]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Grow: analyse_change needs to set new_size even if nothing much is happening.
[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 re->new_size = info->component_size * re->before.data_disks;
1416 if (info->new_chunk == 0)
1417 info->new_chunk = info->array.chunk_size;
1418 if (re->after.data_disks == re->before.data_disks &&
1419 re->after.layout == re->before.layout &&
1420 info->new_chunk == info->array.chunk_size) {
1421 /* Nothing to change, can change level immediately. */
1422 re->level = info->new_level;
1423 re->backup_blocks = 0;
1424 return NULL;
1425 }
1426 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1427 /* chunk and layout changes make no difference */
1428 re->level = info->new_level;
1429 re->backup_blocks = 0;
1430 return NULL;
1431 }
1432
1433 if (re->after.data_disks == re->before.data_disks &&
1434 get_linux_version() < 2006032)
1435 return "in-place reshape is not safe before 2.6.32 - sorry.";
1436
1437 if (re->after.data_disks < re->before.data_disks &&
1438 get_linux_version() < 2006030)
1439 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1440
1441 re->backup_blocks = compute_backup_blocks(
1442 info->new_chunk, info->array.chunk_size,
1443 re->after.data_disks,
1444 re->before.data_disks);
1445 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1446
1447 re->new_size = info->component_size * re->after.data_disks;
1448 return NULL;
1449 }
1450
1451 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1452 char *text_version)
1453 {
1454 struct mdinfo *info;
1455 char *subarray;
1456 int ret_val = -1;
1457
1458 if ((st == NULL) || (sra == NULL))
1459 return ret_val;
1460
1461 if (text_version == NULL)
1462 text_version = sra->text_version;
1463 subarray = strchr(text_version+1, '/')+1;
1464 info = st->ss->container_content(st, subarray);
1465 if (info) {
1466 unsigned long long current_size = 0;
1467 unsigned long long new_size =
1468 info->custom_array_size/2;
1469
1470 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1471 new_size > current_size) {
1472 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1473 < 0)
1474 dprintf("Error: Cannot set array size");
1475 else {
1476 ret_val = 0;
1477 dprintf("Array size changed");
1478 }
1479 dprintf(" from %llu to %llu.\n",
1480 current_size, new_size);
1481 }
1482 sysfs_free(info);
1483 } else
1484 dprintf("Error: set_array_size(): info pointer in NULL\n");
1485
1486 return ret_val;
1487 }
1488
1489 static int reshape_array(char *container, int fd, char *devname,
1490 struct supertype *st, struct mdinfo *info,
1491 int force, struct mddev_dev *devlist,
1492 unsigned long long data_offset,
1493 char *backup_file, int verbose, int forked,
1494 int restart, int freeze_reshape);
1495 static int reshape_container(char *container, char *devname,
1496 int mdfd,
1497 struct supertype *st,
1498 struct mdinfo *info,
1499 int force,
1500 char *backup_file,
1501 int verbose, int restart, int freeze_reshape);
1502
1503 int Grow_reshape(char *devname, int fd,
1504 struct mddev_dev *devlist,
1505 unsigned long long data_offset,
1506 struct context *c, struct shape *s)
1507 {
1508 /* Make some changes in the shape of an array.
1509 * The kernel must support the change.
1510 *
1511 * There are three different changes. Each can trigger
1512 * a resync or recovery so we freeze that until we have
1513 * requested everything (if kernel supports freezing - 2.6.30).
1514 * The steps are:
1515 * - change size (i.e. component_size)
1516 * - change level
1517 * - change layout/chunksize/ndisks
1518 *
1519 * The last can require a reshape. It is different on different
1520 * levels so we need to check the level before actioning it.
1521 * Some times the level change needs to be requested after the
1522 * reshape (e.g. raid6->raid5, raid5->raid0)
1523 *
1524 */
1525 struct mdu_array_info_s array;
1526 int rv = 0;
1527 struct supertype *st;
1528 char *subarray = NULL;
1529
1530 int frozen;
1531 int changed = 0;
1532 char *container = NULL;
1533 int cfd = -1;
1534
1535 struct mddev_dev *dv;
1536 int added_disks;
1537
1538 struct mdinfo info;
1539 struct mdinfo *sra;
1540
1541 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1542 pr_err("%s is not an active md array - aborting\n",
1543 devname);
1544 return 1;
1545 }
1546 if (data_offset != INVALID_SECTORS && array.level != 10
1547 && (array.level < 4 || array.level > 6)) {
1548 pr_err("--grow --data-offset not yet supported\n");
1549 return 1;
1550 }
1551
1552 if (s->size > 0 &&
1553 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1554 pr_err("cannot change component size at the same time "
1555 "as other changes.\n"
1556 " Change size first, then check data is intact before "
1557 "making other changes.\n");
1558 return 1;
1559 }
1560
1561 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1562 get_linux_version() < 2006032 &&
1563 !check_env("MDADM_FORCE_FEWER")) {
1564 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1565 " Please use a newer kernel\n");
1566 return 1;
1567 }
1568
1569 st = super_by_fd(fd, &subarray);
1570 if (!st) {
1571 pr_err("Unable to determine metadata format for %s\n", devname);
1572 return 1;
1573 }
1574 if (s->raiddisks > st->max_devs) {
1575 pr_err("Cannot increase raid-disks on this array"
1576 " beyond %d\n", st->max_devs);
1577 return 1;
1578 }
1579
1580 /* in the external case we need to check that the requested reshape is
1581 * supported, and perform an initial check that the container holds the
1582 * pre-requisite spare devices (mdmon owns final validation)
1583 */
1584 if (st->ss->external) {
1585 int rv;
1586
1587 if (subarray) {
1588 container = st->container_devnm;
1589 cfd = open_dev_excl(st->container_devnm);
1590 } else {
1591 container = st->devnm;
1592 close(fd);
1593 cfd = open_dev_excl(st->devnm);
1594 fd = cfd;
1595 }
1596 if (cfd < 0) {
1597 pr_err("Unable to open container for %s\n",
1598 devname);
1599 free(subarray);
1600 return 1;
1601 }
1602
1603 rv = st->ss->load_container(st, cfd, NULL);
1604
1605 if (rv) {
1606 pr_err("Cannot read superblock for %s\n",
1607 devname);
1608 free(subarray);
1609 return 1;
1610 }
1611
1612 /* check if operation is supported for metadata handler */
1613 if (st->ss->container_content) {
1614 struct mdinfo *cc = NULL;
1615 struct mdinfo *content = NULL;
1616
1617 cc = st->ss->container_content(st, subarray);
1618 for (content = cc; content ; content = content->next) {
1619 int allow_reshape = 1;
1620
1621 /* check if reshape is allowed based on metadata
1622 * indications stored in content.array.status
1623 */
1624 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1625 allow_reshape = 0;
1626 if (content->array.state
1627 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1628 allow_reshape = 0;
1629 if (!allow_reshape) {
1630 pr_err("cannot reshape arrays in"
1631 " container with unsupported"
1632 " metadata: %s(%s)\n",
1633 devname, container);
1634 sysfs_free(cc);
1635 free(subarray);
1636 return 1;
1637 }
1638 }
1639 sysfs_free(cc);
1640 }
1641 if (mdmon_running(container))
1642 st->update_tail = &st->updates;
1643 }
1644
1645 added_disks = 0;
1646 for (dv = devlist; dv; dv = dv->next)
1647 added_disks++;
1648 if (s->raiddisks > array.raid_disks &&
1649 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1650 !c->force) {
1651 pr_err("Need %d spare%s to avoid degraded array,"
1652 " and only have %d.\n"
1653 " Use --force to over-ride this check.\n",
1654 s->raiddisks - array.raid_disks,
1655 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1656 array.spare_disks + added_disks);
1657 return 1;
1658 }
1659
1660 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1661 | GET_STATE | GET_VERSION);
1662 if (sra) {
1663 if (st->ss->external && subarray == NULL) {
1664 array.level = LEVEL_CONTAINER;
1665 sra->array.level = LEVEL_CONTAINER;
1666 }
1667 } else {
1668 pr_err("failed to read sysfs parameters for %s\n",
1669 devname);
1670 return 1;
1671 }
1672 frozen = freeze(st);
1673 if (frozen < -1) {
1674 /* freeze() already spewed the reason */
1675 sysfs_free(sra);
1676 return 1;
1677 } else if (frozen < 0) {
1678 pr_err("%s is performing resync/recovery and cannot"
1679 " be reshaped\n", devname);
1680 sysfs_free(sra);
1681 return 1;
1682 }
1683
1684 /* ========= set size =============== */
1685 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1686 unsigned long long orig_size = get_component_size(fd)/2;
1687 unsigned long long min_csize;
1688 struct mdinfo *mdi;
1689 int raid0_takeover = 0;
1690
1691 if (orig_size == 0)
1692 orig_size = (unsigned) array.size;
1693
1694 if (orig_size == 0) {
1695 pr_err("Cannot set device size in this type of array.\n");
1696 rv = 1;
1697 goto release;
1698 }
1699
1700 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1701 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1702 rv = 1;
1703 goto release;
1704 }
1705 sync_metadata(st);
1706 if (st->ss->external) {
1707 /* metadata can have size limitation
1708 * update size value according to metadata information
1709 */
1710 struct mdinfo *sizeinfo =
1711 st->ss->container_content(st, subarray);
1712 if (sizeinfo) {
1713 unsigned long long new_size =
1714 sizeinfo->custom_array_size/2;
1715 int data_disks = get_data_disks(
1716 sizeinfo->array.level,
1717 sizeinfo->array.layout,
1718 sizeinfo->array.raid_disks);
1719 new_size /= data_disks;
1720 dprintf("Metadata size correction from %llu to "
1721 "%llu (%llu)\n", orig_size, new_size,
1722 new_size * data_disks);
1723 s->size = new_size;
1724 sysfs_free(sizeinfo);
1725 }
1726 }
1727
1728 /* Update the size of each member device in case
1729 * they have been resized. This will never reduce
1730 * below the current used-size. The "size" attribute
1731 * understands '0' to mean 'max'.
1732 */
1733 min_csize = 0;
1734 rv = 0;
1735 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1736 if (sysfs_set_num(sra, mdi, "size",
1737 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1738 /* Probably kernel refusing to let us
1739 * reduce the size - not an error.
1740 */
1741 break;
1742 }
1743 if (array.not_persistent == 0 &&
1744 array.major_version == 0 &&
1745 get_linux_version() < 3001000) {
1746 /* Dangerous to allow size to exceed 2TB */
1747 unsigned long long csize;
1748 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1749 if (csize >= 2ULL*1024*1024*1024)
1750 csize = 2ULL*1024*1024*1024;
1751 if ((min_csize == 0 || (min_csize
1752 > csize)))
1753 min_csize = csize;
1754 }
1755 }
1756 }
1757 if (rv) {
1758 pr_err("Cannot set size on "
1759 "array members.\n");
1760 goto size_change_error;
1761 }
1762 if (min_csize && s->size > min_csize) {
1763 pr_err("Cannot safely make this array "
1764 "use more than 2TB per device on this kernel.\n");
1765 rv = 1;
1766 goto size_change_error;
1767 }
1768 if (min_csize && s->size == MAX_SIZE) {
1769 /* Don't let the kernel choose a size - it will get
1770 * it wrong
1771 */
1772 pr_err("Limited v0.90 array to "
1773 "2TB per device\n");
1774 s->size = min_csize;
1775 }
1776 if (st->ss->external) {
1777 if (sra->array.level == 0) {
1778 rv = sysfs_set_str(sra, NULL, "level",
1779 "raid5");
1780 if (!rv) {
1781 raid0_takeover = 1;
1782 /* get array parametes after takeover
1783 * to chane one parameter at time only
1784 */
1785 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1786 }
1787 }
1788 /* make sure mdmon is
1789 * aware of the new level */
1790 if (!mdmon_running(st->container_devnm))
1791 start_mdmon(st->container_devnm);
1792 ping_monitor(container);
1793 if (mdmon_running(st->container_devnm) &&
1794 st->update_tail == NULL)
1795 st->update_tail = &st->updates;
1796 }
1797
1798 if (s->size == MAX_SIZE)
1799 s->size = 0;
1800 array.size = s->size;
1801 if ((unsigned)array.size != s->size) {
1802 /* got truncated to 32bit, write to
1803 * component_size instead
1804 */
1805 if (sra)
1806 rv = sysfs_set_num(sra, NULL,
1807 "component_size", s->size);
1808 else
1809 rv = -1;
1810 } else {
1811 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1812
1813 /* manage array size when it is managed externally
1814 */
1815 if ((rv == 0) && st->ss->external)
1816 rv = set_array_size(st, sra, sra->text_version);
1817 }
1818
1819 if (raid0_takeover) {
1820 /* do not recync non-existing parity,
1821 * we will drop it anyway
1822 */
1823 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1824 /* go back to raid0, drop parity disk
1825 */
1826 sysfs_set_str(sra, NULL, "level", "raid0");
1827 ioctl(fd, GET_ARRAY_INFO, &array);
1828 }
1829
1830 size_change_error:
1831 if (rv != 0) {
1832 int err = errno;
1833
1834 /* restore metadata */
1835 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1836 UnSet, NULL, devname,
1837 ROLLBACK_METADATA_CHANGES,
1838 c->verbose) == 0)
1839 sync_metadata(st);
1840 pr_err("Cannot set device size for %s: %s\n",
1841 devname, strerror(err));
1842 if (err == EBUSY &&
1843 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1844 cont_err("Bitmap must be removed before size can be changed\n");
1845 rv = 1;
1846 goto release;
1847 }
1848 if (s->assume_clean) {
1849 /* This will fail on kernels older than 3.0 unless
1850 * a backport has been arranged.
1851 */
1852 if (sra == NULL ||
1853 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1854 pr_err("--assume-clean not supported with --grow on this kernel\n");
1855 }
1856 ioctl(fd, GET_ARRAY_INFO, &array);
1857 s->size = get_component_size(fd)/2;
1858 if (s->size == 0)
1859 s->size = array.size;
1860 if (c->verbose >= 0) {
1861 if (s->size == orig_size)
1862 pr_err("component size of %s "
1863 "unchanged at %lluK\n",
1864 devname, s->size);
1865 else
1866 pr_err("component size of %s "
1867 "has been set to %lluK\n",
1868 devname, s->size);
1869 }
1870 changed = 1;
1871 } else if (array.level != LEVEL_CONTAINER) {
1872 s->size = get_component_size(fd)/2;
1873 if (s->size == 0)
1874 s->size = array.size;
1875 }
1876
1877 /* See if there is anything else to do */
1878 if ((s->level == UnSet || s->level == array.level) &&
1879 (s->layout_str == NULL) &&
1880 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1881 data_offset == INVALID_SECTORS &&
1882 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1883 /* Nothing more to do */
1884 if (!changed && c->verbose >= 0)
1885 pr_err("%s: no change requested\n",
1886 devname);
1887 goto release;
1888 }
1889
1890 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1891 * current implementation assumes that following conditions must be met:
1892 * - RAID10:
1893 * - far_copies == 1
1894 * - near_copies == 2
1895 */
1896 if ((s->level == 0 && array.level == 10 && sra &&
1897 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1898 (s->level == 0 && array.level == 1 && sra)) {
1899 int err;
1900 err = remove_disks_for_takeover(st, sra, array.layout);
1901 if (err) {
1902 dprintf(Name": Array cannot be reshaped\n");
1903 if (cfd > -1)
1904 close(cfd);
1905 rv = 1;
1906 goto release;
1907 }
1908 /* Make sure mdmon has seen the device removal
1909 * and updated metadata before we continue with
1910 * level change
1911 */
1912 if (container)
1913 ping_monitor(container);
1914 }
1915
1916 memset(&info, 0, sizeof(info));
1917 info.array = array;
1918 sysfs_init(&info, fd, NULL);
1919 strcpy(info.text_version, sra->text_version);
1920 info.component_size = s->size*2;
1921 info.new_level = s->level;
1922 info.new_chunk = s->chunk * 1024;
1923 if (info.array.level == LEVEL_CONTAINER) {
1924 info.delta_disks = UnSet;
1925 info.array.raid_disks = s->raiddisks;
1926 } else if (s->raiddisks)
1927 info.delta_disks = s->raiddisks - info.array.raid_disks;
1928 else
1929 info.delta_disks = UnSet;
1930 if (s->layout_str == NULL) {
1931 info.new_layout = UnSet;
1932 if (info.array.level == 6 &&
1933 (info.new_level == 6 || info.new_level == UnSet) &&
1934 info.array.layout >= 16) {
1935 pr_err("%s has a non-standard layout. If you"
1936 " wish to preserve this\n", devname);
1937 cont_err("during the reshape, please specify"
1938 " --layout=preserve\n");
1939 cont_err("If you want to change it, specify a"
1940 " layout or use --layout=normalise\n");
1941 rv = 1;
1942 goto release;
1943 }
1944 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1945 strcmp(s->layout_str, "normalize") == 0) {
1946 /* If we have a -6 RAID6 layout, remove the '-6'. */
1947 info.new_layout = UnSet;
1948 if (info.array.level == 6 && info.new_level == UnSet) {
1949 char l[40], *h;
1950 strcpy(l, map_num(r6layout, info.array.layout));
1951 h = strrchr(l, '-');
1952 if (h && strcmp(h, "-6") == 0) {
1953 *h = 0;
1954 info.new_layout = map_name(r6layout, l);
1955 }
1956 } else {
1957 pr_err("%s is only meaningful when reshaping"
1958 " a RAID6 array.\n", s->layout_str);
1959 rv = 1;
1960 goto release;
1961 }
1962 } else if (strcmp(s->layout_str, "preserve") == 0) {
1963 /* This means that a non-standard RAID6 layout
1964 * is OK.
1965 * In particular:
1966 * - When reshape a RAID6 (e.g. adding a device)
1967 * which is in a non-standard layout, it is OK
1968 * to preserve that layout.
1969 * - When converting a RAID5 to RAID6, leave it in
1970 * the XXX-6 layout, don't re-layout.
1971 */
1972 if (info.array.level == 6 && info.new_level == UnSet)
1973 info.new_layout = info.array.layout;
1974 else if (info.array.level == 5 && info.new_level == 6) {
1975 char l[40];
1976 strcpy(l, map_num(r5layout, info.array.layout));
1977 strcat(l, "-6");
1978 info.new_layout = map_name(r6layout, l);
1979 } else {
1980 pr_err("%s in only meaningful when reshaping"
1981 " to RAID6\n", s->layout_str);
1982 rv = 1;
1983 goto release;
1984 }
1985 } else {
1986 int l = info.new_level;
1987 if (l == UnSet)
1988 l = info.array.level;
1989 switch (l) {
1990 case 5:
1991 info.new_layout = map_name(r5layout, s->layout_str);
1992 break;
1993 case 6:
1994 info.new_layout = map_name(r6layout, s->layout_str);
1995 break;
1996 case 10:
1997 info.new_layout = parse_layout_10(s->layout_str);
1998 break;
1999 case LEVEL_FAULTY:
2000 info.new_layout = parse_layout_faulty(s->layout_str);
2001 break;
2002 default:
2003 pr_err("layout not meaningful"
2004 " with this level\n");
2005 rv = 1;
2006 goto release;
2007 }
2008 if (info.new_layout == UnSet) {
2009 pr_err("layout %s not understood"
2010 " for this level\n",
2011 s->layout_str);
2012 rv = 1;
2013 goto release;
2014 }
2015 }
2016
2017 if (array.level == LEVEL_FAULTY) {
2018 if (s->level != UnSet && s->level != array.level) {
2019 pr_err("cannot change level of Faulty device\n");
2020 rv =1 ;
2021 }
2022 if (s->chunk) {
2023 pr_err("cannot set chunksize of Faulty device\n");
2024 rv =1 ;
2025 }
2026 if (s->raiddisks && s->raiddisks != 1) {
2027 pr_err("cannot set raid_disks of Faulty device\n");
2028 rv =1 ;
2029 }
2030 if (s->layout_str) {
2031 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2032 dprintf("Cannot get array information.\n");
2033 goto release;
2034 }
2035 array.layout = info.new_layout;
2036 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2037 pr_err("failed to set new layout\n");
2038 rv = 1;
2039 } else if (c->verbose >= 0)
2040 printf("layout for %s set to %d\n",
2041 devname, array.layout);
2042 }
2043 } else if (array.level == LEVEL_CONTAINER) {
2044 /* This change is to be applied to every array in the
2045 * container. This is only needed when the metadata imposes
2046 * restraints of the various arrays in the container.
2047 * Currently we only know that IMSM requires all arrays
2048 * to have the same number of devices so changing the
2049 * number of devices (On-Line Capacity Expansion) must be
2050 * performed at the level of the container
2051 */
2052 rv = reshape_container(container, devname, -1, st, &info,
2053 c->force, c->backup_file, c->verbose, 0, 0);
2054 frozen = 0;
2055 } else {
2056 /* get spare devices from external metadata
2057 */
2058 if (st->ss->external) {
2059 struct mdinfo *info2;
2060
2061 info2 = st->ss->container_content(st, subarray);
2062 if (info2) {
2063 info.array.spare_disks =
2064 info2->array.spare_disks;
2065 sysfs_free(info2);
2066 }
2067 }
2068
2069 /* Impose these changes on a single array. First
2070 * check that the metadata is OK with the change. */
2071
2072 if (reshape_super(st, 0, info.new_level,
2073 info.new_layout, info.new_chunk,
2074 info.array.raid_disks, info.delta_disks,
2075 c->backup_file, devname, APPLY_METADATA_CHANGES,
2076 c->verbose)) {
2077 rv = 1;
2078 goto release;
2079 }
2080 sync_metadata(st);
2081 rv = reshape_array(container, fd, devname, st, &info, c->force,
2082 devlist, data_offset, c->backup_file, c->verbose,
2083 0, 0, 0);
2084 frozen = 0;
2085 }
2086 release:
2087 sysfs_free(sra);
2088 if (frozen > 0)
2089 unfreeze(st);
2090 return rv;
2091 }
2092
2093 /* verify_reshape_position()
2094 * Function checks if reshape position in metadata is not farther
2095 * than position in md.
2096 * Return value:
2097 * 0 : not valid sysfs entry
2098 * it can be caused by not started reshape, it should be started
2099 * by reshape array or raid0 array is before takeover
2100 * -1 : error, reshape position is obviously wrong
2101 * 1 : success, reshape progress correct or updated
2102 */
2103 static int verify_reshape_position(struct mdinfo *info, int level)
2104 {
2105 int ret_val = 0;
2106 char buf[40];
2107 int rv;
2108
2109 /* read sync_max, failure can mean raid0 array */
2110 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2111
2112 if (rv > 0) {
2113 char *ep;
2114 unsigned long long position = strtoull(buf, &ep, 0);
2115
2116 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2117 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2118 position *= get_data_disks(level,
2119 info->new_layout,
2120 info->array.raid_disks);
2121 if (info->reshape_progress < position) {
2122 dprintf("Corrected reshape progress (%llu) to "
2123 "md position (%llu)\n",
2124 info->reshape_progress, position);
2125 info->reshape_progress = position;
2126 ret_val = 1;
2127 } else if (info->reshape_progress > position) {
2128 pr_err("Fatal error: array "
2129 "reshape was not properly frozen "
2130 "(expected reshape position is %llu, "
2131 "but reshape progress is %llu.\n",
2132 position, info->reshape_progress);
2133 ret_val = -1;
2134 } else {
2135 dprintf("Reshape position in md and metadata "
2136 "are the same;");
2137 ret_val = 1;
2138 }
2139 }
2140 } else if (rv == 0) {
2141 /* for valid sysfs entry, 0-length content
2142 * should be indicated as error
2143 */
2144 ret_val = -1;
2145 }
2146
2147 return ret_val;
2148 }
2149
2150 static unsigned long long choose_offset(unsigned long long lo,
2151 unsigned long long hi,
2152 unsigned long long min,
2153 unsigned long long max)
2154 {
2155 /* Choose a new offset between hi and lo.
2156 * It must be between min and max, but
2157 * we would prefer something near the middle of hi/lo, and also
2158 * prefer to be aligned to a big power of 2.
2159 *
2160 * So we start with the middle, then for each bit,
2161 * starting at '1' and increasing, if it is set, we either
2162 * add it or subtract it if possible, preferring the option
2163 * which is furthest from the boundary.
2164 *
2165 * We stop once we get a 1MB alignment. As units are in sectors,
2166 * 1MB = 2*1024 sectors.
2167 */
2168 unsigned long long choice = (lo + hi) / 2;
2169 unsigned long long bit = 1;
2170
2171 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2172 unsigned long long bigger, smaller;
2173 if (! (bit & choice))
2174 continue;
2175 bigger = choice + bit;
2176 smaller = choice - bit;
2177 if (bigger > max && smaller < min)
2178 break;
2179 if (bigger > max)
2180 choice = smaller;
2181 else if (smaller < min)
2182 choice = bigger;
2183 else if (hi - bigger > smaller - lo)
2184 choice = bigger;
2185 else
2186 choice = smaller;
2187 }
2188 return choice;
2189 }
2190
2191 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2192 char *devname, int delta_disks,
2193 unsigned long long data_offset,
2194 unsigned long long min)
2195 {
2196 struct mdinfo *sd;
2197 int dir = 0;
2198 int err = 0;
2199 unsigned long long before, after;
2200
2201 /* Need to find min space before and after so same is used
2202 * on all devices
2203 */
2204 before = UINT64_MAX;
2205 after = UINT64_MAX;
2206 for (sd = sra->devs; sd; sd = sd->next) {
2207 char *dn;
2208 int dfd;
2209 int rv;
2210 struct supertype *st2;
2211 struct mdinfo info2;
2212
2213 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2214 continue;
2215 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2216 dfd = dev_open(dn, O_RDONLY);
2217 if (dfd < 0) {
2218 pr_err("%s: cannot open component %s\n",
2219 devname, dn ? dn : "-unknown-");
2220 goto release;
2221 }
2222 st2 = dup_super(st);
2223 rv = st2->ss->load_super(st2,dfd, NULL);
2224 close(dfd);
2225 if (rv) {
2226 free(st2);
2227 pr_err("%s: cannot get superblock from %s\n",
2228 devname, dn);
2229 goto release;
2230 }
2231 st2->ss->getinfo_super(st2, &info2, NULL);
2232 st2->ss->free_super(st2);
2233 free(st2);
2234 if (info2.space_before == 0 &&
2235 info2.space_after == 0) {
2236 /* Metadata doesn't support data_offset changes */
2237 return 1;
2238 }
2239 if (before > info2.space_before)
2240 before = info2.space_before;
2241 if (after > info2.space_after)
2242 after = info2.space_after;
2243
2244 if (data_offset != INVALID_SECTORS) {
2245 if (dir == 0) {
2246 if (info2.data_offset == data_offset) {
2247 pr_err("%s: already has that data_offset\n",
2248 dn);
2249 goto release;
2250 }
2251 if (data_offset < info2.data_offset)
2252 dir = -1;
2253 else
2254 dir = 1;
2255 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2256 (data_offset >= info2.data_offset && dir == -1)) {
2257 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2258 dn);
2259 goto release;
2260 }
2261 }
2262 }
2263 if (before == UINT64_MAX)
2264 /* impossible really, there must be no devices */
2265 return 1;
2266
2267 for (sd = sra->devs; sd; sd = sd->next) {
2268 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2269 unsigned long long new_data_offset;
2270
2271 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2272 continue;
2273 if (delta_disks < 0) {
2274 /* Don't need any space as array is shrinking
2275 * just move data_offset up by min
2276 */
2277 if (data_offset == INVALID_SECTORS)
2278 new_data_offset = sd->data_offset + min;
2279 else {
2280 if (data_offset < sd->data_offset + min) {
2281 pr_err("--data-offset too small for %s\n",
2282 dn);
2283 goto release;
2284 }
2285 new_data_offset = data_offset;
2286 }
2287 } else if (delta_disks > 0) {
2288 /* need space before */
2289 if (before < min) {
2290 pr_err("Insufficient head-space for reshape on %s\n",
2291 dn);
2292 goto release;
2293 }
2294 if (data_offset == INVALID_SECTORS)
2295 new_data_offset = sd->data_offset - min;
2296 else {
2297 if (data_offset > sd->data_offset - min) {
2298 pr_err("--data-offset too large for %s\n",
2299 dn);
2300 goto release;
2301 }
2302 new_data_offset = data_offset;
2303 }
2304 } else {
2305 if (dir == 0) {
2306 /* can move up or down. If 'data_offset'
2307 * was set we would have already decided,
2308 * so just choose direction with most space.
2309 */
2310 if (before > after)
2311 dir = -1;
2312 else
2313 dir = 1;
2314 }
2315 sysfs_set_str(sra, NULL, "reshape_direction",
2316 dir == 1 ? "backwards" : "forwards");
2317 if (dir > 0) {
2318 /* Increase data offset */
2319 if (after < min) {
2320 pr_err("Insufficient tail-space for reshape on %s\n",
2321 dn);
2322 goto release;
2323 }
2324 if (data_offset != INVALID_SECTORS &&
2325 data_offset < sd->data_offset + min) {
2326 pr_err("--data-offset too small on %s\n",
2327 dn);
2328 goto release;
2329 }
2330 if (data_offset != INVALID_SECTORS)
2331 new_data_offset = data_offset;
2332 else
2333 new_data_offset = choose_offset(sd->data_offset,
2334 sd->data_offset + after,
2335 sd->data_offset + min,
2336 sd->data_offset + after);
2337 } else {
2338 /* Decrease data offset */
2339 if (before < min) {
2340 pr_err("insufficient head-room on %s\n",
2341 dn);
2342 goto release;
2343 }
2344 if (data_offset != INVALID_SECTORS &&
2345 data_offset < sd->data_offset - min) {
2346 pr_err("--data-offset too small on %s\n",
2347 dn);
2348 goto release;
2349 }
2350 if (data_offset != INVALID_SECTORS)
2351 new_data_offset = data_offset;
2352 else
2353 new_data_offset = choose_offset(sd->data_offset - before,
2354 sd->data_offset,
2355 sd->data_offset - before,
2356 sd->data_offset - min);
2357 }
2358 }
2359 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2360 if (err < 0 && errno == E2BIG) {
2361 /* try again after increasing data size to max */
2362 err = sysfs_set_num(sra, sd, "size", 0);
2363 if (err < 0 && errno == EINVAL &&
2364 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2365 /* some kernels have a bug where you cannot
2366 * use '0' on spare devices. */
2367 sysfs_set_num(sra, sd, "size",
2368 (sra->component_size + after)/2);
2369 }
2370 err = sysfs_set_num(sra, sd, "new_offset",
2371 new_data_offset);
2372 }
2373 if (err < 0) {
2374 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2375 pr_err("data-offset is too big for %s\n",
2376 dn);
2377 goto release;
2378 }
2379 if (sd == sra->devs &&
2380 (errno == ENOENT || errno == E2BIG))
2381 /* Early kernel, no 'new_offset' file,
2382 * or kernel doesn't like us.
2383 * For RAID5/6 this is not fatal
2384 */
2385 return 1;
2386 pr_err("Cannot set new_offset for %s\n",
2387 dn);
2388 break;
2389 }
2390 }
2391 return err;
2392 release:
2393 return -1;
2394 }
2395
2396 static int raid10_reshape(char *container, int fd, char *devname,
2397 struct supertype *st, struct mdinfo *info,
2398 struct reshape *reshape,
2399 unsigned long long data_offset,
2400 int force, int verbose)
2401 {
2402 /* Changing raid_disks, layout, chunksize or possibly
2403 * just data_offset for a RAID10.
2404 * We must always change data_offset. We change by at least
2405 * ->min_offset_change which is the largest of the old and new
2406 * chunk sizes.
2407 * If raid_disks is increasing, then data_offset must decrease
2408 * by at least this copy size.
2409 * If raid_disks is unchanged, data_offset must increase or
2410 * decrease by at least min_offset_change but preferably by much more.
2411 * We choose half of the available space.
2412 * If raid_disks is decreasing, data_offset must increase by
2413 * at least min_offset_change. To allow of this, component_size
2414 * must be decreased by the same amount.
2415 *
2416 * So we calculate the required minimum and direction, possibly
2417 * reduce the component_size, then iterate through the devices
2418 * and set the new_data_offset.
2419 * If that all works, we set chunk_size, layout, raid_disks, and start
2420 * 'reshape'
2421 */
2422 struct mdinfo *sra;
2423 unsigned long long min;
2424 int err = 0;
2425
2426 sra = sysfs_read(fd, NULL,
2427 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2428 );
2429 if (!sra) {
2430 pr_err("%s: Cannot get array details from sysfs\n",
2431 devname);
2432 goto release;
2433 }
2434 min = reshape->min_offset_change;
2435
2436 if (info->delta_disks)
2437 sysfs_set_str(sra, NULL, "reshape_direction",
2438 info->delta_disks < 0 ? "backwards" : "forwards");
2439 if (info->delta_disks < 0 &&
2440 info->space_after < min) {
2441 int rv = sysfs_set_num(sra, NULL, "component_size",
2442 (sra->component_size -
2443 min)/2);
2444 if (rv) {
2445 pr_err("cannot reduce component size\n");
2446 goto release;
2447 }
2448 }
2449 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2450 min);
2451 if (err == 1) {
2452 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2453 cont_err("supported on this kernel\n");
2454 err = -1;
2455 }
2456 if (err < 0)
2457 goto release;
2458
2459 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2460 err = errno;
2461 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2462 err = errno;
2463 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2464 info->array.raid_disks + info->delta_disks) < 0)
2465 err = errno;
2466 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2467 err = errno;
2468 if (err) {
2469 pr_err("Cannot set array shape for %s\n",
2470 devname);
2471 if (err == EBUSY &&
2472 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2473 cont_err(" Bitmap must be removed before"
2474 " shape can be changed\n");
2475 goto release;
2476 }
2477 sysfs_free(sra);
2478 return 0;
2479 release:
2480 sysfs_free(sra);
2481 return 1;
2482 }
2483
2484 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2485 {
2486 struct mdinfo *sra, *sd;
2487 /* Initialisation to silence compiler warning */
2488 unsigned long long min_space_before = 0, min_space_after = 0;
2489 int first = 1;
2490
2491 sra = sysfs_read(fd, NULL, GET_DEVS);
2492 if (!sra)
2493 return;
2494 for (sd = sra->devs; sd; sd = sd->next) {
2495 char *dn;
2496 int dfd;
2497 struct supertype *st2;
2498 struct mdinfo info2;
2499
2500 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2501 continue;
2502 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2503 dfd = dev_open(dn, O_RDONLY);
2504 if (dfd < 0)
2505 break;
2506 st2 = dup_super(st);
2507 if (st2->ss->load_super(st2,dfd, NULL)) {
2508 close(dfd);
2509 free(st2);
2510 break;
2511 }
2512 close(dfd);
2513 st2->ss->getinfo_super(st2, &info2, NULL);
2514 st2->ss->free_super(st2);
2515 free(st2);
2516 if (first ||
2517 min_space_before > info2.space_before)
2518 min_space_before = info2.space_before;
2519 if (first ||
2520 min_space_after > info2.space_after)
2521 min_space_after = info2.space_after;
2522 first = 0;
2523 }
2524 if (sd == NULL && !first) {
2525 info->space_after = min_space_after;
2526 info->space_before = min_space_before;
2527 }
2528 sysfs_free(sra);
2529 }
2530
2531 static void update_cache_size(char *container, struct mdinfo *sra,
2532 struct mdinfo *info,
2533 int disks, unsigned long long blocks)
2534 {
2535 /* Check that the internal stripe cache is
2536 * large enough, or it won't work.
2537 * It must hold at least 4 stripes of the larger
2538 * chunk size
2539 */
2540 unsigned long cache;
2541 cache = max(info->array.chunk_size, info->new_chunk);
2542 cache *= 4; /* 4 stripes minimum */
2543 cache /= 512; /* convert to sectors */
2544 /* make sure there is room for 'blocks' with a bit to spare */
2545 if (cache < 16 + blocks / disks)
2546 cache = 16 + blocks / disks;
2547 cache /= (4096/512); /* Covert from sectors to pages */
2548
2549 if (sra->cache_size < cache)
2550 subarray_set_num(container, sra, "stripe_cache_size",
2551 cache+1);
2552 }
2553
2554 static int impose_reshape(struct mdinfo *sra,
2555 struct mdinfo *info,
2556 struct supertype *st,
2557 int fd,
2558 int restart,
2559 char *devname, char *container,
2560 struct reshape *reshape)
2561 {
2562 struct mdu_array_info_s array;
2563
2564 sra->new_chunk = info->new_chunk;
2565
2566 if (restart) {
2567 /* for external metadata checkpoint saved by mdmon can be lost
2568 * or missed /due to e.g. crash/. Check if md is not during
2569 * restart farther than metadata points to.
2570 * If so, this means metadata information is obsolete.
2571 */
2572 if (st->ss->external)
2573 verify_reshape_position(info, reshape->level);
2574 sra->reshape_progress = info->reshape_progress;
2575 } else {
2576 sra->reshape_progress = 0;
2577 if (reshape->after.data_disks < reshape->before.data_disks)
2578 /* start from the end of the new array */
2579 sra->reshape_progress = (sra->component_size
2580 * reshape->after.data_disks);
2581 }
2582
2583 ioctl(fd, GET_ARRAY_INFO, &array);
2584 if (info->array.chunk_size == info->new_chunk &&
2585 reshape->before.layout == reshape->after.layout &&
2586 st->ss->external == 0) {
2587 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2588 array.raid_disks = reshape->after.data_disks + reshape->parity;
2589 if (!restart &&
2590 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2591 int err = errno;
2592
2593 pr_err("Cannot set device shape for %s: %s\n",
2594 devname, strerror(errno));
2595
2596 if (err == EBUSY &&
2597 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2598 cont_err("Bitmap must be removed before"
2599 " shape can be changed\n");
2600
2601 goto release;
2602 }
2603 } else if (!restart) {
2604 /* set them all just in case some old 'new_*' value
2605 * persists from some earlier problem.
2606 */
2607 int err = 0;
2608 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2609 err = errno;
2610 if (!err && sysfs_set_num(sra, NULL, "layout",
2611 reshape->after.layout) < 0)
2612 err = errno;
2613 if (!err && subarray_set_num(container, sra, "raid_disks",
2614 reshape->after.data_disks +
2615 reshape->parity) < 0)
2616 err = errno;
2617 if (err) {
2618 pr_err("Cannot set device shape for %s\n",
2619 devname);
2620
2621 if (err == EBUSY &&
2622 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2623 cont_err("Bitmap must be removed before"
2624 " shape can be changed\n");
2625 goto release;
2626 }
2627 }
2628 return 0;
2629 release:
2630 return -1;
2631 }
2632
2633 static int reshape_array(char *container, int fd, char *devname,
2634 struct supertype *st, struct mdinfo *info,
2635 int force, struct mddev_dev *devlist,
2636 unsigned long long data_offset,
2637 char *backup_file, int verbose, int forked,
2638 int restart, int freeze_reshape)
2639 {
2640 struct reshape reshape;
2641 int spares_needed;
2642 char *msg;
2643 int orig_level = UnSet;
2644 int odisks;
2645 int delayed;
2646
2647 struct mdu_array_info_s array;
2648 char *c;
2649
2650 struct mddev_dev *dv;
2651 int added_disks;
2652
2653 int *fdlist = NULL;
2654 unsigned long long *offsets = NULL;
2655 int d;
2656 int nrdisks;
2657 int err;
2658 unsigned long blocks;
2659 unsigned long long array_size;
2660 int done;
2661 struct mdinfo *sra = NULL;
2662
2663 /* when reshaping a RAID0, the component_size might be zero.
2664 * So try to fix that up.
2665 */
2666 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2667 dprintf("Cannot get array information.\n");
2668 goto release;
2669 }
2670 if (array.level == 0 && info->component_size == 0) {
2671 get_dev_size(fd, NULL, &array_size);
2672 info->component_size = array_size / array.raid_disks;
2673 }
2674
2675 if (array.level == 10)
2676 /* Need space_after info */
2677 get_space_after(fd, st, info);
2678
2679 if (info->reshape_active) {
2680 int new_level = info->new_level;
2681 info->new_level = UnSet;
2682 if (info->delta_disks > 0)
2683 info->array.raid_disks -= info->delta_disks;
2684 msg = analyse_change(info, &reshape);
2685 info->new_level = new_level;
2686 if (info->delta_disks > 0)
2687 info->array.raid_disks += info->delta_disks;
2688 if (!restart)
2689 /* Make sure the array isn't read-only */
2690 ioctl(fd, RESTART_ARRAY_RW, 0);
2691 } else
2692 msg = analyse_change(info, &reshape);
2693 if (msg) {
2694 pr_err("%s\n", msg);
2695 goto release;
2696 }
2697 if (restart &&
2698 (reshape.level != info->array.level ||
2699 reshape.before.layout != info->array.layout ||
2700 reshape.before.data_disks + reshape.parity
2701 != info->array.raid_disks - max(0, info->delta_disks))) {
2702 pr_err("reshape info is not in native format -"
2703 " cannot continue.\n");
2704 goto release;
2705 }
2706
2707 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2708 /* When reshape is restarted from '0', very begin of array
2709 * it is possible that for external metadata reshape and array
2710 * configuration doesn't happen.
2711 * Check if md has the same opinion, and reshape is restarted
2712 * from 0. If so, this is regular reshape start after reshape
2713 * switch in metadata to next array only.
2714 */
2715 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2716 (info->reshape_progress == 0))
2717 restart = 0;
2718 }
2719 if (restart) {
2720 /* reshape already started. just skip to monitoring the reshape */
2721 if (reshape.backup_blocks == 0)
2722 return 0;
2723 if (restart & RESHAPE_NO_BACKUP)
2724 return 0;
2725 goto started;
2726 }
2727 /* The container is frozen but the array may not be.
2728 * So freeze the array so spares don't get put to the wrong use
2729 * FIXME there should probably be a cleaner separation between
2730 * freeze_array and freeze_container.
2731 */
2732 sysfs_freeze_array(info);
2733 /* Check we have enough spares to not be degraded */
2734 added_disks = 0;
2735 for (dv = devlist; dv ; dv=dv->next)
2736 added_disks++;
2737 spares_needed = max(reshape.before.data_disks,
2738 reshape.after.data_disks)
2739 + reshape.parity - array.raid_disks;
2740
2741 if (!force &&
2742 info->new_level > 1 && info->array.level > 1 &&
2743 spares_needed > info->array.spare_disks + added_disks) {
2744 pr_err("Need %d spare%s to avoid degraded array,"
2745 " and only have %d.\n"
2746 " Use --force to over-ride this check.\n",
2747 spares_needed,
2748 spares_needed == 1 ? "" : "s",
2749 info->array.spare_disks + added_disks);
2750 goto release;
2751 }
2752 /* Check we have enough spares to not fail */
2753 spares_needed = max(reshape.before.data_disks,
2754 reshape.after.data_disks)
2755 - array.raid_disks;
2756 if ((info->new_level > 1 || info->new_level == 0) &&
2757 spares_needed > info->array.spare_disks +added_disks) {
2758 pr_err("Need %d spare%s to create working array,"
2759 " and only have %d.\n",
2760 spares_needed,
2761 spares_needed == 1 ? "" : "s",
2762 info->array.spare_disks + added_disks);
2763 goto release;
2764 }
2765
2766 if (reshape.level != array.level) {
2767 char *c = map_num(pers, reshape.level);
2768 int err;
2769 if (c == NULL)
2770 goto release;
2771
2772 err = sysfs_set_str(info, NULL, "level", c);
2773 if (err) {
2774 err = errno;
2775 pr_err("%s: could not set level to %s\n",
2776 devname, c);
2777 if (err == EBUSY &&
2778 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2779 cont_err("Bitmap must be removed"
2780 " before level can be changed\n");
2781 goto release;
2782 }
2783 if (verbose >= 0)
2784 pr_err("level of %s changed to %s\n",
2785 devname, c);
2786 orig_level = array.level;
2787 sysfs_freeze_array(info);
2788
2789 if (reshape.level > 0 && st->ss->external) {
2790 /* make sure mdmon is aware of the new level */
2791 if (mdmon_running(container))
2792 flush_mdmon(container);
2793
2794 if (!mdmon_running(container))
2795 start_mdmon(container);
2796 ping_monitor(container);
2797 if (mdmon_running(container) &&
2798 st->update_tail == NULL)
2799 st->update_tail = &st->updates;
2800 }
2801 }
2802 /* ->reshape_super might have chosen some spares from the
2803 * container that it wants to be part of the new array.
2804 * We can collect them with ->container_content and give
2805 * them to the kernel.
2806 */
2807 if (st->ss->reshape_super && st->ss->container_content) {
2808 char *subarray = strchr(info->text_version+1, '/')+1;
2809 struct mdinfo *info2 =
2810 st->ss->container_content(st, subarray);
2811 struct mdinfo *d;
2812
2813 if (info2) {
2814 sysfs_init(info2, fd, st->devnm);
2815 /* When increasing number of devices, we need to set
2816 * new raid_disks before adding these, or they might
2817 * be rejected.
2818 */
2819 if (reshape.backup_blocks &&
2820 reshape.after.data_disks > reshape.before.data_disks)
2821 subarray_set_num(container, info2, "raid_disks",
2822 reshape.after.data_disks +
2823 reshape.parity);
2824 for (d = info2->devs; d; d = d->next) {
2825 if (d->disk.state == 0 &&
2826 d->disk.raid_disk >= 0) {
2827 /* This is a spare that wants to
2828 * be part of the array.
2829 */
2830 add_disk(fd, st, info2, d);
2831 }
2832 }
2833 sysfs_free(info2);
2834 }
2835 }
2836 /* We might have been given some devices to add to the
2837 * array. Now that the array has been changed to the right
2838 * level and frozen, we can safely add them.
2839 */
2840 if (devlist)
2841 Manage_subdevs(devname, fd, devlist, verbose,
2842 0,NULL, 0);
2843
2844 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
2845 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
2846 if (reshape.backup_blocks == 0) {
2847 /* No restriping needed, but we might need to impose
2848 * some more changes: layout, raid_disks, chunk_size
2849 */
2850 /* read current array info */
2851 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2852 dprintf("Cannot get array information.\n");
2853 goto release;
2854 }
2855 /* compare current array info with new values and if
2856 * it is different update them to new */
2857 if (info->new_layout != UnSet &&
2858 info->new_layout != array.layout) {
2859 array.layout = info->new_layout;
2860 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2861 pr_err("failed to set new layout\n");
2862 goto release;
2863 } else if (verbose >= 0)
2864 printf("layout for %s set to %d\n",
2865 devname, array.layout);
2866 }
2867 if (info->delta_disks != UnSet &&
2868 info->delta_disks != 0 &&
2869 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
2870 array.raid_disks += info->delta_disks;
2871 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2872 pr_err("failed to set raid disks\n");
2873 goto release;
2874 } else if (verbose >= 0) {
2875 printf("raid_disks for %s set to %d\n",
2876 devname, array.raid_disks);
2877 }
2878 }
2879 if (info->new_chunk != 0 &&
2880 info->new_chunk != array.chunk_size) {
2881 if (sysfs_set_num(info, NULL,
2882 "chunk_size", info->new_chunk) != 0) {
2883 pr_err("failed to set chunk size\n");
2884 goto release;
2885 } else if (verbose >= 0)
2886 printf("chunk size for %s set to %d\n",
2887 devname, array.chunk_size);
2888 }
2889 unfreeze(st);
2890 return 0;
2891 }
2892
2893 /*
2894 * There are three possibilities.
2895 * 1/ The array will shrink.
2896 * We need to ensure the reshape will pause before reaching
2897 * the 'critical section'. We also need to fork and wait for
2898 * that to happen. When it does we
2899 * suspend/backup/complete/unfreeze
2900 *
2901 * 2/ The array will not change size.
2902 * This requires that we keep a backup of a sliding window
2903 * so that we can restore data after a crash. So we need
2904 * to fork and monitor progress.
2905 * In future we will allow the data_offset to change, so
2906 * a sliding backup becomes unnecessary.
2907 *
2908 * 3/ The array will grow. This is relatively easy.
2909 * However the kernel's restripe routines will cheerfully
2910 * overwrite some early data before it is safe. So we
2911 * need to make a backup of the early parts of the array
2912 * and be ready to restore it if rebuild aborts very early.
2913 * For externally managed metadata, we still need a forked
2914 * child to monitor the reshape and suspend IO over the region
2915 * that is being reshaped.
2916 *
2917 * We backup data by writing it to one spare, or to a
2918 * file which was given on command line.
2919 *
2920 * In each case, we first make sure that storage is available
2921 * for the required backup.
2922 * Then we:
2923 * - request the shape change.
2924 * - fork to handle backup etc.
2925 */
2926 /* Check that we can hold all the data */
2927 get_dev_size(fd, NULL, &array_size);
2928 if (reshape.new_size < (array_size/512)) {
2929 pr_err("this change will reduce the size of the array.\n"
2930 " use --grow --array-size first to truncate array.\n"
2931 " e.g. mdadm --grow %s --array-size %llu\n",
2932 devname, reshape.new_size/2);
2933 goto release;
2934 }
2935
2936 if (array.level == 10) {
2937 /* Reshaping RAID10 does not require any data backup by
2938 * user-space. Instead it requires that the data_offset
2939 * is changed to avoid the need for backup.
2940 * So this is handled very separately
2941 */
2942 if (restart)
2943 /* Nothing to do. */
2944 return 0;
2945 return raid10_reshape(container, fd, devname, st, info,
2946 &reshape, data_offset,
2947 force, verbose);
2948 }
2949 sra = sysfs_read(fd, NULL,
2950 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2951 GET_CACHE);
2952 if (!sra) {
2953 pr_err("%s: Cannot get array details from sysfs\n",
2954 devname);
2955 goto release;
2956 }
2957
2958 if (!backup_file)
2959 switch(set_new_data_offset(sra, st, devname,
2960 reshape.after.data_disks - reshape.before.data_disks,
2961 data_offset,
2962 reshape.min_offset_change)) {
2963 case -1:
2964 goto release;
2965 case 0:
2966 /* Updated data_offset, so it's easy now */
2967 update_cache_size(container, sra, info,
2968 min(reshape.before.data_disks,
2969 reshape.after.data_disks),
2970 reshape.backup_blocks);
2971
2972 /* Right, everything seems fine. Let's kick things off.
2973 */
2974 sync_metadata(st);
2975
2976 if (impose_reshape(sra, info, st, fd, restart,
2977 devname, container, &reshape) < 0)
2978 goto release;
2979 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
2980 pr_err("Failed to initiate reshape!\n");
2981 goto release;
2982 }
2983
2984 return 0;
2985 case 1: /* Couldn't set data_offset, try the old way */
2986 if (data_offset != INVALID_SECTORS) {
2987 pr_err("Cannot update data_offset on this array\n");
2988 goto release;
2989 }
2990 break;
2991 }
2992
2993 started:
2994 /* Decide how many blocks (sectors) for a reshape
2995 * unit. The number we have so far is just a minimum
2996 */
2997 blocks = reshape.backup_blocks;
2998 if (reshape.before.data_disks ==
2999 reshape.after.data_disks) {
3000 /* Make 'blocks' bigger for better throughput, but
3001 * not so big that we reject it below.
3002 * Try for 16 megabytes
3003 */
3004 while (blocks * 32 < sra->component_size &&
3005 blocks < 16*1024*2)
3006 blocks *= 2;
3007 } else
3008 pr_err("Need to backup %luK of critical "
3009 "section..\n", blocks/2);
3010
3011 if (blocks >= sra->component_size/2) {
3012 pr_err("%s: Something wrong"
3013 " - reshape aborted\n",
3014 devname);
3015 goto release;
3016 }
3017
3018 /* Now we need to open all these devices so we can read/write.
3019 */
3020 nrdisks = max(reshape.before.data_disks,
3021 reshape.after.data_disks) + reshape.parity
3022 + sra->array.spare_disks;
3023 fdlist = xcalloc((1+nrdisks), sizeof(int));
3024 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3025
3026 odisks = reshape.before.data_disks + reshape.parity;
3027 d = reshape_prepare_fdlist(devname, sra, odisks,
3028 nrdisks, blocks, backup_file,
3029 fdlist, offsets);
3030 if (d < 0) {
3031 goto release;
3032 }
3033 if ((st->ss->manage_reshape == NULL) ||
3034 (st->ss->recover_backup == NULL)) {
3035 if (backup_file == NULL) {
3036 if (reshape.after.data_disks <=
3037 reshape.before.data_disks) {
3038 pr_err("%s: Cannot grow - "
3039 "need backup-file\n", devname);
3040 goto release;
3041 } else if (sra->array.spare_disks == 0) {
3042 pr_err("%s: Cannot grow - "
3043 "need a spare or backup-file to backup "
3044 "critical section\n", devname);
3045 goto release;
3046 }
3047 } else {
3048 if (!reshape_open_backup_file(backup_file, fd, devname,
3049 (signed)blocks,
3050 fdlist+d, offsets+d,
3051 restart)) {
3052 goto release;
3053 }
3054 d++;
3055 }
3056 }
3057
3058 update_cache_size(container, sra, info,
3059 min(reshape.before.data_disks, reshape.after.data_disks),
3060 blocks);
3061
3062 /* Right, everything seems fine. Let's kick things off.
3063 * If only changing raid_disks, use ioctl, else use
3064 * sysfs.
3065 */
3066 sync_metadata(st);
3067
3068 if (impose_reshape(sra, info, st, fd, restart,
3069 devname, container, &reshape) < 0)
3070 goto release;
3071
3072 err = start_reshape(sra, restart, reshape.before.data_disks,
3073 reshape.after.data_disks);
3074 if (err) {
3075 pr_err("Cannot %s reshape for %s\n",
3076 restart ? "continue" : "start",
3077 devname);
3078 goto release;
3079 }
3080 if (restart)
3081 sysfs_set_str(sra, NULL, "array_state", "active");
3082 if (freeze_reshape) {
3083 free(fdlist);
3084 free(offsets);
3085 sysfs_free(sra);
3086 pr_err("Reshape has to be continued from"
3087 " location %llu when root filesystem has been mounted.\n",
3088 sra->reshape_progress);
3089 return 1;
3090 }
3091
3092 /* Now we just need to kick off the reshape and watch, while
3093 * handling backups of the data...
3094 * This is all done by a forked background process.
3095 */
3096 switch(forked ? 0 : fork()) {
3097 case -1:
3098 pr_err("Cannot run child to monitor reshape: %s\n",
3099 strerror(errno));
3100 abort_reshape(sra);
3101 goto release;
3102 default:
3103 free(fdlist);
3104 free(offsets);
3105 sysfs_free(sra);
3106 return 0;
3107 case 0:
3108 map_fork();
3109 break;
3110 }
3111
3112 /* If another array on the same devices is busy, the
3113 * reshape will wait for them. This would mean that
3114 * the first section that we suspend will stay suspended
3115 * for a long time. So check on that possibility
3116 * by looking for "DELAYED" in /proc/mdstat, and if found,
3117 * wait a while
3118 */
3119 do {
3120 struct mdstat_ent *mds, *m;
3121 delayed = 0;
3122 mds = mdstat_read(0, 0);
3123 for (m = mds; m; m = m->next)
3124 if (strcmp(m->devnm, sra->sys_name) == 0) {
3125 if (m->resync &&
3126 m->percent == RESYNC_DELAYED)
3127 delayed = 1;
3128 if (m->resync == 0)
3129 /* Haven't started the reshape thread
3130 * yet, wait a bit
3131 */
3132 delayed = 2;
3133 break;
3134 }
3135 free_mdstat(mds);
3136 if (delayed == 1 && get_linux_version() < 3007000) {
3137 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3138 " You might experience problems until other reshapes complete.\n");
3139 delayed = 0;
3140 }
3141 if (delayed)
3142 sleep(30 - (delayed-1) * 25);
3143 } while (delayed);
3144
3145 close(fd);
3146 if (check_env("MDADM_GROW_VERIFY"))
3147 fd = open(devname, O_RDONLY | O_DIRECT);
3148 else
3149 fd = -1;
3150 mlockall(MCL_FUTURE);
3151
3152 if (st->ss->external) {
3153 /* metadata handler takes it from here */
3154 done = st->ss->manage_reshape(
3155 fd, sra, &reshape, st, blocks,
3156 fdlist, offsets,
3157 d - odisks, fdlist+odisks,
3158 offsets+odisks);
3159 } else
3160 done = child_monitor(
3161 fd, sra, &reshape, st, blocks,
3162 fdlist, offsets,
3163 d - odisks, fdlist+odisks,
3164 offsets+odisks);
3165
3166 free(fdlist);
3167 free(offsets);
3168
3169 if (backup_file && done)
3170 unlink(backup_file);
3171 if (!done) {
3172 abort_reshape(sra);
3173 goto out;
3174 }
3175
3176 if (!st->ss->external &&
3177 !(reshape.before.data_disks != reshape.after.data_disks
3178 && info->custom_array_size) &&
3179 info->new_level == reshape.level &&
3180 !forked) {
3181 /* no need to wait for the reshape to finish as
3182 * there is nothing more to do.
3183 */
3184 sysfs_free(sra);
3185 exit(0);
3186 }
3187 wait_reshape(sra);
3188
3189 if (st->ss->external) {
3190 /* Re-load the metadata as much could have changed */
3191 int cfd = open_dev(st->container_devnm);
3192 if (cfd >= 0) {
3193 flush_mdmon(container);
3194 st->ss->free_super(st);
3195 st->ss->load_container(st, cfd, container);
3196 close(cfd);
3197 }
3198 }
3199
3200 /* set new array size if required customer_array_size is used
3201 * by this metadata.
3202 */
3203 if (reshape.before.data_disks !=
3204 reshape.after.data_disks &&
3205 info->custom_array_size)
3206 set_array_size(st, info, info->text_version);
3207
3208 if (info->new_level != reshape.level) {
3209
3210 c = map_num(pers, info->new_level);
3211 if (c) {
3212 err = sysfs_set_str(sra, NULL, "level", c);
3213 if (err)
3214 pr_err("%s: could not set level "
3215 "to %s\n", devname, c);
3216 }
3217 if (info->new_level == 0)
3218 st->update_tail = NULL;
3219 }
3220 out:
3221 sysfs_free(sra);
3222 if (forked)
3223 return 0;
3224 unfreeze(st);
3225 exit(0);
3226
3227 release:
3228 free(fdlist);
3229 free(offsets);
3230 if (orig_level != UnSet && sra) {
3231 c = map_num(pers, orig_level);
3232 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3233 pr_err("aborting level change\n");
3234 }
3235 sysfs_free(sra);
3236 if (!forked)
3237 unfreeze(st);
3238 return 1;
3239 }
3240
3241 /* mdfd handle is passed to be closed in child process (after fork).
3242 */
3243 int reshape_container(char *container, char *devname,
3244 int mdfd,
3245 struct supertype *st,
3246 struct mdinfo *info,
3247 int force,
3248 char *backup_file,
3249 int verbose, int restart, int freeze_reshape)
3250 {
3251 struct mdinfo *cc = NULL;
3252 int rv = restart;
3253 char last_devnm[32] = "";
3254
3255 /* component_size is not meaningful for a container,
3256 * so pass '0' meaning 'no change'
3257 */
3258 if (!restart &&
3259 reshape_super(st, 0, info->new_level,
3260 info->new_layout, info->new_chunk,
3261 info->array.raid_disks, info->delta_disks,
3262 backup_file, devname, APPLY_METADATA_CHANGES,
3263 verbose)) {
3264 unfreeze(st);
3265 return 1;
3266 }
3267
3268 sync_metadata(st);
3269
3270 /* ping monitor to be sure that update is on disk
3271 */
3272 ping_monitor(container);
3273
3274 switch (fork()) {
3275 case -1: /* error */
3276 perror("Cannot fork to complete reshape\n");
3277 unfreeze(st);
3278 return 1;
3279 default: /* parent */
3280 if (!freeze_reshape)
3281 printf(Name ": multi-array reshape continues"
3282 " in background\n");
3283 return 0;
3284 case 0: /* child */
3285 map_fork();
3286 break;
3287 }
3288
3289 /* close unused handle in child process
3290 */
3291 if (mdfd > -1)
3292 close(mdfd);
3293
3294 while(1) {
3295 /* For each member array with reshape_active,
3296 * we need to perform the reshape.
3297 * We pick the first array that needs reshaping and
3298 * reshape it. reshape_array() will re-read the metadata
3299 * so the next time through a different array should be
3300 * ready for reshape.
3301 * It is possible that the 'different' array will not
3302 * be assembled yet. In that case we simple exit.
3303 * When it is assembled, the mdadm which assembles it
3304 * will take over the reshape.
3305 */
3306 struct mdinfo *content;
3307 int fd;
3308 struct mdstat_ent *mdstat;
3309 char *adev;
3310 int devid;
3311
3312 sysfs_free(cc);
3313
3314 cc = st->ss->container_content(st, NULL);
3315
3316 for (content = cc; content ; content = content->next) {
3317 char *subarray;
3318 if (!content->reshape_active)
3319 continue;
3320
3321 subarray = strchr(content->text_version+1, '/')+1;
3322 mdstat = mdstat_by_subdev(subarray, container);
3323 if (!mdstat)
3324 continue;
3325 if (mdstat->active == 0) {
3326 pr_err("Skipping inactive array %s.\n",
3327 mdstat->devnm);
3328 free_mdstat(mdstat);
3329 mdstat = NULL;
3330 continue;
3331 }
3332 break;
3333 }
3334 if (!content)
3335 break;
3336
3337 devid = devnm2devid(mdstat->devnm);
3338 adev = map_dev(major(devid), minor(devid), 0);
3339 if (!adev)
3340 adev = content->text_version;
3341
3342 fd = open_dev(mdstat->devnm);
3343 if (fd < 0) {
3344 printf(Name ": Device %s cannot be opened for reshape.",
3345 adev);
3346 break;
3347 }
3348
3349 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3350 /* Do not allow for multiple reshape_array() calls for
3351 * the same array.
3352 * It can happen when reshape_array() returns without
3353 * error, when reshape is not finished (wrong reshape
3354 * starting/continuation conditions). Mdmon doesn't
3355 * switch to next array in container and reentry
3356 * conditions for the same array occur.
3357 * This is possibly interim until the behaviour of
3358 * reshape_array is resolved().
3359 */
3360 printf(Name ": Multiple reshape execution detected for "
3361 "device %s.", adev);
3362 close(fd);
3363 break;
3364 }
3365 strcpy(last_devnm, mdstat->devnm);
3366
3367 sysfs_init(content, fd, mdstat->devnm);
3368
3369 if (mdmon_running(container))
3370 flush_mdmon(container);
3371
3372 rv = reshape_array(container, fd, adev, st,
3373 content, force, NULL, 0ULL,
3374 backup_file, verbose, 1, restart,
3375 freeze_reshape);
3376 close(fd);
3377
3378 if (freeze_reshape) {
3379 sysfs_free(cc);
3380 exit(0);
3381 }
3382
3383 restart = 0;
3384 if (rv)
3385 break;
3386
3387 if (mdmon_running(container))
3388 flush_mdmon(container);
3389 }
3390 if (!rv)
3391 unfreeze(st);
3392 sysfs_free(cc);
3393 exit(0);
3394 }
3395
3396 /*
3397 * We run a child process in the background which performs the following
3398 * steps:
3399 * - wait for resync to reach a certain point
3400 * - suspend io to the following section
3401 * - backup that section
3402 * - allow resync to proceed further
3403 * - resume io
3404 * - discard the backup.
3405 *
3406 * When are combined in slightly different ways in the three cases.
3407 * Grow:
3408 * - suspend/backup/allow/wait/resume/discard
3409 * Shrink:
3410 * - allow/wait/suspend/backup/allow/wait/resume/discard
3411 * same-size:
3412 * - wait/resume/discard/suspend/backup/allow
3413 *
3414 * suspend/backup/allow always come together
3415 * wait/resume/discard do too.
3416 * For the same-size case we have two backups to improve flow.
3417 *
3418 */
3419
3420 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3421 unsigned long long backup_point,
3422 unsigned long long wait_point,
3423 unsigned long long *suspend_point,
3424 unsigned long long *reshape_completed)
3425 {
3426 /* This function is called repeatedly by the reshape manager.
3427 * It determines how much progress can safely be made and allows
3428 * that progress.
3429 * - 'info' identifies the array and particularly records in
3430 * ->reshape_progress the metadata's knowledge of progress
3431 * This is a sector offset from the start of the array
3432 * of the next array block to be relocated. This number
3433 * may increase from 0 or decrease from array_size, depending
3434 * on the type of reshape that is happening.
3435 * Note that in contrast, 'sync_completed' is a block count of the
3436 * reshape so far. It gives the distance between the start point
3437 * (head or tail of device) and the next place that data will be
3438 * written. It always increases.
3439 * - 'reshape' is the structure created by analyse_change
3440 * - 'backup_point' shows how much the metadata manager has backed-up
3441 * data. For reshapes with increasing progress, it is the next address
3442 * to be backed up, previous addresses have been backed-up. For
3443 * decreasing progress, it is the earliest address that has been
3444 * backed up - later address are also backed up.
3445 * So addresses between reshape_progress and backup_point are
3446 * backed up providing those are in the 'correct' order.
3447 * - 'wait_point' is an array address. When reshape_completed
3448 * passes this point, progress_reshape should return. It might
3449 * return earlier if it determines that ->reshape_progress needs
3450 * to be updated or further backup is needed.
3451 * - suspend_point is maintained by progress_reshape and the caller
3452 * should not touch it except to initialise to zero.
3453 * It is an array address and it only increases in 2.6.37 and earlier.
3454 * This makes it difficult to handle reducing reshapes with
3455 * external metadata.
3456 * However: it is similar to backup_point in that it records the
3457 * other end of a suspended region from reshape_progress.
3458 * it is moved to extend the region that is safe to backup and/or
3459 * reshape
3460 * - reshape_completed is read from sysfs and returned. The caller
3461 * should copy this into ->reshape_progress when it has reason to
3462 * believe that the metadata knows this, and any backup outside this
3463 * has been erased.
3464 *
3465 * Return value is:
3466 * 1 if more data from backup_point - but only as far as suspend_point,
3467 * should be backed up
3468 * 0 if things are progressing smoothly
3469 * -1 if the reshape is finished because it is all done,
3470 * -2 if the reshape is finished due to an error.
3471 */
3472
3473 int advancing = (reshape->after.data_disks
3474 >= reshape->before.data_disks);
3475 unsigned long long need_backup; /* All data between start of array and
3476 * here will at some point need to
3477 * be backed up.
3478 */
3479 unsigned long long read_offset, write_offset;
3480 unsigned long long write_range;
3481 unsigned long long max_progress, target, completed;
3482 unsigned long long array_size = (info->component_size
3483 * reshape->before.data_disks);
3484 int fd;
3485 char buf[20];
3486
3487 /* First, we unsuspend any region that is now known to be safe.
3488 * If suspend_point is on the 'wrong' side of reshape_progress, then
3489 * we don't have or need suspension at the moment. This is true for
3490 * native metadata when we don't need to back-up.
3491 */
3492 if (advancing) {
3493 if (info->reshape_progress <= *suspend_point)
3494 sysfs_set_num(info, NULL, "suspend_lo",
3495 info->reshape_progress);
3496 } else {
3497 /* Note: this won't work in 2.6.37 and before.
3498 * Something somewhere should make sure we don't need it!
3499 */
3500 if (info->reshape_progress >= *suspend_point)
3501 sysfs_set_num(info, NULL, "suspend_hi",
3502 info->reshape_progress);
3503 }
3504
3505 /* Now work out how far it is safe to progress.
3506 * If the read_offset for ->reshape_progress is less than
3507 * 'blocks' beyond the write_offset, we can only progress as far
3508 * as a backup.
3509 * Otherwise we can progress until the write_offset for the new location
3510 * reaches (within 'blocks' of) the read_offset at the current location.
3511 * However that region must be suspended unless we are using native
3512 * metadata.
3513 * If we need to suspend more, we limit it to 128M per device, which is
3514 * rather arbitrary and should be some time-based calculation.
3515 */
3516 read_offset = info->reshape_progress / reshape->before.data_disks;
3517 write_offset = info->reshape_progress / reshape->after.data_disks;
3518 write_range = info->new_chunk/512;
3519 if (reshape->before.data_disks == reshape->after.data_disks)
3520 need_backup = array_size;
3521 else
3522 need_backup = reshape->backup_blocks;
3523 if (advancing) {
3524 if (read_offset < write_offset + write_range)
3525 max_progress = backup_point;
3526 else
3527 max_progress =
3528 read_offset *
3529 reshape->after.data_disks;
3530 } else {
3531 if (read_offset > write_offset - write_range)
3532 /* Can only progress as far as has been backed up,
3533 * which must be suspended */
3534 max_progress = backup_point;
3535 else if (info->reshape_progress <= need_backup)
3536 max_progress = backup_point;
3537 else {
3538 if (info->array.major_version >= 0)
3539 /* Can progress until backup is needed */
3540 max_progress = need_backup;
3541 else {
3542 /* Can progress until metadata update is required */
3543 max_progress =
3544 read_offset *
3545 reshape->after.data_disks;
3546 /* but data must be suspended */
3547 if (max_progress < *suspend_point)
3548 max_progress = *suspend_point;
3549 }
3550 }
3551 }
3552
3553 /* We know it is safe to progress to 'max_progress' providing
3554 * it is suspended or we are using native metadata.
3555 * Consider extending suspend_point 128M per device if it
3556 * is less than 64M per device beyond reshape_progress.
3557 * But always do a multiple of 'blocks'
3558 * FIXME this is too big - it takes to long to complete
3559 * this much.
3560 */
3561 target = 64*1024*2 * min(reshape->before.data_disks,
3562 reshape->after.data_disks);
3563 target /= reshape->backup_blocks;
3564 if (target < 2)
3565 target = 2;
3566 target *= reshape->backup_blocks;
3567
3568 /* For externally managed metadata we always need to suspend IO to
3569 * the area being reshaped so we regularly push suspend_point forward.
3570 * For native metadata we only need the suspend if we are going to do
3571 * a backup.
3572 */
3573 if (advancing) {
3574 if ((need_backup > info->reshape_progress
3575 || info->array.major_version < 0) &&
3576 *suspend_point < info->reshape_progress + target) {
3577 if (need_backup < *suspend_point + 2 * target)
3578 *suspend_point = need_backup;
3579 else if (*suspend_point + 2 * target < array_size)
3580 *suspend_point += 2 * target;
3581 else
3582 *suspend_point = array_size;
3583 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3584 if (max_progress > *suspend_point)
3585 max_progress = *suspend_point;
3586 }
3587 } else {
3588 if (info->array.major_version >= 0) {
3589 /* Only need to suspend when about to backup */
3590 if (info->reshape_progress < need_backup * 2 &&
3591 *suspend_point > 0) {
3592 *suspend_point = 0;
3593 sysfs_set_num(info, NULL, "suspend_lo", 0);
3594 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3595 }
3596 } else {
3597 /* Need to suspend continually */
3598 if (info->reshape_progress < *suspend_point)
3599 *suspend_point = info->reshape_progress;
3600 if (*suspend_point + target < info->reshape_progress)
3601 /* No need to move suspend region yet */;
3602 else {
3603 if (*suspend_point >= 2 * target)
3604 *suspend_point -= 2 * target;
3605 else
3606 *suspend_point = 0;
3607 sysfs_set_num(info, NULL, "suspend_lo",
3608 *suspend_point);
3609 }
3610 if (max_progress < *suspend_point)
3611 max_progress = *suspend_point;
3612 }
3613 }
3614
3615 /* now set sync_max to allow that progress. sync_max, like
3616 * sync_completed is a count of sectors written per device, so
3617 * we find the difference between max_progress and the start point,
3618 * and divide that by after.data_disks to get a sync_max
3619 * number.
3620 * At the same time we convert wait_point to a similar number
3621 * for comparing against sync_completed.
3622 */
3623 /* scale down max_progress to per_disk */
3624 max_progress /= reshape->after.data_disks;
3625 /* Round to chunk size as some kernels give an erroneously high number */
3626 max_progress /= info->new_chunk/512;
3627 max_progress *= info->new_chunk/512;
3628 /* And round to old chunk size as the kernel wants that */
3629 max_progress /= info->array.chunk_size/512;
3630 max_progress *= info->array.chunk_size/512;
3631 /* Limit progress to the whole device */
3632 if (max_progress > info->component_size)
3633 max_progress = info->component_size;
3634 wait_point /= reshape->after.data_disks;
3635 if (!advancing) {
3636 /* switch from 'device offset' to 'processed block count' */
3637 max_progress = info->component_size - max_progress;
3638 wait_point = info->component_size - wait_point;
3639 }
3640
3641 sysfs_set_num(info, NULL, "sync_max", max_progress);
3642
3643 /* Now wait. If we have already reached the point that we were
3644 * asked to wait to, don't wait at all, else wait for any change.
3645 * We need to select on 'sync_completed' as that is the place that
3646 * notifications happen, but we are really interested in
3647 * 'reshape_position'
3648 */
3649 fd = sysfs_get_fd(info, NULL, "sync_completed");
3650 if (fd < 0)
3651 goto check_progress;
3652
3653 if (sysfs_fd_get_ll(fd, &completed) < 0)
3654 goto check_progress;
3655
3656 while (completed < max_progress && completed < wait_point) {
3657 /* Check that sync_action is still 'reshape' to avoid
3658 * waiting forever on a dead array
3659 */
3660 char action[20];
3661 fd_set rfds;
3662 if (sysfs_get_str(info, NULL, "sync_action",
3663 action, 20) <= 0 ||
3664 strncmp(action, "reshape", 7) != 0)
3665 break;
3666 /* Some kernels reset 'sync_completed' to zero
3667 * before setting 'sync_action' to 'idle'.
3668 * So we need these extra tests.
3669 */
3670 if (completed == 0 && advancing
3671 && info->reshape_progress > 0)
3672 break;
3673 if (completed == 0 && !advancing
3674 && info->reshape_progress < (info->component_size
3675 * reshape->after.data_disks))
3676 break;
3677 FD_ZERO(&rfds);
3678 FD_SET(fd, &rfds);
3679 select(fd+1, NULL, NULL, &rfds, NULL);
3680 if (sysfs_fd_get_ll(fd, &completed) < 0)
3681 goto check_progress;
3682 }
3683 /* Some kernels reset 'sync_completed' to zero,
3684 * we need to have real point we are in md
3685 */
3686 if (completed == 0)
3687 completed = max_progress;
3688
3689 /* some kernels can give an incorrectly high 'completed' number */
3690 completed /= (info->new_chunk/512);
3691 completed *= (info->new_chunk/512);
3692 /* Convert 'completed' back in to a 'progress' number */
3693 completed *= reshape->after.data_disks;
3694 if (!advancing) {
3695 completed = info->component_size * reshape->after.data_disks
3696 - completed;
3697 }
3698 *reshape_completed = completed;
3699
3700 close(fd);
3701
3702 /* We return the need_backup flag. Caller will decide
3703 * how much - a multiple of ->backup_blocks up to *suspend_point
3704 */
3705 if (advancing)
3706 return need_backup > info->reshape_progress;
3707 else
3708 return need_backup >= info->reshape_progress;
3709
3710 check_progress:
3711 /* if we couldn't read a number from sync_completed, then
3712 * either the reshape did complete, or it aborted.
3713 * We can tell which by checking for 'none' in reshape_position.
3714 * If it did abort, then it might immediately restart if it
3715 * it was just a device failure that leaves us degraded but
3716 * functioning.
3717 */
3718 strcpy(buf, "hi");
3719 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3720 || strncmp(buf, "none", 4) != 0) {
3721 /* The abort might only be temporary. Wait up to 10
3722 * seconds for fd to contain a valid number again.
3723 */
3724 struct timeval tv;
3725 int rv = -2;
3726 tv.tv_sec = 10;
3727 tv.tv_usec = 0;
3728 while (fd >= 0 && rv < 0 && tv.tv_sec > 0) {
3729 fd_set rfds;
3730 FD_ZERO(&rfds);
3731 FD_SET(fd, &rfds);
3732 if (select(fd+1, NULL, NULL, &rfds, &tv) != 1)
3733 break;
3734 switch (sysfs_fd_get_ll(fd, &completed)) {
3735 case 0:
3736 /* all good again */
3737 rv = 1;
3738 break;
3739 case -2: /* read error - abort */
3740 tv.tv_sec = 0;
3741 break;
3742 }
3743 }
3744 if (fd >= 0)
3745 close(fd);
3746 return rv; /* abort */
3747 } else {
3748 /* Maybe racing with array shutdown - check state */
3749 if (fd >= 0)
3750 close(fd);
3751 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3752 || strncmp(buf, "inactive", 8) == 0
3753 || strncmp(buf, "clear",5) == 0)
3754 return -2; /* abort */
3755 return -1; /* complete */
3756 }
3757 }
3758
3759 /* FIXME return status is never checked */
3760 static int grow_backup(struct mdinfo *sra,
3761 unsigned long long offset, /* per device */
3762 unsigned long stripes, /* per device, in old chunks */
3763 int *sources, unsigned long long *offsets,
3764 int disks, int chunk, int level, int layout,
3765 int dests, int *destfd, unsigned long long *destoffsets,
3766 int part, int *degraded,
3767 char *buf)
3768 {
3769 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3770 * to storage 'destfd' (offset 'destoffsets'), after first
3771 * suspending IO. Then allow resync to continue
3772 * over the suspended section.
3773 * Use part 'part' of the backup-super-block.
3774 */
3775 int odata = disks;
3776 int rv = 0;
3777 int i;
3778 unsigned long long ll;
3779 int new_degraded;
3780 //printf("offset %llu\n", offset);
3781 if (level >= 4)
3782 odata--;
3783 if (level == 6)
3784 odata--;
3785
3786 /* Check that array hasn't become degraded, else we might backup the wrong data */
3787 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3788 return -1; /* FIXME this error is ignored */
3789 new_degraded = (int)ll;
3790 if (new_degraded != *degraded) {
3791 /* check each device to ensure it is still working */
3792 struct mdinfo *sd;
3793 for (sd = sra->devs ; sd ; sd = sd->next) {
3794 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3795 continue;
3796 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3797 char sbuf[20];
3798 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3799 strstr(sbuf, "faulty") ||
3800 strstr(sbuf, "in_sync") == NULL) {
3801 /* this device is dead */
3802 sd->disk.state = (1<<MD_DISK_FAULTY);
3803 if (sd->disk.raid_disk >= 0 &&
3804 sources[sd->disk.raid_disk] >= 0) {
3805 close(sources[sd->disk.raid_disk]);
3806 sources[sd->disk.raid_disk] = -1;
3807 }
3808 }
3809 }
3810 }
3811 *degraded = new_degraded;
3812 }
3813 if (part) {
3814 bsb.arraystart2 = __cpu_to_le64(offset * odata);
3815 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
3816 } else {
3817 bsb.arraystart = __cpu_to_le64(offset * odata);
3818 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
3819 }
3820 if (part)
3821 bsb.magic[15] = '2';
3822 for (i = 0; i < dests; i++)
3823 if (part)
3824 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
3825 else
3826 lseek64(destfd[i], destoffsets[i], 0);
3827
3828 rv = save_stripes(sources, offsets,
3829 disks, chunk, level, layout,
3830 dests, destfd,
3831 offset*512*odata, stripes * chunk * odata,
3832 buf);
3833
3834 if (rv)
3835 return rv;
3836 bsb.mtime = __cpu_to_le64(time(0));
3837 for (i = 0; i < dests; i++) {
3838 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3839
3840 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3841 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3842 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3843 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3844
3845 rv = -1;
3846 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
3847 != destoffsets[i] - 4096)
3848 break;
3849 if (write(destfd[i], &bsb, 512) != 512)
3850 break;
3851 if (destoffsets[i] > 4096) {
3852 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
3853 destoffsets[i]+stripes*chunk*odata)
3854 break;
3855 if (write(destfd[i], &bsb, 512) != 512)
3856 break;
3857 }
3858 fsync(destfd[i]);
3859 rv = 0;
3860 }
3861
3862 return rv;
3863 }
3864
3865 /* in 2.6.30, the value reported by sync_completed can be
3866 * less that it should be by one stripe.
3867 * This only happens when reshape hits sync_max and pauses.
3868 * So allow wait_backup to either extent sync_max further
3869 * than strictly necessary, or return before the
3870 * sync has got quite as far as we would really like.
3871 * This is what 'blocks2' is for.
3872 * The various caller give appropriate values so that
3873 * every works.
3874 */
3875 /* FIXME return value is often ignored */
3876 static int forget_backup(int dests, int *destfd,
3877 unsigned long long *destoffsets,
3878 int part)
3879 {
3880 /*
3881 * Erase backup 'part' (which is 0 or 1)
3882 */
3883 int i;
3884 int rv;
3885
3886 if (part) {
3887 bsb.arraystart2 = __cpu_to_le64(0);
3888 bsb.length2 = __cpu_to_le64(0);
3889 } else {
3890 bsb.arraystart = __cpu_to_le64(0);
3891 bsb.length = __cpu_to_le64(0);
3892 }
3893 bsb.mtime = __cpu_to_le64(time(0));
3894 rv = 0;
3895 for (i = 0; i < dests; i++) {
3896 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3897 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3898 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3899 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3900 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3901 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
3902 destoffsets[i]-4096)
3903 rv = -1;
3904 if (rv == 0 &&
3905 write(destfd[i], &bsb, 512) != 512)
3906 rv = -1;
3907 fsync(destfd[i]);
3908 }
3909 return rv;
3910 }
3911
3912 static void fail(char *msg)
3913 {
3914 int rv;
3915 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
3916 rv |= (write(2, "\n", 1) != 1);
3917 exit(rv ? 1 : 2);
3918 }
3919
3920 static char *abuf, *bbuf;
3921 static unsigned long long abuflen;
3922 static void validate(int afd, int bfd, unsigned long long offset)
3923 {
3924 /* check that the data in the backup against the array.
3925 * This is only used for regression testing and should not
3926 * be used while the array is active
3927 */
3928 if (afd < 0)
3929 return;
3930 lseek64(bfd, offset - 4096, 0);
3931 if (read(bfd, &bsb2, 512) != 512)
3932 fail("cannot read bsb");
3933 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
3934 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
3935 fail("first csum bad");
3936 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
3937 fail("magic is bad");
3938 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
3939 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
3940 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
3941 fail("second csum bad");
3942
3943 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
3944 fail("devstart is wrong");
3945
3946 if (bsb2.length) {
3947 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
3948
3949 if (abuflen < len) {
3950 free(abuf);
3951 free(bbuf);
3952 abuflen = len;
3953 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
3954 posix_memalign((void**)&bbuf, 4096, abuflen)) {
3955 abuflen = 0;
3956 /* just stop validating on mem-alloc failure */
3957 return;
3958 }
3959 }
3960
3961 lseek64(bfd, offset, 0);
3962 if ((unsigned long long)read(bfd, bbuf, len) != len) {
3963 //printf("len %llu\n", len);
3964 fail("read first backup failed");
3965 }
3966 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
3967 if ((unsigned long long)read(afd, abuf, len) != len)
3968 fail("read first from array failed");
3969 if (memcmp(bbuf, abuf, len) != 0) {
3970 #if 0
3971 int i;
3972 printf("offset=%llu len=%llu\n",
3973 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
3974 for (i=0; i<len; i++)
3975 if (bbuf[i] != abuf[i]) {
3976 printf("first diff byte %d\n", i);
3977 break;
3978 }
3979 #endif
3980 fail("data1 compare failed");
3981 }
3982 }
3983 if (bsb2.length2) {
3984 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
3985
3986 if (abuflen < len) {
3987 free(abuf);
3988 free(bbuf);
3989 abuflen = len;
3990 abuf = xmalloc(abuflen);
3991 bbuf = xmalloc(abuflen);
3992 }
3993
3994 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
3995 if ((unsigned long long)read(bfd, bbuf, len) != len)
3996 fail("read second backup failed");
3997 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
3998 if ((unsigned long long)read(afd, abuf, len) != len)
3999 fail("read second from array failed");
4000 if (memcmp(bbuf, abuf, len) != 0)
4001 fail("data2 compare failed");
4002 }
4003 }
4004
4005 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4006 struct supertype *st, unsigned long blocks,
4007 int *fds, unsigned long long *offsets,
4008 int dests, int *destfd, unsigned long long *destoffsets)
4009 {
4010 /* Monitor a reshape where backup is being performed using
4011 * 'native' mechanism - either to a backup file, or
4012 * to some space in a spare.
4013 */
4014 char *buf;
4015 int degraded = -1;
4016 unsigned long long speed;
4017 unsigned long long suspend_point, array_size;
4018 unsigned long long backup_point, wait_point;
4019 unsigned long long reshape_completed;
4020 int done = 0;
4021 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4022 int part = 0; /* The next part of the backup area to fill. It may already
4023 * be full, so we need to check */
4024 int level = reshape->level;
4025 int layout = reshape->before.layout;
4026 int data = reshape->before.data_disks;
4027 int disks = reshape->before.data_disks + reshape->parity;
4028 int chunk = sra->array.chunk_size;
4029 struct mdinfo *sd;
4030 unsigned long stripes;
4031 int uuid[4];
4032
4033 /* set up the backup-super-block. This requires the
4034 * uuid from the array.
4035 */
4036 /* Find a superblock */
4037 for (sd = sra->devs; sd; sd = sd->next) {
4038 char *dn;
4039 int devfd;
4040 int ok;
4041 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4042 continue;
4043 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4044 devfd = dev_open(dn, O_RDONLY);
4045 if (devfd < 0)
4046 continue;
4047 ok = st->ss->load_super(st, devfd, NULL);
4048 close(devfd);
4049 if (ok == 0)
4050 break;
4051 }
4052 if (!sd) {
4053 pr_err("Cannot find a superblock\n");
4054 return 0;
4055 }
4056
4057 memset(&bsb, 0, 512);
4058 memcpy(bsb.magic, "md_backup_data-1", 16);
4059 st->ss->uuid_from_super(st, uuid);
4060 memcpy(bsb.set_uuid, uuid, 16);
4061 bsb.mtime = __cpu_to_le64(time(0));
4062 bsb.devstart2 = blocks;
4063
4064 stripes = blocks / (sra->array.chunk_size/512) /
4065 reshape->before.data_disks;
4066
4067 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4068 /* Don't start the 'reshape' */
4069 return 0;
4070 if (reshape->before.data_disks == reshape->after.data_disks) {
4071 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4072 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4073 }
4074
4075 if (increasing) {
4076 array_size = sra->component_size * reshape->after.data_disks;
4077 backup_point = sra->reshape_progress;
4078 suspend_point = 0;
4079 } else {
4080 array_size = sra->component_size * reshape->before.data_disks;
4081 backup_point = reshape->backup_blocks;
4082 suspend_point = array_size;
4083 }
4084
4085 while (!done) {
4086 int rv;
4087
4088 /* Want to return as soon the oldest backup slot can
4089 * be released as that allows us to start backing up
4090 * some more, providing suspend_point has been
4091 * advanced, which it should have.
4092 */
4093 if (increasing) {
4094 wait_point = array_size;
4095 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4096 wait_point = (__le64_to_cpu(bsb.arraystart) +
4097 __le64_to_cpu(bsb.length));
4098 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4099 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4100 __le64_to_cpu(bsb.length2));
4101 } else {
4102 wait_point = 0;
4103 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4104 wait_point = __le64_to_cpu(bsb.arraystart);
4105 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4106 wait_point = __le64_to_cpu(bsb.arraystart2);
4107 }
4108
4109 rv = progress_reshape(sra, reshape,
4110 backup_point, wait_point,
4111 &suspend_point, &reshape_completed);
4112 /* external metadata would need to ping_monitor here */
4113 sra->reshape_progress = reshape_completed;
4114
4115 /* Clear any backup region that is before 'here' */
4116 if (increasing) {
4117 if (__le64_to_cpu(bsb.length) > 0 &&
4118 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4119 __le64_to_cpu(bsb.length)))
4120 forget_backup(dests, destfd,
4121 destoffsets, 0);
4122 if (__le64_to_cpu(bsb.length2) > 0 &&
4123 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4124 __le64_to_cpu(bsb.length2)))
4125 forget_backup(dests, destfd,
4126 destoffsets, 1);
4127 } else {
4128 if (__le64_to_cpu(bsb.length) > 0 &&
4129 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4130 forget_backup(dests, destfd,
4131 destoffsets, 0);
4132 if (__le64_to_cpu(bsb.length2) > 0 &&
4133 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4134 forget_backup(dests, destfd,
4135 destoffsets, 1);
4136 }
4137
4138 if (rv < 0) {
4139 if (rv == -1)
4140 done = 1;
4141 break;
4142 }
4143 if (rv == 0 && increasing && !st->ss->external) {
4144 /* No longer need to monitor this reshape */
4145 done = 1;
4146 break;
4147 }
4148
4149 while (rv) {
4150 unsigned long long offset;
4151 unsigned long actual_stripes;
4152 /* Need to backup some data.
4153 * If 'part' is not used and the desired
4154 * backup size is suspended, do a backup,
4155 * then consider the next part.
4156 */
4157 /* Check that 'part' is unused */
4158 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4159 break;
4160 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4161 break;
4162
4163 offset = backup_point / data;
4164 actual_stripes = stripes;
4165 if (increasing) {
4166 if (offset + actual_stripes * (chunk/512) >
4167 sra->component_size)
4168 actual_stripes = ((sra->component_size - offset)
4169 / (chunk/512));
4170 if (offset + actual_stripes * (chunk/512) >
4171 suspend_point/data)
4172 break;
4173 } else {
4174 if (offset < actual_stripes * (chunk/512))
4175 actual_stripes = offset / (chunk/512);
4176 offset -= actual_stripes * (chunk/512);
4177 if (offset < suspend_point/data)
4178 break;
4179 }
4180 if (actual_stripes == 0)
4181 break;
4182 grow_backup(sra, offset, actual_stripes,
4183 fds, offsets,
4184 disks, chunk, level, layout,
4185 dests, destfd, destoffsets,
4186 part, &degraded, buf);
4187 validate(afd, destfd[0], destoffsets[0]);
4188 /* record where 'part' is up to */
4189 part = !part;
4190 if (increasing)
4191 backup_point += actual_stripes * (chunk/512) * data;
4192 else
4193 backup_point -= actual_stripes * (chunk/512) * data;
4194 }
4195 }
4196
4197 /* FIXME maybe call progress_reshape one more time instead */
4198 abort_reshape(sra); /* remove any remaining suspension */
4199 if (reshape->before.data_disks == reshape->after.data_disks)
4200 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4201 free(buf);
4202 return done;
4203 }
4204
4205 /*
4206 * If any spare contains md_back_data-1 which is recent wrt mtime,
4207 * write that data into the array and update the super blocks with
4208 * the new reshape_progress
4209 */
4210 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4211 char *backup_file, int verbose)
4212 {
4213 int i, j;
4214 int old_disks;
4215 unsigned long long *offsets;
4216 unsigned long long nstripe, ostripe;
4217 int ndata, odata;
4218
4219 odata = info->array.raid_disks - info->delta_disks - 1;
4220 if (info->array.level == 6) odata--; /* number of data disks */
4221 ndata = info->array.raid_disks - 1;
4222 if (info->new_level == 6) ndata--;
4223
4224 old_disks = info->array.raid_disks - info->delta_disks;
4225
4226 if (info->delta_disks <= 0)
4227 /* Didn't grow, so the backup file must have
4228 * been used
4229 */
4230 old_disks = cnt;
4231 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4232 struct mdinfo dinfo;
4233 int fd;
4234 int bsbsize;
4235 char *devname, namebuf[20];
4236 unsigned long long lo, hi;
4237
4238 /* This was a spare and may have some saved data on it.
4239 * Load the superblock, find and load the
4240 * backup_super_block.
4241 * If either fail, go on to next device.
4242 * If the backup contains no new info, just return
4243 * else restore data and update all superblocks
4244 */
4245 if (i == old_disks-1) {
4246 fd = open(backup_file, O_RDONLY);
4247 if (fd<0) {
4248 pr_err("backup file %s inaccessible: %s\n",
4249 backup_file, strerror(errno));
4250 continue;
4251 }
4252 devname = backup_file;
4253 } else {
4254 fd = fdlist[i];
4255 if (fd < 0)
4256 continue;
4257 if (st->ss->load_super(st, fd, NULL))
4258 continue;
4259
4260 st->ss->getinfo_super(st, &dinfo, NULL);
4261 st->ss->free_super(st);
4262
4263 if (lseek64(fd,
4264 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4265 0) < 0) {
4266 pr_err("Cannot seek on device %d\n", i);
4267 continue; /* Cannot seek */
4268 }
4269 sprintf(namebuf, "device-%d", i);
4270 devname = namebuf;
4271 }
4272 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4273 if (verbose)
4274 pr_err("Cannot read from %s\n", devname);
4275 continue; /* Cannot read */
4276 }
4277 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4278 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4279 if (verbose)
4280 pr_err("No backup metadata on %s\n", devname);
4281 continue;
4282 }
4283 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4284 if (verbose)
4285 pr_err("Bad backup-metadata checksum on %s\n", devname);
4286 continue; /* bad checksum */
4287 }
4288 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4289 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4290 if (verbose)
4291 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4292 continue; /* Bad second checksum */
4293 }
4294 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4295 if (verbose)
4296 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4297 continue; /* Wrong uuid */
4298 }
4299
4300 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4301 * sometimes they aren't... So allow considerable flexability in matching, and allow
4302 * this test to be overridden by an environment variable.
4303 */
4304 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4305 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4306 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4307 pr_err("accepting backup with timestamp %lu "
4308 "for array with timestamp %lu\n",
4309 (unsigned long)__le64_to_cpu(bsb.mtime),
4310 (unsigned long)info->array.utime);
4311 } else {
4312 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4313 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4314 continue; /* time stamp is too bad */
4315 }
4316 }
4317
4318 if (bsb.magic[15] == '1') {
4319 if (bsb.length == 0)
4320 continue;
4321 if (info->delta_disks >= 0) {
4322 /* reshape_progress is increasing */
4323 if (__le64_to_cpu(bsb.arraystart)
4324 + __le64_to_cpu(bsb.length)
4325 < info->reshape_progress) {
4326 nonew:
4327 if (verbose)
4328 pr_err("backup-metadata found on %s but is not needed\n", devname);
4329 continue; /* No new data here */
4330 }
4331 } else {
4332 /* reshape_progress is decreasing */
4333 if (__le64_to_cpu(bsb.arraystart) >=
4334 info->reshape_progress)
4335 goto nonew; /* No new data here */
4336 }
4337 } else {
4338 if (bsb.length == 0 && bsb.length2 == 0)
4339 continue;
4340 if (info->delta_disks >= 0) {
4341 /* reshape_progress is increasing */
4342 if ((__le64_to_cpu(bsb.arraystart)
4343 + __le64_to_cpu(bsb.length)
4344 < info->reshape_progress)
4345 &&
4346 (__le64_to_cpu(bsb.arraystart2)
4347 + __le64_to_cpu(bsb.length2)
4348 < info->reshape_progress))
4349 goto nonew; /* No new data here */
4350 } else {
4351 /* reshape_progress is decreasing */
4352 if (__le64_to_cpu(bsb.arraystart) >=
4353 info->reshape_progress &&
4354 __le64_to_cpu(bsb.arraystart2) >=
4355 info->reshape_progress)
4356 goto nonew; /* No new data here */
4357 }
4358 }
4359 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4360 second_fail:
4361 if (verbose)
4362 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4363 devname);
4364 continue; /* Cannot seek */
4365 }
4366 /* There should be a duplicate backup superblock 4k before here */
4367 if (lseek64(fd, -4096, 1) < 0 ||
4368 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4369 goto second_fail; /* Cannot find leading superblock */
4370 if (bsb.magic[15] == '1')
4371 bsbsize = offsetof(struct mdp_backup_super, pad1);
4372 else
4373 bsbsize = offsetof(struct mdp_backup_super, pad);
4374 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4375 goto second_fail; /* Cannot find leading superblock */
4376
4377 /* Now need the data offsets for all devices. */
4378 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4379 for(j=0; j<info->array.raid_disks; j++) {
4380 if (fdlist[j] < 0)
4381 continue;
4382 if (st->ss->load_super(st, fdlist[j], NULL))
4383 /* FIXME should be this be an error */
4384 continue;
4385 st->ss->getinfo_super(st, &dinfo, NULL);
4386 st->ss->free_super(st);
4387 offsets[j] = dinfo.data_offset * 512;
4388 }
4389 printf(Name ": restoring critical section\n");
4390
4391 if (restore_stripes(fdlist, offsets,
4392 info->array.raid_disks,
4393 info->new_chunk,
4394 info->new_level,
4395 info->new_layout,
4396 fd, __le64_to_cpu(bsb.devstart)*512,
4397 __le64_to_cpu(bsb.arraystart)*512,
4398 __le64_to_cpu(bsb.length)*512, NULL)) {
4399 /* didn't succeed, so giveup */
4400 if (verbose)
4401 pr_err("Error restoring backup from %s\n",
4402 devname);
4403 free(offsets);
4404 return 1;
4405 }
4406
4407 if (bsb.magic[15] == '2' &&
4408 restore_stripes(fdlist, offsets,
4409 info->array.raid_disks,
4410 info->new_chunk,
4411 info->new_level,
4412 info->new_layout,
4413 fd, __le64_to_cpu(bsb.devstart)*512 +
4414 __le64_to_cpu(bsb.devstart2)*512,
4415 __le64_to_cpu(bsb.arraystart2)*512,
4416 __le64_to_cpu(bsb.length2)*512, NULL)) {
4417 /* didn't succeed, so giveup */
4418 if (verbose)
4419 pr_err("Error restoring second backup from %s\n",
4420 devname);
4421 free(offsets);
4422 return 1;
4423 }
4424
4425 free(offsets);
4426
4427 /* Ok, so the data is restored. Let's update those superblocks. */
4428
4429 lo = hi = 0;
4430 if (bsb.length) {
4431 lo = __le64_to_cpu(bsb.arraystart);
4432 hi = lo + __le64_to_cpu(bsb.length);
4433 }
4434 if (bsb.magic[15] == '2' && bsb.length2) {
4435 unsigned long long lo1, hi1;
4436 lo1 = __le64_to_cpu(bsb.arraystart2);
4437 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4438 if (lo == hi) {
4439 lo = lo1;
4440 hi = hi1;
4441 } else if (lo < lo1)
4442 hi = hi1;
4443 else
4444 lo = lo1;
4445 }
4446 if (lo < hi &&
4447 (info->reshape_progress < lo ||
4448 info->reshape_progress > hi))
4449 /* backup does not affect reshape_progress*/ ;
4450 else if (info->delta_disks >= 0) {
4451 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4452 __le64_to_cpu(bsb.length);
4453 if (bsb.magic[15] == '2') {
4454 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4455 __le64_to_cpu(bsb.length2);
4456 if (p2 > info->reshape_progress)
4457 info->reshape_progress = p2;
4458 }
4459 } else {
4460 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4461 if (bsb.magic[15] == '2') {
4462 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4463 if (p2 < info->reshape_progress)
4464 info->reshape_progress = p2;
4465 }
4466 }
4467 for (j=0; j<info->array.raid_disks; j++) {
4468 if (fdlist[j] < 0)
4469 continue;
4470 if (st->ss->load_super(st, fdlist[j], NULL))
4471 continue;
4472 st->ss->getinfo_super(st, &dinfo, NULL);
4473 dinfo.reshape_progress = info->reshape_progress;
4474 st->ss->update_super(st, &dinfo,
4475 "_reshape_progress",
4476 NULL,0, 0, NULL);
4477 st->ss->store_super(st, fdlist[j]);
4478 st->ss->free_super(st);
4479 }
4480 return 0;
4481 }
4482 /* Didn't find any backup data, try to see if any
4483 * was needed.
4484 */
4485 if (info->delta_disks < 0) {
4486 /* When shrinking, the critical section is at the end.
4487 * So see if we are before the critical section.
4488 */
4489 unsigned long long first_block;
4490 nstripe = ostripe = 0;
4491 first_block = 0;
4492 while (ostripe >= nstripe) {
4493 ostripe += info->array.chunk_size / 512;
4494 first_block = ostripe * odata;
4495 nstripe = first_block / ndata / (info->new_chunk/512) *
4496 (info->new_chunk/512);
4497 }
4498
4499 if (info->reshape_progress >= first_block)
4500 return 0;
4501 }
4502 if (info->delta_disks > 0) {
4503 /* See if we are beyond the critical section. */
4504 unsigned long long last_block;
4505 nstripe = ostripe = 0;
4506 last_block = 0;
4507 while (nstripe >= ostripe) {
4508 nstripe += info->new_chunk / 512;
4509 last_block = nstripe * ndata;
4510 ostripe = last_block / odata / (info->array.chunk_size/512) *
4511 (info->array.chunk_size/512);
4512 }
4513
4514 if (info->reshape_progress >= last_block)
4515 return 0;
4516 }
4517 /* needed to recover critical section! */
4518 if (verbose)
4519 pr_err("Failed to find backup of critical section\n");
4520 return 1;
4521 }
4522
4523 int Grow_continue_command(char *devname, int fd,
4524 char *backup_file, int verbose)
4525 {
4526 int ret_val = 0;
4527 struct supertype *st = NULL;
4528 struct mdinfo *content = NULL;
4529 struct mdinfo array;
4530 char *subarray = NULL;
4531 struct mdinfo *cc = NULL;
4532 struct mdstat_ent *mdstat = NULL;
4533 int cfd = -1;
4534 int fd2 = -1;
4535
4536 dprintf("Grow continue from command line called for %s\n",
4537 devname);
4538
4539 st = super_by_fd(fd, &subarray);
4540 if (!st || !st->ss) {
4541 pr_err("Unable to determine metadata format for %s\n",
4542 devname);
4543 return 1;
4544 }
4545 dprintf("Grow continue is run for ");
4546 if (st->ss->external == 0) {
4547 int d;
4548 dprintf("native array (%s)\n", devname);
4549 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4550 pr_err("%s is not an active md array -"
4551 " aborting\n", devname);
4552 ret_val = 1;
4553 goto Grow_continue_command_exit;
4554 }
4555 content = &array;
4556 /* Need to load a superblock.
4557 * FIXME we should really get what we need from
4558 * sysfs
4559 */
4560 for (d = 0; d < MAX_DISKS; d++) {
4561 mdu_disk_info_t disk;
4562 char *dv;
4563 int err;
4564 disk.number = d;
4565 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4566 continue;
4567 if (disk.major == 0 && disk.minor == 0)
4568 continue;
4569 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4570 continue;
4571 dv = map_dev(disk.major, disk.minor, 1);
4572 if (!dv)
4573 continue;
4574 fd2 = dev_open(dv, O_RDONLY);
4575 if (fd2 < 0)
4576 continue;
4577 err = st->ss->load_super(st, fd2, NULL);
4578 close(fd2);
4579 if (err)
4580 continue;
4581 break;
4582 }
4583 if (d == MAX_DISKS) {
4584 pr_err("Unable to load metadata for %s\n",
4585 devname);
4586 ret_val = 1;
4587 goto Grow_continue_command_exit;
4588 }
4589 st->ss->getinfo_super(st, content, NULL);
4590 } else {
4591 char *container;
4592
4593 if (subarray) {
4594 dprintf("subarray (%s)\n", subarray);
4595 container = st->container_devnm;
4596 cfd = open_dev_excl(st->container_devnm);
4597 } else {
4598 container = st->devnm;
4599 close(fd);
4600 cfd = open_dev_excl(st->devnm);
4601 dprintf("container (%s)\n", container);
4602 fd = cfd;
4603 }
4604 if (cfd < 0) {
4605 pr_err("Unable to open container "
4606 "for %s\n", devname);
4607 ret_val = 1;
4608 goto Grow_continue_command_exit;
4609 }
4610
4611 /* find in container array under reshape
4612 */
4613 ret_val = st->ss->load_container(st, cfd, NULL);
4614 if (ret_val) {
4615 pr_err("Cannot read superblock for %s\n",
4616 devname);
4617 ret_val = 1;
4618 goto Grow_continue_command_exit;
4619 }
4620
4621 cc = st->ss->container_content(st, subarray);
4622 for (content = cc; content ; content = content->next) {
4623 char *array;
4624 int allow_reshape = 1;
4625
4626 if (content->reshape_active == 0)
4627 continue;
4628 /* The decision about array or container wide
4629 * reshape is taken in Grow_continue based
4630 * content->reshape_active state, therefore we
4631 * need to check_reshape based on
4632 * reshape_active and subarray name
4633 */
4634 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4635 allow_reshape = 0;
4636 if (content->reshape_active == CONTAINER_RESHAPE &&
4637 (content->array.state
4638 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4639 allow_reshape = 0;
4640
4641 if (!allow_reshape) {
4642 pr_err("cannot continue reshape of an array"
4643 " in container with unsupported"
4644 " metadata: %s(%s)\n",
4645 devname, container);
4646 ret_val = 1;
4647 goto Grow_continue_command_exit;
4648 }
4649
4650 array = strchr(content->text_version+1, '/')+1;
4651 mdstat = mdstat_by_subdev(array, container);
4652 if (!mdstat)
4653 continue;
4654 if (mdstat->active == 0) {
4655 pr_err("Skipping inactive array %s.\n",
4656 mdstat->devnm);
4657 free_mdstat(mdstat);
4658 mdstat = NULL;
4659 continue;
4660 }
4661 break;
4662 }
4663 if (!content) {
4664 pr_err("Unable to determine reshaped "
4665 "array for %s\n", devname);
4666 ret_val = 1;
4667 goto Grow_continue_command_exit;
4668 }
4669 fd2 = open_dev(mdstat->devnm);
4670 if (fd2 < 0) {
4671 pr_err("cannot open (%s)\n", mdstat->devnm);
4672 ret_val = 1;
4673 goto Grow_continue_command_exit;
4674 }
4675
4676 sysfs_init(content, fd2, mdstat->devnm);
4677
4678 /* start mdmon in case it is not running
4679 */
4680 if (!mdmon_running(container))
4681 start_mdmon(container);
4682 ping_monitor(container);
4683
4684 if (mdmon_running(container))
4685 st->update_tail = &st->updates;
4686 else {
4687 pr_err("No mdmon found. "
4688 "Grow cannot continue.\n");
4689 ret_val = 1;
4690 goto Grow_continue_command_exit;
4691 }
4692 }
4693
4694 /* verify that array under reshape is started from
4695 * correct position
4696 */
4697 if (verify_reshape_position(content, content->array.level) < 0) {
4698 ret_val = 1;
4699 goto Grow_continue_command_exit;
4700 }
4701
4702 /* continue reshape
4703 */
4704 ret_val = Grow_continue(fd, st, content, backup_file, 0);
4705
4706 Grow_continue_command_exit:
4707 if (fd2 > -1)
4708 close(fd2);
4709 if (cfd > -1)
4710 close(cfd);
4711 st->ss->free_super(st);
4712 free_mdstat(mdstat);
4713 sysfs_free(cc);
4714 free(subarray);
4715
4716 return ret_val;
4717 }
4718
4719 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4720 char *backup_file, int freeze_reshape)
4721 {
4722 int ret_val = 2;
4723
4724 if (!info->reshape_active)
4725 return ret_val;
4726
4727 if (st->ss->external) {
4728 int cfd = open_dev(st->container_devnm);
4729
4730 if (cfd < 0)
4731 return 1;
4732
4733 st->ss->load_container(st, cfd, st->container_devnm);
4734 close(cfd);
4735 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4736 st, info, 0, backup_file,
4737 0,
4738 1 | info->reshape_active,
4739 freeze_reshape);
4740 } else
4741 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4742 NULL, 0ULL, backup_file, 0, 0,
4743 1 | info->reshape_active,
4744 freeze_reshape);
4745
4746 return ret_val;
4747 }
Unnamed repository; edit this file 'description' to name the repository.