]> 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
Minor cosmetic fixes in various files.
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2012 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27
28 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
29 #error no endian defined
30 #endif
31 #include "md_u.h"
32 #include "md_p.h"
33
34 #ifndef offsetof
35 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
36 #endif
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char *backup_file,
43 int verbose)
44 {
45 int i;
46 int *fdlist;
47 struct mdinfo *dev;
48 int err;
49 int disk_count = next_spare + working_disks;
50
51 dprintf("Called restore_backup()\n");
52 fdlist = xmalloc(sizeof(int) * disk_count);
53
54 for (i = 0; i < next_spare; i++)
55 fdlist[i] = -1;
56 for (dev = content->devs; dev; dev = dev->next) {
57 char buf[22];
58 int fd;
59 sprintf(buf, "%d:%d",
60 dev->disk.major,
61 dev->disk.minor);
62 fd = dev_open(buf, O_RDWR);
63
64 if (dev->disk.raid_disk >= 0)
65 fdlist[dev->disk.raid_disk] = fd;
66 else
67 fdlist[next_spare++] = fd;
68 }
69
70 if (st->ss->external && st->ss->recover_backup)
71 err = st->ss->recover_backup(st, content);
72 else
73 err = Grow_restart(st, content, fdlist, next_spare,
74 backup_file, verbose > 0);
75
76 while (next_spare > 0) {
77 next_spare--;
78 if (fdlist[next_spare] >= 0)
79 close(fdlist[next_spare]);
80 }
81 free(fdlist);
82 if (err) {
83 pr_err("Failed to restore critical"
84 " section for reshape - sorry.\n");
85 if (!backup_file)
86 pr_err("Possibly you need"
87 " to specify a --backup-file\n");
88 return 1;
89 }
90
91 dprintf("restore_backup() returns status OK.\n");
92 return 0;
93 }
94
95 int Grow_Add_device(char *devname, int fd, char *newdev)
96 {
97 /* Add a device to an active array.
98 * Currently, just extend a linear array.
99 * This requires writing a new superblock on the
100 * new device, calling the kernel to add the device,
101 * and if that succeeds, update the superblock on
102 * all other devices.
103 * This means that we need to *find* all other devices.
104 */
105 struct mdinfo info;
106
107 struct stat stb;
108 int nfd, fd2;
109 int d, nd;
110 struct supertype *st = NULL;
111 char *subarray = NULL;
112
113 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
114 pr_err("cannot get array info for %s\n", devname);
115 return 1;
116 }
117
118 if (info.array.level != -1) {
119 pr_err("can only add devices to linear arrays\n");
120 return 1;
121 }
122
123 st = super_by_fd(fd, &subarray);
124 if (!st) {
125 pr_err("cannot handle arrays with superblock version %d\n",
126 info.array.major_version);
127 return 1;
128 }
129
130 if (subarray) {
131 pr_err("Cannot grow linear sub-arrays yet\n");
132 free(subarray);
133 free(st);
134 return 1;
135 }
136
137 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
138 if (nfd < 0) {
139 pr_err("cannot open %s\n", newdev);
140 free(st);
141 return 1;
142 }
143 fstat(nfd, &stb);
144 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
145 pr_err("%s is not a block device!\n", newdev);
146 close(nfd);
147 free(st);
148 return 1;
149 }
150 /* now check out all the devices and make sure we can read the
151 * superblock */
152 for (d=0 ; d < info.array.raid_disks ; d++) {
153 mdu_disk_info_t disk;
154 char *dv;
155
156 st->ss->free_super(st);
157
158 disk.number = d;
159 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
160 pr_err("cannot get device detail for device %d\n",
161 d);
162 close(nfd);
163 free(st);
164 return 1;
165 }
166 dv = map_dev(disk.major, disk.minor, 1);
167 if (!dv) {
168 pr_err("cannot find device file for device %d\n",
169 d);
170 close(nfd);
171 free(st);
172 return 1;
173 }
174 fd2 = dev_open(dv, O_RDWR);
175 if (fd2 < 0) {
176 pr_err("cannot open device file %s\n", dv);
177 close(nfd);
178 free(st);
179 return 1;
180 }
181
182 if (st->ss->load_super(st, fd2, NULL)) {
183 pr_err("cannot find super block on %s\n", dv);
184 close(nfd);
185 close(fd2);
186 free(st);
187 return 1;
188 }
189 close(fd2);
190 }
191 /* Ok, looks good. Lets update the superblock and write it out to
192 * newdev.
193 */
194
195 info.disk.number = d;
196 info.disk.major = major(stb.st_rdev);
197 info.disk.minor = minor(stb.st_rdev);
198 info.disk.raid_disk = d;
199 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
200 st->ss->update_super(st, &info, "linear-grow-new", newdev,
201 0, 0, NULL);
202
203 if (st->ss->store_super(st, nfd)) {
204 pr_err("Cannot store new superblock on %s\n",
205 newdev);
206 close(nfd);
207 return 1;
208 }
209 close(nfd);
210
211 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
212 pr_err("Cannot add new disk to this array\n");
213 return 1;
214 }
215 /* Well, that seems to have worked.
216 * Now go through and update all superblocks
217 */
218
219 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
220 pr_err("cannot get array info for %s\n", devname);
221 return 1;
222 }
223
224 nd = d;
225 for (d=0 ; d < info.array.raid_disks ; d++) {
226 mdu_disk_info_t disk;
227 char *dv;
228
229 disk.number = d;
230 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
231 pr_err("cannot get device detail for device %d\n",
232 d);
233 return 1;
234 }
235 dv = map_dev(disk.major, disk.minor, 1);
236 if (!dv) {
237 pr_err("cannot find device file for device %d\n",
238 d);
239 return 1;
240 }
241 fd2 = dev_open(dv, O_RDWR);
242 if (fd2 < 0) {
243 pr_err("cannot open device file %s\n", dv);
244 return 1;
245 }
246 if (st->ss->load_super(st, fd2, NULL)) {
247 pr_err("cannot find super block on %s\n", dv);
248 close(fd);
249 return 1;
250 }
251 info.array.raid_disks = nd+1;
252 info.array.nr_disks = nd+1;
253 info.array.active_disks = nd+1;
254 info.array.working_disks = nd+1;
255
256 st->ss->update_super(st, &info, "linear-grow-update", dv,
257 0, 0, NULL);
258
259 if (st->ss->store_super(st, fd2)) {
260 pr_err("Cannot store new superblock on %s\n", dv);
261 close(fd2);
262 return 1;
263 }
264 close(fd2);
265 }
266
267 return 0;
268 }
269
270 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
271 {
272 /*
273 * First check that array doesn't have a bitmap
274 * Then create the bitmap
275 * Then add it
276 *
277 * For internal bitmaps, we need to check the version,
278 * find all the active devices, and write the bitmap block
279 * to all devices
280 */
281 mdu_bitmap_file_t bmf;
282 mdu_array_info_t array;
283 struct supertype *st;
284 char *subarray = NULL;
285 int major = BITMAP_MAJOR_HI;
286 int vers = md_get_version(fd);
287 unsigned long long bitmapsize, array_size;
288
289 if (vers < 9003) {
290 major = BITMAP_MAJOR_HOSTENDIAN;
291 pr_err("Warning - bitmaps created on this kernel"
292 " are not portable\n"
293 " between different architectures. Consider upgrading"
294 " the Linux kernel.\n");
295 }
296
297 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
298 if (errno == ENOMEM)
299 pr_err("Memory allocation failure.\n");
300 else
301 pr_err("bitmaps not supported by this kernel.\n");
302 return 1;
303 }
304 if (bmf.pathname[0]) {
305 if (strcmp(s->bitmap_file,"none")==0) {
306 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
307 pr_err("failed to remove bitmap %s\n",
308 bmf.pathname);
309 return 1;
310 }
311 return 0;
312 }
313 pr_err("%s already has a bitmap (%s)\n",
314 devname, bmf.pathname);
315 return 1;
316 }
317 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
318 pr_err("cannot get array status for %s\n", devname);
319 return 1;
320 }
321 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
322 if (strcmp(s->bitmap_file, "none")==0) {
323 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
324 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
325 pr_err("failed to remove internal bitmap.\n");
326 return 1;
327 }
328 return 0;
329 }
330 pr_err("Internal bitmap already present on %s\n",
331 devname);
332 return 1;
333 }
334
335 if (strcmp(s->bitmap_file, "none") == 0) {
336 pr_err("no bitmap found on %s\n", devname);
337 return 1;
338 }
339 if (array.level <= 0) {
340 pr_err("Bitmaps not meaningful with level %s\n",
341 map_num(pers, array.level)?:"of this array");
342 return 1;
343 }
344 bitmapsize = array.size;
345 bitmapsize <<= 1;
346 if (get_dev_size(fd, NULL, &array_size) &&
347 array_size > (0x7fffffffULL<<9)) {
348 /* Array is big enough that we cannot trust array.size
349 * try other approaches
350 */
351 bitmapsize = get_component_size(fd);
352 }
353 if (bitmapsize == 0) {
354 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
355 return 1;
356 }
357
358 if (array.level == 10) {
359 int ncopies = (array.layout&255)*((array.layout>>8)&255);
360 bitmapsize = bitmapsize * array.raid_disks / ncopies;
361 }
362
363 st = super_by_fd(fd, &subarray);
364 if (!st) {
365 pr_err("Cannot understand version %d.%d\n",
366 array.major_version, array.minor_version);
367 return 1;
368 }
369 if (subarray) {
370 pr_err("Cannot add bitmaps to sub-arrays yet\n");
371 free(subarray);
372 free(st);
373 return 1;
374 }
375 if (strcmp(s->bitmap_file, "internal") == 0) {
376 int rv;
377 int d;
378 int offset_setable = 0;
379 struct mdinfo *mdi;
380 if (st->ss->add_internal_bitmap == NULL) {
381 pr_err("Internal bitmaps not supported "
382 "with %s metadata\n", st->ss->name);
383 return 1;
384 }
385 mdi = sysfs_read(fd, -1, GET_BITMAP_LOCATION);
386 if (mdi)
387 offset_setable = 1;
388 for (d=0; d< st->max_devs; d++) {
389 mdu_disk_info_t disk;
390 char *dv;
391 disk.number = d;
392 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
393 continue;
394 if (disk.major == 0 &&
395 disk.minor == 0)
396 continue;
397 if ((disk.state & (1<<MD_DISK_SYNC))==0)
398 continue;
399 dv = map_dev(disk.major, disk.minor, 1);
400 if (dv) {
401 int fd2 = dev_open(dv, O_RDWR);
402 if (fd2 < 0)
403 continue;
404 if (st->ss->load_super(st, fd2, NULL)==0) {
405 if (st->ss->add_internal_bitmap(
406 st,
407 &s->bitmap_chunk, c->delay, s->write_behind,
408 bitmapsize, offset_setable,
409 major)
410 )
411 st->ss->write_bitmap(st, fd2);
412 else {
413 pr_err("failed to create internal bitmap"
414 " - chunksize problem.\n");
415 close(fd2);
416 return 1;
417 }
418 }
419 close(fd2);
420 }
421 }
422 if (offset_setable) {
423 st->ss->getinfo_super(st, mdi, NULL);
424 sysfs_init(mdi, fd, -1);
425 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
426 mdi->bitmap_offset);
427 } else {
428 array.state |= (1<<MD_SB_BITMAP_PRESENT);
429 rv = ioctl(fd, SET_ARRAY_INFO, &array);
430 }
431 if (rv < 0) {
432 if (errno == EBUSY)
433 pr_err("Cannot add bitmap while array is"
434 " resyncing or reshaping etc.\n");
435 pr_err("failed to set internal bitmap.\n");
436 return 1;
437 }
438 } else {
439 int uuid[4];
440 int bitmap_fd;
441 int d;
442 int max_devs = st->max_devs;
443
444 /* try to load a superblock */
445 for (d = 0; d < max_devs; d++) {
446 mdu_disk_info_t disk;
447 char *dv;
448 int fd2;
449 disk.number = d;
450 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
451 continue;
452 if ((disk.major==0 && disk.minor==0) ||
453 (disk.state & (1<<MD_DISK_REMOVED)))
454 continue;
455 dv = map_dev(disk.major, disk.minor, 1);
456 if (!dv)
457 continue;
458 fd2 = dev_open(dv, O_RDONLY);
459 if (fd2 >= 0) {
460 if (st->ss->load_super(st, fd2, NULL) == 0) {
461 close(fd2);
462 st->ss->uuid_from_super(st, uuid);
463 break;
464 }
465 close(fd2);
466 }
467 }
468 if (d == max_devs) {
469 pr_err("cannot find UUID for array!\n");
470 return 1;
471 }
472 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
473 c->delay, s->write_behind, bitmapsize, major)) {
474 return 1;
475 }
476 bitmap_fd = open(s->bitmap_file, O_RDWR);
477 if (bitmap_fd < 0) {
478 pr_err("weird: %s cannot be opened\n",
479 s->bitmap_file);
480 return 1;
481 }
482 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
483 int err = errno;
484 if (errno == EBUSY)
485 pr_err("Cannot add bitmap while array is"
486 " resyncing or reshaping etc.\n");
487 pr_err("Cannot set bitmap file for %s: %s\n",
488 devname, strerror(err));
489 return 1;
490 }
491 }
492
493 return 0;
494 }
495
496 /*
497 * When reshaping an array we might need to backup some data.
498 * This is written to all spares with a 'super_block' describing it.
499 * The superblock goes 4K from the end of the used space on the
500 * device.
501 * It if written after the backup is complete.
502 * It has the following structure.
503 */
504
505 static struct mdp_backup_super {
506 char magic[16]; /* md_backup_data-1 or -2 */
507 __u8 set_uuid[16];
508 __u64 mtime;
509 /* start/sizes in 512byte sectors */
510 __u64 devstart; /* address on backup device/file of data */
511 __u64 arraystart;
512 __u64 length;
513 __u32 sb_csum; /* csum of preceeding bytes. */
514 __u32 pad1;
515 __u64 devstart2; /* offset in to data of second section */
516 __u64 arraystart2;
517 __u64 length2;
518 __u32 sb_csum2; /* csum of preceeding bytes. */
519 __u8 pad[512-68-32];
520 } __attribute__((aligned(512))) bsb, bsb2;
521
522 static __u32 bsb_csum(char *buf, int len)
523 {
524 int i;
525 int csum = 0;
526 for (i = 0; i < len; i++)
527 csum = (csum<<3) + buf[0];
528 return __cpu_to_le32(csum);
529 }
530
531 static int check_idle(struct supertype *st)
532 {
533 /* Check that all member arrays for this container, or the
534 * container of this array, are idle
535 */
536 int container_dev = (st->container_dev != NoMdDev
537 ? st->container_dev : st->devnum);
538 char container[40];
539 struct mdstat_ent *ent, *e;
540 int is_idle = 1;
541
542 fmt_devname(container, container_dev);
543 ent = mdstat_read(0, 0);
544 for (e = ent ; e; e = e->next) {
545 if (!is_container_member(e, container))
546 continue;
547 if (e->percent >= 0) {
548 is_idle = 0;
549 break;
550 }
551 }
552 free_mdstat(ent);
553 return is_idle;
554 }
555
556 static int freeze_container(struct supertype *st)
557 {
558 int container_dev = (st->container_dev != NoMdDev
559 ? st->container_dev : st->devnum);
560 char container[40];
561
562 if (!check_idle(st))
563 return -1;
564
565 fmt_devname(container, container_dev);
566
567 if (block_monitor(container, 1)) {
568 pr_err("failed to freeze container\n");
569 return -2;
570 }
571
572 return 1;
573 }
574
575 static void unfreeze_container(struct supertype *st)
576 {
577 int container_dev = (st->container_dev != NoMdDev
578 ? st->container_dev : st->devnum);
579 char container[40];
580
581 fmt_devname(container, container_dev);
582
583 unblock_monitor(container, 1);
584 }
585
586 static int freeze(struct supertype *st)
587 {
588 /* Try to freeze resync/rebuild on this array/container.
589 * Return -1 if the array is busy,
590 * return -2 container cannot be frozen,
591 * return 0 if this kernel doesn't support 'frozen'
592 * return 1 if it worked.
593 */
594 if (st->ss->external)
595 return freeze_container(st);
596 else {
597 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
598 int err;
599 char buf[20];
600
601 if (!sra)
602 return -1;
603 /* Need to clear any 'read-auto' status */
604 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
605 strncmp(buf, "read-auto", 9) == 0)
606 sysfs_set_str(sra, NULL, "array_state", "clean");
607
608 err = sysfs_freeze_array(sra);
609 sysfs_free(sra);
610 return err;
611 }
612 }
613
614 static void unfreeze(struct supertype *st)
615 {
616 if (st->ss->external)
617 return unfreeze_container(st);
618 else {
619 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
620
621 if (sra)
622 sysfs_set_str(sra, NULL, "sync_action", "idle");
623 sysfs_free(sra);
624 }
625 }
626
627 static void wait_reshape(struct mdinfo *sra)
628 {
629 int fd = sysfs_get_fd(sra, NULL, "sync_action");
630 char action[20];
631
632 if (fd < 0)
633 return;
634
635 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
636 strncmp(action, "reshape", 7) == 0) {
637 fd_set rfds;
638 FD_ZERO(&rfds);
639 FD_SET(fd, &rfds);
640 select(fd+1, NULL, NULL, &rfds, NULL);
641 }
642 close(fd);
643 }
644
645 static int reshape_super(struct supertype *st, long long size, int level,
646 int layout, int chunksize, int raid_disks,
647 int delta_disks, char *backup_file, char *dev,
648 int direction, int verbose)
649 {
650 /* nothing extra to check in the native case */
651 if (!st->ss->external)
652 return 0;
653 if (!st->ss->reshape_super ||
654 !st->ss->manage_reshape) {
655 pr_err("%s metadata does not support reshape\n",
656 st->ss->name);
657 return 1;
658 }
659
660 return st->ss->reshape_super(st, size, level, layout, chunksize,
661 raid_disks, delta_disks, backup_file, dev,
662 direction, verbose);
663 }
664
665 static void sync_metadata(struct supertype *st)
666 {
667 if (st->ss->external) {
668 if (st->update_tail) {
669 flush_metadata_updates(st);
670 st->update_tail = &st->updates;
671 } else
672 st->ss->sync_metadata(st);
673 }
674 }
675
676 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
677 {
678 /* when dealing with external metadata subarrays we need to be
679 * prepared to handle EAGAIN. The kernel may need to wait for
680 * mdmon to mark the array active so the kernel can handle
681 * allocations/writeback when preparing the reshape action
682 * (md_allow_write()). We temporarily disable safe_mode_delay
683 * to close a race with the array_state going clean before the
684 * next write to raid_disks / stripe_cache_size
685 */
686 char safe[50];
687 int rc;
688
689 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
690 if (!container ||
691 (strcmp(name, "raid_disks") != 0 &&
692 strcmp(name, "stripe_cache_size") != 0))
693 return sysfs_set_num(sra, NULL, name, n);
694
695 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
696 if (rc <= 0)
697 return -1;
698 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
699 rc = sysfs_set_num(sra, NULL, name, n);
700 if (rc < 0 && errno == EAGAIN) {
701 ping_monitor(container);
702 /* if we get EAGAIN here then the monitor is not active
703 * so stop trying
704 */
705 rc = sysfs_set_num(sra, NULL, name, n);
706 }
707 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
708 return rc;
709 }
710
711 int start_reshape(struct mdinfo *sra, int already_running,
712 int before_data_disks, int data_disks)
713 {
714 int err;
715 unsigned long long sync_max_to_set;
716
717 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
718 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
719 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
720 sra->reshape_progress);
721 if (before_data_disks <= data_disks)
722 sync_max_to_set = sra->reshape_progress / data_disks;
723 else
724 sync_max_to_set = (sra->component_size * data_disks
725 - sra->reshape_progress) / data_disks;
726 if (!already_running)
727 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
728 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
729 if (!already_running)
730 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
731
732 return err;
733 }
734
735 void abort_reshape(struct mdinfo *sra)
736 {
737 sysfs_set_str(sra, NULL, "sync_action", "idle");
738 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
739 sysfs_set_num(sra, NULL, "suspend_hi", 0);
740 sysfs_set_num(sra, NULL, "suspend_lo", 0);
741 sysfs_set_num(sra, NULL, "sync_min", 0);
742 sysfs_set_str(sra, NULL, "sync_max", "max");
743 }
744
745 int remove_disks_for_takeover(struct supertype *st,
746 struct mdinfo *sra,
747 int layout)
748 {
749 int nr_of_copies;
750 struct mdinfo *remaining;
751 int slot;
752
753 if (sra->array.level == 10)
754 nr_of_copies = layout & 0xff;
755 else if (sra->array.level == 1)
756 nr_of_copies = sra->array.raid_disks;
757 else
758 return 1;
759
760 remaining = sra->devs;
761 sra->devs = NULL;
762 /* for each 'copy', select one device and remove from the list. */
763 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
764 struct mdinfo **diskp;
765 int found = 0;
766
767 /* Find a working device to keep */
768 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
769 struct mdinfo *disk = *diskp;
770
771 if (disk->disk.raid_disk < slot)
772 continue;
773 if (disk->disk.raid_disk >= slot + nr_of_copies)
774 continue;
775 if (disk->disk.state & (1<<MD_DISK_REMOVED))
776 continue;
777 if (disk->disk.state & (1<<MD_DISK_FAULTY))
778 continue;
779 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
780 continue;
781
782 /* We have found a good disk to use! */
783 *diskp = disk->next;
784 disk->next = sra->devs;
785 sra->devs = disk;
786 found = 1;
787 break;
788 }
789 if (!found)
790 break;
791 }
792
793 if (slot < sra->array.raid_disks) {
794 /* didn't find all slots */
795 struct mdinfo **e;
796 e = &remaining;
797 while (*e)
798 e = &(*e)->next;
799 *e = sra->devs;
800 sra->devs = remaining;
801 return 1;
802 }
803
804 /* Remove all 'remaining' devices from the array */
805 while (remaining) {
806 struct mdinfo *sd = remaining;
807 remaining = sd->next;
808
809 sysfs_set_str(sra, sd, "state", "faulty");
810 sysfs_set_str(sra, sd, "slot", "none");
811 /* for external metadata disks should be removed in mdmon */
812 if (!st->ss->external)
813 sysfs_set_str(sra, sd, "state", "remove");
814 sd->disk.state |= (1<<MD_DISK_REMOVED);
815 sd->disk.state &= ~(1<<MD_DISK_SYNC);
816 sd->next = sra->devs;
817 sra->devs = sd;
818 }
819 return 0;
820 }
821
822 void reshape_free_fdlist(int *fdlist,
823 unsigned long long *offsets,
824 int size)
825 {
826 int i;
827
828 for (i = 0; i < size; i++)
829 if (fdlist[i] >= 0)
830 close(fdlist[i]);
831
832 free(fdlist);
833 free(offsets);
834 }
835
836 int reshape_prepare_fdlist(char *devname,
837 struct mdinfo *sra,
838 int raid_disks,
839 int nrdisks,
840 unsigned long blocks,
841 char *backup_file,
842 int *fdlist,
843 unsigned long long *offsets)
844 {
845 int d = 0;
846 struct mdinfo *sd;
847
848 for (d = 0; d <= nrdisks; d++)
849 fdlist[d] = -1;
850 d = raid_disks;
851 for (sd = sra->devs; sd; sd = sd->next) {
852 if (sd->disk.state & (1<<MD_DISK_FAULTY))
853 continue;
854 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
855 char *dn = map_dev(sd->disk.major,
856 sd->disk.minor, 1);
857 fdlist[sd->disk.raid_disk]
858 = dev_open(dn, O_RDONLY);
859 offsets[sd->disk.raid_disk] = sd->data_offset*512;
860 if (fdlist[sd->disk.raid_disk] < 0) {
861 pr_err("%s: cannot open component %s\n",
862 devname, dn ? dn : "-unknown-");
863 d = -1;
864 goto release;
865 }
866 } else if (backup_file == NULL) {
867 /* spare */
868 char *dn = map_dev(sd->disk.major,
869 sd->disk.minor, 1);
870 fdlist[d] = dev_open(dn, O_RDWR);
871 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
872 if (fdlist[d] < 0) {
873 pr_err("%s: cannot open component %s\n",
874 devname, dn ? dn : "-unknown-");
875 d = -1;
876 goto release;
877 }
878 d++;
879 }
880 }
881 release:
882 return d;
883 }
884
885 int reshape_open_backup_file(char *backup_file,
886 int fd,
887 char *devname,
888 long blocks,
889 int *fdlist,
890 unsigned long long *offsets,
891 int restart)
892 {
893 /* Return 1 on success, 0 on any form of failure */
894 /* need to check backup file is large enough */
895 char buf[512];
896 struct stat stb;
897 unsigned int dev;
898 int i;
899
900 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
901 S_IRUSR | S_IWUSR);
902 *offsets = 8 * 512;
903 if (*fdlist < 0) {
904 pr_err("%s: cannot create backup file %s: %s\n",
905 devname, backup_file, strerror(errno));
906 return 0;
907 }
908 /* Guard against backup file being on array device.
909 * If array is partitioned or if LVM etc is in the
910 * way this will not notice, but it is better than
911 * nothing.
912 */
913 fstat(*fdlist, &stb);
914 dev = stb.st_dev;
915 fstat(fd, &stb);
916 if (stb.st_rdev == dev) {
917 pr_err("backup file must NOT be"
918 " on the array being reshaped.\n");
919 close(*fdlist);
920 return 0;
921 }
922
923 memset(buf, 0, 512);
924 for (i=0; i < blocks + 8 ; i++) {
925 if (write(*fdlist, buf, 512) != 512) {
926 pr_err("%s: cannot create"
927 " backup file %s: %s\n",
928 devname, backup_file, strerror(errno));
929 return 0;
930 }
931 }
932 if (fsync(*fdlist) != 0) {
933 pr_err("%s: cannot create backup file %s: %s\n",
934 devname, backup_file, strerror(errno));
935 return 0;
936 }
937
938 return 1;
939 }
940
941 unsigned long compute_backup_blocks(int nchunk, int ochunk,
942 unsigned int ndata, unsigned int odata)
943 {
944 unsigned long a, b, blocks;
945 /* So how much do we need to backup.
946 * We need an amount of data which is both a whole number of
947 * old stripes and a whole number of new stripes.
948 * So LCM for (chunksize*datadisks).
949 */
950 a = (ochunk/512) * odata;
951 b = (nchunk/512) * ndata;
952 /* Find GCD */
953 while (a != b) {
954 if (a < b)
955 b -= a;
956 if (b < a)
957 a -= b;
958 }
959 /* LCM == product / GCD */
960 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
961
962 return blocks;
963 }
964
965 char *analyse_change(struct mdinfo *info, struct reshape *re)
966 {
967 /* Based on the current array state in info->array and
968 * the changes in info->new_* etc, determine:
969 * - whether the change is possible
970 * - Intermediate level/raid_disks/layout
971 * - whether a restriping reshape is needed
972 * - number of sectors in minimum change unit. This
973 * will cover a whole number of stripes in 'before' and
974 * 'after'.
975 *
976 * Return message if the change should be rejected
977 * NULL if the change can be achieved
978 *
979 * This can be called as part of starting a reshape, or
980 * when assembling an array that is undergoing reshape.
981 */
982 int new_disks;
983 /* delta_parity records change in number of devices
984 * caused by level change
985 */
986 int delta_parity = 0;
987
988 /* If a new level not explicitly given, we assume no-change */
989 if (info->new_level == UnSet)
990 info->new_level = info->array.level;
991
992 if (info->new_chunk)
993 switch (info->new_level) {
994 case 0:
995 case 4:
996 case 5:
997 case 6:
998 case 10:
999 /* chunk size is meaningful, must divide component_size
1000 * evenly
1001 */
1002 if (info->component_size % (info->new_chunk/512))
1003 return "New chunk size does not"
1004 " divide component size";
1005 break;
1006 default:
1007 return "chunk size not meaningful for this level";
1008 }
1009 else
1010 info->new_chunk = info->array.chunk_size;
1011
1012 switch (info->array.level) {
1013 default:
1014 return "Cannot understand this RAID level";
1015 case 1:
1016 /* RAID1 can convert to RAID1 with different disks, or
1017 * raid5 with 2 disks, or
1018 * raid0 with 1 disk
1019 */
1020 if (info->new_level > 1 &&
1021 (info->component_size & 7))
1022 return "Cannot convert RAID1 of this size - "
1023 "reduce size to multiple of 4K first.";
1024 if (info->new_level == 0) {
1025 if (info->delta_disks != UnSet &&
1026 info->delta_disks != 0)
1027 return "Cannot change number of disks "
1028 "with RAID1->RAID0 conversion";
1029 re->level = 0;
1030 re->before.data_disks = 1;
1031 re->after.data_disks = 1;
1032 re->before.layout = 0;
1033 re->backup_blocks = 0;
1034 re->parity = 0;
1035 return NULL;
1036 }
1037 if (info->new_level == 1) {
1038 if (info->delta_disks == UnSet)
1039 /* Don't know what to do */
1040 return "no change requested for Growing RAID1";
1041 re->level = 1;
1042 re->backup_blocks = 0;
1043 re->parity = 0;
1044 return NULL;
1045 }
1046 if (info->array.raid_disks == 2 &&
1047 info->new_level == 5) {
1048
1049 re->level = 5;
1050 re->before.data_disks = 1;
1051 if (info->delta_disks != UnSet &&
1052 info->delta_disks != 0)
1053 re->after.data_disks = 1 + info->delta_disks;
1054 else
1055 re->after.data_disks = 1;
1056 if (re->after.data_disks < 1)
1057 return "Number of disks too small for RAID5";
1058
1059 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1060 info->array.chunk_size = 65536;
1061 break;
1062 }
1063 /* Could do some multi-stage conversions, but leave that to
1064 * later.
1065 */
1066 return "Impossibly level change request for RAID1";
1067
1068 case 10:
1069 /* RAID10 can only be converted from near mode to
1070 * RAID0 by removing some devices
1071 */
1072 if ((info->array.layout & ~0xff) != 0x100)
1073 return "Cannot Grow RAID10 with far/offset layout";
1074 /* number of devices must be multiple of number of copies */
1075 if (info->array.raid_disks % (info->array.layout & 0xff))
1076 return "RAID10 layout too complex for Grow operation";
1077
1078 if (info->new_level != 0)
1079 return "RAID10 can only be changed to RAID0";
1080 new_disks = (info->array.raid_disks
1081 / (info->array.layout & 0xff));
1082 if (info->delta_disks == UnSet)
1083 info->delta_disks = (new_disks
1084 - info->array.raid_disks);
1085
1086 if (info->delta_disks != new_disks - info->array.raid_disks)
1087 return "New number of raid-devices impossible for RAID10";
1088 if (info->new_chunk &&
1089 info->new_chunk != info->array.chunk_size)
1090 return "Cannot change chunk-size with RAID10 Grow";
1091
1092 /* looks good */
1093 re->level = 0;
1094 re->parity = 0;
1095 re->before.data_disks = new_disks;
1096 re->after.data_disks = re->before.data_disks;
1097 re->before.layout = 0;
1098 re->backup_blocks = 0;
1099 return NULL;
1100
1101 case 0:
1102 /* RAID0 can be converted to RAID10, or to RAID456 */
1103 if (info->new_level == 10) {
1104 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1105 /* Assume near=2 layout */
1106 info->new_layout = 0x102;
1107 info->delta_disks = info->array.raid_disks;
1108 }
1109 if (info->new_layout == UnSet) {
1110 int copies = 1 + (info->delta_disks
1111 / info->array.raid_disks);
1112 if (info->array.raid_disks * (copies-1)
1113 != info->delta_disks)
1114 return "Impossible number of devices"
1115 " for RAID0->RAID10";
1116 info->new_layout = 0x100 + copies;
1117 }
1118 if (info->delta_disks == UnSet) {
1119 int copies = info->new_layout & 0xff;
1120 if (info->new_layout != 0x100 + copies)
1121 return "New layout impossible"
1122 " for RAID0->RAID10";;
1123 info->delta_disks = (copies - 1) *
1124 info->array.raid_disks;
1125 }
1126 if (info->new_chunk &&
1127 info->new_chunk != info->array.chunk_size)
1128 return "Cannot change chunk-size with RAID0->RAID10";
1129 /* looks good */
1130 re->level = 10;
1131 re->parity = 0;
1132 re->before.data_disks = (info->array.raid_disks +
1133 info->delta_disks);
1134 re->after.data_disks = re->before.data_disks;
1135 re->before.layout = info->new_layout;
1136 re->backup_blocks = 0;
1137 return NULL;
1138 }
1139
1140 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1141 * a raid4 style layout of the final level.
1142 */
1143 switch (info->new_level) {
1144 case 4:
1145 delta_parity = 1;
1146 case 0:
1147 re->level = 4;
1148 re->before.layout = 0;
1149 break;
1150 case 5:
1151 delta_parity = 1;
1152 re->level = 5;
1153 re->before.layout = ALGORITHM_PARITY_N;
1154 break;
1155 case 6:
1156 delta_parity = 2;
1157 re->level = 6;
1158 re->before.layout = ALGORITHM_PARITY_N;
1159 break;
1160 default:
1161 return "Impossible level change requested";
1162 }
1163 re->before.data_disks = info->array.raid_disks;
1164 /* determining 'after' layout happens outside this 'switch' */
1165 break;
1166
1167 case 4:
1168 info->array.layout = ALGORITHM_PARITY_N;
1169 case 5:
1170 switch (info->new_level) {
1171 case 0:
1172 delta_parity = -1;
1173 case 4:
1174 re->level = info->array.level;
1175 re->before.data_disks = info->array.raid_disks - 1;
1176 re->before.layout = info->array.layout;
1177 break;
1178 case 5:
1179 re->level = 5;
1180 re->before.data_disks = info->array.raid_disks - 1;
1181 re->before.layout = info->array.layout;
1182 break;
1183 case 6:
1184 delta_parity = 1;
1185 re->level = 6;
1186 re->before.data_disks = info->array.raid_disks - 1;
1187 switch (info->array.layout) {
1188 case ALGORITHM_LEFT_ASYMMETRIC:
1189 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1190 break;
1191 case ALGORITHM_RIGHT_ASYMMETRIC:
1192 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1193 break;
1194 case ALGORITHM_LEFT_SYMMETRIC:
1195 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1196 break;
1197 case ALGORITHM_RIGHT_SYMMETRIC:
1198 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1199 break;
1200 case ALGORITHM_PARITY_0:
1201 re->before.layout = ALGORITHM_PARITY_0_6;
1202 break;
1203 case ALGORITHM_PARITY_N:
1204 re->before.layout = ALGORITHM_PARITY_N_6;
1205 break;
1206 default:
1207 return "Cannot convert an array with this layout";
1208 }
1209 break;
1210 case 1:
1211 if (info->array.raid_disks != 2)
1212 return "Can only convert a 2-device array to RAID1";
1213 if (info->delta_disks != UnSet &&
1214 info->delta_disks != 0)
1215 return "Cannot set raid_disk when "
1216 "converting RAID5->RAID1";
1217 re->level = 1;
1218 break;
1219 default:
1220 return "Impossible level change requested";
1221 }
1222 break;
1223 case 6:
1224 switch (info->new_level) {
1225 case 4:
1226 case 5:
1227 delta_parity = -1;
1228 case 6:
1229 re->level = 6;
1230 re->before.data_disks = info->array.raid_disks - 2;
1231 re->before.layout = info->array.layout;
1232 break;
1233 default:
1234 return "Impossible level change requested";
1235 }
1236 break;
1237 }
1238
1239 /* If we reached here then it looks like a re-stripe is
1240 * happening. We have determined the intermediate level
1241 * and initial raid_disks/layout and stored these in 're'.
1242 *
1243 * We need to deduce the final layout that can be atomically
1244 * converted to the end state.
1245 */
1246 switch (info->new_level) {
1247 case 0:
1248 /* We can only get to RAID0 from RAID4 or RAID5
1249 * with appropriate layout and one extra device
1250 */
1251 if (re->level != 4 && re->level != 5)
1252 return "Cannot covert to RAID0 from this level";
1253
1254 switch (re->level) {
1255 case 4:
1256 re->after.layout = 0;
1257 break;
1258 case 5:
1259 re->after.layout = ALGORITHM_PARITY_N;
1260 break;
1261 }
1262 break;
1263
1264 case 4:
1265 /* We can only get to RAID4 from RAID5 */
1266 if (re->level != 4 && re->level != 5)
1267 return "Cannot convert to RAID4 from this level";
1268
1269 switch (re->level) {
1270 case 4:
1271 re->after.layout = 0;
1272 break;
1273 case 5:
1274 re->after.layout = ALGORITHM_PARITY_N;
1275 break;
1276 }
1277 break;
1278
1279 case 5:
1280 /* We get to RAID5 from RAID5 or RAID6 */
1281 if (re->level != 5 && re->level != 6)
1282 return "Cannot convert to RAID5 from this level";
1283
1284 switch (re->level) {
1285 case 5:
1286 if (info->new_layout == UnSet)
1287 re->after.layout = re->before.layout;
1288 else
1289 re->after.layout = info->new_layout;
1290 break;
1291 case 6:
1292 if (info->new_layout == UnSet)
1293 info->new_layout = re->before.layout;
1294
1295 /* after.layout needs to be raid6 version of new_layout */
1296 if (info->new_layout == ALGORITHM_PARITY_N)
1297 re->after.layout = ALGORITHM_PARITY_N;
1298 else {
1299 char layout[40];
1300 char *ls = map_num(r5layout, info->new_layout);
1301 int l;
1302 if (ls) {
1303 /* Current RAID6 layout has a RAID5
1304 * equivalent - good
1305 */
1306 strcat(strcpy(layout, ls), "-6");
1307 l = map_name(r6layout, layout);
1308 if (l == UnSet)
1309 return "Cannot find RAID6 layout"
1310 " to convert to";
1311 } else {
1312 /* Current RAID6 has no equivalent.
1313 * If it is already a '-6' layout we
1314 * can leave it unchanged, else we must
1315 * fail
1316 */
1317 ls = map_num(r6layout, info->new_layout);
1318 if (!ls ||
1319 strcmp(ls+strlen(ls)-2, "-6") != 0)
1320 return "Please specify new layout";
1321 l = info->new_layout;
1322 }
1323 re->after.layout = l;
1324 }
1325 }
1326 break;
1327
1328 case 6:
1329 /* We must already be at level 6 */
1330 if (re->level != 6)
1331 return "Impossible level change";
1332 if (info->new_layout == UnSet)
1333 re->after.layout = info->array.layout;
1334 else
1335 re->after.layout = info->new_layout;
1336 break;
1337 default:
1338 return "Impossible level change requested";
1339 }
1340 if (info->delta_disks == UnSet)
1341 info->delta_disks = delta_parity;
1342
1343 re->after.data_disks = (re->before.data_disks
1344 + info->delta_disks
1345 - delta_parity);
1346 switch (re->level) {
1347 case 6: re->parity = 2;
1348 break;
1349 case 4:
1350 case 5: re->parity = 1;
1351 break;
1352 default: re->parity = 0;
1353 break;
1354 }
1355 /* So we have a restripe operation, we need to calculate the number
1356 * of blocks per reshape operation.
1357 */
1358 if (info->new_chunk == 0)
1359 info->new_chunk = info->array.chunk_size;
1360 if (re->after.data_disks == re->before.data_disks &&
1361 re->after.layout == re->before.layout &&
1362 info->new_chunk == info->array.chunk_size) {
1363 /* Nothing to change */
1364 re->backup_blocks = 0;
1365 return NULL;
1366 }
1367 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1368 /* chunk and layout changes make no difference */
1369 re->backup_blocks = 0;
1370 return NULL;
1371 }
1372
1373 if (re->after.data_disks == re->before.data_disks &&
1374 get_linux_version() < 2006032)
1375 return "in-place reshape is not safe before 2.6.32 - sorry.";
1376
1377 if (re->after.data_disks < re->before.data_disks &&
1378 get_linux_version() < 2006030)
1379 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1380
1381 re->backup_blocks = compute_backup_blocks(
1382 info->new_chunk, info->array.chunk_size,
1383 re->after.data_disks,
1384 re->before.data_disks);
1385
1386 re->new_size = info->component_size * re->after.data_disks;
1387 return NULL;
1388 }
1389
1390 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1391 char *text_version)
1392 {
1393 struct mdinfo *info;
1394 char *subarray;
1395 int ret_val = -1;
1396
1397 if ((st == NULL) || (sra == NULL))
1398 return ret_val;
1399
1400 if (text_version == NULL)
1401 text_version = sra->text_version;
1402 subarray = strchr(text_version+1, '/')+1;
1403 info = st->ss->container_content(st, subarray);
1404 if (info) {
1405 unsigned long long current_size = 0;
1406 unsigned long long new_size =
1407 info->custom_array_size/2;
1408
1409 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1410 new_size > current_size) {
1411 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1412 < 0)
1413 dprintf("Error: Cannot set array size");
1414 else {
1415 ret_val = 0;
1416 dprintf("Array size changed");
1417 }
1418 dprintf(" from %llu to %llu.\n",
1419 current_size, new_size);
1420 }
1421 sysfs_free(info);
1422 } else
1423 dprintf("Error: set_array_size(): info pointer in NULL\n");
1424
1425 return ret_val;
1426 }
1427
1428 static int reshape_array(char *container, int fd, char *devname,
1429 struct supertype *st, struct mdinfo *info,
1430 int force, struct mddev_dev *devlist,
1431 char *backup_file, int verbose, int forked,
1432 int restart, int freeze_reshape);
1433 static int reshape_container(char *container, char *devname,
1434 int mdfd,
1435 struct supertype *st,
1436 struct mdinfo *info,
1437 int force,
1438 char *backup_file,
1439 int verbose, int restart, int freeze_reshape);
1440
1441 int Grow_reshape(char *devname, int fd,
1442 struct mddev_dev *devlist,
1443 struct context *c, struct shape *s)
1444 {
1445 /* Make some changes in the shape of an array.
1446 * The kernel must support the change.
1447 *
1448 * There are three different changes. Each can trigger
1449 * a resync or recovery so we freeze that until we have
1450 * requested everything (if kernel supports freezing - 2.6.30).
1451 * The steps are:
1452 * - change size (i.e. component_size)
1453 * - change level
1454 * - change layout/chunksize/ndisks
1455 *
1456 * The last can require a reshape. It is different on different
1457 * levels so we need to check the level before actioning it.
1458 * Some times the level change needs to be requested after the
1459 * reshape (e.g. raid6->raid5, raid5->raid0)
1460 *
1461 */
1462 struct mdu_array_info_s array;
1463 int rv = 0;
1464 struct supertype *st;
1465 char *subarray = NULL;
1466
1467 int frozen;
1468 int changed = 0;
1469 char *container = NULL;
1470 char container_buf[20];
1471 int cfd = -1;
1472
1473 struct mddev_dev *dv;
1474 int added_disks;
1475
1476 struct mdinfo info;
1477 struct mdinfo *sra;
1478
1479 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1480 pr_err("%s is not an active md array - aborting\n",
1481 devname);
1482 return 1;
1483 }
1484
1485 if (s->size > 0 &&
1486 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1487 pr_err("cannot change component size at the same time "
1488 "as other changes.\n"
1489 " Change size first, then check data is intact before "
1490 "making other changes.\n");
1491 return 1;
1492 }
1493
1494 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1495 get_linux_version() < 2006032 &&
1496 !check_env("MDADM_FORCE_FEWER")) {
1497 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1498 " Please use a newer kernel\n");
1499 return 1;
1500 }
1501
1502 st = super_by_fd(fd, &subarray);
1503 if (!st) {
1504 pr_err("Unable to determine metadata format for %s\n", devname);
1505 return 1;
1506 }
1507 if (s->raiddisks > st->max_devs) {
1508 pr_err("Cannot increase raid-disks on this array"
1509 " beyond %d\n", st->max_devs);
1510 return 1;
1511 }
1512
1513 /* in the external case we need to check that the requested reshape is
1514 * supported, and perform an initial check that the container holds the
1515 * pre-requisite spare devices (mdmon owns final validation)
1516 */
1517 if (st->ss->external) {
1518 int container_dev;
1519 int rv;
1520
1521 if (subarray) {
1522 container_dev = st->container_dev;
1523 cfd = open_dev_excl(st->container_dev);
1524 } else {
1525 container_dev = st->devnum;
1526 close(fd);
1527 cfd = open_dev_excl(st->devnum);
1528 fd = cfd;
1529 }
1530 if (cfd < 0) {
1531 pr_err("Unable to open container for %s\n",
1532 devname);
1533 free(subarray);
1534 return 1;
1535 }
1536
1537 fmt_devname(container_buf, container_dev);
1538 container = container_buf;
1539
1540 rv = st->ss->load_container(st, cfd, NULL);
1541
1542 if (rv) {
1543 pr_err("Cannot read superblock for %s\n",
1544 devname);
1545 free(subarray);
1546 return 1;
1547 }
1548
1549 /* check if operation is supported for metadata handler */
1550 if (st->ss->container_content) {
1551 struct mdinfo *cc = NULL;
1552 struct mdinfo *content = NULL;
1553
1554 cc = st->ss->container_content(st, subarray);
1555 for (content = cc; content ; content = content->next) {
1556 int allow_reshape = 1;
1557
1558 /* check if reshape is allowed based on metadata
1559 * indications stored in content.array.status
1560 */
1561 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1562 allow_reshape = 0;
1563 if (content->array.state
1564 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1565 allow_reshape = 0;
1566 if (!allow_reshape) {
1567 pr_err("cannot reshape arrays in"
1568 " container with unsupported"
1569 " metadata: %s(%s)\n",
1570 devname, container_buf);
1571 sysfs_free(cc);
1572 free(subarray);
1573 return 1;
1574 }
1575 }
1576 sysfs_free(cc);
1577 }
1578 if (mdmon_running(container_dev))
1579 st->update_tail = &st->updates;
1580 }
1581
1582 added_disks = 0;
1583 for (dv = devlist; dv; dv = dv->next)
1584 added_disks++;
1585 if (s->raiddisks > array.raid_disks &&
1586 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1587 !c->force) {
1588 pr_err("Need %d spare%s to avoid degraded array,"
1589 " and only have %d.\n"
1590 " Use --force to over-ride this check.\n",
1591 s->raiddisks - array.raid_disks,
1592 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1593 array.spare_disks + added_disks);
1594 return 1;
1595 }
1596
1597 sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS
1598 | GET_STATE | GET_VERSION);
1599 if (sra) {
1600 if (st->ss->external && subarray == NULL) {
1601 array.level = LEVEL_CONTAINER;
1602 sra->array.level = LEVEL_CONTAINER;
1603 }
1604 } else {
1605 pr_err("failed to read sysfs parameters for %s\n",
1606 devname);
1607 return 1;
1608 }
1609 frozen = freeze(st);
1610 if (frozen < -1) {
1611 /* freeze() already spewed the reason */
1612 sysfs_free(sra);
1613 return 1;
1614 } else if (frozen < 0) {
1615 pr_err("%s is performing resync/recovery and cannot"
1616 " be reshaped\n", devname);
1617 sysfs_free(sra);
1618 return 1;
1619 }
1620
1621 /* ========= set size =============== */
1622 if (s->size > 0 && (s->size == MAX_DISKS || s->size != (unsigned)array.size)) {
1623 unsigned long long orig_size = get_component_size(fd)/2;
1624 unsigned long long min_csize;
1625 struct mdinfo *mdi;
1626 int raid0_takeover = 0;
1627
1628 if (orig_size == 0)
1629 orig_size = (unsigned) array.size;
1630
1631 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1632 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1633 rv = 1;
1634 goto release;
1635 }
1636 sync_metadata(st);
1637 if (st->ss->external) {
1638 /* metadata can have size limitation
1639 * update size value according to metadata information
1640 */
1641 struct mdinfo *sizeinfo =
1642 st->ss->container_content(st, subarray);
1643 if (sizeinfo) {
1644 unsigned long long new_size =
1645 sizeinfo->custom_array_size/2;
1646 int data_disks = get_data_disks(
1647 sizeinfo->array.level,
1648 sizeinfo->array.layout,
1649 sizeinfo->array.raid_disks);
1650 new_size /= data_disks;
1651 dprintf("Metadata size correction from %llu to "
1652 "%llu (%llu)\n", orig_size, new_size,
1653 new_size * data_disks);
1654 s->size = new_size;
1655 sysfs_free(sizeinfo);
1656 }
1657 }
1658
1659 /* Update the size of each member device in case
1660 * they have been resized. This will never reduce
1661 * below the current used-size. The "size" attribute
1662 * understands '0' to mean 'max'.
1663 */
1664 min_csize = 0;
1665 rv = 0;
1666 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1667 if (sysfs_set_num(sra, mdi, "size",
1668 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1669 /* Probably kernel refusing to let us
1670 * reduce the size - not an error.
1671 */
1672 break;
1673 }
1674 if (array.not_persistent == 0 &&
1675 array.major_version == 0 &&
1676 get_linux_version() < 3001000) {
1677 /* Dangerous to allow size to exceed 2TB */
1678 unsigned long long csize;
1679 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1680 if (csize >= 2ULL*1024*1024*1024)
1681 csize = 2ULL*1024*1024*1024;
1682 if ((min_csize == 0 || (min_csize
1683 > csize)))
1684 min_csize = csize;
1685 }
1686 }
1687 }
1688 if (rv) {
1689 pr_err("Cannot set size on "
1690 "array members.\n");
1691 goto size_change_error;
1692 }
1693 if (min_csize && s->size > min_csize) {
1694 pr_err("Cannot safely make this array "
1695 "use more than 2TB per device on this kernel.\n");
1696 rv = 1;
1697 goto size_change_error;
1698 }
1699 if (min_csize && s->size == MAX_SIZE) {
1700 /* Don't let the kernel choose a size - it will get
1701 * it wrong
1702 */
1703 pr_err("Limited v0.90 array to "
1704 "2TB per device\n");
1705 s->size = min_csize;
1706 }
1707 if (st->ss->external) {
1708 if (sra->array.level == 0) {
1709 rv = sysfs_set_str(sra, NULL, "level",
1710 "raid5");
1711 if (!rv) {
1712 raid0_takeover = 1;
1713 /* get array parametes after takeover
1714 * to chane one parameter at time only
1715 */
1716 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1717 }
1718 }
1719 /* make sure mdmon is
1720 * aware of the new level */
1721 if (!mdmon_running(st->container_dev))
1722 start_mdmon(st->container_dev);
1723 ping_monitor(container);
1724 if (mdmon_running(st->container_dev) &&
1725 st->update_tail == NULL)
1726 st->update_tail = &st->updates;
1727 }
1728
1729 if (s->size == MAX_SIZE)
1730 s->size = 0;
1731 array.size = s->size;
1732 if ((unsigned)array.size != s->size) {
1733 /* got truncated to 32bit, write to
1734 * component_size instead
1735 */
1736 if (sra)
1737 rv = sysfs_set_num(sra, NULL,
1738 "component_size", s->size);
1739 else
1740 rv = -1;
1741 } else {
1742 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1743
1744 /* manage array size when it is managed externally
1745 */
1746 if ((rv == 0) && st->ss->external)
1747 rv = set_array_size(st, sra, sra->text_version);
1748 }
1749
1750 if (raid0_takeover) {
1751 /* do not recync non-existing parity,
1752 * we will drop it anyway
1753 */
1754 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1755 /* go back to raid0, drop parity disk
1756 */
1757 sysfs_set_str(sra, NULL, "level", "raid0");
1758 ioctl(fd, GET_ARRAY_INFO, &array);
1759 }
1760
1761 size_change_error:
1762 if (rv != 0) {
1763 int err = errno;
1764
1765 /* restore metadata */
1766 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1767 UnSet, NULL, devname,
1768 ROLLBACK_METADATA_CHANGES,
1769 c->verbose) == 0)
1770 sync_metadata(st);
1771 pr_err("Cannot set device size for %s: %s\n",
1772 devname, strerror(err));
1773 if (err == EBUSY &&
1774 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1775 cont_err("Bitmap must be removed before size can be changed\n");
1776 rv = 1;
1777 goto release;
1778 }
1779 if (s->assume_clean) {
1780 /* This will fail on kernels newer than 3.0 unless
1781 * a backport has been arranged.
1782 */
1783 if (sra == NULL ||
1784 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1785 pr_err("--assume-clean not support with --grow on this kernel\n");
1786 }
1787 ioctl(fd, GET_ARRAY_INFO, &array);
1788 s->size = get_component_size(fd)/2;
1789 if (s->size == 0)
1790 s->size = array.size;
1791 if (c->verbose >= 0) {
1792 if (s->size == orig_size)
1793 pr_err("component size of %s "
1794 "unchanged at %lluK\n",
1795 devname, s->size);
1796 else
1797 pr_err("component size of %s "
1798 "has been set to %lluK\n",
1799 devname, s->size);
1800 }
1801 changed = 1;
1802 } else if (array.level != LEVEL_CONTAINER) {
1803 s->size = get_component_size(fd)/2;
1804 if (s->size == 0)
1805 s->size = array.size;
1806 }
1807
1808 /* See if there is anything else to do */
1809 if ((s->level == UnSet || s->level == array.level) &&
1810 (s->layout_str == NULL) &&
1811 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1812 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1813 /* Nothing more to do */
1814 if (!changed && c->verbose >= 0)
1815 pr_err("%s: no change requested\n",
1816 devname);
1817 goto release;
1818 }
1819
1820 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1821 * current implementation assumes that following conditions must be met:
1822 * - RAID10:
1823 * - far_copies == 1
1824 * - near_copies == 2
1825 */
1826 if ((s->level == 0 && array.level == 10 && sra &&
1827 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1828 (s->level == 0 && array.level == 1 && sra)) {
1829 int err;
1830 err = remove_disks_for_takeover(st, sra, array.layout);
1831 if (err) {
1832 dprintf(Name": Array cannot be reshaped\n");
1833 if (cfd > -1)
1834 close(cfd);
1835 rv = 1;
1836 goto release;
1837 }
1838 /* Make sure mdmon has seen the device removal
1839 * and updated metadata before we continue with
1840 * level change
1841 */
1842 if (container)
1843 ping_monitor(container);
1844 }
1845
1846 memset(&info, 0, sizeof(info));
1847 info.array = array;
1848 sysfs_init(&info, fd, NoMdDev);
1849 strcpy(info.text_version, sra->text_version);
1850 info.component_size = s->size*2;
1851 info.new_level = s->level;
1852 info.new_chunk = s->chunk * 1024;
1853 if (info.array.level == LEVEL_CONTAINER) {
1854 info.delta_disks = UnSet;
1855 info.array.raid_disks = s->raiddisks;
1856 } else if (s->raiddisks)
1857 info.delta_disks = s->raiddisks - info.array.raid_disks;
1858 else
1859 info.delta_disks = UnSet;
1860 if (s->layout_str == NULL) {
1861 info.new_layout = UnSet;
1862 if (info.array.level == 6 &&
1863 (info.new_level == 6 || info.new_level == UnSet) &&
1864 info.array.layout >= 16) {
1865 pr_err("%s has a non-standard layout. If you"
1866 " wish to preserve this\n", devname);
1867 cont_err("during the reshape, please specify"
1868 " --layout=preserve\n");
1869 cont_err("If you want to change it, specify a"
1870 " layout or use --layout=normalise\n");
1871 rv = 1;
1872 goto release;
1873 }
1874 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1875 strcmp(s->layout_str, "normalize") == 0) {
1876 /* If we have a -6 RAID6 layout, remove the '-6'. */
1877 info.new_layout = UnSet;
1878 if (info.array.level == 6 && info.new_level == UnSet) {
1879 char l[40], *h;
1880 strcpy(l, map_num(r6layout, info.array.layout));
1881 h = strrchr(l, '-');
1882 if (h && strcmp(h, "-6") == 0) {
1883 *h = 0;
1884 info.new_layout = map_name(r6layout, l);
1885 }
1886 } else {
1887 pr_err("%s is only meaningful when reshaping"
1888 " a RAID6 array.\n", s->layout_str);
1889 rv = 1;
1890 goto release;
1891 }
1892 } else if (strcmp(s->layout_str, "preserve") == 0) {
1893 /* This means that a non-standard RAID6 layout
1894 * is OK.
1895 * In particular:
1896 * - When reshape a RAID6 (e.g. adding a device)
1897 * which is in a non-standard layout, it is OK
1898 * to preserve that layout.
1899 * - When converting a RAID5 to RAID6, leave it in
1900 * the XXX-6 layout, don't re-layout.
1901 */
1902 if (info.array.level == 6 && info.new_level == UnSet)
1903 info.new_layout = info.array.layout;
1904 else if (info.array.level == 5 && info.new_level == 6) {
1905 char l[40];
1906 strcpy(l, map_num(r5layout, info.array.layout));
1907 strcat(l, "-6");
1908 info.new_layout = map_name(r6layout, l);
1909 } else {
1910 pr_err("%s in only meaningful when reshaping"
1911 " to RAID6\n", s->layout_str);
1912 rv = 1;
1913 goto release;
1914 }
1915 } else {
1916 int l = info.new_level;
1917 if (l == UnSet)
1918 l = info.array.level;
1919 switch (l) {
1920 case 5:
1921 info.new_layout = map_name(r5layout, s->layout_str);
1922 break;
1923 case 6:
1924 info.new_layout = map_name(r6layout, s->layout_str);
1925 break;
1926 case 10:
1927 info.new_layout = parse_layout_10(s->layout_str);
1928 break;
1929 case LEVEL_FAULTY:
1930 info.new_layout = parse_layout_faulty(s->layout_str);
1931 break;
1932 default:
1933 pr_err("layout not meaningful"
1934 " with this level\n");
1935 rv = 1;
1936 goto release;
1937 }
1938 if (info.new_layout == UnSet) {
1939 pr_err("layout %s not understood"
1940 " for this level\n",
1941 s->layout_str);
1942 rv = 1;
1943 goto release;
1944 }
1945 }
1946
1947 if (array.level == LEVEL_FAULTY) {
1948 if (s->level != UnSet && s->level != array.level) {
1949 pr_err("cannot change level of Faulty device\n");
1950 rv =1 ;
1951 }
1952 if (s->chunk) {
1953 pr_err("cannot set chunksize of Faulty device\n");
1954 rv =1 ;
1955 }
1956 if (s->raiddisks && s->raiddisks != 1) {
1957 pr_err("cannot set raid_disks of Faulty device\n");
1958 rv =1 ;
1959 }
1960 if (s->layout_str) {
1961 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
1962 dprintf("Cannot get array information.\n");
1963 goto release;
1964 }
1965 array.layout = info.new_layout;
1966 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1967 pr_err("failed to set new layout\n");
1968 rv = 1;
1969 } else if (c->verbose >= 0)
1970 printf("layout for %s set to %d\n",
1971 devname, array.layout);
1972 }
1973 } else if (array.level == LEVEL_CONTAINER) {
1974 /* This change is to be applied to every array in the
1975 * container. This is only needed when the metadata imposes
1976 * restraints of the various arrays in the container.
1977 * Currently we only know that IMSM requires all arrays
1978 * to have the same number of devices so changing the
1979 * number of devices (On-Line Capacity Expansion) must be
1980 * performed at the level of the container
1981 */
1982 rv = reshape_container(container, devname, -1, st, &info,
1983 c->force, c->backup_file, c->verbose, 0, 0);
1984 frozen = 0;
1985 } else {
1986 /* get spare devices from external metadata
1987 */
1988 if (st->ss->external) {
1989 struct mdinfo *info2;
1990
1991 info2 = st->ss->container_content(st, subarray);
1992 if (info2) {
1993 info.array.spare_disks =
1994 info2->array.spare_disks;
1995 sysfs_free(info2);
1996 }
1997 }
1998
1999 /* Impose these changes on a single array. First
2000 * check that the metadata is OK with the change. */
2001
2002 if (reshape_super(st, -1, info.new_level,
2003 info.new_layout, info.new_chunk,
2004 info.array.raid_disks, info.delta_disks,
2005 c->backup_file, devname, APPLY_METADATA_CHANGES,
2006 c->verbose)) {
2007 rv = 1;
2008 goto release;
2009 }
2010 sync_metadata(st);
2011 rv = reshape_array(container, fd, devname, st, &info, c->force,
2012 devlist, c->backup_file, c->verbose, 0, 0, 0);
2013 frozen = 0;
2014 }
2015 release:
2016 sysfs_free(sra);
2017 if (frozen > 0)
2018 unfreeze(st);
2019 return rv;
2020 }
2021
2022 /* verify_reshape_position()
2023 * Function checks if reshape position in metadata is not farther
2024 * than position in md.
2025 * Return value:
2026 * 0 : not valid sysfs entry
2027 * it can be caused by not started reshape, it should be started
2028 * by reshape array or raid0 array is before takeover
2029 * -1 : error, reshape position is obviously wrong
2030 * 1 : success, reshape progress correct or updated
2031 */
2032 static int verify_reshape_position(struct mdinfo *info, int level)
2033 {
2034 int ret_val = 0;
2035 char buf[40];
2036 int rv;
2037
2038 /* read sync_max, failure can mean raid0 array */
2039 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2040
2041 if (rv > 0) {
2042 char *ep;
2043 unsigned long long position = strtoull(buf, &ep, 0);
2044
2045 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2046 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2047 position *= get_data_disks(level,
2048 info->new_layout,
2049 info->array.raid_disks);
2050 if (info->reshape_progress < position) {
2051 dprintf("Corrected reshape progress (%llu) to "
2052 "md position (%llu)\n",
2053 info->reshape_progress, position);
2054 info->reshape_progress = position;
2055 ret_val = 1;
2056 } else if (info->reshape_progress > position) {
2057 pr_err("Fatal error: array "
2058 "reshape was not properly frozen "
2059 "(expected reshape position is %llu, "
2060 "but reshape progress is %llu.\n",
2061 position, info->reshape_progress);
2062 ret_val = -1;
2063 } else {
2064 dprintf("Reshape position in md and metadata "
2065 "are the same;");
2066 ret_val = 1;
2067 }
2068 }
2069 } else if (rv == 0) {
2070 /* for valid sysfs entry, 0-length content
2071 * should be indicated as error
2072 */
2073 ret_val = -1;
2074 }
2075
2076 return ret_val;
2077 }
2078
2079 static int reshape_array(char *container, int fd, char *devname,
2080 struct supertype *st, struct mdinfo *info,
2081 int force, struct mddev_dev *devlist,
2082 char *backup_file, int verbose, int forked,
2083 int restart, int freeze_reshape)
2084 {
2085 struct reshape reshape;
2086 int spares_needed;
2087 char *msg;
2088 int orig_level = UnSet;
2089 int disks, odisks;
2090
2091 struct mdu_array_info_s array;
2092 char *c;
2093
2094 struct mddev_dev *dv;
2095 int added_disks;
2096
2097 int *fdlist = NULL;
2098 unsigned long long *offsets = NULL;
2099 int d;
2100 int nrdisks;
2101 int err;
2102 unsigned long blocks;
2103 unsigned long cache;
2104 unsigned long long array_size;
2105 int done;
2106 struct mdinfo *sra = NULL;
2107
2108 /* when reshaping a RAID0, the component_size might be zero.
2109 * So try to fix that up.
2110 */
2111 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2112 dprintf("Cannot get array information.\n");
2113 goto release;
2114 }
2115 if (array.level == 0 && info->component_size == 0) {
2116 get_dev_size(fd, NULL, &array_size);
2117 info->component_size = array_size / array.raid_disks;
2118 }
2119
2120 if (info->reshape_active) {
2121 int new_level = info->new_level;
2122 info->new_level = UnSet;
2123 if (info->delta_disks > 0)
2124 info->array.raid_disks -= info->delta_disks;
2125 msg = analyse_change(info, &reshape);
2126 info->new_level = new_level;
2127 if (info->delta_disks > 0)
2128 info->array.raid_disks += info->delta_disks;
2129 if (!restart)
2130 /* Make sure the array isn't read-only */
2131 ioctl(fd, RESTART_ARRAY_RW, 0);
2132 } else
2133 msg = analyse_change(info, &reshape);
2134 if (msg) {
2135 pr_err("%s\n", msg);
2136 goto release;
2137 }
2138 if (restart &&
2139 (reshape.level != info->array.level ||
2140 reshape.before.layout != info->array.layout ||
2141 reshape.before.data_disks + reshape.parity
2142 != info->array.raid_disks - max(0, info->delta_disks))) {
2143 pr_err("reshape info is not in native format -"
2144 " cannot continue.\n");
2145 goto release;
2146 }
2147
2148 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2149 /* When reshape is restarted from '0', very begin of array
2150 * it is possible that for external metadata reshape and array
2151 * configuration doesn't happen.
2152 * Check if md has the same opinion, and reshape is restarted
2153 * from 0. If so, this is regular reshape start after reshape
2154 * switch in metadata to next array only.
2155 */
2156 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2157 (info->reshape_progress == 0))
2158 restart = 0;
2159 }
2160 if (restart) {
2161 /* reshape already started. just skip to monitoring the reshape */
2162 if (reshape.backup_blocks == 0)
2163 return 0;
2164 goto started;
2165 }
2166 /* The container is frozen but the array may not be.
2167 * So freeze the array so spares don't get put to the wrong use
2168 * FIXME there should probably be a cleaner separation between
2169 * freeze_array and freeze_container.
2170 */
2171 sysfs_freeze_array(info);
2172 /* Check we have enough spares to not be degraded */
2173 added_disks = 0;
2174 for (dv = devlist; dv ; dv=dv->next)
2175 added_disks++;
2176 spares_needed = max(reshape.before.data_disks,
2177 reshape.after.data_disks)
2178 + reshape.parity - array.raid_disks;
2179
2180 if (!force &&
2181 info->new_level > 1 && info->array.level > 1 &&
2182 spares_needed > info->array.spare_disks + added_disks) {
2183 pr_err("Need %d spare%s to avoid degraded array,"
2184 " and only have %d.\n"
2185 " Use --force to over-ride this check.\n",
2186 spares_needed,
2187 spares_needed == 1 ? "" : "s",
2188 info->array.spare_disks + added_disks);
2189 goto release;
2190 }
2191 /* Check we have enough spares to not fail */
2192 spares_needed = max(reshape.before.data_disks,
2193 reshape.after.data_disks)
2194 - array.raid_disks;
2195 if ((info->new_level > 1 || info->new_level == 0) &&
2196 spares_needed > info->array.spare_disks +added_disks) {
2197 pr_err("Need %d spare%s to create working array,"
2198 " and only have %d.\n",
2199 spares_needed,
2200 spares_needed == 1 ? "" : "s",
2201 info->array.spare_disks + added_disks);
2202 goto release;
2203 }
2204
2205 if (reshape.level != array.level) {
2206 char *c = map_num(pers, reshape.level);
2207 int err;
2208 if (c == NULL)
2209 goto release;
2210
2211 err = sysfs_set_str(info, NULL, "level", c);
2212 if (err) {
2213 err = errno;
2214 pr_err("%s: could not set level to %s\n",
2215 devname, c);
2216 if (err == EBUSY &&
2217 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2218 cont_err("Bitmap must be removed"
2219 " before level can be changed\n");
2220 goto release;
2221 }
2222 if (verbose >= 0)
2223 pr_err("level of %s changed to %s\n",
2224 devname, c);
2225 orig_level = array.level;
2226 sysfs_freeze_array(info);
2227
2228 if (reshape.level > 0 && st->ss->external) {
2229 /* make sure mdmon is aware of the new level */
2230 if (mdmon_running(st->container_dev))
2231 flush_mdmon(container);
2232
2233 if (!mdmon_running(st->container_dev))
2234 start_mdmon(st->container_dev);
2235 ping_monitor(container);
2236 if (mdmon_running(st->container_dev) &&
2237 st->update_tail == NULL)
2238 st->update_tail = &st->updates;
2239 }
2240 }
2241 /* ->reshape_super might have chosen some spares from the
2242 * container that it wants to be part of the new array.
2243 * We can collect them with ->container_content and give
2244 * them to the kernel.
2245 */
2246 if (st->ss->reshape_super && st->ss->container_content) {
2247 char *subarray = strchr(info->text_version+1, '/')+1;
2248 struct mdinfo *info2 =
2249 st->ss->container_content(st, subarray);
2250 struct mdinfo *d;
2251
2252 if (info2) {
2253 sysfs_init(info2, fd, st->devnum);
2254 /* When increasing number of devices, we need to set
2255 * new raid_disks before adding these, or they might
2256 * be rejected.
2257 */
2258 if (reshape.backup_blocks &&
2259 reshape.after.data_disks > reshape.before.data_disks)
2260 subarray_set_num(container, info2, "raid_disks",
2261 reshape.after.data_disks +
2262 reshape.parity);
2263 for (d = info2->devs; d; d = d->next) {
2264 if (d->disk.state == 0 &&
2265 d->disk.raid_disk >= 0) {
2266 /* This is a spare that wants to
2267 * be part of the array.
2268 */
2269 add_disk(fd, st, info2, d);
2270 }
2271 }
2272 sysfs_free(info2);
2273 }
2274 }
2275 /* We might have been given some devices to add to the
2276 * array. Now that the array has been changed to the right
2277 * level and frozen, we can safely add them.
2278 */
2279 if (devlist)
2280 Manage_subdevs(devname, fd, devlist, verbose,
2281 0,NULL, 0);
2282
2283 if (reshape.backup_blocks == 0) {
2284 /* No restriping needed, but we might need to impose
2285 * some more changes: layout, raid_disks, chunk_size
2286 */
2287 /* read current array info */
2288 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2289 dprintf("Cannot get array information.\n");
2290 goto release;
2291 }
2292 /* compare current array info with new values and if
2293 * it is different update them to new */
2294 if (info->new_layout != UnSet &&
2295 info->new_layout != array.layout) {
2296 array.layout = info->new_layout;
2297 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2298 pr_err("failed to set new layout\n");
2299 goto release;
2300 } else if (verbose >= 0)
2301 printf("layout for %s set to %d\n",
2302 devname, array.layout);
2303 }
2304 if (info->delta_disks != UnSet &&
2305 info->delta_disks != 0 &&
2306 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
2307 array.raid_disks += info->delta_disks;
2308 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2309 pr_err("failed to set raid disks\n");
2310 goto release;
2311 } else if (verbose >= 0) {
2312 printf("raid_disks for %s set to %d\n",
2313 devname, array.raid_disks);
2314 }
2315 }
2316 if (info->new_chunk != 0 &&
2317 info->new_chunk != array.chunk_size) {
2318 if (sysfs_set_num(info, NULL,
2319 "chunk_size", info->new_chunk) != 0) {
2320 pr_err("failed to set chunk size\n");
2321 goto release;
2322 } else if (verbose >= 0)
2323 printf("chunk size for %s set to %d\n",
2324 devname, array.chunk_size);
2325 }
2326 unfreeze(st);
2327 return 0;
2328 }
2329
2330 /*
2331 * There are three possibilities.
2332 * 1/ The array will shrink.
2333 * We need to ensure the reshape will pause before reaching
2334 * the 'critical section'. We also need to fork and wait for
2335 * that to happen. When it does we
2336 * suspend/backup/complete/unfreeze
2337 *
2338 * 2/ The array will not change size.
2339 * This requires that we keep a backup of a sliding window
2340 * so that we can restore data after a crash. So we need
2341 * to fork and monitor progress.
2342 * In future we will allow the data_offset to change, so
2343 * a sliding backup becomes unnecessary.
2344 *
2345 * 3/ The array will grow. This is relatively easy.
2346 * However the kernel's restripe routines will cheerfully
2347 * overwrite some early data before it is safe. So we
2348 * need to make a backup of the early parts of the array
2349 * and be ready to restore it if rebuild aborts very early.
2350 * For externally managed metadata, we still need a forked
2351 * child to monitor the reshape and suspend IO over the region
2352 * that is being reshaped.
2353 *
2354 * We backup data by writing it to one spare, or to a
2355 * file which was given on command line.
2356 *
2357 * In each case, we first make sure that storage is available
2358 * for the required backup.
2359 * Then we:
2360 * - request the shape change.
2361 * - fork to handle backup etc.
2362 */
2363 started:
2364 /* Check that we can hold all the data */
2365 get_dev_size(fd, NULL, &array_size);
2366 if (reshape.new_size < (array_size/512)) {
2367 pr_err("this change will reduce the size of the array.\n"
2368 " use --grow --array-size first to truncate array.\n"
2369 " e.g. mdadm --grow %s --array-size %llu\n",
2370 devname, reshape.new_size/2);
2371 goto release;
2372 }
2373
2374 sra = sysfs_read(fd, 0,
2375 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2376 GET_CACHE);
2377 if (!sra) {
2378 pr_err("%s: Cannot get array details from sysfs\n",
2379 devname);
2380 goto release;
2381 }
2382
2383 /* Decide how many blocks (sectors) for a reshape
2384 * unit. The number we have so far is just a minimum
2385 */
2386 blocks = reshape.backup_blocks;
2387 if (reshape.before.data_disks ==
2388 reshape.after.data_disks) {
2389 /* Make 'blocks' bigger for better throughput, but
2390 * not so big that we reject it below.
2391 * Try for 16 megabytes
2392 */
2393 while (blocks * 32 < sra->component_size &&
2394 blocks < 16*1024*2)
2395 blocks *= 2;
2396 } else
2397 pr_err("Need to backup %luK of critical "
2398 "section..\n", blocks/2);
2399
2400 if (blocks >= sra->component_size/2) {
2401 pr_err("%s: Something wrong"
2402 " - reshape aborted\n",
2403 devname);
2404 goto release;
2405 }
2406
2407 /* Now we need to open all these devices so we can read/write.
2408 */
2409 nrdisks = max(reshape.before.data_disks,
2410 reshape.after.data_disks) + reshape.parity
2411 + sra->array.spare_disks;
2412 fdlist = xcalloc((1+nrdisks), sizeof(int));
2413 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
2414
2415 odisks = reshape.before.data_disks + reshape.parity;
2416 d = reshape_prepare_fdlist(devname, sra, odisks,
2417 nrdisks, blocks, backup_file,
2418 fdlist, offsets);
2419 if (d < 0) {
2420 goto release;
2421 }
2422 if ((st->ss->manage_reshape == NULL) ||
2423 (st->ss->recover_backup == NULL)) {
2424 if (backup_file == NULL) {
2425 if (reshape.after.data_disks <=
2426 reshape.before.data_disks) {
2427 pr_err("%s: Cannot grow - "
2428 "need backup-file\n", devname);
2429 goto release;
2430 } else if (sra->array.spare_disks == 0) {
2431 pr_err("%s: Cannot grow - "
2432 "need a spare or backup-file to backup "
2433 "critical section\n", devname);
2434 goto release;
2435 }
2436 } else {
2437 if (!reshape_open_backup_file(backup_file, fd, devname,
2438 (signed)blocks,
2439 fdlist+d, offsets+d,
2440 restart)) {
2441 goto release;
2442 }
2443 d++;
2444 }
2445 }
2446
2447 /* lastly, check that the internal stripe cache is
2448 * large enough, or it won't work.
2449 * It must hold at least 4 stripes of the larger
2450 * chunk size
2451 */
2452 cache = max(info->array.chunk_size, info->new_chunk);
2453 cache *= 4; /* 4 stripes minimum */
2454 cache /= 512; /* convert to sectors */
2455 disks = min(reshape.before.data_disks, reshape.after.data_disks);
2456 /* make sure there is room for 'blocks' with a bit to spare */
2457 if (cache < 16 + blocks / disks)
2458 cache = 16 + blocks / disks;
2459 cache /= (4096/512); /* Covert from sectors to pages */
2460
2461 if (sra->cache_size < cache)
2462 subarray_set_num(container, sra, "stripe_cache_size",
2463 cache+1);
2464
2465 /* Right, everything seems fine. Let's kick things off.
2466 * If only changing raid_disks, use ioctl, else use
2467 * sysfs.
2468 */
2469 sync_metadata(st);
2470
2471 sra->new_chunk = info->new_chunk;
2472
2473 if (restart) {
2474 /* for external metadata checkpoint saved by mdmon can be lost
2475 * or missed /due to e.g. crash/. Check if md is not during
2476 * restart farther than metadata points to.
2477 * If so, this means metadata information is obsolete.
2478 */
2479 if (st->ss->external)
2480 verify_reshape_position(info, reshape.level);
2481 sra->reshape_progress = info->reshape_progress;
2482 } else {
2483 sra->reshape_progress = 0;
2484 if (reshape.after.data_disks < reshape.before.data_disks)
2485 /* start from the end of the new array */
2486 sra->reshape_progress = (sra->component_size
2487 * reshape.after.data_disks);
2488 }
2489
2490 if (info->array.chunk_size == info->new_chunk &&
2491 reshape.before.layout == reshape.after.layout &&
2492 st->ss->external == 0) {
2493 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2494 ioctl(fd, GET_ARRAY_INFO, &array);
2495 array.raid_disks = reshape.after.data_disks + reshape.parity;
2496 if (!restart &&
2497 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2498 int err = errno;
2499
2500 pr_err("Cannot set device shape for %s: %s\n",
2501 devname, strerror(errno));
2502
2503 if (err == EBUSY &&
2504 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2505 cont_err("Bitmap must be removed before"
2506 " shape can be changed\n");
2507
2508 goto release;
2509 }
2510 } else if (!restart) {
2511 /* set them all just in case some old 'new_*' value
2512 * persists from some earlier problem.
2513 */
2514 int err = 0;
2515 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2516 err = errno;
2517 if (!err && sysfs_set_num(sra, NULL, "layout",
2518 reshape.after.layout) < 0)
2519 err = errno;
2520 if (!err && subarray_set_num(container, sra, "raid_disks",
2521 reshape.after.data_disks +
2522 reshape.parity) < 0)
2523 err = errno;
2524 if (err) {
2525 pr_err("Cannot set device shape for %s\n",
2526 devname);
2527
2528 if (err == EBUSY &&
2529 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2530 cont_err("Bitmap must be removed before"
2531 " shape can be changed\n");
2532 goto release;
2533 }
2534 }
2535
2536 err = start_reshape(sra, restart, reshape.before.data_disks,
2537 reshape.after.data_disks);
2538 if (err) {
2539 pr_err("Cannot %s reshape for %s\n",
2540 restart ? "continue" : "start",
2541 devname);
2542 goto release;
2543 }
2544 if (restart)
2545 sysfs_set_str(sra, NULL, "array_state", "active");
2546 if (freeze_reshape) {
2547 free(fdlist);
2548 free(offsets);
2549 sysfs_free(sra);
2550 pr_err("Reshape has to be continued from"
2551 " location %llu when root filesystem has been mounted.\n",
2552 sra->reshape_progress);
2553 return 1;
2554 }
2555
2556 /* Now we just need to kick off the reshape and watch, while
2557 * handling backups of the data...
2558 * This is all done by a forked background process.
2559 */
2560 switch(forked ? 0 : fork()) {
2561 case -1:
2562 pr_err("Cannot run child to monitor reshape: %s\n",
2563 strerror(errno));
2564 abort_reshape(sra);
2565 goto release;
2566 default:
2567 free(fdlist);
2568 free(offsets);
2569 sysfs_free(sra);
2570 return 0;
2571 case 0:
2572 map_fork();
2573 break;
2574 }
2575
2576 close(fd);
2577 if (check_env("MDADM_GROW_VERIFY"))
2578 fd = open(devname, O_RDONLY | O_DIRECT);
2579 else
2580 fd = -1;
2581 mlockall(MCL_FUTURE);
2582
2583 if (st->ss->external) {
2584 /* metadata handler takes it from here */
2585 done = st->ss->manage_reshape(
2586 fd, sra, &reshape, st, blocks,
2587 fdlist, offsets,
2588 d - odisks, fdlist+odisks,
2589 offsets+odisks);
2590 } else
2591 done = child_monitor(
2592 fd, sra, &reshape, st, blocks,
2593 fdlist, offsets,
2594 d - odisks, fdlist+odisks,
2595 offsets+odisks);
2596
2597 free(fdlist);
2598 free(offsets);
2599
2600 if (backup_file && done)
2601 unlink(backup_file);
2602 if (!done) {
2603 abort_reshape(sra);
2604 goto out;
2605 }
2606
2607 if (!st->ss->external &&
2608 !(reshape.before.data_disks != reshape.after.data_disks
2609 && info->custom_array_size) &&
2610 info->new_level == reshape.level &&
2611 !forked) {
2612 /* no need to wait for the reshape to finish as
2613 * there is nothing more to do.
2614 */
2615 sysfs_free(sra);
2616 exit(0);
2617 }
2618 wait_reshape(sra);
2619
2620 if (st->ss->external) {
2621 /* Re-load the metadata as much could have changed */
2622 int cfd = open_dev(st->container_dev);
2623 if (cfd >= 0) {
2624 flush_mdmon(container);
2625 st->ss->free_super(st);
2626 st->ss->load_container(st, cfd, container);
2627 close(cfd);
2628 }
2629 }
2630
2631 /* set new array size if required customer_array_size is used
2632 * by this metadata.
2633 */
2634 if (reshape.before.data_disks !=
2635 reshape.after.data_disks &&
2636 info->custom_array_size)
2637 set_array_size(st, info, info->text_version);
2638
2639 if (info->new_level != reshape.level) {
2640
2641 c = map_num(pers, info->new_level);
2642 if (c) {
2643 err = sysfs_set_str(sra, NULL, "level", c);
2644 if (err)
2645 pr_err("%s: could not set level "
2646 "to %s\n", devname, c);
2647 }
2648 if (info->new_level == 0)
2649 st->update_tail = NULL;
2650 }
2651 out:
2652 sysfs_free(sra);
2653 if (forked)
2654 return 0;
2655 unfreeze(st);
2656 exit(0);
2657
2658 release:
2659 free(fdlist);
2660 free(offsets);
2661 if (orig_level != UnSet && sra) {
2662 c = map_num(pers, orig_level);
2663 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
2664 pr_err("aborting level change\n");
2665 }
2666 sysfs_free(sra);
2667 if (!forked)
2668 unfreeze(st);
2669 return 1;
2670 }
2671
2672 /* mdfd handle is passed to be closed in child process (after fork).
2673 */
2674 int reshape_container(char *container, char *devname,
2675 int mdfd,
2676 struct supertype *st,
2677 struct mdinfo *info,
2678 int force,
2679 char *backup_file,
2680 int verbose, int restart, int freeze_reshape)
2681 {
2682 struct mdinfo *cc = NULL;
2683 int rv = restart;
2684 int last_devnum = -1;
2685
2686 /* component_size is not meaningful for a container,
2687 * so pass '-1' meaning 'no change'
2688 */
2689 if (!restart &&
2690 reshape_super(st, -1, info->new_level,
2691 info->new_layout, info->new_chunk,
2692 info->array.raid_disks, info->delta_disks,
2693 backup_file, devname, APPLY_METADATA_CHANGES,
2694 verbose)) {
2695 unfreeze(st);
2696 return 1;
2697 }
2698
2699 sync_metadata(st);
2700
2701 /* ping monitor to be sure that update is on disk
2702 */
2703 ping_monitor(container);
2704
2705 switch (fork()) {
2706 case -1: /* error */
2707 perror("Cannot fork to complete reshape\n");
2708 unfreeze(st);
2709 return 1;
2710 default: /* parent */
2711 if (!freeze_reshape)
2712 printf(Name ": multi-array reshape continues"
2713 " in background\n");
2714 return 0;
2715 case 0: /* child */
2716 map_fork();
2717 break;
2718 }
2719
2720 /* close unused handle in child process
2721 */
2722 if (mdfd > -1)
2723 close(mdfd);
2724
2725 while(1) {
2726 /* For each member array with reshape_active,
2727 * we need to perform the reshape.
2728 * We pick the first array that needs reshaping and
2729 * reshape it. reshape_array() will re-read the metadata
2730 * so the next time through a different array should be
2731 * ready for reshape.
2732 * It is possible that the 'different' array will not
2733 * be assembled yet. In that case we simple exit.
2734 * When it is assembled, the mdadm which assembles it
2735 * will take over the reshape.
2736 */
2737 struct mdinfo *content;
2738 int fd;
2739 struct mdstat_ent *mdstat;
2740 char *adev;
2741
2742 sysfs_free(cc);
2743
2744 cc = st->ss->container_content(st, NULL);
2745
2746 for (content = cc; content ; content = content->next) {
2747 char *subarray;
2748 if (!content->reshape_active)
2749 continue;
2750
2751 subarray = strchr(content->text_version+1, '/')+1;
2752 mdstat = mdstat_by_subdev(subarray,
2753 devname2devnum(container));
2754 if (!mdstat)
2755 continue;
2756 if (mdstat->active == 0) {
2757 pr_err("Skipping inactive "
2758 "array md%i.\n", mdstat->devnum);
2759 free_mdstat(mdstat);
2760 mdstat = NULL;
2761 continue;
2762 }
2763 break;
2764 }
2765 if (!content)
2766 break;
2767
2768 adev = map_dev(dev2major(mdstat->devnum),
2769 dev2minor(mdstat->devnum),
2770 0);
2771 if (!adev)
2772 adev = content->text_version;
2773
2774 fd = open_dev(mdstat->devnum);
2775 if (fd < 0) {
2776 printf(Name ": Device %s cannot be opened for reshape.",
2777 adev);
2778 break;
2779 }
2780
2781 if (last_devnum == mdstat->devnum) {
2782 /* Do not allow for multiple reshape_array() calls for
2783 * the same array.
2784 * It can happen when reshape_array() returns without
2785 * error, when reshape is not finished (wrong reshape
2786 * starting/continuation conditions). Mdmon doesn't
2787 * switch to next array in container and reentry
2788 * conditions for the same array occur.
2789 * This is possibly interim until the behaviour of
2790 * reshape_array is resolved().
2791 */
2792 printf(Name ": Multiple reshape execution detected for "
2793 "device %s.", adev);
2794 close(fd);
2795 break;
2796 }
2797 last_devnum = mdstat->devnum;
2798
2799 sysfs_init(content, fd, mdstat->devnum);
2800
2801 if (mdmon_running(devname2devnum(container)))
2802 flush_mdmon(container);
2803
2804 rv = reshape_array(container, fd, adev, st,
2805 content, force, NULL,
2806 backup_file, verbose, 1, restart,
2807 freeze_reshape);
2808 close(fd);
2809
2810 if (freeze_reshape) {
2811 sysfs_free(cc);
2812 exit(0);
2813 }
2814
2815 restart = 0;
2816 if (rv)
2817 break;
2818
2819 if (mdmon_running(devname2devnum(container)))
2820 flush_mdmon(container);
2821 }
2822 if (!rv)
2823 unfreeze(st);
2824 sysfs_free(cc);
2825 exit(0);
2826 }
2827
2828 /*
2829 * We run a child process in the background which performs the following
2830 * steps:
2831 * - wait for resync to reach a certain point
2832 * - suspend io to the following section
2833 * - backup that section
2834 * - allow resync to proceed further
2835 * - resume io
2836 * - discard the backup.
2837 *
2838 * When are combined in slightly different ways in the three cases.
2839 * Grow:
2840 * - suspend/backup/allow/wait/resume/discard
2841 * Shrink:
2842 * - allow/wait/suspend/backup/allow/wait/resume/discard
2843 * same-size:
2844 * - wait/resume/discard/suspend/backup/allow
2845 *
2846 * suspend/backup/allow always come together
2847 * wait/resume/discard do too.
2848 * For the same-size case we have two backups to improve flow.
2849 *
2850 */
2851
2852 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
2853 unsigned long long backup_point,
2854 unsigned long long wait_point,
2855 unsigned long long *suspend_point,
2856 unsigned long long *reshape_completed)
2857 {
2858 /* This function is called repeatedly by the reshape manager.
2859 * It determines how much progress can safely be made and allows
2860 * that progress.
2861 * - 'info' identifies the array and particularly records in
2862 * ->reshape_progress the metadata's knowledge of progress
2863 * This is a sector offset from the start of the array
2864 * of the next array block to be relocated. This number
2865 * may increase from 0 or decrease from array_size, depending
2866 * on the type of reshape that is happening.
2867 * Note that in contrast, 'sync_completed' is a block count of the
2868 * reshape so far. It gives the distance between the start point
2869 * (head or tail of device) and the next place that data will be
2870 * written. It always increases.
2871 * - 'reshape' is the structure created by analyse_change
2872 * - 'backup_point' shows how much the metadata manager has backed-up
2873 * data. For reshapes with increasing progress, it is the next address
2874 * to be backed up, previous addresses have been backed-up. For
2875 * decreasing progress, it is the earliest address that has been
2876 * backed up - later address are also backed up.
2877 * So addresses between reshape_progress and backup_point are
2878 * backed up providing those are in the 'correct' order.
2879 * - 'wait_point' is an array address. When reshape_completed
2880 * passes this point, progress_reshape should return. It might
2881 * return earlier if it determines that ->reshape_progress needs
2882 * to be updated or further backup is needed.
2883 * - suspend_point is maintained by progress_reshape and the caller
2884 * should not touch it except to initialise to zero.
2885 * It is an array address and it only increases in 2.6.37 and earlier.
2886 * This makes it difficult to handle reducing reshapes with
2887 * external metadata.
2888 * However: it is similar to backup_point in that it records the
2889 * other end of a suspended region from reshape_progress.
2890 * it is moved to extend the region that is safe to backup and/or
2891 * reshape
2892 * - reshape_completed is read from sysfs and returned. The caller
2893 * should copy this into ->reshape_progress when it has reason to
2894 * believe that the metadata knows this, and any backup outside this
2895 * has been erased.
2896 *
2897 * Return value is:
2898 * 1 if more data from backup_point - but only as far as suspend_point,
2899 * should be backed up
2900 * 0 if things are progressing smoothly
2901 * -1 if the reshape is finished because it is all done,
2902 * -2 if the reshape is finished due to an error.
2903 */
2904
2905 int advancing = (reshape->after.data_disks
2906 >= reshape->before.data_disks);
2907 unsigned long long need_backup; /* All data between start of array and
2908 * here will at some point need to
2909 * be backed up.
2910 */
2911 unsigned long long read_offset, write_offset;
2912 unsigned long long write_range;
2913 unsigned long long max_progress, target, completed;
2914 unsigned long long array_size = (info->component_size
2915 * reshape->before.data_disks);
2916 int fd;
2917 char buf[20];
2918
2919 /* First, we unsuspend any region that is now known to be safe.
2920 * If suspend_point is on the 'wrong' side of reshape_progress, then
2921 * we don't have or need suspension at the moment. This is true for
2922 * native metadata when we don't need to back-up.
2923 */
2924 if (advancing) {
2925 if (info->reshape_progress <= *suspend_point)
2926 sysfs_set_num(info, NULL, "suspend_lo",
2927 info->reshape_progress);
2928 } else {
2929 /* Note: this won't work in 2.6.37 and before.
2930 * Something somewhere should make sure we don't need it!
2931 */
2932 if (info->reshape_progress >= *suspend_point)
2933 sysfs_set_num(info, NULL, "suspend_hi",
2934 info->reshape_progress);
2935 }
2936
2937 /* Now work out how far it is safe to progress.
2938 * If the read_offset for ->reshape_progress is less than
2939 * 'blocks' beyond the write_offset, we can only progress as far
2940 * as a backup.
2941 * Otherwise we can progress until the write_offset for the new location
2942 * reaches (within 'blocks' of) the read_offset at the current location.
2943 * However that region must be suspended unless we are using native
2944 * metadata.
2945 * If we need to suspend more, we limit it to 128M per device, which is
2946 * rather arbitrary and should be some time-based calculation.
2947 */
2948 read_offset = info->reshape_progress / reshape->before.data_disks;
2949 write_offset = info->reshape_progress / reshape->after.data_disks;
2950 write_range = info->new_chunk/512;
2951 if (reshape->before.data_disks == reshape->after.data_disks)
2952 need_backup = array_size;
2953 else
2954 need_backup = reshape->backup_blocks;
2955 if (advancing) {
2956 if (read_offset < write_offset + write_range)
2957 max_progress = backup_point;
2958 else
2959 max_progress =
2960 read_offset *
2961 reshape->after.data_disks;
2962 } else {
2963 if (read_offset > write_offset - write_range)
2964 /* Can only progress as far as has been backed up,
2965 * which must be suspended */
2966 max_progress = backup_point;
2967 else if (info->reshape_progress <= need_backup)
2968 max_progress = backup_point;
2969 else {
2970 if (info->array.major_version >= 0)
2971 /* Can progress until backup is needed */
2972 max_progress = need_backup;
2973 else {
2974 /* Can progress until metadata update is required */
2975 max_progress =
2976 read_offset *
2977 reshape->after.data_disks;
2978 /* but data must be suspended */
2979 if (max_progress < *suspend_point)
2980 max_progress = *suspend_point;
2981 }
2982 }
2983 }
2984
2985 /* We know it is safe to progress to 'max_progress' providing
2986 * it is suspended or we are using native metadata.
2987 * Consider extending suspend_point 128M per device if it
2988 * is less than 64M per device beyond reshape_progress.
2989 * But always do a multiple of 'blocks'
2990 * FIXME this is too big - it takes to long to complete
2991 * this much.
2992 */
2993 target = 64*1024*2 * min(reshape->before.data_disks,
2994 reshape->after.data_disks);
2995 target /= reshape->backup_blocks;
2996 if (target < 2)
2997 target = 2;
2998 target *= reshape->backup_blocks;
2999
3000 /* For externally managed metadata we always need to suspend IO to
3001 * the area being reshaped so we regularly push suspend_point forward.
3002 * For native metadata we only need the suspend if we are going to do
3003 * a backup.
3004 */
3005 if (advancing) {
3006 if ((need_backup > info->reshape_progress
3007 || info->array.major_version < 0) &&
3008 *suspend_point < info->reshape_progress + target) {
3009 if (need_backup < *suspend_point + 2 * target)
3010 *suspend_point = need_backup;
3011 else if (*suspend_point + 2 * target < array_size)
3012 *suspend_point += 2 * target;
3013 else
3014 *suspend_point = array_size;
3015 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3016 if (max_progress > *suspend_point)
3017 max_progress = *suspend_point;
3018 }
3019 } else {
3020 if (info->array.major_version >= 0) {
3021 /* Only need to suspend when about to backup */
3022 if (info->reshape_progress < need_backup * 2 &&
3023 *suspend_point > 0) {
3024 *suspend_point = 0;
3025 sysfs_set_num(info, NULL, "suspend_lo", 0);
3026 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3027 }
3028 } else {
3029 /* Need to suspend continually */
3030 if (info->reshape_progress < *suspend_point)
3031 *suspend_point = info->reshape_progress;
3032 if (*suspend_point + target < info->reshape_progress)
3033 /* No need to move suspend region yet */;
3034 else {
3035 if (*suspend_point >= 2 * target)
3036 *suspend_point -= 2 * target;
3037 else
3038 *suspend_point = 0;
3039 sysfs_set_num(info, NULL, "suspend_lo",
3040 *suspend_point);
3041 }
3042 if (max_progress < *suspend_point)
3043 max_progress = *suspend_point;
3044 }
3045 }
3046
3047 /* now set sync_max to allow that progress. sync_max, like
3048 * sync_completed is a count of sectors written per device, so
3049 * we find the difference between max_progress and the start point,
3050 * and divide that by after.data_disks to get a sync_max
3051 * number.
3052 * At the same time we convert wait_point to a similar number
3053 * for comparing against sync_completed.
3054 */
3055 /* scale down max_progress to per_disk */
3056 max_progress /= reshape->after.data_disks;
3057 /* Round to chunk size as some kernels give an erroneously high number */
3058 max_progress /= info->new_chunk/512;
3059 max_progress *= info->new_chunk/512;
3060 /* And round to old chunk size as the kernel wants that */
3061 max_progress /= info->array.chunk_size/512;
3062 max_progress *= info->array.chunk_size/512;
3063 /* Limit progress to the whole device */
3064 if (max_progress > info->component_size)
3065 max_progress = info->component_size;
3066 wait_point /= reshape->after.data_disks;
3067 if (!advancing) {
3068 /* switch from 'device offset' to 'processed block count' */
3069 max_progress = info->component_size - max_progress;
3070 wait_point = info->component_size - wait_point;
3071 }
3072
3073 sysfs_set_num(info, NULL, "sync_max", max_progress);
3074
3075 /* Now wait. If we have already reached the point that we were
3076 * asked to wait to, don't wait at all, else wait for any change.
3077 * We need to select on 'sync_completed' as that is the place that
3078 * notifications happen, but we are really interested in
3079 * 'reshape_position'
3080 */
3081 fd = sysfs_get_fd(info, NULL, "sync_completed");
3082 if (fd < 0)
3083 goto check_progress;
3084
3085 if (sysfs_fd_get_ll(fd, &completed) < 0)
3086 goto check_progress;
3087
3088 while (completed < max_progress && completed < wait_point) {
3089 /* Check that sync_action is still 'reshape' to avoid
3090 * waiting forever on a dead array
3091 */
3092 char action[20];
3093 fd_set rfds;
3094 if (sysfs_get_str(info, NULL, "sync_action",
3095 action, 20) <= 0 ||
3096 strncmp(action, "reshape", 7) != 0)
3097 break;
3098 /* Some kernels reset 'sync_completed' to zero
3099 * before setting 'sync_action' to 'idle'.
3100 * So we need these extra tests.
3101 */
3102 if (completed == 0 && advancing
3103 && info->reshape_progress > 0)
3104 break;
3105 if (completed == 0 && !advancing
3106 && info->reshape_progress < (info->component_size
3107 * reshape->after.data_disks))
3108 break;
3109 FD_ZERO(&rfds);
3110 FD_SET(fd, &rfds);
3111 select(fd+1, NULL, NULL, &rfds, NULL);
3112 if (sysfs_fd_get_ll(fd, &completed) < 0)
3113 goto check_progress;
3114 }
3115 /* Some kernels reset 'sync_completed' to zero,
3116 * we need to have real point we are in md
3117 */
3118 if (completed == 0)
3119 completed = max_progress;
3120
3121 /* some kernels can give an incorrectly high 'completed' number */
3122 completed /= (info->new_chunk/512);
3123 completed *= (info->new_chunk/512);
3124 /* Convert 'completed' back in to a 'progress' number */
3125 completed *= reshape->after.data_disks;
3126 if (!advancing) {
3127 completed = info->component_size * reshape->after.data_disks
3128 - completed;
3129 }
3130 *reshape_completed = completed;
3131
3132 close(fd);
3133
3134 /* We return the need_backup flag. Caller will decide
3135 * how much - a multiple of ->backup_blocks up to *suspend_point
3136 */
3137 if (advancing)
3138 return need_backup > info->reshape_progress;
3139 else
3140 return need_backup >= info->reshape_progress;
3141
3142 check_progress:
3143 /* if we couldn't read a number from sync_completed, then
3144 * either the reshape did complete, or it aborted.
3145 * We can tell which by checking for 'none' in reshape_position.
3146 * If it did abort, then it might immediately restart if it
3147 * it was just a device failure that leaves us degraded but
3148 * functioning.
3149 */
3150 strcpy(buf, "hi");
3151 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3152 || strncmp(buf, "none", 4) != 0) {
3153 /* The abort might only be temporary. Wait up to 10
3154 * seconds for fd to contain a valid number again.
3155 */
3156 struct timeval tv;
3157 int rv = -2;
3158 tv.tv_sec = 10;
3159 tv.tv_usec = 0;
3160 while (fd >= 0 && rv < 0 && tv.tv_sec > 0) {
3161 fd_set rfds;
3162 FD_ZERO(&rfds);
3163 FD_SET(fd, &rfds);
3164 if (select(fd+1, NULL, NULL, &rfds, &tv) != 1)
3165 break;
3166 switch (sysfs_fd_get_ll(fd, &completed)) {
3167 case 0:
3168 /* all good again */
3169 rv = 1;
3170 break;
3171 case -2: /* read error - abort */
3172 tv.tv_sec = 0;
3173 break;
3174 }
3175 }
3176 if (fd >= 0)
3177 close(fd);
3178 return rv; /* abort */
3179 } else {
3180 /* Maybe racing with array shutdown - check state */
3181 if (fd >= 0)
3182 close(fd);
3183 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3184 || strncmp(buf, "inactive", 8) == 0
3185 || strncmp(buf, "clear",5) == 0)
3186 return -2; /* abort */
3187 return -1; /* complete */
3188 }
3189 }
3190
3191 /* FIXME return status is never checked */
3192 static int grow_backup(struct mdinfo *sra,
3193 unsigned long long offset, /* per device */
3194 unsigned long stripes, /* per device, in old chunks */
3195 int *sources, unsigned long long *offsets,
3196 int disks, int chunk, int level, int layout,
3197 int dests, int *destfd, unsigned long long *destoffsets,
3198 int part, int *degraded,
3199 char *buf)
3200 {
3201 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3202 * to storage 'destfd' (offset 'destoffsets'), after first
3203 * suspending IO. Then allow resync to continue
3204 * over the suspended section.
3205 * Use part 'part' of the backup-super-block.
3206 */
3207 int odata = disks;
3208 int rv = 0;
3209 int i;
3210 unsigned long long ll;
3211 int new_degraded;
3212 //printf("offset %llu\n", offset);
3213 if (level >= 4)
3214 odata--;
3215 if (level == 6)
3216 odata--;
3217
3218 /* Check that array hasn't become degraded, else we might backup the wrong data */
3219 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3220 return -1; /* FIXME this error is ignored */
3221 new_degraded = (int)ll;
3222 if (new_degraded != *degraded) {
3223 /* check each device to ensure it is still working */
3224 struct mdinfo *sd;
3225 for (sd = sra->devs ; sd ; sd = sd->next) {
3226 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3227 continue;
3228 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3229 char sbuf[20];
3230 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3231 strstr(sbuf, "faulty") ||
3232 strstr(sbuf, "in_sync") == NULL) {
3233 /* this device is dead */
3234 sd->disk.state = (1<<MD_DISK_FAULTY);
3235 if (sd->disk.raid_disk >= 0 &&
3236 sources[sd->disk.raid_disk] >= 0) {
3237 close(sources[sd->disk.raid_disk]);
3238 sources[sd->disk.raid_disk] = -1;
3239 }
3240 }
3241 }
3242 }
3243 *degraded = new_degraded;
3244 }
3245 if (part) {
3246 bsb.arraystart2 = __cpu_to_le64(offset * odata);
3247 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
3248 } else {
3249 bsb.arraystart = __cpu_to_le64(offset * odata);
3250 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
3251 }
3252 if (part)
3253 bsb.magic[15] = '2';
3254 for (i = 0; i < dests; i++)
3255 if (part)
3256 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
3257 else
3258 lseek64(destfd[i], destoffsets[i], 0);
3259
3260 rv = save_stripes(sources, offsets,
3261 disks, chunk, level, layout,
3262 dests, destfd,
3263 offset*512*odata, stripes * chunk * odata,
3264 buf);
3265
3266 if (rv)
3267 return rv;
3268 bsb.mtime = __cpu_to_le64(time(0));
3269 for (i = 0; i < dests; i++) {
3270 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3271
3272 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3273 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3274 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3275 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3276
3277 rv = -1;
3278 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
3279 != destoffsets[i] - 4096)
3280 break;
3281 if (write(destfd[i], &bsb, 512) != 512)
3282 break;
3283 if (destoffsets[i] > 4096) {
3284 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
3285 destoffsets[i]+stripes*chunk*odata)
3286 break;
3287 if (write(destfd[i], &bsb, 512) != 512)
3288 break;
3289 }
3290 fsync(destfd[i]);
3291 rv = 0;
3292 }
3293
3294 return rv;
3295 }
3296
3297 /* in 2.6.30, the value reported by sync_completed can be
3298 * less that it should be by one stripe.
3299 * This only happens when reshape hits sync_max and pauses.
3300 * So allow wait_backup to either extent sync_max further
3301 * than strictly necessary, or return before the
3302 * sync has got quite as far as we would really like.
3303 * This is what 'blocks2' is for.
3304 * The various caller give appropriate values so that
3305 * every works.
3306 */
3307 /* FIXME return value is often ignored */
3308 static int forget_backup(int dests, int *destfd,
3309 unsigned long long *destoffsets,
3310 int part)
3311 {
3312 /*
3313 * Erase backup 'part' (which is 0 or 1)
3314 */
3315 int i;
3316 int rv;
3317
3318 if (part) {
3319 bsb.arraystart2 = __cpu_to_le64(0);
3320 bsb.length2 = __cpu_to_le64(0);
3321 } else {
3322 bsb.arraystart = __cpu_to_le64(0);
3323 bsb.length = __cpu_to_le64(0);
3324 }
3325 bsb.mtime = __cpu_to_le64(time(0));
3326 rv = 0;
3327 for (i = 0; i < dests; i++) {
3328 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3329 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3330 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3331 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3332 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3333 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
3334 destoffsets[i]-4096)
3335 rv = -1;
3336 if (rv == 0 &&
3337 write(destfd[i], &bsb, 512) != 512)
3338 rv = -1;
3339 fsync(destfd[i]);
3340 }
3341 return rv;
3342 }
3343
3344 static void fail(char *msg)
3345 {
3346 int rv;
3347 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
3348 rv |= (write(2, "\n", 1) != 1);
3349 exit(rv ? 1 : 2);
3350 }
3351
3352 static char *abuf, *bbuf;
3353 static unsigned long long abuflen;
3354 static void validate(int afd, int bfd, unsigned long long offset)
3355 {
3356 /* check that the data in the backup against the array.
3357 * This is only used for regression testing and should not
3358 * be used while the array is active
3359 */
3360 if (afd < 0)
3361 return;
3362 lseek64(bfd, offset - 4096, 0);
3363 if (read(bfd, &bsb2, 512) != 512)
3364 fail("cannot read bsb");
3365 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
3366 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
3367 fail("first csum bad");
3368 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
3369 fail("magic is bad");
3370 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
3371 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
3372 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
3373 fail("second csum bad");
3374
3375 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
3376 fail("devstart is wrong");
3377
3378 if (bsb2.length) {
3379 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
3380
3381 if (abuflen < len) {
3382 free(abuf);
3383 free(bbuf);
3384 abuflen = len;
3385 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
3386 posix_memalign((void**)&bbuf, 4096, abuflen)) {
3387 abuflen = 0;
3388 /* just stop validating on mem-alloc failure */
3389 return;
3390 }
3391 }
3392
3393 lseek64(bfd, offset, 0);
3394 if ((unsigned long long)read(bfd, bbuf, len) != len) {
3395 //printf("len %llu\n", len);
3396 fail("read first backup failed");
3397 }
3398 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
3399 if ((unsigned long long)read(afd, abuf, len) != len)
3400 fail("read first from array failed");
3401 if (memcmp(bbuf, abuf, len) != 0) {
3402 #if 0
3403 int i;
3404 printf("offset=%llu len=%llu\n",
3405 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
3406 for (i=0; i<len; i++)
3407 if (bbuf[i] != abuf[i]) {
3408 printf("first diff byte %d\n", i);
3409 break;
3410 }
3411 #endif
3412 fail("data1 compare failed");
3413 }
3414 }
3415 if (bsb2.length2) {
3416 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
3417
3418 if (abuflen < len) {
3419 free(abuf);
3420 free(bbuf);
3421 abuflen = len;
3422 abuf = xmalloc(abuflen);
3423 bbuf = xmalloc(abuflen);
3424 }
3425
3426 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
3427 if ((unsigned long long)read(bfd, bbuf, len) != len)
3428 fail("read second backup failed");
3429 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
3430 if ((unsigned long long)read(afd, abuf, len) != len)
3431 fail("read second from array failed");
3432 if (memcmp(bbuf, abuf, len) != 0)
3433 fail("data2 compare failed");
3434 }
3435 }
3436
3437 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
3438 struct supertype *st, unsigned long blocks,
3439 int *fds, unsigned long long *offsets,
3440 int dests, int *destfd, unsigned long long *destoffsets)
3441 {
3442 /* Monitor a reshape where backup is being performed using
3443 * 'native' mechanism - either to a backup file, or
3444 * to some space in a spare.
3445 */
3446 char *buf;
3447 int degraded = -1;
3448 unsigned long long speed;
3449 unsigned long long suspend_point, array_size;
3450 unsigned long long backup_point, wait_point;
3451 unsigned long long reshape_completed;
3452 int done = 0;
3453 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
3454 int part = 0; /* The next part of the backup area to fill. It may already
3455 * be full, so we need to check */
3456 int level = reshape->level;
3457 int layout = reshape->before.layout;
3458 int data = reshape->before.data_disks;
3459 int disks = reshape->before.data_disks + reshape->parity;
3460 int chunk = sra->array.chunk_size;
3461 struct mdinfo *sd;
3462 unsigned long stripes;
3463 int uuid[4];
3464
3465 /* set up the backup-super-block. This requires the
3466 * uuid from the array.
3467 */
3468 /* Find a superblock */
3469 for (sd = sra->devs; sd; sd = sd->next) {
3470 char *dn;
3471 int devfd;
3472 int ok;
3473 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3474 continue;
3475 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
3476 devfd = dev_open(dn, O_RDONLY);
3477 if (devfd < 0)
3478 continue;
3479 ok = st->ss->load_super(st, devfd, NULL);
3480 close(devfd);
3481 if (ok == 0)
3482 break;
3483 }
3484 if (!sd) {
3485 pr_err("Cannot find a superblock\n");
3486 return 0;
3487 }
3488
3489 memset(&bsb, 0, 512);
3490 memcpy(bsb.magic, "md_backup_data-1", 16);
3491 st->ss->uuid_from_super(st, uuid);
3492 memcpy(bsb.set_uuid, uuid, 16);
3493 bsb.mtime = __cpu_to_le64(time(0));
3494 bsb.devstart2 = blocks;
3495
3496 stripes = blocks / (sra->array.chunk_size/512) /
3497 reshape->before.data_disks;
3498
3499 if (posix_memalign((void**)&buf, 4096, disks * chunk))
3500 /* Don't start the 'reshape' */
3501 return 0;
3502 if (reshape->before.data_disks == reshape->after.data_disks) {
3503 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
3504 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
3505 }
3506
3507 if (increasing) {
3508 array_size = sra->component_size * reshape->after.data_disks;
3509 backup_point = sra->reshape_progress;
3510 suspend_point = 0;
3511 } else {
3512 array_size = sra->component_size * reshape->before.data_disks;
3513 backup_point = reshape->backup_blocks;
3514 suspend_point = array_size;
3515 }
3516
3517 while (!done) {
3518 int rv;
3519
3520 /* Want to return as soon the oldest backup slot can
3521 * be released as that allows us to start backing up
3522 * some more, providing suspend_point has been
3523 * advanced, which it should have.
3524 */
3525 if (increasing) {
3526 wait_point = array_size;
3527 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
3528 wait_point = (__le64_to_cpu(bsb.arraystart) +
3529 __le64_to_cpu(bsb.length));
3530 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
3531 wait_point = (__le64_to_cpu(bsb.arraystart2) +
3532 __le64_to_cpu(bsb.length2));
3533 } else {
3534 wait_point = 0;
3535 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
3536 wait_point = __le64_to_cpu(bsb.arraystart);
3537 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
3538 wait_point = __le64_to_cpu(bsb.arraystart2);
3539 }
3540
3541 rv = progress_reshape(sra, reshape,
3542 backup_point, wait_point,
3543 &suspend_point, &reshape_completed);
3544 /* external metadata would need to ping_monitor here */
3545 sra->reshape_progress = reshape_completed;
3546
3547 /* Clear any backup region that is before 'here' */
3548 if (increasing) {
3549 if (__le64_to_cpu(bsb.length) > 0 &&
3550 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
3551 __le64_to_cpu(bsb.length)))
3552 forget_backup(dests, destfd,
3553 destoffsets, 0);
3554 if (__le64_to_cpu(bsb.length2) > 0 &&
3555 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
3556 __le64_to_cpu(bsb.length2)))
3557 forget_backup(dests, destfd,
3558 destoffsets, 1);
3559 } else {
3560 if (__le64_to_cpu(bsb.length) > 0 &&
3561 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
3562 forget_backup(dests, destfd,
3563 destoffsets, 0);
3564 if (__le64_to_cpu(bsb.length2) > 0 &&
3565 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
3566 forget_backup(dests, destfd,
3567 destoffsets, 1);
3568 }
3569
3570 if (rv < 0) {
3571 if (rv == -1)
3572 done = 1;
3573 break;
3574 }
3575 if (rv == 0 && increasing && !st->ss->external) {
3576 /* No longer need to monitor this reshape */
3577 done = 1;
3578 break;
3579 }
3580
3581 while (rv) {
3582 unsigned long long offset;
3583 unsigned long actual_stripes;
3584 /* Need to backup some data.
3585 * If 'part' is not used and the desired
3586 * backup size is suspended, do a backup,
3587 * then consider the next part.
3588 */
3589 /* Check that 'part' is unused */
3590 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
3591 break;
3592 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
3593 break;
3594
3595 offset = backup_point / data;
3596 actual_stripes = stripes;
3597 if (increasing) {
3598 if (offset + actual_stripes * (chunk/512) >
3599 sra->component_size)
3600 actual_stripes = ((sra->component_size - offset)
3601 / (chunk/512));
3602 if (offset + actual_stripes * (chunk/512) >
3603 suspend_point/data)
3604 break;
3605 } else {
3606 if (offset < actual_stripes * (chunk/512))
3607 actual_stripes = offset / (chunk/512);
3608 offset -= actual_stripes * (chunk/512);
3609 if (offset < suspend_point/data)
3610 break;
3611 }
3612 if (actual_stripes == 0)
3613 break;
3614 grow_backup(sra, offset, actual_stripes,
3615 fds, offsets,
3616 disks, chunk, level, layout,
3617 dests, destfd, destoffsets,
3618 part, &degraded, buf);
3619 validate(afd, destfd[0], destoffsets[0]);
3620 /* record where 'part' is up to */
3621 part = !part;
3622 if (increasing)
3623 backup_point += actual_stripes * (chunk/512) * data;
3624 else
3625 backup_point -= actual_stripes * (chunk/512) * data;
3626 }
3627 }
3628
3629 /* FIXME maybe call progress_reshape one more time instead */
3630 abort_reshape(sra); /* remove any remaining suspension */
3631 if (reshape->before.data_disks == reshape->after.data_disks)
3632 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
3633 free(buf);
3634 return done;
3635 }
3636
3637 /*
3638 * If any spare contains md_back_data-1 which is recent wrt mtime,
3639 * write that data into the array and update the super blocks with
3640 * the new reshape_progress
3641 */
3642 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
3643 char *backup_file, int verbose)
3644 {
3645 int i, j;
3646 int old_disks;
3647 unsigned long long *offsets;
3648 unsigned long long nstripe, ostripe;
3649 int ndata, odata;
3650
3651 odata = info->array.raid_disks - info->delta_disks - 1;
3652 if (info->array.level == 6) odata--; /* number of data disks */
3653 ndata = info->array.raid_disks - 1;
3654 if (info->new_level == 6) ndata--;
3655
3656 old_disks = info->array.raid_disks - info->delta_disks;
3657
3658 if (info->delta_disks <= 0)
3659 /* Didn't grow, so the backup file must have
3660 * been used
3661 */
3662 old_disks = cnt;
3663 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
3664 struct mdinfo dinfo;
3665 int fd;
3666 int bsbsize;
3667 char *devname, namebuf[20];
3668 unsigned long long lo, hi;
3669
3670 /* This was a spare and may have some saved data on it.
3671 * Load the superblock, find and load the
3672 * backup_super_block.
3673 * If either fail, go on to next device.
3674 * If the backup contains no new info, just return
3675 * else restore data and update all superblocks
3676 */
3677 if (i == old_disks-1) {
3678 fd = open(backup_file, O_RDONLY);
3679 if (fd<0) {
3680 pr_err("backup file %s inaccessible: %s\n",
3681 backup_file, strerror(errno));
3682 continue;
3683 }
3684 devname = backup_file;
3685 } else {
3686 fd = fdlist[i];
3687 if (fd < 0)
3688 continue;
3689 if (st->ss->load_super(st, fd, NULL))
3690 continue;
3691
3692 st->ss->getinfo_super(st, &dinfo, NULL);
3693 st->ss->free_super(st);
3694
3695 if (lseek64(fd,
3696 (dinfo.data_offset + dinfo.component_size - 8) <<9,
3697 0) < 0) {
3698 pr_err("Cannot seek on device %d\n", i);
3699 continue; /* Cannot seek */
3700 }
3701 sprintf(namebuf, "device-%d", i);
3702 devname = namebuf;
3703 }
3704 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
3705 if (verbose)
3706 pr_err("Cannot read from %s\n", devname);
3707 continue; /* Cannot read */
3708 }
3709 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
3710 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
3711 if (verbose)
3712 pr_err("No backup metadata on %s\n", devname);
3713 continue;
3714 }
3715 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
3716 if (verbose)
3717 pr_err("Bad backup-metadata checksum on %s\n", devname);
3718 continue; /* bad checksum */
3719 }
3720 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
3721 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
3722 if (verbose)
3723 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
3724 continue; /* Bad second checksum */
3725 }
3726 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
3727 if (verbose)
3728 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
3729 continue; /* Wrong uuid */
3730 }
3731
3732 /* array utime and backup-mtime should be updated at much the same time, but it seems that
3733 * sometimes they aren't... So allow considerable flexability in matching, and allow
3734 * this test to be overridden by an environment variable.
3735 */
3736 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
3737 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
3738 if (check_env("MDADM_GROW_ALLOW_OLD")) {
3739 pr_err("accepting backup with timestamp %lu "
3740 "for array with timestamp %lu\n",
3741 (unsigned long)__le64_to_cpu(bsb.mtime),
3742 (unsigned long)info->array.utime);
3743 } else {
3744 if (verbose)
3745 pr_err("too-old timestamp on "
3746 "backup-metadata on %s\n", devname);
3747 continue; /* time stamp is too bad */
3748 }
3749 }
3750
3751 if (bsb.magic[15] == '1') {
3752 if (bsb.length == 0)
3753 continue;
3754 if (info->delta_disks >= 0) {
3755 /* reshape_progress is increasing */
3756 if (__le64_to_cpu(bsb.arraystart)
3757 + __le64_to_cpu(bsb.length)
3758 < info->reshape_progress) {
3759 nonew:
3760 if (verbose)
3761 pr_err("backup-metadata found on %s but is not needed\n", devname);
3762 continue; /* No new data here */
3763 }
3764 } else {
3765 /* reshape_progress is decreasing */
3766 if (__le64_to_cpu(bsb.arraystart) >=
3767 info->reshape_progress)
3768 goto nonew; /* No new data here */
3769 }
3770 } else {
3771 if (bsb.length == 0 && bsb.length2 == 0)
3772 continue;
3773 if (info->delta_disks >= 0) {
3774 /* reshape_progress is increasing */
3775 if ((__le64_to_cpu(bsb.arraystart)
3776 + __le64_to_cpu(bsb.length)
3777 < info->reshape_progress)
3778 &&
3779 (__le64_to_cpu(bsb.arraystart2)
3780 + __le64_to_cpu(bsb.length2)
3781 < info->reshape_progress))
3782 goto nonew; /* No new data here */
3783 } else {
3784 /* reshape_progress is decreasing */
3785 if (__le64_to_cpu(bsb.arraystart) >=
3786 info->reshape_progress &&
3787 __le64_to_cpu(bsb.arraystart2) >=
3788 info->reshape_progress)
3789 goto nonew; /* No new data here */
3790 }
3791 }
3792 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
3793 second_fail:
3794 if (verbose)
3795 pr_err("Failed to verify secondary backup-metadata block on %s\n",
3796 devname);
3797 continue; /* Cannot seek */
3798 }
3799 /* There should be a duplicate backup superblock 4k before here */
3800 if (lseek64(fd, -4096, 1) < 0 ||
3801 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
3802 goto second_fail; /* Cannot find leading superblock */
3803 if (bsb.magic[15] == '1')
3804 bsbsize = offsetof(struct mdp_backup_super, pad1);
3805 else
3806 bsbsize = offsetof(struct mdp_backup_super, pad);
3807 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
3808 goto second_fail; /* Cannot find leading superblock */
3809
3810 /* Now need the data offsets for all devices. */
3811 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
3812 for(j=0; j<info->array.raid_disks; j++) {
3813 if (fdlist[j] < 0)
3814 continue;
3815 if (st->ss->load_super(st, fdlist[j], NULL))
3816 /* FIXME should be this be an error */
3817 continue;
3818 st->ss->getinfo_super(st, &dinfo, NULL);
3819 st->ss->free_super(st);
3820 offsets[j] = dinfo.data_offset * 512;
3821 }
3822 printf(Name ": restoring critical section\n");
3823
3824 if (restore_stripes(fdlist, offsets,
3825 info->array.raid_disks,
3826 info->new_chunk,
3827 info->new_level,
3828 info->new_layout,
3829 fd, __le64_to_cpu(bsb.devstart)*512,
3830 __le64_to_cpu(bsb.arraystart)*512,
3831 __le64_to_cpu(bsb.length)*512, NULL)) {
3832 /* didn't succeed, so giveup */
3833 if (verbose)
3834 pr_err("Error restoring backup from %s\n",
3835 devname);
3836 free(offsets);
3837 return 1;
3838 }
3839
3840 if (bsb.magic[15] == '2' &&
3841 restore_stripes(fdlist, offsets,
3842 info->array.raid_disks,
3843 info->new_chunk,
3844 info->new_level,
3845 info->new_layout,
3846 fd, __le64_to_cpu(bsb.devstart)*512 +
3847 __le64_to_cpu(bsb.devstart2)*512,
3848 __le64_to_cpu(bsb.arraystart2)*512,
3849 __le64_to_cpu(bsb.length2)*512, NULL)) {
3850 /* didn't succeed, so giveup */
3851 if (verbose)
3852 pr_err("Error restoring second backup from %s\n",
3853 devname);
3854 free(offsets);
3855 return 1;
3856 }
3857
3858 free(offsets);
3859
3860 /* Ok, so the data is restored. Let's update those superblocks. */
3861
3862 lo = hi = 0;
3863 if (bsb.length) {
3864 lo = __le64_to_cpu(bsb.arraystart);
3865 hi = lo + __le64_to_cpu(bsb.length);
3866 }
3867 if (bsb.magic[15] == '2' && bsb.length2) {
3868 unsigned long long lo1, hi1;
3869 lo1 = __le64_to_cpu(bsb.arraystart2);
3870 hi1 = lo1 + __le64_to_cpu(bsb.length2);
3871 if (lo == hi) {
3872 lo = lo1;
3873 hi = hi1;
3874 } else if (lo < lo1)
3875 hi = hi1;
3876 else
3877 lo = lo1;
3878 }
3879 if (lo < hi &&
3880 (info->reshape_progress < lo ||
3881 info->reshape_progress > hi))
3882 /* backup does not affect reshape_progress*/ ;
3883 else if (info->delta_disks >= 0) {
3884 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
3885 __le64_to_cpu(bsb.length);
3886 if (bsb.magic[15] == '2') {
3887 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
3888 __le64_to_cpu(bsb.length2);
3889 if (p2 > info->reshape_progress)
3890 info->reshape_progress = p2;
3891 }
3892 } else {
3893 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
3894 if (bsb.magic[15] == '2') {
3895 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
3896 if (p2 < info->reshape_progress)
3897 info->reshape_progress = p2;
3898 }
3899 }
3900 for (j=0; j<info->array.raid_disks; j++) {
3901 if (fdlist[j] < 0)
3902 continue;
3903 if (st->ss->load_super(st, fdlist[j], NULL))
3904 continue;
3905 st->ss->getinfo_super(st, &dinfo, NULL);
3906 dinfo.reshape_progress = info->reshape_progress;
3907 st->ss->update_super(st, &dinfo,
3908 "_reshape_progress",
3909 NULL,0, 0, NULL);
3910 st->ss->store_super(st, fdlist[j]);
3911 st->ss->free_super(st);
3912 }
3913 return 0;
3914 }
3915 /* Didn't find any backup data, try to see if any
3916 * was needed.
3917 */
3918 if (info->delta_disks < 0) {
3919 /* When shrinking, the critical section is at the end.
3920 * So see if we are before the critical section.
3921 */
3922 unsigned long long first_block;
3923 nstripe = ostripe = 0;
3924 first_block = 0;
3925 while (ostripe >= nstripe) {
3926 ostripe += info->array.chunk_size / 512;
3927 first_block = ostripe * odata;
3928 nstripe = first_block / ndata / (info->new_chunk/512) *
3929 (info->new_chunk/512);
3930 }
3931
3932 if (info->reshape_progress >= first_block)
3933 return 0;
3934 }
3935 if (info->delta_disks > 0) {
3936 /* See if we are beyond the critical section. */
3937 unsigned long long last_block;
3938 nstripe = ostripe = 0;
3939 last_block = 0;
3940 while (nstripe >= ostripe) {
3941 nstripe += info->new_chunk / 512;
3942 last_block = nstripe * ndata;
3943 ostripe = last_block / odata / (info->array.chunk_size/512) *
3944 (info->array.chunk_size/512);
3945 }
3946
3947 if (info->reshape_progress >= last_block)
3948 return 0;
3949 }
3950 /* needed to recover critical section! */
3951 if (verbose)
3952 pr_err("Failed to find backup of critical section\n");
3953 return 1;
3954 }
3955
3956 int Grow_continue_command(char *devname, int fd,
3957 char *backup_file, int verbose)
3958 {
3959 int ret_val = 0;
3960 struct supertype *st = NULL;
3961 struct mdinfo *content = NULL;
3962 struct mdinfo array;
3963 char *subarray = NULL;
3964 struct mdinfo *cc = NULL;
3965 struct mdstat_ent *mdstat = NULL;
3966 char buf[40];
3967 int cfd = -1;
3968 int fd2 = -1;
3969
3970 dprintf("Grow continue from command line called for %s\n",
3971 devname);
3972
3973 st = super_by_fd(fd, &subarray);
3974 if (!st || !st->ss) {
3975 pr_err("Unable to determine metadata format for %s\n",
3976 devname);
3977 return 1;
3978 }
3979 dprintf("Grow continue is run for ");
3980 if (st->ss->external == 0) {
3981 dprintf("native array (%s)\n", devname);
3982 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
3983 pr_err("%s is not an active md array -"
3984 " aborting\n", devname);
3985 ret_val = 1;
3986 goto Grow_continue_command_exit;
3987 }
3988 content = &array;
3989 sysfs_init(content, fd, st->devnum);
3990 } else {
3991 int container_dev;
3992
3993 if (subarray) {
3994 dprintf("subarray (%s)\n", subarray);
3995 container_dev = st->container_dev;
3996 cfd = open_dev_excl(st->container_dev);
3997 } else {
3998 container_dev = st->devnum;
3999 close(fd);
4000 cfd = open_dev_excl(st->devnum);
4001 dprintf("container (%i)\n", container_dev);
4002 fd = cfd;
4003 }
4004 if (cfd < 0) {
4005 pr_err("Unable to open container "
4006 "for %s\n", devname);
4007 ret_val = 1;
4008 goto Grow_continue_command_exit;
4009 }
4010 fmt_devname(buf, container_dev);
4011
4012 /* find in container array under reshape
4013 */
4014 ret_val = st->ss->load_container(st, cfd, NULL);
4015 if (ret_val) {
4016 pr_err("Cannot read superblock for %s\n",
4017 devname);
4018 ret_val = 1;
4019 goto Grow_continue_command_exit;
4020 }
4021
4022 cc = st->ss->container_content(st, subarray);
4023 for (content = cc; content ; content = content->next) {
4024 char *array;
4025 int allow_reshape = 1;
4026
4027 if (content->reshape_active == 0)
4028 continue;
4029 /* The decision about array or container wide
4030 * reshape is taken in Grow_continue based
4031 * content->reshape_active state, therefore we
4032 * need to check_reshape based on
4033 * reshape_active and subarray name
4034 */
4035 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4036 allow_reshape = 0;
4037 if (content->reshape_active == CONTAINER_RESHAPE &&
4038 (content->array.state
4039 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4040 allow_reshape = 0;
4041
4042 if (!allow_reshape) {
4043 pr_err("cannot continue reshape of an array"
4044 " in container with unsupported"
4045 " metadata: %s(%s)\n",
4046 devname, buf);
4047 ret_val = 1;
4048 goto Grow_continue_command_exit;
4049 }
4050
4051 array = strchr(content->text_version+1, '/')+1;
4052 mdstat = mdstat_by_subdev(array, container_dev);
4053 if (!mdstat)
4054 continue;
4055 if (mdstat->active == 0) {
4056 pr_err("Skipping inactive "
4057 "array md%i.\n", mdstat->devnum);
4058 free_mdstat(mdstat);
4059 mdstat = NULL;
4060 continue;
4061 }
4062 break;
4063 }
4064 if (!content) {
4065 pr_err("Unable to determine reshaped "
4066 "array for %s\n", devname);
4067 ret_val = 1;
4068 goto Grow_continue_command_exit;
4069 }
4070 fd2 = open_dev(mdstat->devnum);
4071 if (fd2 < 0) {
4072 pr_err("cannot open (md%i)\n",
4073 mdstat->devnum);
4074 ret_val = 1;
4075 goto Grow_continue_command_exit;
4076 }
4077
4078 sysfs_init(content, fd2, mdstat->devnum);
4079
4080 /* start mdmon in case it is not running
4081 */
4082 if (!mdmon_running(container_dev))
4083 start_mdmon(container_dev);
4084 ping_monitor(buf);
4085
4086 if (mdmon_running(container_dev))
4087 st->update_tail = &st->updates;
4088 else {
4089 pr_err("No mdmon found. "
4090 "Grow cannot continue.\n");
4091 ret_val = 1;
4092 goto Grow_continue_command_exit;
4093 }
4094 }
4095
4096 /* verify that array under reshape is started from
4097 * correct position
4098 */
4099 if (verify_reshape_position(content,
4100 map_name(pers, mdstat->level)) < 0) {
4101 ret_val = 1;
4102 goto Grow_continue_command_exit;
4103 }
4104
4105 /* continue reshape
4106 */
4107 ret_val = Grow_continue(fd, st, content, backup_file, 0);
4108
4109 Grow_continue_command_exit:
4110 if (fd2 > -1)
4111 close(fd2);
4112 if (cfd > -1)
4113 close(cfd);
4114 st->ss->free_super(st);
4115 free_mdstat(mdstat);
4116 sysfs_free(cc);
4117 free(subarray);
4118
4119 return ret_val;
4120 }
4121
4122 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4123 char *backup_file, int freeze_reshape)
4124 {
4125 int ret_val = 2;
4126
4127 if (!info->reshape_active)
4128 return ret_val;
4129
4130 if (st->ss->external) {
4131 char container[40];
4132 int cfd = open_dev(st->container_dev);
4133
4134 if (cfd < 0)
4135 return 1;
4136
4137 fmt_devname(container, st->container_dev);
4138 st->ss->load_container(st, cfd, container);
4139 close(cfd);
4140 ret_val = reshape_container(container, NULL, mdfd,
4141 st, info, 0, backup_file,
4142 0, 1, freeze_reshape);
4143 } else
4144 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4145 NULL, backup_file, 0, 0, 1,
4146 freeze_reshape);
4147
4148 return ret_val;
4149 }
Unnamed repository; edit this file 'description' to name the repository.