]> 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
DDF: remove "BVD xx is missing".
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27 #include <stddef.h>
28 #include <stdint.h>
29 #include <signal.h>
30 #include <sys/wait.h>
31
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
34 #endif
35 #include "md_u.h"
36 #include "md_p.h"
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char **backup_filep,
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 char *backup_file = *backup_filep;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61 sprintf(buf, "%d:%d",
62 dev->disk.major,
63 dev->disk.minor);
64 fd = dev_open(buf, O_RDWR);
65
66 if (dev->disk.raid_disk >= 0)
67 fdlist[dev->disk.raid_disk] = fd;
68 else
69 fdlist[next_spare++] = fd;
70 }
71
72 if (!backup_file) {
73 backup_file = locate_backup(content->sys_name);
74 *backup_filep = backup_file;
75 }
76
77 if (st->ss->external && st->ss->recover_backup)
78 err = st->ss->recover_backup(st, content);
79 else
80 err = Grow_restart(st, content, fdlist, next_spare,
81 backup_file, verbose > 0);
82
83 while (next_spare > 0) {
84 next_spare--;
85 if (fdlist[next_spare] >= 0)
86 close(fdlist[next_spare]);
87 }
88 free(fdlist);
89 if (err) {
90 pr_err("Failed to restore critical"
91 " section for reshape - sorry.\n");
92 if (!backup_file)
93 pr_err("Possibly you need"
94 " to specify a --backup-file\n");
95 return 1;
96 }
97
98 dprintf("restore_backup() returns status OK.\n");
99 return 0;
100 }
101
102 int Grow_Add_device(char *devname, int fd, char *newdev)
103 {
104 /* Add a device to an active array.
105 * Currently, just extend a linear array.
106 * This requires writing a new superblock on the
107 * new device, calling the kernel to add the device,
108 * and if that succeeds, update the superblock on
109 * all other devices.
110 * This means that we need to *find* all other devices.
111 */
112 struct mdinfo info;
113
114 struct stat stb;
115 int nfd, fd2;
116 int d, nd;
117 struct supertype *st = NULL;
118 char *subarray = NULL;
119
120 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
121 pr_err("cannot get array info for %s\n", devname);
122 return 1;
123 }
124
125 if (info.array.level != -1) {
126 pr_err("can only add devices to linear arrays\n");
127 return 1;
128 }
129
130 st = super_by_fd(fd, &subarray);
131 if (!st) {
132 pr_err("cannot handle arrays with superblock version %d\n",
133 info.array.major_version);
134 return 1;
135 }
136
137 if (subarray) {
138 pr_err("Cannot grow linear sub-arrays yet\n");
139 free(subarray);
140 free(st);
141 return 1;
142 }
143
144 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
145 if (nfd < 0) {
146 pr_err("cannot open %s\n", newdev);
147 free(st);
148 return 1;
149 }
150 fstat(nfd, &stb);
151 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
152 pr_err("%s is not a block device!\n", newdev);
153 close(nfd);
154 free(st);
155 return 1;
156 }
157 /* now check out all the devices and make sure we can read the
158 * superblock */
159 for (d=0 ; d < info.array.raid_disks ; d++) {
160 mdu_disk_info_t disk;
161 char *dv;
162
163 st->ss->free_super(st);
164
165 disk.number = d;
166 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
167 pr_err("cannot get device detail for device %d\n",
168 d);
169 close(nfd);
170 free(st);
171 return 1;
172 }
173 dv = map_dev(disk.major, disk.minor, 1);
174 if (!dv) {
175 pr_err("cannot find device file for device %d\n",
176 d);
177 close(nfd);
178 free(st);
179 return 1;
180 }
181 fd2 = dev_open(dv, O_RDWR);
182 if (fd2 < 0) {
183 pr_err("cannot open device file %s\n", dv);
184 close(nfd);
185 free(st);
186 return 1;
187 }
188
189 if (st->ss->load_super(st, fd2, NULL)) {
190 pr_err("cannot find super block on %s\n", dv);
191 close(nfd);
192 close(fd2);
193 free(st);
194 return 1;
195 }
196 close(fd2);
197 }
198 /* Ok, looks good. Lets update the superblock and write it out to
199 * newdev.
200 */
201
202 info.disk.number = d;
203 info.disk.major = major(stb.st_rdev);
204 info.disk.minor = minor(stb.st_rdev);
205 info.disk.raid_disk = d;
206 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
207 st->ss->update_super(st, &info, "linear-grow-new", newdev,
208 0, 0, NULL);
209
210 if (st->ss->store_super(st, nfd)) {
211 pr_err("Cannot store new superblock on %s\n",
212 newdev);
213 close(nfd);
214 return 1;
215 }
216 close(nfd);
217
218 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
219 pr_err("Cannot add new disk to this array\n");
220 return 1;
221 }
222 /* Well, that seems to have worked.
223 * Now go through and update all superblocks
224 */
225
226 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
227 pr_err("cannot get array info for %s\n", devname);
228 return 1;
229 }
230
231 nd = d;
232 for (d=0 ; d < info.array.raid_disks ; d++) {
233 mdu_disk_info_t disk;
234 char *dv;
235
236 disk.number = d;
237 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
238 pr_err("cannot get device detail for device %d\n",
239 d);
240 return 1;
241 }
242 dv = map_dev(disk.major, disk.minor, 1);
243 if (!dv) {
244 pr_err("cannot find device file for device %d\n",
245 d);
246 return 1;
247 }
248 fd2 = dev_open(dv, O_RDWR);
249 if (fd2 < 0) {
250 pr_err("cannot open device file %s\n", dv);
251 return 1;
252 }
253 if (st->ss->load_super(st, fd2, NULL)) {
254 pr_err("cannot find super block on %s\n", dv);
255 close(fd);
256 return 1;
257 }
258 info.array.raid_disks = nd+1;
259 info.array.nr_disks = nd+1;
260 info.array.active_disks = nd+1;
261 info.array.working_disks = nd+1;
262
263 st->ss->update_super(st, &info, "linear-grow-update", dv,
264 0, 0, NULL);
265
266 if (st->ss->store_super(st, fd2)) {
267 pr_err("Cannot store new superblock on %s\n", dv);
268 close(fd2);
269 return 1;
270 }
271 close(fd2);
272 }
273
274 return 0;
275 }
276
277 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
278 {
279 /*
280 * First check that array doesn't have a bitmap
281 * Then create the bitmap
282 * Then add it
283 *
284 * For internal bitmaps, we need to check the version,
285 * find all the active devices, and write the bitmap block
286 * to all devices
287 */
288 mdu_bitmap_file_t bmf;
289 mdu_array_info_t array;
290 struct supertype *st;
291 char *subarray = NULL;
292 int major = BITMAP_MAJOR_HI;
293 int vers = md_get_version(fd);
294 unsigned long long bitmapsize, array_size;
295
296 if (vers < 9003) {
297 major = BITMAP_MAJOR_HOSTENDIAN;
298 pr_err("Warning - bitmaps created on this kernel"
299 " are not portable\n"
300 " between different architectures. Consider upgrading"
301 " the Linux kernel.\n");
302 }
303
304 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
305 if (errno == ENOMEM)
306 pr_err("Memory allocation failure.\n");
307 else
308 pr_err("bitmaps not supported by this kernel.\n");
309 return 1;
310 }
311 if (bmf.pathname[0]) {
312 if (strcmp(s->bitmap_file,"none")==0) {
313 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
314 pr_err("failed to remove bitmap %s\n",
315 bmf.pathname);
316 return 1;
317 }
318 return 0;
319 }
320 pr_err("%s already has a bitmap (%s)\n",
321 devname, bmf.pathname);
322 return 1;
323 }
324 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
325 pr_err("cannot get array status for %s\n", devname);
326 return 1;
327 }
328 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
329 if (strcmp(s->bitmap_file, "none")==0) {
330 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
331 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
332 pr_err("failed to remove internal bitmap.\n");
333 return 1;
334 }
335 return 0;
336 }
337 pr_err("Internal bitmap already present on %s\n",
338 devname);
339 return 1;
340 }
341
342 if (strcmp(s->bitmap_file, "none") == 0) {
343 pr_err("no bitmap found on %s\n", devname);
344 return 1;
345 }
346 if (array.level <= 0) {
347 pr_err("Bitmaps not meaningful with level %s\n",
348 map_num(pers, array.level)?:"of this array");
349 return 1;
350 }
351 bitmapsize = array.size;
352 bitmapsize <<= 1;
353 if (get_dev_size(fd, NULL, &array_size) &&
354 array_size > (0x7fffffffULL<<9)) {
355 /* Array is big enough that we cannot trust array.size
356 * try other approaches
357 */
358 bitmapsize = get_component_size(fd);
359 }
360 if (bitmapsize == 0) {
361 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
362 return 1;
363 }
364
365 if (array.level == 10) {
366 int ncopies = (array.layout&255)*((array.layout>>8)&255);
367 bitmapsize = bitmapsize * array.raid_disks / ncopies;
368 }
369
370 st = super_by_fd(fd, &subarray);
371 if (!st) {
372 pr_err("Cannot understand version %d.%d\n",
373 array.major_version, array.minor_version);
374 return 1;
375 }
376 if (subarray) {
377 pr_err("Cannot add bitmaps to sub-arrays yet\n");
378 free(subarray);
379 free(st);
380 return 1;
381 }
382 if (strcmp(s->bitmap_file, "internal") == 0) {
383 int rv;
384 int d;
385 int offset_setable = 0;
386 struct mdinfo *mdi;
387 if (st->ss->add_internal_bitmap == NULL) {
388 pr_err("Internal bitmaps not supported "
389 "with %s metadata\n", st->ss->name);
390 return 1;
391 }
392 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
393 if (mdi)
394 offset_setable = 1;
395 for (d=0; d< st->max_devs; d++) {
396 mdu_disk_info_t disk;
397 char *dv;
398 disk.number = d;
399 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
400 continue;
401 if (disk.major == 0 &&
402 disk.minor == 0)
403 continue;
404 if ((disk.state & (1<<MD_DISK_SYNC))==0)
405 continue;
406 dv = map_dev(disk.major, disk.minor, 1);
407 if (dv) {
408 int fd2 = dev_open(dv, O_RDWR);
409 if (fd2 < 0)
410 continue;
411 if (st->ss->load_super(st, fd2, NULL)==0) {
412 if (st->ss->add_internal_bitmap(
413 st,
414 &s->bitmap_chunk, c->delay, s->write_behind,
415 bitmapsize, offset_setable,
416 major)
417 )
418 st->ss->write_bitmap(st, fd2);
419 else {
420 pr_err("failed to create internal bitmap"
421 " - chunksize problem.\n");
422 close(fd2);
423 return 1;
424 }
425 }
426 close(fd2);
427 }
428 }
429 if (offset_setable) {
430 st->ss->getinfo_super(st, mdi, NULL);
431 sysfs_init(mdi, fd, NULL);
432 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
433 mdi->bitmap_offset);
434 } else {
435 array.state |= (1<<MD_SB_BITMAP_PRESENT);
436 rv = ioctl(fd, SET_ARRAY_INFO, &array);
437 }
438 if (rv < 0) {
439 if (errno == EBUSY)
440 pr_err("Cannot add bitmap while array is"
441 " resyncing or reshaping etc.\n");
442 pr_err("failed to set internal bitmap.\n");
443 return 1;
444 }
445 } else {
446 int uuid[4];
447 int bitmap_fd;
448 int d;
449 int max_devs = st->max_devs;
450
451 /* try to load a superblock */
452 for (d = 0; d < max_devs; d++) {
453 mdu_disk_info_t disk;
454 char *dv;
455 int fd2;
456 disk.number = d;
457 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
458 continue;
459 if ((disk.major==0 && disk.minor==0) ||
460 (disk.state & (1<<MD_DISK_REMOVED)))
461 continue;
462 dv = map_dev(disk.major, disk.minor, 1);
463 if (!dv)
464 continue;
465 fd2 = dev_open(dv, O_RDONLY);
466 if (fd2 >= 0) {
467 if (st->ss->load_super(st, fd2, NULL) == 0) {
468 close(fd2);
469 st->ss->uuid_from_super(st, uuid);
470 break;
471 }
472 close(fd2);
473 }
474 }
475 if (d == max_devs) {
476 pr_err("cannot find UUID for array!\n");
477 return 1;
478 }
479 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
480 c->delay, s->write_behind, bitmapsize, major)) {
481 return 1;
482 }
483 bitmap_fd = open(s->bitmap_file, O_RDWR);
484 if (bitmap_fd < 0) {
485 pr_err("weird: %s cannot be opened\n",
486 s->bitmap_file);
487 return 1;
488 }
489 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
490 int err = errno;
491 if (errno == EBUSY)
492 pr_err("Cannot add bitmap while array is"
493 " resyncing or reshaping etc.\n");
494 pr_err("Cannot set bitmap file for %s: %s\n",
495 devname, strerror(err));
496 return 1;
497 }
498 }
499
500 return 0;
501 }
502
503 /*
504 * When reshaping an array we might need to backup some data.
505 * This is written to all spares with a 'super_block' describing it.
506 * The superblock goes 4K from the end of the used space on the
507 * device.
508 * It if written after the backup is complete.
509 * It has the following structure.
510 */
511
512 static struct mdp_backup_super {
513 char magic[16]; /* md_backup_data-1 or -2 */
514 __u8 set_uuid[16];
515 __u64 mtime;
516 /* start/sizes in 512byte sectors */
517 __u64 devstart; /* address on backup device/file of data */
518 __u64 arraystart;
519 __u64 length;
520 __u32 sb_csum; /* csum of preceeding bytes. */
521 __u32 pad1;
522 __u64 devstart2; /* offset in to data of second section */
523 __u64 arraystart2;
524 __u64 length2;
525 __u32 sb_csum2; /* csum of preceeding bytes. */
526 __u8 pad[512-68-32];
527 } __attribute__((aligned(512))) bsb, bsb2;
528
529 static __u32 bsb_csum(char *buf, int len)
530 {
531 int i;
532 int csum = 0;
533 for (i = 0; i < len; i++)
534 csum = (csum<<3) + buf[0];
535 return __cpu_to_le32(csum);
536 }
537
538 static int check_idle(struct supertype *st)
539 {
540 /* Check that all member arrays for this container, or the
541 * container of this array, are idle
542 */
543 char *container = (st->container_devnm[0]
544 ? st->container_devnm : st->devnm);
545 struct mdstat_ent *ent, *e;
546 int is_idle = 1;
547
548 ent = mdstat_read(0, 0);
549 for (e = ent ; e; e = e->next) {
550 if (!is_container_member(e, container))
551 continue;
552 if (e->percent >= 0) {
553 is_idle = 0;
554 break;
555 }
556 }
557 free_mdstat(ent);
558 return is_idle;
559 }
560
561 static int freeze_container(struct supertype *st)
562 {
563 char *container = (st->container_devnm[0]
564 ? st->container_devnm : st->devnm);
565
566 if (!check_idle(st))
567 return -1;
568
569 if (block_monitor(container, 1)) {
570 pr_err("failed to freeze container\n");
571 return -2;
572 }
573
574 return 1;
575 }
576
577 static void unfreeze_container(struct supertype *st)
578 {
579 char *container = (st->container_devnm[0]
580 ? st->container_devnm : st->devnm);
581
582 unblock_monitor(container, 1);
583 }
584
585 static int freeze(struct supertype *st)
586 {
587 /* Try to freeze resync/rebuild on this array/container.
588 * Return -1 if the array is busy,
589 * return -2 container cannot be frozen,
590 * return 0 if this kernel doesn't support 'frozen'
591 * return 1 if it worked.
592 */
593 if (st->ss->external)
594 return freeze_container(st);
595 else {
596 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
597 int err;
598 char buf[20];
599
600 if (!sra)
601 return -1;
602 /* Need to clear any 'read-auto' status */
603 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
604 strncmp(buf, "read-auto", 9) == 0)
605 sysfs_set_str(sra, NULL, "array_state", "clean");
606
607 err = sysfs_freeze_array(sra);
608 sysfs_free(sra);
609 return err;
610 }
611 }
612
613 static void unfreeze(struct supertype *st)
614 {
615 if (st->ss->external)
616 return unfreeze_container(st);
617 else {
618 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
619 char buf[20];
620
621 if (sra &&
622 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
623 && strcmp(buf, "frozen\n") == 0) {
624 printf("unfreeze\n");
625 sysfs_set_str(sra, NULL, "sync_action", "idle");
626 }
627 sysfs_free(sra);
628 }
629 }
630
631 static void wait_reshape(struct mdinfo *sra)
632 {
633 int fd = sysfs_get_fd(sra, NULL, "sync_action");
634 char action[20];
635
636 if (fd < 0)
637 return;
638
639 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
640 strncmp(action, "reshape", 7) == 0)
641 sysfs_wait(fd, NULL);
642 close(fd);
643 }
644
645 static int reshape_super(struct supertype *st, unsigned long long size,
646 int level, 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 // It isn't safe to reset sync_max as we aren't monitoring.
743 // Array really should be stopped at this point.
744 }
745
746 int remove_disks_for_takeover(struct supertype *st,
747 struct mdinfo *sra,
748 int layout)
749 {
750 int nr_of_copies;
751 struct mdinfo *remaining;
752 int slot;
753
754 if (sra->array.level == 10)
755 nr_of_copies = layout & 0xff;
756 else if (sra->array.level == 1)
757 nr_of_copies = sra->array.raid_disks;
758 else
759 return 1;
760
761 remaining = sra->devs;
762 sra->devs = NULL;
763 /* for each 'copy', select one device and remove from the list. */
764 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
765 struct mdinfo **diskp;
766 int found = 0;
767
768 /* Find a working device to keep */
769 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
770 struct mdinfo *disk = *diskp;
771
772 if (disk->disk.raid_disk < slot)
773 continue;
774 if (disk->disk.raid_disk >= slot + nr_of_copies)
775 continue;
776 if (disk->disk.state & (1<<MD_DISK_REMOVED))
777 continue;
778 if (disk->disk.state & (1<<MD_DISK_FAULTY))
779 continue;
780 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
781 continue;
782
783 /* We have found a good disk to use! */
784 *diskp = disk->next;
785 disk->next = sra->devs;
786 sra->devs = disk;
787 found = 1;
788 break;
789 }
790 if (!found)
791 break;
792 }
793
794 if (slot < sra->array.raid_disks) {
795 /* didn't find all slots */
796 struct mdinfo **e;
797 e = &remaining;
798 while (*e)
799 e = &(*e)->next;
800 *e = sra->devs;
801 sra->devs = remaining;
802 return 1;
803 }
804
805 /* Remove all 'remaining' devices from the array */
806 while (remaining) {
807 struct mdinfo *sd = remaining;
808 remaining = sd->next;
809
810 sysfs_set_str(sra, sd, "state", "faulty");
811 sysfs_set_str(sra, sd, "slot", "none");
812 /* for external metadata disks should be removed in mdmon */
813 if (!st->ss->external)
814 sysfs_set_str(sra, sd, "state", "remove");
815 sd->disk.state |= (1<<MD_DISK_REMOVED);
816 sd->disk.state &= ~(1<<MD_DISK_SYNC);
817 sd->next = sra->devs;
818 sra->devs = sd;
819 }
820 return 0;
821 }
822
823 void reshape_free_fdlist(int *fdlist,
824 unsigned long long *offsets,
825 int size)
826 {
827 int i;
828
829 for (i = 0; i < size; i++)
830 if (fdlist[i] >= 0)
831 close(fdlist[i]);
832
833 free(fdlist);
834 free(offsets);
835 }
836
837 int reshape_prepare_fdlist(char *devname,
838 struct mdinfo *sra,
839 int raid_disks,
840 int nrdisks,
841 unsigned long blocks,
842 char *backup_file,
843 int *fdlist,
844 unsigned long long *offsets)
845 {
846 int d = 0;
847 struct mdinfo *sd;
848
849 enable_fds(nrdisks);
850 for (d = 0; d <= nrdisks; d++)
851 fdlist[d] = -1;
852 d = raid_disks;
853 for (sd = sra->devs; sd; sd = sd->next) {
854 if (sd->disk.state & (1<<MD_DISK_FAULTY))
855 continue;
856 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
857 char *dn = map_dev(sd->disk.major,
858 sd->disk.minor, 1);
859 fdlist[sd->disk.raid_disk]
860 = dev_open(dn, O_RDONLY);
861 offsets[sd->disk.raid_disk] = sd->data_offset*512;
862 if (fdlist[sd->disk.raid_disk] < 0) {
863 pr_err("%s: cannot open component %s\n",
864 devname, dn ? dn : "-unknown-");
865 d = -1;
866 goto release;
867 }
868 } else if (backup_file == NULL) {
869 /* spare */
870 char *dn = map_dev(sd->disk.major,
871 sd->disk.minor, 1);
872 fdlist[d] = dev_open(dn, O_RDWR);
873 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
874 if (fdlist[d] < 0) {
875 pr_err("%s: cannot open component %s\n",
876 devname, dn ? dn : "-unknown-");
877 d = -1;
878 goto release;
879 }
880 d++;
881 }
882 }
883 release:
884 return d;
885 }
886
887 int reshape_open_backup_file(char *backup_file,
888 int fd,
889 char *devname,
890 long blocks,
891 int *fdlist,
892 unsigned long long *offsets,
893 char *sys_name,
894 int restart)
895 {
896 /* Return 1 on success, 0 on any form of failure */
897 /* need to check backup file is large enough */
898 char buf[512];
899 struct stat stb;
900 unsigned int dev;
901 int i;
902
903 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
904 S_IRUSR | S_IWUSR);
905 *offsets = 8 * 512;
906 if (*fdlist < 0) {
907 pr_err("%s: cannot create backup file %s: %s\n",
908 devname, backup_file, strerror(errno));
909 return 0;
910 }
911 /* Guard against backup file being on array device.
912 * If array is partitioned or if LVM etc is in the
913 * way this will not notice, but it is better than
914 * nothing.
915 */
916 fstat(*fdlist, &stb);
917 dev = stb.st_dev;
918 fstat(fd, &stb);
919 if (stb.st_rdev == dev) {
920 pr_err("backup file must NOT be"
921 " on the array being reshaped.\n");
922 close(*fdlist);
923 return 0;
924 }
925
926 memset(buf, 0, 512);
927 for (i=0; i < blocks + 8 ; i++) {
928 if (write(*fdlist, buf, 512) != 512) {
929 pr_err("%s: cannot create"
930 " backup file %s: %s\n",
931 devname, backup_file, strerror(errno));
932 return 0;
933 }
934 }
935 if (fsync(*fdlist) != 0) {
936 pr_err("%s: cannot create backup file %s: %s\n",
937 devname, backup_file, strerror(errno));
938 return 0;
939 }
940
941 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
942 char *bu = make_backup(sys_name);
943 symlink(backup_file, bu);
944 free(bu);
945 }
946
947 return 1;
948 }
949
950 unsigned long compute_backup_blocks(int nchunk, int ochunk,
951 unsigned int ndata, unsigned int odata)
952 {
953 unsigned long a, b, blocks;
954 /* So how much do we need to backup.
955 * We need an amount of data which is both a whole number of
956 * old stripes and a whole number of new stripes.
957 * So LCM for (chunksize*datadisks).
958 */
959 a = (ochunk/512) * odata;
960 b = (nchunk/512) * ndata;
961 /* Find GCD */
962 a = GCD(a, b);
963 /* LCM == product / GCD */
964 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
965
966 return blocks;
967 }
968
969 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
970 {
971 /* Based on the current array state in info->array and
972 * the changes in info->new_* etc, determine:
973 * - whether the change is possible
974 * - Intermediate level/raid_disks/layout
975 * - whether a restriping reshape is needed
976 * - number of sectors in minimum change unit. This
977 * will cover a whole number of stripes in 'before' and
978 * 'after'.
979 *
980 * Return message if the change should be rejected
981 * NULL if the change can be achieved
982 *
983 * This can be called as part of starting a reshape, or
984 * when assembling an array that is undergoing reshape.
985 */
986 int near, far, offset, copies;
987 int new_disks;
988 int old_chunk, new_chunk;
989 /* delta_parity records change in number of devices
990 * caused by level change
991 */
992 int delta_parity = 0;
993
994 memset(re, 0, sizeof(*re));
995
996 /* If a new level not explicitly given, we assume no-change */
997 if (info->new_level == UnSet)
998 info->new_level = info->array.level;
999
1000 if (info->new_chunk)
1001 switch (info->new_level) {
1002 case 0:
1003 case 4:
1004 case 5:
1005 case 6:
1006 case 10:
1007 /* chunk size is meaningful, must divide component_size
1008 * evenly
1009 */
1010 if (info->component_size % (info->new_chunk/512)) {
1011 unsigned long long shrink = info->component_size;
1012 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1013 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1014 info->new_chunk/1024, info->component_size/2);
1015 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1016 devname, shrink/2);
1017 pr_err("will shrink the array so the given chunk size would work.\n");
1018 return "";
1019 }
1020 break;
1021 default:
1022 return "chunk size not meaningful for this level";
1023 }
1024 else
1025 info->new_chunk = info->array.chunk_size;
1026
1027 switch (info->array.level) {
1028 default:
1029 return "Cannot understand this RAID level";
1030 case 1:
1031 /* RAID1 can convert to RAID1 with different disks, or
1032 * raid5 with 2 disks, or
1033 * raid0 with 1 disk
1034 */
1035 if (info->new_level > 1 &&
1036 (info->component_size & 7))
1037 return "Cannot convert RAID1 of this size - "
1038 "reduce size to multiple of 4K first.";
1039 if (info->new_level == 0) {
1040 if (info->delta_disks != UnSet &&
1041 info->delta_disks != 0)
1042 return "Cannot change number of disks "
1043 "with RAID1->RAID0 conversion";
1044 re->level = 0;
1045 re->before.data_disks = 1;
1046 re->after.data_disks = 1;
1047 return NULL;
1048 }
1049 if (info->new_level == 1) {
1050 if (info->delta_disks == UnSet)
1051 /* Don't know what to do */
1052 return "no change requested for Growing RAID1";
1053 re->level = 1;
1054 return NULL;
1055 }
1056 if (info->array.raid_disks == 2 &&
1057 info->new_level == 5) {
1058
1059 re->level = 5;
1060 re->before.data_disks = 1;
1061 if (info->delta_disks != UnSet &&
1062 info->delta_disks != 0)
1063 re->after.data_disks = 1 + info->delta_disks;
1064 else
1065 re->after.data_disks = 1;
1066 if (re->after.data_disks < 1)
1067 return "Number of disks too small for RAID5";
1068
1069 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1070 info->array.chunk_size = 65536;
1071 break;
1072 }
1073 /* Could do some multi-stage conversions, but leave that to
1074 * later.
1075 */
1076 return "Impossibly level change request for RAID1";
1077
1078 case 10:
1079 /* RAID10 can be converted from near mode to
1080 * RAID0 by removing some devices.
1081 * It can also be reshaped if the kernel supports
1082 * new_data_offset.
1083 */
1084 switch (info->new_level) {
1085 case 0:
1086 if ((info->array.layout & ~0xff) != 0x100)
1087 return "Cannot Grow RAID10 with far/offset layout";
1088 /* number of devices must be multiple of number of copies */
1089 if (info->array.raid_disks % (info->array.layout & 0xff))
1090 return "RAID10 layout too complex for Grow operation";
1091
1092 new_disks = (info->array.raid_disks
1093 / (info->array.layout & 0xff));
1094 if (info->delta_disks == UnSet)
1095 info->delta_disks = (new_disks
1096 - info->array.raid_disks);
1097
1098 if (info->delta_disks != new_disks - info->array.raid_disks)
1099 return "New number of raid-devices impossible for RAID10";
1100 if (info->new_chunk &&
1101 info->new_chunk != info->array.chunk_size)
1102 return "Cannot change chunk-size with RAID10 Grow";
1103
1104 /* looks good */
1105 re->level = 0;
1106 re->before.data_disks = new_disks;
1107 re->after.data_disks = re->before.data_disks;
1108 return NULL;
1109
1110 case 10:
1111 near = info->array.layout & 0xff;
1112 far = (info->array.layout >> 8) & 0xff;
1113 offset = info->array.layout & 0x10000;
1114 if (far > 1 && !offset)
1115 return "Cannot reshape RAID10 in far-mode";
1116 copies = near * far;
1117
1118 old_chunk = info->array.chunk_size * far;
1119
1120 if (info->new_layout == UnSet)
1121 info->new_layout = info->array.layout;
1122 else {
1123 near = info->new_layout & 0xff;
1124 far = (info->new_layout >> 8) & 0xff;
1125 offset = info->new_layout & 0x10000;
1126 if (far > 1 && !offset)
1127 return "Cannot reshape RAID10 to far-mode";
1128 if (near * far != copies)
1129 return "Cannot change number of copies"
1130 " when reshaping RAID10";
1131 }
1132 if (info->delta_disks == UnSet)
1133 info->delta_disks = 0;
1134 new_disks = (info->array.raid_disks +
1135 info->delta_disks);
1136
1137 new_chunk = info->new_chunk * far;
1138
1139 re->level = 10;
1140 re->before.layout = info->array.layout;
1141 re->before.data_disks = info->array.raid_disks;
1142 re->after.layout = info->new_layout;
1143 re->after.data_disks = new_disks;
1144 /* For RAID10 we don't do backup but do allow reshape,
1145 * so set backup_blocks to INVALID_SECTORS rather than
1146 * zero.
1147 * And there is no need to synchronise stripes on both
1148 * 'old' and 'new'. So the important
1149 * number is the minimum data_offset difference
1150 * which is the larger of (offset copies * chunk).
1151 */
1152 re->backup_blocks = INVALID_SECTORS;
1153 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1154 if (new_disks < re->before.data_disks &&
1155 info->space_after < re->min_offset_change)
1156 /* Reduce component size by one chunk */
1157 re->new_size = (info->component_size -
1158 re->min_offset_change);
1159 else
1160 re->new_size = info->component_size;
1161 re->new_size = re->new_size * new_disks / copies;
1162 return NULL;
1163
1164 default:
1165 return "RAID10 can only be changed to RAID0";
1166 }
1167 case 0:
1168 /* RAID0 can be converted to RAID10, or to RAID456 */
1169 if (info->new_level == 10) {
1170 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1171 /* Assume near=2 layout */
1172 info->new_layout = 0x102;
1173 info->delta_disks = info->array.raid_disks;
1174 }
1175 if (info->new_layout == UnSet) {
1176 int copies = 1 + (info->delta_disks
1177 / info->array.raid_disks);
1178 if (info->array.raid_disks * (copies-1)
1179 != info->delta_disks)
1180 return "Impossible number of devices"
1181 " for RAID0->RAID10";
1182 info->new_layout = 0x100 + copies;
1183 }
1184 if (info->delta_disks == UnSet) {
1185 int copies = info->new_layout & 0xff;
1186 if (info->new_layout != 0x100 + copies)
1187 return "New layout impossible"
1188 " for RAID0->RAID10";;
1189 info->delta_disks = (copies - 1) *
1190 info->array.raid_disks;
1191 }
1192 if (info->new_chunk &&
1193 info->new_chunk != info->array.chunk_size)
1194 return "Cannot change chunk-size with RAID0->RAID10";
1195 /* looks good */
1196 re->level = 10;
1197 re->before.data_disks = (info->array.raid_disks +
1198 info->delta_disks);
1199 re->after.data_disks = re->before.data_disks;
1200 re->before.layout = info->new_layout;
1201 return NULL;
1202 }
1203
1204 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1205 * a raid4 style layout of the final level.
1206 */
1207 switch (info->new_level) {
1208 case 4:
1209 delta_parity = 1;
1210 case 0:
1211 re->level = 4;
1212 re->before.layout = 0;
1213 break;
1214 case 5:
1215 delta_parity = 1;
1216 re->level = 5;
1217 re->before.layout = ALGORITHM_PARITY_N;
1218 if (info->new_layout == UnSet)
1219 info->new_layout = map_name(r5layout, "default");
1220 break;
1221 case 6:
1222 delta_parity = 2;
1223 re->level = 6;
1224 re->before.layout = ALGORITHM_PARITY_N;
1225 if (info->new_layout == UnSet)
1226 info->new_layout = map_name(r6layout, "default");
1227 break;
1228 default:
1229 return "Impossible level change requested";
1230 }
1231 re->before.data_disks = info->array.raid_disks;
1232 /* determining 'after' layout happens outside this 'switch' */
1233 break;
1234
1235 case 4:
1236 info->array.layout = ALGORITHM_PARITY_N;
1237 case 5:
1238 switch (info->new_level) {
1239 case 0:
1240 delta_parity = -1;
1241 case 4:
1242 re->level = info->array.level;
1243 re->before.data_disks = info->array.raid_disks - 1;
1244 re->before.layout = info->array.layout;
1245 break;
1246 case 5:
1247 re->level = 5;
1248 re->before.data_disks = info->array.raid_disks - 1;
1249 re->before.layout = info->array.layout;
1250 break;
1251 case 6:
1252 delta_parity = 1;
1253 re->level = 6;
1254 re->before.data_disks = info->array.raid_disks - 1;
1255 switch (info->array.layout) {
1256 case ALGORITHM_LEFT_ASYMMETRIC:
1257 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1258 break;
1259 case ALGORITHM_RIGHT_ASYMMETRIC:
1260 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1261 break;
1262 case ALGORITHM_LEFT_SYMMETRIC:
1263 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1264 break;
1265 case ALGORITHM_RIGHT_SYMMETRIC:
1266 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1267 break;
1268 case ALGORITHM_PARITY_0:
1269 re->before.layout = ALGORITHM_PARITY_0_6;
1270 break;
1271 case ALGORITHM_PARITY_N:
1272 re->before.layout = ALGORITHM_PARITY_N_6;
1273 break;
1274 default:
1275 return "Cannot convert an array with this layout";
1276 }
1277 break;
1278 case 1:
1279 if (info->array.raid_disks != 2)
1280 return "Can only convert a 2-device array to RAID1";
1281 if (info->delta_disks != UnSet &&
1282 info->delta_disks != 0)
1283 return "Cannot set raid_disk when "
1284 "converting RAID5->RAID1";
1285 re->level = 1;
1286 info->new_chunk = 0;
1287 return NULL;
1288 default:
1289 return "Impossible level change requested";
1290 }
1291 break;
1292 case 6:
1293 switch (info->new_level) {
1294 case 4:
1295 case 5:
1296 delta_parity = -1;
1297 case 6:
1298 re->level = 6;
1299 re->before.data_disks = info->array.raid_disks - 2;
1300 re->before.layout = info->array.layout;
1301 break;
1302 default:
1303 return "Impossible level change requested";
1304 }
1305 break;
1306 }
1307
1308 /* If we reached here then it looks like a re-stripe is
1309 * happening. We have determined the intermediate level
1310 * and initial raid_disks/layout and stored these in 're'.
1311 *
1312 * We need to deduce the final layout that can be atomically
1313 * converted to the end state.
1314 */
1315 switch (info->new_level) {
1316 case 0:
1317 /* We can only get to RAID0 from RAID4 or RAID5
1318 * with appropriate layout and one extra device
1319 */
1320 if (re->level != 4 && re->level != 5)
1321 return "Cannot covert to RAID0 from this level";
1322
1323 switch (re->level) {
1324 case 4:
1325 re->before.layout = 0;
1326 re->after.layout = 0;
1327 break;
1328 case 5:
1329 re->after.layout = ALGORITHM_PARITY_N;
1330 break;
1331 }
1332 break;
1333
1334 case 4:
1335 /* We can only get to RAID4 from RAID5 */
1336 if (re->level != 4 && re->level != 5)
1337 return "Cannot convert to RAID4 from this level";
1338
1339 switch (re->level) {
1340 case 4:
1341 re->before.layout = 0;
1342 re->after.layout = 0;
1343 break;
1344 case 5:
1345 re->after.layout = ALGORITHM_PARITY_N;
1346 break;
1347 }
1348 break;
1349
1350 case 5:
1351 /* We get to RAID5 from RAID5 or RAID6 */
1352 if (re->level != 5 && re->level != 6)
1353 return "Cannot convert to RAID5 from this level";
1354
1355 switch (re->level) {
1356 case 5:
1357 if (info->new_layout == UnSet)
1358 re->after.layout = re->before.layout;
1359 else
1360 re->after.layout = info->new_layout;
1361 break;
1362 case 6:
1363 if (info->new_layout == UnSet)
1364 info->new_layout = re->before.layout;
1365
1366 /* after.layout needs to be raid6 version of new_layout */
1367 if (info->new_layout == ALGORITHM_PARITY_N)
1368 re->after.layout = ALGORITHM_PARITY_N;
1369 else {
1370 char layout[40];
1371 char *ls = map_num(r5layout, info->new_layout);
1372 int l;
1373 if (ls) {
1374 /* Current RAID6 layout has a RAID5
1375 * equivalent - good
1376 */
1377 strcat(strcpy(layout, ls), "-6");
1378 l = map_name(r6layout, layout);
1379 if (l == UnSet)
1380 return "Cannot find RAID6 layout"
1381 " to convert to";
1382 } else {
1383 /* Current RAID6 has no equivalent.
1384 * If it is already a '-6' layout we
1385 * can leave it unchanged, else we must
1386 * fail
1387 */
1388 ls = map_num(r6layout, info->new_layout);
1389 if (!ls ||
1390 strcmp(ls+strlen(ls)-2, "-6") != 0)
1391 return "Please specify new layout";
1392 l = info->new_layout;
1393 }
1394 re->after.layout = l;
1395 }
1396 }
1397 break;
1398
1399 case 6:
1400 /* We must already be at level 6 */
1401 if (re->level != 6)
1402 return "Impossible level change";
1403 if (info->new_layout == UnSet)
1404 re->after.layout = info->array.layout;
1405 else
1406 re->after.layout = info->new_layout;
1407 break;
1408 default:
1409 return "Impossible level change requested";
1410 }
1411 if (info->delta_disks == UnSet)
1412 info->delta_disks = delta_parity;
1413
1414 re->after.data_disks = (re->before.data_disks
1415 + info->delta_disks
1416 - delta_parity);
1417 switch (re->level) {
1418 case 6: re->parity = 2;
1419 break;
1420 case 4:
1421 case 5: re->parity = 1;
1422 break;
1423 default: re->parity = 0;
1424 break;
1425 }
1426 /* So we have a restripe operation, we need to calculate the number
1427 * of blocks per reshape operation.
1428 */
1429 re->new_size = info->component_size * re->before.data_disks;
1430 if (info->new_chunk == 0)
1431 info->new_chunk = info->array.chunk_size;
1432 if (re->after.data_disks == re->before.data_disks &&
1433 re->after.layout == re->before.layout &&
1434 info->new_chunk == info->array.chunk_size) {
1435 /* Nothing to change, can change level immediately. */
1436 re->level = info->new_level;
1437 re->backup_blocks = 0;
1438 return NULL;
1439 }
1440 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1441 /* chunk and layout changes make no difference */
1442 re->level = info->new_level;
1443 re->backup_blocks = 0;
1444 return NULL;
1445 }
1446
1447 if (re->after.data_disks == re->before.data_disks &&
1448 get_linux_version() < 2006032)
1449 return "in-place reshape is not safe before 2.6.32 - sorry.";
1450
1451 if (re->after.data_disks < re->before.data_disks &&
1452 get_linux_version() < 2006030)
1453 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1454
1455 re->backup_blocks = compute_backup_blocks(
1456 info->new_chunk, info->array.chunk_size,
1457 re->after.data_disks,
1458 re->before.data_disks);
1459 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1460
1461 re->new_size = info->component_size * re->after.data_disks;
1462 return NULL;
1463 }
1464
1465 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1466 char *text_version)
1467 {
1468 struct mdinfo *info;
1469 char *subarray;
1470 int ret_val = -1;
1471
1472 if ((st == NULL) || (sra == NULL))
1473 return ret_val;
1474
1475 if (text_version == NULL)
1476 text_version = sra->text_version;
1477 subarray = strchr(text_version+1, '/')+1;
1478 info = st->ss->container_content(st, subarray);
1479 if (info) {
1480 unsigned long long current_size = 0;
1481 unsigned long long new_size =
1482 info->custom_array_size/2;
1483
1484 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1485 new_size > current_size) {
1486 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1487 < 0)
1488 dprintf("Error: Cannot set array size");
1489 else {
1490 ret_val = 0;
1491 dprintf("Array size changed");
1492 }
1493 dprintf(" from %llu to %llu.\n",
1494 current_size, new_size);
1495 }
1496 sysfs_free(info);
1497 } else
1498 dprintf("Error: set_array_size(): info pointer in NULL\n");
1499
1500 return ret_val;
1501 }
1502
1503 static int reshape_array(char *container, int fd, char *devname,
1504 struct supertype *st, struct mdinfo *info,
1505 int force, struct mddev_dev *devlist,
1506 unsigned long long data_offset,
1507 char *backup_file, int verbose, int forked,
1508 int restart, int freeze_reshape);
1509 static int reshape_container(char *container, char *devname,
1510 int mdfd,
1511 struct supertype *st,
1512 struct mdinfo *info,
1513 int force,
1514 char *backup_file, int verbose,
1515 int forked, int restart, int freeze_reshape);
1516
1517 int Grow_reshape(char *devname, int fd,
1518 struct mddev_dev *devlist,
1519 unsigned long long data_offset,
1520 struct context *c, struct shape *s)
1521 {
1522 /* Make some changes in the shape of an array.
1523 * The kernel must support the change.
1524 *
1525 * There are three different changes. Each can trigger
1526 * a resync or recovery so we freeze that until we have
1527 * requested everything (if kernel supports freezing - 2.6.30).
1528 * The steps are:
1529 * - change size (i.e. component_size)
1530 * - change level
1531 * - change layout/chunksize/ndisks
1532 *
1533 * The last can require a reshape. It is different on different
1534 * levels so we need to check the level before actioning it.
1535 * Some times the level change needs to be requested after the
1536 * reshape (e.g. raid6->raid5, raid5->raid0)
1537 *
1538 */
1539 struct mdu_array_info_s array;
1540 int rv = 0;
1541 struct supertype *st;
1542 char *subarray = NULL;
1543
1544 int frozen;
1545 int changed = 0;
1546 char *container = NULL;
1547 int cfd = -1;
1548
1549 struct mddev_dev *dv;
1550 int added_disks;
1551
1552 struct mdinfo info;
1553 struct mdinfo *sra;
1554
1555 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1556 pr_err("%s is not an active md array - aborting\n",
1557 devname);
1558 return 1;
1559 }
1560 if (data_offset != INVALID_SECTORS && array.level != 10
1561 && (array.level < 4 || array.level > 6)) {
1562 pr_err("--grow --data-offset not yet supported\n");
1563 return 1;
1564 }
1565
1566 if (s->size > 0 &&
1567 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1568 pr_err("cannot change component size at the same time "
1569 "as other changes.\n"
1570 " Change size first, then check data is intact before "
1571 "making other changes.\n");
1572 return 1;
1573 }
1574
1575 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1576 get_linux_version() < 2006032 &&
1577 !check_env("MDADM_FORCE_FEWER")) {
1578 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1579 " Please use a newer kernel\n");
1580 return 1;
1581 }
1582
1583 st = super_by_fd(fd, &subarray);
1584 if (!st) {
1585 pr_err("Unable to determine metadata format for %s\n", devname);
1586 return 1;
1587 }
1588 if (s->raiddisks > st->max_devs) {
1589 pr_err("Cannot increase raid-disks on this array"
1590 " beyond %d\n", st->max_devs);
1591 return 1;
1592 }
1593
1594 /* in the external case we need to check that the requested reshape is
1595 * supported, and perform an initial check that the container holds the
1596 * pre-requisite spare devices (mdmon owns final validation)
1597 */
1598 if (st->ss->external) {
1599 int rv;
1600
1601 if (subarray) {
1602 container = st->container_devnm;
1603 cfd = open_dev_excl(st->container_devnm);
1604 } else {
1605 container = st->devnm;
1606 close(fd);
1607 cfd = open_dev_excl(st->devnm);
1608 fd = cfd;
1609 }
1610 if (cfd < 0) {
1611 pr_err("Unable to open container for %s\n",
1612 devname);
1613 free(subarray);
1614 return 1;
1615 }
1616
1617 rv = st->ss->load_container(st, cfd, NULL);
1618
1619 if (rv) {
1620 pr_err("Cannot read superblock for %s\n",
1621 devname);
1622 free(subarray);
1623 return 1;
1624 }
1625
1626 /* check if operation is supported for metadata handler */
1627 if (st->ss->container_content) {
1628 struct mdinfo *cc = NULL;
1629 struct mdinfo *content = NULL;
1630
1631 cc = st->ss->container_content(st, subarray);
1632 for (content = cc; content ; content = content->next) {
1633 int allow_reshape = 1;
1634
1635 /* check if reshape is allowed based on metadata
1636 * indications stored in content.array.status
1637 */
1638 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1639 allow_reshape = 0;
1640 if (content->array.state
1641 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1642 allow_reshape = 0;
1643 if (!allow_reshape) {
1644 pr_err("cannot reshape arrays in"
1645 " container with unsupported"
1646 " metadata: %s(%s)\n",
1647 devname, container);
1648 sysfs_free(cc);
1649 free(subarray);
1650 return 1;
1651 }
1652 }
1653 sysfs_free(cc);
1654 }
1655 if (mdmon_running(container))
1656 st->update_tail = &st->updates;
1657 }
1658
1659 added_disks = 0;
1660 for (dv = devlist; dv; dv = dv->next)
1661 added_disks++;
1662 if (s->raiddisks > array.raid_disks &&
1663 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1664 !c->force) {
1665 pr_err("Need %d spare%s to avoid degraded array,"
1666 " and only have %d.\n"
1667 " Use --force to over-ride this check.\n",
1668 s->raiddisks - array.raid_disks,
1669 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1670 array.spare_disks + added_disks);
1671 return 1;
1672 }
1673
1674 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1675 | GET_STATE | GET_VERSION);
1676 if (sra) {
1677 if (st->ss->external && subarray == NULL) {
1678 array.level = LEVEL_CONTAINER;
1679 sra->array.level = LEVEL_CONTAINER;
1680 }
1681 } else {
1682 pr_err("failed to read sysfs parameters for %s\n",
1683 devname);
1684 return 1;
1685 }
1686 frozen = freeze(st);
1687 if (frozen < -1) {
1688 /* freeze() already spewed the reason */
1689 sysfs_free(sra);
1690 return 1;
1691 } else if (frozen < 0) {
1692 pr_err("%s is performing resync/recovery and cannot"
1693 " be reshaped\n", devname);
1694 sysfs_free(sra);
1695 return 1;
1696 }
1697
1698 /* ========= set size =============== */
1699 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1700 unsigned long long orig_size = get_component_size(fd)/2;
1701 unsigned long long min_csize;
1702 struct mdinfo *mdi;
1703 int raid0_takeover = 0;
1704
1705 if (orig_size == 0)
1706 orig_size = (unsigned) array.size;
1707
1708 if (orig_size == 0) {
1709 pr_err("Cannot set device size in this type of array.\n");
1710 rv = 1;
1711 goto release;
1712 }
1713
1714 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1715 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1716 rv = 1;
1717 goto release;
1718 }
1719 sync_metadata(st);
1720 if (st->ss->external) {
1721 /* metadata can have size limitation
1722 * update size value according to metadata information
1723 */
1724 struct mdinfo *sizeinfo =
1725 st->ss->container_content(st, subarray);
1726 if (sizeinfo) {
1727 unsigned long long new_size =
1728 sizeinfo->custom_array_size/2;
1729 int data_disks = get_data_disks(
1730 sizeinfo->array.level,
1731 sizeinfo->array.layout,
1732 sizeinfo->array.raid_disks);
1733 new_size /= data_disks;
1734 dprintf("Metadata size correction from %llu to "
1735 "%llu (%llu)\n", orig_size, new_size,
1736 new_size * data_disks);
1737 s->size = new_size;
1738 sysfs_free(sizeinfo);
1739 }
1740 }
1741
1742 /* Update the size of each member device in case
1743 * they have been resized. This will never reduce
1744 * below the current used-size. The "size" attribute
1745 * understands '0' to mean 'max'.
1746 */
1747 min_csize = 0;
1748 rv = 0;
1749 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1750 if (sysfs_set_num(sra, mdi, "size",
1751 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1752 /* Probably kernel refusing to let us
1753 * reduce the size - not an error.
1754 */
1755 break;
1756 }
1757 if (array.not_persistent == 0 &&
1758 array.major_version == 0 &&
1759 get_linux_version() < 3001000) {
1760 /* Dangerous to allow size to exceed 2TB */
1761 unsigned long long csize;
1762 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1763 if (csize >= 2ULL*1024*1024*1024)
1764 csize = 2ULL*1024*1024*1024;
1765 if ((min_csize == 0 || (min_csize
1766 > csize)))
1767 min_csize = csize;
1768 }
1769 }
1770 }
1771 if (rv) {
1772 pr_err("Cannot set size on "
1773 "array members.\n");
1774 goto size_change_error;
1775 }
1776 if (min_csize && s->size > min_csize) {
1777 pr_err("Cannot safely make this array "
1778 "use more than 2TB per device on this kernel.\n");
1779 rv = 1;
1780 goto size_change_error;
1781 }
1782 if (min_csize && s->size == MAX_SIZE) {
1783 /* Don't let the kernel choose a size - it will get
1784 * it wrong
1785 */
1786 pr_err("Limited v0.90 array to "
1787 "2TB per device\n");
1788 s->size = min_csize;
1789 }
1790 if (st->ss->external) {
1791 if (sra->array.level == 0) {
1792 rv = sysfs_set_str(sra, NULL, "level",
1793 "raid5");
1794 if (!rv) {
1795 raid0_takeover = 1;
1796 /* get array parametes after takeover
1797 * to chane one parameter at time only
1798 */
1799 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1800 }
1801 }
1802 /* make sure mdmon is
1803 * aware of the new level */
1804 if (!mdmon_running(st->container_devnm))
1805 start_mdmon(st->container_devnm);
1806 ping_monitor(container);
1807 if (mdmon_running(st->container_devnm) &&
1808 st->update_tail == NULL)
1809 st->update_tail = &st->updates;
1810 }
1811
1812 if (s->size == MAX_SIZE)
1813 s->size = 0;
1814 array.size = s->size;
1815 if ((unsigned)array.size != s->size) {
1816 /* got truncated to 32bit, write to
1817 * component_size instead
1818 */
1819 if (sra)
1820 rv = sysfs_set_num(sra, NULL,
1821 "component_size", s->size);
1822 else
1823 rv = -1;
1824 } else {
1825 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1826
1827 /* manage array size when it is managed externally
1828 */
1829 if ((rv == 0) && st->ss->external)
1830 rv = set_array_size(st, sra, sra->text_version);
1831 }
1832
1833 if (raid0_takeover) {
1834 /* do not recync non-existing parity,
1835 * we will drop it anyway
1836 */
1837 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1838 /* go back to raid0, drop parity disk
1839 */
1840 sysfs_set_str(sra, NULL, "level", "raid0");
1841 ioctl(fd, GET_ARRAY_INFO, &array);
1842 }
1843
1844 size_change_error:
1845 if (rv != 0) {
1846 int err = errno;
1847
1848 /* restore metadata */
1849 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1850 UnSet, NULL, devname,
1851 ROLLBACK_METADATA_CHANGES,
1852 c->verbose) == 0)
1853 sync_metadata(st);
1854 pr_err("Cannot set device size for %s: %s\n",
1855 devname, strerror(err));
1856 if (err == EBUSY &&
1857 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1858 cont_err("Bitmap must be removed before size can be changed\n");
1859 rv = 1;
1860 goto release;
1861 }
1862 if (s->assume_clean) {
1863 /* This will fail on kernels older than 3.0 unless
1864 * a backport has been arranged.
1865 */
1866 if (sra == NULL ||
1867 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1868 pr_err("--assume-clean not supported with --grow on this kernel\n");
1869 }
1870 ioctl(fd, GET_ARRAY_INFO, &array);
1871 s->size = get_component_size(fd)/2;
1872 if (s->size == 0)
1873 s->size = array.size;
1874 if (c->verbose >= 0) {
1875 if (s->size == orig_size)
1876 pr_err("component size of %s "
1877 "unchanged at %lluK\n",
1878 devname, s->size);
1879 else
1880 pr_err("component size of %s "
1881 "has been set to %lluK\n",
1882 devname, s->size);
1883 }
1884 changed = 1;
1885 } else if (array.level != LEVEL_CONTAINER) {
1886 s->size = get_component_size(fd)/2;
1887 if (s->size == 0)
1888 s->size = array.size;
1889 }
1890
1891 /* See if there is anything else to do */
1892 if ((s->level == UnSet || s->level == array.level) &&
1893 (s->layout_str == NULL) &&
1894 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1895 data_offset == INVALID_SECTORS &&
1896 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1897 /* Nothing more to do */
1898 if (!changed && c->verbose >= 0)
1899 pr_err("%s: no change requested\n",
1900 devname);
1901 goto release;
1902 }
1903
1904 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1905 * current implementation assumes that following conditions must be met:
1906 * - RAID10:
1907 * - far_copies == 1
1908 * - near_copies == 2
1909 */
1910 if ((s->level == 0 && array.level == 10 && sra &&
1911 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1912 (s->level == 0 && array.level == 1 && sra)) {
1913 int err;
1914 err = remove_disks_for_takeover(st, sra, array.layout);
1915 if (err) {
1916 dprintf(Name": Array cannot be reshaped\n");
1917 if (cfd > -1)
1918 close(cfd);
1919 rv = 1;
1920 goto release;
1921 }
1922 /* Make sure mdmon has seen the device removal
1923 * and updated metadata before we continue with
1924 * level change
1925 */
1926 if (container)
1927 ping_monitor(container);
1928 }
1929
1930 memset(&info, 0, sizeof(info));
1931 info.array = array;
1932 sysfs_init(&info, fd, NULL);
1933 strcpy(info.text_version, sra->text_version);
1934 info.component_size = s->size*2;
1935 info.new_level = s->level;
1936 info.new_chunk = s->chunk * 1024;
1937 if (info.array.level == LEVEL_CONTAINER) {
1938 info.delta_disks = UnSet;
1939 info.array.raid_disks = s->raiddisks;
1940 } else if (s->raiddisks)
1941 info.delta_disks = s->raiddisks - info.array.raid_disks;
1942 else
1943 info.delta_disks = UnSet;
1944 if (s->layout_str == NULL) {
1945 info.new_layout = UnSet;
1946 if (info.array.level == 6 &&
1947 (info.new_level == 6 || info.new_level == UnSet) &&
1948 info.array.layout >= 16) {
1949 pr_err("%s has a non-standard layout. If you"
1950 " wish to preserve this\n", devname);
1951 cont_err("during the reshape, please specify"
1952 " --layout=preserve\n");
1953 cont_err("If you want to change it, specify a"
1954 " layout or use --layout=normalise\n");
1955 rv = 1;
1956 goto release;
1957 }
1958 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1959 strcmp(s->layout_str, "normalize") == 0) {
1960 /* If we have a -6 RAID6 layout, remove the '-6'. */
1961 info.new_layout = UnSet;
1962 if (info.array.level == 6 && info.new_level == UnSet) {
1963 char l[40], *h;
1964 strcpy(l, map_num(r6layout, info.array.layout));
1965 h = strrchr(l, '-');
1966 if (h && strcmp(h, "-6") == 0) {
1967 *h = 0;
1968 info.new_layout = map_name(r6layout, l);
1969 }
1970 } else {
1971 pr_err("%s is only meaningful when reshaping"
1972 " a RAID6 array.\n", s->layout_str);
1973 rv = 1;
1974 goto release;
1975 }
1976 } else if (strcmp(s->layout_str, "preserve") == 0) {
1977 /* This means that a non-standard RAID6 layout
1978 * is OK.
1979 * In particular:
1980 * - When reshape a RAID6 (e.g. adding a device)
1981 * which is in a non-standard layout, it is OK
1982 * to preserve that layout.
1983 * - When converting a RAID5 to RAID6, leave it in
1984 * the XXX-6 layout, don't re-layout.
1985 */
1986 if (info.array.level == 6 && info.new_level == UnSet)
1987 info.new_layout = info.array.layout;
1988 else if (info.array.level == 5 && info.new_level == 6) {
1989 char l[40];
1990 strcpy(l, map_num(r5layout, info.array.layout));
1991 strcat(l, "-6");
1992 info.new_layout = map_name(r6layout, l);
1993 } else {
1994 pr_err("%s in only meaningful when reshaping"
1995 " to RAID6\n", s->layout_str);
1996 rv = 1;
1997 goto release;
1998 }
1999 } else {
2000 int l = info.new_level;
2001 if (l == UnSet)
2002 l = info.array.level;
2003 switch (l) {
2004 case 5:
2005 info.new_layout = map_name(r5layout, s->layout_str);
2006 break;
2007 case 6:
2008 info.new_layout = map_name(r6layout, s->layout_str);
2009 break;
2010 case 10:
2011 info.new_layout = parse_layout_10(s->layout_str);
2012 break;
2013 case LEVEL_FAULTY:
2014 info.new_layout = parse_layout_faulty(s->layout_str);
2015 break;
2016 default:
2017 pr_err("layout not meaningful"
2018 " with this level\n");
2019 rv = 1;
2020 goto release;
2021 }
2022 if (info.new_layout == UnSet) {
2023 pr_err("layout %s not understood"
2024 " for this level\n",
2025 s->layout_str);
2026 rv = 1;
2027 goto release;
2028 }
2029 }
2030
2031 if (array.level == LEVEL_FAULTY) {
2032 if (s->level != UnSet && s->level != array.level) {
2033 pr_err("cannot change level of Faulty device\n");
2034 rv =1 ;
2035 }
2036 if (s->chunk) {
2037 pr_err("cannot set chunksize of Faulty device\n");
2038 rv =1 ;
2039 }
2040 if (s->raiddisks && s->raiddisks != 1) {
2041 pr_err("cannot set raid_disks of Faulty device\n");
2042 rv =1 ;
2043 }
2044 if (s->layout_str) {
2045 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2046 dprintf("Cannot get array information.\n");
2047 goto release;
2048 }
2049 array.layout = info.new_layout;
2050 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2051 pr_err("failed to set new layout\n");
2052 rv = 1;
2053 } else if (c->verbose >= 0)
2054 printf("layout for %s set to %d\n",
2055 devname, array.layout);
2056 }
2057 } else if (array.level == LEVEL_CONTAINER) {
2058 /* This change is to be applied to every array in the
2059 * container. This is only needed when the metadata imposes
2060 * restraints of the various arrays in the container.
2061 * Currently we only know that IMSM requires all arrays
2062 * to have the same number of devices so changing the
2063 * number of devices (On-Line Capacity Expansion) must be
2064 * performed at the level of the container
2065 */
2066 rv = reshape_container(container, devname, -1, st, &info,
2067 c->force, c->backup_file, c->verbose, 0, 0, 0);
2068 frozen = 0;
2069 } else {
2070 /* get spare devices from external metadata
2071 */
2072 if (st->ss->external) {
2073 struct mdinfo *info2;
2074
2075 info2 = st->ss->container_content(st, subarray);
2076 if (info2) {
2077 info.array.spare_disks =
2078 info2->array.spare_disks;
2079 sysfs_free(info2);
2080 }
2081 }
2082
2083 /* Impose these changes on a single array. First
2084 * check that the metadata is OK with the change. */
2085
2086 if (reshape_super(st, 0, info.new_level,
2087 info.new_layout, info.new_chunk,
2088 info.array.raid_disks, info.delta_disks,
2089 c->backup_file, devname, APPLY_METADATA_CHANGES,
2090 c->verbose)) {
2091 rv = 1;
2092 goto release;
2093 }
2094 sync_metadata(st);
2095 rv = reshape_array(container, fd, devname, st, &info, c->force,
2096 devlist, data_offset, c->backup_file, c->verbose,
2097 0, 0, 0);
2098 frozen = 0;
2099 }
2100 release:
2101 sysfs_free(sra);
2102 if (frozen > 0)
2103 unfreeze(st);
2104 return rv;
2105 }
2106
2107 /* verify_reshape_position()
2108 * Function checks if reshape position in metadata is not farther
2109 * than position in md.
2110 * Return value:
2111 * 0 : not valid sysfs entry
2112 * it can be caused by not started reshape, it should be started
2113 * by reshape array or raid0 array is before takeover
2114 * -1 : error, reshape position is obviously wrong
2115 * 1 : success, reshape progress correct or updated
2116 */
2117 static int verify_reshape_position(struct mdinfo *info, int level)
2118 {
2119 int ret_val = 0;
2120 char buf[40];
2121 int rv;
2122
2123 /* read sync_max, failure can mean raid0 array */
2124 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2125
2126 if (rv > 0) {
2127 char *ep;
2128 unsigned long long position = strtoull(buf, &ep, 0);
2129
2130 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2131 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2132 position *= get_data_disks(level,
2133 info->new_layout,
2134 info->array.raid_disks);
2135 if (info->reshape_progress < position) {
2136 dprintf("Corrected reshape progress (%llu) to "
2137 "md position (%llu)\n",
2138 info->reshape_progress, position);
2139 info->reshape_progress = position;
2140 ret_val = 1;
2141 } else if (info->reshape_progress > position) {
2142 pr_err("Fatal error: array "
2143 "reshape was not properly frozen "
2144 "(expected reshape position is %llu, "
2145 "but reshape progress is %llu.\n",
2146 position, info->reshape_progress);
2147 ret_val = -1;
2148 } else {
2149 dprintf("Reshape position in md and metadata "
2150 "are the same;");
2151 ret_val = 1;
2152 }
2153 }
2154 } else if (rv == 0) {
2155 /* for valid sysfs entry, 0-length content
2156 * should be indicated as error
2157 */
2158 ret_val = -1;
2159 }
2160
2161 return ret_val;
2162 }
2163
2164 static unsigned long long choose_offset(unsigned long long lo,
2165 unsigned long long hi,
2166 unsigned long long min,
2167 unsigned long long max)
2168 {
2169 /* Choose a new offset between hi and lo.
2170 * It must be between min and max, but
2171 * we would prefer something near the middle of hi/lo, and also
2172 * prefer to be aligned to a big power of 2.
2173 *
2174 * So we start with the middle, then for each bit,
2175 * starting at '1' and increasing, if it is set, we either
2176 * add it or subtract it if possible, preferring the option
2177 * which is furthest from the boundary.
2178 *
2179 * We stop once we get a 1MB alignment. As units are in sectors,
2180 * 1MB = 2*1024 sectors.
2181 */
2182 unsigned long long choice = (lo + hi) / 2;
2183 unsigned long long bit = 1;
2184
2185 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2186 unsigned long long bigger, smaller;
2187 if (! (bit & choice))
2188 continue;
2189 bigger = choice + bit;
2190 smaller = choice - bit;
2191 if (bigger > max && smaller < min)
2192 break;
2193 if (bigger > max)
2194 choice = smaller;
2195 else if (smaller < min)
2196 choice = bigger;
2197 else if (hi - bigger > smaller - lo)
2198 choice = bigger;
2199 else
2200 choice = smaller;
2201 }
2202 return choice;
2203 }
2204
2205 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2206 char *devname, int delta_disks,
2207 unsigned long long data_offset,
2208 unsigned long long min,
2209 int can_fallback)
2210 {
2211 struct mdinfo *sd;
2212 int dir = 0;
2213 int err = 0;
2214 unsigned long long before, after;
2215
2216 /* Need to find min space before and after so same is used
2217 * on all devices
2218 */
2219 before = UINT64_MAX;
2220 after = UINT64_MAX;
2221 for (sd = sra->devs; sd; sd = sd->next) {
2222 char *dn;
2223 int dfd;
2224 int rv;
2225 struct supertype *st2;
2226 struct mdinfo info2;
2227
2228 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2229 continue;
2230 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2231 dfd = dev_open(dn, O_RDONLY);
2232 if (dfd < 0) {
2233 pr_err("%s: cannot open component %s\n",
2234 devname, dn ? dn : "-unknown-");
2235 goto release;
2236 }
2237 st2 = dup_super(st);
2238 rv = st2->ss->load_super(st2,dfd, NULL);
2239 close(dfd);
2240 if (rv) {
2241 free(st2);
2242 pr_err("%s: cannot get superblock from %s\n",
2243 devname, dn);
2244 goto release;
2245 }
2246 st2->ss->getinfo_super(st2, &info2, NULL);
2247 st2->ss->free_super(st2);
2248 free(st2);
2249 if (info2.space_before == 0 &&
2250 info2.space_after == 0) {
2251 /* Metadata doesn't support data_offset changes */
2252 return 1;
2253 }
2254 if (before > info2.space_before)
2255 before = info2.space_before;
2256 if (after > info2.space_after)
2257 after = info2.space_after;
2258
2259 if (data_offset != INVALID_SECTORS) {
2260 if (dir == 0) {
2261 if (info2.data_offset == data_offset) {
2262 pr_err("%s: already has that data_offset\n",
2263 dn);
2264 goto release;
2265 }
2266 if (data_offset < info2.data_offset)
2267 dir = -1;
2268 else
2269 dir = 1;
2270 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2271 (data_offset >= info2.data_offset && dir == -1)) {
2272 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2273 dn);
2274 goto release;
2275 }
2276 }
2277 }
2278 if (before == UINT64_MAX)
2279 /* impossible really, there must be no devices */
2280 return 1;
2281
2282 for (sd = sra->devs; sd; sd = sd->next) {
2283 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2284 unsigned long long new_data_offset;
2285
2286 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2287 continue;
2288 if (delta_disks < 0) {
2289 /* Don't need any space as array is shrinking
2290 * just move data_offset up by min
2291 */
2292 if (data_offset == INVALID_SECTORS)
2293 new_data_offset = sd->data_offset + min;
2294 else {
2295 if (data_offset < sd->data_offset + min) {
2296 pr_err("--data-offset too small for %s\n",
2297 dn);
2298 goto release;
2299 }
2300 new_data_offset = data_offset;
2301 }
2302 } else if (delta_disks > 0) {
2303 /* need space before */
2304 if (before < min) {
2305 if (can_fallback)
2306 goto fallback;
2307 pr_err("Insufficient head-space for reshape on %s\n",
2308 dn);
2309 goto release;
2310 }
2311 if (data_offset == INVALID_SECTORS)
2312 new_data_offset = sd->data_offset - min;
2313 else {
2314 if (data_offset > sd->data_offset - min) {
2315 pr_err("--data-offset too large for %s\n",
2316 dn);
2317 goto release;
2318 }
2319 new_data_offset = data_offset;
2320 }
2321 } else {
2322 if (dir == 0) {
2323 /* can move up or down. If 'data_offset'
2324 * was set we would have already decided,
2325 * so just choose direction with most space.
2326 */
2327 if (before > after)
2328 dir = -1;
2329 else
2330 dir = 1;
2331 }
2332 sysfs_set_str(sra, NULL, "reshape_direction",
2333 dir == 1 ? "backwards" : "forwards");
2334 if (dir > 0) {
2335 /* Increase data offset */
2336 if (after < min) {
2337 if (can_fallback)
2338 goto fallback;
2339 pr_err("Insufficient tail-space for reshape on %s\n",
2340 dn);
2341 goto release;
2342 }
2343 if (data_offset != INVALID_SECTORS &&
2344 data_offset < sd->data_offset + min) {
2345 pr_err("--data-offset too small on %s\n",
2346 dn);
2347 goto release;
2348 }
2349 if (data_offset != INVALID_SECTORS)
2350 new_data_offset = data_offset;
2351 else
2352 new_data_offset = choose_offset(sd->data_offset,
2353 sd->data_offset + after,
2354 sd->data_offset + min,
2355 sd->data_offset + after);
2356 } else {
2357 /* Decrease data offset */
2358 if (before < min) {
2359 if (can_fallback)
2360 goto fallback;
2361 pr_err("insufficient head-room on %s\n",
2362 dn);
2363 goto release;
2364 }
2365 if (data_offset != INVALID_SECTORS &&
2366 data_offset < sd->data_offset - min) {
2367 pr_err("--data-offset too small on %s\n",
2368 dn);
2369 goto release;
2370 }
2371 if (data_offset != INVALID_SECTORS)
2372 new_data_offset = data_offset;
2373 else
2374 new_data_offset = choose_offset(sd->data_offset - before,
2375 sd->data_offset,
2376 sd->data_offset - before,
2377 sd->data_offset - min);
2378 }
2379 }
2380 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2381 if (err < 0 && errno == E2BIG) {
2382 /* try again after increasing data size to max */
2383 err = sysfs_set_num(sra, sd, "size", 0);
2384 if (err < 0 && errno == EINVAL &&
2385 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2386 /* some kernels have a bug where you cannot
2387 * use '0' on spare devices. */
2388 sysfs_set_num(sra, sd, "size",
2389 (sra->component_size + after)/2);
2390 }
2391 err = sysfs_set_num(sra, sd, "new_offset",
2392 new_data_offset);
2393 }
2394 if (err < 0) {
2395 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2396 pr_err("data-offset is too big for %s\n",
2397 dn);
2398 goto release;
2399 }
2400 if (sd == sra->devs &&
2401 (errno == ENOENT || errno == E2BIG))
2402 /* Early kernel, no 'new_offset' file,
2403 * or kernel doesn't like us.
2404 * For RAID5/6 this is not fatal
2405 */
2406 return 1;
2407 pr_err("Cannot set new_offset for %s\n",
2408 dn);
2409 break;
2410 }
2411 }
2412 return err;
2413 release:
2414 return -1;
2415 fallback:
2416 /* Just use a backup file */
2417 return 1;
2418 }
2419
2420 static int raid10_reshape(char *container, int fd, char *devname,
2421 struct supertype *st, struct mdinfo *info,
2422 struct reshape *reshape,
2423 unsigned long long data_offset,
2424 int force, int verbose)
2425 {
2426 /* Changing raid_disks, layout, chunksize or possibly
2427 * just data_offset for a RAID10.
2428 * We must always change data_offset. We change by at least
2429 * ->min_offset_change which is the largest of the old and new
2430 * chunk sizes.
2431 * If raid_disks is increasing, then data_offset must decrease
2432 * by at least this copy size.
2433 * If raid_disks is unchanged, data_offset must increase or
2434 * decrease by at least min_offset_change but preferably by much more.
2435 * We choose half of the available space.
2436 * If raid_disks is decreasing, data_offset must increase by
2437 * at least min_offset_change. To allow of this, component_size
2438 * must be decreased by the same amount.
2439 *
2440 * So we calculate the required minimum and direction, possibly
2441 * reduce the component_size, then iterate through the devices
2442 * and set the new_data_offset.
2443 * If that all works, we set chunk_size, layout, raid_disks, and start
2444 * 'reshape'
2445 */
2446 struct mdinfo *sra;
2447 unsigned long long min;
2448 int err = 0;
2449
2450 sra = sysfs_read(fd, NULL,
2451 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2452 );
2453 if (!sra) {
2454 pr_err("%s: Cannot get array details from sysfs\n",
2455 devname);
2456 goto release;
2457 }
2458 min = reshape->min_offset_change;
2459
2460 if (info->delta_disks)
2461 sysfs_set_str(sra, NULL, "reshape_direction",
2462 info->delta_disks < 0 ? "backwards" : "forwards");
2463 if (info->delta_disks < 0 &&
2464 info->space_after < min) {
2465 int rv = sysfs_set_num(sra, NULL, "component_size",
2466 (sra->component_size -
2467 min)/2);
2468 if (rv) {
2469 pr_err("cannot reduce component size\n");
2470 goto release;
2471 }
2472 }
2473 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2474 min, 0);
2475 if (err == 1) {
2476 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2477 cont_err("supported on this kernel\n");
2478 err = -1;
2479 }
2480 if (err < 0)
2481 goto release;
2482
2483 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2484 err = errno;
2485 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2486 err = errno;
2487 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2488 info->array.raid_disks + info->delta_disks) < 0)
2489 err = errno;
2490 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2491 err = errno;
2492 if (err) {
2493 pr_err("Cannot set array shape for %s\n",
2494 devname);
2495 if (err == EBUSY &&
2496 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2497 cont_err(" Bitmap must be removed before"
2498 " shape can be changed\n");
2499 goto release;
2500 }
2501 sysfs_free(sra);
2502 return 0;
2503 release:
2504 sysfs_free(sra);
2505 return 1;
2506 }
2507
2508 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2509 {
2510 struct mdinfo *sra, *sd;
2511 /* Initialisation to silence compiler warning */
2512 unsigned long long min_space_before = 0, min_space_after = 0;
2513 int first = 1;
2514
2515 sra = sysfs_read(fd, NULL, GET_DEVS);
2516 if (!sra)
2517 return;
2518 for (sd = sra->devs; sd; sd = sd->next) {
2519 char *dn;
2520 int dfd;
2521 struct supertype *st2;
2522 struct mdinfo info2;
2523
2524 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2525 continue;
2526 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2527 dfd = dev_open(dn, O_RDONLY);
2528 if (dfd < 0)
2529 break;
2530 st2 = dup_super(st);
2531 if (st2->ss->load_super(st2,dfd, NULL)) {
2532 close(dfd);
2533 free(st2);
2534 break;
2535 }
2536 close(dfd);
2537 st2->ss->getinfo_super(st2, &info2, NULL);
2538 st2->ss->free_super(st2);
2539 free(st2);
2540 if (first ||
2541 min_space_before > info2.space_before)
2542 min_space_before = info2.space_before;
2543 if (first ||
2544 min_space_after > info2.space_after)
2545 min_space_after = info2.space_after;
2546 first = 0;
2547 }
2548 if (sd == NULL && !first) {
2549 info->space_after = min_space_after;
2550 info->space_before = min_space_before;
2551 }
2552 sysfs_free(sra);
2553 }
2554
2555 static void update_cache_size(char *container, struct mdinfo *sra,
2556 struct mdinfo *info,
2557 int disks, unsigned long long blocks)
2558 {
2559 /* Check that the internal stripe cache is
2560 * large enough, or it won't work.
2561 * It must hold at least 4 stripes of the larger
2562 * chunk size
2563 */
2564 unsigned long cache;
2565 cache = max(info->array.chunk_size, info->new_chunk);
2566 cache *= 4; /* 4 stripes minimum */
2567 cache /= 512; /* convert to sectors */
2568 /* make sure there is room for 'blocks' with a bit to spare */
2569 if (cache < 16 + blocks / disks)
2570 cache = 16 + blocks / disks;
2571 cache /= (4096/512); /* Covert from sectors to pages */
2572
2573 if (sra->cache_size < cache)
2574 subarray_set_num(container, sra, "stripe_cache_size",
2575 cache+1);
2576 }
2577
2578 static int impose_reshape(struct mdinfo *sra,
2579 struct mdinfo *info,
2580 struct supertype *st,
2581 int fd,
2582 int restart,
2583 char *devname, char *container,
2584 struct reshape *reshape)
2585 {
2586 struct mdu_array_info_s array;
2587
2588 sra->new_chunk = info->new_chunk;
2589
2590 if (restart) {
2591 /* for external metadata checkpoint saved by mdmon can be lost
2592 * or missed /due to e.g. crash/. Check if md is not during
2593 * restart farther than metadata points to.
2594 * If so, this means metadata information is obsolete.
2595 */
2596 if (st->ss->external)
2597 verify_reshape_position(info, reshape->level);
2598 sra->reshape_progress = info->reshape_progress;
2599 } else {
2600 sra->reshape_progress = 0;
2601 if (reshape->after.data_disks < reshape->before.data_disks)
2602 /* start from the end of the new array */
2603 sra->reshape_progress = (sra->component_size
2604 * reshape->after.data_disks);
2605 }
2606
2607 ioctl(fd, GET_ARRAY_INFO, &array);
2608 if (info->array.chunk_size == info->new_chunk &&
2609 reshape->before.layout == reshape->after.layout &&
2610 st->ss->external == 0) {
2611 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2612 array.raid_disks = reshape->after.data_disks + reshape->parity;
2613 if (!restart &&
2614 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2615 int err = errno;
2616
2617 pr_err("Cannot set device shape for %s: %s\n",
2618 devname, strerror(errno));
2619
2620 if (err == EBUSY &&
2621 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2622 cont_err("Bitmap must be removed before"
2623 " shape can be changed\n");
2624
2625 goto release;
2626 }
2627 } else if (!restart) {
2628 /* set them all just in case some old 'new_*' value
2629 * persists from some earlier problem.
2630 */
2631 int err = 0;
2632 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2633 err = errno;
2634 if (!err && sysfs_set_num(sra, NULL, "layout",
2635 reshape->after.layout) < 0)
2636 err = errno;
2637 if (!err && subarray_set_num(container, sra, "raid_disks",
2638 reshape->after.data_disks +
2639 reshape->parity) < 0)
2640 err = errno;
2641 if (err) {
2642 pr_err("Cannot set device shape for %s\n",
2643 devname);
2644
2645 if (err == EBUSY &&
2646 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2647 cont_err("Bitmap must be removed before"
2648 " shape can be changed\n");
2649 goto release;
2650 }
2651 }
2652 return 0;
2653 release:
2654 return -1;
2655 }
2656
2657 static int impose_level(int fd, int level, char *devname, int verbose)
2658 {
2659 char *c;
2660 struct mdu_array_info_s array;
2661 struct mdinfo info;
2662 sysfs_init(&info, fd, NULL);
2663
2664 ioctl(fd, GET_ARRAY_INFO, &array);
2665 if (level == 0 &&
2666 (array.level >= 4 && array.level <= 6)) {
2667 /* To convert to RAID0 we need to fail and
2668 * remove any non-data devices. */
2669 int found = 0;
2670 int d;
2671 int data_disks = array.raid_disks - 1;
2672 if (array.level == 6)
2673 data_disks -= 1;
2674 if (array.level == 5 &&
2675 array.layout != ALGORITHM_PARITY_N)
2676 return -1;
2677 if (array.level == 6 &&
2678 array.layout != ALGORITHM_PARITY_N_6)
2679 return -1;
2680 sysfs_set_str(&info, NULL,"sync_action", "idle");
2681 /* First remove any spares so no recovery starts */
2682 for (d = 0, found = 0;
2683 d < MAX_DISKS && found < array.nr_disks;
2684 d++) {
2685 mdu_disk_info_t disk;
2686 disk.number = d;
2687 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2688 continue;
2689 if (disk.major == 0 && disk.minor == 0)
2690 continue;
2691 found++;
2692 if ((disk.state & (1 << MD_DISK_ACTIVE))
2693 && disk.raid_disk < data_disks)
2694 /* keep this */
2695 continue;
2696 ioctl(fd, HOT_REMOVE_DISK,
2697 makedev(disk.major, disk.minor));
2698 }
2699 /* Now fail anything left */
2700 ioctl(fd, GET_ARRAY_INFO, &array);
2701 for (d = 0, found = 0;
2702 d < MAX_DISKS && found < array.nr_disks;
2703 d++) {
2704 int cnt;
2705 mdu_disk_info_t disk;
2706 disk.number = d;
2707 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2708 continue;
2709 if (disk.major == 0 && disk.minor == 0)
2710 continue;
2711 found++;
2712 if ((disk.state & (1 << MD_DISK_ACTIVE))
2713 && disk.raid_disk < data_disks)
2714 /* keep this */
2715 continue;
2716 ioctl(fd, SET_DISK_FAULTY,
2717 makedev(disk.major, disk.minor));
2718 cnt = 5;
2719 while (ioctl(fd, HOT_REMOVE_DISK,
2720 makedev(disk.major, disk.minor)) < 0
2721 && errno == EBUSY
2722 && cnt--) {
2723 usleep(10000);
2724 }
2725 }
2726 }
2727 c = map_num(pers, level);
2728 if (c) {
2729 int err = sysfs_set_str(&info, NULL, "level", c);
2730 if (err) {
2731 err = errno;
2732 pr_err("%s: could not set level to %s\n",
2733 devname, c);
2734 if (err == EBUSY &&
2735 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2736 cont_err("Bitmap must be removed"
2737 " before level can be changed\n");
2738 return err;
2739 }
2740 if (verbose >= 0)
2741 pr_err("level of %s changed to %s\n",
2742 devname, c);
2743 }
2744 return 0;
2745 }
2746
2747 int sigterm = 0;
2748 static void catch_term(int sig)
2749 {
2750 sigterm = 1;
2751 }
2752
2753 static int continue_via_systemd(char *devnm)
2754 {
2755 int skipped, i, pid, status;
2756 char pathbuf[1024];
2757 /* In a systemd/udev world, it is best to get systemd to
2758 * run "mdadm --grow --continue" rather than running in the
2759 * background.
2760 */
2761 switch(fork()) {
2762 case 0:
2763 /* FIXME yuk. CLOSE_EXEC?? */
2764 skipped = 0;
2765 for (i = 3; skipped < 20; i++)
2766 if (close(i) < 0)
2767 skipped++;
2768 else
2769 skipped = 0;
2770
2771 /* Don't want to see error messages from
2772 * systemctl. If the service doesn't exist,
2773 * we fork ourselves.
2774 */
2775 close(2);
2776 open("/dev/null", O_WRONLY);
2777 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2778 devnm);
2779 status = execl("/usr/bin/systemctl", "systemctl",
2780 "start",
2781 pathbuf, NULL);
2782 status = execl("/bin/systemctl", "systemctl", "start",
2783 pathbuf, NULL);
2784 exit(1);
2785 case -1: /* Just do it ourselves. */
2786 break;
2787 default: /* parent - good */
2788 pid = wait(&status);
2789 if (pid >= 0 && status == 0)
2790 return 1;
2791 }
2792 return 0;
2793 }
2794
2795 static int reshape_array(char *container, int fd, char *devname,
2796 struct supertype *st, struct mdinfo *info,
2797 int force, struct mddev_dev *devlist,
2798 unsigned long long data_offset,
2799 char *backup_file, int verbose, int forked,
2800 int restart, int freeze_reshape)
2801 {
2802 struct reshape reshape;
2803 int spares_needed;
2804 char *msg;
2805 int orig_level = UnSet;
2806 int odisks;
2807 int delayed;
2808
2809 struct mdu_array_info_s array;
2810 char *c;
2811
2812 struct mddev_dev *dv;
2813 int added_disks;
2814
2815 int *fdlist = NULL;
2816 unsigned long long *offsets = NULL;
2817 int d;
2818 int nrdisks;
2819 int err;
2820 unsigned long blocks;
2821 unsigned long long array_size;
2822 int done;
2823 struct mdinfo *sra = NULL;
2824
2825 /* when reshaping a RAID0, the component_size might be zero.
2826 * So try to fix that up.
2827 */
2828 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2829 dprintf("Cannot get array information.\n");
2830 goto release;
2831 }
2832 if (array.level == 0 && info->component_size == 0) {
2833 get_dev_size(fd, NULL, &array_size);
2834 info->component_size = array_size / array.raid_disks;
2835 }
2836
2837 if (array.level == 10)
2838 /* Need space_after info */
2839 get_space_after(fd, st, info);
2840
2841 if (info->reshape_active) {
2842 int new_level = info->new_level;
2843 info->new_level = UnSet;
2844 if (info->delta_disks > 0)
2845 info->array.raid_disks -= info->delta_disks;
2846 msg = analyse_change(devname, info, &reshape);
2847 info->new_level = new_level;
2848 if (info->delta_disks > 0)
2849 info->array.raid_disks += info->delta_disks;
2850 if (!restart)
2851 /* Make sure the array isn't read-only */
2852 ioctl(fd, RESTART_ARRAY_RW, 0);
2853 } else
2854 msg = analyse_change(devname, info, &reshape);
2855 if (msg) {
2856 /* if msg == "", error has already been printed */
2857 if (msg[0])
2858 pr_err("%s\n", msg);
2859 goto release;
2860 }
2861 if (restart &&
2862 (reshape.level != info->array.level ||
2863 reshape.before.layout != info->array.layout ||
2864 reshape.before.data_disks + reshape.parity
2865 != info->array.raid_disks - max(0, info->delta_disks))) {
2866 pr_err("reshape info is not in native format -"
2867 " cannot continue.\n");
2868 goto release;
2869 }
2870
2871 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2872 /* When reshape is restarted from '0', very begin of array
2873 * it is possible that for external metadata reshape and array
2874 * configuration doesn't happen.
2875 * Check if md has the same opinion, and reshape is restarted
2876 * from 0. If so, this is regular reshape start after reshape
2877 * switch in metadata to next array only.
2878 */
2879 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2880 (info->reshape_progress == 0))
2881 restart = 0;
2882 }
2883 if (restart) {
2884 /* reshape already started. just skip to monitoring the reshape */
2885 if (reshape.backup_blocks == 0)
2886 return 0;
2887 if (restart & RESHAPE_NO_BACKUP)
2888 return 0;
2889
2890 /* Need 'sra' down at 'started:' */
2891 sra = sysfs_read(fd, NULL,
2892 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2893 GET_CACHE);
2894 if (!sra) {
2895 pr_err("%s: Cannot get array details from sysfs\n",
2896 devname);
2897 goto release;
2898 }
2899
2900 if (!backup_file)
2901 backup_file = locate_backup(sra->sys_name);
2902
2903 goto started;
2904 }
2905 /* The container is frozen but the array may not be.
2906 * So freeze the array so spares don't get put to the wrong use
2907 * FIXME there should probably be a cleaner separation between
2908 * freeze_array and freeze_container.
2909 */
2910 sysfs_freeze_array(info);
2911 /* Check we have enough spares to not be degraded */
2912 added_disks = 0;
2913 for (dv = devlist; dv ; dv=dv->next)
2914 added_disks++;
2915 spares_needed = max(reshape.before.data_disks,
2916 reshape.after.data_disks)
2917 + reshape.parity - array.raid_disks;
2918
2919 if (!force &&
2920 info->new_level > 1 && info->array.level > 1 &&
2921 spares_needed > info->array.spare_disks + added_disks) {
2922 pr_err("Need %d spare%s to avoid degraded array,"
2923 " and only have %d.\n"
2924 " Use --force to over-ride this check.\n",
2925 spares_needed,
2926 spares_needed == 1 ? "" : "s",
2927 info->array.spare_disks + added_disks);
2928 goto release;
2929 }
2930 /* Check we have enough spares to not fail */
2931 spares_needed = max(reshape.before.data_disks,
2932 reshape.after.data_disks)
2933 - array.raid_disks;
2934 if ((info->new_level > 1 || info->new_level == 0) &&
2935 spares_needed > info->array.spare_disks +added_disks) {
2936 pr_err("Need %d spare%s to create working array,"
2937 " and only have %d.\n",
2938 spares_needed,
2939 spares_needed == 1 ? "" : "s",
2940 info->array.spare_disks + added_disks);
2941 goto release;
2942 }
2943
2944 if (reshape.level != array.level) {
2945 int err = impose_level(fd, reshape.level, devname, verbose);
2946 if (err)
2947 goto release;
2948 info->new_layout = UnSet; /* after level change,
2949 * layout is meaningless */
2950 orig_level = array.level;
2951 sysfs_freeze_array(info);
2952
2953 if (reshape.level > 0 && st->ss->external) {
2954 /* make sure mdmon is aware of the new level */
2955 if (mdmon_running(container))
2956 flush_mdmon(container);
2957
2958 if (!mdmon_running(container))
2959 start_mdmon(container);
2960 ping_monitor(container);
2961 if (mdmon_running(container) &&
2962 st->update_tail == NULL)
2963 st->update_tail = &st->updates;
2964 }
2965 }
2966 /* ->reshape_super might have chosen some spares from the
2967 * container that it wants to be part of the new array.
2968 * We can collect them with ->container_content and give
2969 * them to the kernel.
2970 */
2971 if (st->ss->reshape_super && st->ss->container_content) {
2972 char *subarray = strchr(info->text_version+1, '/')+1;
2973 struct mdinfo *info2 =
2974 st->ss->container_content(st, subarray);
2975 struct mdinfo *d;
2976
2977 if (info2) {
2978 sysfs_init(info2, fd, st->devnm);
2979 /* When increasing number of devices, we need to set
2980 * new raid_disks before adding these, or they might
2981 * be rejected.
2982 */
2983 if (reshape.backup_blocks &&
2984 reshape.after.data_disks > reshape.before.data_disks)
2985 subarray_set_num(container, info2, "raid_disks",
2986 reshape.after.data_disks +
2987 reshape.parity);
2988 for (d = info2->devs; d; d = d->next) {
2989 if (d->disk.state == 0 &&
2990 d->disk.raid_disk >= 0) {
2991 /* This is a spare that wants to
2992 * be part of the array.
2993 */
2994 add_disk(fd, st, info2, d);
2995 }
2996 }
2997 sysfs_free(info2);
2998 }
2999 }
3000 /* We might have been given some devices to add to the
3001 * array. Now that the array has been changed to the right
3002 * level and frozen, we can safely add them.
3003 */
3004 if (devlist)
3005 Manage_subdevs(devname, fd, devlist, verbose,
3006 0,NULL, 0);
3007
3008 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3009 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3010 if (reshape.backup_blocks == 0) {
3011 /* No restriping needed, but we might need to impose
3012 * some more changes: layout, raid_disks, chunk_size
3013 */
3014 /* read current array info */
3015 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
3016 dprintf("Cannot get array information.\n");
3017 goto release;
3018 }
3019 /* compare current array info with new values and if
3020 * it is different update them to new */
3021 if (info->new_layout != UnSet &&
3022 info->new_layout != array.layout) {
3023 array.layout = info->new_layout;
3024 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3025 pr_err("failed to set new layout\n");
3026 goto release;
3027 } else if (verbose >= 0)
3028 printf("layout for %s set to %d\n",
3029 devname, array.layout);
3030 }
3031 if (info->delta_disks != UnSet &&
3032 info->delta_disks != 0 &&
3033 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3034 array.raid_disks += info->delta_disks;
3035 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3036 pr_err("failed to set raid disks\n");
3037 goto release;
3038 } else if (verbose >= 0) {
3039 printf("raid_disks for %s set to %d\n",
3040 devname, array.raid_disks);
3041 }
3042 }
3043 if (info->new_chunk != 0 &&
3044 info->new_chunk != array.chunk_size) {
3045 if (sysfs_set_num(info, NULL,
3046 "chunk_size", info->new_chunk) != 0) {
3047 pr_err("failed to set chunk size\n");
3048 goto release;
3049 } else if (verbose >= 0)
3050 printf("chunk size for %s set to %d\n",
3051 devname, array.chunk_size);
3052 }
3053 unfreeze(st);
3054 return 0;
3055 }
3056
3057 /*
3058 * There are three possibilities.
3059 * 1/ The array will shrink.
3060 * We need to ensure the reshape will pause before reaching
3061 * the 'critical section'. We also need to fork and wait for
3062 * that to happen. When it does we
3063 * suspend/backup/complete/unfreeze
3064 *
3065 * 2/ The array will not change size.
3066 * This requires that we keep a backup of a sliding window
3067 * so that we can restore data after a crash. So we need
3068 * to fork and monitor progress.
3069 * In future we will allow the data_offset to change, so
3070 * a sliding backup becomes unnecessary.
3071 *
3072 * 3/ The array will grow. This is relatively easy.
3073 * However the kernel's restripe routines will cheerfully
3074 * overwrite some early data before it is safe. So we
3075 * need to make a backup of the early parts of the array
3076 * and be ready to restore it if rebuild aborts very early.
3077 * For externally managed metadata, we still need a forked
3078 * child to monitor the reshape and suspend IO over the region
3079 * that is being reshaped.
3080 *
3081 * We backup data by writing it to one spare, or to a
3082 * file which was given on command line.
3083 *
3084 * In each case, we first make sure that storage is available
3085 * for the required backup.
3086 * Then we:
3087 * - request the shape change.
3088 * - fork to handle backup etc.
3089 */
3090 /* Check that we can hold all the data */
3091 get_dev_size(fd, NULL, &array_size);
3092 if (reshape.new_size < (array_size/512)) {
3093 pr_err("this change will reduce the size of the array.\n"
3094 " use --grow --array-size first to truncate array.\n"
3095 " e.g. mdadm --grow %s --array-size %llu\n",
3096 devname, reshape.new_size/2);
3097 goto release;
3098 }
3099
3100 if (array.level == 10) {
3101 /* Reshaping RAID10 does not require any data backup by
3102 * user-space. Instead it requires that the data_offset
3103 * is changed to avoid the need for backup.
3104 * So this is handled very separately
3105 */
3106 if (restart)
3107 /* Nothing to do. */
3108 return 0;
3109 return raid10_reshape(container, fd, devname, st, info,
3110 &reshape, data_offset,
3111 force, verbose);
3112 }
3113 sra = sysfs_read(fd, NULL,
3114 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3115 GET_CACHE);
3116 if (!sra) {
3117 pr_err("%s: Cannot get array details from sysfs\n",
3118 devname);
3119 goto release;
3120 }
3121
3122 if (!backup_file)
3123 switch(set_new_data_offset(sra, st, devname,
3124 reshape.after.data_disks - reshape.before.data_disks,
3125 data_offset,
3126 reshape.min_offset_change, 1)) {
3127 case -1:
3128 goto release;
3129 case 0:
3130 /* Updated data_offset, so it's easy now */
3131 update_cache_size(container, sra, info,
3132 min(reshape.before.data_disks,
3133 reshape.after.data_disks),
3134 reshape.backup_blocks);
3135
3136 /* Right, everything seems fine. Let's kick things off.
3137 */
3138 sync_metadata(st);
3139
3140 if (impose_reshape(sra, info, st, fd, restart,
3141 devname, container, &reshape) < 0)
3142 goto release;
3143 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3144 pr_err("Failed to initiate reshape!\n");
3145 goto release;
3146 }
3147 if (info->new_level == reshape.level)
3148 return 0;
3149 /* need to adjust level when reshape completes */
3150 switch(fork()) {
3151 case -1: /* ignore error, but don't wait */
3152 return 0;
3153 default: /* parent */
3154 return 0;
3155 case 0:
3156 map_fork();
3157 break;
3158 }
3159 close(fd);
3160 wait_reshape(sra);
3161 fd = open_dev(sra->sys_name);
3162 if (fd >= 0)
3163 impose_level(fd, info->new_level, devname, verbose);
3164 return 0;
3165 case 1: /* Couldn't set data_offset, try the old way */
3166 if (data_offset != INVALID_SECTORS) {
3167 pr_err("Cannot update data_offset on this array\n");
3168 goto release;
3169 }
3170 break;
3171 }
3172
3173 started:
3174 /* Decide how many blocks (sectors) for a reshape
3175 * unit. The number we have so far is just a minimum
3176 */
3177 blocks = reshape.backup_blocks;
3178 if (reshape.before.data_disks ==
3179 reshape.after.data_disks) {
3180 /* Make 'blocks' bigger for better throughput, but
3181 * not so big that we reject it below.
3182 * Try for 16 megabytes
3183 */
3184 while (blocks * 32 < sra->component_size &&
3185 blocks < 16*1024*2)
3186 blocks *= 2;
3187 } else
3188 pr_err("Need to backup %luK of critical "
3189 "section..\n", blocks/2);
3190
3191 if (blocks >= sra->component_size/2) {
3192 pr_err("%s: Something wrong"
3193 " - reshape aborted\n",
3194 devname);
3195 goto release;
3196 }
3197
3198 /* Now we need to open all these devices so we can read/write.
3199 */
3200 nrdisks = max(reshape.before.data_disks,
3201 reshape.after.data_disks) + reshape.parity
3202 + sra->array.spare_disks;
3203 fdlist = xcalloc((1+nrdisks), sizeof(int));
3204 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3205
3206 odisks = reshape.before.data_disks + reshape.parity;
3207 d = reshape_prepare_fdlist(devname, sra, odisks,
3208 nrdisks, blocks, backup_file,
3209 fdlist, offsets);
3210 if (d < 0) {
3211 goto release;
3212 }
3213 if ((st->ss->manage_reshape == NULL) ||
3214 (st->ss->recover_backup == NULL)) {
3215 if (backup_file == NULL) {
3216 if (reshape.after.data_disks <=
3217 reshape.before.data_disks) {
3218 pr_err("%s: Cannot grow - need backup-file\n",
3219 devname);
3220 pr_err(" Please provide one with \"--backup=...\"\n");
3221 goto release;
3222 } else if (sra->array.spare_disks == 0) {
3223 pr_err("%s: Cannot grow - "
3224 "need a spare or backup-file to backup "
3225 "critical section\n", devname);
3226 goto release;
3227 }
3228 } else {
3229 if (!reshape_open_backup_file(backup_file, fd, devname,
3230 (signed)blocks,
3231 fdlist+d, offsets+d,
3232 sra->sys_name,
3233 restart)) {
3234 goto release;
3235 }
3236 d++;
3237 }
3238 }
3239
3240 update_cache_size(container, sra, info,
3241 min(reshape.before.data_disks, reshape.after.data_disks),
3242 blocks);
3243
3244 /* Right, everything seems fine. Let's kick things off.
3245 * If only changing raid_disks, use ioctl, else use
3246 * sysfs.
3247 */
3248 sync_metadata(st);
3249
3250 if (impose_reshape(sra, info, st, fd, restart,
3251 devname, container, &reshape) < 0)
3252 goto release;
3253
3254 err = start_reshape(sra, restart, reshape.before.data_disks,
3255 reshape.after.data_disks);
3256 if (err) {
3257 pr_err("Cannot %s reshape for %s\n",
3258 restart ? "continue" : "start",
3259 devname);
3260 goto release;
3261 }
3262 if (restart)
3263 sysfs_set_str(sra, NULL, "array_state", "active");
3264 if (freeze_reshape) {
3265 free(fdlist);
3266 free(offsets);
3267 sysfs_free(sra);
3268 pr_err("Reshape has to be continued from"
3269 " location %llu when root filesystem has been mounted.\n",
3270 sra->reshape_progress);
3271 return 1;
3272 }
3273
3274 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3275 if (continue_via_systemd(container ?: sra->sys_name)) {
3276 free(fdlist);
3277 free(offsets);
3278 sysfs_free(sra);
3279 return 0;
3280 }
3281
3282 /* Now we just need to kick off the reshape and watch, while
3283 * handling backups of the data...
3284 * This is all done by a forked background process.
3285 */
3286 switch(forked ? 0 : fork()) {
3287 case -1:
3288 pr_err("Cannot run child to monitor reshape: %s\n",
3289 strerror(errno));
3290 abort_reshape(sra);
3291 goto release;
3292 default:
3293 free(fdlist);
3294 free(offsets);
3295 sysfs_free(sra);
3296 return 0;
3297 case 0:
3298 map_fork();
3299 break;
3300 }
3301
3302 /* If another array on the same devices is busy, the
3303 * reshape will wait for them. This would mean that
3304 * the first section that we suspend will stay suspended
3305 * for a long time. So check on that possibility
3306 * by looking for "DELAYED" in /proc/mdstat, and if found,
3307 * wait a while
3308 */
3309 do {
3310 struct mdstat_ent *mds, *m;
3311 delayed = 0;
3312 mds = mdstat_read(1, 0);
3313 for (m = mds; m; m = m->next)
3314 if (strcmp(m->devnm, sra->sys_name) == 0) {
3315 if (m->resync &&
3316 m->percent == RESYNC_DELAYED)
3317 delayed = 1;
3318 if (m->resync == 0)
3319 /* Haven't started the reshape thread
3320 * yet, wait a bit
3321 */
3322 delayed = 2;
3323 break;
3324 }
3325 free_mdstat(mds);
3326 if (delayed == 1 && get_linux_version() < 3007000) {
3327 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3328 " You might experience problems until other reshapes complete.\n");
3329 delayed = 0;
3330 }
3331 if (delayed)
3332 mdstat_wait(30 - (delayed-1) * 25);
3333 } while (delayed);
3334 mdstat_close();
3335 close(fd);
3336 if (check_env("MDADM_GROW_VERIFY"))
3337 fd = open(devname, O_RDONLY | O_DIRECT);
3338 else
3339 fd = -1;
3340 mlockall(MCL_FUTURE);
3341
3342 signal(SIGTERM, catch_term);
3343
3344 if (st->ss->external) {
3345 /* metadata handler takes it from here */
3346 done = st->ss->manage_reshape(
3347 fd, sra, &reshape, st, blocks,
3348 fdlist, offsets,
3349 d - odisks, fdlist+odisks,
3350 offsets+odisks);
3351 } else
3352 done = child_monitor(
3353 fd, sra, &reshape, st, blocks,
3354 fdlist, offsets,
3355 d - odisks, fdlist+odisks,
3356 offsets+odisks);
3357
3358 free(fdlist);
3359 free(offsets);
3360
3361 if (backup_file && done) {
3362 char *bul;
3363 bul = make_backup(sra->sys_name);
3364 if (bul) {
3365 char buf[1024];
3366 int l = readlink(bul, buf, sizeof(buf));
3367 if (l > 0) {
3368 buf[l]=0;
3369 unlink(buf);
3370 }
3371 unlink(bul);
3372 free(bul);
3373 }
3374 unlink(backup_file);
3375 }
3376 if (!done) {
3377 abort_reshape(sra);
3378 goto out;
3379 }
3380
3381 if (!st->ss->external &&
3382 !(reshape.before.data_disks != reshape.after.data_disks
3383 && info->custom_array_size) &&
3384 info->new_level == reshape.level &&
3385 !forked) {
3386 /* no need to wait for the reshape to finish as
3387 * there is nothing more to do.
3388 */
3389 sysfs_free(sra);
3390 exit(0);
3391 }
3392 wait_reshape(sra);
3393
3394 if (st->ss->external) {
3395 /* Re-load the metadata as much could have changed */
3396 int cfd = open_dev(st->container_devnm);
3397 if (cfd >= 0) {
3398 flush_mdmon(container);
3399 st->ss->free_super(st);
3400 st->ss->load_container(st, cfd, container);
3401 close(cfd);
3402 }
3403 }
3404
3405 /* set new array size if required customer_array_size is used
3406 * by this metadata.
3407 */
3408 if (reshape.before.data_disks !=
3409 reshape.after.data_disks &&
3410 info->custom_array_size)
3411 set_array_size(st, info, info->text_version);
3412
3413 if (info->new_level != reshape.level) {
3414 if (fd < 0)
3415 fd = open(devname, O_RDONLY);
3416 impose_level(fd, info->new_level, devname, verbose);
3417 close(fd);
3418 if (info->new_level == 0)
3419 st->update_tail = NULL;
3420 }
3421 out:
3422 sysfs_free(sra);
3423 if (forked)
3424 return 0;
3425 unfreeze(st);
3426 exit(0);
3427
3428 release:
3429 free(fdlist);
3430 free(offsets);
3431 if (orig_level != UnSet && sra) {
3432 c = map_num(pers, orig_level);
3433 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3434 pr_err("aborting level change\n");
3435 }
3436 sysfs_free(sra);
3437 if (!forked)
3438 unfreeze(st);
3439 return 1;
3440 }
3441
3442 /* mdfd handle is passed to be closed in child process (after fork).
3443 */
3444 int reshape_container(char *container, char *devname,
3445 int mdfd,
3446 struct supertype *st,
3447 struct mdinfo *info,
3448 int force,
3449 char *backup_file, int verbose,
3450 int forked, int restart, int freeze_reshape)
3451 {
3452 struct mdinfo *cc = NULL;
3453 int rv = restart;
3454 char last_devnm[32] = "";
3455
3456 /* component_size is not meaningful for a container,
3457 * so pass '0' meaning 'no change'
3458 */
3459 if (!restart &&
3460 reshape_super(st, 0, info->new_level,
3461 info->new_layout, info->new_chunk,
3462 info->array.raid_disks, info->delta_disks,
3463 backup_file, devname, APPLY_METADATA_CHANGES,
3464 verbose)) {
3465 unfreeze(st);
3466 return 1;
3467 }
3468
3469 sync_metadata(st);
3470
3471 /* ping monitor to be sure that update is on disk
3472 */
3473 ping_monitor(container);
3474
3475 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3476 if (continue_via_systemd(container))
3477 return 0;
3478
3479 switch (forked ? 0 : fork()) {
3480 case -1: /* error */
3481 perror("Cannot fork to complete reshape\n");
3482 unfreeze(st);
3483 return 1;
3484 default: /* parent */
3485 if (!freeze_reshape)
3486 printf(Name ": multi-array reshape continues"
3487 " in background\n");
3488 return 0;
3489 case 0: /* child */
3490 map_fork();
3491 break;
3492 }
3493
3494 /* close unused handle in child process
3495 */
3496 if (mdfd > -1)
3497 close(mdfd);
3498
3499 while(1) {
3500 /* For each member array with reshape_active,
3501 * we need to perform the reshape.
3502 * We pick the first array that needs reshaping and
3503 * reshape it. reshape_array() will re-read the metadata
3504 * so the next time through a different array should be
3505 * ready for reshape.
3506 * It is possible that the 'different' array will not
3507 * be assembled yet. In that case we simple exit.
3508 * When it is assembled, the mdadm which assembles it
3509 * will take over the reshape.
3510 */
3511 struct mdinfo *content;
3512 int fd;
3513 struct mdstat_ent *mdstat;
3514 char *adev;
3515 int devid;
3516
3517 sysfs_free(cc);
3518
3519 cc = st->ss->container_content(st, NULL);
3520
3521 for (content = cc; content ; content = content->next) {
3522 char *subarray;
3523 if (!content->reshape_active)
3524 continue;
3525
3526 subarray = strchr(content->text_version+1, '/')+1;
3527 mdstat = mdstat_by_subdev(subarray, container);
3528 if (!mdstat)
3529 continue;
3530 if (mdstat->active == 0) {
3531 pr_err("Skipping inactive array %s.\n",
3532 mdstat->devnm);
3533 free_mdstat(mdstat);
3534 mdstat = NULL;
3535 continue;
3536 }
3537 break;
3538 }
3539 if (!content)
3540 break;
3541
3542 devid = devnm2devid(mdstat->devnm);
3543 adev = map_dev(major(devid), minor(devid), 0);
3544 if (!adev)
3545 adev = content->text_version;
3546
3547 fd = open_dev(mdstat->devnm);
3548 if (fd < 0) {
3549 printf(Name ": Device %s cannot be opened for reshape.",
3550 adev);
3551 break;
3552 }
3553
3554 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3555 /* Do not allow for multiple reshape_array() calls for
3556 * the same array.
3557 * It can happen when reshape_array() returns without
3558 * error, when reshape is not finished (wrong reshape
3559 * starting/continuation conditions). Mdmon doesn't
3560 * switch to next array in container and reentry
3561 * conditions for the same array occur.
3562 * This is possibly interim until the behaviour of
3563 * reshape_array is resolved().
3564 */
3565 printf(Name ": Multiple reshape execution detected for "
3566 "device %s.", adev);
3567 close(fd);
3568 break;
3569 }
3570 strcpy(last_devnm, mdstat->devnm);
3571
3572 sysfs_init(content, fd, mdstat->devnm);
3573
3574 if (mdmon_running(container))
3575 flush_mdmon(container);
3576
3577 rv = reshape_array(container, fd, adev, st,
3578 content, force, NULL, INVALID_SECTORS,
3579 backup_file, verbose, 1, restart,
3580 freeze_reshape);
3581 close(fd);
3582
3583 if (freeze_reshape) {
3584 sysfs_free(cc);
3585 exit(0);
3586 }
3587
3588 restart = 0;
3589 if (rv)
3590 break;
3591
3592 if (mdmon_running(container))
3593 flush_mdmon(container);
3594 }
3595 if (!rv)
3596 unfreeze(st);
3597 sysfs_free(cc);
3598 exit(0);
3599 }
3600
3601 /*
3602 * We run a child process in the background which performs the following
3603 * steps:
3604 * - wait for resync to reach a certain point
3605 * - suspend io to the following section
3606 * - backup that section
3607 * - allow resync to proceed further
3608 * - resume io
3609 * - discard the backup.
3610 *
3611 * When are combined in slightly different ways in the three cases.
3612 * Grow:
3613 * - suspend/backup/allow/wait/resume/discard
3614 * Shrink:
3615 * - allow/wait/suspend/backup/allow/wait/resume/discard
3616 * same-size:
3617 * - wait/resume/discard/suspend/backup/allow
3618 *
3619 * suspend/backup/allow always come together
3620 * wait/resume/discard do too.
3621 * For the same-size case we have two backups to improve flow.
3622 *
3623 */
3624
3625 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3626 unsigned long long backup_point,
3627 unsigned long long wait_point,
3628 unsigned long long *suspend_point,
3629 unsigned long long *reshape_completed, int *frozen)
3630 {
3631 /* This function is called repeatedly by the reshape manager.
3632 * It determines how much progress can safely be made and allows
3633 * that progress.
3634 * - 'info' identifies the array and particularly records in
3635 * ->reshape_progress the metadata's knowledge of progress
3636 * This is a sector offset from the start of the array
3637 * of the next array block to be relocated. This number
3638 * may increase from 0 or decrease from array_size, depending
3639 * on the type of reshape that is happening.
3640 * Note that in contrast, 'sync_completed' is a block count of the
3641 * reshape so far. It gives the distance between the start point
3642 * (head or tail of device) and the next place that data will be
3643 * written. It always increases.
3644 * - 'reshape' is the structure created by analyse_change
3645 * - 'backup_point' shows how much the metadata manager has backed-up
3646 * data. For reshapes with increasing progress, it is the next address
3647 * to be backed up, previous addresses have been backed-up. For
3648 * decreasing progress, it is the earliest address that has been
3649 * backed up - later address are also backed up.
3650 * So addresses between reshape_progress and backup_point are
3651 * backed up providing those are in the 'correct' order.
3652 * - 'wait_point' is an array address. When reshape_completed
3653 * passes this point, progress_reshape should return. It might
3654 * return earlier if it determines that ->reshape_progress needs
3655 * to be updated or further backup is needed.
3656 * - suspend_point is maintained by progress_reshape and the caller
3657 * should not touch it except to initialise to zero.
3658 * It is an array address and it only increases in 2.6.37 and earlier.
3659 * This makes it difficult to handle reducing reshapes with
3660 * external metadata.
3661 * However: it is similar to backup_point in that it records the
3662 * other end of a suspended region from reshape_progress.
3663 * it is moved to extend the region that is safe to backup and/or
3664 * reshape
3665 * - reshape_completed is read from sysfs and returned. The caller
3666 * should copy this into ->reshape_progress when it has reason to
3667 * believe that the metadata knows this, and any backup outside this
3668 * has been erased.
3669 *
3670 * Return value is:
3671 * 1 if more data from backup_point - but only as far as suspend_point,
3672 * should be backed up
3673 * 0 if things are progressing smoothly
3674 * -1 if the reshape is finished because it is all done,
3675 * -2 if the reshape is finished due to an error.
3676 */
3677
3678 int advancing = (reshape->after.data_disks
3679 >= reshape->before.data_disks);
3680 unsigned long long need_backup; /* All data between start of array and
3681 * here will at some point need to
3682 * be backed up.
3683 */
3684 unsigned long long read_offset, write_offset;
3685 unsigned long long write_range;
3686 unsigned long long max_progress, target, completed;
3687 unsigned long long array_size = (info->component_size
3688 * reshape->before.data_disks);
3689 int fd;
3690 char buf[20];
3691
3692 /* First, we unsuspend any region that is now known to be safe.
3693 * If suspend_point is on the 'wrong' side of reshape_progress, then
3694 * we don't have or need suspension at the moment. This is true for
3695 * native metadata when we don't need to back-up.
3696 */
3697 if (advancing) {
3698 if (info->reshape_progress <= *suspend_point)
3699 sysfs_set_num(info, NULL, "suspend_lo",
3700 info->reshape_progress);
3701 } else {
3702 /* Note: this won't work in 2.6.37 and before.
3703 * Something somewhere should make sure we don't need it!
3704 */
3705 if (info->reshape_progress >= *suspend_point)
3706 sysfs_set_num(info, NULL, "suspend_hi",
3707 info->reshape_progress);
3708 }
3709
3710 /* Now work out how far it is safe to progress.
3711 * If the read_offset for ->reshape_progress is less than
3712 * 'blocks' beyond the write_offset, we can only progress as far
3713 * as a backup.
3714 * Otherwise we can progress until the write_offset for the new location
3715 * reaches (within 'blocks' of) the read_offset at the current location.
3716 * However that region must be suspended unless we are using native
3717 * metadata.
3718 * If we need to suspend more, we limit it to 128M per device, which is
3719 * rather arbitrary and should be some time-based calculation.
3720 */
3721 read_offset = info->reshape_progress / reshape->before.data_disks;
3722 write_offset = info->reshape_progress / reshape->after.data_disks;
3723 write_range = info->new_chunk/512;
3724 if (reshape->before.data_disks == reshape->after.data_disks)
3725 need_backup = array_size;
3726 else
3727 need_backup = reshape->backup_blocks;
3728 if (advancing) {
3729 if (read_offset < write_offset + write_range)
3730 max_progress = backup_point;
3731 else
3732 max_progress =
3733 read_offset *
3734 reshape->after.data_disks;
3735 } else {
3736 if (read_offset > write_offset - write_range)
3737 /* Can only progress as far as has been backed up,
3738 * which must be suspended */
3739 max_progress = backup_point;
3740 else if (info->reshape_progress <= need_backup)
3741 max_progress = backup_point;
3742 else {
3743 if (info->array.major_version >= 0)
3744 /* Can progress until backup is needed */
3745 max_progress = need_backup;
3746 else {
3747 /* Can progress until metadata update is required */
3748 max_progress =
3749 read_offset *
3750 reshape->after.data_disks;
3751 /* but data must be suspended */
3752 if (max_progress < *suspend_point)
3753 max_progress = *suspend_point;
3754 }
3755 }
3756 }
3757
3758 /* We know it is safe to progress to 'max_progress' providing
3759 * it is suspended or we are using native metadata.
3760 * Consider extending suspend_point 128M per device if it
3761 * is less than 64M per device beyond reshape_progress.
3762 * But always do a multiple of 'blocks'
3763 * FIXME this is too big - it takes to long to complete
3764 * this much.
3765 */
3766 target = 64*1024*2 * min(reshape->before.data_disks,
3767 reshape->after.data_disks);
3768 target /= reshape->backup_blocks;
3769 if (target < 2)
3770 target = 2;
3771 target *= reshape->backup_blocks;
3772
3773 /* For externally managed metadata we always need to suspend IO to
3774 * the area being reshaped so we regularly push suspend_point forward.
3775 * For native metadata we only need the suspend if we are going to do
3776 * a backup.
3777 */
3778 if (advancing) {
3779 if ((need_backup > info->reshape_progress
3780 || info->array.major_version < 0) &&
3781 *suspend_point < info->reshape_progress + target) {
3782 if (need_backup < *suspend_point + 2 * target)
3783 *suspend_point = need_backup;
3784 else if (*suspend_point + 2 * target < array_size)
3785 *suspend_point += 2 * target;
3786 else
3787 *suspend_point = array_size;
3788 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3789 if (max_progress > *suspend_point)
3790 max_progress = *suspend_point;
3791 }
3792 } else {
3793 if (info->array.major_version >= 0) {
3794 /* Only need to suspend when about to backup */
3795 if (info->reshape_progress < need_backup * 2 &&
3796 *suspend_point > 0) {
3797 *suspend_point = 0;
3798 sysfs_set_num(info, NULL, "suspend_lo", 0);
3799 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3800 }
3801 } else {
3802 /* Need to suspend continually */
3803 if (info->reshape_progress < *suspend_point)
3804 *suspend_point = info->reshape_progress;
3805 if (*suspend_point + target < info->reshape_progress)
3806 /* No need to move suspend region yet */;
3807 else {
3808 if (*suspend_point >= 2 * target)
3809 *suspend_point -= 2 * target;
3810 else
3811 *suspend_point = 0;
3812 sysfs_set_num(info, NULL, "suspend_lo",
3813 *suspend_point);
3814 }
3815 if (max_progress < *suspend_point)
3816 max_progress = *suspend_point;
3817 }
3818 }
3819
3820 /* now set sync_max to allow that progress. sync_max, like
3821 * sync_completed is a count of sectors written per device, so
3822 * we find the difference between max_progress and the start point,
3823 * and divide that by after.data_disks to get a sync_max
3824 * number.
3825 * At the same time we convert wait_point to a similar number
3826 * for comparing against sync_completed.
3827 */
3828 /* scale down max_progress to per_disk */
3829 max_progress /= reshape->after.data_disks;
3830 /* Round to chunk size as some kernels give an erroneously high number */
3831 max_progress /= info->new_chunk/512;
3832 max_progress *= info->new_chunk/512;
3833 /* And round to old chunk size as the kernel wants that */
3834 max_progress /= info->array.chunk_size/512;
3835 max_progress *= info->array.chunk_size/512;
3836 /* Limit progress to the whole device */
3837 if (max_progress > info->component_size)
3838 max_progress = info->component_size;
3839 wait_point /= reshape->after.data_disks;
3840 if (!advancing) {
3841 /* switch from 'device offset' to 'processed block count' */
3842 max_progress = info->component_size - max_progress;
3843 wait_point = info->component_size - wait_point;
3844 }
3845
3846 if (!*frozen)
3847 sysfs_set_num(info, NULL, "sync_max", max_progress);
3848
3849 /* Now wait. If we have already reached the point that we were
3850 * asked to wait to, don't wait at all, else wait for any change.
3851 * We need to select on 'sync_completed' as that is the place that
3852 * notifications happen, but we are really interested in
3853 * 'reshape_position'
3854 */
3855 fd = sysfs_get_fd(info, NULL, "sync_completed");
3856 if (fd < 0)
3857 goto check_progress;
3858
3859 if (sysfs_fd_get_ll(fd, &completed) < 0)
3860 goto check_progress;
3861
3862 while (completed < max_progress && completed < wait_point) {
3863 /* Check that sync_action is still 'reshape' to avoid
3864 * waiting forever on a dead array
3865 */
3866 char action[20];
3867 if (sysfs_get_str(info, NULL, "sync_action",
3868 action, 20) <= 0 ||
3869 strncmp(action, "reshape", 7) != 0)
3870 break;
3871 /* Some kernels reset 'sync_completed' to zero
3872 * before setting 'sync_action' to 'idle'.
3873 * So we need these extra tests.
3874 */
3875 if (completed == 0 && advancing
3876 && info->reshape_progress > 0)
3877 break;
3878 if (completed == 0 && !advancing
3879 && info->reshape_progress < (info->component_size
3880 * reshape->after.data_disks))
3881 break;
3882 sysfs_wait(fd, NULL);
3883 if (sysfs_fd_get_ll(fd, &completed) < 0)
3884 goto check_progress;
3885 }
3886 /* Some kernels reset 'sync_completed' to zero,
3887 * we need to have real point we are in md
3888 */
3889 if (completed == 0)
3890 completed = max_progress;
3891
3892 /* some kernels can give an incorrectly high 'completed' number */
3893 completed /= (info->new_chunk/512);
3894 completed *= (info->new_chunk/512);
3895 /* Convert 'completed' back in to a 'progress' number */
3896 completed *= reshape->after.data_disks;
3897 if (!advancing) {
3898 completed = info->component_size * reshape->after.data_disks
3899 - completed;
3900 }
3901 *reshape_completed = completed;
3902
3903 close(fd);
3904
3905 /* We return the need_backup flag. Caller will decide
3906 * how much - a multiple of ->backup_blocks up to *suspend_point
3907 */
3908 if (advancing)
3909 return need_backup > info->reshape_progress;
3910 else
3911 return need_backup >= info->reshape_progress;
3912
3913 check_progress:
3914 /* if we couldn't read a number from sync_completed, then
3915 * either the reshape did complete, or it aborted.
3916 * We can tell which by checking for 'none' in reshape_position.
3917 * If it did abort, then it might immediately restart if it
3918 * it was just a device failure that leaves us degraded but
3919 * functioning.
3920 */
3921 strcpy(buf, "hi");
3922 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3923 || strncmp(buf, "none", 4) != 0) {
3924 /* The abort might only be temporary. Wait up to 10
3925 * seconds for fd to contain a valid number again.
3926 */
3927 int wait = 10000;
3928 int rv = -2;
3929 unsigned long long new_sync_max;
3930 while (fd >= 0 && rv < 0 && wait > 0) {
3931 if (sysfs_wait(fd, &wait) != 1)
3932 break;
3933 switch (sysfs_fd_get_ll(fd, &completed)) {
3934 case 0:
3935 /* all good again */
3936 rv = 1;
3937 /* If "sync_max" is no longer max_progress
3938 * we need to freeze things
3939 */
3940 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3941 *frozen = (new_sync_max != max_progress);
3942 break;
3943 case -2: /* read error - abort */
3944 wait = 0;
3945 break;
3946 }
3947 }
3948 if (fd >= 0)
3949 close(fd);
3950 return rv; /* abort */
3951 } else {
3952 /* Maybe racing with array shutdown - check state */
3953 if (fd >= 0)
3954 close(fd);
3955 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3956 || strncmp(buf, "inactive", 8) == 0
3957 || strncmp(buf, "clear",5) == 0)
3958 return -2; /* abort */
3959 return -1; /* complete */
3960 }
3961 }
3962
3963 /* FIXME return status is never checked */
3964 static int grow_backup(struct mdinfo *sra,
3965 unsigned long long offset, /* per device */
3966 unsigned long stripes, /* per device, in old chunks */
3967 int *sources, unsigned long long *offsets,
3968 int disks, int chunk, int level, int layout,
3969 int dests, int *destfd, unsigned long long *destoffsets,
3970 int part, int *degraded,
3971 char *buf)
3972 {
3973 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3974 * to storage 'destfd' (offset 'destoffsets'), after first
3975 * suspending IO. Then allow resync to continue
3976 * over the suspended section.
3977 * Use part 'part' of the backup-super-block.
3978 */
3979 int odata = disks;
3980 int rv = 0;
3981 int i;
3982 unsigned long long ll;
3983 int new_degraded;
3984 //printf("offset %llu\n", offset);
3985 if (level >= 4)
3986 odata--;
3987 if (level == 6)
3988 odata--;
3989
3990 /* Check that array hasn't become degraded, else we might backup the wrong data */
3991 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3992 return -1; /* FIXME this error is ignored */
3993 new_degraded = (int)ll;
3994 if (new_degraded != *degraded) {
3995 /* check each device to ensure it is still working */
3996 struct mdinfo *sd;
3997 for (sd = sra->devs ; sd ; sd = sd->next) {
3998 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3999 continue;
4000 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4001 char sbuf[20];
4002 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
4003 strstr(sbuf, "faulty") ||
4004 strstr(sbuf, "in_sync") == NULL) {
4005 /* this device is dead */
4006 sd->disk.state = (1<<MD_DISK_FAULTY);
4007 if (sd->disk.raid_disk >= 0 &&
4008 sources[sd->disk.raid_disk] >= 0) {
4009 close(sources[sd->disk.raid_disk]);
4010 sources[sd->disk.raid_disk] = -1;
4011 }
4012 }
4013 }
4014 }
4015 *degraded = new_degraded;
4016 }
4017 if (part) {
4018 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4019 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4020 } else {
4021 bsb.arraystart = __cpu_to_le64(offset * odata);
4022 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4023 }
4024 if (part)
4025 bsb.magic[15] = '2';
4026 for (i = 0; i < dests; i++)
4027 if (part)
4028 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4029 else
4030 lseek64(destfd[i], destoffsets[i], 0);
4031
4032 rv = save_stripes(sources, offsets,
4033 disks, chunk, level, layout,
4034 dests, destfd,
4035 offset*512*odata, stripes * chunk * odata,
4036 buf);
4037
4038 if (rv)
4039 return rv;
4040 bsb.mtime = __cpu_to_le64(time(0));
4041 for (i = 0; i < dests; i++) {
4042 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4043
4044 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4045 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4046 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4047 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4048
4049 rv = -1;
4050 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4051 != destoffsets[i] - 4096)
4052 break;
4053 if (write(destfd[i], &bsb, 512) != 512)
4054 break;
4055 if (destoffsets[i] > 4096) {
4056 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4057 destoffsets[i]+stripes*chunk*odata)
4058 break;
4059 if (write(destfd[i], &bsb, 512) != 512)
4060 break;
4061 }
4062 fsync(destfd[i]);
4063 rv = 0;
4064 }
4065
4066 return rv;
4067 }
4068
4069 /* in 2.6.30, the value reported by sync_completed can be
4070 * less that it should be by one stripe.
4071 * This only happens when reshape hits sync_max and pauses.
4072 * So allow wait_backup to either extent sync_max further
4073 * than strictly necessary, or return before the
4074 * sync has got quite as far as we would really like.
4075 * This is what 'blocks2' is for.
4076 * The various caller give appropriate values so that
4077 * every works.
4078 */
4079 /* FIXME return value is often ignored */
4080 static int forget_backup(int dests, int *destfd,
4081 unsigned long long *destoffsets,
4082 int part)
4083 {
4084 /*
4085 * Erase backup 'part' (which is 0 or 1)
4086 */
4087 int i;
4088 int rv;
4089
4090 if (part) {
4091 bsb.arraystart2 = __cpu_to_le64(0);
4092 bsb.length2 = __cpu_to_le64(0);
4093 } else {
4094 bsb.arraystart = __cpu_to_le64(0);
4095 bsb.length = __cpu_to_le64(0);
4096 }
4097 bsb.mtime = __cpu_to_le64(time(0));
4098 rv = 0;
4099 for (i = 0; i < dests; i++) {
4100 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4101 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4102 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4103 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4104 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4105 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4106 destoffsets[i]-4096)
4107 rv = -1;
4108 if (rv == 0 &&
4109 write(destfd[i], &bsb, 512) != 512)
4110 rv = -1;
4111 fsync(destfd[i]);
4112 }
4113 return rv;
4114 }
4115
4116 static void fail(char *msg)
4117 {
4118 int rv;
4119 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4120 rv |= (write(2, "\n", 1) != 1);
4121 exit(rv ? 1 : 2);
4122 }
4123
4124 static char *abuf, *bbuf;
4125 static unsigned long long abuflen;
4126 static void validate(int afd, int bfd, unsigned long long offset)
4127 {
4128 /* check that the data in the backup against the array.
4129 * This is only used for regression testing and should not
4130 * be used while the array is active
4131 */
4132 if (afd < 0)
4133 return;
4134 lseek64(bfd, offset - 4096, 0);
4135 if (read(bfd, &bsb2, 512) != 512)
4136 fail("cannot read bsb");
4137 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4138 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4139 fail("first csum bad");
4140 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4141 fail("magic is bad");
4142 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4143 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4144 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4145 fail("second csum bad");
4146
4147 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4148 fail("devstart is wrong");
4149
4150 if (bsb2.length) {
4151 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4152
4153 if (abuflen < len) {
4154 free(abuf);
4155 free(bbuf);
4156 abuflen = len;
4157 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4158 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4159 abuflen = 0;
4160 /* just stop validating on mem-alloc failure */
4161 return;
4162 }
4163 }
4164
4165 lseek64(bfd, offset, 0);
4166 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4167 //printf("len %llu\n", len);
4168 fail("read first backup failed");
4169 }
4170 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4171 if ((unsigned long long)read(afd, abuf, len) != len)
4172 fail("read first from array failed");
4173 if (memcmp(bbuf, abuf, len) != 0) {
4174 #if 0
4175 int i;
4176 printf("offset=%llu len=%llu\n",
4177 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4178 for (i=0; i<len; i++)
4179 if (bbuf[i] != abuf[i]) {
4180 printf("first diff byte %d\n", i);
4181 break;
4182 }
4183 #endif
4184 fail("data1 compare failed");
4185 }
4186 }
4187 if (bsb2.length2) {
4188 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4189
4190 if (abuflen < len) {
4191 free(abuf);
4192 free(bbuf);
4193 abuflen = len;
4194 abuf = xmalloc(abuflen);
4195 bbuf = xmalloc(abuflen);
4196 }
4197
4198 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4199 if ((unsigned long long)read(bfd, bbuf, len) != len)
4200 fail("read second backup failed");
4201 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4202 if ((unsigned long long)read(afd, abuf, len) != len)
4203 fail("read second from array failed");
4204 if (memcmp(bbuf, abuf, len) != 0)
4205 fail("data2 compare failed");
4206 }
4207 }
4208
4209 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4210 struct supertype *st, unsigned long blocks,
4211 int *fds, unsigned long long *offsets,
4212 int dests, int *destfd, unsigned long long *destoffsets)
4213 {
4214 /* Monitor a reshape where backup is being performed using
4215 * 'native' mechanism - either to a backup file, or
4216 * to some space in a spare.
4217 */
4218 char *buf;
4219 int degraded = -1;
4220 unsigned long long speed;
4221 unsigned long long suspend_point, array_size;
4222 unsigned long long backup_point, wait_point;
4223 unsigned long long reshape_completed;
4224 int done = 0;
4225 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4226 int part = 0; /* The next part of the backup area to fill. It may already
4227 * be full, so we need to check */
4228 int level = reshape->level;
4229 int layout = reshape->before.layout;
4230 int data = reshape->before.data_disks;
4231 int disks = reshape->before.data_disks + reshape->parity;
4232 int chunk = sra->array.chunk_size;
4233 struct mdinfo *sd;
4234 unsigned long stripes;
4235 int uuid[4];
4236 int frozen = 0;
4237
4238 /* set up the backup-super-block. This requires the
4239 * uuid from the array.
4240 */
4241 /* Find a superblock */
4242 for (sd = sra->devs; sd; sd = sd->next) {
4243 char *dn;
4244 int devfd;
4245 int ok;
4246 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4247 continue;
4248 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4249 devfd = dev_open(dn, O_RDONLY);
4250 if (devfd < 0)
4251 continue;
4252 ok = st->ss->load_super(st, devfd, NULL);
4253 close(devfd);
4254 if (ok == 0)
4255 break;
4256 }
4257 if (!sd) {
4258 pr_err("Cannot find a superblock\n");
4259 return 0;
4260 }
4261
4262 memset(&bsb, 0, 512);
4263 memcpy(bsb.magic, "md_backup_data-1", 16);
4264 st->ss->uuid_from_super(st, uuid);
4265 memcpy(bsb.set_uuid, uuid, 16);
4266 bsb.mtime = __cpu_to_le64(time(0));
4267 bsb.devstart2 = blocks;
4268
4269 stripes = blocks / (sra->array.chunk_size/512) /
4270 reshape->before.data_disks;
4271
4272 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4273 /* Don't start the 'reshape' */
4274 return 0;
4275 if (reshape->before.data_disks == reshape->after.data_disks) {
4276 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4277 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4278 }
4279
4280 if (increasing) {
4281 array_size = sra->component_size * reshape->after.data_disks;
4282 backup_point = sra->reshape_progress;
4283 suspend_point = 0;
4284 } else {
4285 array_size = sra->component_size * reshape->before.data_disks;
4286 backup_point = reshape->backup_blocks;
4287 suspend_point = array_size;
4288 }
4289
4290 while (!done) {
4291 int rv;
4292
4293 /* Want to return as soon the oldest backup slot can
4294 * be released as that allows us to start backing up
4295 * some more, providing suspend_point has been
4296 * advanced, which it should have.
4297 */
4298 if (increasing) {
4299 wait_point = array_size;
4300 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4301 wait_point = (__le64_to_cpu(bsb.arraystart) +
4302 __le64_to_cpu(bsb.length));
4303 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4304 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4305 __le64_to_cpu(bsb.length2));
4306 } else {
4307 wait_point = 0;
4308 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4309 wait_point = __le64_to_cpu(bsb.arraystart);
4310 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4311 wait_point = __le64_to_cpu(bsb.arraystart2);
4312 }
4313
4314 reshape_completed = sra->reshape_progress;
4315 rv = progress_reshape(sra, reshape,
4316 backup_point, wait_point,
4317 &suspend_point, &reshape_completed,
4318 &frozen);
4319 /* external metadata would need to ping_monitor here */
4320 sra->reshape_progress = reshape_completed;
4321
4322 /* Clear any backup region that is before 'here' */
4323 if (increasing) {
4324 if (__le64_to_cpu(bsb.length) > 0 &&
4325 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4326 __le64_to_cpu(bsb.length)))
4327 forget_backup(dests, destfd,
4328 destoffsets, 0);
4329 if (__le64_to_cpu(bsb.length2) > 0 &&
4330 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4331 __le64_to_cpu(bsb.length2)))
4332 forget_backup(dests, destfd,
4333 destoffsets, 1);
4334 } else {
4335 if (__le64_to_cpu(bsb.length) > 0 &&
4336 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4337 forget_backup(dests, destfd,
4338 destoffsets, 0);
4339 if (__le64_to_cpu(bsb.length2) > 0 &&
4340 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4341 forget_backup(dests, destfd,
4342 destoffsets, 1);
4343 }
4344 if (sigterm)
4345 rv = -2;
4346 if (rv < 0) {
4347 if (rv == -1)
4348 done = 1;
4349 break;
4350 }
4351 if (rv == 0 && increasing && !st->ss->external) {
4352 /* No longer need to monitor this reshape */
4353 sysfs_set_str(sra, NULL, "sync_max", "max");
4354 done = 1;
4355 break;
4356 }
4357
4358 while (rv) {
4359 unsigned long long offset;
4360 unsigned long actual_stripes;
4361 /* Need to backup some data.
4362 * If 'part' is not used and the desired
4363 * backup size is suspended, do a backup,
4364 * then consider the next part.
4365 */
4366 /* Check that 'part' is unused */
4367 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4368 break;
4369 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4370 break;
4371
4372 offset = backup_point / data;
4373 actual_stripes = stripes;
4374 if (increasing) {
4375 if (offset + actual_stripes * (chunk/512) >
4376 sra->component_size)
4377 actual_stripes = ((sra->component_size - offset)
4378 / (chunk/512));
4379 if (offset + actual_stripes * (chunk/512) >
4380 suspend_point/data)
4381 break;
4382 } else {
4383 if (offset < actual_stripes * (chunk/512))
4384 actual_stripes = offset / (chunk/512);
4385 offset -= actual_stripes * (chunk/512);
4386 if (offset < suspend_point/data)
4387 break;
4388 }
4389 if (actual_stripes == 0)
4390 break;
4391 grow_backup(sra, offset, actual_stripes,
4392 fds, offsets,
4393 disks, chunk, level, layout,
4394 dests, destfd, destoffsets,
4395 part, &degraded, buf);
4396 validate(afd, destfd[0], destoffsets[0]);
4397 /* record where 'part' is up to */
4398 part = !part;
4399 if (increasing)
4400 backup_point += actual_stripes * (chunk/512) * data;
4401 else
4402 backup_point -= actual_stripes * (chunk/512) * data;
4403 }
4404 }
4405
4406 /* FIXME maybe call progress_reshape one more time instead */
4407 /* remove any remaining suspension */
4408 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4409 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4410 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4411 sysfs_set_num(sra, NULL, "sync_min", 0);
4412
4413 if (reshape->before.data_disks == reshape->after.data_disks)
4414 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4415 free(buf);
4416 return done;
4417 }
4418
4419 /*
4420 * If any spare contains md_back_data-1 which is recent wrt mtime,
4421 * write that data into the array and update the super blocks with
4422 * the new reshape_progress
4423 */
4424 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4425 char *backup_file, int verbose)
4426 {
4427 int i, j;
4428 int old_disks;
4429 unsigned long long *offsets;
4430 unsigned long long nstripe, ostripe;
4431 int ndata, odata;
4432
4433 odata = info->array.raid_disks - info->delta_disks - 1;
4434 if (info->array.level == 6) odata--; /* number of data disks */
4435 ndata = info->array.raid_disks - 1;
4436 if (info->new_level == 6) ndata--;
4437
4438 old_disks = info->array.raid_disks - info->delta_disks;
4439
4440 if (info->delta_disks <= 0)
4441 /* Didn't grow, so the backup file must have
4442 * been used
4443 */
4444 old_disks = cnt;
4445 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4446 struct mdinfo dinfo;
4447 int fd;
4448 int bsbsize;
4449 char *devname, namebuf[20];
4450 unsigned long long lo, hi;
4451
4452 /* This was a spare and may have some saved data on it.
4453 * Load the superblock, find and load the
4454 * backup_super_block.
4455 * If either fail, go on to next device.
4456 * If the backup contains no new info, just return
4457 * else restore data and update all superblocks
4458 */
4459 if (i == old_disks-1) {
4460 fd = open(backup_file, O_RDONLY);
4461 if (fd<0) {
4462 pr_err("backup file %s inaccessible: %s\n",
4463 backup_file, strerror(errno));
4464 continue;
4465 }
4466 devname = backup_file;
4467 } else {
4468 fd = fdlist[i];
4469 if (fd < 0)
4470 continue;
4471 if (st->ss->load_super(st, fd, NULL))
4472 continue;
4473
4474 st->ss->getinfo_super(st, &dinfo, NULL);
4475 st->ss->free_super(st);
4476
4477 if (lseek64(fd,
4478 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4479 0) < 0) {
4480 pr_err("Cannot seek on device %d\n", i);
4481 continue; /* Cannot seek */
4482 }
4483 sprintf(namebuf, "device-%d", i);
4484 devname = namebuf;
4485 }
4486 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4487 if (verbose)
4488 pr_err("Cannot read from %s\n", devname);
4489 continue; /* Cannot read */
4490 }
4491 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4492 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4493 if (verbose)
4494 pr_err("No backup metadata on %s\n", devname);
4495 continue;
4496 }
4497 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4498 if (verbose)
4499 pr_err("Bad backup-metadata checksum on %s\n", devname);
4500 continue; /* bad checksum */
4501 }
4502 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4503 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4504 if (verbose)
4505 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4506 continue; /* Bad second checksum */
4507 }
4508 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4509 if (verbose)
4510 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4511 continue; /* Wrong uuid */
4512 }
4513
4514 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4515 * sometimes they aren't... So allow considerable flexability in matching, and allow
4516 * this test to be overridden by an environment variable.
4517 */
4518 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4519 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4520 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4521 pr_err("accepting backup with timestamp %lu "
4522 "for array with timestamp %lu\n",
4523 (unsigned long)__le64_to_cpu(bsb.mtime),
4524 (unsigned long)info->array.utime);
4525 } else {
4526 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4527 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4528 continue; /* time stamp is too bad */
4529 }
4530 }
4531
4532 if (bsb.magic[15] == '1') {
4533 if (bsb.length == 0)
4534 continue;
4535 if (info->delta_disks >= 0) {
4536 /* reshape_progress is increasing */
4537 if (__le64_to_cpu(bsb.arraystart)
4538 + __le64_to_cpu(bsb.length)
4539 < info->reshape_progress) {
4540 nonew:
4541 if (verbose)
4542 pr_err("backup-metadata found on %s but is not needed\n", devname);
4543 continue; /* No new data here */
4544 }
4545 } else {
4546 /* reshape_progress is decreasing */
4547 if (__le64_to_cpu(bsb.arraystart) >=
4548 info->reshape_progress)
4549 goto nonew; /* No new data here */
4550 }
4551 } else {
4552 if (bsb.length == 0 && bsb.length2 == 0)
4553 continue;
4554 if (info->delta_disks >= 0) {
4555 /* reshape_progress is increasing */
4556 if ((__le64_to_cpu(bsb.arraystart)
4557 + __le64_to_cpu(bsb.length)
4558 < info->reshape_progress)
4559 &&
4560 (__le64_to_cpu(bsb.arraystart2)
4561 + __le64_to_cpu(bsb.length2)
4562 < info->reshape_progress))
4563 goto nonew; /* No new data here */
4564 } else {
4565 /* reshape_progress is decreasing */
4566 if (__le64_to_cpu(bsb.arraystart) >=
4567 info->reshape_progress &&
4568 __le64_to_cpu(bsb.arraystart2) >=
4569 info->reshape_progress)
4570 goto nonew; /* No new data here */
4571 }
4572 }
4573 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4574 second_fail:
4575 if (verbose)
4576 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4577 devname);
4578 continue; /* Cannot seek */
4579 }
4580 /* There should be a duplicate backup superblock 4k before here */
4581 if (lseek64(fd, -4096, 1) < 0 ||
4582 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4583 goto second_fail; /* Cannot find leading superblock */
4584 if (bsb.magic[15] == '1')
4585 bsbsize = offsetof(struct mdp_backup_super, pad1);
4586 else
4587 bsbsize = offsetof(struct mdp_backup_super, pad);
4588 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4589 goto second_fail; /* Cannot find leading superblock */
4590
4591 /* Now need the data offsets for all devices. */
4592 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4593 for(j=0; j<info->array.raid_disks; j++) {
4594 if (fdlist[j] < 0)
4595 continue;
4596 if (st->ss->load_super(st, fdlist[j], NULL))
4597 /* FIXME should be this be an error */
4598 continue;
4599 st->ss->getinfo_super(st, &dinfo, NULL);
4600 st->ss->free_super(st);
4601 offsets[j] = dinfo.data_offset * 512;
4602 }
4603 printf(Name ": restoring critical section\n");
4604
4605 if (restore_stripes(fdlist, offsets,
4606 info->array.raid_disks,
4607 info->new_chunk,
4608 info->new_level,
4609 info->new_layout,
4610 fd, __le64_to_cpu(bsb.devstart)*512,
4611 __le64_to_cpu(bsb.arraystart)*512,
4612 __le64_to_cpu(bsb.length)*512, NULL)) {
4613 /* didn't succeed, so giveup */
4614 if (verbose)
4615 pr_err("Error restoring backup from %s\n",
4616 devname);
4617 free(offsets);
4618 return 1;
4619 }
4620
4621 if (bsb.magic[15] == '2' &&
4622 restore_stripes(fdlist, offsets,
4623 info->array.raid_disks,
4624 info->new_chunk,
4625 info->new_level,
4626 info->new_layout,
4627 fd, __le64_to_cpu(bsb.devstart)*512 +
4628 __le64_to_cpu(bsb.devstart2)*512,
4629 __le64_to_cpu(bsb.arraystart2)*512,
4630 __le64_to_cpu(bsb.length2)*512, NULL)) {
4631 /* didn't succeed, so giveup */
4632 if (verbose)
4633 pr_err("Error restoring second backup from %s\n",
4634 devname);
4635 free(offsets);
4636 return 1;
4637 }
4638
4639 free(offsets);
4640
4641 /* Ok, so the data is restored. Let's update those superblocks. */
4642
4643 lo = hi = 0;
4644 if (bsb.length) {
4645 lo = __le64_to_cpu(bsb.arraystart);
4646 hi = lo + __le64_to_cpu(bsb.length);
4647 }
4648 if (bsb.magic[15] == '2' && bsb.length2) {
4649 unsigned long long lo1, hi1;
4650 lo1 = __le64_to_cpu(bsb.arraystart2);
4651 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4652 if (lo == hi) {
4653 lo = lo1;
4654 hi = hi1;
4655 } else if (lo < lo1)
4656 hi = hi1;
4657 else
4658 lo = lo1;
4659 }
4660 if (lo < hi &&
4661 (info->reshape_progress < lo ||
4662 info->reshape_progress > hi))
4663 /* backup does not affect reshape_progress*/ ;
4664 else if (info->delta_disks >= 0) {
4665 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4666 __le64_to_cpu(bsb.length);
4667 if (bsb.magic[15] == '2') {
4668 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4669 __le64_to_cpu(bsb.length2);
4670 if (p2 > info->reshape_progress)
4671 info->reshape_progress = p2;
4672 }
4673 } else {
4674 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4675 if (bsb.magic[15] == '2') {
4676 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4677 if (p2 < info->reshape_progress)
4678 info->reshape_progress = p2;
4679 }
4680 }
4681 for (j=0; j<info->array.raid_disks; j++) {
4682 if (fdlist[j] < 0)
4683 continue;
4684 if (st->ss->load_super(st, fdlist[j], NULL))
4685 continue;
4686 st->ss->getinfo_super(st, &dinfo, NULL);
4687 dinfo.reshape_progress = info->reshape_progress;
4688 st->ss->update_super(st, &dinfo,
4689 "_reshape_progress",
4690 NULL,0, 0, NULL);
4691 st->ss->store_super(st, fdlist[j]);
4692 st->ss->free_super(st);
4693 }
4694 return 0;
4695 }
4696 /* Didn't find any backup data, try to see if any
4697 * was needed.
4698 */
4699 if (info->delta_disks < 0) {
4700 /* When shrinking, the critical section is at the end.
4701 * So see if we are before the critical section.
4702 */
4703 unsigned long long first_block;
4704 nstripe = ostripe = 0;
4705 first_block = 0;
4706 while (ostripe >= nstripe) {
4707 ostripe += info->array.chunk_size / 512;
4708 first_block = ostripe * odata;
4709 nstripe = first_block / ndata / (info->new_chunk/512) *
4710 (info->new_chunk/512);
4711 }
4712
4713 if (info->reshape_progress >= first_block)
4714 return 0;
4715 }
4716 if (info->delta_disks > 0) {
4717 /* See if we are beyond the critical section. */
4718 unsigned long long last_block;
4719 nstripe = ostripe = 0;
4720 last_block = 0;
4721 while (nstripe >= ostripe) {
4722 nstripe += info->new_chunk / 512;
4723 last_block = nstripe * ndata;
4724 ostripe = last_block / odata / (info->array.chunk_size/512) *
4725 (info->array.chunk_size/512);
4726 }
4727
4728 if (info->reshape_progress >= last_block)
4729 return 0;
4730 }
4731 /* needed to recover critical section! */
4732 if (verbose)
4733 pr_err("Failed to find backup of critical section\n");
4734 return 1;
4735 }
4736
4737 int Grow_continue_command(char *devname, int fd,
4738 char *backup_file, int verbose)
4739 {
4740 int ret_val = 0;
4741 struct supertype *st = NULL;
4742 struct mdinfo *content = NULL;
4743 struct mdinfo array;
4744 char *subarray = NULL;
4745 struct mdinfo *cc = NULL;
4746 struct mdstat_ent *mdstat = NULL;
4747 int cfd = -1;
4748 int fd2 = -1;
4749
4750 dprintf("Grow continue from command line called for %s\n",
4751 devname);
4752
4753 st = super_by_fd(fd, &subarray);
4754 if (!st || !st->ss) {
4755 pr_err("Unable to determine metadata format for %s\n",
4756 devname);
4757 return 1;
4758 }
4759 dprintf("Grow continue is run for ");
4760 if (st->ss->external == 0) {
4761 int d;
4762 dprintf("native array (%s)\n", devname);
4763 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4764 pr_err("%s is not an active md array -"
4765 " aborting\n", devname);
4766 ret_val = 1;
4767 goto Grow_continue_command_exit;
4768 }
4769 content = &array;
4770 /* Need to load a superblock.
4771 * FIXME we should really get what we need from
4772 * sysfs
4773 */
4774 for (d = 0; d < MAX_DISKS; d++) {
4775 mdu_disk_info_t disk;
4776 char *dv;
4777 int err;
4778 disk.number = d;
4779 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4780 continue;
4781 if (disk.major == 0 && disk.minor == 0)
4782 continue;
4783 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4784 continue;
4785 dv = map_dev(disk.major, disk.minor, 1);
4786 if (!dv)
4787 continue;
4788 fd2 = dev_open(dv, O_RDONLY);
4789 if (fd2 < 0)
4790 continue;
4791 err = st->ss->load_super(st, fd2, NULL);
4792 close(fd2);
4793 /* invalidate fd2 to avoid possible double close() */
4794 fd2 = -1;
4795 if (err)
4796 continue;
4797 break;
4798 }
4799 if (d == MAX_DISKS) {
4800 pr_err("Unable to load metadata for %s\n",
4801 devname);
4802 ret_val = 1;
4803 goto Grow_continue_command_exit;
4804 }
4805 st->ss->getinfo_super(st, content, NULL);
4806 } else {
4807 char *container;
4808
4809 if (subarray) {
4810 dprintf("subarray (%s)\n", subarray);
4811 container = st->container_devnm;
4812 cfd = open_dev_excl(st->container_devnm);
4813 } else {
4814 container = st->devnm;
4815 close(fd);
4816 cfd = open_dev_excl(st->devnm);
4817 dprintf("container (%s)\n", container);
4818 fd = cfd;
4819 }
4820 if (cfd < 0) {
4821 pr_err("Unable to open container "
4822 "for %s\n", devname);
4823 ret_val = 1;
4824 goto Grow_continue_command_exit;
4825 }
4826
4827 /* find in container array under reshape
4828 */
4829 ret_val = st->ss->load_container(st, cfd, NULL);
4830 if (ret_val) {
4831 pr_err("Cannot read superblock for %s\n",
4832 devname);
4833 ret_val = 1;
4834 goto Grow_continue_command_exit;
4835 }
4836
4837 cc = st->ss->container_content(st, subarray);
4838 for (content = cc; content ; content = content->next) {
4839 char *array;
4840 int allow_reshape = 1;
4841
4842 if (content->reshape_active == 0)
4843 continue;
4844 /* The decision about array or container wide
4845 * reshape is taken in Grow_continue based
4846 * content->reshape_active state, therefore we
4847 * need to check_reshape based on
4848 * reshape_active and subarray name
4849 */
4850 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4851 allow_reshape = 0;
4852 if (content->reshape_active == CONTAINER_RESHAPE &&
4853 (content->array.state
4854 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4855 allow_reshape = 0;
4856
4857 if (!allow_reshape) {
4858 pr_err("cannot continue reshape of an array"
4859 " in container with unsupported"
4860 " metadata: %s(%s)\n",
4861 devname, container);
4862 ret_val = 1;
4863 goto Grow_continue_command_exit;
4864 }
4865
4866 array = strchr(content->text_version+1, '/')+1;
4867 mdstat = mdstat_by_subdev(array, container);
4868 if (!mdstat)
4869 continue;
4870 if (mdstat->active == 0) {
4871 pr_err("Skipping inactive array %s.\n",
4872 mdstat->devnm);
4873 free_mdstat(mdstat);
4874 mdstat = NULL;
4875 continue;
4876 }
4877 break;
4878 }
4879 if (!content) {
4880 pr_err("Unable to determine reshaped "
4881 "array for %s\n", devname);
4882 ret_val = 1;
4883 goto Grow_continue_command_exit;
4884 }
4885 fd2 = open_dev(mdstat->devnm);
4886 if (fd2 < 0) {
4887 pr_err("cannot open (%s)\n", mdstat->devnm);
4888 ret_val = 1;
4889 goto Grow_continue_command_exit;
4890 }
4891
4892 sysfs_init(content, fd2, mdstat->devnm);
4893
4894 /* start mdmon in case it is not running
4895 */
4896 if (!mdmon_running(container))
4897 start_mdmon(container);
4898 ping_monitor(container);
4899
4900 if (mdmon_running(container))
4901 st->update_tail = &st->updates;
4902 else {
4903 pr_err("No mdmon found. "
4904 "Grow cannot continue.\n");
4905 ret_val = 1;
4906 goto Grow_continue_command_exit;
4907 }
4908 }
4909
4910 /* verify that array under reshape is started from
4911 * correct position
4912 */
4913 if (verify_reshape_position(content, content->array.level) < 0) {
4914 ret_val = 1;
4915 goto Grow_continue_command_exit;
4916 }
4917
4918 /* continue reshape
4919 */
4920 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4921
4922 Grow_continue_command_exit:
4923 if (fd2 > -1)
4924 close(fd2);
4925 if (cfd > -1)
4926 close(cfd);
4927 st->ss->free_super(st);
4928 free_mdstat(mdstat);
4929 sysfs_free(cc);
4930 free(subarray);
4931
4932 return ret_val;
4933 }
4934
4935 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4936 char *backup_file, int forked, int freeze_reshape)
4937 {
4938 int ret_val = 2;
4939
4940 if (!info->reshape_active)
4941 return ret_val;
4942
4943 if (st->ss->external) {
4944 int cfd = open_dev(st->container_devnm);
4945
4946 if (cfd < 0)
4947 return 1;
4948
4949 st->ss->load_container(st, cfd, st->container_devnm);
4950 close(cfd);
4951 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4952 st, info, 0, backup_file,
4953 0, 1,
4954 1 | info->reshape_active,
4955 freeze_reshape);
4956 } else
4957 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4958 NULL, INVALID_SECTORS,
4959 backup_file, 0, forked,
4960 1 | info->reshape_active,
4961 freeze_reshape);
4962
4963 return ret_val;
4964 }
4965
4966 char *make_backup(char *name)
4967 {
4968 char *base = "backup_file-";
4969 int len;
4970 char *fname;
4971
4972 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4973 fname = xmalloc(len);
4974 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4975 return fname;
4976 }
4977
4978 char *locate_backup(char *name)
4979 {
4980 char *fl = make_backup(name);
4981 struct stat stb;
4982
4983 if (stat(fl, &stb) == 0 &&
4984 S_ISREG(stb.st_mode))
4985 return fl;
4986
4987 free(fl);
4988 return NULL;
4989 }
Unnamed repository; edit this file 'description' to name the repository.