->external && subarray == NULL) {
1663 array.level = LEVEL_CONTAINER;
1664 sra->array.level = LEVEL_CONTAINER;
1665 }
1666 } else {
1667 pr_err("failed to read sysfs parameters for %s\n",
1668 devname);
1669 return 1;
1670 }
1671 frozen = freeze(st);
1672 if (frozen < -1) {
1673 /* freeze() already spewed the reason */
1674 sysfs_free(sra);
1675 return 1;
1676 } else if (frozen < 0) {
1677 pr_err("%s is performing resync/recovery and cannot"
1678 " be reshaped\n", devname);
1679 sysfs_free(sra);
1680 return 1;
1681 }
1682
1683 /* ========= set size =============== */
1684 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1685 unsigned long long orig_size = get_component_size(fd)/2;
1686 unsigned long long min_csize;
1687 struct mdinfo *mdi;
1688 int raid0_takeover = 0;
1689
1690 if (orig_size == 0)
1691 orig_size = (unsigned) array.size;
1692
1693 if (orig_size == 0) {
1694 pr_err("Cannot set device size in this type of array.\n");
1695 rv = 1;
1696 goto release;
1697 }
1698
1699 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1700 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1701 rv = 1;
1702 goto release;
1703 }
1704 sync_metadata(st);
1705 if (st->ss->external) {
1706 /* metadata can have size limitation
1707 * update size value according to metadata information
1708 */
1709 struct mdinfo *sizeinfo =
1710 st->ss->container_content(st, subarray);
1711 if (sizeinfo) {
1712 unsigned long long new_size =
1713 sizeinfo->custom_array_size/2;
1714 int data_disks = get_data_disks(
1715 sizeinfo->array.level,
1716 sizeinfo->array.layout,
1717 sizeinfo->array.raid_disks);
1718 new_size /= data_disks;
1719 dprintf("Metadata size correction from %llu to "
1720 "%llu (%llu)\n", orig_size, new_size,
1721 new_size * data_disks);
1722 s->size = new_size;
1723 sysfs_free(sizeinfo);
1724 }
1725 }
1726
1727 /* Update the size of each member device in case
1728 * they have been resized. This will never reduce
1729 * below the current used-size. The "size" attribute
1730 * understands '0' to mean 'max'.
1731 */
1732 min_csize = 0;
1733 rv = 0;
1734 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1735 if (sysfs_set_num(sra, mdi, "size",
1736 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1737 /* Probably kernel refusing to let us
1738 * reduce the size - not an error.
1739 */
1740 break;
1741 }
1742 if (array.not_persistent == 0 &&
1743 array.major_version == 0 &&
1744 get_linux_version() < 3001000) {
1745 /* Dangerous to allow size to exceed 2TB */
1746 unsigned long long csize;
1747 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1748 if (csize >= 2ULL*1024*1024*1024)
1749 csize = 2ULL*1024*1024*1024;
1750 if ((min_csize == 0 || (min_csize
1751 > csize)))
1752 min_csize = csize;
1753 }
1754 }
1755 }
1756 if (rv) {
1757 pr_err("Cannot set size on "
1758 "array members.\n");
1759 goto size_change_error;
1760 }
1761 if (min_csize && s->size > min_csize) {
1762 pr_err("Cannot safely make this array "
1763 "use more than 2TB per device on this kernel.\n");
1764 rv = 1;
1765 goto size_change_error;
1766 }
1767 if (min_csize && s->size == MAX_SIZE) {
1768 /* Don't let the kernel choose a size - it will get
1769 * it wrong
1770 */
1771 pr_err("Limited v0.90 array to "
1772 "2TB per device\n");
1773 s->size = min_csize;
1774 }
1775 if (st->ss->external) {
1776 if (sra->array.level == 0) {
1777 rv = sysfs_set_str(sra, NULL, "level",
1778 "raid5");
1779 if (!rv) {
1780 raid0_takeover = 1;
1781 /* get array parametes after takeover
1782 * to chane one parameter at time only
1783 */
1784 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1785 }
1786 }
1787 /* make sure mdmon is
1788 * aware of the new level */
1789 if (!mdmon_running(st->container_devnm))
1790 start_mdmon(st->container_devnm);
1791 ping_monitor(container);
1792 if (mdmon_running(st->container_devnm) &&
1793 st->update_tail == NULL)
1794 st->update_tail = &st->updates;
1795 }
1796
1797 if (s->size == MAX_SIZE)
1798 s->size = 0;
1799 array.size = s->size;
1800 if ((unsigned)array.size != s->size) {
1801 /* got truncated to 32bit, write to
1802 * component_size instead
1803 */
1804 if (sra)
1805 rv = sysfs_set_num(sra, NULL,
1806 "component_size", s->size);
1807 else
1808 rv = -1;
1809 } else {
1810 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1811
1812 /* manage array size when it is managed externally
1813 */
1814 if ((rv == 0) && st->ss->external)
1815 rv = set_array_size(st, sra, sra->text_version);
1816 }
1817
1818 if (raid0_takeover) {
1819 /* do not recync non-existing parity,
1820 * we will drop it anyway
1821 */
1822 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1823 /* go back to raid0, drop parity disk
1824 */
1825 sysfs_set_str(sra, NULL, "level", "raid0");
1826 ioctl(fd, GET_ARRAY_INFO, &array);
1827 }
1828
1829 size_change_error:
1830 if (rv != 0) {
1831 int err = errno;
1832
1833 /* restore metadata */
1834 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1835 UnSet, NULL, devname,
1836 ROLLBACK_METADATA_CHANGES,
1837 c->verbose) == 0)
1838 sync_metadata(st);
1839 pr_err("Cannot set device size for %s: %s\n",
1840 devname, strerror(err));
1841 if (err == EBUSY &&
1842 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1843 cont_err("Bitmap must be removed before size can be changed\n");
1844 rv = 1;
1845 goto release;
1846 }
1847 if (s->assume_clean) {
1848 /* This will fail on kernels older than 3.0 unless
1849 * a backport has been arranged.
1850 */
1851 if (sra == NULL ||
1852 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1853 pr_err("--assume-clean not supported with --grow on this kernel\n");
1854 }
1855 ioctl(fd, GET_ARRAY_INFO, &array);
1856 s->size = get_component_size(fd)/2;
1857 if (s->size == 0)
1858 s->size = array.size;
1859 if (c->verbose >= 0) {
1860 if (s->size == orig_size)
1861 pr_err("component size of %s "
1862 "unchanged at %lluK\n",
1863 devname, s->size);
1864 else
1865 pr_err("component size of %s "
1866 "has been set to %lluK\n",
1867 devname, s->size);
1868 }
1869 changed = 1;
1870 } else if (array.level != LEVEL_CONTAINER) {
1871 s->size = get_component_size(fd)/2;
1872 if (s->size == 0)
1873 s->size = array.size;
1874 }
1875
1876 /* See if there is anything else to do */
1877 if ((s->level == UnSet || s->level == array.level) &&
1878 (s->layout_str == NULL) &&
1879 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1880 data_offset == INVALID_SECTORS &&
1881 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1882 /* Nothing more to do */
1883 if (!changed && c->verbose >= 0)
1884 pr_err("%s: no change requested\n",
1885 devname);
1886 goto release;
1887 }
1888
1889 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1890 * current implementation assumes that following conditions must be met:
1891 * - RAID10:
1892 * - far_copies == 1
1893 * - near_copies == 2
1894 */
1895 if ((s->level == 0 && array.level == 10 && sra &&
1896 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1897 (s->level == 0 && array.level == 1 && sra)) {
1898 int err;
1899 err = remove_disks_for_takeover(st, sra, array.layout);
1900 if (err) {
1901 dprintf(Name": Array cannot be reshaped\n");
1902 if (cfd > -1)
1903 close(cfd);
1904 rv = 1;
1905 goto release;
1906 }
1907 /* Make sure mdmon has seen the device removal
1908 * and updated metadata before we continue with
1909 * level change
1910 */
1911 if (container)
1912 ping_monitor(container);
1913 }
1914
1915 memset(&info, 0, sizeof(info));
1916 info.array = array;
1917 sysfs_init(&info, fd, NULL);
1918 strcpy(info.text_version, sra->text_version);
1919 info.component_size = s->size*2;
1920 info.new_level = s->level;
1921 info.new_chunk = s->chunk * 1024;
1922 if (info.array.level == LEVEL_CONTAINER) {
1923 info.delta_disks = UnSet;
1924 info.array.raid_disks = s->raiddisks;
1925 } else if (s->raiddisks)
1926 info.delta_disks = s->raiddisks - info.array.raid_disks;
1927 else
1928 info.delta_disks = UnSet;
1929 if (s->layout_str == NULL) {
1930 info.new_layout = UnSet;
1931 if (info.array.level == 6 &&
1932 (info.new_level == 6 || info.new_level == UnSet) &&
1933 info.array.layout >= 16) {
1934 pr_err("%s has a non-standard layout. If you"
1935 " wish to preserve this\n", devname);
1936 cont_err("during the reshape, please specify"
1937 " --layout=preserve\n");
1938 cont_err("If you want to change it, specify a"
1939 " layout or use --layout=normalise\n");
1940 rv = 1;
1941 goto release;
1942 }
1943 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1944 strcmp(s->layout_str, "normalize") == 0) {
1945 /* If we have a -6 RAID6 layout, remove the '-6'. */
1946 info.new_layout = UnSet;
1947 if (info.array.level == 6 && info.new_level == UnSet) {
1948 char l[40], *h;
1949 strcpy(l, map_num(r6layout, info.array.layout));
1950 h = strrchr(l, '-');
1951 if (h && strcmp(h, "-6") == 0) {
1952 *h = 0;
1953 info.new_layout = map_name(r6layout, l);
1954 }
1955 } else {
1956 pr_err("%s is only meaningful when reshaping"
1957 " a RAID6 array.\n", s->layout_str);
1958 rv = 1;
1959 goto release;
1960 }
1961 } else if (strcmp(s->layout_str, "preserve") == 0) {
1962 /* This means that a non-standard RAID6 layout
1963 * is OK.
1964 * In particular:
1965 * - When reshape a RAID6 (e.g. adding a device)
1966 * which is in a non-standard layout, it is OK
1967 * to preserve that layout.
1968 * - When converting a RAID5 to RAID6, leave it in
1969 * the XXX-6 layout, don't re-layout.
1970 */
1971 if (info.array.level == 6 && info.new_level == UnSet)
1972 info.new_layout = info.array.layout;
1973 else if (info.array.level == 5 && info.new_level == 6) {
1974 char l[40];
1975 strcpy(l, map_num(r5layout, info.array.layout));
1976 strcat(l, "-6");
1977 info.new_layout = map_name(r6layout, l);
1978 } else {
1979 pr_err("%s in only meaningful when reshaping"
1980 " to RAID6\n", s->layout_str);
1981 rv = 1;
1982 goto release;
1983 }
1984 } else {
1985 int l = info.new_level;
1986 if (l == UnSet)
1987 l = info.array.level;
1988 switch (l) {
1989 case 5:
1990 info.new_layout = map_name(r5layout, s->layout_str);
1991 break;
1992 case 6:
1993 info.new_layout = map_name(r6layout, s->layout_str);
1994 break;
1995 case 10:
1996 info.new_layout = parse_layout_10(s->layout_str);
1997 break;
1998 case LEVEL_FAULTY:
1999 info.new_layout = parse_layout_faulty(s->layout_str);
2000 break;
2001 default:
2002 pr_err("layout not meaningful"
2003 " with this level\n");
2004 rv = 1;
2005 goto release;
2006 }
2007 if (info.new_layout == UnSet) {
2008 pr_err("layout %s not understood"
2009 " for this level\n",
2010 s->layout_str);
2011 rv = 1;
2012 goto release;
2013 }
2014 }
2015
2016 if (array.level == LEVEL_FAULTY) {
2017 if (s->level != UnSet && s->level != array.level) {
2018 pr_err("cannot change level of Faulty device\n");
2019 rv =1 ;
2020 }
2021 if (s->chunk) {
2022 pr_err("cannot set chunksize of Faulty device\n");
2023 rv =1 ;
2024 }
2025 if (s->raiddisks && s->raiddisks != 1) {
2026 pr_err("cannot set raid_disks of Faulty device\n");
2027 rv =1 ;
2028 }
2029 if (s->layout_str) {
2030 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2031 dprintf("Cannot get array information.\n");
2032 goto release;
2033 }
2034 array.layout = info.new_layout;
2035 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2036 pr_err("failed to set new layout\n");
2037 rv = 1;
2038 } else if (c->verbose >= 0)
2039 printf("layout for %s set to %d\n",
2040 devname, array.layout);
2041 }
2042 } else if (array.level == LEVEL_CONTAINER) {
2043 /* This change is to be applied to every array in the
2044 * container. This is only needed when the metadata imposes
2045 * restraints of the various arrays in the container.
2046 * Currently we only know that IMSM requires all arrays
2047 * to have the same number of devices so changing the
2048 * number of devices (On-Line Capacity Expansion) must be
2049 * performed at the level of the container
2050 */
2051 rv = reshape_container(container, devname, -1, st, &info,
2052 c->force, c->backup_file, c->verbose, 0, 0);
2053 frozen = 0;
2054 } else {
2055 /* get spare devices from external metadata
2056 */
2057 if (st->ss->external) {
2058 struct mdinfo *info2;
2059
2060 info2 = st->ss->container_content(st, subarray);
2061 if (info2) {
2062 info.array.spare_disks =
2063 info2->array.spare_disks;
2064 sysfs_free(info2);
2065 }
2066 }
2067
2068 /* Impose these changes on a single array. First
2069 * check that the metadata is OK with the change. */
2070
2071 if (reshape_super(st, 0, info.new_level,
2072 info.new_layout, info.new_chunk,
2073 info.array.raid_disks, info.delta_disks,
2074 c->backup_file, devname, APPLY_METADATA_CHANGES,
2075 c->verbose)) {
2076 rv = 1;
2077 goto release;
2078 }
2079 sync_metadata(st);
2080 rv = reshape_array(container, fd, devname, st, &info, c->force,
2081 devlist, data_offset, c->backup_file, c->verbose,
2082 0, 0, 0);
2083 frozen = 0;
2084 }
2085 release:
2086 sysfs_free(sra);
2087 if (frozen > 0)
2088 unfreeze(st);
2089 return rv;
2090 }
2091
2092 /* verify_reshape_position()
2093 * Function checks if reshape position in metadata is not farther
2094 * than position in md.
2095 * Return value:
2096 * 0 : not valid sysfs entry
2097 * it can be caused by not started reshape, it should be started
2098 * by reshape array or raid0 array is before takeover
2099 * -1 : error, reshape position is obviously wrong
2100 * 1 : success, reshape progress correct or updated
2101 */
2102 static int verify_reshape_position(struct mdinfo *info, int level)
2103 {
2104 int ret_val = 0;
2105 char buf[40];
2106 int rv;
2107
2108 /* read sync_max, failure can mean raid0 array */
2109 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2110
2111 if (rv > 0) {
2112 char *ep;
2113 unsigned long long position = strtoull(buf, &ep, 0);
2114
2115 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2116 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2117 position *= get_data_disks(level,
2118 info->new_layout,
2119 info->array.raid_disks);
2120 if (info->reshape_progress < position) {
2121 dprintf("Corrected reshape progress (%llu) to "
2122 "md position (%llu)\n",
2123 info->reshape_progress, position);
2124 info->reshape_progress = position;
2125 ret_val = 1;
2126 } else if (info->reshape_progress > position) {
2127 pr_err("Fatal error: array "
2128 "reshape was not properly frozen "
2129 "(expected reshape position is %llu, "
2130 "but reshape progress is %llu.\n",
2131 position, info->reshape_progress);
2132 ret_val = -1;
2133 } else {
2134 dprintf("Reshape position in md and metadata "
2135 "are the same;");
2136 ret_val = 1;
2137 }
2138 }
2139 } else if (rv == 0) {
2140 /* for valid sysfs entry, 0-length content
2141 * should be indicated as error
2142 */
2143 ret_val = -1;
2144 }
2145
2146 return ret_val;
2147 }
2148
2149 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2150 char *devname, int delta_disks,
2151 unsigned long long data_offset,
2152 unsigned long long min)
2153 {
2154 struct mdinfo *sd;
2155 int dir = 0;
2156 int err = 0;
2157 unsigned long long before, after;
2158
2159 /* Need to find min space before and after so same is used
2160 * on all devices
2161 */
2162 before = UINT64_MAX;
2163 after = UINT64_MAX;
2164 for (sd = sra->devs; sd; sd = sd->next) {
2165 char *dn;
2166 int dfd;
2167 int rv;
2168 struct supertype *st2;
2169 struct mdinfo info2;
2170
2171 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2172 continue;
2173 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2174 dfd = dev_open(dn, O_RDONLY);
2175 if (dfd < 0) {
2176 pr_err("%s: cannot open component %s\n",
2177 devname, dn ? dn : "-unknown-");
2178 goto release;
2179 }
2180 st2 = dup_super(st);
2181 rv = st2->ss->load_super(st2,dfd, NULL);
2182 close(dfd);
2183 if (rv) {
2184 free(st2);
2185 pr_err("%s: cannot get superblock from %s\n",
2186 devname, dn);
2187 goto release;
2188 }
2189 st2->ss->getinfo_super(st2, &info2, NULL);
2190 st2->ss->free_super(st2);
2191 free(st2);
2192 if (info2.space_before == 0 &&
2193 info2.space_after == 0) {
2194 /* Metadata doesn't support data_offset changes */
2195 return 1;
2196 }
2197 if (before > info2.space_before)
2198 before = info2.space_before;
2199 if (after > info2.space_after)
2200 after = info2.space_after;
2201
2202 if (data_offset != INVALID_SECTORS) {
2203 if (dir == 0) {
2204 if (info2.data_offset == data_offset) {
2205 pr_err("%s: already has that data_offset\n",
2206 dn);
2207 goto release;
2208 }
2209 if (data_offset < info2.data_offset)
2210 dir = -1;
2211 else
2212 dir = 1;
2213 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2214 (data_offset >= info2.data_offset && dir == -1)) {
2215 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2216 dn);
2217 goto release;
2218 }
2219 }
2220 }
2221 if (before == UINT64_MAX)
2222 /* impossible really, there must be no devices */
2223 return 1;
2224
2225 for (sd = sra->devs; sd; sd = sd->next) {
2226 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2227 unsigned long long new_data_offset;
2228
2229 if (delta_disks < 0) {
2230 /* Don't need any space as array is shrinking
2231 * just move data_offset up by min
2232 */
2233 if (data_offset == INVALID_SECTORS)
2234 new_data_offset = sd->data_offset + min;
2235 else {
2236 if (data_offset < sd->data_offset + min) {
2237 pr_err("--data-offset too small for %s\n",
2238 dn);
2239 goto release;
2240 }
2241 new_data_offset = data_offset;
2242 }
2243 } else if (delta_disks > 0) {
2244 /* need space before */
2245 if (before < min) {
2246 pr_err("Insufficient head-space for reshape on %s\n",
2247 dn);
2248 goto release;
2249 }
2250 if (data_offset == INVALID_SECTORS)
2251 new_data_offset = sd->data_offset - min;
2252 else {
2253 if (data_offset > sd->data_offset - min) {
2254 pr_err("--data-offset too large for %s\n",
2255 dn);
2256 goto release;
2257 }
2258 new_data_offset = data_offset;
2259 }
2260 } else {
2261 if (dir == 0) {
2262 /* can move up or down. If 'data_offset'
2263 * was set we would have already decided,
2264 * so just choose direction with most space.
2265 */
2266 if (before > after)
2267 dir = -1;
2268 else
2269 dir = 1;
2270 sysfs_set_str(sra, NULL, "reshape_direction",
2271 dir == 1 ? "backwards" : "forwards");
2272 }
2273 if (dir > 0) {
2274 /* Increase data offset */
2275 if (after < min) {
2276 pr_err("Insufficient tail-space for reshape on %s\n",
2277 dn);
2278 goto release;
2279 }
2280 if (data_offset != INVALID_SECTORS &&
2281 data_offset < sd->data_offset + min) {
2282 pr_err("--data-offset too small on %s\n",
2283 dn);
2284 goto release;
2285 }
2286 if (data_offset != INVALID_SECTORS)
2287 new_data_offset = data_offset;
2288 else {
2289 unsigned long long off = after / 2;
2290 off &= ~7ULL;
2291 if (off < min)
2292 off = min;
2293 new_data_offset =
2294 sd->data_offset + off;
2295 }
2296 } else {
2297 /* Decrease data offset */
2298 if (before < min) {
2299 pr_err("insufficient head-room on %s\n",
2300 dn);
2301 goto release;
2302 }
2303 if (data_offset != INVALID_SECTORS &&
2304 data_offset < sd->data_offset - min) {
2305 pr_err("--data-offset too small on %s\n",
2306 dn);
2307 goto release;
2308 }
2309 if (data_offset != INVALID_SECTORS)
2310 new_data_offset = data_offset;
2311 else {
2312 unsigned long long off = before / 2;
2313 off &= ~7ULL;
2314 if (off < min)
2315 off = min;
2316 new_data_offset =
2317 sd->data_offset - off;
2318 }
2319 }
2320 }
2321 if (sysfs_set_num(sra, sd, "new_offset",
2322 new_data_offset) < 0) {
2323 err = errno;
2324 err = -1;
2325 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2326 pr_err("data-offset is too big for %s\n",
2327 dn);
2328 goto release;
2329 }
2330 if (sd == sra->devs &&
2331 (errno == ENOENT || errno == E2BIG))
2332 /* Early kernel, no 'new_offset' file,
2333 * or kernel doesn't like us.
2334 * For RAID5/6 this is not fatal
2335 */
2336 return 1;
2337 pr_err("Cannot set new_offset for %s\n",
2338 dn);
2339 break;
2340 }
2341 }
2342 return err;
2343 release:
2344 return -1;
2345 }
2346
2347 static int raid10_reshape(char *container, int fd, char *devname,
2348 struct supertype *st, struct mdinfo *info,
2349 struct reshape *reshape,
2350 unsigned long long data_offset,
2351 int force, int verbose)
2352 {
2353 /* Changing raid_disks, layout, chunksize or possibly
2354 * just data_offset for a RAID10.
2355 * We must always change data_offset. We change by at least
2356 * ->min_offset_change which is the largest of the old and new
2357 * chunk sizes.
2358 * If raid_disks is increasing, then data_offset must decrease
2359 * by at least this copy size.
2360 * If raid_disks is unchanged, data_offset must increase or
2361 * decrease by at least min_offset_change but preferably by much more.
2362 * We choose half of the available space.
2363 * If raid_disks is decreasing, data_offset must increase by
2364 * at least min_offset_change. To allow of this, component_size
2365 * must be decreased by the same amount.
2366 *
2367 * So we calculate the required minimum and direction, possibly
2368 * reduce the component_size, then iterate through the devices
2369 * and set the new_data_offset.
2370 * If that all works, we set chunk_size, layout, raid_disks, and start
2371 * 'reshape'
2372 */
2373 struct mdinfo *sra;
2374 unsigned long long min;
2375 int err = 0;
2376
2377 sra = sysfs_read(fd, NULL,
2378 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2379 );
2380 if (!sra) {
2381 pr_err("%s: Cannot get array details from sysfs\n",
2382 devname);
2383 goto release;
2384 }
2385 min = reshape->min_offset_change;
2386
2387 if (info->delta_disks)
2388 sysfs_set_str(sra, NULL, "reshape_direction",
2389 info->delta_disks < 0 ? "backwards" : "forwards");
2390 if (info->delta_disks < 0 &&
2391 info->space_after < min) {
2392 int rv = sysfs_set_num(sra, NULL, "component_size",
2393 (sra->component_size -
2394 min)/2);
2395 if (rv) {
2396 pr_err("cannot reduce component size\n");
2397 goto release;
2398 }
2399 }
2400 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2401 min);
2402 if (err == 1) {
2403 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2404 cont_err("supported on this kernel\n");
2405 err = -1;
2406 }
2407 if (err < 0)
2408 goto release;
2409
2410 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2411 err = errno;
2412 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2413 err = errno;
2414 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2415 info->array.raid_disks + info->delta_disks) < 0)
2416 err = errno;
2417 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2418 err = errno;
2419 if (err) {
2420 pr_err("Cannot set array shape for %s\n",
2421 devname);
2422 if (err == EBUSY &&
2423 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2424 cont_err(" Bitmap must be removed before"
2425 " shape can be changed\n");
2426 goto release;
2427 }
2428 sysfs_free(sra);
2429 return 0;
2430 release:
2431 sysfs_free(sra);
2432 return 1;
2433 }
2434
2435 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2436 {
2437 struct mdinfo *sra, *sd;
2438 /* Initialisation to silence compiler warning */
2439 unsigned long long min_space_before = 0, min_space_after = 0;
2440 int first = 1;
2441
2442 sra = sysfs_read(fd, NULL, GET_DEVS);
2443 if (!sra)
2444 return;
2445 for (sd = sra->devs; sd; sd = sd->next) {
2446 char *dn;
2447 int dfd;
2448 struct supertype *st2;
2449 struct mdinfo info2;
2450
2451 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2452 continue;
2453 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2454 dfd = dev_open(dn, O_RDONLY);
2455 if (dfd < 0)
2456 break;
2457 st2 = dup_super(st);
2458 if (st2->ss->load_super(st2,dfd, NULL)) {
2459 close(dfd);
2460 free(st2);
2461 break;
2462 }
2463 close(dfd);
2464 st2->ss->getinfo_super(st2, &info2, NULL);
2465 st2->ss->free_super(st2);
2466 free(st2);
2467 if (first ||
2468 min_space_before > info2.space_before)
2469 min_space_before = info2.space_before;
2470 if (first ||
2471 min_space_after > info2.space_after)
2472 min_space_after = info2.space_after;
2473 first = 0;
2474 }
2475 if (sd == NULL && !first) {
2476 info->space_after = min_space_after;
2477 info->space_before = min_space_before;
2478 }
2479 sysfs_free(sra);
2480 }
2481
2482 static void update_cache_size(char *container, struct mdinfo *sra,
2483 struct mdinfo *info,
2484 int disks, unsigned long long blocks)
2485 {
2486 /* Check that the internal stripe cache is
2487 * large enough, or it won't work.
2488 * It must hold at least 4 stripes of the larger
2489 * chunk size
2490 */
2491 unsigned long cache;
2492 cache = max(info->array.chunk_size, info->new_chunk);
2493 cache *= 4; /* 4 stripes minimum */
2494 cache /= 512; /* convert to sectors */
2495 /* make sure there is room for 'blocks' with a bit to spare */
2496 if (cache < 16 + blocks / disks)
2497 cache = 16 + blocks / disks;
2498 cache /= (4096/512); /* Covert from sectors to pages */
2499
2500 if (sra->cache_size < cache)
2501 subarray_set_num(container, sra, "stripe_cache_size",
2502 cache+1);
2503 }
2504
2505 static int impose_reshape(struct mdinfo *sra,
2506 struct mdinfo *info,
2507 struct supertype *st,
2508 int fd,
2509 int restart,
2510 char *devname, char *container,
2511 struct reshape *reshape)
2512 {
2513 struct mdu_array_info_s array;
2514
2515 sra->new_chunk = info->new_chunk;
2516
2517 if (restart) {
2518 /* for external metadata checkpoint saved by mdmon can be lost
2519 * or missed /due to e.g. crash/. Check if md is not during
2520 * restart farther than metadata points to.
2521 * If so, this means metadata information is obsolete.
2522 */
2523 if (st->ss->external)
2524 verify_reshape_position(info, reshape->level);
2525 sra->reshape_progress = info->reshape_progress;
2526 } else {
2527 sra->reshape_progress = 0;
2528 if (reshape->after.data_disks < reshape->before.data_disks)
2529 /* start from the end of the new array */
2530 sra->reshape_progress = (sra->component_size
2531 * reshape->after.data_disks);
2532 }
2533
2534 ioctl(fd, GET_ARRAY_INFO, &array);
2535 if (info->array.chunk_size == info->new_chunk &&
2536 reshape->before.layout == reshape->after.layout &&
2537 st->ss->external == 0) {
2538 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2539 array.raid_disks = reshape->after.data_disks + reshape->parity;
2540 if (!restart &&
2541 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2542 int err = errno;
2543
2544 pr_err("Cannot set device shape for %s: %s\n",
2545 devname, strerror(errno));
2546
2547 if (err == EBUSY &&
2548 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2549 cont_err("Bitmap must be removed before"
2550 " shape can be changed\n");
2551
2552 goto release;
2553 }
2554 } else if (!restart) {
2555 /* set them all just in case some old 'new_*' value
2556 * persists from some earlier problem.
2557 */
2558 int err = 0;
2559 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2560 err = errno;
2561 if (!err && sysfs_set_num(sra, NULL, "layout",
2562 reshape->after.layout) < 0)
2563 err = errno;
2564 if (!err && subarray_set_num(container, sra, "raid_disks",
2565 reshape->after.data_disks +
2566 reshape->parity) < 0)
2567 err = errno;
2568 if (err) {
2569 pr_err("Cannot set device shape for %s\n",
2570 devname);
2571
2572 if (err == EBUSY &&
2573 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2574 cont_err("Bitmap must be removed before"
2575 " shape can be changed\n");
2576 goto release;
2577 }
2578 }
2579 return 0;
2580 release:
2581 return -1;
2582 }
2583
2584 static int reshape_array(char *container, int fd, char *devname,
2585 struct supertype *st, struct mdinfo *info,
2586 int force, struct mddev_dev *devlist,
2587 unsigned long long data_offset,
2588 char *backup_file, int verbose, int forked,
2589 int restart, int freeze_reshape)
2590 {
2591 struct reshape reshape;
2592 int spares_needed;
2593 char *msg;
2594 int orig_level = UnSet;
2595 int odisks;
2596 int delayed;
2597
2598 struct mdu_array_info_s array;
2599 char *c;
2600
2601 struct mddev_dev *dv;
2602 int added_disks;
2603
2604 int *fdlist = NULL;
2605 unsigned long long *offsets = NULL;
2606 int d;
2607 int nrdisks;
2608 int err;
2609 unsigned long blocks;
2610 unsigned long long array_size;
2611 int done;
2612 struct mdinfo *sra = NULL;
2613
2614 /* when reshaping a RAID0, the component_size might be zero.
2615 * So try to fix that up.
2616 */
2617 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2618 dprintf("Cannot get array information.\n");
2619 goto release;
2620 }
2621 if (array.level == 0 && info->component_size == 0) {
2622 get_dev_size(fd, NULL, &array_size);
2623 info->component_size = array_size / array.raid_disks;
2624 }
2625
2626 if (array.level == 10)
2627 /* Need space_after info */
2628 get_space_after(fd, st, info);
2629
2630 if (info->reshape_active) {
2631 int new_level = info->new_level;
2632 info->new_level = UnSet;
2633 if (info->delta_disks > 0)
2634 info->array.raid_disks -= info->delta_disks;
2635 msg = analyse_change(info, &reshape);
2636 info->new_level = new_level;
2637 if (info->delta_disks > 0)
2638 info->array.raid_disks += info->delta_disks;
2639 if (!restart)
2640 /* Make sure the array isn't read-only */
2641 ioctl(fd, RESTART_ARRAY_RW, 0);
2642 } else
2643 msg = analyse_change(info, &reshape);
2644 if (msg) {
2645 pr_err("%s\n", msg);
2646 goto release;
2647 }
2648 if (restart &&
2649 (reshape.level != info->array.level ||
2650 reshape.before.layout != info->array.layout ||
2651 reshape.before.data_disks + reshape.parity
2652 != info->array.raid_disks - max(0, info->delta_disks))) {
2653 pr_err("reshape info is not in native format -"
2654 " cannot continue.\n");
2655 goto release;
2656 }
2657
2658 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2659 /* When reshape is restarted from '0', very begin of array
2660 * it is possible that for external metadata reshape and array
2661 * configuration doesn't happen.
2662 * Check if md has the same opinion, and reshape is restarted
2663 * from 0. If so, this is regular reshape start after reshape
2664 * switch in metadata to next array only.
2665 */
2666 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2667 (info->reshape_progress == 0))
2668 restart = 0;
2669 }
2670 if (restart) {
2671 /* reshape already started. just skip to monitoring the reshape */
2672 if (reshape.backup_blocks == 0)
2673 return 0;
2674 if (restart & RESHAPE_NO_BACKUP)
2675 return 0;
2676 goto started;
2677 }
2678 /* The container is frozen but the array may not be.
2679 * So freeze the array so spares don't get put to the wrong use
2680 * FIXME there should probably be a cleaner separation between
2681 * freeze_array and freeze_container.
2682 */
2683 sysfs_freeze_array(info);
2684 /* Check we have enough spares to not be degraded */
2685 added_disks = 0;
2686 for (dv = devlist; dv ; dv=dv->next)
2687 added_disks++;
2688 spares_needed = max(reshape.before.data_disks,
2689 reshape.after.data_disks)
2690 + reshape.parity - array.raid_disks;
2691
2692 if (!force &&
2693 info->new_level > 1 && info->array.level > 1 &&
2694 spares_needed > info->array.spare_disks + added_disks) {
2695 pr_err("Need %d spare%s to avoid degraded array,"
2696 " and only have %d.\n"
2697 " Use --force to over-ride this check.\n",
2698 spares_needed,
2699 spares_needed == 1 ? "" : "s",
2700 info->array.spare_disks + added_disks);
2701 goto release;
2702 }
2703 /* Check we have enough spares to not fail */
2704 spares_needed = max(reshape.before.data_disks,
2705 reshape.after.data_disks)
2706 - array.raid_disks;
2707 if ((info->new_level > 1 || info->new_level == 0) &&
2708 spares_needed > info->array.spare_disks +added_disks) {
2709 pr_err("Need %d spare%s to create working array,"
2710 " and only have %d.\n",
2711 spares_needed,
2712 spares_needed == 1 ? "" : "s",
2713 info->array.spare_disks + added_disks);
2714 goto release;
2715 }
2716
2717 if (reshape.level != array.level) {
2718 char *c = map_num(pers, reshape.level);
2719 int err;
2720 if (c == NULL)
2721 goto release;
2722
2723 err = sysfs_set_str(info, NULL, "level", c);
2724 if (err) {
2725 err = errno;
2726 pr_err("%s: could not set level to %s\n",
2727 devname, c);
2728 if (err == EBUSY &&
2729 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2730 cont_err("Bitmap must be removed"
2731 " before level can be changed\n");
2732 goto release;
2733 }
2734 if (verbose >= 0)
2735 pr_err("level of %s changed to %s\n",
2736 devname, c);
2737 orig_level = array.level;
2738 sysfs_freeze_array(info);
2739
2740 if (reshape.level > 0 && st->ss->external) {
2741 /* make sure mdmon is aware of the new level */
2742 if (mdmon_running(container))
2743 flush_mdmon(container);
2744
2745 if (!mdmon_running(container))
2746 start_mdmon(container);
2747 ping_monitor(container);
2748 if (mdmon_running(container) &&
2749 st->update_tail == NULL)
2750 st->update_tail = &st->updates;
2751 }
2752 }
2753 /* ->reshape_super might have chosen some spares from the
2754 * container that it wants to be part of the new array.
2755 * We can collect them with ->container_content and give
2756 * them to the kernel.
2757 */
2758 if (st->ss->reshape_super && st->ss->container_content) {
2759 char *subarray = strchr(info->text_version+1, '/')+1;
2760 struct mdinfo *info2 =
2761 st->ss->container_content(st, subarray);
2762 struct mdinfo *d;
2763
2764 if (info2) {
2765 sysfs_init(info2, fd, st->devnm);
2766 /* When increasing number of devices, we need to set
2767 * new raid_disks before adding these, or they might
2768 * be rejected.
2769 */
2770 if (reshape.backup_blocks &&
2771 reshape.after.data_disks > reshape.before.data_disks)
2772 subarray_set_num(container, info2, "raid_disks",
2773 reshape.after.data_disks +
2774 reshape.parity);
2775 for (d = info2->devs; d; d = d->next) {
2776 if (d->disk.state == 0 &&
2777 d->disk.raid_disk >= 0) {
2778 /* This is a spare that wants to
2779 * be part of the array.
2780 */
2781 add_disk(fd, st, info2, d);
2782 }
2783 }
2784 sysfs_free(info2);
2785 }
2786 }
2787 /* We might have been given some devices to add to the
2788 * array. Now that the array has been changed to the right
2789 * level and frozen, we can safely add them.
2790 */
2791 if (devlist)
2792 Manage_subdevs(devname, fd, devlist, verbose,
2793 0,NULL, 0);
2794
2795 if (reshape.backup_blocks == 0 && data_offset)
2796 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
2797 if (reshape.backup_blocks == 0) {
2798 /* No restriping needed, but we might need to impose
2799 * some more changes: layout, raid_disks, chunk_size
2800 */
2801 /* read current array info */
2802 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2803 dprintf("Cannot get array information.\n");
2804 goto release;
2805 }
2806 /* compare current array info with new values and if
2807 * it is different update them to new */
2808 if (info->new_layout != UnSet &&
2809 info->new_layout != array.layout) {
2810 array.layout = info->new_layout;
2811 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2812 pr_err("failed to set new layout\n");
2813 goto release;
2814 } else if (verbose >= 0)
2815 printf("layout for %s set to %d\n",
2816 devname, array.layout);
2817 }
2818 if (info->delta_disks != UnSet &&
2819 info->delta_disks != 0 &&
2820 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
2821 array.raid_disks += info->delta_disks;
2822 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2823 pr_err("failed to set raid disks\n");
2824 goto release;
2825 } else if (verbose >= 0) {
2826 printf("raid_disks for %s set to %d\n",
2827 devname, array.raid_disks);
2828 }
2829 }
2830 if (info->new_chunk != 0 &&
2831 info->new_chunk != array.chunk_size) {
2832 if (sysfs_set_num(info, NULL,
2833 "chunk_size", info->new_chunk) != 0) {
2834 pr_err("failed to set chunk size\n");
2835 goto release;
2836 } else if (verbose >= 0)
2837 printf("chunk size for %s set to %d\n",
2838 devname, array.chunk_size);
2839 }
2840 unfreeze(st);
2841 return 0;
2842 }
2843
2844 /*
2845 * There are three possibilities.
2846 * 1/ The array will shrink.
2847 * We need to ensure the reshape will pause before reaching
2848 * the 'critical section'. We also need to fork and wait for
2849 * that to happen. When it does we
2850 * suspend/backup/complete/unfreeze
2851 *
2852 * 2/ The array will not change size.
2853 * This requires that we keep a backup of a sliding window
2854 * so that we can restore data after a crash. So we need
2855 * to fork and monitor progress.
2856 * In future we will allow the data_offset to change, so
2857 * a sliding backup becomes unnecessary.
2858 *
2859 * 3/ The array will grow. This is relatively easy.
2860 * However the kernel's restripe routines will cheerfully
2861 * overwrite some early data before it is safe. So we
2862 * need to make a backup of the early parts of the array
2863 * and be ready to restore it if rebuild aborts very early.
2864 * For externally managed metadata, we still need a forked
2865 * child to monitor the reshape and suspend IO over the region
2866 * that is being reshaped.
2867 *
2868 * We backup data by writing it to one spare, or to a
2869 * file which was given on command line.
2870 *
2871 * In each case, we first make sure that storage is available
2872 * for the required backup.
2873 * Then we:
2874 * - request the shape change.
2875 * - fork to handle backup etc.
2876 */
2877 /* Check that we can hold all the data */
2878 get_dev_size(fd, NULL, &array_size);
2879 if (reshape.new_size < (array_size/512)) {
2880 pr_err("this change will reduce the size of the array.\n"
2881 " use --grow --array-size first to truncate array.\n"
2882 " e.g. mdadm --grow %s --array-size %llu\n",
2883 devname, reshape.new_size/2);
2884 goto release;
2885 }
2886
2887 if (array.level == 10) {
2888 /* Reshaping RAID10 does not require any data backup by
2889 * user-space. Instead it requires that the data_offset
2890 * is changed to avoid the need for backup.
2891 * So this is handled very separately
2892 */
2893 if (restart)
2894 /* Nothing to do. */
2895 return 0;
2896 return raid10_reshape(container, fd, devname, st, info,
2897 &reshape, data_offset,
2898 force, verbose);
2899 }
2900 sra = sysfs_read(fd, NULL,
2901 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2902 GET_CACHE);
2903 if (!sra) {
2904 pr_err("%s: Cannot get array details from sysfs\n",
2905 devname);
2906 goto release;
2907 }
2908
2909 if (!backup_file)
2910 switch(set_new_data_offset(sra, st, devname,
2911 reshape.after.data_disks - reshape.before.data_disks,
2912 data_offset,
2913 reshape.min_offset_change)) {
2914 case -1:
2915 goto release;
2916 case 0:
2917 /* Updated data_offset, so it's easy now */
2918 update_cache_size(container, sra, info,
2919 min(reshape.before.data_disks,
2920 reshape.after.data_disks),
2921 reshape.backup_blocks);
2922
2923 /* Right, everything seems fine. Let's kick things off.
2924 */
2925 sync_metadata(st);
2926
2927 if (impose_reshape(sra, info, st, fd, restart,
2928 devname, container, &reshape) < 0)
2929 goto release;
2930 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
2931 pr_err("Failed to initiate reshape!\n");
2932 goto release;
2933 }
2934
2935 return 0;
2936 case 1: /* Couldn't set data_offset, try the old way */
2937 if (data_offset != INVALID_SECTORS) {
2938 pr_err("Cannot update data_offset on this array\n");
2939 goto release;
2940 }
2941 break;
2942 }
2943
2944 started:
2945 /* Decide how many blocks (sectors) for a reshape
2946 * unit. The number we have so far is just a minimum
2947 */
2948 blocks = reshape.backup_blocks;
2949 if (reshape.before.data_disks ==
2950 reshape.after.data_disks) {
2951 /* Make 'blocks' bigger for better throughput, but
2952 * not so big that we reject it below.
2953 * Try for 16 megabytes
2954 */
2955 while (blocks * 32 < sra->component_size &&
2956 blocks < 16*1024*2)
2957 blocks *= 2;
2958 } else
2959 pr_err("Need to backup %luK of critical "
2960 "section..\n", blocks/2);
2961
2962 if (blocks >= sra->component_size/2) {
2963 pr_err("%s: Something wrong"
2964 " - reshape aborted\n",
2965 devname);
2966 goto release;
2967 }
2968
2969 /* Now we need to open all these devices so we can read/write.
2970 */
2971 nrdisks = max(reshape.before.data_disks,
2972 reshape.after.data_disks) + reshape.parity
2973 + sra->array.spare_disks;
2974 fdlist = xcalloc((1+nrdisks), sizeof(int));
2975 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
2976
2977 odisks = reshape.before.data_disks + reshape.parity;
2978 d = reshape_prepare_fdlist(devname, sra, odisks,
2979 nrdisks, blocks, backup_file,
2980 fdlist, offsets);
2981 if (d < 0) {
2982 goto release;
2983 }
2984 if ((st->ss->manage_reshape == NULL) ||
2985 (st->ss->recover_backup == NULL)) {
2986 if (backup_file == NULL) {
2987 if (reshape.after.data_disks <=
2988 reshape.before.data_disks) {
2989 pr_err("%s: Cannot grow - "
2990 "need backup-file\n", devname);
2991 goto release;
2992 } else if (sra->array.spare_disks == 0) {
2993 pr_err("%s: Cannot grow - "
2994 "need a spare or backup-file to backup "
2995 "critical section\n", devname);
2996 goto release;
2997 }
2998 } else {
2999 if (!reshape_open_backup_file(backup_file, fd, devname,
3000 (signed)blocks,
3001 fdlist+d, offsets+d,
3002 restart)) {
3003 goto release;
3004 }
3005 d++;
3006 }
3007 }
3008
3009 update_cache_size(container, sra, info,
3010 min(reshape.before.data_disks, reshape.after.data_disks),
3011 blocks);
3012
3013 /* Right, everything seems fine. Let's kick things off.
3014 * If only changing raid_disks, use ioctl, else use
3015 * sysfs.
3016 */
3017 sync_metadata(st);
3018
3019 if (impose_reshape(sra, info, st, fd, restart,
3020 devname, container, &reshape) < 0)
3021 goto release;
3022
3023 err = start_reshape(sra, restart, reshape.before.data_disks,
3024 reshape.after.data_disks);
3025 if (err) {
3026 pr_err("Cannot %s reshape for %s\n",
3027 restart ? "continue" : "start",
3028 devname);
3029 goto release;
3030 }
3031 if (restart)
3032 sysfs_set_str(sra, NULL, "array_state", "active");
3033 if (freeze_reshape) {
3034 free(fdlist);
3035 free(offsets);
3036 sysfs_free(sra);
3037 pr_err("Reshape has to be continued from"
3038 " location %llu when root filesystem has been mounted.\n",
3039 sra->reshape_progress);
3040 return 1;
3041 }
3042
3043 /* Now we just need to kick off the reshape and watch, while
3044 * handling backups of the data...
3045 * This is all done by a forked background process.
3046 */
3047 switch(forked ? 0 : fork()) {
3048 case -1:
3049 pr_err("Cannot run child to monitor reshape: %s\n",
3050 strerror(errno));
3051 abort_reshape(sra);
3052 goto release;
3053 default:
3054 free(fdlist);
3055 free(offsets);
3056 sysfs_free(sra);
3057 return 0;
3058 case 0:
3059 map_fork();
3060 break;
3061 }
3062
3063 /* If another array on the same devices is busy, the
3064 * reshape will wait for them. This would mean that
3065 * the first section that we suspend will stay suspended
3066 * for a long time. So check on that possibility
3067 * by looking for "DELAYED" in /proc/mdstat, and if found,
3068 * wait a while
3069 */
3070 do {
3071 struct mdstat_ent *mds, *m;
3072 delayed = 0;
3073 mds = mdstat_read(0, 0);
3074 for (m = mds; m; m = m->next)
3075 if (strcmp(m->devnm, sra->sys_name) == 0) {
3076 if (m->resync &&
3077 m->percent == RESYNC_DELAYED)
3078 delayed = 1;
3079 if (m->resync == 0)
3080 /* Haven't started the reshape thread
3081 * yet, wait a bit
3082 */
3083 delayed = 2;
3084 break;
3085 }
3086 free_mdstat(mds);
3087 if (delayed == 1 && get_linux_version() < 3007000) {
3088 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3089 " You might experience problems until other reshapes complete.\n");
3090 delayed = 0;
3091 }
3092 if (delayed)
3093 sleep(30 - (delayed-1) * 25);
3094 } while (delayed);
3095
3096 close(fd);
3097 if (check_env("MDADM_GROW_VERIFY"))
3098 fd = open(devname, O_RDONLY | O_DIRECT);
3099 else
3100 fd = -1;
3101 mlockall(MCL_FUTURE);
3102
3103 if (st->ss->external) {
3104 /* metadata handler takes it from here */
3105 done = st->ss->manage_reshape(
3106 fd, sra, &reshape, st, blocks,
3107 fdlist, offsets,
3108 d - odisks, fdlist+odisks,
3109 offsets+odisks);
3110 } else
3111 done = child_monitor(
3112 fd, sra, &reshape, st, blocks,
3113 fdlist, offsets,
3114 d - odisks, fdlist+odisks,
3115 offsets+odisks);
3116
3117 free(fdlist);
3118 free(offsets);
3119
3120 if (backup_file && done)
3121 unlink(backup_file);
3122 if (!done) {
3123 abort_reshape(sra);
3124 goto out;
3125 }
3126
3127 if (!st->ss->external &&
3128 !(reshape.before.data_disks != reshape.after.data_disks
3129 && info->custom_array_size) &&
3130 info->new_level == reshape.level &&
3131 !forked) {
3132 /* no need to wait for the reshape to finish as
3133 * there is nothing more to do.
3134 */
3135 sysfs_free(sra);
3136 exit(0);
3137 }
3138 wait_reshape(sra);
3139
3140 if (st->ss->external) {
3141 /* Re-load the metadata as much could have changed */
3142 int cfd = open_dev(st->container_devnm);
3143 if (cfd >= 0) {
3144 flush_mdmon(container);
3145 st->ss->free_super(st);
3146 st->ss->load_container(st, cfd, container);
3147 close(cfd);
3148 }
3149 }
3150
3151 /* set new array size if required customer_array_size is used
3152 * by this metadata.
3153 */
3154 if (reshape.before.data_disks !=
3155 reshape.after.data_disks &&
3156 info->custom_array_size)
3157 set_array_size(st, info, info->text_version);
3158
3159 if (info->new_level != reshape.level) {
3160
3161 c = map_num(pers, info->new_level);
3162 if (c) {
3163 err = sysfs_set_str(sra, NULL, "level", c);
3164 if (err)
3165 pr_err("%s: could not set level "
3166 "to %s\n", devname, c);
3167 }
3168 if (info->new_level == 0)
3169 st->update_tail = NULL;
3170 }
3171 out:
3172 sysfs_free(sra);
3173 if (forked)
3174 return 0;
3175 unfreeze(st);
3176 exit(0);
3177
3178 release:
3179 free(fdlist);
3180 free(offsets);
3181 if (orig_level != UnSet && sra) {
3182 c = map_num(pers, orig_level);
3183 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3184 pr_err("aborting level change\n");
3185 }
3186 sysfs_free(sra);
3187 if (!forked)
3188 unfreeze(st);
3189 return 1;
3190 }
3191
3192 /* mdfd handle is passed to be closed in child process (after fork).
3193 */
3194 int reshape_container(char *container, char *devname,
3195 int mdfd,
3196 struct supertype *st,
3197 struct mdinfo *info,
3198 int force,
3199 char *backup_file,
3200 int verbose, int restart, int freeze_reshape)
3201 {
3202 struct mdinfo *cc = NULL;
3203 int rv = restart;
3204 char last_devnm[32] = "";
3205
3206 /* component_size is not meaningful for a container,
3207 * so pass '0' meaning 'no change'
3208 */
3209 if (!restart &&
3210 reshape_super(st, 0, info->new_level,
3211 info->new_layout, info->new_chunk,
3212 info->array.raid_disks, info->delta_disks,
3213 backup_file, devname, APPLY_METADATA_CHANGES,
3214 verbose)) {
3215 unfreeze(st);
3216 return 1;
3217 }
3218
3219 sync_metadata(st);
3220
3221 /* ping monitor to be sure that update is on disk
3222 */
3223 ping_monitor(container);
3224
3225 switch (fork()) {
3226 case -1: /* error */
3227 perror("Cannot fork to complete reshape\n");
3228 unfreeze(st);
3229 return 1;
3230 default: /* parent */
3231 if (!freeze_reshape)
3232 printf(Name ": multi-array reshape continues"
3233 " in background\n");
3234 return 0;
3235 case 0: /* child */
3236 map_fork();
3237 break;
3238 }
3239
3240 /* close unused handle in child process
3241 */
3242 if (mdfd > -1)
3243 close(mdfd);
3244
3245 while(1) {
3246 /* For each member array with reshape_active,
3247 * we need to perform the reshape.
3248 * We pick the first array that needs reshaping and
3249 * reshape it. reshape_array() will re-read the metadata
3250 * so the next time through a different array should be
3251 * ready for reshape.
3252 * It is possible that the 'different' array will not
3253 * be assembled yet. In that case we simple exit.
3254 * When it is assembled, the mdadm which assembles it
3255 * will take over the reshape.
3256 */
3257 struct mdinfo *content;
3258 int fd;
3259 struct mdstat_ent *mdstat;
3260 char *adev;
3261 int devid;
3262
3263 sysfs_free(cc);
3264
3265 cc = st->ss->container_content(st, NULL);
3266
3267 for (content = cc; content ; content = content->next) {
3268 char *subarray;
3269 if (!content->reshape_active)
3270 continue;
3271
3272 subarray = strchr(content->text_version+1, '/')+1;
3273 mdstat = mdstat_by_subdev(subarray, container);
3274 if (!mdstat)
3275 continue;
3276 if (mdstat->active == 0) {
3277 pr_err("Skipping inactive array %s.\n",
3278 mdstat->devnm);
3279 free_mdstat(mdstat);
3280 mdstat = NULL;
3281 continue;
3282 }
3283 break;
3284 }
3285 if (!content)
3286 break;
3287
3288 devid = devnm2devid(mdstat->devnm);
3289 adev = map_dev(major(devid), minor(devid), 0);
3290 if (!adev)
3291 adev = content->text_version;
3292
3293 fd = open_dev(mdstat->devnm);
3294 if (fd < 0) {
3295 printf(Name ": Device %s cannot be opened for reshape.",
3296 adev);
3297 break;
3298 }
3299
3300 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3301 /* Do not allow for multiple reshape_array() calls for
3302 * the same array.
3303 * It can happen when reshape_array() returns without
3304 * error, when reshape is not finished (wrong reshape
3305 * starting/continuation conditions). Mdmon doesn't
3306 * switch to next array in container and reentry
3307 * conditions for the same array occur.
3308 * This is possibly interim until the behaviour of
3309 * reshape_array is resolved().
3310 */
3311 printf(Name ": Multiple reshape execution detected for "
3312 "device %s.", adev);
3313 close(fd);
3314 break;
3315 }
3316 strcpy(last_devnm, mdstat->devnm);
3317
3318 sysfs_init(content, fd, mdstat->devnm);
3319
3320 if (mdmon_running(container))
3321 flush_mdmon(container);
3322
3323 rv = reshape_array(container, fd, adev, st,
3324 content, force, NULL, 0ULL,
3325 backup_file, verbose, 1, restart,
3326 freeze_reshape);
3327 close(fd);
3328
3329 if (freeze_reshape) {
3330 sysfs_free(cc);
3331 exit(0);
3332 }
3333
3334 restart = 0;
3335 if (rv)
3336 break;
3337
3338 if (mdmon_running(container))
3339 flush_mdmon(container);
3340 }
3341 if (!rv)
3342 unfreeze(st);
3343 sysfs_free(cc);
3344 exit(0);
3345 }
3346
3347 /*
3348 * We run a child process in the background which performs the following
3349 * steps:
3350 * - wait for resync to reach a certain point
3351 * - suspend io to the following section
3352 * - backup that section
3353 * - allow resync to proceed further
3354 * - resume io
3355 * - discard the backup.
3356 *
3357 * When are combined in slightly different ways in the three cases.
3358 * Grow:
3359 * - suspend/backup/allow/wait/resume/discard
3360 * Shrink:
3361 * - allow/wait/suspend/backup/allow/wait/resume/discard
3362 * same-size:
3363 * - wait/resume/discard/suspend/backup/allow
3364 *
3365 * suspend/backup/allow always come together
3366 * wait/resume/discard do too.
3367 * For the same-size case we have two backups to improve flow.
3368 *
3369 */
3370
3371 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3372 unsigned long long backup_point,
3373 unsigned long long wait_point,
3374 unsigned long long *suspend_point,
3375 unsigned long long *reshape_completed)
3376 {
3377 /* This function is called repeatedly by the reshape manager.
3378 * It determines how much progress can safely be made and allows
3379 * that progress.
3380 * - 'info' identifies the array and particularly records in
3381 * ->reshape_progress the metadata's knowledge of progress
3382 * This is a sector offset from the start of the array
3383 * of the next array block to be relocated. This number
3384 * may increase from 0 or decrease from array_size, depending
3385 * on the type of reshape that is happening.
3386 * Note that in contrast, 'sync_completed' is a block count of the
3387 * reshape so far. It gives the distance between the start point
3388 * (head or tail of device) and the next place that data will be
3389 * written. It always increases.
3390 * - 'reshape' is the structure created by analyse_change
3391 * - 'backup_point' shows how much the metadata manager has backed-up
3392 * data. For reshapes with increasing progress, it is the next address
3393 * to be backed up, previous addresses have been backed-up. For
3394 * decreasing progress, it is the earliest address that has been
3395 * backed up - later address are also backed up.
3396 * So addresses between reshape_progress and backup_point are
3397 * backed up providing those are in the 'correct' order.
3398 * - 'wait_point' is an array address. When reshape_completed
3399 * passes this point, progress_reshape should return. It might
3400 * return earlier if it determines that ->reshape_progress needs
3401 * to be updated or further backup is needed.
3402 * - suspend_point is maintained by progress_reshape and the caller
3403 * should not touch it except to initialise to zero.
3404 * It is an array address and it only increases in 2.6.37 and earlier.
3405 * This makes it difficult to handle reducing reshapes with
3406 * external metadata.
3407 * However: it is similar to backup_point in that it records the
3408 * other end of a suspended region from reshape_progress.
3409 * it is moved to extend the region that is safe to backup and/or
3410 * reshape
3411 * - reshape_completed is read from sysfs and returned. The caller
3412 * should copy this into ->reshape_progress when it has reason to
3413 * believe that the metadata knows this, and any backup outside this
3414 * has been erased.
3415 *
3416 * Return value is:
3417 * 1 if more data from backup_point - but only as far as suspend_point,
3418 * should be backed up
3419 * 0 if things are progressing smoothly
3420 * -1 if the reshape is finished because it is all done,
3421 * -2 if the reshape is finished due to an error.
3422 */
3423
3424 int advancing = (reshape->after.data_disks
3425 >= reshape->before.data_disks);
3426 unsigned long long need_backup; /* All data between start of array and
3427 * here will at some point need to
3428 * be backed up.
3429 */
3430 unsigned long long read_offset, write_offset;
3431 unsigned long long write_range;
3432 unsigned long long max_progress, target, completed;
3433 unsigned long long array_size = (info->component_size
3434 * reshape->before.data_disks);
3435 int fd;
3436 char buf[20];
3437
3438 /* First, we unsuspend any region that is now known to be safe.
3439 * If suspend_point is on the 'wrong' side of reshape_progress, then
3440 * we don't have or need suspension at the moment. This is true for
3441 * native metadata when we don't need to back-up.
3442 */
3443 if (advancing) {
3444 if (info->reshape_progress <= *suspend_point)
3445 sysfs_set_num(info, NULL, "suspend_lo",
3446 info->reshape_progress);
3447 } else {
3448 /* Note: this won't work in 2.6.37 and before.
3449 * Something somewhere should make sure we don't need it!
3450 */
3451 if (info->reshape_progress >= *suspend_point)
3452 sysfs_set_num(info, NULL, "suspend_hi",
3453 info->reshape_progress);
3454 }
3455
3456 /* Now work out how far it is safe to progress.
3457 * If the read_offset for ->reshape_progress is less than
3458 * 'blocks' beyond the write_offset, we can only progress as far
3459 * as a backup.
3460 * Otherwise we can progress until the write_offset for the new location
3461 * reaches (within 'blocks' of) the read_offset at the current location.
3462 * However that region must be suspended unless we are using native
3463 * metadata.
3464 * If we need to suspend more, we limit it to 128M per device, which is
3465 * rather arbitrary and should be some time-based calculation.
3466 */
3467 read_offset = info->reshape_progress / reshape->before.data_disks;
3468 write_offset = info->reshape_progress / reshape->after.data_disks;
3469 write_range = info->new_chunk/512;
3470 if (reshape->before.data_disks == reshape->after.data_disks)
3471 need_backup = array_size;
3472 else
3473 need_backup = reshape->backup_blocks;
3474 if (advancing) {
3475 if (read_offset < write_offset + write_range)
3476 max_progress = backup_point;
3477 else
3478 max_progress =
3479 read_offset *
3480 reshape->after.data_disks;
3481 } else {
3482 if (read_offset > write_offset - write_range)
3483 /* Can only progress as far as has been backed up,
3484 * which must be suspended */
3485 max_progress = backup_point;
3486 else if (info->reshape_progress <= need_backup)
3487 max_progress = backup_point;
3488 else {
3489 if (info->array.major_version >= 0)
3490 /* Can progress until backup is needed */
3491 max_progress = need_backup;
3492 else {
3493 /* Can progress until metadata update is required */
3494 max_progress =
3495 read_offset *
3496 reshape->after.data_disks;
3497 /* but data must be suspended */
3498 if (max_progress < *suspend_point)
3499 max_progress = *suspend_point;
3500 }
3501 }
3502 }
3503
3504 /* We know it is safe to progress to 'max_progress' providing
3505 * it is suspended or we are using native metadata.
3506 * Consider extending suspend_point 128M per device if it
3507 * is less than 64M per device beyond reshape_progress.
3508 * But always do a multiple of 'blocks'
3509 * FIXME this is too big - it takes to long to complete
3510 * this much.
3511 */
3512 target = 64*1024*2 * min(reshape->before.data_disks,
3513 reshape->after.data_disks);
3514 target /= reshape->backup_blocks;
3515 if (target < 2)
3516 target = 2;
3517 target *= reshape->backup_blocks;
3518
3519 /* For externally managed metadata we always need to suspend IO to
3520 * the area being reshaped so we regularly push suspend_point forward.
3521 * For native metadata we only need the suspend if we are going to do
3522 * a backup.
3523 */
3524 if (advancing) {
3525 if ((need_backup > info->reshape_progress
3526 || info->array.major_version < 0) &&
3527 *suspend_point < info->reshape_progress + target) {
3528 if (need_backup < *suspend_point + 2 * target)
3529 *suspend_point = need_backup;
3530 else if (*suspend_point + 2 * target < array_size)
3531 *suspend_point += 2 * target;
3532 else
3533 *suspend_point = array_size;
3534 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3535 if (max_progress > *suspend_point)
3536 max_progress = *suspend_point;
3537 }
3538 } else {
3539 if (info->array.major_version >= 0) {
3540 /* Only need to suspend when about to backup */
3541 if (info->reshape_progress < need_backup * 2 &&
3542 *suspend_point > 0) {
3543 *suspend_point = 0;
3544 sysfs_set_num(info, NULL, "suspend_lo", 0);
3545 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3546 }
3547 } else {
3548 /* Need to suspend continually */
3549 if (info->reshape_progress < *suspend_point)
3550 *suspend_point = info->reshape_progress;
3551 if (*suspend_point + target < info->reshape_progress)
3552 /* No need to move suspend region yet */;
3553 else {
3554 if (*suspend_point >= 2 * target)
3555 *suspend_point -= 2 * target;
3556 else
3557 *suspend_point = 0;
3558 sysfs_set_num(info, NULL, "suspend_lo",
3559 *suspend_point);
3560 }
3561 if (max_progress < *suspend_point)
3562 max_progress = *suspend_point;
3563 }
3564 }
3565
3566 /* now set sync_max to allow that progress. sync_max, like
3567 * sync_completed is a count of sectors written per device, so
3568 * we find the difference between max_progress and the start point,
3569 * and divide that by after.data_disks to get a sync_max
3570 * number.
3571 * At the same time we convert wait_point to a similar number
3572 * for comparing against sync_completed.
3573 */
3574 /* scale down max_progress to per_disk */
3575 max_progress /= reshape->after.data_disks;
3576 /* Round to chunk size as some kernels give an erroneously high number */
3577 max_progress /= info->new_chunk/512;
3578 max_progress *= info->new_chunk/512;
3579 /* And round to old chunk size as the kernel wants that */
3580 max_progress /= info->array.chunk_size/512;
3581 max_progress *= info->array.chunk_size/512;
3582 /* Limit progress to the whole device */
3583 if (max_progress > info->component_size)
3584 max_progress = info->component_size;
3585 wait_point /= reshape->after.data_disks;
3586 if (!advancing) {
3587 /* switch from 'device offset' to 'processed block count' */
3588 max_progress = info->component_size - max_progress;
3589 wait_point = info->component_size - wait_point;
3590 }
3591
3592 sysfs_set_num(info, NULL, "sync_max", max_progress);
3593
3594 /* Now wait. If we have already reached the point that we were
3595 * asked to wait to, don't wait at all, else wait for any change.
3596 * We need to select on 'sync_completed' as that is the place that
3597 * notifications happen, but we are really interested in
3598 * 'reshape_position'
3599 */
3600 fd = sysfs_get_fd(info, NULL, "sync_completed");
3601 if (fd < 0)
3602 goto check_progress;
3603
3604 if (sysfs_fd_get_ll(fd, &completed) < 0)
3605 goto check_progress;
3606
3607 while (completed < max_progress && completed < wait_point) {
3608 /* Check that sync_action is still 'reshape' to avoid
3609 * waiting forever on a dead array
3610 */
3611 char action[20];
3612 fd_set rfds;
3613 if (sysfs_get_str(info, NULL, "sync_action",
3614 action, 20) <= 0 ||
3615 strncmp(action, "reshape", 7) != 0)
3616 break;
3617 /* Some kernels reset 'sync_completed' to zero
3618 * before setting 'sync_action' to 'idle'.
3619 * So we need these extra tests.
3620 */
3621 if (completed == 0 && advancing
3622 && info->reshape_progress > 0)
3623 break;
3624 if (completed == 0 && !advancing
3625 && info->reshape_progress < (info->component_size
3626 * reshape->after.data_disks))
3627 break;
3628 FD_ZERO(&rfds);
3629 FD_SET(fd, &rfds);
3630 select(fd+1, NULL, NULL, &rfds, NULL);
3631 if (sysfs_fd_get_ll(fd, &completed) < 0)
3632 goto check_progress;
3633 }
3634 /* Some kernels reset 'sync_completed' to zero,
3635 * we need to have real point we are in md
3636 */
3637 if (completed == 0)
3638 completed = max_progress;
3639
3640 /* some kernels can give an incorrectly high 'completed' number */
3641 completed /= (info->new_chunk/512);
3642 completed *= (info->new_chunk/512);
3643 /* Convert 'completed' back in to a 'progress' number */
3644 completed *= reshape->after.data_disks;
3645 if (!advancing) {
3646 completed = info->component_size * reshape->after.data_disks
3647 - completed;
3648 }
3649 *reshape_completed = completed;
3650
3651 close(fd);
3652
3653 /* We return the need_backup flag. Caller will decide
3654 * how much - a multiple of ->backup_blocks up to *suspend_point
3655 */
3656 if (advancing)
3657 return need_backup > info->reshape_progress;
3658 else
3659 return need_backup >= info->reshape_progress;
3660
3661 check_progress:
3662 /* if we couldn't read a number from sync_completed, then
3663 * either the reshape did complete, or it aborted.
3664 * We can tell which by checking for 'none' in reshape_position.
3665 * If it did abort, then it might immediately restart if it
3666 * it was just a device failure that leaves us degraded but
3667 * functioning.
3668 */
3669 strcpy(buf, "hi");
3670 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3671 || strncmp(buf, "none", 4) != 0) {
3672 /* The abort might only be temporary. Wait up to 10
3673 * seconds for fd to contain a valid number again.
3674 */
3675 struct timeval tv;
3676 int rv = -2;
3677 tv.tv_sec = 10;
3678 tv.tv_usec = 0;
3679 while (fd >= 0 && rv < 0 && tv.tv_sec > 0) {
3680 fd_set rfds;
3681 FD_ZERO(&rfds);
3682 FD_SET(fd, &rfds);
3683 if (select(fd+1, NULL, NULL, &rfds, &tv) != 1)
3684 break;
3685 switch (sysfs_fd_get_ll(fd, &completed)) {
3686 case 0:
3687 /* all good again */
3688 rv = 1;
3689 break;
3690 case -2: /* read error - abort */
3691 tv.tv_sec = 0;
3692 break;
3693 }
3694 }
3695 if (fd >= 0)
3696 close(fd);
3697 return rv; /* abort */
3698 } else {
3699 /* Maybe racing with array shutdown - check state */
3700 if (fd >= 0)
3701 close(fd);
3702 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3703 || strncmp(buf, "inactive", 8) == 0
3704 || strncmp(buf, "clear",5) == 0)
3705 return -2; /* abort */
3706 return -1; /* complete */
3707 }
3708 }
3709
3710 /* FIXME return status is never checked */
3711 static int grow_backup(struct mdinfo *sra,
3712 unsigned long long offset, /* per device */
3713 unsigned long stripes, /* per device, in old chunks */
3714 int *sources, unsigned long long *offsets,
3715 int disks, int chunk, int level, int layout,
3716 int dests, int *destfd, unsigned long long *destoffsets,
3717 int part, int *degraded,
3718 char *buf)
3719 {
3720 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3721 * to storage 'destfd' (offset 'destoffsets'), after first
3722 * suspending IO. Then allow resync to continue
3723 * over the suspended section.
3724 * Use part 'part' of the backup-super-block.
3725 */
3726 int odata = disks;
3727 int rv = 0;
3728 int i;
3729 unsigned long long ll;
3730 int new_degraded;
3731 //printf("offset %llu\n", offset);
3732 if (level >= 4)
3733 odata--;
3734 if (level == 6)
3735 odata--;
3736
3737 /* Check that array hasn't become degraded, else we might backup the wrong data */
3738 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3739 return -1; /* FIXME this error is ignored */
3740 new_degraded = (int)ll;
3741 if (new_degraded != *degraded) {
3742 /* check each device to ensure it is still working */
3743 struct mdinfo *sd;
3744 for (sd = sra->devs ; sd ; sd = sd->next) {
3745 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3746 continue;
3747 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3748 char sbuf[20];
3749 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3750 strstr(sbuf, "faulty") ||
3751 strstr(sbuf, "in_sync") == NULL) {
3752 /* this device is dead */
3753 sd->disk.state = (1<<MD_DISK_FAULTY);
3754 if (sd->disk.raid_disk >= 0 &&
3755 sources[sd->disk.raid_disk] >= 0) {
3756 close(sources[sd->disk.raid_disk]);
3757 sources[sd->disk.raid_disk] = -1;
3758 }
3759 }
3760 }
3761 }
3762 *degraded = new_degraded;
3763 }
3764 if (part) {
3765 bsb.arraystart2 = __cpu_to_le64(offset * odata);
3766 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
3767 } else {
3768 bsb.arraystart = __cpu_to_le64(offset * odata);
3769 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
3770 }
3771 if (part)
3772 bsb.magic[15] = '2';
3773 for (i = 0; i < dests; i++)
3774 if (part)
3775 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
3776 else
3777 lseek64(destfd[i], destoffsets[i], 0);
3778
3779 rv = save_stripes(sources, offsets,
3780 disks, chunk, level, layout,
3781 dests, destfd,
3782 offset*512*odata, stripes * chunk * odata,
3783 buf);
3784
3785 if (rv)
3786 return rv;
3787 bsb.mtime = __cpu_to_le64(time(0));
3788 for (i = 0; i < dests; i++) {
3789 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3790
3791 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3792 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3793 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3794 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3795
3796 rv = -1;
3797 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
3798 != destoffsets[i] - 4096)
3799 break;
3800 if (write(destfd[i], &bsb, 512) != 512)
3801 break;
3802 if (destoffsets[i] > 4096) {
3803 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
3804 destoffsets[i]+stripes*chunk*odata)
3805 break;
3806 if (write(destfd[i], &bsb, 512) != 512)
3807 break;
3808 }
3809 fsync(destfd[i]);
3810 rv = 0;
3811 }
3812
3813 return rv;
3814 }
3815
3816 /* in 2.6.30, the value reported by sync_completed can be
3817 * less that it should be by one stripe.
3818 * This only happens when reshape hits sync_max and pauses.
3819 * So allow wait_backup to either extent sync_max further
3820 * than strictly necessary, or return before the
3821 * sync has got quite as far as we would really like.
3822 * This is what 'blocks2' is for.
3823 * The various caller give appropriate values so that
3824 * every works.
3825 */
3826 /* FIXME return value is often ignored */
3827 static int forget_backup(int dests, int *destfd,
3828 unsigned long long *destoffsets,
3829 int part)
3830 {
3831 /*
3832 * Erase backup 'part' (which is 0 or 1)
3833 */
3834 int i;
3835 int rv;
3836
3837 if (part) {
3838 bsb.arraystart2 = __cpu_to_le64(0);
3839 bsb.length2 = __cpu_to_le64(0);
3840 } else {
3841 bsb.arraystart = __cpu_to_le64(0);
3842 bsb.length = __cpu_to_le64(0);
3843 }
3844 bsb.mtime = __cpu_to_le64(time(0));
3845 rv = 0;
3846 for (i = 0; i < dests; i++) {
3847 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3848 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3849 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3850 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3851 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3852 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
3853 destoffsets[i]-4096)
3854 rv = -1;
3855 if (rv == 0 &&
3856 write(destfd[i], &bsb, 512) != 512)
3857 rv = -1;
3858 fsync(destfd[i]);
3859 }
3860 return rv;
3861 }
3862
3863 static void fail(char *msg)
3864 {
3865 int rv;
3866 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
3867 rv |= (write(2, "\n", 1) != 1);
3868 exit(rv ? 1 : 2);
3869 }
3870
3871 static char *abuf, *bbuf;
3872 static unsigned long long abuflen;
3873 static void validate(int afd, int bfd, unsigned long long offset)
3874 {
3875 /* check that the data in the backup against the array.
3876 * This is only used for regression testing and should not
3877 * be used while the array is active
3878 */
3879 if (afd < 0)
3880 return;
3881 lseek64(bfd, offset - 4096, 0);
3882 if (read(bfd, &bsb2, 512) != 512)
3883 fail("cannot read bsb");
3884 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
3885 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
3886 fail("first csum bad");
3887 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
3888 fail("magic is bad");
3889 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
3890 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
3891 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
3892 fail("second csum bad");
3893
3894 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
3895 fail("devstart is wrong");
3896
3897 if (bsb2.length) {
3898 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
3899
3900 if (abuflen < len) {
3901 free(abuf);
3902 free(bbuf);
3903 abuflen = len;
3904 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
3905 posix_memalign((void**)&bbuf, 4096, abuflen)) {
3906 abuflen = 0;
3907 /* just stop validating on mem-alloc failure */
3908 return;
3909 }
3910 }
3911
3912 lseek64(bfd, offset, 0);
3913 if ((unsigned long long)read(bfd, bbuf, len) != len) {
3914 //printf("len %llu\n", len);
3915 fail("read first backup failed");
3916 }
3917 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
3918 if ((unsigned long long)read(afd, abuf, len) != len)
3919 fail("read first from array failed");
3920 if (memcmp(bbuf, abuf, len) != 0) {
3921 #if 0
3922 int i;
3923 printf("offset=%llu len=%llu\n",
3924 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
3925 for (i=0; i<len; i++)
3926 if (bbuf[i] != abuf[i]) {
3927 printf("first diff byte %d\n", i);
3928 break;
3929 }
3930 #endif
3931 fail("data1 compare failed");
3932 }
3933 }
3934 if (bsb2.length2) {
3935 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
3936
3937 if (abuflen < len) {
3938 free(abuf);
3939 free(bbuf);
3940 abuflen = len;
3941 abuf = xmalloc(abuflen);
3942 bbuf = xmalloc(abuflen);
3943 }
3944
3945 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
3946 if ((unsigned long long)read(bfd, bbuf, len) != len)
3947 fail("read second backup failed");
3948 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
3949 if ((unsigned long long)read(afd, abuf, len) != len)
3950 fail("read second from array failed");
3951 if (memcmp(bbuf, abuf, len) != 0)
3952 fail("data2 compare failed");
3953 }
3954 }
3955
3956 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
3957 struct supertype *st, unsigned long blocks,
3958 int *fds, unsigned long long *offsets,
3959 int dests, int *destfd, unsigned long long *destoffsets)
3960 {
3961 /* Monitor a reshape where backup is being performed using
3962 * 'native' mechanism - either to a backup file, or
3963 * to some space in a spare.
3964 */
3965 char *buf;
3966 int degraded = -1;
3967 unsigned long long speed;
3968 unsigned long long suspend_point, array_size;
3969 unsigned long long backup_point, wait_point;
3970 unsigned long long reshape_completed;
3971 int done = 0;
3972 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
3973 int part = 0; /* The next part of the backup area to fill. It may already
3974 * be full, so we need to check */
3975 int level = reshape->level;
3976 int layout = reshape->before.layout;
3977 int data = reshape->before.data_disks;
3978 int disks = reshape->before.data_disks + reshape->parity;
3979 int chunk = sra->array.chunk_size;
3980 struct mdinfo *sd;
3981 unsigned long stripes;
3982 int uuid[4];
3983
3984 /* set up the backup-super-block. This requires the
3985 * uuid from the array.
3986 */
3987 /* Find a superblock */
3988 for (sd = sra->devs; sd; sd = sd->next) {
3989 char *dn;
3990 int devfd;
3991 int ok;
3992 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3993 continue;
3994 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
3995 devfd = dev_open(dn, O_RDONLY);
3996 if (devfd < 0)
3997 continue;
3998 ok = st->ss->load_super(st, devfd, NULL);
3999 close(devfd);
4000 if (ok == 0)
4001 break;
4002 }
4003 if (!sd) {
4004 pr_err("Cannot find a superblock\n");
4005 return 0;
4006 }
4007
4008 memset(&bsb, 0, 512);
4009 memcpy(bsb.magic, "md_backup_data-1", 16);
4010 st->ss->uuid_from_super(st, uuid);
4011 memcpy(bsb.set_uuid, uuid, 16);
4012 bsb.mtime = __cpu_to_le64(time(0));
4013 bsb.devstart2 = blocks;
4014
4015 stripes = blocks / (sra->array.chunk_size/512) /
4016 reshape->before.data_disks;
4017
4018 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4019 /* Don't start the 'reshape' */
4020 return 0;
4021 if (reshape->before.data_disks == reshape->after.data_disks) {
4022 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4023 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4024 }
4025
4026 if (increasing) {
4027 array_size = sra->component_size * reshape->after.data_disks;
4028 backup_point = sra->reshape_progress;
4029 suspend_point = 0;
4030 } else {
4031 array_size = sra->component_size * reshape->before.data_disks;
4032 backup_point = reshape->backup_blocks;
4033 suspend_point = array_size;
4034 }
4035
4036 while (!done) {
4037 int rv;
4038
4039 /* Want to return as soon the oldest backup slot can
4040 * be released as that allows us to start backing up
4041 * some more, providing suspend_point has been
4042 * advanced, which it should have.
4043 */
4044 if (increasing) {
4045 wait_point = array_size;
4046 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4047 wait_point = (__le64_to_cpu(bsb.arraystart) +
4048 __le64_to_cpu(bsb.length));
4049 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4050 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4051 __le64_to_cpu(bsb.length2));
4052 } else {
4053 wait_point = 0;
4054 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4055 wait_point = __le64_to_cpu(bsb.arraystart);
4056 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4057 wait_point = __le64_to_cpu(bsb.arraystart2);
4058 }
4059
4060 rv = progress_reshape(sra, reshape,
4061 backup_point, wait_point,
4062 &suspend_point, &reshape_completed);
4063 /* external metadata would need to ping_monitor here */
4064 sra->reshape_progress = reshape_completed;
4065
4066 /* Clear any backup region that is before 'here' */
4067 if (increasing) {
4068 if (__le64_to_cpu(bsb.length) > 0 &&
4069 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4070 __le64_to_cpu(bsb.length)))
4071 forget_backup(dests, destfd,
4072 destoffsets, 0);
4073 if (__le64_to_cpu(bsb.length2) > 0 &&
4074 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4075 __le64_to_cpu(bsb.length2)))
4076 forget_backup(dests, destfd,
4077 destoffsets, 1);
4078 } else {
4079 if (__le64_to_cpu(bsb.length) > 0 &&
4080 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4081 forget_backup(dests, destfd,
4082 destoffsets, 0);
4083 if (__le64_to_cpu(bsb.length2) > 0 &&
4084 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4085 forget_backup(dests, destfd,
4086 destoffsets, 1);
4087 }
4088
4089 if (rv < 0) {
4090 if (rv == -1)
4091 done = 1;
4092 break;
4093 }
4094 if (rv == 0 && increasing && !st->ss->external) {
4095 /* No longer need to monitor this reshape */
4096 done = 1;
4097 break;
4098 }
4099
4100 while (rv) {
4101 unsigned long long offset;
4102 unsigned long actual_stripes;
4103 /* Need to backup some data.
4104 * If 'part' is not used and the desired
4105 * backup size is suspended, do a backup,
4106 * then consider the next part.
4107 */
4108 /* Check that 'part' is unused */
4109 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4110 break;
4111 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4112 break;
4113
4114 offset = backup_point / data;
4115 actual_stripes = stripes;
4116 if (increasing) {
4117 if (offset + actual_stripes * (chunk/512) >
4118 sra->component_size)
4119 actual_stripes = ((sra->component_size - offset)
4120 / (chunk/512));
4121 if (offset + actual_stripes * (chunk/512) >
4122 suspend_point/data)
4123 break;
4124 } else {
4125 if (offset < actual_stripes * (chunk/512))
4126 actual_stripes = offset / (chunk/512);
4127 offset -= actual_stripes * (chunk/512);
4128 if (offset < suspend_point/data)
4129 break;
4130 }
4131 if (actual_stripes == 0)
4132 break;
4133 grow_backup(sra, offset, actual_stripes,
4134 fds, offsets,
4135 disks, chunk, level, layout,
4136 dests, destfd, destoffsets,
4137 part, &degraded, buf);
4138 validate(afd, destfd[0], destoffsets[0]);
4139 /* record where 'part' is up to */
4140 part = !part;
4141 if (increasing)
4142 backup_point += actual_stripes * (chunk/512) * data;
4143 else
4144 backup_point -= actual_stripes * (chunk/512) * data;
4145 }
4146 }
4147
4148 /* FIXME maybe call progress_reshape one more time instead */
4149 abort_reshape(sra); /* remove any remaining suspension */
4150 if (reshape->before.data_disks == reshape->after.data_disks)
4151 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4152 free(buf);
4153 return done;
4154 }
4155
4156 /*
4157 * If any spare contains md_back_data-1 which is recent wrt mtime,
4158 * write that data into the array and update the super blocks with
4159 * the new reshape_progress
4160 */
4161 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4162 char *backup_file, int verbose)
4163 {
4164 int i, j;
4165 int old_disks;
4166 unsigned long long *offsets;
4167 unsigned long long nstripe, ostripe;
4168 int ndata, odata;
4169
4170 odata = info->array.raid_disks - info->delta_disks - 1;
4171 if (info->array.level == 6) odata--; /* number of data disks */
4172 ndata = info->array.raid_disks - 1;
4173 if (info->new_level == 6) ndata--;
4174
4175 old_disks = info->array.raid_disks - info->delta_disks;
4176
4177 if (info->delta_disks <= 0)
4178 /* Didn't grow, so the backup file must have
4179 * been used
4180 */
4181 old_disks = cnt;
4182 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4183 struct mdinfo dinfo;
4184 int fd;
4185 int bsbsize;
4186 char *devname, namebuf[20];
4187 unsigned long long lo, hi;
4188
4189 /* This was a spare and may have some saved data on it.
4190 * Load the superblock, find and load the
4191 * backup_super_block.
4192 * If either fail, go on to next device.
4193 * If the backup contains no new info, just return
4194 * else restore data and update all superblocks
4195 */
4196 if (i == old_disks-1) {
4197 fd = open(backup_file, O_RDONLY);
4198 if (fd<0) {
4199 pr_err("backup file %s inaccessible: %s\n",
4200 backup_file, strerror(errno));
4201 continue;
4202 }
4203 devname = backup_file;
4204 } else {
4205 fd = fdlist[i];
4206 if (fd < 0)
4207 continue;
4208 if (st->ss->load_super(st, fd, NULL))
4209 continue;
4210
4211 st->ss->getinfo_super(st, &dinfo, NULL);
4212 st->ss->free_super(st);
4213
4214 if (lseek64(fd,
4215 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4216 0) < 0) {
4217 pr_err("Cannot seek on device %d\n", i);
4218 continue; /* Cannot seek */
4219 }
4220 sprintf(namebuf, "device-%d", i);
4221 devname = namebuf;
4222 }
4223 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4224 if (verbose)
4225 pr_err("Cannot read from %s\n", devname);
4226 continue; /* Cannot read */
4227 }
4228 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4229 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4230 if (verbose)
4231 pr_err("No backup metadata on %s\n", devname);
4232 continue;
4233 }
4234 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4235 if (verbose)
4236 pr_err("Bad backup-metadata checksum on %s\n", devname);
4237 continue; /* bad checksum */
4238 }
4239 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4240 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4241 if (verbose)
4242 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4243 continue; /* Bad second checksum */
4244 }
4245 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4246 if (verbose)
4247 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4248 continue; /* Wrong uuid */
4249 }
4250
4251 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4252 * sometimes they aren't... So allow considerable flexability in matching, and allow
4253 * this test to be overridden by an environment variable.
4254 */
4255 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4256 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4257 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4258 pr_err("accepting backup with timestamp %lu "
4259 "for array with timestamp %lu\n",
4260 (unsigned long)__le64_to_cpu(bsb.mtime),
4261 (unsigned long)info->array.utime);
4262 } else {
4263 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4264 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4265 continue; /* time stamp is too bad */
4266 }
4267 }
4268
4269 if (bsb.magic[15] == '1') {
4270 if (bsb.length == 0)
4271 continue;
4272 if (info->delta_disks >= 0) {
4273 /* reshape_progress is increasing */
4274 if (__le64_to_cpu(bsb.arraystart)
4275 + __le64_to_cpu(bsb.length)
4276 < info->reshape_progress) {
4277 nonew:
4278 if (verbose)
4279 pr_err("backup-metadata found on %s but is not needed\n", devname);
4280 continue; /* No new data here */
4281 }
4282 } else {
4283 /* reshape_progress is decreasing */
4284 if (__le64_to_cpu(bsb.arraystart) >=
4285 info->reshape_progress)
4286 goto nonew; /* No new data here */
4287 }
4288 } else {
4289 if (bsb.length == 0 && bsb.length2 == 0)
4290 continue;
4291 if (info->delta_disks >= 0) {
4292 /* reshape_progress is increasing */
4293 if ((__le64_to_cpu(bsb.arraystart)
4294 + __le64_to_cpu(bsb.length)
4295 < info->reshape_progress)
4296 &&
4297 (__le64_to_cpu(bsb.arraystart2)
4298 + __le64_to_cpu(bsb.length2)
4299 < info->reshape_progress))
4300 goto nonew; /* No new data here */
4301 } else {
4302 /* reshape_progress is decreasing */
4303 if (__le64_to_cpu(bsb.arraystart) >=
4304 info->reshape_progress &&
4305 __le64_to_cpu(bsb.arraystart2) >=
4306 info->reshape_progress)
4307 goto nonew; /* No new data here */
4308 }
4309 }
4310 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4311 second_fail:
4312 if (verbose)
4313 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4314 devname);
4315 continue; /* Cannot seek */
4316 }
4317 /* There should be a duplicate backup superblock 4k before here */
4318 if (lseek64(fd, -4096, 1) < 0 ||
4319 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4320 goto second_fail; /* Cannot find leading superblock */
4321 if (bsb.magic[15] == '1')
4322 bsbsize = offsetof(struct mdp_backup_super, pad1);
4323 else
4324 bsbsize = offsetof(struct mdp_backup_super, pad);
4325 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4326 goto second_fail; /* Cannot find leading superblock */
4327
4328 /* Now need the data offsets for all devices. */
4329 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4330 for(j=0; j<info->array.raid_disks; j++) {
4331 if (fdlist[j] < 0)
4332 continue;
4333 if (st->ss->load_super(st, fdlist[j], NULL))
4334 /* FIXME should be this be an error */
4335 continue;
4336 st->ss->getinfo_super(st, &dinfo, NULL);
4337 st->ss->free_super(st);
4338 offsets[j] = dinfo.data_offset * 512;
4339 }
4340 printf(Name ": restoring critical section\n");
4341
4342 if (restore_stripes(fdlist, offsets,
4343 info->array.raid_disks,
4344 info->new_chunk,
4345 info->new_level,
4346 info->new_layout,
4347 fd, __le64_to_cpu(bsb.devstart)*512,
4348 __le64_to_cpu(bsb.arraystart)*512,
4349 __le64_to_cpu(bsb.length)*512, NULL)) {
4350 /* didn't succeed, so giveup */
4351 if (verbose)
4352 pr_err("Error restoring backup from %s\n",
4353 devname);
4354 free(offsets);
4355 return 1;
4356 }
4357
4358 if (bsb.magic[15] == '2' &&
4359 restore_stripes(fdlist, offsets,
4360 info->array.raid_disks,
4361 info->new_chunk,
4362 info->new_level,
4363 info->new_layout,
4364 fd, __le64_to_cpu(bsb.devstart)*512 +
4365 __le64_to_cpu(bsb.devstart2)*512,
4366 __le64_to_cpu(bsb.arraystart2)*512,
4367 __le64_to_cpu(bsb.length2)*512, NULL)) {
4368 /* didn't succeed, so giveup */
4369 if (verbose)
4370 pr_err("Error restoring second backup from %s\n",
4371 devname);
4372 free(offsets);
4373 return 1;
4374 }
4375
4376 free(offsets);
4377
4378 /* Ok, so the data is restored. Let's update those superblocks. */
4379
4380 lo = hi = 0;
4381 if (bsb.length) {
4382 lo = __le64_to_cpu(bsb.arraystart);
4383 hi = lo + __le64_to_cpu(bsb.length);
4384 }
4385 if (bsb.magic[15] == '2' && bsb.length2) {
4386 unsigned long long lo1, hi1;
4387 lo1 = __le64_to_cpu(bsb.arraystart2);
4388 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4389 if (lo == hi) {
4390 lo = lo1;
4391 hi = hi1;
4392 } else if (lo < lo1)
4393 hi = hi1;
4394 else
4395 lo = lo1;
4396 }
4397 if (lo < hi &&
4398 (info->reshape_progress < lo ||
4399 info->reshape_progress > hi))
4400 /* backup does not affect reshape_progress*/ ;
4401 else if (info->delta_disks >= 0) {
4402 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4403 __le64_to_cpu(bsb.length);
4404 if (bsb.magic[15] == '2') {
4405 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4406 __le64_to_cpu(bsb.length2);
4407 if (p2 > info->reshape_progress)
4408 info->reshape_progress = p2;
4409 }
4410 } else {
4411 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4412 if (bsb.magic[15] == '2') {
4413 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4414 if (p2 < info->reshape_progress)
4415 info->reshape_progress = p2;
4416 }
4417 }
4418 for (j=0; j<info->array.raid_disks; j++) {
4419 if (fdlist[j] < 0)
4420 continue;
4421 if (st->ss->load_super(st, fdlist[j], NULL))
4422 continue;
4423 st->ss->getinfo_super(st, &dinfo, NULL);
4424 dinfo.reshape_progress = info->reshape_progress;
4425 st->ss->update_super(st, &dinfo,
4426 "_reshape_progress",
4427 NULL,0, 0, NULL);
4428 st->ss->store_super(st, fdlist[j]);
4429 st->ss->free_super(st);
4430 }
4431 return 0;
4432 }
4433 /* Didn't find any backup data, try to see if any
4434 * was needed.
4435 */
4436 if (info->delta_disks < 0) {
4437 /* When shrinking, the critical section is at the end.
4438 * So see if we are before the critical section.
4439 */
4440 unsigned long long first_block;
4441 nstripe = ostripe = 0;
4442 first_block = 0;
4443 while (ostripe >= nstripe) {
4444 ostripe += info->array.chunk_size / 512;
4445 first_block = ostripe * odata;
4446 nstripe = first_block / ndata / (info->new_chunk/512) *
4447 (info->new_chunk/512);
4448 }
4449
4450 if (info->reshape_progress >= first_block)
4451 return 0;
4452 }
4453 if (info->delta_disks > 0) {
4454 /* See if we are beyond the critical section. */
4455 unsigned long long last_block;
4456 nstripe = ostripe = 0;
4457 last_block = 0;
4458 while (nstripe >= ostripe) {
4459 nstripe += info->new_chunk / 512;
4460 last_block = nstripe * ndata;
4461 ostripe = last_block / odata / (info->array.chunk_size/512) *
4462 (info->array.chunk_size/512);
4463 }
4464
4465 if (info->reshape_progress >= last_block)
4466 return 0;
4467 }
4468 /* needed to recover critical section! */
4469 if (verbose)
4470 pr_err("Failed to find backup of critical section\n");
4471 return 1;
4472 }
4473
4474 int Grow_continue_command(char *devname, int fd,
4475 char *backup_file, int verbose)
4476 {
4477 int ret_val = 0;
4478 struct supertype *st = NULL;
4479 struct mdinfo *content = NULL;
4480 struct mdinfo array;
4481 char *subarray = NULL;
4482 struct mdinfo *cc = NULL;
4483 struct mdstat_ent *mdstat = NULL;
4484 int cfd = -1;
4485 int fd2 = -1;
4486
4487 dprintf("Grow continue from command line called for %s\n",
4488 devname);
4489
4490 st = super_by_fd(fd, &subarray);
4491 if (!st || !st->ss) {
4492 pr_err("Unable to determine metadata format for %s\n",
4493 devname);
4494 return 1;
4495 }
4496 dprintf("Grow continue is run for ");
4497 if (st->ss->external == 0) {
4498 int d;
4499 dprintf("native array (%s)\n", devname);
4500 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4501 pr_err("%s is not an active md array -"
4502 " aborting\n", devname);
4503 ret_val = 1;
4504 goto Grow_continue_command_exit;
4505 }
4506 content = &array;
4507 /* Need to load a superblock.
4508 * FIXME we should really get what we need from
4509 * sysfs
4510 */
4511 for (d = 0; d < MAX_DISKS; d++) {
4512 mdu_disk_info_t disk;
4513 char *dv;
4514 int err;
4515 disk.number = d;
4516 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4517 continue;
4518 if (disk.major == 0 && disk.minor == 0)
4519 continue;
4520 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4521 continue;
4522 dv = map_dev(disk.major, disk.minor, 1);
4523 if (!dv)
4524 continue;
4525 fd2 = dev_open(dv, O_RDONLY);
4526 if (fd2 < 0)
4527 continue;
4528 err = st->ss->load_super(st, fd2, NULL);
4529 close(fd2);
4530 if (err)
4531 continue;
4532 break;
4533 }
4534 if (d == MAX_DISKS) {
4535 pr_err("Unable to load metadata for %s\n",
4536 devname);
4537 ret_val = 1;
4538 goto Grow_continue_command_exit;
4539 }
4540 st->ss->getinfo_super(st, content, NULL);
4541 } else {
4542 char *container;
4543
4544 if (subarray) {
4545 dprintf("subarray (%s)\n", subarray);
4546 container = st->container_devnm;
4547 cfd = open_dev_excl(st->container_devnm);
4548 } else {
4549 container = st->devnm;
4550 close(fd);
4551 cfd = open_dev_excl(st->devnm);
4552 dprintf("container (%s)\n", container);
4553 fd = cfd;
4554 }
4555 if (cfd < 0) {
4556 pr_err("Unable to open container "
4557 "for %s\n", devname);
4558 ret_val = 1;
4559 goto Grow_continue_command_exit;
4560 }
4561
4562 /* find in container array under reshape
4563 */
4564 ret_val = st->ss->load_container(st, cfd, NULL);
4565 if (ret_val) {
4566 pr_err("Cannot read superblock for %s\n",
4567 devname);
4568 ret_val = 1;
4569 goto Grow_continue_command_exit;
4570 }
4571
4572 cc = st->ss->container_content(st, subarray);
4573 for (content = cc; content ; content = content->next) {
4574 char *array;
4575 int allow_reshape = 1;
4576
4577 if (content->reshape_active == 0)
4578 continue;
4579 /* The decision about array or container wide
4580 * reshape is taken in Grow_continue based
4581 * content->reshape_active state, therefore we
4582 * need to check_reshape based on
4583 * reshape_active and subarray name
4584 */
4585 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4586 allow_reshape = 0;
4587 if (content->reshape_active == CONTAINER_RESHAPE &&
4588 (content->array.state
4589 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4590 allow_reshape = 0;
4591
4592 if (!allow_reshape) {
4593 pr_err("cannot continue reshape of an array"
4594 " in container with unsupported"
4595 " metadata: %s(%s)\n",
4596 devname, container);
4597 ret_val = 1;
4598 goto Grow_continue_command_exit;
4599 }
4600
4601 array = strchr(content->text_version+1, '/')+1;
4602 mdstat = mdstat_by_subdev(array, container);
4603 if (!mdstat)
4604 continue;
4605 if (mdstat->active == 0) {
4606 pr_err("Skipping inactive array %s.\n",
4607 mdstat->devnm);
4608 free_mdstat(mdstat);
4609 mdstat = NULL;
4610 continue;
4611 }
4612 break;
4613 }
4614 if (!content) {
4615 pr_err("Unable to determine reshaped "
4616 "array for %s\n", devname);
4617 ret_val = 1;
4618 goto Grow_continue_command_exit;
4619 }
4620 fd2 = open_dev(mdstat->devnm);
4621 if (fd2 < 0) {
4622 pr_err("cannot open (%s)\n", mdstat->devnm);
4623 ret_val = 1;
4624 goto Grow_continue_command_exit;
4625 }
4626
4627 sysfs_init(content, fd2, mdstat->devnm);
4628
4629 /* start mdmon in case it is not running
4630 */
4631 if (!mdmon_running(container))
4632 start_mdmon(container);
4633 ping_monitor(container);
4634
4635 if (mdmon_running(container))
4636 st->update_tail = &st->updates;
4637 else {
4638 pr_err("No mdmon found. "
4639 "Grow cannot continue.\n");
4640 ret_val = 1;
4641 goto Grow_continue_command_exit;
4642 }
4643 }
4644
4645 /* verify that array under reshape is started from
4646 * correct position
4647 */
4648 if (verify_reshape_position(content, content->array.level) < 0) {
4649 ret_val = 1;
4650 goto Grow_continue_command_exit;
4651 }
4652
4653 /* continue reshape
4654 */
4655 ret_val = Grow_continue(fd, st, content, backup_file, 0);
4656
4657 Grow_continue_command_exit:
4658 if (fd2 > -1)
4659 close(fd2);
4660 if (cfd > -1)
4661 close(cfd);
4662 st->ss->free_super(st);
4663 free_mdstat(mdstat);
4664 sysfs_free(cc);
4665 free(subarray);
4666
4667 return ret_val;
4668 }
4669
4670 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4671 char *backup_file, int freeze_reshape)
4672 {
4673 int ret_val = 2;
4674
4675 if (!info->reshape_active)
4676 return ret_val;
4677
4678 if (st->ss->external) {
4679 int cfd = open_dev(st->container_devnm);
4680
4681 if (cfd < 0)
4682 return 1;
4683
4684 st->ss->load_container(st, cfd, st->container_devnm);
4685 close(cfd);
4686 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4687 st, info, 0, backup_file,
4688 0,
4689 1 | info->reshape_active,
4690 freeze_reshape);
4691 } else
4692 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4693 NULL, 0ULL, backup_file, 0, 0,
4694 1 | info->reshape_active,
4695 freeze_reshape);
4696
4697 return ret_val;
4698 }