]> 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
Makefile: install mdadm-grow-continue@.service
[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 if (symlink(backup_file, bu))
944 pr_err("Recording backup file in " MAP_DIR "failed: %s\n",
945 strerror(errno));
946 free(bu);
947 }
948
949 return 1;
950 }
951
952 unsigned long compute_backup_blocks(int nchunk, int ochunk,
953 unsigned int ndata, unsigned int odata)
954 {
955 unsigned long a, b, blocks;
956 /* So how much do we need to backup.
957 * We need an amount of data which is both a whole number of
958 * old stripes and a whole number of new stripes.
959 * So LCM for (chunksize*datadisks).
960 */
961 a = (ochunk/512) * odata;
962 b = (nchunk/512) * ndata;
963 /* Find GCD */
964 a = GCD(a, b);
965 /* LCM == product / GCD */
966 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
967
968 return blocks;
969 }
970
971 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
972 {
973 /* Based on the current array state in info->array and
974 * the changes in info->new_* etc, determine:
975 * - whether the change is possible
976 * - Intermediate level/raid_disks/layout
977 * - whether a restriping reshape is needed
978 * - number of sectors in minimum change unit. This
979 * will cover a whole number of stripes in 'before' and
980 * 'after'.
981 *
982 * Return message if the change should be rejected
983 * NULL if the change can be achieved
984 *
985 * This can be called as part of starting a reshape, or
986 * when assembling an array that is undergoing reshape.
987 */
988 int near, far, offset, copies;
989 int new_disks;
990 int old_chunk, new_chunk;
991 /* delta_parity records change in number of devices
992 * caused by level change
993 */
994 int delta_parity = 0;
995
996 memset(re, 0, sizeof(*re));
997
998 /* If a new level not explicitly given, we assume no-change */
999 if (info->new_level == UnSet)
1000 info->new_level = info->array.level;
1001
1002 if (info->new_chunk)
1003 switch (info->new_level) {
1004 case 0:
1005 case 4:
1006 case 5:
1007 case 6:
1008 case 10:
1009 /* chunk size is meaningful, must divide component_size
1010 * evenly
1011 */
1012 if (info->component_size % (info->new_chunk/512)) {
1013 unsigned long long shrink = info->component_size;
1014 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1015 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1016 info->new_chunk/1024, info->component_size/2);
1017 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1018 devname, shrink/2);
1019 pr_err("will shrink the array so the given chunk size would work.\n");
1020 return "";
1021 }
1022 break;
1023 default:
1024 return "chunk size not meaningful for this level";
1025 }
1026 else
1027 info->new_chunk = info->array.chunk_size;
1028
1029 switch (info->array.level) {
1030 default:
1031 return "Cannot understand this RAID level";
1032 case 1:
1033 /* RAID1 can convert to RAID1 with different disks, or
1034 * raid5 with 2 disks, or
1035 * raid0 with 1 disk
1036 */
1037 if (info->new_level > 1 &&
1038 (info->component_size & 7))
1039 return "Cannot convert RAID1 of this size - "
1040 "reduce size to multiple of 4K first.";
1041 if (info->new_level == 0) {
1042 if (info->delta_disks != UnSet &&
1043 info->delta_disks != 0)
1044 return "Cannot change number of disks "
1045 "with RAID1->RAID0 conversion";
1046 re->level = 0;
1047 re->before.data_disks = 1;
1048 re->after.data_disks = 1;
1049 return NULL;
1050 }
1051 if (info->new_level == 1) {
1052 if (info->delta_disks == UnSet)
1053 /* Don't know what to do */
1054 return "no change requested for Growing RAID1";
1055 re->level = 1;
1056 return NULL;
1057 }
1058 if (info->array.raid_disks == 2 &&
1059 info->new_level == 5) {
1060
1061 re->level = 5;
1062 re->before.data_disks = 1;
1063 if (info->delta_disks != UnSet &&
1064 info->delta_disks != 0)
1065 re->after.data_disks = 1 + info->delta_disks;
1066 else
1067 re->after.data_disks = 1;
1068 if (re->after.data_disks < 1)
1069 return "Number of disks too small for RAID5";
1070
1071 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1072 info->array.chunk_size = 65536;
1073 break;
1074 }
1075 /* Could do some multi-stage conversions, but leave that to
1076 * later.
1077 */
1078 return "Impossibly level change request for RAID1";
1079
1080 case 10:
1081 /* RAID10 can be converted from near mode to
1082 * RAID0 by removing some devices.
1083 * It can also be reshaped if the kernel supports
1084 * new_data_offset.
1085 */
1086 switch (info->new_level) {
1087 case 0:
1088 if ((info->array.layout & ~0xff) != 0x100)
1089 return "Cannot Grow RAID10 with far/offset layout";
1090 /* number of devices must be multiple of number of copies */
1091 if (info->array.raid_disks % (info->array.layout & 0xff))
1092 return "RAID10 layout too complex for Grow operation";
1093
1094 new_disks = (info->array.raid_disks
1095 / (info->array.layout & 0xff));
1096 if (info->delta_disks == UnSet)
1097 info->delta_disks = (new_disks
1098 - info->array.raid_disks);
1099
1100 if (info->delta_disks != new_disks - info->array.raid_disks)
1101 return "New number of raid-devices impossible for RAID10";
1102 if (info->new_chunk &&
1103 info->new_chunk != info->array.chunk_size)
1104 return "Cannot change chunk-size with RAID10 Grow";
1105
1106 /* looks good */
1107 re->level = 0;
1108 re->before.data_disks = new_disks;
1109 re->after.data_disks = re->before.data_disks;
1110 return NULL;
1111
1112 case 10:
1113 near = info->array.layout & 0xff;
1114 far = (info->array.layout >> 8) & 0xff;
1115 offset = info->array.layout & 0x10000;
1116 if (far > 1 && !offset)
1117 return "Cannot reshape RAID10 in far-mode";
1118 copies = near * far;
1119
1120 old_chunk = info->array.chunk_size * far;
1121
1122 if (info->new_layout == UnSet)
1123 info->new_layout = info->array.layout;
1124 else {
1125 near = info->new_layout & 0xff;
1126 far = (info->new_layout >> 8) & 0xff;
1127 offset = info->new_layout & 0x10000;
1128 if (far > 1 && !offset)
1129 return "Cannot reshape RAID10 to far-mode";
1130 if (near * far != copies)
1131 return "Cannot change number of copies"
1132 " when reshaping RAID10";
1133 }
1134 if (info->delta_disks == UnSet)
1135 info->delta_disks = 0;
1136 new_disks = (info->array.raid_disks +
1137 info->delta_disks);
1138
1139 new_chunk = info->new_chunk * far;
1140
1141 re->level = 10;
1142 re->before.layout = info->array.layout;
1143 re->before.data_disks = info->array.raid_disks;
1144 re->after.layout = info->new_layout;
1145 re->after.data_disks = new_disks;
1146 /* For RAID10 we don't do backup but do allow reshape,
1147 * so set backup_blocks to INVALID_SECTORS rather than
1148 * zero.
1149 * And there is no need to synchronise stripes on both
1150 * 'old' and 'new'. So the important
1151 * number is the minimum data_offset difference
1152 * which is the larger of (offset copies * chunk).
1153 */
1154 re->backup_blocks = INVALID_SECTORS;
1155 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1156 if (new_disks < re->before.data_disks &&
1157 info->space_after < re->min_offset_change)
1158 /* Reduce component size by one chunk */
1159 re->new_size = (info->component_size -
1160 re->min_offset_change);
1161 else
1162 re->new_size = info->component_size;
1163 re->new_size = re->new_size * new_disks / copies;
1164 return NULL;
1165
1166 default:
1167 return "RAID10 can only be changed to RAID0";
1168 }
1169 case 0:
1170 /* RAID0 can be converted to RAID10, or to RAID456 */
1171 if (info->new_level == 10) {
1172 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1173 /* Assume near=2 layout */
1174 info->new_layout = 0x102;
1175 info->delta_disks = info->array.raid_disks;
1176 }
1177 if (info->new_layout == UnSet) {
1178 int copies = 1 + (info->delta_disks
1179 / info->array.raid_disks);
1180 if (info->array.raid_disks * (copies-1)
1181 != info->delta_disks)
1182 return "Impossible number of devices"
1183 " for RAID0->RAID10";
1184 info->new_layout = 0x100 + copies;
1185 }
1186 if (info->delta_disks == UnSet) {
1187 int copies = info->new_layout & 0xff;
1188 if (info->new_layout != 0x100 + copies)
1189 return "New layout impossible"
1190 " for RAID0->RAID10";;
1191 info->delta_disks = (copies - 1) *
1192 info->array.raid_disks;
1193 }
1194 if (info->new_chunk &&
1195 info->new_chunk != info->array.chunk_size)
1196 return "Cannot change chunk-size with RAID0->RAID10";
1197 /* looks good */
1198 re->level = 10;
1199 re->before.data_disks = (info->array.raid_disks +
1200 info->delta_disks);
1201 re->after.data_disks = re->before.data_disks;
1202 re->before.layout = info->new_layout;
1203 return NULL;
1204 }
1205
1206 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1207 * a raid4 style layout of the final level.
1208 */
1209 switch (info->new_level) {
1210 case 4:
1211 delta_parity = 1;
1212 case 0:
1213 re->level = 4;
1214 re->before.layout = 0;
1215 break;
1216 case 5:
1217 delta_parity = 1;
1218 re->level = 5;
1219 re->before.layout = ALGORITHM_PARITY_N;
1220 if (info->new_layout == UnSet)
1221 info->new_layout = map_name(r5layout, "default");
1222 break;
1223 case 6:
1224 delta_parity = 2;
1225 re->level = 6;
1226 re->before.layout = ALGORITHM_PARITY_N;
1227 if (info->new_layout == UnSet)
1228 info->new_layout = map_name(r6layout, "default");
1229 break;
1230 default:
1231 return "Impossible level change requested";
1232 }
1233 re->before.data_disks = info->array.raid_disks;
1234 /* determining 'after' layout happens outside this 'switch' */
1235 break;
1236
1237 case 4:
1238 info->array.layout = ALGORITHM_PARITY_N;
1239 case 5:
1240 switch (info->new_level) {
1241 case 0:
1242 delta_parity = -1;
1243 case 4:
1244 re->level = info->array.level;
1245 re->before.data_disks = info->array.raid_disks - 1;
1246 re->before.layout = info->array.layout;
1247 break;
1248 case 5:
1249 re->level = 5;
1250 re->before.data_disks = info->array.raid_disks - 1;
1251 re->before.layout = info->array.layout;
1252 break;
1253 case 6:
1254 delta_parity = 1;
1255 re->level = 6;
1256 re->before.data_disks = info->array.raid_disks - 1;
1257 switch (info->array.layout) {
1258 case ALGORITHM_LEFT_ASYMMETRIC:
1259 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1260 break;
1261 case ALGORITHM_RIGHT_ASYMMETRIC:
1262 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1263 break;
1264 case ALGORITHM_LEFT_SYMMETRIC:
1265 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1266 break;
1267 case ALGORITHM_RIGHT_SYMMETRIC:
1268 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1269 break;
1270 case ALGORITHM_PARITY_0:
1271 re->before.layout = ALGORITHM_PARITY_0_6;
1272 break;
1273 case ALGORITHM_PARITY_N:
1274 re->before.layout = ALGORITHM_PARITY_N_6;
1275 break;
1276 default:
1277 return "Cannot convert an array with this layout";
1278 }
1279 break;
1280 case 1:
1281 if (info->array.raid_disks != 2)
1282 return "Can only convert a 2-device array to RAID1";
1283 if (info->delta_disks != UnSet &&
1284 info->delta_disks != 0)
1285 return "Cannot set raid_disk when "
1286 "converting RAID5->RAID1";
1287 re->level = 1;
1288 info->new_chunk = 0;
1289 return NULL;
1290 default:
1291 return "Impossible level change requested";
1292 }
1293 break;
1294 case 6:
1295 switch (info->new_level) {
1296 case 4:
1297 case 5:
1298 delta_parity = -1;
1299 case 6:
1300 re->level = 6;
1301 re->before.data_disks = info->array.raid_disks - 2;
1302 re->before.layout = info->array.layout;
1303 break;
1304 default:
1305 return "Impossible level change requested";
1306 }
1307 break;
1308 }
1309
1310 /* If we reached here then it looks like a re-stripe is
1311 * happening. We have determined the intermediate level
1312 * and initial raid_disks/layout and stored these in 're'.
1313 *
1314 * We need to deduce the final layout that can be atomically
1315 * converted to the end state.
1316 */
1317 switch (info->new_level) {
1318 case 0:
1319 /* We can only get to RAID0 from RAID4 or RAID5
1320 * with appropriate layout and one extra device
1321 */
1322 if (re->level != 4 && re->level != 5)
1323 return "Cannot covert to RAID0 from this level";
1324
1325 switch (re->level) {
1326 case 4:
1327 re->after.layout = 0;
1328 break;
1329 case 5:
1330 re->after.layout = ALGORITHM_PARITY_N;
1331 break;
1332 }
1333 break;
1334
1335 case 4:
1336 /* We can only get to RAID4 from RAID5 */
1337 if (re->level != 4 && re->level != 5)
1338 return "Cannot convert to RAID4 from this level";
1339
1340 switch (re->level) {
1341 case 4:
1342 re->before.layout = 0;
1343 re->after.layout = 0;
1344 break;
1345 case 5:
1346 re->after.layout = ALGORITHM_PARITY_N;
1347 break;
1348 }
1349 break;
1350
1351 case 5:
1352 /* We get to RAID5 from RAID5 or RAID6 */
1353 if (re->level != 5 && re->level != 6)
1354 return "Cannot convert to RAID5 from this level";
1355
1356 switch (re->level) {
1357 case 5:
1358 if (info->new_layout == UnSet)
1359 re->after.layout = re->before.layout;
1360 else
1361 re->after.layout = info->new_layout;
1362 break;
1363 case 6:
1364 if (info->new_layout == UnSet)
1365 info->new_layout = re->before.layout;
1366
1367 /* after.layout needs to be raid6 version of new_layout */
1368 if (info->new_layout == ALGORITHM_PARITY_N)
1369 re->after.layout = ALGORITHM_PARITY_N;
1370 else {
1371 char layout[40];
1372 char *ls = map_num(r5layout, info->new_layout);
1373 int l;
1374 if (ls) {
1375 /* Current RAID6 layout has a RAID5
1376 * equivalent - good
1377 */
1378 strcat(strcpy(layout, ls), "-6");
1379 l = map_name(r6layout, layout);
1380 if (l == UnSet)
1381 return "Cannot find RAID6 layout"
1382 " to convert to";
1383 } else {
1384 /* Current RAID6 has no equivalent.
1385 * If it is already a '-6' layout we
1386 * can leave it unchanged, else we must
1387 * fail
1388 */
1389 ls = map_num(r6layout, info->new_layout);
1390 if (!ls ||
1391 strcmp(ls+strlen(ls)-2, "-6") != 0)
1392 return "Please specify new layout";
1393 l = info->new_layout;
1394 }
1395 re->after.layout = l;
1396 }
1397 }
1398 break;
1399
1400 case 6:
1401 /* We must already be at level 6 */
1402 if (re->level != 6)
1403 return "Impossible level change";
1404 if (info->new_layout == UnSet)
1405 re->after.layout = info->array.layout;
1406 else
1407 re->after.layout = info->new_layout;
1408 break;
1409 default:
1410 return "Impossible level change requested";
1411 }
1412 if (info->delta_disks == UnSet)
1413 info->delta_disks = delta_parity;
1414
1415 re->after.data_disks = (re->before.data_disks
1416 + info->delta_disks
1417 - delta_parity);
1418 switch (re->level) {
1419 case 6: re->parity = 2;
1420 break;
1421 case 4:
1422 case 5: re->parity = 1;
1423 break;
1424 default: re->parity = 0;
1425 break;
1426 }
1427 /* So we have a restripe operation, we need to calculate the number
1428 * of blocks per reshape operation.
1429 */
1430 re->new_size = info->component_size * re->before.data_disks;
1431 if (info->new_chunk == 0)
1432 info->new_chunk = info->array.chunk_size;
1433 if (re->after.data_disks == re->before.data_disks &&
1434 re->after.layout == re->before.layout &&
1435 info->new_chunk == info->array.chunk_size) {
1436 /* Nothing to change, can change level immediately. */
1437 re->level = info->new_level;
1438 re->backup_blocks = 0;
1439 return NULL;
1440 }
1441 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1442 /* chunk and layout changes make no difference */
1443 re->level = info->new_level;
1444 re->backup_blocks = 0;
1445 return NULL;
1446 }
1447
1448 if (re->after.data_disks == re->before.data_disks &&
1449 get_linux_version() < 2006032)
1450 return "in-place reshape is not safe before 2.6.32 - sorry.";
1451
1452 if (re->after.data_disks < re->before.data_disks &&
1453 get_linux_version() < 2006030)
1454 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1455
1456 re->backup_blocks = compute_backup_blocks(
1457 info->new_chunk, info->array.chunk_size,
1458 re->after.data_disks,
1459 re->before.data_disks);
1460 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1461
1462 re->new_size = info->component_size * re->after.data_disks;
1463 return NULL;
1464 }
1465
1466 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1467 char *text_version)
1468 {
1469 struct mdinfo *info;
1470 char *subarray;
1471 int ret_val = -1;
1472
1473 if ((st == NULL) || (sra == NULL))
1474 return ret_val;
1475
1476 if (text_version == NULL)
1477 text_version = sra->text_version;
1478 subarray = strchr(text_version+1, '/')+1;
1479 info = st->ss->container_content(st, subarray);
1480 if (info) {
1481 unsigned long long current_size = 0;
1482 unsigned long long new_size =
1483 info->custom_array_size/2;
1484
1485 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1486 new_size > current_size) {
1487 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1488 < 0)
1489 dprintf("Error: Cannot set array size");
1490 else {
1491 ret_val = 0;
1492 dprintf("Array size changed");
1493 }
1494 dprintf(" from %llu to %llu.\n",
1495 current_size, new_size);
1496 }
1497 sysfs_free(info);
1498 } else
1499 dprintf("Error: set_array_size(): info pointer in NULL\n");
1500
1501 return ret_val;
1502 }
1503
1504 static int reshape_array(char *container, int fd, char *devname,
1505 struct supertype *st, struct mdinfo *info,
1506 int force, struct mddev_dev *devlist,
1507 unsigned long long data_offset,
1508 char *backup_file, int verbose, int forked,
1509 int restart, int freeze_reshape);
1510 static int reshape_container(char *container, char *devname,
1511 int mdfd,
1512 struct supertype *st,
1513 struct mdinfo *info,
1514 int force,
1515 char *backup_file, int verbose,
1516 int forked, int restart, int freeze_reshape);
1517
1518 int Grow_reshape(char *devname, int fd,
1519 struct mddev_dev *devlist,
1520 unsigned long long data_offset,
1521 struct context *c, struct shape *s)
1522 {
1523 /* Make some changes in the shape of an array.
1524 * The kernel must support the change.
1525 *
1526 * There are three different changes. Each can trigger
1527 * a resync or recovery so we freeze that until we have
1528 * requested everything (if kernel supports freezing - 2.6.30).
1529 * The steps are:
1530 * - change size (i.e. component_size)
1531 * - change level
1532 * - change layout/chunksize/ndisks
1533 *
1534 * The last can require a reshape. It is different on different
1535 * levels so we need to check the level before actioning it.
1536 * Some times the level change needs to be requested after the
1537 * reshape (e.g. raid6->raid5, raid5->raid0)
1538 *
1539 */
1540 struct mdu_array_info_s array;
1541 int rv = 0;
1542 struct supertype *st;
1543 char *subarray = NULL;
1544
1545 int frozen;
1546 int changed = 0;
1547 char *container = NULL;
1548 int cfd = -1;
1549
1550 struct mddev_dev *dv;
1551 int added_disks;
1552
1553 struct mdinfo info;
1554 struct mdinfo *sra;
1555
1556 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1557 pr_err("%s is not an active md array - aborting\n",
1558 devname);
1559 return 1;
1560 }
1561 if (data_offset != INVALID_SECTORS && array.level != 10
1562 && (array.level < 4 || array.level > 6)) {
1563 pr_err("--grow --data-offset not yet supported\n");
1564 return 1;
1565 }
1566
1567 if (s->size > 0 &&
1568 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1569 pr_err("cannot change component size at the same time "
1570 "as other changes.\n"
1571 " Change size first, then check data is intact before "
1572 "making other changes.\n");
1573 return 1;
1574 }
1575
1576 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1577 get_linux_version() < 2006032 &&
1578 !check_env("MDADM_FORCE_FEWER")) {
1579 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1580 " Please use a newer kernel\n");
1581 return 1;
1582 }
1583
1584 st = super_by_fd(fd, &subarray);
1585 if (!st) {
1586 pr_err("Unable to determine metadata format for %s\n", devname);
1587 return 1;
1588 }
1589 if (s->raiddisks > st->max_devs) {
1590 pr_err("Cannot increase raid-disks on this array"
1591 " beyond %d\n", st->max_devs);
1592 return 1;
1593 }
1594
1595 /* in the external case we need to check that the requested reshape is
1596 * supported, and perform an initial check that the container holds the
1597 * pre-requisite spare devices (mdmon owns final validation)
1598 */
1599 if (st->ss->external) {
1600 int rv;
1601
1602 if (subarray) {
1603 container = st->container_devnm;
1604 cfd = open_dev_excl(st->container_devnm);
1605 } else {
1606 container = st->devnm;
1607 close(fd);
1608 cfd = open_dev_excl(st->devnm);
1609 fd = cfd;
1610 }
1611 if (cfd < 0) {
1612 pr_err("Unable to open container for %s\n",
1613 devname);
1614 free(subarray);
1615 return 1;
1616 }
1617
1618 rv = st->ss->load_container(st, cfd, NULL);
1619
1620 if (rv) {
1621 pr_err("Cannot read superblock for %s\n",
1622 devname);
1623 free(subarray);
1624 return 1;
1625 }
1626
1627 /* check if operation is supported for metadata handler */
1628 if (st->ss->container_content) {
1629 struct mdinfo *cc = NULL;
1630 struct mdinfo *content = NULL;
1631
1632 cc = st->ss->container_content(st, subarray);
1633 for (content = cc; content ; content = content->next) {
1634 int allow_reshape = 1;
1635
1636 /* check if reshape is allowed based on metadata
1637 * indications stored in content.array.status
1638 */
1639 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1640 allow_reshape = 0;
1641 if (content->array.state
1642 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1643 allow_reshape = 0;
1644 if (!allow_reshape) {
1645 pr_err("cannot reshape arrays in"
1646 " container with unsupported"
1647 " metadata: %s(%s)\n",
1648 devname, container);
1649 sysfs_free(cc);
1650 free(subarray);
1651 return 1;
1652 }
1653 }
1654 sysfs_free(cc);
1655 }
1656 if (mdmon_running(container))
1657 st->update_tail = &st->updates;
1658 }
1659
1660 added_disks = 0;
1661 for (dv = devlist; dv; dv = dv->next)
1662 added_disks++;
1663 if (s->raiddisks > array.raid_disks &&
1664 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1665 !c->force) {
1666 pr_err("Need %d spare%s to avoid degraded array,"
1667 " and only have %d.\n"
1668 " Use --force to over-ride this check.\n",
1669 s->raiddisks - array.raid_disks,
1670 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1671 array.spare_disks + added_disks);
1672 return 1;
1673 }
1674
1675 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1676 | GET_STATE | GET_VERSION);
1677 if (sra) {
1678 if (st->ss->external && subarray == NULL) {
1679 array.level = LEVEL_CONTAINER;
1680 sra->array.level = LEVEL_CONTAINER;
1681 }
1682 } else {
1683 pr_err("failed to read sysfs parameters for %s\n",
1684 devname);
1685 return 1;
1686 }
1687 frozen = freeze(st);
1688 if (frozen < -1) {
1689 /* freeze() already spewed the reason */
1690 sysfs_free(sra);
1691 return 1;
1692 } else if (frozen < 0) {
1693 pr_err("%s is performing resync/recovery and cannot"
1694 " be reshaped\n", devname);
1695 sysfs_free(sra);
1696 return 1;
1697 }
1698
1699 /* ========= set size =============== */
1700 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1701 unsigned long long orig_size = get_component_size(fd)/2;
1702 unsigned long long min_csize;
1703 struct mdinfo *mdi;
1704 int raid0_takeover = 0;
1705
1706 if (orig_size == 0)
1707 orig_size = (unsigned) array.size;
1708
1709 if (orig_size == 0) {
1710 pr_err("Cannot set device size in this type of array.\n");
1711 rv = 1;
1712 goto release;
1713 }
1714
1715 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1716 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1717 rv = 1;
1718 goto release;
1719 }
1720 sync_metadata(st);
1721 if (st->ss->external) {
1722 /* metadata can have size limitation
1723 * update size value according to metadata information
1724 */
1725 struct mdinfo *sizeinfo =
1726 st->ss->container_content(st, subarray);
1727 if (sizeinfo) {
1728 unsigned long long new_size =
1729 sizeinfo->custom_array_size/2;
1730 int data_disks = get_data_disks(
1731 sizeinfo->array.level,
1732 sizeinfo->array.layout,
1733 sizeinfo->array.raid_disks);
1734 new_size /= data_disks;
1735 dprintf("Metadata size correction from %llu to "
1736 "%llu (%llu)\n", orig_size, new_size,
1737 new_size * data_disks);
1738 s->size = new_size;
1739 sysfs_free(sizeinfo);
1740 }
1741 }
1742
1743 /* Update the size of each member device in case
1744 * they have been resized. This will never reduce
1745 * below the current used-size. The "size" attribute
1746 * understands '0' to mean 'max'.
1747 */
1748 min_csize = 0;
1749 rv = 0;
1750 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1751 if (sysfs_set_num(sra, mdi, "size",
1752 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1753 /* Probably kernel refusing to let us
1754 * reduce the size - not an error.
1755 */
1756 break;
1757 }
1758 if (array.not_persistent == 0 &&
1759 array.major_version == 0 &&
1760 get_linux_version() < 3001000) {
1761 /* Dangerous to allow size to exceed 2TB */
1762 unsigned long long csize;
1763 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1764 if (csize >= 2ULL*1024*1024*1024)
1765 csize = 2ULL*1024*1024*1024;
1766 if ((min_csize == 0 || (min_csize
1767 > csize)))
1768 min_csize = csize;
1769 }
1770 }
1771 }
1772 if (rv) {
1773 pr_err("Cannot set size on "
1774 "array members.\n");
1775 goto size_change_error;
1776 }
1777 if (min_csize && s->size > min_csize) {
1778 pr_err("Cannot safely make this array "
1779 "use more than 2TB per device on this kernel.\n");
1780 rv = 1;
1781 goto size_change_error;
1782 }
1783 if (min_csize && s->size == MAX_SIZE) {
1784 /* Don't let the kernel choose a size - it will get
1785 * it wrong
1786 */
1787 pr_err("Limited v0.90 array to "
1788 "2TB per device\n");
1789 s->size = min_csize;
1790 }
1791 if (st->ss->external) {
1792 if (sra->array.level == 0) {
1793 rv = sysfs_set_str(sra, NULL, "level",
1794 "raid5");
1795 if (!rv) {
1796 raid0_takeover = 1;
1797 /* get array parametes after takeover
1798 * to chane one parameter at time only
1799 */
1800 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1801 }
1802 }
1803 /* make sure mdmon is
1804 * aware of the new level */
1805 if (!mdmon_running(st->container_devnm))
1806 start_mdmon(st->container_devnm);
1807 ping_monitor(container);
1808 if (mdmon_running(st->container_devnm) &&
1809 st->update_tail == NULL)
1810 st->update_tail = &st->updates;
1811 }
1812
1813 if (s->size == MAX_SIZE)
1814 s->size = 0;
1815 array.size = s->size;
1816 if ((unsigned)array.size != s->size) {
1817 /* got truncated to 32bit, write to
1818 * component_size instead
1819 */
1820 if (sra)
1821 rv = sysfs_set_num(sra, NULL,
1822 "component_size", s->size);
1823 else
1824 rv = -1;
1825 } else {
1826 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1827
1828 /* manage array size when it is managed externally
1829 */
1830 if ((rv == 0) && st->ss->external)
1831 rv = set_array_size(st, sra, sra->text_version);
1832 }
1833
1834 if (raid0_takeover) {
1835 /* do not recync non-existing parity,
1836 * we will drop it anyway
1837 */
1838 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1839 /* go back to raid0, drop parity disk
1840 */
1841 sysfs_set_str(sra, NULL, "level", "raid0");
1842 ioctl(fd, GET_ARRAY_INFO, &array);
1843 }
1844
1845 size_change_error:
1846 if (rv != 0) {
1847 int err = errno;
1848
1849 /* restore metadata */
1850 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1851 UnSet, NULL, devname,
1852 ROLLBACK_METADATA_CHANGES,
1853 c->verbose) == 0)
1854 sync_metadata(st);
1855 pr_err("Cannot set device size for %s: %s\n",
1856 devname, strerror(err));
1857 if (err == EBUSY &&
1858 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1859 cont_err("Bitmap must be removed before size can be changed\n");
1860 rv = 1;
1861 goto release;
1862 }
1863 if (s->assume_clean) {
1864 /* This will fail on kernels older than 3.0 unless
1865 * a backport has been arranged.
1866 */
1867 if (sra == NULL ||
1868 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1869 pr_err("--assume-clean not supported with --grow on this kernel\n");
1870 }
1871 ioctl(fd, GET_ARRAY_INFO, &array);
1872 s->size = get_component_size(fd)/2;
1873 if (s->size == 0)
1874 s->size = array.size;
1875 if (c->verbose >= 0) {
1876 if (s->size == orig_size)
1877 pr_err("component size of %s "
1878 "unchanged at %lluK\n",
1879 devname, s->size);
1880 else
1881 pr_err("component size of %s "
1882 "has been set to %lluK\n",
1883 devname, s->size);
1884 }
1885 changed = 1;
1886 } else if (array.level != LEVEL_CONTAINER) {
1887 s->size = get_component_size(fd)/2;
1888 if (s->size == 0)
1889 s->size = array.size;
1890 }
1891
1892 /* See if there is anything else to do */
1893 if ((s->level == UnSet || s->level == array.level) &&
1894 (s->layout_str == NULL) &&
1895 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1896 data_offset == INVALID_SECTORS &&
1897 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1898 /* Nothing more to do */
1899 if (!changed && c->verbose >= 0)
1900 pr_err("%s: no change requested\n",
1901 devname);
1902 goto release;
1903 }
1904
1905 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1906 * current implementation assumes that following conditions must be met:
1907 * - RAID10:
1908 * - far_copies == 1
1909 * - near_copies == 2
1910 */
1911 if ((s->level == 0 && array.level == 10 && sra &&
1912 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1913 (s->level == 0 && array.level == 1 && sra)) {
1914 int err;
1915 err = remove_disks_for_takeover(st, sra, array.layout);
1916 if (err) {
1917 dprintf(Name": Array cannot be reshaped\n");
1918 if (cfd > -1)
1919 close(cfd);
1920 rv = 1;
1921 goto release;
1922 }
1923 /* Make sure mdmon has seen the device removal
1924 * and updated metadata before we continue with
1925 * level change
1926 */
1927 if (container)
1928 ping_monitor(container);
1929 }
1930
1931 memset(&info, 0, sizeof(info));
1932 info.array = array;
1933 sysfs_init(&info, fd, NULL);
1934 strcpy(info.text_version, sra->text_version);
1935 info.component_size = s->size*2;
1936 info.new_level = s->level;
1937 info.new_chunk = s->chunk * 1024;
1938 if (info.array.level == LEVEL_CONTAINER) {
1939 info.delta_disks = UnSet;
1940 info.array.raid_disks = s->raiddisks;
1941 } else if (s->raiddisks)
1942 info.delta_disks = s->raiddisks - info.array.raid_disks;
1943 else
1944 info.delta_disks = UnSet;
1945 if (s->layout_str == NULL) {
1946 info.new_layout = UnSet;
1947 if (info.array.level == 6 &&
1948 (info.new_level == 6 || info.new_level == UnSet) &&
1949 info.array.layout >= 16) {
1950 pr_err("%s has a non-standard layout. If you"
1951 " wish to preserve this\n", devname);
1952 cont_err("during the reshape, please specify"
1953 " --layout=preserve\n");
1954 cont_err("If you want to change it, specify a"
1955 " layout or use --layout=normalise\n");
1956 rv = 1;
1957 goto release;
1958 }
1959 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1960 strcmp(s->layout_str, "normalize") == 0) {
1961 /* If we have a -6 RAID6 layout, remove the '-6'. */
1962 info.new_layout = UnSet;
1963 if (info.array.level == 6 && info.new_level == UnSet) {
1964 char l[40], *h;
1965 strcpy(l, map_num(r6layout, info.array.layout));
1966 h = strrchr(l, '-');
1967 if (h && strcmp(h, "-6") == 0) {
1968 *h = 0;
1969 info.new_layout = map_name(r6layout, l);
1970 }
1971 } else {
1972 pr_err("%s is only meaningful when reshaping"
1973 " a RAID6 array.\n", s->layout_str);
1974 rv = 1;
1975 goto release;
1976 }
1977 } else if (strcmp(s->layout_str, "preserve") == 0) {
1978 /* This means that a non-standard RAID6 layout
1979 * is OK.
1980 * In particular:
1981 * - When reshape a RAID6 (e.g. adding a device)
1982 * which is in a non-standard layout, it is OK
1983 * to preserve that layout.
1984 * - When converting a RAID5 to RAID6, leave it in
1985 * the XXX-6 layout, don't re-layout.
1986 */
1987 if (info.array.level == 6 && info.new_level == UnSet)
1988 info.new_layout = info.array.layout;
1989 else if (info.array.level == 5 && info.new_level == 6) {
1990 char l[40];
1991 strcpy(l, map_num(r5layout, info.array.layout));
1992 strcat(l, "-6");
1993 info.new_layout = map_name(r6layout, l);
1994 } else {
1995 pr_err("%s in only meaningful when reshaping"
1996 " to RAID6\n", s->layout_str);
1997 rv = 1;
1998 goto release;
1999 }
2000 } else {
2001 int l = info.new_level;
2002 if (l == UnSet)
2003 l = info.array.level;
2004 switch (l) {
2005 case 5:
2006 info.new_layout = map_name(r5layout, s->layout_str);
2007 break;
2008 case 6:
2009 info.new_layout = map_name(r6layout, s->layout_str);
2010 break;
2011 case 10:
2012 info.new_layout = parse_layout_10(s->layout_str);
2013 break;
2014 case LEVEL_FAULTY:
2015 info.new_layout = parse_layout_faulty(s->layout_str);
2016 break;
2017 default:
2018 pr_err("layout not meaningful"
2019 " with this level\n");
2020 rv = 1;
2021 goto release;
2022 }
2023 if (info.new_layout == UnSet) {
2024 pr_err("layout %s not understood"
2025 " for this level\n",
2026 s->layout_str);
2027 rv = 1;
2028 goto release;
2029 }
2030 }
2031
2032 if (array.level == LEVEL_FAULTY) {
2033 if (s->level != UnSet && s->level != array.level) {
2034 pr_err("cannot change level of Faulty device\n");
2035 rv =1 ;
2036 }
2037 if (s->chunk) {
2038 pr_err("cannot set chunksize of Faulty device\n");
2039 rv =1 ;
2040 }
2041 if (s->raiddisks && s->raiddisks != 1) {
2042 pr_err("cannot set raid_disks of Faulty device\n");
2043 rv =1 ;
2044 }
2045 if (s->layout_str) {
2046 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2047 dprintf("Cannot get array information.\n");
2048 goto release;
2049 }
2050 array.layout = info.new_layout;
2051 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2052 pr_err("failed to set new layout\n");
2053 rv = 1;
2054 } else if (c->verbose >= 0)
2055 printf("layout for %s set to %d\n",
2056 devname, array.layout);
2057 }
2058 } else if (array.level == LEVEL_CONTAINER) {
2059 /* This change is to be applied to every array in the
2060 * container. This is only needed when the metadata imposes
2061 * restraints of the various arrays in the container.
2062 * Currently we only know that IMSM requires all arrays
2063 * to have the same number of devices so changing the
2064 * number of devices (On-Line Capacity Expansion) must be
2065 * performed at the level of the container
2066 */
2067 rv = reshape_container(container, devname, -1, st, &info,
2068 c->force, c->backup_file, c->verbose, 0, 0, 0);
2069 frozen = 0;
2070 } else {
2071 /* get spare devices from external metadata
2072 */
2073 if (st->ss->external) {
2074 struct mdinfo *info2;
2075
2076 info2 = st->ss->container_content(st, subarray);
2077 if (info2) {
2078 info.array.spare_disks =
2079 info2->array.spare_disks;
2080 sysfs_free(info2);
2081 }
2082 }
2083
2084 /* Impose these changes on a single array. First
2085 * check that the metadata is OK with the change. */
2086
2087 if (reshape_super(st, 0, info.new_level,
2088 info.new_layout, info.new_chunk,
2089 info.array.raid_disks, info.delta_disks,
2090 c->backup_file, devname, APPLY_METADATA_CHANGES,
2091 c->verbose)) {
2092 rv = 1;
2093 goto release;
2094 }
2095 sync_metadata(st);
2096 rv = reshape_array(container, fd, devname, st, &info, c->force,
2097 devlist, data_offset, c->backup_file, c->verbose,
2098 0, 0, 0);
2099 frozen = 0;
2100 }
2101 release:
2102 sysfs_free(sra);
2103 if (frozen > 0)
2104 unfreeze(st);
2105 return rv;
2106 }
2107
2108 /* verify_reshape_position()
2109 * Function checks if reshape position in metadata is not farther
2110 * than position in md.
2111 * Return value:
2112 * 0 : not valid sysfs entry
2113 * it can be caused by not started reshape, it should be started
2114 * by reshape array or raid0 array is before takeover
2115 * -1 : error, reshape position is obviously wrong
2116 * 1 : success, reshape progress correct or updated
2117 */
2118 static int verify_reshape_position(struct mdinfo *info, int level)
2119 {
2120 int ret_val = 0;
2121 char buf[40];
2122 int rv;
2123
2124 /* read sync_max, failure can mean raid0 array */
2125 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2126
2127 if (rv > 0) {
2128 char *ep;
2129 unsigned long long position = strtoull(buf, &ep, 0);
2130
2131 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2132 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2133 position *= get_data_disks(level,
2134 info->new_layout,
2135 info->array.raid_disks);
2136 if (info->reshape_progress < position) {
2137 dprintf("Corrected reshape progress (%llu) to "
2138 "md position (%llu)\n",
2139 info->reshape_progress, position);
2140 info->reshape_progress = position;
2141 ret_val = 1;
2142 } else if (info->reshape_progress > position) {
2143 pr_err("Fatal error: array "
2144 "reshape was not properly frozen "
2145 "(expected reshape position is %llu, "
2146 "but reshape progress is %llu.\n",
2147 position, info->reshape_progress);
2148 ret_val = -1;
2149 } else {
2150 dprintf("Reshape position in md and metadata "
2151 "are the same;");
2152 ret_val = 1;
2153 }
2154 }
2155 } else if (rv == 0) {
2156 /* for valid sysfs entry, 0-length content
2157 * should be indicated as error
2158 */
2159 ret_val = -1;
2160 }
2161
2162 return ret_val;
2163 }
2164
2165 static unsigned long long choose_offset(unsigned long long lo,
2166 unsigned long long hi,
2167 unsigned long long min,
2168 unsigned long long max)
2169 {
2170 /* Choose a new offset between hi and lo.
2171 * It must be between min and max, but
2172 * we would prefer something near the middle of hi/lo, and also
2173 * prefer to be aligned to a big power of 2.
2174 *
2175 * So we start with the middle, then for each bit,
2176 * starting at '1' and increasing, if it is set, we either
2177 * add it or subtract it if possible, preferring the option
2178 * which is furthest from the boundary.
2179 *
2180 * We stop once we get a 1MB alignment. As units are in sectors,
2181 * 1MB = 2*1024 sectors.
2182 */
2183 unsigned long long choice = (lo + hi) / 2;
2184 unsigned long long bit = 1;
2185
2186 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2187 unsigned long long bigger, smaller;
2188 if (! (bit & choice))
2189 continue;
2190 bigger = choice + bit;
2191 smaller = choice - bit;
2192 if (bigger > max && smaller < min)
2193 break;
2194 if (bigger > max)
2195 choice = smaller;
2196 else if (smaller < min)
2197 choice = bigger;
2198 else if (hi - bigger > smaller - lo)
2199 choice = bigger;
2200 else
2201 choice = smaller;
2202 }
2203 return choice;
2204 }
2205
2206 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2207 char *devname, int delta_disks,
2208 unsigned long long data_offset,
2209 unsigned long long min,
2210 int can_fallback)
2211 {
2212 struct mdinfo *sd;
2213 int dir = 0;
2214 int err = 0;
2215 unsigned long long before, after;
2216
2217 /* Need to find min space before and after so same is used
2218 * on all devices
2219 */
2220 before = UINT64_MAX;
2221 after = UINT64_MAX;
2222 for (sd = sra->devs; sd; sd = sd->next) {
2223 char *dn;
2224 int dfd;
2225 int rv;
2226 struct supertype *st2;
2227 struct mdinfo info2;
2228
2229 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2230 continue;
2231 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2232 dfd = dev_open(dn, O_RDONLY);
2233 if (dfd < 0) {
2234 pr_err("%s: cannot open component %s\n",
2235 devname, dn ? dn : "-unknown-");
2236 goto release;
2237 }
2238 st2 = dup_super(st);
2239 rv = st2->ss->load_super(st2,dfd, NULL);
2240 close(dfd);
2241 if (rv) {
2242 free(st2);
2243 pr_err("%s: cannot get superblock from %s\n",
2244 devname, dn);
2245 goto release;
2246 }
2247 st2->ss->getinfo_super(st2, &info2, NULL);
2248 st2->ss->free_super(st2);
2249 free(st2);
2250 if (info2.space_before == 0 &&
2251 info2.space_after == 0) {
2252 /* Metadata doesn't support data_offset changes */
2253 return 1;
2254 }
2255 if (before > info2.space_before)
2256 before = info2.space_before;
2257 if (after > info2.space_after)
2258 after = info2.space_after;
2259
2260 if (data_offset != INVALID_SECTORS) {
2261 if (dir == 0) {
2262 if (info2.data_offset == data_offset) {
2263 pr_err("%s: already has that data_offset\n",
2264 dn);
2265 goto release;
2266 }
2267 if (data_offset < info2.data_offset)
2268 dir = -1;
2269 else
2270 dir = 1;
2271 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2272 (data_offset >= info2.data_offset && dir == -1)) {
2273 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2274 dn);
2275 goto release;
2276 }
2277 }
2278 }
2279 if (before == UINT64_MAX)
2280 /* impossible really, there must be no devices */
2281 return 1;
2282
2283 for (sd = sra->devs; sd; sd = sd->next) {
2284 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2285 unsigned long long new_data_offset;
2286
2287 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2288 continue;
2289 if (delta_disks < 0) {
2290 /* Don't need any space as array is shrinking
2291 * just move data_offset up by min
2292 */
2293 if (data_offset == INVALID_SECTORS)
2294 new_data_offset = sd->data_offset + min;
2295 else {
2296 if (data_offset < sd->data_offset + min) {
2297 pr_err("--data-offset too small for %s\n",
2298 dn);
2299 goto release;
2300 }
2301 new_data_offset = data_offset;
2302 }
2303 } else if (delta_disks > 0) {
2304 /* need space before */
2305 if (before < min) {
2306 if (can_fallback)
2307 goto fallback;
2308 pr_err("Insufficient head-space for reshape on %s\n",
2309 dn);
2310 goto release;
2311 }
2312 if (data_offset == INVALID_SECTORS)
2313 new_data_offset = sd->data_offset - min;
2314 else {
2315 if (data_offset > sd->data_offset - min) {
2316 pr_err("--data-offset too large for %s\n",
2317 dn);
2318 goto release;
2319 }
2320 new_data_offset = data_offset;
2321 }
2322 } else {
2323 if (dir == 0) {
2324 /* can move up or down. If 'data_offset'
2325 * was set we would have already decided,
2326 * so just choose direction with most space.
2327 */
2328 if (before > after)
2329 dir = -1;
2330 else
2331 dir = 1;
2332 }
2333 sysfs_set_str(sra, NULL, "reshape_direction",
2334 dir == 1 ? "backwards" : "forwards");
2335 if (dir > 0) {
2336 /* Increase data offset */
2337 if (after < min) {
2338 if (can_fallback)
2339 goto fallback;
2340 pr_err("Insufficient tail-space for reshape on %s\n",
2341 dn);
2342 goto release;
2343 }
2344 if (data_offset != INVALID_SECTORS &&
2345 data_offset < sd->data_offset + min) {
2346 pr_err("--data-offset too small on %s\n",
2347 dn);
2348 goto release;
2349 }
2350 if (data_offset != INVALID_SECTORS)
2351 new_data_offset = data_offset;
2352 else
2353 new_data_offset = choose_offset(sd->data_offset,
2354 sd->data_offset + after,
2355 sd->data_offset + min,
2356 sd->data_offset + after);
2357 } else {
2358 /* Decrease data offset */
2359 if (before < min) {
2360 if (can_fallback)
2361 goto fallback;
2362 pr_err("insufficient head-room on %s\n",
2363 dn);
2364 goto release;
2365 }
2366 if (data_offset != INVALID_SECTORS &&
2367 data_offset < sd->data_offset - min) {
2368 pr_err("--data-offset too small on %s\n",
2369 dn);
2370 goto release;
2371 }
2372 if (data_offset != INVALID_SECTORS)
2373 new_data_offset = data_offset;
2374 else
2375 new_data_offset = choose_offset(sd->data_offset - before,
2376 sd->data_offset,
2377 sd->data_offset - before,
2378 sd->data_offset - min);
2379 }
2380 }
2381 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2382 if (err < 0 && errno == E2BIG) {
2383 /* try again after increasing data size to max */
2384 err = sysfs_set_num(sra, sd, "size", 0);
2385 if (err < 0 && errno == EINVAL &&
2386 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2387 /* some kernels have a bug where you cannot
2388 * use '0' on spare devices. */
2389 sysfs_set_num(sra, sd, "size",
2390 (sra->component_size + after)/2);
2391 }
2392 err = sysfs_set_num(sra, sd, "new_offset",
2393 new_data_offset);
2394 }
2395 if (err < 0) {
2396 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2397 pr_err("data-offset is too big for %s\n",
2398 dn);
2399 goto release;
2400 }
2401 if (sd == sra->devs &&
2402 (errno == ENOENT || errno == E2BIG))
2403 /* Early kernel, no 'new_offset' file,
2404 * or kernel doesn't like us.
2405 * For RAID5/6 this is not fatal
2406 */
2407 return 1;
2408 pr_err("Cannot set new_offset for %s\n",
2409 dn);
2410 break;
2411 }
2412 }
2413 return err;
2414 release:
2415 return -1;
2416 fallback:
2417 /* Just use a backup file */
2418 return 1;
2419 }
2420
2421 static int raid10_reshape(char *container, int fd, char *devname,
2422 struct supertype *st, struct mdinfo *info,
2423 struct reshape *reshape,
2424 unsigned long long data_offset,
2425 int force, int verbose)
2426 {
2427 /* Changing raid_disks, layout, chunksize or possibly
2428 * just data_offset for a RAID10.
2429 * We must always change data_offset. We change by at least
2430 * ->min_offset_change which is the largest of the old and new
2431 * chunk sizes.
2432 * If raid_disks is increasing, then data_offset must decrease
2433 * by at least this copy size.
2434 * If raid_disks is unchanged, data_offset must increase or
2435 * decrease by at least min_offset_change but preferably by much more.
2436 * We choose half of the available space.
2437 * If raid_disks is decreasing, data_offset must increase by
2438 * at least min_offset_change. To allow of this, component_size
2439 * must be decreased by the same amount.
2440 *
2441 * So we calculate the required minimum and direction, possibly
2442 * reduce the component_size, then iterate through the devices
2443 * and set the new_data_offset.
2444 * If that all works, we set chunk_size, layout, raid_disks, and start
2445 * 'reshape'
2446 */
2447 struct mdinfo *sra;
2448 unsigned long long min;
2449 int err = 0;
2450
2451 sra = sysfs_read(fd, NULL,
2452 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2453 );
2454 if (!sra) {
2455 pr_err("%s: Cannot get array details from sysfs\n",
2456 devname);
2457 goto release;
2458 }
2459 min = reshape->min_offset_change;
2460
2461 if (info->delta_disks)
2462 sysfs_set_str(sra, NULL, "reshape_direction",
2463 info->delta_disks < 0 ? "backwards" : "forwards");
2464 if (info->delta_disks < 0 &&
2465 info->space_after < min) {
2466 int rv = sysfs_set_num(sra, NULL, "component_size",
2467 (sra->component_size -
2468 min)/2);
2469 if (rv) {
2470 pr_err("cannot reduce component size\n");
2471 goto release;
2472 }
2473 }
2474 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2475 min, 0);
2476 if (err == 1) {
2477 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2478 cont_err("supported on this kernel\n");
2479 err = -1;
2480 }
2481 if (err < 0)
2482 goto release;
2483
2484 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2485 err = errno;
2486 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2487 err = errno;
2488 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2489 info->array.raid_disks + info->delta_disks) < 0)
2490 err = errno;
2491 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2492 err = errno;
2493 if (err) {
2494 pr_err("Cannot set array shape for %s\n",
2495 devname);
2496 if (err == EBUSY &&
2497 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2498 cont_err(" Bitmap must be removed before"
2499 " shape can be changed\n");
2500 goto release;
2501 }
2502 sysfs_free(sra);
2503 return 0;
2504 release:
2505 sysfs_free(sra);
2506 return 1;
2507 }
2508
2509 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2510 {
2511 struct mdinfo *sra, *sd;
2512 /* Initialisation to silence compiler warning */
2513 unsigned long long min_space_before = 0, min_space_after = 0;
2514 int first = 1;
2515
2516 sra = sysfs_read(fd, NULL, GET_DEVS);
2517 if (!sra)
2518 return;
2519 for (sd = sra->devs; sd; sd = sd->next) {
2520 char *dn;
2521 int dfd;
2522 struct supertype *st2;
2523 struct mdinfo info2;
2524
2525 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2526 continue;
2527 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2528 dfd = dev_open(dn, O_RDONLY);
2529 if (dfd < 0)
2530 break;
2531 st2 = dup_super(st);
2532 if (st2->ss->load_super(st2,dfd, NULL)) {
2533 close(dfd);
2534 free(st2);
2535 break;
2536 }
2537 close(dfd);
2538 st2->ss->getinfo_super(st2, &info2, NULL);
2539 st2->ss->free_super(st2);
2540 free(st2);
2541 if (first ||
2542 min_space_before > info2.space_before)
2543 min_space_before = info2.space_before;
2544 if (first ||
2545 min_space_after > info2.space_after)
2546 min_space_after = info2.space_after;
2547 first = 0;
2548 }
2549 if (sd == NULL && !first) {
2550 info->space_after = min_space_after;
2551 info->space_before = min_space_before;
2552 }
2553 sysfs_free(sra);
2554 }
2555
2556 static void update_cache_size(char *container, struct mdinfo *sra,
2557 struct mdinfo *info,
2558 int disks, unsigned long long blocks)
2559 {
2560 /* Check that the internal stripe cache is
2561 * large enough, or it won't work.
2562 * It must hold at least 4 stripes of the larger
2563 * chunk size
2564 */
2565 unsigned long cache;
2566 cache = max(info->array.chunk_size, info->new_chunk);
2567 cache *= 4; /* 4 stripes minimum */
2568 cache /= 512; /* convert to sectors */
2569 /* make sure there is room for 'blocks' with a bit to spare */
2570 if (cache < 16 + blocks / disks)
2571 cache = 16 + blocks / disks;
2572 cache /= (4096/512); /* Covert from sectors to pages */
2573
2574 if (sra->cache_size < cache)
2575 subarray_set_num(container, sra, "stripe_cache_size",
2576 cache+1);
2577 }
2578
2579 static int impose_reshape(struct mdinfo *sra,
2580 struct mdinfo *info,
2581 struct supertype *st,
2582 int fd,
2583 int restart,
2584 char *devname, char *container,
2585 struct reshape *reshape)
2586 {
2587 struct mdu_array_info_s array;
2588
2589 sra->new_chunk = info->new_chunk;
2590
2591 if (restart) {
2592 /* for external metadata checkpoint saved by mdmon can be lost
2593 * or missed /due to e.g. crash/. Check if md is not during
2594 * restart farther than metadata points to.
2595 * If so, this means metadata information is obsolete.
2596 */
2597 if (st->ss->external)
2598 verify_reshape_position(info, reshape->level);
2599 sra->reshape_progress = info->reshape_progress;
2600 } else {
2601 sra->reshape_progress = 0;
2602 if (reshape->after.data_disks < reshape->before.data_disks)
2603 /* start from the end of the new array */
2604 sra->reshape_progress = (sra->component_size
2605 * reshape->after.data_disks);
2606 }
2607
2608 ioctl(fd, GET_ARRAY_INFO, &array);
2609 if (info->array.chunk_size == info->new_chunk &&
2610 reshape->before.layout == reshape->after.layout &&
2611 st->ss->external == 0) {
2612 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2613 array.raid_disks = reshape->after.data_disks + reshape->parity;
2614 if (!restart &&
2615 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2616 int err = errno;
2617
2618 pr_err("Cannot set device shape for %s: %s\n",
2619 devname, strerror(errno));
2620
2621 if (err == EBUSY &&
2622 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2623 cont_err("Bitmap must be removed before"
2624 " shape can be changed\n");
2625
2626 goto release;
2627 }
2628 } else if (!restart) {
2629 /* set them all just in case some old 'new_*' value
2630 * persists from some earlier problem.
2631 */
2632 int err = 0;
2633 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2634 err = errno;
2635 if (!err && sysfs_set_num(sra, NULL, "layout",
2636 reshape->after.layout) < 0)
2637 err = errno;
2638 if (!err && subarray_set_num(container, sra, "raid_disks",
2639 reshape->after.data_disks +
2640 reshape->parity) < 0)
2641 err = errno;
2642 if (err) {
2643 pr_err("Cannot set device shape for %s\n",
2644 devname);
2645
2646 if (err == EBUSY &&
2647 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2648 cont_err("Bitmap must be removed before"
2649 " shape can be changed\n");
2650 goto release;
2651 }
2652 }
2653 return 0;
2654 release:
2655 return -1;
2656 }
2657
2658 static int impose_level(int fd, int level, char *devname, int verbose)
2659 {
2660 char *c;
2661 struct mdu_array_info_s array;
2662 struct mdinfo info;
2663 sysfs_init(&info, fd, NULL);
2664
2665 ioctl(fd, GET_ARRAY_INFO, &array);
2666 if (level == 0 &&
2667 (array.level >= 4 && array.level <= 6)) {
2668 /* To convert to RAID0 we need to fail and
2669 * remove any non-data devices. */
2670 int found = 0;
2671 int d;
2672 int data_disks = array.raid_disks - 1;
2673 if (array.level == 6)
2674 data_disks -= 1;
2675 if (array.level == 5 &&
2676 array.layout != ALGORITHM_PARITY_N)
2677 return -1;
2678 if (array.level == 6 &&
2679 array.layout != ALGORITHM_PARITY_N_6)
2680 return -1;
2681 sysfs_set_str(&info, NULL,"sync_action", "idle");
2682 /* First remove any spares so no recovery starts */
2683 for (d = 0, found = 0;
2684 d < MAX_DISKS && found < array.nr_disks;
2685 d++) {
2686 mdu_disk_info_t disk;
2687 disk.number = d;
2688 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2689 continue;
2690 if (disk.major == 0 && disk.minor == 0)
2691 continue;
2692 found++;
2693 if ((disk.state & (1 << MD_DISK_ACTIVE))
2694 && disk.raid_disk < data_disks)
2695 /* keep this */
2696 continue;
2697 ioctl(fd, HOT_REMOVE_DISK,
2698 makedev(disk.major, disk.minor));
2699 }
2700 /* Now fail anything left */
2701 ioctl(fd, GET_ARRAY_INFO, &array);
2702 for (d = 0, found = 0;
2703 d < MAX_DISKS && found < array.nr_disks;
2704 d++) {
2705 int cnt;
2706 mdu_disk_info_t disk;
2707 disk.number = d;
2708 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2709 continue;
2710 if (disk.major == 0 && disk.minor == 0)
2711 continue;
2712 found++;
2713 if ((disk.state & (1 << MD_DISK_ACTIVE))
2714 && disk.raid_disk < data_disks)
2715 /* keep this */
2716 continue;
2717 ioctl(fd, SET_DISK_FAULTY,
2718 makedev(disk.major, disk.minor));
2719 cnt = 5;
2720 while (ioctl(fd, HOT_REMOVE_DISK,
2721 makedev(disk.major, disk.minor)) < 0
2722 && errno == EBUSY
2723 && cnt--) {
2724 usleep(10000);
2725 }
2726 }
2727 }
2728 c = map_num(pers, level);
2729 if (c) {
2730 int err = sysfs_set_str(&info, NULL, "level", c);
2731 if (err) {
2732 err = errno;
2733 pr_err("%s: could not set level to %s\n",
2734 devname, c);
2735 if (err == EBUSY &&
2736 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2737 cont_err("Bitmap must be removed"
2738 " before level can be changed\n");
2739 return err;
2740 }
2741 if (verbose >= 0)
2742 pr_err("level of %s changed to %s\n",
2743 devname, c);
2744 }
2745 return 0;
2746 }
2747
2748 int sigterm = 0;
2749 static void catch_term(int sig)
2750 {
2751 sigterm = 1;
2752 }
2753
2754 static int continue_via_systemd(char *devnm)
2755 {
2756 int skipped, i, pid, status;
2757 char pathbuf[1024];
2758 /* In a systemd/udev world, it is best to get systemd to
2759 * run "mdadm --grow --continue" rather than running in the
2760 * background.
2761 */
2762 switch(fork()) {
2763 case 0:
2764 /* FIXME yuk. CLOSE_EXEC?? */
2765 skipped = 0;
2766 for (i = 3; skipped < 20; i++)
2767 if (close(i) < 0)
2768 skipped++;
2769 else
2770 skipped = 0;
2771
2772 /* Don't want to see error messages from
2773 * systemctl. If the service doesn't exist,
2774 * we fork ourselves.
2775 */
2776 close(2);
2777 open("/dev/null", O_WRONLY);
2778 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2779 devnm);
2780 status = execl("/usr/bin/systemctl", "systemctl",
2781 "start",
2782 pathbuf, NULL);
2783 status = execl("/bin/systemctl", "systemctl", "start",
2784 pathbuf, NULL);
2785 exit(1);
2786 case -1: /* Just do it ourselves. */
2787 break;
2788 default: /* parent - good */
2789 pid = wait(&status);
2790 if (pid >= 0 && status == 0)
2791 return 1;
2792 }
2793 return 0;
2794 }
2795
2796 static int reshape_array(char *container, int fd, char *devname,
2797 struct supertype *st, struct mdinfo *info,
2798 int force, struct mddev_dev *devlist,
2799 unsigned long long data_offset,
2800 char *backup_file, int verbose, int forked,
2801 int restart, int freeze_reshape)
2802 {
2803 struct reshape reshape;
2804 int spares_needed;
2805 char *msg;
2806 int orig_level = UnSet;
2807 int odisks;
2808 int delayed;
2809
2810 struct mdu_array_info_s array;
2811 char *c;
2812
2813 struct mddev_dev *dv;
2814 int added_disks;
2815
2816 int *fdlist = NULL;
2817 unsigned long long *offsets = NULL;
2818 int d;
2819 int nrdisks;
2820 int err;
2821 unsigned long blocks;
2822 unsigned long long array_size;
2823 int done;
2824 struct mdinfo *sra = NULL;
2825
2826 /* when reshaping a RAID0, the component_size might be zero.
2827 * So try to fix that up.
2828 */
2829 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2830 dprintf("Cannot get array information.\n");
2831 goto release;
2832 }
2833 if (array.level == 0 && info->component_size == 0) {
2834 get_dev_size(fd, NULL, &array_size);
2835 info->component_size = array_size / array.raid_disks;
2836 }
2837
2838 if (array.level == 10)
2839 /* Need space_after info */
2840 get_space_after(fd, st, info);
2841
2842 if (info->reshape_active) {
2843 int new_level = info->new_level;
2844 info->new_level = UnSet;
2845 if (info->delta_disks > 0)
2846 info->array.raid_disks -= info->delta_disks;
2847 msg = analyse_change(devname, info, &reshape);
2848 info->new_level = new_level;
2849 if (info->delta_disks > 0)
2850 info->array.raid_disks += info->delta_disks;
2851 if (!restart)
2852 /* Make sure the array isn't read-only */
2853 ioctl(fd, RESTART_ARRAY_RW, 0);
2854 } else
2855 msg = analyse_change(devname, info, &reshape);
2856 if (msg) {
2857 /* if msg == "", error has already been printed */
2858 if (msg[0])
2859 pr_err("%s\n", msg);
2860 goto release;
2861 }
2862 if (restart &&
2863 (reshape.level != info->array.level ||
2864 reshape.before.layout != info->array.layout ||
2865 reshape.before.data_disks + reshape.parity
2866 != info->array.raid_disks - max(0, info->delta_disks))) {
2867 pr_err("reshape info is not in native format -"
2868 " cannot continue.\n");
2869 goto release;
2870 }
2871
2872 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2873 /* When reshape is restarted from '0', very begin of array
2874 * it is possible that for external metadata reshape and array
2875 * configuration doesn't happen.
2876 * Check if md has the same opinion, and reshape is restarted
2877 * from 0. If so, this is regular reshape start after reshape
2878 * switch in metadata to next array only.
2879 */
2880 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2881 (info->reshape_progress == 0))
2882 restart = 0;
2883 }
2884 if (restart) {
2885 /* reshape already started. just skip to monitoring the reshape */
2886 if (reshape.backup_blocks == 0)
2887 return 0;
2888 if (restart & RESHAPE_NO_BACKUP)
2889 return 0;
2890
2891 /* Need 'sra' down at 'started:' */
2892 sra = sysfs_read(fd, NULL,
2893 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2894 GET_CACHE);
2895 if (!sra) {
2896 pr_err("%s: Cannot get array details from sysfs\n",
2897 devname);
2898 goto release;
2899 }
2900
2901 if (!backup_file)
2902 backup_file = locate_backup(sra->sys_name);
2903
2904 goto started;
2905 }
2906 /* The container is frozen but the array may not be.
2907 * So freeze the array so spares don't get put to the wrong use
2908 * FIXME there should probably be a cleaner separation between
2909 * freeze_array and freeze_container.
2910 */
2911 sysfs_freeze_array(info);
2912 /* Check we have enough spares to not be degraded */
2913 added_disks = 0;
2914 for (dv = devlist; dv ; dv=dv->next)
2915 added_disks++;
2916 spares_needed = max(reshape.before.data_disks,
2917 reshape.after.data_disks)
2918 + reshape.parity - array.raid_disks;
2919
2920 if (!force &&
2921 info->new_level > 1 && info->array.level > 1 &&
2922 spares_needed > info->array.spare_disks + added_disks) {
2923 pr_err("Need %d spare%s to avoid degraded array,"
2924 " and only have %d.\n"
2925 " Use --force to over-ride this check.\n",
2926 spares_needed,
2927 spares_needed == 1 ? "" : "s",
2928 info->array.spare_disks + added_disks);
2929 goto release;
2930 }
2931 /* Check we have enough spares to not fail */
2932 spares_needed = max(reshape.before.data_disks,
2933 reshape.after.data_disks)
2934 - array.raid_disks;
2935 if ((info->new_level > 1 || info->new_level == 0) &&
2936 spares_needed > info->array.spare_disks +added_disks) {
2937 pr_err("Need %d spare%s to create working array,"
2938 " and only have %d.\n",
2939 spares_needed,
2940 spares_needed == 1 ? "" : "s",
2941 info->array.spare_disks + added_disks);
2942 goto release;
2943 }
2944
2945 if (reshape.level != array.level) {
2946 int err = impose_level(fd, reshape.level, devname, verbose);
2947 if (err)
2948 goto release;
2949 info->new_layout = UnSet; /* after level change,
2950 * layout is meaningless */
2951 orig_level = array.level;
2952 sysfs_freeze_array(info);
2953
2954 if (reshape.level > 0 && st->ss->external) {
2955 /* make sure mdmon is aware of the new level */
2956 if (mdmon_running(container))
2957 flush_mdmon(container);
2958
2959 if (!mdmon_running(container))
2960 start_mdmon(container);
2961 ping_monitor(container);
2962 if (mdmon_running(container) &&
2963 st->update_tail == NULL)
2964 st->update_tail = &st->updates;
2965 }
2966 }
2967 /* ->reshape_super might have chosen some spares from the
2968 * container that it wants to be part of the new array.
2969 * We can collect them with ->container_content and give
2970 * them to the kernel.
2971 */
2972 if (st->ss->reshape_super && st->ss->container_content) {
2973 char *subarray = strchr(info->text_version+1, '/')+1;
2974 struct mdinfo *info2 =
2975 st->ss->container_content(st, subarray);
2976 struct mdinfo *d;
2977
2978 if (info2) {
2979 sysfs_init(info2, fd, st->devnm);
2980 /* When increasing number of devices, we need to set
2981 * new raid_disks before adding these, or they might
2982 * be rejected.
2983 */
2984 if (reshape.backup_blocks &&
2985 reshape.after.data_disks > reshape.before.data_disks)
2986 subarray_set_num(container, info2, "raid_disks",
2987 reshape.after.data_disks +
2988 reshape.parity);
2989 for (d = info2->devs; d; d = d->next) {
2990 if (d->disk.state == 0 &&
2991 d->disk.raid_disk >= 0) {
2992 /* This is a spare that wants to
2993 * be part of the array.
2994 */
2995 add_disk(fd, st, info2, d);
2996 }
2997 }
2998 sysfs_free(info2);
2999 }
3000 }
3001 /* We might have been given some devices to add to the
3002 * array. Now that the array has been changed to the right
3003 * level and frozen, we can safely add them.
3004 */
3005 if (devlist)
3006 Manage_subdevs(devname, fd, devlist, verbose,
3007 0,NULL, 0);
3008
3009 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3010 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3011 if (reshape.backup_blocks == 0) {
3012 /* No restriping needed, but we might need to impose
3013 * some more changes: layout, raid_disks, chunk_size
3014 */
3015 /* read current array info */
3016 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
3017 dprintf("Cannot get array information.\n");
3018 goto release;
3019 }
3020 /* compare current array info with new values and if
3021 * it is different update them to new */
3022 if (info->new_layout != UnSet &&
3023 info->new_layout != array.layout) {
3024 array.layout = info->new_layout;
3025 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3026 pr_err("failed to set new layout\n");
3027 goto release;
3028 } else if (verbose >= 0)
3029 printf("layout for %s set to %d\n",
3030 devname, array.layout);
3031 }
3032 if (info->delta_disks != UnSet &&
3033 info->delta_disks != 0 &&
3034 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3035 array.raid_disks += info->delta_disks;
3036 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3037 pr_err("failed to set raid disks\n");
3038 goto release;
3039 } else if (verbose >= 0) {
3040 printf("raid_disks for %s set to %d\n",
3041 devname, array.raid_disks);
3042 }
3043 }
3044 if (info->new_chunk != 0 &&
3045 info->new_chunk != array.chunk_size) {
3046 if (sysfs_set_num(info, NULL,
3047 "chunk_size", info->new_chunk) != 0) {
3048 pr_err("failed to set chunk size\n");
3049 goto release;
3050 } else if (verbose >= 0)
3051 printf("chunk size for %s set to %d\n",
3052 devname, array.chunk_size);
3053 }
3054 unfreeze(st);
3055 return 0;
3056 }
3057
3058 /*
3059 * There are three possibilities.
3060 * 1/ The array will shrink.
3061 * We need to ensure the reshape will pause before reaching
3062 * the 'critical section'. We also need to fork and wait for
3063 * that to happen. When it does we
3064 * suspend/backup/complete/unfreeze
3065 *
3066 * 2/ The array will not change size.
3067 * This requires that we keep a backup of a sliding window
3068 * so that we can restore data after a crash. So we need
3069 * to fork and monitor progress.
3070 * In future we will allow the data_offset to change, so
3071 * a sliding backup becomes unnecessary.
3072 *
3073 * 3/ The array will grow. This is relatively easy.
3074 * However the kernel's restripe routines will cheerfully
3075 * overwrite some early data before it is safe. So we
3076 * need to make a backup of the early parts of the array
3077 * and be ready to restore it if rebuild aborts very early.
3078 * For externally managed metadata, we still need a forked
3079 * child to monitor the reshape and suspend IO over the region
3080 * that is being reshaped.
3081 *
3082 * We backup data by writing it to one spare, or to a
3083 * file which was given on command line.
3084 *
3085 * In each case, we first make sure that storage is available
3086 * for the required backup.
3087 * Then we:
3088 * - request the shape change.
3089 * - fork to handle backup etc.
3090 */
3091 /* Check that we can hold all the data */
3092 get_dev_size(fd, NULL, &array_size);
3093 if (reshape.new_size < (array_size/512)) {
3094 pr_err("this change will reduce the size of the array.\n"
3095 " use --grow --array-size first to truncate array.\n"
3096 " e.g. mdadm --grow %s --array-size %llu\n",
3097 devname, reshape.new_size/2);
3098 goto release;
3099 }
3100
3101 if (array.level == 10) {
3102 /* Reshaping RAID10 does not require any data backup by
3103 * user-space. Instead it requires that the data_offset
3104 * is changed to avoid the need for backup.
3105 * So this is handled very separately
3106 */
3107 if (restart)
3108 /* Nothing to do. */
3109 return 0;
3110 return raid10_reshape(container, fd, devname, st, info,
3111 &reshape, data_offset,
3112 force, verbose);
3113 }
3114 sra = sysfs_read(fd, NULL,
3115 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3116 GET_CACHE);
3117 if (!sra) {
3118 pr_err("%s: Cannot get array details from sysfs\n",
3119 devname);
3120 goto release;
3121 }
3122
3123 if (!backup_file)
3124 switch(set_new_data_offset(sra, st, devname,
3125 reshape.after.data_disks - reshape.before.data_disks,
3126 data_offset,
3127 reshape.min_offset_change, 1)) {
3128 case -1:
3129 goto release;
3130 case 0:
3131 /* Updated data_offset, so it's easy now */
3132 update_cache_size(container, sra, info,
3133 min(reshape.before.data_disks,
3134 reshape.after.data_disks),
3135 reshape.backup_blocks);
3136
3137 /* Right, everything seems fine. Let's kick things off.
3138 */
3139 sync_metadata(st);
3140
3141 if (impose_reshape(sra, info, st, fd, restart,
3142 devname, container, &reshape) < 0)
3143 goto release;
3144 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3145 pr_err("Failed to initiate reshape!\n");
3146 goto release;
3147 }
3148 if (info->new_level == reshape.level)
3149 return 0;
3150 /* need to adjust level when reshape completes */
3151 switch(fork()) {
3152 case -1: /* ignore error, but don't wait */
3153 return 0;
3154 default: /* parent */
3155 return 0;
3156 case 0:
3157 map_fork();
3158 break;
3159 }
3160 close(fd);
3161 wait_reshape(sra);
3162 fd = open_dev(sra->sys_name);
3163 if (fd >= 0)
3164 impose_level(fd, info->new_level, devname, verbose);
3165 return 0;
3166 case 1: /* Couldn't set data_offset, try the old way */
3167 if (data_offset != INVALID_SECTORS) {
3168 pr_err("Cannot update data_offset on this array\n");
3169 goto release;
3170 }
3171 break;
3172 }
3173
3174 started:
3175 /* Decide how many blocks (sectors) for a reshape
3176 * unit. The number we have so far is just a minimum
3177 */
3178 blocks = reshape.backup_blocks;
3179 if (reshape.before.data_disks ==
3180 reshape.after.data_disks) {
3181 /* Make 'blocks' bigger for better throughput, but
3182 * not so big that we reject it below.
3183 * Try for 16 megabytes
3184 */
3185 while (blocks * 32 < sra->component_size &&
3186 blocks < 16*1024*2)
3187 blocks *= 2;
3188 } else
3189 pr_err("Need to backup %luK of critical "
3190 "section..\n", blocks/2);
3191
3192 if (blocks >= sra->component_size/2) {
3193 pr_err("%s: Something wrong"
3194 " - reshape aborted\n",
3195 devname);
3196 goto release;
3197 }
3198
3199 /* Now we need to open all these devices so we can read/write.
3200 */
3201 nrdisks = max(reshape.before.data_disks,
3202 reshape.after.data_disks) + reshape.parity
3203 + sra->array.spare_disks;
3204 fdlist = xcalloc((1+nrdisks), sizeof(int));
3205 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3206
3207 odisks = reshape.before.data_disks + reshape.parity;
3208 d = reshape_prepare_fdlist(devname, sra, odisks,
3209 nrdisks, blocks, backup_file,
3210 fdlist, offsets);
3211 if (d < 0) {
3212 goto release;
3213 }
3214 if ((st->ss->manage_reshape == NULL) ||
3215 (st->ss->recover_backup == NULL)) {
3216 if (backup_file == NULL) {
3217 if (reshape.after.data_disks <=
3218 reshape.before.data_disks) {
3219 pr_err("%s: Cannot grow - need backup-file\n",
3220 devname);
3221 pr_err(" Please provide one with \"--backup=...\"\n");
3222 goto release;
3223 } else if (sra->array.spare_disks == 0) {
3224 pr_err("%s: Cannot grow - "
3225 "need a spare or backup-file to backup "
3226 "critical section\n", devname);
3227 goto release;
3228 }
3229 } else {
3230 if (!reshape_open_backup_file(backup_file, fd, devname,
3231 (signed)blocks,
3232 fdlist+d, offsets+d,
3233 sra->sys_name,
3234 restart)) {
3235 goto release;
3236 }
3237 d++;
3238 }
3239 }
3240
3241 update_cache_size(container, sra, info,
3242 min(reshape.before.data_disks, reshape.after.data_disks),
3243 blocks);
3244
3245 /* Right, everything seems fine. Let's kick things off.
3246 * If only changing raid_disks, use ioctl, else use
3247 * sysfs.
3248 */
3249 sync_metadata(st);
3250
3251 if (impose_reshape(sra, info, st, fd, restart,
3252 devname, container, &reshape) < 0)
3253 goto release;
3254
3255 err = start_reshape(sra, restart, reshape.before.data_disks,
3256 reshape.after.data_disks);
3257 if (err) {
3258 pr_err("Cannot %s reshape for %s\n",
3259 restart ? "continue" : "start",
3260 devname);
3261 goto release;
3262 }
3263 if (restart)
3264 sysfs_set_str(sra, NULL, "array_state", "active");
3265 if (freeze_reshape) {
3266 free(fdlist);
3267 free(offsets);
3268 sysfs_free(sra);
3269 pr_err("Reshape has to be continued from"
3270 " location %llu when root filesystem has been mounted.\n",
3271 sra->reshape_progress);
3272 return 1;
3273 }
3274
3275 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3276 if (continue_via_systemd(container ?: sra->sys_name)) {
3277 free(fdlist);
3278 free(offsets);
3279 sysfs_free(sra);
3280 return 0;
3281 }
3282
3283 /* Now we just need to kick off the reshape and watch, while
3284 * handling backups of the data...
3285 * This is all done by a forked background process.
3286 */
3287 switch(forked ? 0 : fork()) {
3288 case -1:
3289 pr_err("Cannot run child to monitor reshape: %s\n",
3290 strerror(errno));
3291 abort_reshape(sra);
3292 goto release;
3293 default:
3294 free(fdlist);
3295 free(offsets);
3296 sysfs_free(sra);
3297 return 0;
3298 case 0:
3299 map_fork();
3300 break;
3301 }
3302
3303 /* If another array on the same devices is busy, the
3304 * reshape will wait for them. This would mean that
3305 * the first section that we suspend will stay suspended
3306 * for a long time. So check on that possibility
3307 * by looking for "DELAYED" in /proc/mdstat, and if found,
3308 * wait a while
3309 */
3310 do {
3311 struct mdstat_ent *mds, *m;
3312 delayed = 0;
3313 mds = mdstat_read(1, 0);
3314 for (m = mds; m; m = m->next)
3315 if (strcmp(m->devnm, sra->sys_name) == 0) {
3316 if (m->resync &&
3317 m->percent == RESYNC_DELAYED)
3318 delayed = 1;
3319 if (m->resync == 0)
3320 /* Haven't started the reshape thread
3321 * yet, wait a bit
3322 */
3323 delayed = 2;
3324 break;
3325 }
3326 free_mdstat(mds);
3327 if (delayed == 1 && get_linux_version() < 3007000) {
3328 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3329 " You might experience problems until other reshapes complete.\n");
3330 delayed = 0;
3331 }
3332 if (delayed)
3333 mdstat_wait(30 - (delayed-1) * 25);
3334 } while (delayed);
3335 mdstat_close();
3336 close(fd);
3337 if (check_env("MDADM_GROW_VERIFY"))
3338 fd = open(devname, O_RDONLY | O_DIRECT);
3339 else
3340 fd = -1;
3341 mlockall(MCL_FUTURE);
3342
3343 signal(SIGTERM, catch_term);
3344
3345 if (st->ss->external) {
3346 /* metadata handler takes it from here */
3347 done = st->ss->manage_reshape(
3348 fd, sra, &reshape, st, blocks,
3349 fdlist, offsets,
3350 d - odisks, fdlist+odisks,
3351 offsets+odisks);
3352 } else
3353 done = child_monitor(
3354 fd, sra, &reshape, st, blocks,
3355 fdlist, offsets,
3356 d - odisks, fdlist+odisks,
3357 offsets+odisks);
3358
3359 free(fdlist);
3360 free(offsets);
3361
3362 if (backup_file && done) {
3363 char *bul;
3364 bul = make_backup(sra->sys_name);
3365 if (bul) {
3366 char buf[1024];
3367 int l = readlink(bul, buf, sizeof(buf));
3368 if (l > 0) {
3369 buf[l]=0;
3370 unlink(buf);
3371 }
3372 unlink(bul);
3373 free(bul);
3374 }
3375 unlink(backup_file);
3376 }
3377 if (!done) {
3378 abort_reshape(sra);
3379 goto out;
3380 }
3381
3382 if (!st->ss->external &&
3383 !(reshape.before.data_disks != reshape.after.data_disks
3384 && info->custom_array_size) &&
3385 info->new_level == reshape.level &&
3386 !forked) {
3387 /* no need to wait for the reshape to finish as
3388 * there is nothing more to do.
3389 */
3390 sysfs_free(sra);
3391 exit(0);
3392 }
3393 wait_reshape(sra);
3394
3395 if (st->ss->external) {
3396 /* Re-load the metadata as much could have changed */
3397 int cfd = open_dev(st->container_devnm);
3398 if (cfd >= 0) {
3399 flush_mdmon(container);
3400 st->ss->free_super(st);
3401 st->ss->load_container(st, cfd, container);
3402 close(cfd);
3403 }
3404 }
3405
3406 /* set new array size if required customer_array_size is used
3407 * by this metadata.
3408 */
3409 if (reshape.before.data_disks !=
3410 reshape.after.data_disks &&
3411 info->custom_array_size)
3412 set_array_size(st, info, info->text_version);
3413
3414 if (info->new_level != reshape.level) {
3415 if (fd < 0)
3416 fd = open(devname, O_RDONLY);
3417 impose_level(fd, info->new_level, devname, verbose);
3418 close(fd);
3419 if (info->new_level == 0)
3420 st->update_tail = NULL;
3421 }
3422 out:
3423 sysfs_free(sra);
3424 if (forked)
3425 return 0;
3426 unfreeze(st);
3427 exit(0);
3428
3429 release:
3430 free(fdlist);
3431 free(offsets);
3432 if (orig_level != UnSet && sra) {
3433 c = map_num(pers, orig_level);
3434 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3435 pr_err("aborting level change\n");
3436 }
3437 sysfs_free(sra);
3438 if (!forked)
3439 unfreeze(st);
3440 return 1;
3441 }
3442
3443 /* mdfd handle is passed to be closed in child process (after fork).
3444 */
3445 int reshape_container(char *container, char *devname,
3446 int mdfd,
3447 struct supertype *st,
3448 struct mdinfo *info,
3449 int force,
3450 char *backup_file, int verbose,
3451 int forked, int restart, int freeze_reshape)
3452 {
3453 struct mdinfo *cc = NULL;
3454 int rv = restart;
3455 char last_devnm[32] = "";
3456
3457 /* component_size is not meaningful for a container,
3458 * so pass '0' meaning 'no change'
3459 */
3460 if (!restart &&
3461 reshape_super(st, 0, info->new_level,
3462 info->new_layout, info->new_chunk,
3463 info->array.raid_disks, info->delta_disks,
3464 backup_file, devname, APPLY_METADATA_CHANGES,
3465 verbose)) {
3466 unfreeze(st);
3467 return 1;
3468 }
3469
3470 sync_metadata(st);
3471
3472 /* ping monitor to be sure that update is on disk
3473 */
3474 ping_monitor(container);
3475
3476 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3477 if (continue_via_systemd(container))
3478 return 0;
3479
3480 switch (forked ? 0 : fork()) {
3481 case -1: /* error */
3482 perror("Cannot fork to complete reshape\n");
3483 unfreeze(st);
3484 return 1;
3485 default: /* parent */
3486 if (!freeze_reshape)
3487 printf(Name ": multi-array reshape continues"
3488 " in background\n");
3489 return 0;
3490 case 0: /* child */
3491 map_fork();
3492 break;
3493 }
3494
3495 /* close unused handle in child process
3496 */
3497 if (mdfd > -1)
3498 close(mdfd);
3499
3500 while(1) {
3501 /* For each member array with reshape_active,
3502 * we need to perform the reshape.
3503 * We pick the first array that needs reshaping and
3504 * reshape it. reshape_array() will re-read the metadata
3505 * so the next time through a different array should be
3506 * ready for reshape.
3507 * It is possible that the 'different' array will not
3508 * be assembled yet. In that case we simple exit.
3509 * When it is assembled, the mdadm which assembles it
3510 * will take over the reshape.
3511 */
3512 struct mdinfo *content;
3513 int fd;
3514 struct mdstat_ent *mdstat;
3515 char *adev;
3516 int devid;
3517
3518 sysfs_free(cc);
3519
3520 cc = st->ss->container_content(st, NULL);
3521
3522 for (content = cc; content ; content = content->next) {
3523 char *subarray;
3524 if (!content->reshape_active)
3525 continue;
3526
3527 subarray = strchr(content->text_version+1, '/')+1;
3528 mdstat = mdstat_by_subdev(subarray, container);
3529 if (!mdstat)
3530 continue;
3531 if (mdstat->active == 0) {
3532 pr_err("Skipping inactive array %s.\n",
3533 mdstat->devnm);
3534 free_mdstat(mdstat);
3535 mdstat = NULL;
3536 continue;
3537 }
3538 break;
3539 }
3540 if (!content)
3541 break;
3542
3543 devid = devnm2devid(mdstat->devnm);
3544 adev = map_dev(major(devid), minor(devid), 0);
3545 if (!adev)
3546 adev = content->text_version;
3547
3548 fd = open_dev(mdstat->devnm);
3549 if (fd < 0) {
3550 printf(Name ": Device %s cannot be opened for reshape.",
3551 adev);
3552 break;
3553 }
3554
3555 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3556 /* Do not allow for multiple reshape_array() calls for
3557 * the same array.
3558 * It can happen when reshape_array() returns without
3559 * error, when reshape is not finished (wrong reshape
3560 * starting/continuation conditions). Mdmon doesn't
3561 * switch to next array in container and reentry
3562 * conditions for the same array occur.
3563 * This is possibly interim until the behaviour of
3564 * reshape_array is resolved().
3565 */
3566 printf(Name ": Multiple reshape execution detected for "
3567 "device %s.", adev);
3568 close(fd);
3569 break;
3570 }
3571 strcpy(last_devnm, mdstat->devnm);
3572
3573 sysfs_init(content, fd, mdstat->devnm);
3574
3575 if (mdmon_running(container))
3576 flush_mdmon(container);
3577
3578 rv = reshape_array(container, fd, adev, st,
3579 content, force, NULL, INVALID_SECTORS,
3580 backup_file, verbose, 1, restart,
3581 freeze_reshape);
3582 close(fd);
3583
3584 if (freeze_reshape) {
3585 sysfs_free(cc);
3586 exit(0);
3587 }
3588
3589 restart = 0;
3590 if (rv)
3591 break;
3592
3593 if (mdmon_running(container))
3594 flush_mdmon(container);
3595 }
3596 if (!rv)
3597 unfreeze(st);
3598 sysfs_free(cc);
3599 exit(0);
3600 }
3601
3602 /*
3603 * We run a child process in the background which performs the following
3604 * steps:
3605 * - wait for resync to reach a certain point
3606 * - suspend io to the following section
3607 * - backup that section
3608 * - allow resync to proceed further
3609 * - resume io
3610 * - discard the backup.
3611 *
3612 * When are combined in slightly different ways in the three cases.
3613 * Grow:
3614 * - suspend/backup/allow/wait/resume/discard
3615 * Shrink:
3616 * - allow/wait/suspend/backup/allow/wait/resume/discard
3617 * same-size:
3618 * - wait/resume/discard/suspend/backup/allow
3619 *
3620 * suspend/backup/allow always come together
3621 * wait/resume/discard do too.
3622 * For the same-size case we have two backups to improve flow.
3623 *
3624 */
3625
3626 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3627 unsigned long long backup_point,
3628 unsigned long long wait_point,
3629 unsigned long long *suspend_point,
3630 unsigned long long *reshape_completed, int *frozen)
3631 {
3632 /* This function is called repeatedly by the reshape manager.
3633 * It determines how much progress can safely be made and allows
3634 * that progress.
3635 * - 'info' identifies the array and particularly records in
3636 * ->reshape_progress the metadata's knowledge of progress
3637 * This is a sector offset from the start of the array
3638 * of the next array block to be relocated. This number
3639 * may increase from 0 or decrease from array_size, depending
3640 * on the type of reshape that is happening.
3641 * Note that in contrast, 'sync_completed' is a block count of the
3642 * reshape so far. It gives the distance between the start point
3643 * (head or tail of device) and the next place that data will be
3644 * written. It always increases.
3645 * - 'reshape' is the structure created by analyse_change
3646 * - 'backup_point' shows how much the metadata manager has backed-up
3647 * data. For reshapes with increasing progress, it is the next address
3648 * to be backed up, previous addresses have been backed-up. For
3649 * decreasing progress, it is the earliest address that has been
3650 * backed up - later address are also backed up.
3651 * So addresses between reshape_progress and backup_point are
3652 * backed up providing those are in the 'correct' order.
3653 * - 'wait_point' is an array address. When reshape_completed
3654 * passes this point, progress_reshape should return. It might
3655 * return earlier if it determines that ->reshape_progress needs
3656 * to be updated or further backup is needed.
3657 * - suspend_point is maintained by progress_reshape and the caller
3658 * should not touch it except to initialise to zero.
3659 * It is an array address and it only increases in 2.6.37 and earlier.
3660 * This makes it difficult to handle reducing reshapes with
3661 * external metadata.
3662 * However: it is similar to backup_point in that it records the
3663 * other end of a suspended region from reshape_progress.
3664 * it is moved to extend the region that is safe to backup and/or
3665 * reshape
3666 * - reshape_completed is read from sysfs and returned. The caller
3667 * should copy this into ->reshape_progress when it has reason to
3668 * believe that the metadata knows this, and any backup outside this
3669 * has been erased.
3670 *
3671 * Return value is:
3672 * 1 if more data from backup_point - but only as far as suspend_point,
3673 * should be backed up
3674 * 0 if things are progressing smoothly
3675 * -1 if the reshape is finished because it is all done,
3676 * -2 if the reshape is finished due to an error.
3677 */
3678
3679 int advancing = (reshape->after.data_disks
3680 >= reshape->before.data_disks);
3681 unsigned long long need_backup; /* All data between start of array and
3682 * here will at some point need to
3683 * be backed up.
3684 */
3685 unsigned long long read_offset, write_offset;
3686 unsigned long long write_range;
3687 unsigned long long max_progress, target, completed;
3688 unsigned long long array_size = (info->component_size
3689 * reshape->before.data_disks);
3690 int fd;
3691 char buf[20];
3692
3693 /* First, we unsuspend any region that is now known to be safe.
3694 * If suspend_point is on the 'wrong' side of reshape_progress, then
3695 * we don't have or need suspension at the moment. This is true for
3696 * native metadata when we don't need to back-up.
3697 */
3698 if (advancing) {
3699 if (info->reshape_progress <= *suspend_point)
3700 sysfs_set_num(info, NULL, "suspend_lo",
3701 info->reshape_progress);
3702 } else {
3703 /* Note: this won't work in 2.6.37 and before.
3704 * Something somewhere should make sure we don't need it!
3705 */
3706 if (info->reshape_progress >= *suspend_point)
3707 sysfs_set_num(info, NULL, "suspend_hi",
3708 info->reshape_progress);
3709 }
3710
3711 /* Now work out how far it is safe to progress.
3712 * If the read_offset for ->reshape_progress is less than
3713 * 'blocks' beyond the write_offset, we can only progress as far
3714 * as a backup.
3715 * Otherwise we can progress until the write_offset for the new location
3716 * reaches (within 'blocks' of) the read_offset at the current location.
3717 * However that region must be suspended unless we are using native
3718 * metadata.
3719 * If we need to suspend more, we limit it to 128M per device, which is
3720 * rather arbitrary and should be some time-based calculation.
3721 */
3722 read_offset = info->reshape_progress / reshape->before.data_disks;
3723 write_offset = info->reshape_progress / reshape->after.data_disks;
3724 write_range = info->new_chunk/512;
3725 if (reshape->before.data_disks == reshape->after.data_disks)
3726 need_backup = array_size;
3727 else
3728 need_backup = reshape->backup_blocks;
3729 if (advancing) {
3730 if (read_offset < write_offset + write_range)
3731 max_progress = backup_point;
3732 else
3733 max_progress =
3734 read_offset *
3735 reshape->after.data_disks;
3736 } else {
3737 if (read_offset > write_offset - write_range)
3738 /* Can only progress as far as has been backed up,
3739 * which must be suspended */
3740 max_progress = backup_point;
3741 else if (info->reshape_progress <= need_backup)
3742 max_progress = backup_point;
3743 else {
3744 if (info->array.major_version >= 0)
3745 /* Can progress until backup is needed */
3746 max_progress = need_backup;
3747 else {
3748 /* Can progress until metadata update is required */
3749 max_progress =
3750 read_offset *
3751 reshape->after.data_disks;
3752 /* but data must be suspended */
3753 if (max_progress < *suspend_point)
3754 max_progress = *suspend_point;
3755 }
3756 }
3757 }
3758
3759 /* We know it is safe to progress to 'max_progress' providing
3760 * it is suspended or we are using native metadata.
3761 * Consider extending suspend_point 128M per device if it
3762 * is less than 64M per device beyond reshape_progress.
3763 * But always do a multiple of 'blocks'
3764 * FIXME this is too big - it takes to long to complete
3765 * this much.
3766 */
3767 target = 64*1024*2 * min(reshape->before.data_disks,
3768 reshape->after.data_disks);
3769 target /= reshape->backup_blocks;
3770 if (target < 2)
3771 target = 2;
3772 target *= reshape->backup_blocks;
3773
3774 /* For externally managed metadata we always need to suspend IO to
3775 * the area being reshaped so we regularly push suspend_point forward.
3776 * For native metadata we only need the suspend if we are going to do
3777 * a backup.
3778 */
3779 if (advancing) {
3780 if ((need_backup > info->reshape_progress
3781 || info->array.major_version < 0) &&
3782 *suspend_point < info->reshape_progress + target) {
3783 if (need_backup < *suspend_point + 2 * target)
3784 *suspend_point = need_backup;
3785 else if (*suspend_point + 2 * target < array_size)
3786 *suspend_point += 2 * target;
3787 else
3788 *suspend_point = array_size;
3789 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3790 if (max_progress > *suspend_point)
3791 max_progress = *suspend_point;
3792 }
3793 } else {
3794 if (info->array.major_version >= 0) {
3795 /* Only need to suspend when about to backup */
3796 if (info->reshape_progress < need_backup * 2 &&
3797 *suspend_point > 0) {
3798 *suspend_point = 0;
3799 sysfs_set_num(info, NULL, "suspend_lo", 0);
3800 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3801 }
3802 } else {
3803 /* Need to suspend continually */
3804 if (info->reshape_progress < *suspend_point)
3805 *suspend_point = info->reshape_progress;
3806 if (*suspend_point + target < info->reshape_progress)
3807 /* No need to move suspend region yet */;
3808 else {
3809 if (*suspend_point >= 2 * target)
3810 *suspend_point -= 2 * target;
3811 else
3812 *suspend_point = 0;
3813 sysfs_set_num(info, NULL, "suspend_lo",
3814 *suspend_point);
3815 }
3816 if (max_progress < *suspend_point)
3817 max_progress = *suspend_point;
3818 }
3819 }
3820
3821 /* now set sync_max to allow that progress. sync_max, like
3822 * sync_completed is a count of sectors written per device, so
3823 * we find the difference between max_progress and the start point,
3824 * and divide that by after.data_disks to get a sync_max
3825 * number.
3826 * At the same time we convert wait_point to a similar number
3827 * for comparing against sync_completed.
3828 */
3829 /* scale down max_progress to per_disk */
3830 max_progress /= reshape->after.data_disks;
3831 /* Round to chunk size as some kernels give an erroneously high number */
3832 max_progress /= info->new_chunk/512;
3833 max_progress *= info->new_chunk/512;
3834 /* And round to old chunk size as the kernel wants that */
3835 max_progress /= info->array.chunk_size/512;
3836 max_progress *= info->array.chunk_size/512;
3837 /* Limit progress to the whole device */
3838 if (max_progress > info->component_size)
3839 max_progress = info->component_size;
3840 wait_point /= reshape->after.data_disks;
3841 if (!advancing) {
3842 /* switch from 'device offset' to 'processed block count' */
3843 max_progress = info->component_size - max_progress;
3844 wait_point = info->component_size - wait_point;
3845 }
3846
3847 if (!*frozen)
3848 sysfs_set_num(info, NULL, "sync_max", max_progress);
3849
3850 /* Now wait. If we have already reached the point that we were
3851 * asked to wait to, don't wait at all, else wait for any change.
3852 * We need to select on 'sync_completed' as that is the place that
3853 * notifications happen, but we are really interested in
3854 * 'reshape_position'
3855 */
3856 fd = sysfs_get_fd(info, NULL, "sync_completed");
3857 if (fd < 0)
3858 goto check_progress;
3859
3860 if (sysfs_fd_get_ll(fd, &completed) < 0)
3861 goto check_progress;
3862
3863 while (completed < max_progress && completed < wait_point) {
3864 /* Check that sync_action is still 'reshape' to avoid
3865 * waiting forever on a dead array
3866 */
3867 char action[20];
3868 if (sysfs_get_str(info, NULL, "sync_action",
3869 action, 20) <= 0 ||
3870 strncmp(action, "reshape", 7) != 0)
3871 break;
3872 /* Some kernels reset 'sync_completed' to zero
3873 * before setting 'sync_action' to 'idle'.
3874 * So we need these extra tests.
3875 */
3876 if (completed == 0 && advancing
3877 && info->reshape_progress > 0)
3878 break;
3879 if (completed == 0 && !advancing
3880 && info->reshape_progress < (info->component_size
3881 * reshape->after.data_disks))
3882 break;
3883 sysfs_wait(fd, NULL);
3884 if (sysfs_fd_get_ll(fd, &completed) < 0)
3885 goto check_progress;
3886 }
3887 /* Some kernels reset 'sync_completed' to zero,
3888 * we need to have real point we are in md
3889 */
3890 if (completed == 0)
3891 completed = max_progress;
3892
3893 /* some kernels can give an incorrectly high 'completed' number */
3894 completed /= (info->new_chunk/512);
3895 completed *= (info->new_chunk/512);
3896 /* Convert 'completed' back in to a 'progress' number */
3897 completed *= reshape->after.data_disks;
3898 if (!advancing) {
3899 completed = info->component_size * reshape->after.data_disks
3900 - completed;
3901 }
3902 *reshape_completed = completed;
3903
3904 close(fd);
3905
3906 /* We return the need_backup flag. Caller will decide
3907 * how much - a multiple of ->backup_blocks up to *suspend_point
3908 */
3909 if (advancing)
3910 return need_backup > info->reshape_progress;
3911 else
3912 return need_backup >= info->reshape_progress;
3913
3914 check_progress:
3915 /* if we couldn't read a number from sync_completed, then
3916 * either the reshape did complete, or it aborted.
3917 * We can tell which by checking for 'none' in reshape_position.
3918 * If it did abort, then it might immediately restart if it
3919 * it was just a device failure that leaves us degraded but
3920 * functioning.
3921 */
3922 strcpy(buf, "hi");
3923 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3924 || strncmp(buf, "none", 4) != 0) {
3925 /* The abort might only be temporary. Wait up to 10
3926 * seconds for fd to contain a valid number again.
3927 */
3928 int wait = 10000;
3929 int rv = -2;
3930 unsigned long long new_sync_max;
3931 while (fd >= 0 && rv < 0 && wait > 0) {
3932 if (sysfs_wait(fd, &wait) != 1)
3933 break;
3934 switch (sysfs_fd_get_ll(fd, &completed)) {
3935 case 0:
3936 /* all good again */
3937 rv = 1;
3938 /* If "sync_max" is no longer max_progress
3939 * we need to freeze things
3940 */
3941 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3942 *frozen = (new_sync_max != max_progress);
3943 break;
3944 case -2: /* read error - abort */
3945 wait = 0;
3946 break;
3947 }
3948 }
3949 if (fd >= 0)
3950 close(fd);
3951 return rv; /* abort */
3952 } else {
3953 /* Maybe racing with array shutdown - check state */
3954 if (fd >= 0)
3955 close(fd);
3956 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3957 || strncmp(buf, "inactive", 8) == 0
3958 || strncmp(buf, "clear",5) == 0)
3959 return -2; /* abort */
3960 return -1; /* complete */
3961 }
3962 }
3963
3964 /* FIXME return status is never checked */
3965 static int grow_backup(struct mdinfo *sra,
3966 unsigned long long offset, /* per device */
3967 unsigned long stripes, /* per device, in old chunks */
3968 int *sources, unsigned long long *offsets,
3969 int disks, int chunk, int level, int layout,
3970 int dests, int *destfd, unsigned long long *destoffsets,
3971 int part, int *degraded,
3972 char *buf)
3973 {
3974 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3975 * to storage 'destfd' (offset 'destoffsets'), after first
3976 * suspending IO. Then allow resync to continue
3977 * over the suspended section.
3978 * Use part 'part' of the backup-super-block.
3979 */
3980 int odata = disks;
3981 int rv = 0;
3982 int i;
3983 unsigned long long ll;
3984 int new_degraded;
3985 //printf("offset %llu\n", offset);
3986 if (level >= 4)
3987 odata--;
3988 if (level == 6)
3989 odata--;
3990
3991 /* Check that array hasn't become degraded, else we might backup the wrong data */
3992 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3993 return -1; /* FIXME this error is ignored */
3994 new_degraded = (int)ll;
3995 if (new_degraded != *degraded) {
3996 /* check each device to ensure it is still working */
3997 struct mdinfo *sd;
3998 for (sd = sra->devs ; sd ; sd = sd->next) {
3999 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4000 continue;
4001 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4002 char sbuf[20];
4003 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
4004 strstr(sbuf, "faulty") ||
4005 strstr(sbuf, "in_sync") == NULL) {
4006 /* this device is dead */
4007 sd->disk.state = (1<<MD_DISK_FAULTY);
4008 if (sd->disk.raid_disk >= 0 &&
4009 sources[sd->disk.raid_disk] >= 0) {
4010 close(sources[sd->disk.raid_disk]);
4011 sources[sd->disk.raid_disk] = -1;
4012 }
4013 }
4014 }
4015 }
4016 *degraded = new_degraded;
4017 }
4018 if (part) {
4019 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4020 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4021 } else {
4022 bsb.arraystart = __cpu_to_le64(offset * odata);
4023 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4024 }
4025 if (part)
4026 bsb.magic[15] = '2';
4027 for (i = 0; i < dests; i++)
4028 if (part)
4029 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4030 else
4031 lseek64(destfd[i], destoffsets[i], 0);
4032
4033 rv = save_stripes(sources, offsets,
4034 disks, chunk, level, layout,
4035 dests, destfd,
4036 offset*512*odata, stripes * chunk * odata,
4037 buf);
4038
4039 if (rv)
4040 return rv;
4041 bsb.mtime = __cpu_to_le64(time(0));
4042 for (i = 0; i < dests; i++) {
4043 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4044
4045 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4046 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4047 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4048 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4049
4050 rv = -1;
4051 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4052 != destoffsets[i] - 4096)
4053 break;
4054 if (write(destfd[i], &bsb, 512) != 512)
4055 break;
4056 if (destoffsets[i] > 4096) {
4057 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4058 destoffsets[i]+stripes*chunk*odata)
4059 break;
4060 if (write(destfd[i], &bsb, 512) != 512)
4061 break;
4062 }
4063 fsync(destfd[i]);
4064 rv = 0;
4065 }
4066
4067 return rv;
4068 }
4069
4070 /* in 2.6.30, the value reported by sync_completed can be
4071 * less that it should be by one stripe.
4072 * This only happens when reshape hits sync_max and pauses.
4073 * So allow wait_backup to either extent sync_max further
4074 * than strictly necessary, or return before the
4075 * sync has got quite as far as we would really like.
4076 * This is what 'blocks2' is for.
4077 * The various caller give appropriate values so that
4078 * every works.
4079 */
4080 /* FIXME return value is often ignored */
4081 static int forget_backup(int dests, int *destfd,
4082 unsigned long long *destoffsets,
4083 int part)
4084 {
4085 /*
4086 * Erase backup 'part' (which is 0 or 1)
4087 */
4088 int i;
4089 int rv;
4090
4091 if (part) {
4092 bsb.arraystart2 = __cpu_to_le64(0);
4093 bsb.length2 = __cpu_to_le64(0);
4094 } else {
4095 bsb.arraystart = __cpu_to_le64(0);
4096 bsb.length = __cpu_to_le64(0);
4097 }
4098 bsb.mtime = __cpu_to_le64(time(0));
4099 rv = 0;
4100 for (i = 0; i < dests; i++) {
4101 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4102 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4103 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4104 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4105 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4106 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4107 destoffsets[i]-4096)
4108 rv = -1;
4109 if (rv == 0 &&
4110 write(destfd[i], &bsb, 512) != 512)
4111 rv = -1;
4112 fsync(destfd[i]);
4113 }
4114 return rv;
4115 }
4116
4117 static void fail(char *msg)
4118 {
4119 int rv;
4120 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4121 rv |= (write(2, "\n", 1) != 1);
4122 exit(rv ? 1 : 2);
4123 }
4124
4125 static char *abuf, *bbuf;
4126 static unsigned long long abuflen;
4127 static void validate(int afd, int bfd, unsigned long long offset)
4128 {
4129 /* check that the data in the backup against the array.
4130 * This is only used for regression testing and should not
4131 * be used while the array is active
4132 */
4133 if (afd < 0)
4134 return;
4135 lseek64(bfd, offset - 4096, 0);
4136 if (read(bfd, &bsb2, 512) != 512)
4137 fail("cannot read bsb");
4138 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4139 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4140 fail("first csum bad");
4141 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4142 fail("magic is bad");
4143 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4144 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4145 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4146 fail("second csum bad");
4147
4148 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4149 fail("devstart is wrong");
4150
4151 if (bsb2.length) {
4152 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4153
4154 if (abuflen < len) {
4155 free(abuf);
4156 free(bbuf);
4157 abuflen = len;
4158 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4159 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4160 abuflen = 0;
4161 /* just stop validating on mem-alloc failure */
4162 return;
4163 }
4164 }
4165
4166 lseek64(bfd, offset, 0);
4167 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4168 //printf("len %llu\n", len);
4169 fail("read first backup failed");
4170 }
4171 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4172 if ((unsigned long long)read(afd, abuf, len) != len)
4173 fail("read first from array failed");
4174 if (memcmp(bbuf, abuf, len) != 0) {
4175 #if 0
4176 int i;
4177 printf("offset=%llu len=%llu\n",
4178 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4179 for (i=0; i<len; i++)
4180 if (bbuf[i] != abuf[i]) {
4181 printf("first diff byte %d\n", i);
4182 break;
4183 }
4184 #endif
4185 fail("data1 compare failed");
4186 }
4187 }
4188 if (bsb2.length2) {
4189 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4190
4191 if (abuflen < len) {
4192 free(abuf);
4193 free(bbuf);
4194 abuflen = len;
4195 abuf = xmalloc(abuflen);
4196 bbuf = xmalloc(abuflen);
4197 }
4198
4199 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4200 if ((unsigned long long)read(bfd, bbuf, len) != len)
4201 fail("read second backup failed");
4202 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4203 if ((unsigned long long)read(afd, abuf, len) != len)
4204 fail("read second from array failed");
4205 if (memcmp(bbuf, abuf, len) != 0)
4206 fail("data2 compare failed");
4207 }
4208 }
4209
4210 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4211 struct supertype *st, unsigned long blocks,
4212 int *fds, unsigned long long *offsets,
4213 int dests, int *destfd, unsigned long long *destoffsets)
4214 {
4215 /* Monitor a reshape where backup is being performed using
4216 * 'native' mechanism - either to a backup file, or
4217 * to some space in a spare.
4218 */
4219 char *buf;
4220 int degraded = -1;
4221 unsigned long long speed;
4222 unsigned long long suspend_point, array_size;
4223 unsigned long long backup_point, wait_point;
4224 unsigned long long reshape_completed;
4225 int done = 0;
4226 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4227 int part = 0; /* The next part of the backup area to fill. It may already
4228 * be full, so we need to check */
4229 int level = reshape->level;
4230 int layout = reshape->before.layout;
4231 int data = reshape->before.data_disks;
4232 int disks = reshape->before.data_disks + reshape->parity;
4233 int chunk = sra->array.chunk_size;
4234 struct mdinfo *sd;
4235 unsigned long stripes;
4236 int uuid[4];
4237 int frozen = 0;
4238
4239 /* set up the backup-super-block. This requires the
4240 * uuid from the array.
4241 */
4242 /* Find a superblock */
4243 for (sd = sra->devs; sd; sd = sd->next) {
4244 char *dn;
4245 int devfd;
4246 int ok;
4247 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4248 continue;
4249 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4250 devfd = dev_open(dn, O_RDONLY);
4251 if (devfd < 0)
4252 continue;
4253 ok = st->ss->load_super(st, devfd, NULL);
4254 close(devfd);
4255 if (ok == 0)
4256 break;
4257 }
4258 if (!sd) {
4259 pr_err("Cannot find a superblock\n");
4260 return 0;
4261 }
4262
4263 memset(&bsb, 0, 512);
4264 memcpy(bsb.magic, "md_backup_data-1", 16);
4265 st->ss->uuid_from_super(st, uuid);
4266 memcpy(bsb.set_uuid, uuid, 16);
4267 bsb.mtime = __cpu_to_le64(time(0));
4268 bsb.devstart2 = blocks;
4269
4270 stripes = blocks / (sra->array.chunk_size/512) /
4271 reshape->before.data_disks;
4272
4273 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4274 /* Don't start the 'reshape' */
4275 return 0;
4276 if (reshape->before.data_disks == reshape->after.data_disks) {
4277 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4278 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4279 }
4280
4281 if (increasing) {
4282 array_size = sra->component_size * reshape->after.data_disks;
4283 backup_point = sra->reshape_progress;
4284 suspend_point = 0;
4285 } else {
4286 array_size = sra->component_size * reshape->before.data_disks;
4287 backup_point = reshape->backup_blocks;
4288 suspend_point = array_size;
4289 }
4290
4291 while (!done) {
4292 int rv;
4293
4294 /* Want to return as soon the oldest backup slot can
4295 * be released as that allows us to start backing up
4296 * some more, providing suspend_point has been
4297 * advanced, which it should have.
4298 */
4299 if (increasing) {
4300 wait_point = array_size;
4301 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4302 wait_point = (__le64_to_cpu(bsb.arraystart) +
4303 __le64_to_cpu(bsb.length));
4304 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4305 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4306 __le64_to_cpu(bsb.length2));
4307 } else {
4308 wait_point = 0;
4309 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4310 wait_point = __le64_to_cpu(bsb.arraystart);
4311 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4312 wait_point = __le64_to_cpu(bsb.arraystart2);
4313 }
4314
4315 reshape_completed = sra->reshape_progress;
4316 rv = progress_reshape(sra, reshape,
4317 backup_point, wait_point,
4318 &suspend_point, &reshape_completed,
4319 &frozen);
4320 /* external metadata would need to ping_monitor here */
4321 sra->reshape_progress = reshape_completed;
4322
4323 /* Clear any backup region that is before 'here' */
4324 if (increasing) {
4325 if (__le64_to_cpu(bsb.length) > 0 &&
4326 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4327 __le64_to_cpu(bsb.length)))
4328 forget_backup(dests, destfd,
4329 destoffsets, 0);
4330 if (__le64_to_cpu(bsb.length2) > 0 &&
4331 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4332 __le64_to_cpu(bsb.length2)))
4333 forget_backup(dests, destfd,
4334 destoffsets, 1);
4335 } else {
4336 if (__le64_to_cpu(bsb.length) > 0 &&
4337 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4338 forget_backup(dests, destfd,
4339 destoffsets, 0);
4340 if (__le64_to_cpu(bsb.length2) > 0 &&
4341 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4342 forget_backup(dests, destfd,
4343 destoffsets, 1);
4344 }
4345 if (sigterm)
4346 rv = -2;
4347 if (rv < 0) {
4348 if (rv == -1)
4349 done = 1;
4350 break;
4351 }
4352 if (rv == 0 && increasing && !st->ss->external) {
4353 /* No longer need to monitor this reshape */
4354 sysfs_set_str(sra, NULL, "sync_max", "max");
4355 done = 1;
4356 break;
4357 }
4358
4359 while (rv) {
4360 unsigned long long offset;
4361 unsigned long actual_stripes;
4362 /* Need to backup some data.
4363 * If 'part' is not used and the desired
4364 * backup size is suspended, do a backup,
4365 * then consider the next part.
4366 */
4367 /* Check that 'part' is unused */
4368 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4369 break;
4370 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4371 break;
4372
4373 offset = backup_point / data;
4374 actual_stripes = stripes;
4375 if (increasing) {
4376 if (offset + actual_stripes * (chunk/512) >
4377 sra->component_size)
4378 actual_stripes = ((sra->component_size - offset)
4379 / (chunk/512));
4380 if (offset + actual_stripes * (chunk/512) >
4381 suspend_point/data)
4382 break;
4383 } else {
4384 if (offset < actual_stripes * (chunk/512))
4385 actual_stripes = offset / (chunk/512);
4386 offset -= actual_stripes * (chunk/512);
4387 if (offset < suspend_point/data)
4388 break;
4389 }
4390 if (actual_stripes == 0)
4391 break;
4392 grow_backup(sra, offset, actual_stripes,
4393 fds, offsets,
4394 disks, chunk, level, layout,
4395 dests, destfd, destoffsets,
4396 part, &degraded, buf);
4397 validate(afd, destfd[0], destoffsets[0]);
4398 /* record where 'part' is up to */
4399 part = !part;
4400 if (increasing)
4401 backup_point += actual_stripes * (chunk/512) * data;
4402 else
4403 backup_point -= actual_stripes * (chunk/512) * data;
4404 }
4405 }
4406
4407 /* FIXME maybe call progress_reshape one more time instead */
4408 /* remove any remaining suspension */
4409 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4410 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4411 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4412 sysfs_set_num(sra, NULL, "sync_min", 0);
4413
4414 if (reshape->before.data_disks == reshape->after.data_disks)
4415 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4416 free(buf);
4417 return done;
4418 }
4419
4420 /*
4421 * If any spare contains md_back_data-1 which is recent wrt mtime,
4422 * write that data into the array and update the super blocks with
4423 * the new reshape_progress
4424 */
4425 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4426 char *backup_file, int verbose)
4427 {
4428 int i, j;
4429 int old_disks;
4430 unsigned long long *offsets;
4431 unsigned long long nstripe, ostripe;
4432 int ndata, odata;
4433
4434 odata = info->array.raid_disks - info->delta_disks - 1;
4435 if (info->array.level == 6) odata--; /* number of data disks */
4436 ndata = info->array.raid_disks - 1;
4437 if (info->new_level == 6) ndata--;
4438
4439 old_disks = info->array.raid_disks - info->delta_disks;
4440
4441 if (info->delta_disks <= 0)
4442 /* Didn't grow, so the backup file must have
4443 * been used
4444 */
4445 old_disks = cnt;
4446 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4447 struct mdinfo dinfo;
4448 int fd;
4449 int bsbsize;
4450 char *devname, namebuf[20];
4451 unsigned long long lo, hi;
4452
4453 /* This was a spare and may have some saved data on it.
4454 * Load the superblock, find and load the
4455 * backup_super_block.
4456 * If either fail, go on to next device.
4457 * If the backup contains no new info, just return
4458 * else restore data and update all superblocks
4459 */
4460 if (i == old_disks-1) {
4461 fd = open(backup_file, O_RDONLY);
4462 if (fd<0) {
4463 pr_err("backup file %s inaccessible: %s\n",
4464 backup_file, strerror(errno));
4465 continue;
4466 }
4467 devname = backup_file;
4468 } else {
4469 fd = fdlist[i];
4470 if (fd < 0)
4471 continue;
4472 if (st->ss->load_super(st, fd, NULL))
4473 continue;
4474
4475 st->ss->getinfo_super(st, &dinfo, NULL);
4476 st->ss->free_super(st);
4477
4478 if (lseek64(fd,
4479 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4480 0) < 0) {
4481 pr_err("Cannot seek on device %d\n", i);
4482 continue; /* Cannot seek */
4483 }
4484 sprintf(namebuf, "device-%d", i);
4485 devname = namebuf;
4486 }
4487 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4488 if (verbose)
4489 pr_err("Cannot read from %s\n", devname);
4490 continue; /* Cannot read */
4491 }
4492 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4493 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4494 if (verbose)
4495 pr_err("No backup metadata on %s\n", devname);
4496 continue;
4497 }
4498 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4499 if (verbose)
4500 pr_err("Bad backup-metadata checksum on %s\n", devname);
4501 continue; /* bad checksum */
4502 }
4503 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4504 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4505 if (verbose)
4506 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4507 continue; /* Bad second checksum */
4508 }
4509 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4510 if (verbose)
4511 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4512 continue; /* Wrong uuid */
4513 }
4514
4515 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4516 * sometimes they aren't... So allow considerable flexability in matching, and allow
4517 * this test to be overridden by an environment variable.
4518 */
4519 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4520 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4521 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4522 pr_err("accepting backup with timestamp %lu "
4523 "for array with timestamp %lu\n",
4524 (unsigned long)__le64_to_cpu(bsb.mtime),
4525 (unsigned long)info->array.utime);
4526 } else {
4527 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4528 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4529 continue; /* time stamp is too bad */
4530 }
4531 }
4532
4533 if (bsb.magic[15] == '1') {
4534 if (bsb.length == 0)
4535 continue;
4536 if (info->delta_disks >= 0) {
4537 /* reshape_progress is increasing */
4538 if (__le64_to_cpu(bsb.arraystart)
4539 + __le64_to_cpu(bsb.length)
4540 < info->reshape_progress) {
4541 nonew:
4542 if (verbose)
4543 pr_err("backup-metadata found on %s but is not needed\n", devname);
4544 continue; /* No new data here */
4545 }
4546 } else {
4547 /* reshape_progress is decreasing */
4548 if (__le64_to_cpu(bsb.arraystart) >=
4549 info->reshape_progress)
4550 goto nonew; /* No new data here */
4551 }
4552 } else {
4553 if (bsb.length == 0 && bsb.length2 == 0)
4554 continue;
4555 if (info->delta_disks >= 0) {
4556 /* reshape_progress is increasing */
4557 if ((__le64_to_cpu(bsb.arraystart)
4558 + __le64_to_cpu(bsb.length)
4559 < info->reshape_progress)
4560 &&
4561 (__le64_to_cpu(bsb.arraystart2)
4562 + __le64_to_cpu(bsb.length2)
4563 < info->reshape_progress))
4564 goto nonew; /* No new data here */
4565 } else {
4566 /* reshape_progress is decreasing */
4567 if (__le64_to_cpu(bsb.arraystart) >=
4568 info->reshape_progress &&
4569 __le64_to_cpu(bsb.arraystart2) >=
4570 info->reshape_progress)
4571 goto nonew; /* No new data here */
4572 }
4573 }
4574 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4575 second_fail:
4576 if (verbose)
4577 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4578 devname);
4579 continue; /* Cannot seek */
4580 }
4581 /* There should be a duplicate backup superblock 4k before here */
4582 if (lseek64(fd, -4096, 1) < 0 ||
4583 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4584 goto second_fail; /* Cannot find leading superblock */
4585 if (bsb.magic[15] == '1')
4586 bsbsize = offsetof(struct mdp_backup_super, pad1);
4587 else
4588 bsbsize = offsetof(struct mdp_backup_super, pad);
4589 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4590 goto second_fail; /* Cannot find leading superblock */
4591
4592 /* Now need the data offsets for all devices. */
4593 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4594 for(j=0; j<info->array.raid_disks; j++) {
4595 if (fdlist[j] < 0)
4596 continue;
4597 if (st->ss->load_super(st, fdlist[j], NULL))
4598 /* FIXME should be this be an error */
4599 continue;
4600 st->ss->getinfo_super(st, &dinfo, NULL);
4601 st->ss->free_super(st);
4602 offsets[j] = dinfo.data_offset * 512;
4603 }
4604 printf(Name ": restoring critical section\n");
4605
4606 if (restore_stripes(fdlist, offsets,
4607 info->array.raid_disks,
4608 info->new_chunk,
4609 info->new_level,
4610 info->new_layout,
4611 fd, __le64_to_cpu(bsb.devstart)*512,
4612 __le64_to_cpu(bsb.arraystart)*512,
4613 __le64_to_cpu(bsb.length)*512, NULL)) {
4614 /* didn't succeed, so giveup */
4615 if (verbose)
4616 pr_err("Error restoring backup from %s\n",
4617 devname);
4618 free(offsets);
4619 return 1;
4620 }
4621
4622 if (bsb.magic[15] == '2' &&
4623 restore_stripes(fdlist, offsets,
4624 info->array.raid_disks,
4625 info->new_chunk,
4626 info->new_level,
4627 info->new_layout,
4628 fd, __le64_to_cpu(bsb.devstart)*512 +
4629 __le64_to_cpu(bsb.devstart2)*512,
4630 __le64_to_cpu(bsb.arraystart2)*512,
4631 __le64_to_cpu(bsb.length2)*512, NULL)) {
4632 /* didn't succeed, so giveup */
4633 if (verbose)
4634 pr_err("Error restoring second backup from %s\n",
4635 devname);
4636 free(offsets);
4637 return 1;
4638 }
4639
4640 free(offsets);
4641
4642 /* Ok, so the data is restored. Let's update those superblocks. */
4643
4644 lo = hi = 0;
4645 if (bsb.length) {
4646 lo = __le64_to_cpu(bsb.arraystart);
4647 hi = lo + __le64_to_cpu(bsb.length);
4648 }
4649 if (bsb.magic[15] == '2' && bsb.length2) {
4650 unsigned long long lo1, hi1;
4651 lo1 = __le64_to_cpu(bsb.arraystart2);
4652 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4653 if (lo == hi) {
4654 lo = lo1;
4655 hi = hi1;
4656 } else if (lo < lo1)
4657 hi = hi1;
4658 else
4659 lo = lo1;
4660 }
4661 if (lo < hi &&
4662 (info->reshape_progress < lo ||
4663 info->reshape_progress > hi))
4664 /* backup does not affect reshape_progress*/ ;
4665 else if (info->delta_disks >= 0) {
4666 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4667 __le64_to_cpu(bsb.length);
4668 if (bsb.magic[15] == '2') {
4669 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4670 __le64_to_cpu(bsb.length2);
4671 if (p2 > info->reshape_progress)
4672 info->reshape_progress = p2;
4673 }
4674 } else {
4675 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4676 if (bsb.magic[15] == '2') {
4677 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4678 if (p2 < info->reshape_progress)
4679 info->reshape_progress = p2;
4680 }
4681 }
4682 for (j=0; j<info->array.raid_disks; j++) {
4683 if (fdlist[j] < 0)
4684 continue;
4685 if (st->ss->load_super(st, fdlist[j], NULL))
4686 continue;
4687 st->ss->getinfo_super(st, &dinfo, NULL);
4688 dinfo.reshape_progress = info->reshape_progress;
4689 st->ss->update_super(st, &dinfo,
4690 "_reshape_progress",
4691 NULL,0, 0, NULL);
4692 st->ss->store_super(st, fdlist[j]);
4693 st->ss->free_super(st);
4694 }
4695 return 0;
4696 }
4697 /* Didn't find any backup data, try to see if any
4698 * was needed.
4699 */
4700 if (info->delta_disks < 0) {
4701 /* When shrinking, the critical section is at the end.
4702 * So see if we are before the critical section.
4703 */
4704 unsigned long long first_block;
4705 nstripe = ostripe = 0;
4706 first_block = 0;
4707 while (ostripe >= nstripe) {
4708 ostripe += info->array.chunk_size / 512;
4709 first_block = ostripe * odata;
4710 nstripe = first_block / ndata / (info->new_chunk/512) *
4711 (info->new_chunk/512);
4712 }
4713
4714 if (info->reshape_progress >= first_block)
4715 return 0;
4716 }
4717 if (info->delta_disks > 0) {
4718 /* See if we are beyond the critical section. */
4719 unsigned long long last_block;
4720 nstripe = ostripe = 0;
4721 last_block = 0;
4722 while (nstripe >= ostripe) {
4723 nstripe += info->new_chunk / 512;
4724 last_block = nstripe * ndata;
4725 ostripe = last_block / odata / (info->array.chunk_size/512) *
4726 (info->array.chunk_size/512);
4727 }
4728
4729 if (info->reshape_progress >= last_block)
4730 return 0;
4731 }
4732 /* needed to recover critical section! */
4733 if (verbose)
4734 pr_err("Failed to find backup of critical section\n");
4735 return 1;
4736 }
4737
4738 int Grow_continue_command(char *devname, int fd,
4739 char *backup_file, int verbose)
4740 {
4741 int ret_val = 0;
4742 struct supertype *st = NULL;
4743 struct mdinfo *content = NULL;
4744 struct mdinfo array;
4745 char *subarray = NULL;
4746 struct mdinfo *cc = NULL;
4747 struct mdstat_ent *mdstat = NULL;
4748 int cfd = -1;
4749 int fd2 = -1;
4750
4751 dprintf("Grow continue from command line called for %s\n",
4752 devname);
4753
4754 st = super_by_fd(fd, &subarray);
4755 if (!st || !st->ss) {
4756 pr_err("Unable to determine metadata format for %s\n",
4757 devname);
4758 return 1;
4759 }
4760 dprintf("Grow continue is run for ");
4761 if (st->ss->external == 0) {
4762 int d;
4763 dprintf("native array (%s)\n", devname);
4764 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4765 pr_err("%s is not an active md array -"
4766 " aborting\n", devname);
4767 ret_val = 1;
4768 goto Grow_continue_command_exit;
4769 }
4770 content = &array;
4771 /* Need to load a superblock.
4772 * FIXME we should really get what we need from
4773 * sysfs
4774 */
4775 for (d = 0; d < MAX_DISKS; d++) {
4776 mdu_disk_info_t disk;
4777 char *dv;
4778 int err;
4779 disk.number = d;
4780 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4781 continue;
4782 if (disk.major == 0 && disk.minor == 0)
4783 continue;
4784 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4785 continue;
4786 dv = map_dev(disk.major, disk.minor, 1);
4787 if (!dv)
4788 continue;
4789 fd2 = dev_open(dv, O_RDONLY);
4790 if (fd2 < 0)
4791 continue;
4792 err = st->ss->load_super(st, fd2, NULL);
4793 close(fd2);
4794 /* invalidate fd2 to avoid possible double close() */
4795 fd2 = -1;
4796 if (err)
4797 continue;
4798 break;
4799 }
4800 if (d == MAX_DISKS) {
4801 pr_err("Unable to load metadata for %s\n",
4802 devname);
4803 ret_val = 1;
4804 goto Grow_continue_command_exit;
4805 }
4806 st->ss->getinfo_super(st, content, NULL);
4807 } else {
4808 char *container;
4809
4810 if (subarray) {
4811 dprintf("subarray (%s)\n", subarray);
4812 container = st->container_devnm;
4813 cfd = open_dev_excl(st->container_devnm);
4814 } else {
4815 container = st->devnm;
4816 close(fd);
4817 cfd = open_dev_excl(st->devnm);
4818 dprintf("container (%s)\n", container);
4819 fd = cfd;
4820 }
4821 if (cfd < 0) {
4822 pr_err("Unable to open container "
4823 "for %s\n", devname);
4824 ret_val = 1;
4825 goto Grow_continue_command_exit;
4826 }
4827
4828 /* find in container array under reshape
4829 */
4830 ret_val = st->ss->load_container(st, cfd, NULL);
4831 if (ret_val) {
4832 pr_err("Cannot read superblock for %s\n",
4833 devname);
4834 ret_val = 1;
4835 goto Grow_continue_command_exit;
4836 }
4837
4838 cc = st->ss->container_content(st, subarray);
4839 for (content = cc; content ; content = content->next) {
4840 char *array;
4841 int allow_reshape = 1;
4842
4843 if (content->reshape_active == 0)
4844 continue;
4845 /* The decision about array or container wide
4846 * reshape is taken in Grow_continue based
4847 * content->reshape_active state, therefore we
4848 * need to check_reshape based on
4849 * reshape_active and subarray name
4850 */
4851 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4852 allow_reshape = 0;
4853 if (content->reshape_active == CONTAINER_RESHAPE &&
4854 (content->array.state
4855 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4856 allow_reshape = 0;
4857
4858 if (!allow_reshape) {
4859 pr_err("cannot continue reshape of an array"
4860 " in container with unsupported"
4861 " metadata: %s(%s)\n",
4862 devname, container);
4863 ret_val = 1;
4864 goto Grow_continue_command_exit;
4865 }
4866
4867 array = strchr(content->text_version+1, '/')+1;
4868 mdstat = mdstat_by_subdev(array, container);
4869 if (!mdstat)
4870 continue;
4871 if (mdstat->active == 0) {
4872 pr_err("Skipping inactive array %s.\n",
4873 mdstat->devnm);
4874 free_mdstat(mdstat);
4875 mdstat = NULL;
4876 continue;
4877 }
4878 break;
4879 }
4880 if (!content) {
4881 pr_err("Unable to determine reshaped "
4882 "array for %s\n", devname);
4883 ret_val = 1;
4884 goto Grow_continue_command_exit;
4885 }
4886 fd2 = open_dev(mdstat->devnm);
4887 if (fd2 < 0) {
4888 pr_err("cannot open (%s)\n", mdstat->devnm);
4889 ret_val = 1;
4890 goto Grow_continue_command_exit;
4891 }
4892
4893 sysfs_init(content, fd2, mdstat->devnm);
4894
4895 /* start mdmon in case it is not running
4896 */
4897 if (!mdmon_running(container))
4898 start_mdmon(container);
4899 ping_monitor(container);
4900
4901 if (mdmon_running(container))
4902 st->update_tail = &st->updates;
4903 else {
4904 pr_err("No mdmon found. "
4905 "Grow cannot continue.\n");
4906 ret_val = 1;
4907 goto Grow_continue_command_exit;
4908 }
4909 }
4910
4911 /* verify that array under reshape is started from
4912 * correct position
4913 */
4914 if (verify_reshape_position(content, content->array.level) < 0) {
4915 ret_val = 1;
4916 goto Grow_continue_command_exit;
4917 }
4918
4919 /* continue reshape
4920 */
4921 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4922
4923 Grow_continue_command_exit:
4924 if (fd2 > -1)
4925 close(fd2);
4926 if (cfd > -1)
4927 close(cfd);
4928 st->ss->free_super(st);
4929 free_mdstat(mdstat);
4930 sysfs_free(cc);
4931 free(subarray);
4932
4933 return ret_val;
4934 }
4935
4936 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4937 char *backup_file, int forked, int freeze_reshape)
4938 {
4939 int ret_val = 2;
4940
4941 if (!info->reshape_active)
4942 return ret_val;
4943
4944 if (st->ss->external) {
4945 int cfd = open_dev(st->container_devnm);
4946
4947 if (cfd < 0)
4948 return 1;
4949
4950 st->ss->load_container(st, cfd, st->container_devnm);
4951 close(cfd);
4952 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4953 st, info, 0, backup_file,
4954 0, forked,
4955 1 | info->reshape_active,
4956 freeze_reshape);
4957 } else
4958 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4959 NULL, INVALID_SECTORS,
4960 backup_file, 0, forked,
4961 1 | info->reshape_active,
4962 freeze_reshape);
4963
4964 return ret_val;
4965 }
4966
4967 char *make_backup(char *name)
4968 {
4969 char *base = "backup_file-";
4970 int len;
4971 char *fname;
4972
4973 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4974 fname = xmalloc(len);
4975 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4976 return fname;
4977 }
4978
4979 char *locate_backup(char *name)
4980 {
4981 char *fl = make_backup(name);
4982 struct stat stb;
4983
4984 if (stat(fl, &stb) == 0 &&
4985 S_ISREG(stb.st_mode))
4986 return fl;
4987
4988 free(fl);
4989 return NULL;
4990 }
Unnamed repository; edit this file 'description' to name the repository.