]> 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
{platform,super}-intel: Fix two resource leaks
[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 section for reshape - sorry.\n");
91 if (!backup_file)
92 pr_err("Possibly you need to specify a --backup-file\n");
93 return 1;
94 }
95
96 dprintf("restore_backup() returns status OK.\n");
97 return 0;
98 }
99
100 int Grow_Add_device(char *devname, int fd, char *newdev)
101 {
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
107 * all other devices.
108 * This means that we need to *find* all other devices.
109 */
110 struct mdinfo info;
111
112 struct stat stb;
113 int nfd, fd2;
114 int d, nd;
115 struct supertype *st = NULL;
116 char *subarray = NULL;
117
118 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
119 pr_err("cannot get array info for %s\n", devname);
120 return 1;
121 }
122
123 if (info.array.level != -1) {
124 pr_err("can only add devices to linear arrays\n");
125 return 1;
126 }
127
128 st = super_by_fd(fd, &subarray);
129 if (!st) {
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info.array.major_version);
132 return 1;
133 }
134
135 if (subarray) {
136 pr_err("Cannot grow linear sub-arrays yet\n");
137 free(subarray);
138 free(st);
139 return 1;
140 }
141
142 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
143 if (nfd < 0) {
144 pr_err("cannot open %s\n", newdev);
145 free(st);
146 return 1;
147 }
148 fstat(nfd, &stb);
149 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
150 pr_err("%s is not a block device!\n", newdev);
151 close(nfd);
152 free(st);
153 return 1;
154 }
155 /* now check out all the devices and make sure we can read the
156 * superblock */
157 for (d=0 ; d < info.array.raid_disks ; d++) {
158 mdu_disk_info_t disk;
159 char *dv;
160
161 st->ss->free_super(st);
162
163 disk.number = d;
164 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
165 pr_err("cannot get device detail for device %d\n",
166 d);
167 close(nfd);
168 free(st);
169 return 1;
170 }
171 dv = map_dev(disk.major, disk.minor, 1);
172 if (!dv) {
173 pr_err("cannot find device file for device %d\n",
174 d);
175 close(nfd);
176 free(st);
177 return 1;
178 }
179 fd2 = dev_open(dv, O_RDWR);
180 if (fd2 < 0) {
181 pr_err("cannot open device file %s\n", dv);
182 close(nfd);
183 free(st);
184 return 1;
185 }
186
187 if (st->ss->load_super(st, fd2, NULL)) {
188 pr_err("cannot find super block on %s\n", dv);
189 close(nfd);
190 close(fd2);
191 free(st);
192 return 1;
193 }
194 close(fd2);
195 }
196 /* Ok, looks good. Lets update the superblock and write it out to
197 * newdev.
198 */
199
200 info.disk.number = d;
201 info.disk.major = major(stb.st_rdev);
202 info.disk.minor = minor(stb.st_rdev);
203 info.disk.raid_disk = d;
204 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
205 st->ss->update_super(st, &info, "linear-grow-new", newdev,
206 0, 0, NULL);
207
208 if (st->ss->store_super(st, nfd)) {
209 pr_err("Cannot store new superblock on %s\n",
210 newdev);
211 close(nfd);
212 return 1;
213 }
214 close(nfd);
215
216 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
217 pr_err("Cannot add new disk to this array\n");
218 return 1;
219 }
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
222 */
223
224 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
225 pr_err("cannot get array info for %s\n", devname);
226 return 1;
227 }
228
229 nd = d;
230 for (d=0 ; d < info.array.raid_disks ; d++) {
231 mdu_disk_info_t disk;
232 char *dv;
233
234 disk.number = d;
235 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
236 pr_err("cannot get device detail for device %d\n",
237 d);
238 return 1;
239 }
240 dv = map_dev(disk.major, disk.minor, 1);
241 if (!dv) {
242 pr_err("cannot find device file for device %d\n",
243 d);
244 return 1;
245 }
246 fd2 = dev_open(dv, O_RDWR);
247 if (fd2 < 0) {
248 pr_err("cannot open device file %s\n", dv);
249 return 1;
250 }
251 if (st->ss->load_super(st, fd2, NULL)) {
252 pr_err("cannot find super block on %s\n", dv);
253 close(fd);
254 return 1;
255 }
256 info.array.raid_disks = nd+1;
257 info.array.nr_disks = nd+1;
258 info.array.active_disks = nd+1;
259 info.array.working_disks = nd+1;
260
261 st->ss->update_super(st, &info, "linear-grow-update", dv,
262 0, 0, NULL);
263
264 if (st->ss->store_super(st, fd2)) {
265 pr_err("Cannot store new superblock on %s\n", dv);
266 close(fd2);
267 return 1;
268 }
269 close(fd2);
270 }
271
272 return 0;
273 }
274
275 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
276 {
277 /*
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
280 * Then add it
281 *
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
284 * to all devices
285 */
286 mdu_bitmap_file_t bmf;
287 mdu_array_info_t array;
288 struct supertype *st;
289 char *subarray = NULL;
290 int major = BITMAP_MAJOR_HI;
291 int vers = md_get_version(fd);
292 unsigned long long bitmapsize, array_size;
293
294 if (vers < 9003) {
295 major = BITMAP_MAJOR_HOSTENDIAN;
296 pr_err("Warning - bitmaps created on this kernel are not portable\n"
297 " between different architectures. Consider upgrading the Linux kernel.\n");
298 }
299
300 /*
301 * We only ever get called if s->bitmap_file is != NULL, so this check
302 * is just here to quiet down static code checkers.
303 */
304 if (!s->bitmap_file)
305 return 1;
306
307 if (strcmp(s->bitmap_file, "clustered") == 0)
308 major = BITMAP_MAJOR_CLUSTERED;
309
310 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
311 if (errno == ENOMEM)
312 pr_err("Memory allocation failure.\n");
313 else
314 pr_err("bitmaps not supported by this kernel.\n");
315 return 1;
316 }
317 if (bmf.pathname[0]) {
318 if (strcmp(s->bitmap_file,"none")==0) {
319 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
320 pr_err("failed to remove bitmap %s\n",
321 bmf.pathname);
322 return 1;
323 }
324 return 0;
325 }
326 pr_err("%s already has a bitmap (%s)\n",
327 devname, bmf.pathname);
328 return 1;
329 }
330 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
331 pr_err("cannot get array status for %s\n", devname);
332 return 1;
333 }
334 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
335 if (strcmp(s->bitmap_file, "none")==0) {
336 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
337 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
338 if (array.state & (1<<MD_SB_CLUSTERED))
339 pr_err("failed to remove clustered bitmap.\n");
340 else
341 pr_err("failed to remove internal bitmap.\n");
342 return 1;
343 }
344 return 0;
345 }
346 pr_err("bitmap already present on %s\n", devname);
347 return 1;
348 }
349
350 if (strcmp(s->bitmap_file, "none") == 0) {
351 pr_err("no bitmap found on %s\n", devname);
352 return 1;
353 }
354 if (array.level <= 0) {
355 pr_err("Bitmaps not meaningful with level %s\n",
356 map_num(pers, array.level)?:"of this array");
357 return 1;
358 }
359 bitmapsize = array.size;
360 bitmapsize <<= 1;
361 if (get_dev_size(fd, NULL, &array_size) &&
362 array_size > (0x7fffffffULL<<9)) {
363 /* Array is big enough that we cannot trust array.size
364 * try other approaches
365 */
366 bitmapsize = get_component_size(fd);
367 }
368 if (bitmapsize == 0) {
369 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
370 return 1;
371 }
372
373 if (array.level == 10) {
374 int ncopies = (array.layout&255)*((array.layout>>8)&255);
375 bitmapsize = bitmapsize * array.raid_disks / ncopies;
376 }
377
378 st = super_by_fd(fd, &subarray);
379 if (!st) {
380 pr_err("Cannot understand version %d.%d\n",
381 array.major_version, array.minor_version);
382 return 1;
383 }
384 if (subarray) {
385 pr_err("Cannot add bitmaps to sub-arrays yet\n");
386 free(subarray);
387 free(st);
388 return 1;
389 }
390 if (strcmp(s->bitmap_file, "internal") == 0 ||
391 strcmp(s->bitmap_file, "clustered") == 0) {
392 int rv;
393 int d;
394 int offset_setable = 0;
395 struct mdinfo *mdi;
396 if (st->ss->add_internal_bitmap == NULL) {
397 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
398 return 1;
399 }
400 st->nodes = c->nodes;
401 st->cluster_name = c->homecluster;
402 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
403 if (mdi)
404 offset_setable = 1;
405 for (d=0; d< st->max_devs; d++) {
406 mdu_disk_info_t disk;
407 char *dv;
408 disk.number = d;
409 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
410 continue;
411 if (disk.major == 0 &&
412 disk.minor == 0)
413 continue;
414 if ((disk.state & (1<<MD_DISK_SYNC))==0)
415 continue;
416 dv = map_dev(disk.major, disk.minor, 1);
417 if (dv) {
418 int fd2 = dev_open(dv, O_RDWR);
419 if (fd2 < 0)
420 continue;
421 if (st->ss->load_super(st, fd2, NULL)==0) {
422 if (st->ss->add_internal_bitmap(
423 st,
424 &s->bitmap_chunk, c->delay, s->write_behind,
425 bitmapsize, offset_setable,
426 major)
427 )
428 st->ss->write_bitmap(st, fd2, NoUpdate);
429 else {
430 pr_err("failed to create internal bitmap - chunksize problem.\n");
431 close(fd2);
432 return 1;
433 }
434 }
435 close(fd2);
436 }
437 }
438 if (offset_setable) {
439 st->ss->getinfo_super(st, mdi, NULL);
440 sysfs_init(mdi, fd, NULL);
441 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
442 mdi->bitmap_offset);
443 } else {
444 if (strcmp(s->bitmap_file, "clustered") == 0)
445 array.state |= (1<<MD_SB_CLUSTERED);
446 array.state |= (1<<MD_SB_BITMAP_PRESENT);
447 rv = ioctl(fd, SET_ARRAY_INFO, &array);
448 }
449 if (rv < 0) {
450 if (errno == EBUSY)
451 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
452 pr_err("failed to set internal bitmap.\n");
453 return 1;
454 }
455 } else {
456 int uuid[4];
457 int bitmap_fd;
458 int d;
459 int max_devs = st->max_devs;
460
461 /* try to load a superblock */
462 for (d = 0; d < max_devs; d++) {
463 mdu_disk_info_t disk;
464 char *dv;
465 int fd2;
466 disk.number = d;
467 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
468 continue;
469 if ((disk.major==0 && disk.minor==0) ||
470 (disk.state & (1<<MD_DISK_REMOVED)))
471 continue;
472 dv = map_dev(disk.major, disk.minor, 1);
473 if (!dv)
474 continue;
475 fd2 = dev_open(dv, O_RDONLY);
476 if (fd2 >= 0) {
477 if (st->ss->load_super(st, fd2, NULL) == 0) {
478 close(fd2);
479 st->ss->uuid_from_super(st, uuid);
480 break;
481 }
482 close(fd2);
483 }
484 }
485 if (d == max_devs) {
486 pr_err("cannot find UUID for array!\n");
487 return 1;
488 }
489 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
490 c->delay, s->write_behind, bitmapsize, major)) {
491 return 1;
492 }
493 bitmap_fd = open(s->bitmap_file, O_RDWR);
494 if (bitmap_fd < 0) {
495 pr_err("weird: %s cannot be opened\n",
496 s->bitmap_file);
497 return 1;
498 }
499 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
500 int err = errno;
501 if (errno == EBUSY)
502 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
503 pr_err("Cannot set bitmap file for %s: %s\n",
504 devname, strerror(err));
505 return 1;
506 }
507 }
508
509 return 0;
510 }
511
512 /*
513 * When reshaping an array we might need to backup some data.
514 * This is written to all spares with a 'super_block' describing it.
515 * The superblock goes 4K from the end of the used space on the
516 * device.
517 * It if written after the backup is complete.
518 * It has the following structure.
519 */
520
521 static struct mdp_backup_super {
522 char magic[16]; /* md_backup_data-1 or -2 */
523 __u8 set_uuid[16];
524 __u64 mtime;
525 /* start/sizes in 512byte sectors */
526 __u64 devstart; /* address on backup device/file of data */
527 __u64 arraystart;
528 __u64 length;
529 __u32 sb_csum; /* csum of preceeding bytes. */
530 __u32 pad1;
531 __u64 devstart2; /* offset in to data of second section */
532 __u64 arraystart2;
533 __u64 length2;
534 __u32 sb_csum2; /* csum of preceeding bytes. */
535 __u8 pad[512-68-32];
536 } __attribute__((aligned(512))) bsb, bsb2;
537
538 static __u32 bsb_csum(char *buf, int len)
539 {
540 int i;
541 int csum = 0;
542 for (i = 0; i < len; i++)
543 csum = (csum<<3) + buf[0];
544 return __cpu_to_le32(csum);
545 }
546
547 static int check_idle(struct supertype *st)
548 {
549 /* Check that all member arrays for this container, or the
550 * container of this array, are idle
551 */
552 char *container = (st->container_devnm[0]
553 ? st->container_devnm : st->devnm);
554 struct mdstat_ent *ent, *e;
555 int is_idle = 1;
556
557 ent = mdstat_read(0, 0);
558 for (e = ent ; e; e = e->next) {
559 if (!is_container_member(e, container))
560 continue;
561 if (e->percent >= 0) {
562 is_idle = 0;
563 break;
564 }
565 }
566 free_mdstat(ent);
567 return is_idle;
568 }
569
570 static int freeze_container(struct supertype *st)
571 {
572 char *container = (st->container_devnm[0]
573 ? st->container_devnm : st->devnm);
574
575 if (!check_idle(st))
576 return -1;
577
578 if (block_monitor(container, 1)) {
579 pr_err("failed to freeze container\n");
580 return -2;
581 }
582
583 return 1;
584 }
585
586 static void unfreeze_container(struct supertype *st)
587 {
588 char *container = (st->container_devnm[0]
589 ? st->container_devnm : st->devnm);
590
591 unblock_monitor(container, 1);
592 }
593
594 static int freeze(struct supertype *st)
595 {
596 /* Try to freeze resync/rebuild on this array/container.
597 * Return -1 if the array is busy,
598 * return -2 container cannot be frozen,
599 * return 0 if this kernel doesn't support 'frozen'
600 * return 1 if it worked.
601 */
602 if (st->ss->external)
603 return freeze_container(st);
604 else {
605 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
606 int err;
607 char buf[20];
608
609 if (!sra)
610 return -1;
611 /* Need to clear any 'read-auto' status */
612 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
613 strncmp(buf, "read-auto", 9) == 0)
614 sysfs_set_str(sra, NULL, "array_state", "clean");
615
616 err = sysfs_freeze_array(sra);
617 sysfs_free(sra);
618 return err;
619 }
620 }
621
622 static void unfreeze(struct supertype *st)
623 {
624 if (st->ss->external)
625 return unfreeze_container(st);
626 else {
627 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
628 char buf[20];
629
630 if (sra &&
631 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
632 && strcmp(buf, "frozen\n") == 0)
633 sysfs_set_str(sra, NULL, "sync_action", "idle");
634 sysfs_free(sra);
635 }
636 }
637
638 static void wait_reshape(struct mdinfo *sra)
639 {
640 int fd = sysfs_get_fd(sra, NULL, "sync_action");
641 char action[20];
642
643 if (fd < 0)
644 return;
645
646 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
647 strncmp(action, "reshape", 7) == 0)
648 sysfs_wait(fd, NULL);
649 close(fd);
650 }
651
652 static int reshape_super(struct supertype *st, unsigned long long size,
653 int level, int layout, int chunksize, int raid_disks,
654 int delta_disks, char *backup_file, char *dev,
655 int direction, int verbose)
656 {
657 /* nothing extra to check in the native case */
658 if (!st->ss->external)
659 return 0;
660 if (!st->ss->reshape_super ||
661 !st->ss->manage_reshape) {
662 pr_err("%s metadata does not support reshape\n",
663 st->ss->name);
664 return 1;
665 }
666
667 return st->ss->reshape_super(st, size, level, layout, chunksize,
668 raid_disks, delta_disks, backup_file, dev,
669 direction, verbose);
670 }
671
672 static void sync_metadata(struct supertype *st)
673 {
674 if (st->ss->external) {
675 if (st->update_tail) {
676 flush_metadata_updates(st);
677 st->update_tail = &st->updates;
678 } else
679 st->ss->sync_metadata(st);
680 }
681 }
682
683 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
684 {
685 /* when dealing with external metadata subarrays we need to be
686 * prepared to handle EAGAIN. The kernel may need to wait for
687 * mdmon to mark the array active so the kernel can handle
688 * allocations/writeback when preparing the reshape action
689 * (md_allow_write()). We temporarily disable safe_mode_delay
690 * to close a race with the array_state going clean before the
691 * next write to raid_disks / stripe_cache_size
692 */
693 char safe[50];
694 int rc;
695
696 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
697 if (!container ||
698 (strcmp(name, "raid_disks") != 0 &&
699 strcmp(name, "stripe_cache_size") != 0))
700 return sysfs_set_num(sra, NULL, name, n);
701
702 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
703 if (rc <= 0)
704 return -1;
705 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
706 rc = sysfs_set_num(sra, NULL, name, n);
707 if (rc < 0 && errno == EAGAIN) {
708 ping_monitor(container);
709 /* if we get EAGAIN here then the monitor is not active
710 * so stop trying
711 */
712 rc = sysfs_set_num(sra, NULL, name, n);
713 }
714 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
715 return rc;
716 }
717
718 int start_reshape(struct mdinfo *sra, int already_running,
719 int before_data_disks, int data_disks)
720 {
721 int err;
722 unsigned long long sync_max_to_set;
723
724 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
725 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
726 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
727 sra->reshape_progress);
728 if (before_data_disks <= data_disks)
729 sync_max_to_set = sra->reshape_progress / data_disks;
730 else
731 sync_max_to_set = (sra->component_size * data_disks
732 - sra->reshape_progress) / data_disks;
733 if (!already_running)
734 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
735 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
736 if (!already_running && err == 0) {
737 int cnt = 5;
738 do {
739 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
740 if (err)
741 sleep(1);
742 } while (err && errno == EBUSY && cnt-- > 0);
743 }
744 return err;
745 }
746
747 void abort_reshape(struct mdinfo *sra)
748 {
749 sysfs_set_str(sra, NULL, "sync_action", "idle");
750 /*
751 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
752 * suspend_hi to decrease as well as increase.")
753 * you could only increase suspend_{lo,hi} unless the region they
754 * covered was empty. So to reset to 0, you need to push suspend_lo
755 * up past suspend_hi first. So to maximize the chance of mdadm
756 * working on all kernels, we want to keep doing that.
757 */
758 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
759 sysfs_set_num(sra, NULL, "suspend_hi", 0);
760 sysfs_set_num(sra, NULL, "suspend_lo", 0);
761 sysfs_set_num(sra, NULL, "sync_min", 0);
762 // It isn't safe to reset sync_max as we aren't monitoring.
763 // Array really should be stopped at this point.
764 }
765
766 int remove_disks_for_takeover(struct supertype *st,
767 struct mdinfo *sra,
768 int layout)
769 {
770 int nr_of_copies;
771 struct mdinfo *remaining;
772 int slot;
773
774 if (sra->array.level == 10)
775 nr_of_copies = layout & 0xff;
776 else if (sra->array.level == 1)
777 nr_of_copies = sra->array.raid_disks;
778 else
779 return 1;
780
781 remaining = sra->devs;
782 sra->devs = NULL;
783 /* for each 'copy', select one device and remove from the list. */
784 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
785 struct mdinfo **diskp;
786 int found = 0;
787
788 /* Find a working device to keep */
789 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
790 struct mdinfo *disk = *diskp;
791
792 if (disk->disk.raid_disk < slot)
793 continue;
794 if (disk->disk.raid_disk >= slot + nr_of_copies)
795 continue;
796 if (disk->disk.state & (1<<MD_DISK_REMOVED))
797 continue;
798 if (disk->disk.state & (1<<MD_DISK_FAULTY))
799 continue;
800 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
801 continue;
802
803 /* We have found a good disk to use! */
804 *diskp = disk->next;
805 disk->next = sra->devs;
806 sra->devs = disk;
807 found = 1;
808 break;
809 }
810 if (!found)
811 break;
812 }
813
814 if (slot < sra->array.raid_disks) {
815 /* didn't find all slots */
816 struct mdinfo **e;
817 e = &remaining;
818 while (*e)
819 e = &(*e)->next;
820 *e = sra->devs;
821 sra->devs = remaining;
822 return 1;
823 }
824
825 /* Remove all 'remaining' devices from the array */
826 while (remaining) {
827 struct mdinfo *sd = remaining;
828 remaining = sd->next;
829
830 sysfs_set_str(sra, sd, "state", "faulty");
831 sysfs_set_str(sra, sd, "slot", "none");
832 /* for external metadata disks should be removed in mdmon */
833 if (!st->ss->external)
834 sysfs_set_str(sra, sd, "state", "remove");
835 sd->disk.state |= (1<<MD_DISK_REMOVED);
836 sd->disk.state &= ~(1<<MD_DISK_SYNC);
837 sd->next = sra->devs;
838 sra->devs = sd;
839 }
840 return 0;
841 }
842
843 void reshape_free_fdlist(int *fdlist,
844 unsigned long long *offsets,
845 int size)
846 {
847 int i;
848
849 for (i = 0; i < size; i++)
850 if (fdlist[i] >= 0)
851 close(fdlist[i]);
852
853 free(fdlist);
854 free(offsets);
855 }
856
857 int reshape_prepare_fdlist(char *devname,
858 struct mdinfo *sra,
859 int raid_disks,
860 int nrdisks,
861 unsigned long blocks,
862 char *backup_file,
863 int *fdlist,
864 unsigned long long *offsets)
865 {
866 int d = 0;
867 struct mdinfo *sd;
868
869 enable_fds(nrdisks);
870 for (d = 0; d <= nrdisks; d++)
871 fdlist[d] = -1;
872 d = raid_disks;
873 for (sd = sra->devs; sd; sd = sd->next) {
874 if (sd->disk.state & (1<<MD_DISK_FAULTY))
875 continue;
876 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
877 sd->disk.raid_disk < raid_disks) {
878 char *dn = map_dev(sd->disk.major,
879 sd->disk.minor, 1);
880 fdlist[sd->disk.raid_disk]
881 = dev_open(dn, O_RDONLY);
882 offsets[sd->disk.raid_disk] = sd->data_offset*512;
883 if (fdlist[sd->disk.raid_disk] < 0) {
884 pr_err("%s: cannot open component %s\n",
885 devname, dn ? dn : "-unknown-");
886 d = -1;
887 goto release;
888 }
889 } else if (backup_file == NULL) {
890 /* spare */
891 char *dn = map_dev(sd->disk.major,
892 sd->disk.minor, 1);
893 fdlist[d] = dev_open(dn, O_RDWR);
894 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
895 if (fdlist[d] < 0) {
896 pr_err("%s: cannot open component %s\n",
897 devname, dn ? dn : "-unknown-");
898 d = -1;
899 goto release;
900 }
901 d++;
902 }
903 }
904 release:
905 return d;
906 }
907
908 int reshape_open_backup_file(char *backup_file,
909 int fd,
910 char *devname,
911 long blocks,
912 int *fdlist,
913 unsigned long long *offsets,
914 char *sys_name,
915 int restart)
916 {
917 /* Return 1 on success, 0 on any form of failure */
918 /* need to check backup file is large enough */
919 char buf[512];
920 struct stat stb;
921 unsigned int dev;
922 int i;
923
924 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
925 S_IRUSR | S_IWUSR);
926 *offsets = 8 * 512;
927 if (*fdlist < 0) {
928 pr_err("%s: cannot create backup file %s: %s\n",
929 devname, backup_file, strerror(errno));
930 return 0;
931 }
932 /* Guard against backup file being on array device.
933 * If array is partitioned or if LVM etc is in the
934 * way this will not notice, but it is better than
935 * nothing.
936 */
937 fstat(*fdlist, &stb);
938 dev = stb.st_dev;
939 fstat(fd, &stb);
940 if (stb.st_rdev == dev) {
941 pr_err("backup file must NOT be on the array being reshaped.\n");
942 close(*fdlist);
943 return 0;
944 }
945
946 memset(buf, 0, 512);
947 for (i=0; i < blocks + 8 ; i++) {
948 if (write(*fdlist, buf, 512) != 512) {
949 pr_err("%s: cannot create backup file %s: %s\n",
950 devname, backup_file, strerror(errno));
951 return 0;
952 }
953 }
954 if (fsync(*fdlist) != 0) {
955 pr_err("%s: cannot create backup file %s: %s\n",
956 devname, backup_file, strerror(errno));
957 return 0;
958 }
959
960 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
961 char *bu = make_backup(sys_name);
962 if (symlink(backup_file, bu))
963 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
964 strerror(errno));
965 free(bu);
966 }
967
968 return 1;
969 }
970
971 unsigned long compute_backup_blocks(int nchunk, int ochunk,
972 unsigned int ndata, unsigned int odata)
973 {
974 unsigned long a, b, blocks;
975 /* So how much do we need to backup.
976 * We need an amount of data which is both a whole number of
977 * old stripes and a whole number of new stripes.
978 * So LCM for (chunksize*datadisks).
979 */
980 a = (ochunk/512) * odata;
981 b = (nchunk/512) * ndata;
982 /* Find GCD */
983 a = GCD(a, b);
984 /* LCM == product / GCD */
985 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
986
987 return blocks;
988 }
989
990 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
991 {
992 /* Based on the current array state in info->array and
993 * the changes in info->new_* etc, determine:
994 * - whether the change is possible
995 * - Intermediate level/raid_disks/layout
996 * - whether a restriping reshape is needed
997 * - number of sectors in minimum change unit. This
998 * will cover a whole number of stripes in 'before' and
999 * 'after'.
1000 *
1001 * Return message if the change should be rejected
1002 * NULL if the change can be achieved
1003 *
1004 * This can be called as part of starting a reshape, or
1005 * when assembling an array that is undergoing reshape.
1006 */
1007 int near, far, offset, copies;
1008 int new_disks;
1009 int old_chunk, new_chunk;
1010 /* delta_parity records change in number of devices
1011 * caused by level change
1012 */
1013 int delta_parity = 0;
1014
1015 memset(re, 0, sizeof(*re));
1016
1017 /* If a new level not explicitly given, we assume no-change */
1018 if (info->new_level == UnSet)
1019 info->new_level = info->array.level;
1020
1021 if (info->new_chunk)
1022 switch (info->new_level) {
1023 case 0:
1024 case 4:
1025 case 5:
1026 case 6:
1027 case 10:
1028 /* chunk size is meaningful, must divide component_size
1029 * evenly
1030 */
1031 if (info->component_size % (info->new_chunk/512)) {
1032 unsigned long long shrink = info->component_size;
1033 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1034 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1035 info->new_chunk/1024, info->component_size/2);
1036 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1037 devname, shrink/2);
1038 pr_err("will shrink the array so the given chunk size would work.\n");
1039 return "";
1040 }
1041 break;
1042 default:
1043 return "chunk size not meaningful for this level";
1044 }
1045 else
1046 info->new_chunk = info->array.chunk_size;
1047
1048 switch (info->array.level) {
1049 default:
1050 return "No reshape is possibly for this RAID level";
1051 case LEVEL_LINEAR:
1052 if (info->delta_disks != UnSet)
1053 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1054 else
1055 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1056 case 1:
1057 /* RAID1 can convert to RAID1 with different disks, or
1058 * raid5 with 2 disks, or
1059 * raid0 with 1 disk
1060 */
1061 if (info->new_level > 1 &&
1062 (info->component_size & 7))
1063 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1064 if (info->new_level == 0) {
1065 if (info->delta_disks != UnSet &&
1066 info->delta_disks != 0)
1067 return "Cannot change number of disks with RAID1->RAID0 conversion";
1068 re->level = 0;
1069 re->before.data_disks = 1;
1070 re->after.data_disks = 1;
1071 return NULL;
1072 }
1073 if (info->new_level == 1) {
1074 if (info->delta_disks == UnSet)
1075 /* Don't know what to do */
1076 return "no change requested for Growing RAID1";
1077 re->level = 1;
1078 return NULL;
1079 }
1080 if (info->array.raid_disks == 2 &&
1081 info->new_level == 5) {
1082
1083 re->level = 5;
1084 re->before.data_disks = 1;
1085 if (info->delta_disks != UnSet &&
1086 info->delta_disks != 0)
1087 re->after.data_disks = 1 + info->delta_disks;
1088 else
1089 re->after.data_disks = 1;
1090 if (re->after.data_disks < 1)
1091 return "Number of disks too small for RAID5";
1092
1093 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1094 info->array.chunk_size = 65536;
1095 break;
1096 }
1097 /* Could do some multi-stage conversions, but leave that to
1098 * later.
1099 */
1100 return "Impossibly level change request for RAID1";
1101
1102 case 10:
1103 /* RAID10 can be converted from near mode to
1104 * RAID0 by removing some devices.
1105 * It can also be reshaped if the kernel supports
1106 * new_data_offset.
1107 */
1108 switch (info->new_level) {
1109 case 0:
1110 if ((info->array.layout & ~0xff) != 0x100)
1111 return "Cannot Grow RAID10 with far/offset layout";
1112 /* number of devices must be multiple of number of copies */
1113 if (info->array.raid_disks % (info->array.layout & 0xff))
1114 return "RAID10 layout too complex for Grow operation";
1115
1116 new_disks = (info->array.raid_disks
1117 / (info->array.layout & 0xff));
1118 if (info->delta_disks == UnSet)
1119 info->delta_disks = (new_disks
1120 - info->array.raid_disks);
1121
1122 if (info->delta_disks != new_disks - info->array.raid_disks)
1123 return "New number of raid-devices impossible for RAID10";
1124 if (info->new_chunk &&
1125 info->new_chunk != info->array.chunk_size)
1126 return "Cannot change chunk-size with RAID10 Grow";
1127
1128 /* looks good */
1129 re->level = 0;
1130 re->before.data_disks = new_disks;
1131 re->after.data_disks = re->before.data_disks;
1132 return NULL;
1133
1134 case 10:
1135 near = info->array.layout & 0xff;
1136 far = (info->array.layout >> 8) & 0xff;
1137 offset = info->array.layout & 0x10000;
1138 if (far > 1 && !offset)
1139 return "Cannot reshape RAID10 in far-mode";
1140 copies = near * far;
1141
1142 old_chunk = info->array.chunk_size * far;
1143
1144 if (info->new_layout == UnSet)
1145 info->new_layout = info->array.layout;
1146 else {
1147 near = info->new_layout & 0xff;
1148 far = (info->new_layout >> 8) & 0xff;
1149 offset = info->new_layout & 0x10000;
1150 if (far > 1 && !offset)
1151 return "Cannot reshape RAID10 to far-mode";
1152 if (near * far != copies)
1153 return "Cannot change number of copies when reshaping RAID10";
1154 }
1155 if (info->delta_disks == UnSet)
1156 info->delta_disks = 0;
1157 new_disks = (info->array.raid_disks +
1158 info->delta_disks);
1159
1160 new_chunk = info->new_chunk * far;
1161
1162 re->level = 10;
1163 re->before.layout = info->array.layout;
1164 re->before.data_disks = info->array.raid_disks;
1165 re->after.layout = info->new_layout;
1166 re->after.data_disks = new_disks;
1167 /* For RAID10 we don't do backup but do allow reshape,
1168 * so set backup_blocks to INVALID_SECTORS rather than
1169 * zero.
1170 * And there is no need to synchronise stripes on both
1171 * 'old' and 'new'. So the important
1172 * number is the minimum data_offset difference
1173 * which is the larger of (offset copies * chunk).
1174 */
1175 re->backup_blocks = INVALID_SECTORS;
1176 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1177 if (new_disks < re->before.data_disks &&
1178 info->space_after < re->min_offset_change)
1179 /* Reduce component size by one chunk */
1180 re->new_size = (info->component_size -
1181 re->min_offset_change);
1182 else
1183 re->new_size = info->component_size;
1184 re->new_size = re->new_size * new_disks / copies;
1185 return NULL;
1186
1187 default:
1188 return "RAID10 can only be changed to RAID0";
1189 }
1190 case 0:
1191 /* RAID0 can be converted to RAID10, or to RAID456 */
1192 if (info->new_level == 10) {
1193 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1194 /* Assume near=2 layout */
1195 info->new_layout = 0x102;
1196 info->delta_disks = info->array.raid_disks;
1197 }
1198 if (info->new_layout == UnSet) {
1199 int copies = 1 + (info->delta_disks
1200 / info->array.raid_disks);
1201 if (info->array.raid_disks * (copies-1)
1202 != info->delta_disks)
1203 return "Impossible number of devices for RAID0->RAID10";
1204 info->new_layout = 0x100 + copies;
1205 }
1206 if (info->delta_disks == UnSet) {
1207 int copies = info->new_layout & 0xff;
1208 if (info->new_layout != 0x100 + copies)
1209 return "New layout impossible for RAID0->RAID10";;
1210 info->delta_disks = (copies - 1) *
1211 info->array.raid_disks;
1212 }
1213 if (info->new_chunk &&
1214 info->new_chunk != info->array.chunk_size)
1215 return "Cannot change chunk-size with RAID0->RAID10";
1216 /* looks good */
1217 re->level = 10;
1218 re->before.data_disks = (info->array.raid_disks +
1219 info->delta_disks);
1220 re->after.data_disks = re->before.data_disks;
1221 re->before.layout = info->new_layout;
1222 return NULL;
1223 }
1224
1225 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1226 * a raid4 style layout of the final level.
1227 */
1228 switch (info->new_level) {
1229 case 4:
1230 delta_parity = 1;
1231 case 0:
1232 re->level = 4;
1233 re->before.layout = 0;
1234 break;
1235 case 5:
1236 delta_parity = 1;
1237 re->level = 5;
1238 re->before.layout = ALGORITHM_PARITY_N;
1239 if (info->new_layout == UnSet)
1240 info->new_layout = map_name(r5layout, "default");
1241 break;
1242 case 6:
1243 delta_parity = 2;
1244 re->level = 6;
1245 re->before.layout = ALGORITHM_PARITY_N;
1246 if (info->new_layout == UnSet)
1247 info->new_layout = map_name(r6layout, "default");
1248 break;
1249 default:
1250 return "Impossible level change requested";
1251 }
1252 re->before.data_disks = info->array.raid_disks;
1253 /* determining 'after' layout happens outside this 'switch' */
1254 break;
1255
1256 case 4:
1257 info->array.layout = ALGORITHM_PARITY_N;
1258 case 5:
1259 switch (info->new_level) {
1260 case 0:
1261 delta_parity = -1;
1262 case 4:
1263 re->level = info->array.level;
1264 re->before.data_disks = info->array.raid_disks - 1;
1265 re->before.layout = info->array.layout;
1266 break;
1267 case 5:
1268 re->level = 5;
1269 re->before.data_disks = info->array.raid_disks - 1;
1270 re->before.layout = info->array.layout;
1271 break;
1272 case 6:
1273 delta_parity = 1;
1274 re->level = 6;
1275 re->before.data_disks = info->array.raid_disks - 1;
1276 switch (info->array.layout) {
1277 case ALGORITHM_LEFT_ASYMMETRIC:
1278 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1279 break;
1280 case ALGORITHM_RIGHT_ASYMMETRIC:
1281 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1282 break;
1283 case ALGORITHM_LEFT_SYMMETRIC:
1284 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1285 break;
1286 case ALGORITHM_RIGHT_SYMMETRIC:
1287 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1288 break;
1289 case ALGORITHM_PARITY_0:
1290 re->before.layout = ALGORITHM_PARITY_0_6;
1291 break;
1292 case ALGORITHM_PARITY_N:
1293 re->before.layout = ALGORITHM_PARITY_N_6;
1294 break;
1295 default:
1296 return "Cannot convert an array with this layout";
1297 }
1298 break;
1299 case 1:
1300 if (info->array.raid_disks != 2)
1301 return "Can only convert a 2-device array to RAID1";
1302 if (info->delta_disks != UnSet &&
1303 info->delta_disks != 0)
1304 return "Cannot set raid_disk when converting RAID5->RAID1";
1305 re->level = 1;
1306 info->new_chunk = 0;
1307 return NULL;
1308 default:
1309 return "Impossible level change requested";
1310 }
1311 break;
1312 case 6:
1313 switch (info->new_level) {
1314 case 4:
1315 case 5:
1316 delta_parity = -1;
1317 case 6:
1318 re->level = 6;
1319 re->before.data_disks = info->array.raid_disks - 2;
1320 re->before.layout = info->array.layout;
1321 break;
1322 default:
1323 return "Impossible level change requested";
1324 }
1325 break;
1326 }
1327
1328 /* If we reached here then it looks like a re-stripe is
1329 * happening. We have determined the intermediate level
1330 * and initial raid_disks/layout and stored these in 're'.
1331 *
1332 * We need to deduce the final layout that can be atomically
1333 * converted to the end state.
1334 */
1335 switch (info->new_level) {
1336 case 0:
1337 /* We can only get to RAID0 from RAID4 or RAID5
1338 * with appropriate layout and one extra device
1339 */
1340 if (re->level != 4 && re->level != 5)
1341 return "Cannot covert to RAID0 from this level";
1342
1343 switch (re->level) {
1344 case 4:
1345 re->before.layout = 0;
1346 re->after.layout = 0;
1347 break;
1348 case 5:
1349 re->after.layout = ALGORITHM_PARITY_N;
1350 break;
1351 }
1352 break;
1353
1354 case 4:
1355 /* We can only get to RAID4 from RAID5 */
1356 if (re->level != 4 && re->level != 5)
1357 return "Cannot convert to RAID4 from this level";
1358
1359 switch (re->level) {
1360 case 4:
1361 re->after.layout = 0;
1362 break;
1363 case 5:
1364 re->after.layout = ALGORITHM_PARITY_N;
1365 break;
1366 }
1367 break;
1368
1369 case 5:
1370 /* We get to RAID5 from RAID5 or RAID6 */
1371 if (re->level != 5 && re->level != 6)
1372 return "Cannot convert to RAID5 from this level";
1373
1374 switch (re->level) {
1375 case 5:
1376 if (info->new_layout == UnSet)
1377 re->after.layout = re->before.layout;
1378 else
1379 re->after.layout = info->new_layout;
1380 break;
1381 case 6:
1382 if (info->new_layout == UnSet)
1383 info->new_layout = re->before.layout;
1384
1385 /* after.layout needs to be raid6 version of new_layout */
1386 if (info->new_layout == ALGORITHM_PARITY_N)
1387 re->after.layout = ALGORITHM_PARITY_N;
1388 else {
1389 char layout[40];
1390 char *ls = map_num(r5layout, info->new_layout);
1391 int l;
1392 if (ls) {
1393 /* Current RAID6 layout has a RAID5
1394 * equivalent - good
1395 */
1396 strcat(strcpy(layout, ls), "-6");
1397 l = map_name(r6layout, layout);
1398 if (l == UnSet)
1399 return "Cannot find RAID6 layout to convert to";
1400 } else {
1401 /* Current RAID6 has no equivalent.
1402 * If it is already a '-6' layout we
1403 * can leave it unchanged, else we must
1404 * fail
1405 */
1406 ls = map_num(r6layout, info->new_layout);
1407 if (!ls ||
1408 strcmp(ls+strlen(ls)-2, "-6") != 0)
1409 return "Please specify new layout";
1410 l = info->new_layout;
1411 }
1412 re->after.layout = l;
1413 }
1414 }
1415 break;
1416
1417 case 6:
1418 /* We must already be at level 6 */
1419 if (re->level != 6)
1420 return "Impossible level change";
1421 if (info->new_layout == UnSet)
1422 re->after.layout = info->array.layout;
1423 else
1424 re->after.layout = info->new_layout;
1425 break;
1426 default:
1427 return "Impossible level change requested";
1428 }
1429 if (info->delta_disks == UnSet)
1430 info->delta_disks = delta_parity;
1431
1432 re->after.data_disks = (re->before.data_disks
1433 + info->delta_disks
1434 - delta_parity);
1435 switch (re->level) {
1436 case 6: re->parity = 2;
1437 break;
1438 case 4:
1439 case 5: re->parity = 1;
1440 break;
1441 default: re->parity = 0;
1442 break;
1443 }
1444 /* So we have a restripe operation, we need to calculate the number
1445 * of blocks per reshape operation.
1446 */
1447 re->new_size = info->component_size * re->before.data_disks;
1448 if (info->new_chunk == 0)
1449 info->new_chunk = info->array.chunk_size;
1450 if (re->after.data_disks == re->before.data_disks &&
1451 re->after.layout == re->before.layout &&
1452 info->new_chunk == info->array.chunk_size) {
1453 /* Nothing to change, can change level immediately. */
1454 re->level = info->new_level;
1455 re->backup_blocks = 0;
1456 return NULL;
1457 }
1458 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1459 /* chunk and layout changes make no difference */
1460 re->level = info->new_level;
1461 re->backup_blocks = 0;
1462 return NULL;
1463 }
1464
1465 if (re->after.data_disks == re->before.data_disks &&
1466 get_linux_version() < 2006032)
1467 return "in-place reshape is not safe before 2.6.32 - sorry.";
1468
1469 if (re->after.data_disks < re->before.data_disks &&
1470 get_linux_version() < 2006030)
1471 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1472
1473 re->backup_blocks = compute_backup_blocks(
1474 info->new_chunk, info->array.chunk_size,
1475 re->after.data_disks,
1476 re->before.data_disks);
1477 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1478
1479 re->new_size = info->component_size * re->after.data_disks;
1480 return NULL;
1481 }
1482
1483 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1484 char *text_version)
1485 {
1486 struct mdinfo *info;
1487 char *subarray;
1488 int ret_val = -1;
1489
1490 if ((st == NULL) || (sra == NULL))
1491 return ret_val;
1492
1493 if (text_version == NULL)
1494 text_version = sra->text_version;
1495 subarray = strchr(text_version+1, '/')+1;
1496 info = st->ss->container_content(st, subarray);
1497 if (info) {
1498 unsigned long long current_size = 0;
1499 unsigned long long new_size =
1500 info->custom_array_size/2;
1501
1502 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1503 new_size > current_size) {
1504 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1505 < 0)
1506 dprintf("Error: Cannot set array size");
1507 else {
1508 ret_val = 0;
1509 dprintf("Array size changed");
1510 }
1511 dprintf_cont(" from %llu to %llu.\n",
1512 current_size, new_size);
1513 }
1514 sysfs_free(info);
1515 } else
1516 dprintf("Error: set_array_size(): info pointer in NULL\n");
1517
1518 return ret_val;
1519 }
1520
1521 static int reshape_array(char *container, int fd, char *devname,
1522 struct supertype *st, struct mdinfo *info,
1523 int force, struct mddev_dev *devlist,
1524 unsigned long long data_offset,
1525 char *backup_file, int verbose, int forked,
1526 int restart, int freeze_reshape);
1527 static int reshape_container(char *container, char *devname,
1528 int mdfd,
1529 struct supertype *st,
1530 struct mdinfo *info,
1531 int force,
1532 char *backup_file, int verbose,
1533 int forked, int restart, int freeze_reshape);
1534
1535 int Grow_reshape(char *devname, int fd,
1536 struct mddev_dev *devlist,
1537 unsigned long long data_offset,
1538 struct context *c, struct shape *s)
1539 {
1540 /* Make some changes in the shape of an array.
1541 * The kernel must support the change.
1542 *
1543 * There are three different changes. Each can trigger
1544 * a resync or recovery so we freeze that until we have
1545 * requested everything (if kernel supports freezing - 2.6.30).
1546 * The steps are:
1547 * - change size (i.e. component_size)
1548 * - change level
1549 * - change layout/chunksize/ndisks
1550 *
1551 * The last can require a reshape. It is different on different
1552 * levels so we need to check the level before actioning it.
1553 * Some times the level change needs to be requested after the
1554 * reshape (e.g. raid6->raid5, raid5->raid0)
1555 *
1556 */
1557 struct mdu_array_info_s array;
1558 int rv = 0;
1559 struct supertype *st;
1560 char *subarray = NULL;
1561
1562 int frozen;
1563 int changed = 0;
1564 char *container = NULL;
1565 int cfd = -1;
1566
1567 struct mddev_dev *dv;
1568 int added_disks;
1569
1570 struct mdinfo info;
1571 struct mdinfo *sra;
1572
1573 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1574 pr_err("%s is not an active md array - aborting\n",
1575 devname);
1576 return 1;
1577 }
1578 if (data_offset != INVALID_SECTORS && array.level != 10
1579 && (array.level < 4 || array.level > 6)) {
1580 pr_err("--grow --data-offset not yet supported\n");
1581 return 1;
1582 }
1583
1584 if (s->size > 0 &&
1585 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1586 pr_err("cannot change component size at the same time as other changes.\n"
1587 " Change size first, then check data is intact before making other changes.\n");
1588 return 1;
1589 }
1590
1591 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1592 get_linux_version() < 2006032 &&
1593 !check_env("MDADM_FORCE_FEWER")) {
1594 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1595 " Please use a newer kernel\n");
1596 return 1;
1597 }
1598
1599 st = super_by_fd(fd, &subarray);
1600 if (!st) {
1601 pr_err("Unable to determine metadata format for %s\n", devname);
1602 return 1;
1603 }
1604 if (s->raiddisks > st->max_devs) {
1605 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1606 return 1;
1607 }
1608 if (s->level == 0 &&
1609 (array.state & (1<<MD_SB_BITMAP_PRESENT)) &&
1610 !(array.state & (1<<MD_SB_CLUSTERED))) {
1611 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
1612 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
1613 pr_err("failed to remove internal bitmap.\n");
1614 return 1;
1615 }
1616 }
1617
1618 /* in the external case we need to check that the requested reshape is
1619 * supported, and perform an initial check that the container holds the
1620 * pre-requisite spare devices (mdmon owns final validation)
1621 */
1622 if (st->ss->external) {
1623 int rv;
1624
1625 if (subarray) {
1626 container = st->container_devnm;
1627 cfd = open_dev_excl(st->container_devnm);
1628 } else {
1629 container = st->devnm;
1630 close(fd);
1631 cfd = open_dev_excl(st->devnm);
1632 fd = cfd;
1633 }
1634 if (cfd < 0) {
1635 pr_err("Unable to open container for %s\n",
1636 devname);
1637 free(subarray);
1638 return 1;
1639 }
1640
1641 rv = st->ss->load_container(st, cfd, NULL);
1642
1643 if (rv) {
1644 pr_err("Cannot read superblock for %s\n",
1645 devname);
1646 free(subarray);
1647 return 1;
1648 }
1649
1650 /* check if operation is supported for metadata handler */
1651 if (st->ss->container_content) {
1652 struct mdinfo *cc = NULL;
1653 struct mdinfo *content = NULL;
1654
1655 cc = st->ss->container_content(st, subarray);
1656 for (content = cc; content ; content = content->next) {
1657 int allow_reshape = 1;
1658
1659 /* check if reshape is allowed based on metadata
1660 * indications stored in content.array.status
1661 */
1662 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1663 allow_reshape = 0;
1664 if (content->array.state
1665 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1666 allow_reshape = 0;
1667 if (!allow_reshape) {
1668 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1669 devname, container);
1670 sysfs_free(cc);
1671 free(subarray);
1672 return 1;
1673 }
1674 }
1675 sysfs_free(cc);
1676 }
1677 if (mdmon_running(container))
1678 st->update_tail = &st->updates;
1679 }
1680
1681 added_disks = 0;
1682 for (dv = devlist; dv; dv = dv->next)
1683 added_disks++;
1684 if (s->raiddisks > array.raid_disks &&
1685 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1686 !c->force) {
1687 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1688 " Use --force to over-ride this check.\n",
1689 s->raiddisks - array.raid_disks,
1690 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1691 array.spare_disks + added_disks);
1692 return 1;
1693 }
1694
1695 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1696 | GET_STATE | GET_VERSION);
1697 if (sra) {
1698 if (st->ss->external && subarray == NULL) {
1699 array.level = LEVEL_CONTAINER;
1700 sra->array.level = LEVEL_CONTAINER;
1701 }
1702 } else {
1703 pr_err("failed to read sysfs parameters for %s\n",
1704 devname);
1705 return 1;
1706 }
1707 frozen = freeze(st);
1708 if (frozen < -1) {
1709 /* freeze() already spewed the reason */
1710 sysfs_free(sra);
1711 return 1;
1712 } else if (frozen < 0) {
1713 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1714 sysfs_free(sra);
1715 return 1;
1716 }
1717
1718 /* ========= set size =============== */
1719 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1720 unsigned long long orig_size = get_component_size(fd)/2;
1721 unsigned long long min_csize;
1722 struct mdinfo *mdi;
1723 int raid0_takeover = 0;
1724
1725 if (orig_size == 0)
1726 orig_size = (unsigned) array.size;
1727
1728 if (orig_size == 0) {
1729 pr_err("Cannot set device size in this type of array.\n");
1730 rv = 1;
1731 goto release;
1732 }
1733
1734 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1735 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1736 rv = 1;
1737 goto release;
1738 }
1739 sync_metadata(st);
1740 if (st->ss->external) {
1741 /* metadata can have size limitation
1742 * update size value according to metadata information
1743 */
1744 struct mdinfo *sizeinfo =
1745 st->ss->container_content(st, subarray);
1746 if (sizeinfo) {
1747 unsigned long long new_size =
1748 sizeinfo->custom_array_size/2;
1749 int data_disks = get_data_disks(
1750 sizeinfo->array.level,
1751 sizeinfo->array.layout,
1752 sizeinfo->array.raid_disks);
1753 new_size /= data_disks;
1754 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size, new_size,
1755 new_size * data_disks);
1756 s->size = new_size;
1757 sysfs_free(sizeinfo);
1758 }
1759 }
1760
1761 /* Update the size of each member device in case
1762 * they have been resized. This will never reduce
1763 * below the current used-size. The "size" attribute
1764 * understands '0' to mean 'max'.
1765 */
1766 min_csize = 0;
1767 rv = 0;
1768 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1769 if (sysfs_set_num(sra, mdi, "size",
1770 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1771 /* Probably kernel refusing to let us
1772 * reduce the size - not an error.
1773 */
1774 break;
1775 }
1776 if (array.not_persistent == 0 &&
1777 array.major_version == 0 &&
1778 get_linux_version() < 3001000) {
1779 /* Dangerous to allow size to exceed 2TB */
1780 unsigned long long csize;
1781 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1782 if (csize >= 2ULL*1024*1024*1024)
1783 csize = 2ULL*1024*1024*1024;
1784 if ((min_csize == 0 || (min_csize
1785 > csize)))
1786 min_csize = csize;
1787 }
1788 }
1789 }
1790 if (rv) {
1791 pr_err("Cannot set size on array members.\n");
1792 goto size_change_error;
1793 }
1794 if (min_csize && s->size > min_csize) {
1795 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1796 rv = 1;
1797 goto size_change_error;
1798 }
1799 if (min_csize && s->size == MAX_SIZE) {
1800 /* Don't let the kernel choose a size - it will get
1801 * it wrong
1802 */
1803 pr_err("Limited v0.90 array to 2TB per device\n");
1804 s->size = min_csize;
1805 }
1806 if (st->ss->external) {
1807 if (sra->array.level == 0) {
1808 rv = sysfs_set_str(sra, NULL, "level",
1809 "raid5");
1810 if (!rv) {
1811 raid0_takeover = 1;
1812 /* get array parameters after takeover
1813 * to change one parameter at time only
1814 */
1815 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1816 }
1817 }
1818 /* make sure mdmon is
1819 * aware of the new level */
1820 if (!mdmon_running(st->container_devnm))
1821 start_mdmon(st->container_devnm);
1822 ping_monitor(container);
1823 if (mdmon_running(st->container_devnm) &&
1824 st->update_tail == NULL)
1825 st->update_tail = &st->updates;
1826 }
1827
1828 if (s->size == MAX_SIZE)
1829 s->size = 0;
1830 array.size = s->size;
1831 if (s->size & ~INT32_MAX) {
1832 /* got truncated to 32bit, write to
1833 * component_size instead
1834 */
1835 if (sra)
1836 rv = sysfs_set_num(sra, NULL,
1837 "component_size", s->size);
1838 else
1839 rv = -1;
1840 } else {
1841 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1842
1843 /* manage array size when it is managed externally
1844 */
1845 if ((rv == 0) && st->ss->external)
1846 rv = set_array_size(st, sra, sra->text_version);
1847 }
1848
1849 if (raid0_takeover) {
1850 /* do not recync non-existing parity,
1851 * we will drop it anyway
1852 */
1853 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1854 /* go back to raid0, drop parity disk
1855 */
1856 sysfs_set_str(sra, NULL, "level", "raid0");
1857 ioctl(fd, GET_ARRAY_INFO, &array);
1858 }
1859
1860 size_change_error:
1861 if (rv != 0) {
1862 int err = errno;
1863
1864 /* restore metadata */
1865 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1866 UnSet, NULL, devname,
1867 ROLLBACK_METADATA_CHANGES,
1868 c->verbose) == 0)
1869 sync_metadata(st);
1870 pr_err("Cannot set device size for %s: %s\n",
1871 devname, strerror(err));
1872 if (err == EBUSY &&
1873 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1874 cont_err("Bitmap must be removed before size can be changed\n");
1875 rv = 1;
1876 goto release;
1877 }
1878 if (s->assume_clean) {
1879 /* This will fail on kernels older than 3.0 unless
1880 * a backport has been arranged.
1881 */
1882 if (sra == NULL ||
1883 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1884 pr_err("--assume-clean not supported with --grow on this kernel\n");
1885 }
1886 ioctl(fd, GET_ARRAY_INFO, &array);
1887 s->size = get_component_size(fd)/2;
1888 if (s->size == 0)
1889 s->size = array.size;
1890 if (c->verbose >= 0) {
1891 if (s->size == orig_size)
1892 pr_err("component size of %s unchanged at %lluK\n",
1893 devname, s->size);
1894 else
1895 pr_err("component size of %s has been set to %lluK\n",
1896 devname, s->size);
1897 }
1898 changed = 1;
1899 } else if (array.level != LEVEL_CONTAINER) {
1900 s->size = get_component_size(fd)/2;
1901 if (s->size == 0)
1902 s->size = array.size;
1903 }
1904
1905 /* See if there is anything else to do */
1906 if ((s->level == UnSet || s->level == array.level) &&
1907 (s->layout_str == NULL) &&
1908 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1909 data_offset == INVALID_SECTORS &&
1910 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1911 /* Nothing more to do */
1912 if (!changed && c->verbose >= 0)
1913 pr_err("%s: no change requested\n",
1914 devname);
1915 goto release;
1916 }
1917
1918 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1919 * current implementation assumes that following conditions must be met:
1920 * - RAID10:
1921 * - far_copies == 1
1922 * - near_copies == 2
1923 */
1924 if ((s->level == 0 && array.level == 10 && sra &&
1925 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1926 (s->level == 0 && array.level == 1 && sra)) {
1927 int err;
1928 err = remove_disks_for_takeover(st, sra, array.layout);
1929 if (err) {
1930 dprintf("Array cannot be reshaped\n");
1931 if (cfd > -1)
1932 close(cfd);
1933 rv = 1;
1934 goto release;
1935 }
1936 /* Make sure mdmon has seen the device removal
1937 * and updated metadata before we continue with
1938 * level change
1939 */
1940 if (container)
1941 ping_monitor(container);
1942 }
1943
1944 memset(&info, 0, sizeof(info));
1945 info.array = array;
1946 sysfs_init(&info, fd, NULL);
1947 strcpy(info.text_version, sra->text_version);
1948 info.component_size = s->size*2;
1949 info.new_level = s->level;
1950 info.new_chunk = s->chunk * 1024;
1951 if (info.array.level == LEVEL_CONTAINER) {
1952 info.delta_disks = UnSet;
1953 info.array.raid_disks = s->raiddisks;
1954 } else if (s->raiddisks)
1955 info.delta_disks = s->raiddisks - info.array.raid_disks;
1956 else
1957 info.delta_disks = UnSet;
1958 if (s->layout_str == NULL) {
1959 info.new_layout = UnSet;
1960 if (info.array.level == 6 &&
1961 (info.new_level == 6 || info.new_level == UnSet) &&
1962 info.array.layout >= 16) {
1963 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
1964 cont_err("during the reshape, please specify --layout=preserve\n");
1965 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1966 rv = 1;
1967 goto release;
1968 }
1969 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1970 strcmp(s->layout_str, "normalize") == 0) {
1971 /* If we have a -6 RAID6 layout, remove the '-6'. */
1972 info.new_layout = UnSet;
1973 if (info.array.level == 6 && info.new_level == UnSet) {
1974 char l[40], *h;
1975 strcpy(l, map_num(r6layout, info.array.layout));
1976 h = strrchr(l, '-');
1977 if (h && strcmp(h, "-6") == 0) {
1978 *h = 0;
1979 info.new_layout = map_name(r6layout, l);
1980 }
1981 } else {
1982 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
1983 rv = 1;
1984 goto release;
1985 }
1986 } else if (strcmp(s->layout_str, "preserve") == 0) {
1987 /* This means that a non-standard RAID6 layout
1988 * is OK.
1989 * In particular:
1990 * - When reshape a RAID6 (e.g. adding a device)
1991 * which is in a non-standard layout, it is OK
1992 * to preserve that layout.
1993 * - When converting a RAID5 to RAID6, leave it in
1994 * the XXX-6 layout, don't re-layout.
1995 */
1996 if (info.array.level == 6 && info.new_level == UnSet)
1997 info.new_layout = info.array.layout;
1998 else if (info.array.level == 5 && info.new_level == 6) {
1999 char l[40];
2000 strcpy(l, map_num(r5layout, info.array.layout));
2001 strcat(l, "-6");
2002 info.new_layout = map_name(r6layout, l);
2003 } else {
2004 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
2005 rv = 1;
2006 goto release;
2007 }
2008 } else {
2009 int l = info.new_level;
2010 if (l == UnSet)
2011 l = info.array.level;
2012 switch (l) {
2013 case 5:
2014 info.new_layout = map_name(r5layout, s->layout_str);
2015 break;
2016 case 6:
2017 info.new_layout = map_name(r6layout, s->layout_str);
2018 break;
2019 case 10:
2020 info.new_layout = parse_layout_10(s->layout_str);
2021 break;
2022 case LEVEL_FAULTY:
2023 info.new_layout = parse_layout_faulty(s->layout_str);
2024 break;
2025 default:
2026 pr_err("layout not meaningful with this level\n");
2027 rv = 1;
2028 goto release;
2029 }
2030 if (info.new_layout == UnSet) {
2031 pr_err("layout %s not understood for this level\n",
2032 s->layout_str);
2033 rv = 1;
2034 goto release;
2035 }
2036 }
2037
2038 if (array.level == LEVEL_FAULTY) {
2039 if (s->level != UnSet && s->level != array.level) {
2040 pr_err("cannot change level of Faulty device\n");
2041 rv =1 ;
2042 }
2043 if (s->chunk) {
2044 pr_err("cannot set chunksize of Faulty device\n");
2045 rv =1 ;
2046 }
2047 if (s->raiddisks && s->raiddisks != 1) {
2048 pr_err("cannot set raid_disks of Faulty device\n");
2049 rv =1 ;
2050 }
2051 if (s->layout_str) {
2052 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2053 dprintf("Cannot get array information.\n");
2054 goto release;
2055 }
2056 array.layout = info.new_layout;
2057 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2058 pr_err("failed to set new layout\n");
2059 rv = 1;
2060 } else if (c->verbose >= 0)
2061 printf("layout for %s set to %d\n",
2062 devname, array.layout);
2063 }
2064 } else if (array.level == LEVEL_CONTAINER) {
2065 /* This change is to be applied to every array in the
2066 * container. This is only needed when the metadata imposes
2067 * restraints of the various arrays in the container.
2068 * Currently we only know that IMSM requires all arrays
2069 * to have the same number of devices so changing the
2070 * number of devices (On-Line Capacity Expansion) must be
2071 * performed at the level of the container
2072 */
2073 rv = reshape_container(container, devname, -1, st, &info,
2074 c->force, c->backup_file, c->verbose, 0, 0, 0);
2075 frozen = 0;
2076 } else {
2077 /* get spare devices from external metadata
2078 */
2079 if (st->ss->external) {
2080 struct mdinfo *info2;
2081
2082 info2 = st->ss->container_content(st, subarray);
2083 if (info2) {
2084 info.array.spare_disks =
2085 info2->array.spare_disks;
2086 sysfs_free(info2);
2087 }
2088 }
2089
2090 /* Impose these changes on a single array. First
2091 * check that the metadata is OK with the change. */
2092
2093 if (reshape_super(st, 0, info.new_level,
2094 info.new_layout, info.new_chunk,
2095 info.array.raid_disks, info.delta_disks,
2096 c->backup_file, devname, APPLY_METADATA_CHANGES,
2097 c->verbose)) {
2098 rv = 1;
2099 goto release;
2100 }
2101 sync_metadata(st);
2102 rv = reshape_array(container, fd, devname, st, &info, c->force,
2103 devlist, data_offset, c->backup_file, c->verbose,
2104 0, 0, 0);
2105 frozen = 0;
2106 }
2107 release:
2108 sysfs_free(sra);
2109 if (frozen > 0)
2110 unfreeze(st);
2111 return rv;
2112 }
2113
2114 /* verify_reshape_position()
2115 * Function checks if reshape position in metadata is not farther
2116 * than position in md.
2117 * Return value:
2118 * 0 : not valid sysfs entry
2119 * it can be caused by not started reshape, it should be started
2120 * by reshape array or raid0 array is before takeover
2121 * -1 : error, reshape position is obviously wrong
2122 * 1 : success, reshape progress correct or updated
2123 */
2124 static int verify_reshape_position(struct mdinfo *info, int level)
2125 {
2126 int ret_val = 0;
2127 char buf[40];
2128 int rv;
2129
2130 /* read sync_max, failure can mean raid0 array */
2131 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2132
2133 if (rv > 0) {
2134 char *ep;
2135 unsigned long long position = strtoull(buf, &ep, 0);
2136
2137 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2138 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2139 position *= get_data_disks(level,
2140 info->new_layout,
2141 info->array.raid_disks);
2142 if (info->reshape_progress < position) {
2143 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2144 info->reshape_progress, position);
2145 info->reshape_progress = position;
2146 ret_val = 1;
2147 } else if (info->reshape_progress > position) {
2148 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2149 position, info->reshape_progress);
2150 ret_val = -1;
2151 } else {
2152 dprintf("Reshape position in md and metadata are the same;");
2153 ret_val = 1;
2154 }
2155 }
2156 } else if (rv == 0) {
2157 /* for valid sysfs entry, 0-length content
2158 * should be indicated as error
2159 */
2160 ret_val = -1;
2161 }
2162
2163 return ret_val;
2164 }
2165
2166 static unsigned long long choose_offset(unsigned long long lo,
2167 unsigned long long hi,
2168 unsigned long long min,
2169 unsigned long long max)
2170 {
2171 /* Choose a new offset between hi and lo.
2172 * It must be between min and max, but
2173 * we would prefer something near the middle of hi/lo, and also
2174 * prefer to be aligned to a big power of 2.
2175 *
2176 * So we start with the middle, then for each bit,
2177 * starting at '1' and increasing, if it is set, we either
2178 * add it or subtract it if possible, preferring the option
2179 * which is furthest from the boundary.
2180 *
2181 * We stop once we get a 1MB alignment. As units are in sectors,
2182 * 1MB = 2*1024 sectors.
2183 */
2184 unsigned long long choice = (lo + hi) / 2;
2185 unsigned long long bit = 1;
2186
2187 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2188 unsigned long long bigger, smaller;
2189 if (! (bit & choice))
2190 continue;
2191 bigger = choice + bit;
2192 smaller = choice - bit;
2193 if (bigger > max && smaller < min)
2194 break;
2195 if (bigger > max)
2196 choice = smaller;
2197 else if (smaller < min)
2198 choice = bigger;
2199 else if (hi - bigger > smaller - lo)
2200 choice = bigger;
2201 else
2202 choice = smaller;
2203 }
2204 return choice;
2205 }
2206
2207 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2208 char *devname, int delta_disks,
2209 unsigned long long data_offset,
2210 unsigned long long min,
2211 int can_fallback)
2212 {
2213 struct mdinfo *sd;
2214 int dir = 0;
2215 int err = 0;
2216 unsigned long long before, after;
2217
2218 /* Need to find min space before and after so same is used
2219 * on all devices
2220 */
2221 before = UINT64_MAX;
2222 after = UINT64_MAX;
2223 for (sd = sra->devs; sd; sd = sd->next) {
2224 char *dn;
2225 int dfd;
2226 int rv;
2227 struct supertype *st2;
2228 struct mdinfo info2;
2229
2230 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2231 continue;
2232 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2233 dfd = dev_open(dn, O_RDONLY);
2234 if (dfd < 0) {
2235 pr_err("%s: cannot open component %s\n",
2236 devname, dn ? dn : "-unknown-");
2237 goto release;
2238 }
2239 st2 = dup_super(st);
2240 rv = st2->ss->load_super(st2,dfd, NULL);
2241 close(dfd);
2242 if (rv) {
2243 free(st2);
2244 pr_err("%s: cannot get superblock from %s\n",
2245 devname, dn);
2246 goto release;
2247 }
2248 st2->ss->getinfo_super(st2, &info2, NULL);
2249 st2->ss->free_super(st2);
2250 free(st2);
2251 if (info2.space_before == 0 &&
2252 info2.space_after == 0) {
2253 /* Metadata doesn't support data_offset changes */
2254 if (!can_fallback)
2255 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2256 devname);
2257 goto fallback;
2258 }
2259 if (before > info2.space_before)
2260 before = info2.space_before;
2261 if (after > info2.space_after)
2262 after = info2.space_after;
2263
2264 if (data_offset != INVALID_SECTORS) {
2265 if (dir == 0) {
2266 if (info2.data_offset == data_offset) {
2267 pr_err("%s: already has that data_offset\n",
2268 dn);
2269 goto release;
2270 }
2271 if (data_offset < info2.data_offset)
2272 dir = -1;
2273 else
2274 dir = 1;
2275 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2276 (data_offset >= info2.data_offset && dir == -1)) {
2277 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2278 dn);
2279 goto release;
2280 }
2281 }
2282 }
2283 if (before == UINT64_MAX)
2284 /* impossible really, there must be no devices */
2285 return 1;
2286
2287 for (sd = sra->devs; sd; sd = sd->next) {
2288 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2289 unsigned long long new_data_offset;
2290
2291 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2292 continue;
2293 if (delta_disks < 0) {
2294 /* Don't need any space as array is shrinking
2295 * just move data_offset up by min
2296 */
2297 if (data_offset == INVALID_SECTORS)
2298 new_data_offset = sd->data_offset + min;
2299 else {
2300 if (data_offset < sd->data_offset + min) {
2301 pr_err("--data-offset too small for %s\n",
2302 dn);
2303 goto release;
2304 }
2305 new_data_offset = data_offset;
2306 }
2307 } else if (delta_disks > 0) {
2308 /* need space before */
2309 if (before < min) {
2310 if (can_fallback)
2311 goto fallback;
2312 pr_err("Insufficient head-space for reshape on %s\n",
2313 dn);
2314 goto release;
2315 }
2316 if (data_offset == INVALID_SECTORS)
2317 new_data_offset = sd->data_offset - min;
2318 else {
2319 if (data_offset > sd->data_offset - min) {
2320 pr_err("--data-offset too large for %s\n",
2321 dn);
2322 goto release;
2323 }
2324 new_data_offset = data_offset;
2325 }
2326 } else {
2327 if (dir == 0) {
2328 /* can move up or down. If 'data_offset'
2329 * was set we would have already decided,
2330 * so just choose direction with most space.
2331 */
2332 if (before > after)
2333 dir = -1;
2334 else
2335 dir = 1;
2336 }
2337 sysfs_set_str(sra, NULL, "reshape_direction",
2338 dir == 1 ? "backwards" : "forwards");
2339 if (dir > 0) {
2340 /* Increase data offset */
2341 if (after < min) {
2342 if (can_fallback)
2343 goto fallback;
2344 pr_err("Insufficient tail-space for reshape on %s\n",
2345 dn);
2346 goto release;
2347 }
2348 if (data_offset != INVALID_SECTORS &&
2349 data_offset < sd->data_offset + min) {
2350 pr_err("--data-offset too small on %s\n",
2351 dn);
2352 goto release;
2353 }
2354 if (data_offset != INVALID_SECTORS)
2355 new_data_offset = data_offset;
2356 else
2357 new_data_offset = choose_offset(sd->data_offset,
2358 sd->data_offset + after,
2359 sd->data_offset + min,
2360 sd->data_offset + after);
2361 } else {
2362 /* Decrease data offset */
2363 if (before < min) {
2364 if (can_fallback)
2365 goto fallback;
2366 pr_err("insufficient head-room on %s\n",
2367 dn);
2368 goto release;
2369 }
2370 if (data_offset != INVALID_SECTORS &&
2371 data_offset < sd->data_offset - min) {
2372 pr_err("--data-offset too small on %s\n",
2373 dn);
2374 goto release;
2375 }
2376 if (data_offset != INVALID_SECTORS)
2377 new_data_offset = data_offset;
2378 else
2379 new_data_offset = choose_offset(sd->data_offset - before,
2380 sd->data_offset,
2381 sd->data_offset - before,
2382 sd->data_offset - min);
2383 }
2384 }
2385 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2386 if (err < 0 && errno == E2BIG) {
2387 /* try again after increasing data size to max */
2388 err = sysfs_set_num(sra, sd, "size", 0);
2389 if (err < 0 && errno == EINVAL &&
2390 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2391 /* some kernels have a bug where you cannot
2392 * use '0' on spare devices. */
2393 sysfs_set_num(sra, sd, "size",
2394 (sra->component_size + after)/2);
2395 }
2396 err = sysfs_set_num(sra, sd, "new_offset",
2397 new_data_offset);
2398 }
2399 if (err < 0) {
2400 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2401 pr_err("data-offset is too big for %s\n",
2402 dn);
2403 goto release;
2404 }
2405 if (sd == sra->devs &&
2406 (errno == ENOENT || errno == E2BIG))
2407 /* Early kernel, no 'new_offset' file,
2408 * or kernel doesn't like us.
2409 * For RAID5/6 this is not fatal
2410 */
2411 return 1;
2412 pr_err("Cannot set new_offset for %s\n",
2413 dn);
2414 break;
2415 }
2416 }
2417 return err;
2418 release:
2419 return -1;
2420 fallback:
2421 /* Just use a backup file */
2422 return 1;
2423 }
2424
2425 static int raid10_reshape(char *container, int fd, char *devname,
2426 struct supertype *st, struct mdinfo *info,
2427 struct reshape *reshape,
2428 unsigned long long data_offset,
2429 int force, int verbose)
2430 {
2431 /* Changing raid_disks, layout, chunksize or possibly
2432 * just data_offset for a RAID10.
2433 * We must always change data_offset. We change by at least
2434 * ->min_offset_change which is the largest of the old and new
2435 * chunk sizes.
2436 * If raid_disks is increasing, then data_offset must decrease
2437 * by at least this copy size.
2438 * If raid_disks is unchanged, data_offset must increase or
2439 * decrease by at least min_offset_change but preferably by much more.
2440 * We choose half of the available space.
2441 * If raid_disks is decreasing, data_offset must increase by
2442 * at least min_offset_change. To allow of this, component_size
2443 * must be decreased by the same amount.
2444 *
2445 * So we calculate the required minimum and direction, possibly
2446 * reduce the component_size, then iterate through the devices
2447 * and set the new_data_offset.
2448 * If that all works, we set chunk_size, layout, raid_disks, and start
2449 * 'reshape'
2450 */
2451 struct mdinfo *sra;
2452 unsigned long long min;
2453 int err = 0;
2454
2455 sra = sysfs_read(fd, NULL,
2456 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2457 );
2458 if (!sra) {
2459 pr_err("%s: Cannot get array details from sysfs\n",
2460 devname);
2461 goto release;
2462 }
2463 min = reshape->min_offset_change;
2464
2465 if (info->delta_disks)
2466 sysfs_set_str(sra, NULL, "reshape_direction",
2467 info->delta_disks < 0 ? "backwards" : "forwards");
2468 if (info->delta_disks < 0 &&
2469 info->space_after < min) {
2470 int rv = sysfs_set_num(sra, NULL, "component_size",
2471 (sra->component_size -
2472 min)/2);
2473 if (rv) {
2474 pr_err("cannot reduce component size\n");
2475 goto release;
2476 }
2477 }
2478 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2479 min, 0);
2480 if (err == 1) {
2481 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2482 cont_err("supported on this kernel\n");
2483 err = -1;
2484 }
2485 if (err < 0)
2486 goto release;
2487
2488 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2489 err = errno;
2490 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2491 err = errno;
2492 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2493 info->array.raid_disks + info->delta_disks) < 0)
2494 err = errno;
2495 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2496 err = errno;
2497 if (err) {
2498 pr_err("Cannot set array shape for %s\n",
2499 devname);
2500 if (err == EBUSY &&
2501 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2502 cont_err(" Bitmap must be removed before shape can be changed\n");
2503 goto release;
2504 }
2505 sysfs_free(sra);
2506 return 0;
2507 release:
2508 sysfs_free(sra);
2509 return 1;
2510 }
2511
2512 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2513 {
2514 struct mdinfo *sra, *sd;
2515 /* Initialisation to silence compiler warning */
2516 unsigned long long min_space_before = 0, min_space_after = 0;
2517 int first = 1;
2518
2519 sra = sysfs_read(fd, NULL, GET_DEVS);
2520 if (!sra)
2521 return;
2522 for (sd = sra->devs; sd; sd = sd->next) {
2523 char *dn;
2524 int dfd;
2525 struct supertype *st2;
2526 struct mdinfo info2;
2527
2528 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2529 continue;
2530 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2531 dfd = dev_open(dn, O_RDONLY);
2532 if (dfd < 0)
2533 break;
2534 st2 = dup_super(st);
2535 if (st2->ss->load_super(st2,dfd, NULL)) {
2536 close(dfd);
2537 free(st2);
2538 break;
2539 }
2540 close(dfd);
2541 st2->ss->getinfo_super(st2, &info2, NULL);
2542 st2->ss->free_super(st2);
2543 free(st2);
2544 if (first ||
2545 min_space_before > info2.space_before)
2546 min_space_before = info2.space_before;
2547 if (first ||
2548 min_space_after > info2.space_after)
2549 min_space_after = info2.space_after;
2550 first = 0;
2551 }
2552 if (sd == NULL && !first) {
2553 info->space_after = min_space_after;
2554 info->space_before = min_space_before;
2555 }
2556 sysfs_free(sra);
2557 }
2558
2559 static void update_cache_size(char *container, struct mdinfo *sra,
2560 struct mdinfo *info,
2561 int disks, unsigned long long blocks)
2562 {
2563 /* Check that the internal stripe cache is
2564 * large enough, or it won't work.
2565 * It must hold at least 4 stripes of the larger
2566 * chunk size
2567 */
2568 unsigned long cache;
2569 cache = max(info->array.chunk_size, info->new_chunk);
2570 cache *= 4; /* 4 stripes minimum */
2571 cache /= 512; /* convert to sectors */
2572 /* make sure there is room for 'blocks' with a bit to spare */
2573 if (cache < 16 + blocks / disks)
2574 cache = 16 + blocks / disks;
2575 cache /= (4096/512); /* Convert from sectors to pages */
2576
2577 if (sra->cache_size < cache)
2578 subarray_set_num(container, sra, "stripe_cache_size",
2579 cache+1);
2580 }
2581
2582 static int impose_reshape(struct mdinfo *sra,
2583 struct mdinfo *info,
2584 struct supertype *st,
2585 int fd,
2586 int restart,
2587 char *devname, char *container,
2588 struct reshape *reshape)
2589 {
2590 struct mdu_array_info_s array;
2591
2592 sra->new_chunk = info->new_chunk;
2593
2594 if (restart) {
2595 /* for external metadata checkpoint saved by mdmon can be lost
2596 * or missed /due to e.g. crash/. Check if md is not during
2597 * restart farther than metadata points to.
2598 * If so, this means metadata information is obsolete.
2599 */
2600 if (st->ss->external)
2601 verify_reshape_position(info, reshape->level);
2602 sra->reshape_progress = info->reshape_progress;
2603 } else {
2604 sra->reshape_progress = 0;
2605 if (reshape->after.data_disks < reshape->before.data_disks)
2606 /* start from the end of the new array */
2607 sra->reshape_progress = (sra->component_size
2608 * reshape->after.data_disks);
2609 }
2610
2611 ioctl(fd, GET_ARRAY_INFO, &array);
2612 if (info->array.chunk_size == info->new_chunk &&
2613 reshape->before.layout == reshape->after.layout &&
2614 st->ss->external == 0) {
2615 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2616 array.raid_disks = reshape->after.data_disks + reshape->parity;
2617 if (!restart &&
2618 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2619 int err = errno;
2620
2621 pr_err("Cannot set device shape for %s: %s\n",
2622 devname, strerror(errno));
2623
2624 if (err == EBUSY &&
2625 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2626 cont_err("Bitmap must be removed before shape can be changed\n");
2627
2628 goto release;
2629 }
2630 } else if (!restart) {
2631 /* set them all just in case some old 'new_*' value
2632 * persists from some earlier problem.
2633 */
2634 int err = 0;
2635 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2636 err = errno;
2637 if (!err && sysfs_set_num(sra, NULL, "layout",
2638 reshape->after.layout) < 0)
2639 err = errno;
2640 if (!err && subarray_set_num(container, sra, "raid_disks",
2641 reshape->after.data_disks +
2642 reshape->parity) < 0)
2643 err = errno;
2644 if (err) {
2645 pr_err("Cannot set device shape for %s\n",
2646 devname);
2647
2648 if (err == EBUSY &&
2649 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2650 cont_err("Bitmap must be removed before shape can be changed\n");
2651 goto release;
2652 }
2653 }
2654 return 0;
2655 release:
2656 return -1;
2657 }
2658
2659 static int impose_level(int fd, int level, char *devname, int verbose)
2660 {
2661 char *c;
2662 struct mdu_array_info_s array;
2663 struct mdinfo info;
2664 sysfs_init(&info, fd, NULL);
2665
2666 ioctl(fd, GET_ARRAY_INFO, &array);
2667 if (level == 0 &&
2668 (array.level >= 4 && array.level <= 6)) {
2669 /* To convert to RAID0 we need to fail and
2670 * remove any non-data devices. */
2671 int found = 0;
2672 int d;
2673 int data_disks = array.raid_disks - 1;
2674 if (array.level == 6)
2675 data_disks -= 1;
2676 if (array.level == 5 &&
2677 array.layout != ALGORITHM_PARITY_N)
2678 return -1;
2679 if (array.level == 6 &&
2680 array.layout != ALGORITHM_PARITY_N_6)
2681 return -1;
2682 sysfs_set_str(&info, NULL,"sync_action", "idle");
2683 /* First remove any spares so no recovery starts */
2684 for (d = 0, found = 0;
2685 d < MAX_DISKS && found < array.nr_disks;
2686 d++) {
2687 mdu_disk_info_t disk;
2688 disk.number = d;
2689 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2690 continue;
2691 if (disk.major == 0 && disk.minor == 0)
2692 continue;
2693 found++;
2694 if ((disk.state & (1 << MD_DISK_ACTIVE))
2695 && disk.raid_disk < data_disks)
2696 /* keep this */
2697 continue;
2698 ioctl(fd, HOT_REMOVE_DISK,
2699 makedev(disk.major, disk.minor));
2700 }
2701 /* Now fail anything left */
2702 ioctl(fd, GET_ARRAY_INFO, &array);
2703 for (d = 0, found = 0;
2704 d < MAX_DISKS && found < array.nr_disks;
2705 d++) {
2706 int cnt;
2707 mdu_disk_info_t disk;
2708 disk.number = d;
2709 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2710 continue;
2711 if (disk.major == 0 && disk.minor == 0)
2712 continue;
2713 found++;
2714 if ((disk.state & (1 << MD_DISK_ACTIVE))
2715 && disk.raid_disk < data_disks)
2716 /* keep this */
2717 continue;
2718 ioctl(fd, SET_DISK_FAULTY,
2719 makedev(disk.major, disk.minor));
2720 cnt = 5;
2721 while (ioctl(fd, HOT_REMOVE_DISK,
2722 makedev(disk.major, disk.minor)) < 0
2723 && errno == EBUSY
2724 && cnt--) {
2725 usleep(10000);
2726 }
2727 }
2728 }
2729 c = map_num(pers, level);
2730 if (c) {
2731 int err = sysfs_set_str(&info, NULL, "level", c);
2732 if (err) {
2733 err = errno;
2734 pr_err("%s: could not set level to %s\n",
2735 devname, c);
2736 if (err == EBUSY &&
2737 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2738 cont_err("Bitmap must be removed 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 char buf[20];
2826
2827 /* when reshaping a RAID0, the component_size might be zero.
2828 * So try to fix that up.
2829 */
2830 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2831 dprintf("Cannot get array information.\n");
2832 goto release;
2833 }
2834 if (array.level == 0 && info->component_size == 0) {
2835 get_dev_size(fd, NULL, &array_size);
2836 info->component_size = array_size / array.raid_disks;
2837 }
2838
2839 if (array.level == 10)
2840 /* Need space_after info */
2841 get_space_after(fd, st, info);
2842
2843 if (info->reshape_active) {
2844 int new_level = info->new_level;
2845 info->new_level = UnSet;
2846 if (info->delta_disks > 0)
2847 info->array.raid_disks -= info->delta_disks;
2848 msg = analyse_change(devname, info, &reshape);
2849 info->new_level = new_level;
2850 if (info->delta_disks > 0)
2851 info->array.raid_disks += info->delta_disks;
2852 if (!restart)
2853 /* Make sure the array isn't read-only */
2854 ioctl(fd, RESTART_ARRAY_RW, 0);
2855 } else
2856 msg = analyse_change(devname, info, &reshape);
2857 if (msg) {
2858 /* if msg == "", error has already been printed */
2859 if (msg[0])
2860 pr_err("%s\n", msg);
2861 goto release;
2862 }
2863 if (restart &&
2864 (reshape.level != info->array.level ||
2865 reshape.before.layout != info->array.layout ||
2866 reshape.before.data_disks + reshape.parity
2867 != info->array.raid_disks - max(0, info->delta_disks))) {
2868 pr_err("reshape info is not in native format - cannot continue.\n");
2869 goto release;
2870 }
2871
2872 if (st->ss->external && restart && (info->reshape_progress == 0) &&
2873 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
2874 (strncmp(buf, "reshape", 7) == 0))) {
2875 /* When reshape is restarted from '0', very begin of array
2876 * it is possible that for external metadata reshape and array
2877 * configuration doesn't happen.
2878 * Check if md has the same opinion, and reshape is restarted
2879 * from 0. If so, this is regular reshape start after reshape
2880 * switch in metadata to next array only.
2881 */
2882 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2883 (info->reshape_progress == 0))
2884 restart = 0;
2885 }
2886 if (restart) {
2887 /* reshape already started. just skip to monitoring the reshape */
2888 if (reshape.backup_blocks == 0)
2889 return 0;
2890 if (restart & RESHAPE_NO_BACKUP)
2891 return 0;
2892
2893 /* Need 'sra' down at 'started:' */
2894 sra = sysfs_read(fd, NULL,
2895 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2896 GET_CACHE);
2897 if (!sra) {
2898 pr_err("%s: Cannot get array details from sysfs\n",
2899 devname);
2900 goto release;
2901 }
2902
2903 if (!backup_file)
2904 backup_file = locate_backup(sra->sys_name);
2905
2906 goto started;
2907 }
2908 /* The container is frozen but the array may not be.
2909 * So freeze the array so spares don't get put to the wrong use
2910 * FIXME there should probably be a cleaner separation between
2911 * freeze_array and freeze_container.
2912 */
2913 sysfs_freeze_array(info);
2914 /* Check we have enough spares to not be degraded */
2915 added_disks = 0;
2916 for (dv = devlist; dv ; dv=dv->next)
2917 added_disks++;
2918 spares_needed = max(reshape.before.data_disks,
2919 reshape.after.data_disks)
2920 + reshape.parity - array.raid_disks;
2921
2922 if (!force &&
2923 info->new_level > 1 && info->array.level > 1 &&
2924 spares_needed > info->array.spare_disks + added_disks) {
2925 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2926 " Use --force to over-ride this check.\n",
2927 spares_needed,
2928 spares_needed == 1 ? "" : "s",
2929 info->array.spare_disks + added_disks);
2930 goto release;
2931 }
2932 /* Check we have enough spares to not fail */
2933 spares_needed = max(reshape.before.data_disks,
2934 reshape.after.data_disks)
2935 - array.raid_disks;
2936 if ((info->new_level > 1 || info->new_level == 0) &&
2937 spares_needed > info->array.spare_disks +added_disks) {
2938 pr_err("Need %d spare%s to create working array, 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 struct mdinfo *sd;
3146 if (errno != EINVAL) {
3147 pr_err("Failed to initiate reshape!\n");
3148 goto release;
3149 }
3150 /* revert data_offset and try the old way */
3151 for (sd = sra->devs; sd; sd = sd->next) {
3152 sysfs_set_num(sra, sd, "new_offset",
3153 sd->data_offset);
3154 sysfs_set_str(sra, NULL, "reshape_direction",
3155 "forwards");
3156 }
3157 break;
3158 }
3159 if (info->new_level == reshape.level)
3160 return 0;
3161 /* need to adjust level when reshape completes */
3162 switch(fork()) {
3163 case -1: /* ignore error, but don't wait */
3164 return 0;
3165 default: /* parent */
3166 return 0;
3167 case 0:
3168 map_fork();
3169 break;
3170 }
3171 close(fd);
3172 wait_reshape(sra);
3173 fd = open_dev(sra->sys_name);
3174 if (fd >= 0)
3175 impose_level(fd, info->new_level, devname, verbose);
3176 return 0;
3177 case 1: /* Couldn't set data_offset, try the old way */
3178 if (data_offset != INVALID_SECTORS) {
3179 pr_err("Cannot update data_offset on this array\n");
3180 goto release;
3181 }
3182 break;
3183 }
3184
3185 started:
3186 /* Decide how many blocks (sectors) for a reshape
3187 * unit. The number we have so far is just a minimum
3188 */
3189 blocks = reshape.backup_blocks;
3190 if (reshape.before.data_disks ==
3191 reshape.after.data_disks) {
3192 /* Make 'blocks' bigger for better throughput, but
3193 * not so big that we reject it below.
3194 * Try for 16 megabytes
3195 */
3196 while (blocks * 32 < sra->component_size &&
3197 blocks < 16*1024*2)
3198 blocks *= 2;
3199 } else
3200 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3201
3202 if (blocks >= sra->component_size/2) {
3203 pr_err("%s: Something wrong - reshape aborted\n",
3204 devname);
3205 goto release;
3206 }
3207
3208 /* Now we need to open all these devices so we can read/write.
3209 */
3210 nrdisks = max(reshape.before.data_disks,
3211 reshape.after.data_disks) + reshape.parity
3212 + sra->array.spare_disks;
3213 fdlist = xcalloc((1+nrdisks), sizeof(int));
3214 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3215
3216 odisks = reshape.before.data_disks + reshape.parity;
3217 d = reshape_prepare_fdlist(devname, sra, odisks,
3218 nrdisks, blocks, backup_file,
3219 fdlist, offsets);
3220 if (d < odisks) {
3221 goto release;
3222 }
3223 if ((st->ss->manage_reshape == NULL) ||
3224 (st->ss->recover_backup == NULL)) {
3225 if (backup_file == NULL) {
3226 if (reshape.after.data_disks <=
3227 reshape.before.data_disks) {
3228 pr_err("%s: Cannot grow - need backup-file\n",
3229 devname);
3230 pr_err(" Please provide one with \"--backup=...\"\n");
3231 goto release;
3232 } else if (d == odisks) {
3233 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3234 goto release;
3235 }
3236 } else {
3237 if (!reshape_open_backup_file(backup_file, fd, devname,
3238 (signed)blocks,
3239 fdlist+d, offsets+d,
3240 sra->sys_name,
3241 restart)) {
3242 goto release;
3243 }
3244 d++;
3245 }
3246 }
3247
3248 update_cache_size(container, sra, info,
3249 min(reshape.before.data_disks, reshape.after.data_disks),
3250 blocks);
3251
3252 /* Right, everything seems fine. Let's kick things off.
3253 * If only changing raid_disks, use ioctl, else use
3254 * sysfs.
3255 */
3256 sync_metadata(st);
3257
3258 if (impose_reshape(sra, info, st, fd, restart,
3259 devname, container, &reshape) < 0)
3260 goto release;
3261
3262 err = start_reshape(sra, restart, reshape.before.data_disks,
3263 reshape.after.data_disks);
3264 if (err) {
3265 pr_err("Cannot %s reshape for %s\n",
3266 restart ? "continue" : "start",
3267 devname);
3268 goto release;
3269 }
3270 if (restart)
3271 sysfs_set_str(sra, NULL, "array_state", "active");
3272 if (freeze_reshape) {
3273 free(fdlist);
3274 free(offsets);
3275 sysfs_free(sra);
3276 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3277 sra->reshape_progress);
3278 return 1;
3279 }
3280
3281 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3282 if (continue_via_systemd(container ?: sra->sys_name)) {
3283 free(fdlist);
3284 free(offsets);
3285 sysfs_free(sra);
3286 return 0;
3287 }
3288
3289 /* Now we just need to kick off the reshape and watch, while
3290 * handling backups of the data...
3291 * This is all done by a forked background process.
3292 */
3293 switch(forked ? 0 : fork()) {
3294 case -1:
3295 pr_err("Cannot run child to monitor reshape: %s\n",
3296 strerror(errno));
3297 abort_reshape(sra);
3298 goto release;
3299 default:
3300 free(fdlist);
3301 free(offsets);
3302 sysfs_free(sra);
3303 return 0;
3304 case 0:
3305 map_fork();
3306 break;
3307 }
3308
3309 /* If another array on the same devices is busy, the
3310 * reshape will wait for them. This would mean that
3311 * the first section that we suspend will stay suspended
3312 * for a long time. So check on that possibility
3313 * by looking for "DELAYED" in /proc/mdstat, and if found,
3314 * wait a while
3315 */
3316 do {
3317 struct mdstat_ent *mds, *m;
3318 delayed = 0;
3319 mds = mdstat_read(1, 0);
3320 for (m = mds; m; m = m->next)
3321 if (strcmp(m->devnm, sra->sys_name) == 0) {
3322 if (m->resync &&
3323 m->percent == RESYNC_DELAYED)
3324 delayed = 1;
3325 if (m->resync == 0)
3326 /* Haven't started the reshape thread
3327 * yet, wait a bit
3328 */
3329 delayed = 2;
3330 break;
3331 }
3332 free_mdstat(mds);
3333 if (delayed == 1 && get_linux_version() < 3007000) {
3334 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3335 " You might experience problems until other reshapes complete.\n");
3336 delayed = 0;
3337 }
3338 if (delayed)
3339 mdstat_wait(30 - (delayed-1) * 25);
3340 } while (delayed);
3341 mdstat_close();
3342 close(fd);
3343 if (check_env("MDADM_GROW_VERIFY"))
3344 fd = open(devname, O_RDONLY | O_DIRECT);
3345 else
3346 fd = -1;
3347 mlockall(MCL_FUTURE);
3348
3349 signal(SIGTERM, catch_term);
3350
3351 if (st->ss->external) {
3352 /* metadata handler takes it from here */
3353 done = st->ss->manage_reshape(
3354 fd, sra, &reshape, st, blocks,
3355 fdlist, offsets,
3356 d - odisks, fdlist+odisks,
3357 offsets+odisks);
3358 } else
3359 done = child_monitor(
3360 fd, sra, &reshape, st, blocks,
3361 fdlist, offsets,
3362 d - odisks, fdlist+odisks,
3363 offsets+odisks);
3364
3365 free(fdlist);
3366 free(offsets);
3367
3368 if (backup_file && done) {
3369 char *bul;
3370 bul = make_backup(sra->sys_name);
3371 if (bul) {
3372 char buf[1024];
3373 int l = readlink(bul, buf, sizeof(buf) - 1);
3374 if (l > 0) {
3375 buf[l]=0;
3376 unlink(buf);
3377 }
3378 unlink(bul);
3379 free(bul);
3380 }
3381 unlink(backup_file);
3382 }
3383 if (!done) {
3384 abort_reshape(sra);
3385 goto out;
3386 }
3387
3388 if (!st->ss->external &&
3389 !(reshape.before.data_disks != reshape.after.data_disks
3390 && info->custom_array_size) &&
3391 info->new_level == reshape.level &&
3392 !forked) {
3393 /* no need to wait for the reshape to finish as
3394 * there is nothing more to do.
3395 */
3396 sysfs_free(sra);
3397 exit(0);
3398 }
3399 wait_reshape(sra);
3400
3401 if (st->ss->external) {
3402 /* Re-load the metadata as much could have changed */
3403 int cfd = open_dev(st->container_devnm);
3404 if (cfd >= 0) {
3405 flush_mdmon(container);
3406 st->ss->free_super(st);
3407 st->ss->load_container(st, cfd, container);
3408 close(cfd);
3409 }
3410 }
3411
3412 /* set new array size if required customer_array_size is used
3413 * by this metadata.
3414 */
3415 if (reshape.before.data_disks !=
3416 reshape.after.data_disks &&
3417 info->custom_array_size)
3418 set_array_size(st, info, info->text_version);
3419
3420 if (info->new_level != reshape.level) {
3421 if (fd < 0)
3422 fd = open(devname, O_RDONLY);
3423 impose_level(fd, info->new_level, devname, verbose);
3424 close(fd);
3425 if (info->new_level == 0)
3426 st->update_tail = NULL;
3427 }
3428 out:
3429 sysfs_free(sra);
3430 if (forked)
3431 return 0;
3432 unfreeze(st);
3433 exit(0);
3434
3435 release:
3436 free(fdlist);
3437 free(offsets);
3438 if (orig_level != UnSet && sra) {
3439 c = map_num(pers, orig_level);
3440 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3441 pr_err("aborting level change\n");
3442 }
3443 sysfs_free(sra);
3444 if (!forked)
3445 unfreeze(st);
3446 return 1;
3447 }
3448
3449 /* mdfd handle is passed to be closed in child process (after fork).
3450 */
3451 int reshape_container(char *container, char *devname,
3452 int mdfd,
3453 struct supertype *st,
3454 struct mdinfo *info,
3455 int force,
3456 char *backup_file, int verbose,
3457 int forked, int restart, int freeze_reshape)
3458 {
3459 struct mdinfo *cc = NULL;
3460 int rv = restart;
3461 char last_devnm[32] = "";
3462
3463 /* component_size is not meaningful for a container,
3464 * so pass '0' meaning 'no change'
3465 */
3466 if (!restart &&
3467 reshape_super(st, 0, info->new_level,
3468 info->new_layout, info->new_chunk,
3469 info->array.raid_disks, info->delta_disks,
3470 backup_file, devname, APPLY_METADATA_CHANGES,
3471 verbose)) {
3472 unfreeze(st);
3473 return 1;
3474 }
3475
3476 sync_metadata(st);
3477
3478 /* ping monitor to be sure that update is on disk
3479 */
3480 ping_monitor(container);
3481
3482 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3483 if (continue_via_systemd(container))
3484 return 0;
3485
3486 switch (forked ? 0 : fork()) {
3487 case -1: /* error */
3488 perror("Cannot fork to complete reshape\n");
3489 unfreeze(st);
3490 return 1;
3491 default: /* parent */
3492 if (!freeze_reshape)
3493 printf("%s: multi-array reshape continues in background\n", Name);
3494 return 0;
3495 case 0: /* child */
3496 map_fork();
3497 break;
3498 }
3499
3500 /* close unused handle in child process
3501 */
3502 if (mdfd > -1)
3503 close(mdfd);
3504
3505 while(1) {
3506 /* For each member array with reshape_active,
3507 * we need to perform the reshape.
3508 * We pick the first array that needs reshaping and
3509 * reshape it. reshape_array() will re-read the metadata
3510 * so the next time through a different array should be
3511 * ready for reshape.
3512 * It is possible that the 'different' array will not
3513 * be assembled yet. In that case we simple exit.
3514 * When it is assembled, the mdadm which assembles it
3515 * will take over the reshape.
3516 */
3517 struct mdinfo *content;
3518 int fd;
3519 struct mdstat_ent *mdstat;
3520 char *adev;
3521 int devid;
3522
3523 sysfs_free(cc);
3524
3525 cc = st->ss->container_content(st, NULL);
3526
3527 for (content = cc; content ; content = content->next) {
3528 char *subarray;
3529 if (!content->reshape_active)
3530 continue;
3531
3532 subarray = strchr(content->text_version+1, '/')+1;
3533 mdstat = mdstat_by_subdev(subarray, container);
3534 if (!mdstat)
3535 continue;
3536 if (mdstat->active == 0) {
3537 pr_err("Skipping inactive array %s.\n",
3538 mdstat->devnm);
3539 free_mdstat(mdstat);
3540 mdstat = NULL;
3541 continue;
3542 }
3543 break;
3544 }
3545 if (!content)
3546 break;
3547
3548 devid = devnm2devid(mdstat->devnm);
3549 adev = map_dev(major(devid), minor(devid), 0);
3550 if (!adev)
3551 adev = content->text_version;
3552
3553 fd = open_dev(mdstat->devnm);
3554 if (fd < 0) {
3555 pr_err("Device %s cannot be opened for reshape.\n", adev);
3556 break;
3557 }
3558
3559 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3560 /* Do not allow for multiple reshape_array() calls for
3561 * the same array.
3562 * It can happen when reshape_array() returns without
3563 * error, when reshape is not finished (wrong reshape
3564 * starting/continuation conditions). Mdmon doesn't
3565 * switch to next array in container and reentry
3566 * conditions for the same array occur.
3567 * This is possibly interim until the behaviour of
3568 * reshape_array is resolved().
3569 */
3570 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3571 close(fd);
3572 break;
3573 }
3574 strcpy(last_devnm, mdstat->devnm);
3575
3576 sysfs_init(content, fd, mdstat->devnm);
3577
3578 if (mdmon_running(container))
3579 flush_mdmon(container);
3580
3581 rv = reshape_array(container, fd, adev, st,
3582 content, force, NULL, INVALID_SECTORS,
3583 backup_file, verbose, 1, restart,
3584 freeze_reshape);
3585 close(fd);
3586
3587 if (freeze_reshape) {
3588 sysfs_free(cc);
3589 exit(0);
3590 }
3591
3592 restart = 0;
3593 if (rv)
3594 break;
3595
3596 if (mdmon_running(container))
3597 flush_mdmon(container);
3598 }
3599 if (!rv)
3600 unfreeze(st);
3601 sysfs_free(cc);
3602 exit(0);
3603 }
3604
3605 /*
3606 * We run a child process in the background which performs the following
3607 * steps:
3608 * - wait for resync to reach a certain point
3609 * - suspend io to the following section
3610 * - backup that section
3611 * - allow resync to proceed further
3612 * - resume io
3613 * - discard the backup.
3614 *
3615 * When are combined in slightly different ways in the three cases.
3616 * Grow:
3617 * - suspend/backup/allow/wait/resume/discard
3618 * Shrink:
3619 * - allow/wait/suspend/backup/allow/wait/resume/discard
3620 * same-size:
3621 * - wait/resume/discard/suspend/backup/allow
3622 *
3623 * suspend/backup/allow always come together
3624 * wait/resume/discard do too.
3625 * For the same-size case we have two backups to improve flow.
3626 *
3627 */
3628
3629 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3630 unsigned long long backup_point,
3631 unsigned long long wait_point,
3632 unsigned long long *suspend_point,
3633 unsigned long long *reshape_completed, int *frozen)
3634 {
3635 /* This function is called repeatedly by the reshape manager.
3636 * It determines how much progress can safely be made and allows
3637 * that progress.
3638 * - 'info' identifies the array and particularly records in
3639 * ->reshape_progress the metadata's knowledge of progress
3640 * This is a sector offset from the start of the array
3641 * of the next array block to be relocated. This number
3642 * may increase from 0 or decrease from array_size, depending
3643 * on the type of reshape that is happening.
3644 * Note that in contrast, 'sync_completed' is a block count of the
3645 * reshape so far. It gives the distance between the start point
3646 * (head or tail of device) and the next place that data will be
3647 * written. It always increases.
3648 * - 'reshape' is the structure created by analyse_change
3649 * - 'backup_point' shows how much the metadata manager has backed-up
3650 * data. For reshapes with increasing progress, it is the next address
3651 * to be backed up, previous addresses have been backed-up. For
3652 * decreasing progress, it is the earliest address that has been
3653 * backed up - later address are also backed up.
3654 * So addresses between reshape_progress and backup_point are
3655 * backed up providing those are in the 'correct' order.
3656 * - 'wait_point' is an array address. When reshape_completed
3657 * passes this point, progress_reshape should return. It might
3658 * return earlier if it determines that ->reshape_progress needs
3659 * to be updated or further backup is needed.
3660 * - suspend_point is maintained by progress_reshape and the caller
3661 * should not touch it except to initialise to zero.
3662 * It is an array address and it only increases in 2.6.37 and earlier.
3663 * This makes it difficult to handle reducing reshapes with
3664 * external metadata.
3665 * However: it is similar to backup_point in that it records the
3666 * other end of a suspended region from reshape_progress.
3667 * it is moved to extend the region that is safe to backup and/or
3668 * reshape
3669 * - reshape_completed is read from sysfs and returned. The caller
3670 * should copy this into ->reshape_progress when it has reason to
3671 * believe that the metadata knows this, and any backup outside this
3672 * has been erased.
3673 *
3674 * Return value is:
3675 * 1 if more data from backup_point - but only as far as suspend_point,
3676 * should be backed up
3677 * 0 if things are progressing smoothly
3678 * -1 if the reshape is finished because it is all done,
3679 * -2 if the reshape is finished due to an error.
3680 */
3681
3682 int advancing = (reshape->after.data_disks
3683 >= reshape->before.data_disks);
3684 unsigned long long need_backup; /* All data between start of array and
3685 * here will at some point need to
3686 * be backed up.
3687 */
3688 unsigned long long read_offset, write_offset;
3689 unsigned long long write_range;
3690 unsigned long long max_progress, target, completed;
3691 unsigned long long array_size = (info->component_size
3692 * reshape->before.data_disks);
3693 int fd;
3694 char buf[20];
3695
3696 /* First, we unsuspend any region that is now known to be safe.
3697 * If suspend_point is on the 'wrong' side of reshape_progress, then
3698 * we don't have or need suspension at the moment. This is true for
3699 * native metadata when we don't need to back-up.
3700 */
3701 if (advancing) {
3702 if (info->reshape_progress <= *suspend_point)
3703 sysfs_set_num(info, NULL, "suspend_lo",
3704 info->reshape_progress);
3705 } else {
3706 /* Note: this won't work in 2.6.37 and before.
3707 * Something somewhere should make sure we don't need it!
3708 */
3709 if (info->reshape_progress >= *suspend_point)
3710 sysfs_set_num(info, NULL, "suspend_hi",
3711 info->reshape_progress);
3712 }
3713
3714 /* Now work out how far it is safe to progress.
3715 * If the read_offset for ->reshape_progress is less than
3716 * 'blocks' beyond the write_offset, we can only progress as far
3717 * as a backup.
3718 * Otherwise we can progress until the write_offset for the new location
3719 * reaches (within 'blocks' of) the read_offset at the current location.
3720 * However that region must be suspended unless we are using native
3721 * metadata.
3722 * If we need to suspend more, we limit it to 128M per device, which is
3723 * rather arbitrary and should be some time-based calculation.
3724 */
3725 read_offset = info->reshape_progress / reshape->before.data_disks;
3726 write_offset = info->reshape_progress / reshape->after.data_disks;
3727 write_range = info->new_chunk/512;
3728 if (reshape->before.data_disks == reshape->after.data_disks)
3729 need_backup = array_size;
3730 else
3731 need_backup = reshape->backup_blocks;
3732 if (advancing) {
3733 if (read_offset < write_offset + write_range)
3734 max_progress = backup_point;
3735 else
3736 max_progress =
3737 read_offset *
3738 reshape->after.data_disks;
3739 } else {
3740 if (read_offset > write_offset - write_range)
3741 /* Can only progress as far as has been backed up,
3742 * which must be suspended */
3743 max_progress = backup_point;
3744 else if (info->reshape_progress <= need_backup)
3745 max_progress = backup_point;
3746 else {
3747 if (info->array.major_version >= 0)
3748 /* Can progress until backup is needed */
3749 max_progress = need_backup;
3750 else {
3751 /* Can progress until metadata update is required */
3752 max_progress =
3753 read_offset *
3754 reshape->after.data_disks;
3755 /* but data must be suspended */
3756 if (max_progress < *suspend_point)
3757 max_progress = *suspend_point;
3758 }
3759 }
3760 }
3761
3762 /* We know it is safe to progress to 'max_progress' providing
3763 * it is suspended or we are using native metadata.
3764 * Consider extending suspend_point 128M per device if it
3765 * is less than 64M per device beyond reshape_progress.
3766 * But always do a multiple of 'blocks'
3767 * FIXME this is too big - it takes to long to complete
3768 * this much.
3769 */
3770 target = 64*1024*2 * min(reshape->before.data_disks,
3771 reshape->after.data_disks);
3772 target /= reshape->backup_blocks;
3773 if (target < 2)
3774 target = 2;
3775 target *= reshape->backup_blocks;
3776
3777 /* For externally managed metadata we always need to suspend IO to
3778 * the area being reshaped so we regularly push suspend_point forward.
3779 * For native metadata we only need the suspend if we are going to do
3780 * a backup.
3781 */
3782 if (advancing) {
3783 if ((need_backup > info->reshape_progress
3784 || info->array.major_version < 0) &&
3785 *suspend_point < info->reshape_progress + target) {
3786 if (need_backup < *suspend_point + 2 * target)
3787 *suspend_point = need_backup;
3788 else if (*suspend_point + 2 * target < array_size)
3789 *suspend_point += 2 * target;
3790 else
3791 *suspend_point = array_size;
3792 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3793 if (max_progress > *suspend_point)
3794 max_progress = *suspend_point;
3795 }
3796 } else {
3797 if (info->array.major_version >= 0) {
3798 /* Only need to suspend when about to backup */
3799 if (info->reshape_progress < need_backup * 2 &&
3800 *suspend_point > 0) {
3801 *suspend_point = 0;
3802 sysfs_set_num(info, NULL, "suspend_lo", 0);
3803 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3804 }
3805 } else {
3806 /* Need to suspend continually */
3807 if (info->reshape_progress < *suspend_point)
3808 *suspend_point = info->reshape_progress;
3809 if (*suspend_point + target < info->reshape_progress)
3810 /* No need to move suspend region yet */;
3811 else {
3812 if (*suspend_point >= 2 * target)
3813 *suspend_point -= 2 * target;
3814 else
3815 *suspend_point = 0;
3816 sysfs_set_num(info, NULL, "suspend_lo",
3817 *suspend_point);
3818 }
3819 if (max_progress < *suspend_point)
3820 max_progress = *suspend_point;
3821 }
3822 }
3823
3824 /* now set sync_max to allow that progress. sync_max, like
3825 * sync_completed is a count of sectors written per device, so
3826 * we find the difference between max_progress and the start point,
3827 * and divide that by after.data_disks to get a sync_max
3828 * number.
3829 * At the same time we convert wait_point to a similar number
3830 * for comparing against sync_completed.
3831 */
3832 /* scale down max_progress to per_disk */
3833 max_progress /= reshape->after.data_disks;
3834 /* Round to chunk size as some kernels give an erroneously high number */
3835 max_progress /= info->new_chunk/512;
3836 max_progress *= info->new_chunk/512;
3837 /* And round to old chunk size as the kernel wants that */
3838 max_progress /= info->array.chunk_size/512;
3839 max_progress *= info->array.chunk_size/512;
3840 /* Limit progress to the whole device */
3841 if (max_progress > info->component_size)
3842 max_progress = info->component_size;
3843 wait_point /= reshape->after.data_disks;
3844 if (!advancing) {
3845 /* switch from 'device offset' to 'processed block count' */
3846 max_progress = info->component_size - max_progress;
3847 wait_point = info->component_size - wait_point;
3848 }
3849
3850 if (!*frozen)
3851 sysfs_set_num(info, NULL, "sync_max", max_progress);
3852
3853 /* Now wait. If we have already reached the point that we were
3854 * asked to wait to, don't wait at all, else wait for any change.
3855 * We need to select on 'sync_completed' as that is the place that
3856 * notifications happen, but we are really interested in
3857 * 'reshape_position'
3858 */
3859 fd = sysfs_get_fd(info, NULL, "sync_completed");
3860 if (fd < 0)
3861 goto check_progress;
3862
3863 if (sysfs_fd_get_ll(fd, &completed) < 0)
3864 goto check_progress;
3865
3866 while (completed < max_progress && completed < wait_point) {
3867 /* Check that sync_action is still 'reshape' to avoid
3868 * waiting forever on a dead array
3869 */
3870 char action[20];
3871 if (sysfs_get_str(info, NULL, "sync_action",
3872 action, 20) <= 0 ||
3873 strncmp(action, "reshape", 7) != 0)
3874 break;
3875 /* Some kernels reset 'sync_completed' to zero
3876 * before setting 'sync_action' to 'idle'.
3877 * So we need these extra tests.
3878 */
3879 if (completed == 0 && advancing
3880 && strncmp(action, "idle", 4) == 0
3881 && info->reshape_progress > 0)
3882 break;
3883 if (completed == 0 && !advancing
3884 && strncmp(action, "idle", 4) == 0
3885 && info->reshape_progress < (info->component_size
3886 * reshape->after.data_disks))
3887 break;
3888 sysfs_wait(fd, NULL);
3889 if (sysfs_fd_get_ll(fd, &completed) < 0)
3890 goto check_progress;
3891 }
3892 /* Some kernels reset 'sync_completed' to zero,
3893 * we need to have real point we are in md.
3894 * So in that case, read 'reshape_position' from sysfs.
3895 */
3896 if (completed == 0) {
3897 unsigned long long reshapep;
3898 char action[20];
3899 if (sysfs_get_str(info, NULL, "sync_action",
3900 action, 20) > 0 &&
3901 strncmp(action, "idle", 4) == 0 &&
3902 sysfs_get_ll(info, NULL,
3903 "reshape_position", &reshapep) == 0)
3904 *reshape_completed = reshapep;
3905 } else {
3906 /* some kernels can give an incorrectly high
3907 * 'completed' number, so round down */
3908 completed /= (info->new_chunk/512);
3909 completed *= (info->new_chunk/512);
3910 /* Convert 'completed' back in to a 'progress' number */
3911 completed *= reshape->after.data_disks;
3912 if (!advancing)
3913 completed = (info->component_size
3914 * reshape->after.data_disks
3915 - completed);
3916 *reshape_completed = completed;
3917 }
3918
3919 close(fd);
3920
3921 /* We return the need_backup flag. Caller will decide
3922 * how much - a multiple of ->backup_blocks up to *suspend_point
3923 */
3924 if (advancing)
3925 return need_backup > info->reshape_progress;
3926 else
3927 return need_backup >= info->reshape_progress;
3928
3929 check_progress:
3930 /* if we couldn't read a number from sync_completed, then
3931 * either the reshape did complete, or it aborted.
3932 * We can tell which by checking for 'none' in reshape_position.
3933 * If it did abort, then it might immediately restart if it
3934 * it was just a device failure that leaves us degraded but
3935 * functioning.
3936 */
3937 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3938 || strncmp(buf, "none", 4) != 0) {
3939 /* The abort might only be temporary. Wait up to 10
3940 * seconds for fd to contain a valid number again.
3941 */
3942 int wait = 10000;
3943 int rv = -2;
3944 unsigned long long new_sync_max;
3945 while (fd >= 0 && rv < 0 && wait > 0) {
3946 if (sysfs_wait(fd, &wait) != 1)
3947 break;
3948 switch (sysfs_fd_get_ll(fd, &completed)) {
3949 case 0:
3950 /* all good again */
3951 rv = 1;
3952 /* If "sync_max" is no longer max_progress
3953 * we need to freeze things
3954 */
3955 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3956 *frozen = (new_sync_max != max_progress);
3957 break;
3958 case -2: /* read error - abort */
3959 wait = 0;
3960 break;
3961 }
3962 }
3963 if (fd >= 0)
3964 close(fd);
3965 return rv; /* abort */
3966 } else {
3967 /* Maybe racing with array shutdown - check state */
3968 if (fd >= 0)
3969 close(fd);
3970 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3971 || strncmp(buf, "inactive", 8) == 0
3972 || strncmp(buf, "clear",5) == 0)
3973 return -2; /* abort */
3974 return -1; /* complete */
3975 }
3976 }
3977
3978 /* FIXME return status is never checked */
3979 static int grow_backup(struct mdinfo *sra,
3980 unsigned long long offset, /* per device */
3981 unsigned long stripes, /* per device, in old chunks */
3982 int *sources, unsigned long long *offsets,
3983 int disks, int chunk, int level, int layout,
3984 int dests, int *destfd, unsigned long long *destoffsets,
3985 int part, int *degraded,
3986 char *buf)
3987 {
3988 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3989 * to storage 'destfd' (offset 'destoffsets'), after first
3990 * suspending IO. Then allow resync to continue
3991 * over the suspended section.
3992 * Use part 'part' of the backup-super-block.
3993 */
3994 int odata = disks;
3995 int rv = 0;
3996 int i;
3997 unsigned long long ll;
3998 int new_degraded;
3999 //printf("offset %llu\n", offset);
4000 if (level >= 4)
4001 odata--;
4002 if (level == 6)
4003 odata--;
4004
4005 /* Check that array hasn't become degraded, else we might backup the wrong data */
4006 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4007 return -1; /* FIXME this error is ignored */
4008 new_degraded = (int)ll;
4009 if (new_degraded != *degraded) {
4010 /* check each device to ensure it is still working */
4011 struct mdinfo *sd;
4012 for (sd = sra->devs ; sd ; sd = sd->next) {
4013 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4014 continue;
4015 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4016 char sbuf[20];
4017 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
4018 strstr(sbuf, "faulty") ||
4019 strstr(sbuf, "in_sync") == NULL) {
4020 /* this device is dead */
4021 sd->disk.state = (1<<MD_DISK_FAULTY);
4022 if (sd->disk.raid_disk >= 0 &&
4023 sources[sd->disk.raid_disk] >= 0) {
4024 close(sources[sd->disk.raid_disk]);
4025 sources[sd->disk.raid_disk] = -1;
4026 }
4027 }
4028 }
4029 }
4030 *degraded = new_degraded;
4031 }
4032 if (part) {
4033 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4034 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4035 } else {
4036 bsb.arraystart = __cpu_to_le64(offset * odata);
4037 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4038 }
4039 if (part)
4040 bsb.magic[15] = '2';
4041 for (i = 0; i < dests; i++)
4042 if (part)
4043 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4044 else
4045 lseek64(destfd[i], destoffsets[i], 0);
4046
4047 rv = save_stripes(sources, offsets,
4048 disks, chunk, level, layout,
4049 dests, destfd,
4050 offset*512*odata, stripes * chunk * odata,
4051 buf);
4052
4053 if (rv)
4054 return rv;
4055 bsb.mtime = __cpu_to_le64(time(0));
4056 for (i = 0; i < dests; i++) {
4057 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4058
4059 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4060 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4061 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4062 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4063
4064 rv = -1;
4065 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4066 != destoffsets[i] - 4096)
4067 break;
4068 if (write(destfd[i], &bsb, 512) != 512)
4069 break;
4070 if (destoffsets[i] > 4096) {
4071 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4072 destoffsets[i]+stripes*chunk*odata)
4073 break;
4074 if (write(destfd[i], &bsb, 512) != 512)
4075 break;
4076 }
4077 fsync(destfd[i]);
4078 rv = 0;
4079 }
4080
4081 return rv;
4082 }
4083
4084 /* in 2.6.30, the value reported by sync_completed can be
4085 * less that it should be by one stripe.
4086 * This only happens when reshape hits sync_max and pauses.
4087 * So allow wait_backup to either extent sync_max further
4088 * than strictly necessary, or return before the
4089 * sync has got quite as far as we would really like.
4090 * This is what 'blocks2' is for.
4091 * The various caller give appropriate values so that
4092 * every works.
4093 */
4094 /* FIXME return value is often ignored */
4095 static int forget_backup(int dests, int *destfd,
4096 unsigned long long *destoffsets,
4097 int part)
4098 {
4099 /*
4100 * Erase backup 'part' (which is 0 or 1)
4101 */
4102 int i;
4103 int rv;
4104
4105 if (part) {
4106 bsb.arraystart2 = __cpu_to_le64(0);
4107 bsb.length2 = __cpu_to_le64(0);
4108 } else {
4109 bsb.arraystart = __cpu_to_le64(0);
4110 bsb.length = __cpu_to_le64(0);
4111 }
4112 bsb.mtime = __cpu_to_le64(time(0));
4113 rv = 0;
4114 for (i = 0; i < dests; i++) {
4115 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4116 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4117 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4118 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4119 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4120 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4121 destoffsets[i]-4096)
4122 rv = -1;
4123 if (rv == 0 &&
4124 write(destfd[i], &bsb, 512) != 512)
4125 rv = -1;
4126 fsync(destfd[i]);
4127 }
4128 return rv;
4129 }
4130
4131 static void fail(char *msg)
4132 {
4133 int rv;
4134 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4135 rv |= (write(2, "\n", 1) != 1);
4136 exit(rv ? 1 : 2);
4137 }
4138
4139 static char *abuf, *bbuf;
4140 static unsigned long long abuflen;
4141 static void validate(int afd, int bfd, unsigned long long offset)
4142 {
4143 /* check that the data in the backup against the array.
4144 * This is only used for regression testing and should not
4145 * be used while the array is active
4146 */
4147 if (afd < 0)
4148 return;
4149 lseek64(bfd, offset - 4096, 0);
4150 if (read(bfd, &bsb2, 512) != 512)
4151 fail("cannot read bsb");
4152 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4153 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4154 fail("first csum bad");
4155 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4156 fail("magic is bad");
4157 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4158 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4159 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4160 fail("second csum bad");
4161
4162 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4163 fail("devstart is wrong");
4164
4165 if (bsb2.length) {
4166 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4167
4168 if (abuflen < len) {
4169 free(abuf);
4170 free(bbuf);
4171 abuflen = len;
4172 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4173 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4174 abuflen = 0;
4175 /* just stop validating on mem-alloc failure */
4176 return;
4177 }
4178 }
4179
4180 lseek64(bfd, offset, 0);
4181 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4182 //printf("len %llu\n", len);
4183 fail("read first backup failed");
4184 }
4185 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4186 if ((unsigned long long)read(afd, abuf, len) != len)
4187 fail("read first from array failed");
4188 if (memcmp(bbuf, abuf, len) != 0) {
4189 #if 0
4190 int i;
4191 printf("offset=%llu len=%llu\n",
4192 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4193 for (i=0; i<len; i++)
4194 if (bbuf[i] != abuf[i]) {
4195 printf("first diff byte %d\n", i);
4196 break;
4197 }
4198 #endif
4199 fail("data1 compare failed");
4200 }
4201 }
4202 if (bsb2.length2) {
4203 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4204
4205 if (abuflen < len) {
4206 free(abuf);
4207 free(bbuf);
4208 abuflen = len;
4209 abuf = xmalloc(abuflen);
4210 bbuf = xmalloc(abuflen);
4211 }
4212
4213 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4214 if ((unsigned long long)read(bfd, bbuf, len) != len)
4215 fail("read second backup failed");
4216 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4217 if ((unsigned long long)read(afd, abuf, len) != len)
4218 fail("read second from array failed");
4219 if (memcmp(bbuf, abuf, len) != 0)
4220 fail("data2 compare failed");
4221 }
4222 }
4223
4224 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4225 struct supertype *st, unsigned long blocks,
4226 int *fds, unsigned long long *offsets,
4227 int dests, int *destfd, unsigned long long *destoffsets)
4228 {
4229 /* Monitor a reshape where backup is being performed using
4230 * 'native' mechanism - either to a backup file, or
4231 * to some space in a spare.
4232 */
4233 char *buf;
4234 int degraded = -1;
4235 unsigned long long speed;
4236 unsigned long long suspend_point, array_size;
4237 unsigned long long backup_point, wait_point;
4238 unsigned long long reshape_completed;
4239 int done = 0;
4240 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4241 int part = 0; /* The next part of the backup area to fill. It may already
4242 * be full, so we need to check */
4243 int level = reshape->level;
4244 int layout = reshape->before.layout;
4245 int data = reshape->before.data_disks;
4246 int disks = reshape->before.data_disks + reshape->parity;
4247 int chunk = sra->array.chunk_size;
4248 struct mdinfo *sd;
4249 unsigned long stripes;
4250 int uuid[4];
4251 int frozen = 0;
4252
4253 /* set up the backup-super-block. This requires the
4254 * uuid from the array.
4255 */
4256 /* Find a superblock */
4257 for (sd = sra->devs; sd; sd = sd->next) {
4258 char *dn;
4259 int devfd;
4260 int ok;
4261 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4262 continue;
4263 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4264 devfd = dev_open(dn, O_RDONLY);
4265 if (devfd < 0)
4266 continue;
4267 ok = st->ss->load_super(st, devfd, NULL);
4268 close(devfd);
4269 if (ok == 0)
4270 break;
4271 }
4272 if (!sd) {
4273 pr_err("Cannot find a superblock\n");
4274 return 0;
4275 }
4276
4277 memset(&bsb, 0, 512);
4278 memcpy(bsb.magic, "md_backup_data-1", 16);
4279 st->ss->uuid_from_super(st, uuid);
4280 memcpy(bsb.set_uuid, uuid, 16);
4281 bsb.mtime = __cpu_to_le64(time(0));
4282 bsb.devstart2 = blocks;
4283
4284 stripes = blocks / (sra->array.chunk_size/512) /
4285 reshape->before.data_disks;
4286
4287 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4288 /* Don't start the 'reshape' */
4289 return 0;
4290 if (reshape->before.data_disks == reshape->after.data_disks) {
4291 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4292 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4293 }
4294
4295 if (increasing) {
4296 array_size = sra->component_size * reshape->after.data_disks;
4297 backup_point = sra->reshape_progress;
4298 suspend_point = 0;
4299 } else {
4300 array_size = sra->component_size * reshape->before.data_disks;
4301 backup_point = reshape->backup_blocks;
4302 suspend_point = array_size;
4303 }
4304
4305 while (!done) {
4306 int rv;
4307
4308 /* Want to return as soon the oldest backup slot can
4309 * be released as that allows us to start backing up
4310 * some more, providing suspend_point has been
4311 * advanced, which it should have.
4312 */
4313 if (increasing) {
4314 wait_point = array_size;
4315 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4316 wait_point = (__le64_to_cpu(bsb.arraystart) +
4317 __le64_to_cpu(bsb.length));
4318 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4319 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4320 __le64_to_cpu(bsb.length2));
4321 } else {
4322 wait_point = 0;
4323 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4324 wait_point = __le64_to_cpu(bsb.arraystart);
4325 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4326 wait_point = __le64_to_cpu(bsb.arraystart2);
4327 }
4328
4329 reshape_completed = sra->reshape_progress;
4330 rv = progress_reshape(sra, reshape,
4331 backup_point, wait_point,
4332 &suspend_point, &reshape_completed,
4333 &frozen);
4334 /* external metadata would need to ping_monitor here */
4335 sra->reshape_progress = reshape_completed;
4336
4337 /* Clear any backup region that is before 'here' */
4338 if (increasing) {
4339 if (__le64_to_cpu(bsb.length) > 0 &&
4340 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4341 __le64_to_cpu(bsb.length)))
4342 forget_backup(dests, destfd,
4343 destoffsets, 0);
4344 if (__le64_to_cpu(bsb.length2) > 0 &&
4345 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4346 __le64_to_cpu(bsb.length2)))
4347 forget_backup(dests, destfd,
4348 destoffsets, 1);
4349 } else {
4350 if (__le64_to_cpu(bsb.length) > 0 &&
4351 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4352 forget_backup(dests, destfd,
4353 destoffsets, 0);
4354 if (__le64_to_cpu(bsb.length2) > 0 &&
4355 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4356 forget_backup(dests, destfd,
4357 destoffsets, 1);
4358 }
4359 if (sigterm)
4360 rv = -2;
4361 if (rv < 0) {
4362 if (rv == -1)
4363 done = 1;
4364 break;
4365 }
4366 if (rv == 0 && increasing && !st->ss->external) {
4367 /* No longer need to monitor this reshape */
4368 sysfs_set_str(sra, NULL, "sync_max", "max");
4369 done = 1;
4370 break;
4371 }
4372
4373 while (rv) {
4374 unsigned long long offset;
4375 unsigned long actual_stripes;
4376 /* Need to backup some data.
4377 * If 'part' is not used and the desired
4378 * backup size is suspended, do a backup,
4379 * then consider the next part.
4380 */
4381 /* Check that 'part' is unused */
4382 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4383 break;
4384 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4385 break;
4386
4387 offset = backup_point / data;
4388 actual_stripes = stripes;
4389 if (increasing) {
4390 if (offset + actual_stripes * (chunk/512) >
4391 sra->component_size)
4392 actual_stripes = ((sra->component_size - offset)
4393 / (chunk/512));
4394 if (offset + actual_stripes * (chunk/512) >
4395 suspend_point/data)
4396 break;
4397 } else {
4398 if (offset < actual_stripes * (chunk/512))
4399 actual_stripes = offset / (chunk/512);
4400 offset -= actual_stripes * (chunk/512);
4401 if (offset < suspend_point/data)
4402 break;
4403 }
4404 if (actual_stripes == 0)
4405 break;
4406 grow_backup(sra, offset, actual_stripes,
4407 fds, offsets,
4408 disks, chunk, level, layout,
4409 dests, destfd, destoffsets,
4410 part, &degraded, buf);
4411 validate(afd, destfd[0], destoffsets[0]);
4412 /* record where 'part' is up to */
4413 part = !part;
4414 if (increasing)
4415 backup_point += actual_stripes * (chunk/512) * data;
4416 else
4417 backup_point -= actual_stripes * (chunk/512) * data;
4418 }
4419 }
4420
4421 /* FIXME maybe call progress_reshape one more time instead */
4422 /* remove any remaining suspension */
4423 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4424 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4425 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4426 sysfs_set_num(sra, NULL, "sync_min", 0);
4427
4428 if (reshape->before.data_disks == reshape->after.data_disks)
4429 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4430 free(buf);
4431 return done;
4432 }
4433
4434 /*
4435 * If any spare contains md_back_data-1 which is recent wrt mtime,
4436 * write that data into the array and update the super blocks with
4437 * the new reshape_progress
4438 */
4439 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4440 char *backup_file, int verbose)
4441 {
4442 int i, j;
4443 int old_disks;
4444 unsigned long long *offsets;
4445 unsigned long long nstripe, ostripe;
4446 int ndata, odata;
4447
4448 odata = info->array.raid_disks - info->delta_disks - 1;
4449 if (info->array.level == 6) odata--; /* number of data disks */
4450 ndata = info->array.raid_disks - 1;
4451 if (info->new_level == 6) ndata--;
4452
4453 old_disks = info->array.raid_disks - info->delta_disks;
4454
4455 if (info->delta_disks <= 0)
4456 /* Didn't grow, so the backup file must have
4457 * been used
4458 */
4459 old_disks = cnt;
4460 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4461 struct mdinfo dinfo;
4462 int fd;
4463 int bsbsize;
4464 char *devname, namebuf[20];
4465 unsigned long long lo, hi;
4466
4467 /* This was a spare and may have some saved data on it.
4468 * Load the superblock, find and load the
4469 * backup_super_block.
4470 * If either fail, go on to next device.
4471 * If the backup contains no new info, just return
4472 * else restore data and update all superblocks
4473 */
4474 if (i == old_disks-1) {
4475 fd = open(backup_file, O_RDONLY);
4476 if (fd<0) {
4477 pr_err("backup file %s inaccessible: %s\n",
4478 backup_file, strerror(errno));
4479 continue;
4480 }
4481 devname = backup_file;
4482 } else {
4483 fd = fdlist[i];
4484 if (fd < 0)
4485 continue;
4486 if (st->ss->load_super(st, fd, NULL))
4487 continue;
4488
4489 st->ss->getinfo_super(st, &dinfo, NULL);
4490 st->ss->free_super(st);
4491
4492 if (lseek64(fd,
4493 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4494 0) < 0) {
4495 pr_err("Cannot seek on device %d\n", i);
4496 continue; /* Cannot seek */
4497 }
4498 sprintf(namebuf, "device-%d", i);
4499 devname = namebuf;
4500 }
4501 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4502 if (verbose)
4503 pr_err("Cannot read from %s\n", devname);
4504 continue; /* Cannot read */
4505 }
4506 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4507 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4508 if (verbose)
4509 pr_err("No backup metadata on %s\n", devname);
4510 continue;
4511 }
4512 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4513 if (verbose)
4514 pr_err("Bad backup-metadata checksum on %s\n", devname);
4515 continue; /* bad checksum */
4516 }
4517 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4518 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4519 if (verbose)
4520 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4521 continue; /* Bad second checksum */
4522 }
4523 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4524 if (verbose)
4525 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4526 continue; /* Wrong uuid */
4527 }
4528
4529 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4530 * sometimes they aren't... So allow considerable flexability in matching, and allow
4531 * this test to be overridden by an environment variable.
4532 */
4533 if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) ||
4534 time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) {
4535 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4536 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4537 (unsigned long)__le64_to_cpu(bsb.mtime),
4538 (unsigned long)info->array.utime);
4539 } else {
4540 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4541 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4542 continue; /* time stamp is too bad */
4543 }
4544 }
4545
4546 if (bsb.magic[15] == '1') {
4547 if (bsb.length == 0)
4548 continue;
4549 if (info->delta_disks >= 0) {
4550 /* reshape_progress is increasing */
4551 if (__le64_to_cpu(bsb.arraystart)
4552 + __le64_to_cpu(bsb.length)
4553 < info->reshape_progress) {
4554 nonew:
4555 if (verbose)
4556 pr_err("backup-metadata found on %s but is not needed\n", devname);
4557 continue; /* No new data here */
4558 }
4559 } else {
4560 /* reshape_progress is decreasing */
4561 if (__le64_to_cpu(bsb.arraystart) >=
4562 info->reshape_progress)
4563 goto nonew; /* No new data here */
4564 }
4565 } else {
4566 if (bsb.length == 0 && bsb.length2 == 0)
4567 continue;
4568 if (info->delta_disks >= 0) {
4569 /* reshape_progress is increasing */
4570 if ((__le64_to_cpu(bsb.arraystart)
4571 + __le64_to_cpu(bsb.length)
4572 < info->reshape_progress)
4573 &&
4574 (__le64_to_cpu(bsb.arraystart2)
4575 + __le64_to_cpu(bsb.length2)
4576 < info->reshape_progress))
4577 goto nonew; /* No new data here */
4578 } else {
4579 /* reshape_progress is decreasing */
4580 if (__le64_to_cpu(bsb.arraystart) >=
4581 info->reshape_progress &&
4582 __le64_to_cpu(bsb.arraystart2) >=
4583 info->reshape_progress)
4584 goto nonew; /* No new data here */
4585 }
4586 }
4587 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4588 second_fail:
4589 if (verbose)
4590 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4591 devname);
4592 continue; /* Cannot seek */
4593 }
4594 /* There should be a duplicate backup superblock 4k before here */
4595 if (lseek64(fd, -4096, 1) < 0 ||
4596 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4597 goto second_fail; /* Cannot find leading superblock */
4598 if (bsb.magic[15] == '1')
4599 bsbsize = offsetof(struct mdp_backup_super, pad1);
4600 else
4601 bsbsize = offsetof(struct mdp_backup_super, pad);
4602 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4603 goto second_fail; /* Cannot find leading superblock */
4604
4605 /* Now need the data offsets for all devices. */
4606 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4607 for(j=0; j<info->array.raid_disks; j++) {
4608 if (fdlist[j] < 0)
4609 continue;
4610 if (st->ss->load_super(st, fdlist[j], NULL))
4611 /* FIXME should be this be an error */
4612 continue;
4613 st->ss->getinfo_super(st, &dinfo, NULL);
4614 st->ss->free_super(st);
4615 offsets[j] = dinfo.data_offset * 512;
4616 }
4617 printf("%s: restoring critical section\n", Name);
4618
4619 if (restore_stripes(fdlist, offsets,
4620 info->array.raid_disks,
4621 info->new_chunk,
4622 info->new_level,
4623 info->new_layout,
4624 fd, __le64_to_cpu(bsb.devstart)*512,
4625 __le64_to_cpu(bsb.arraystart)*512,
4626 __le64_to_cpu(bsb.length)*512, NULL)) {
4627 /* didn't succeed, so giveup */
4628 if (verbose)
4629 pr_err("Error restoring backup from %s\n",
4630 devname);
4631 free(offsets);
4632 return 1;
4633 }
4634
4635 if (bsb.magic[15] == '2' &&
4636 restore_stripes(fdlist, offsets,
4637 info->array.raid_disks,
4638 info->new_chunk,
4639 info->new_level,
4640 info->new_layout,
4641 fd, __le64_to_cpu(bsb.devstart)*512 +
4642 __le64_to_cpu(bsb.devstart2)*512,
4643 __le64_to_cpu(bsb.arraystart2)*512,
4644 __le64_to_cpu(bsb.length2)*512, NULL)) {
4645 /* didn't succeed, so giveup */
4646 if (verbose)
4647 pr_err("Error restoring second backup from %s\n",
4648 devname);
4649 free(offsets);
4650 return 1;
4651 }
4652
4653 free(offsets);
4654
4655 /* Ok, so the data is restored. Let's update those superblocks. */
4656
4657 lo = hi = 0;
4658 if (bsb.length) {
4659 lo = __le64_to_cpu(bsb.arraystart);
4660 hi = lo + __le64_to_cpu(bsb.length);
4661 }
4662 if (bsb.magic[15] == '2' && bsb.length2) {
4663 unsigned long long lo1, hi1;
4664 lo1 = __le64_to_cpu(bsb.arraystart2);
4665 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4666 if (lo == hi) {
4667 lo = lo1;
4668 hi = hi1;
4669 } else if (lo < lo1)
4670 hi = hi1;
4671 else
4672 lo = lo1;
4673 }
4674 if (lo < hi &&
4675 (info->reshape_progress < lo ||
4676 info->reshape_progress > hi))
4677 /* backup does not affect reshape_progress*/ ;
4678 else if (info->delta_disks >= 0) {
4679 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4680 __le64_to_cpu(bsb.length);
4681 if (bsb.magic[15] == '2') {
4682 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4683 __le64_to_cpu(bsb.length2);
4684 if (p2 > info->reshape_progress)
4685 info->reshape_progress = p2;
4686 }
4687 } else {
4688 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4689 if (bsb.magic[15] == '2') {
4690 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4691 if (p2 < info->reshape_progress)
4692 info->reshape_progress = p2;
4693 }
4694 }
4695 for (j=0; j<info->array.raid_disks; j++) {
4696 if (fdlist[j] < 0)
4697 continue;
4698 if (st->ss->load_super(st, fdlist[j], NULL))
4699 continue;
4700 st->ss->getinfo_super(st, &dinfo, NULL);
4701 dinfo.reshape_progress = info->reshape_progress;
4702 st->ss->update_super(st, &dinfo,
4703 "_reshape_progress",
4704 NULL,0, 0, NULL);
4705 st->ss->store_super(st, fdlist[j]);
4706 st->ss->free_super(st);
4707 }
4708 return 0;
4709 }
4710 /* Didn't find any backup data, try to see if any
4711 * was needed.
4712 */
4713 if (info->delta_disks < 0) {
4714 /* When shrinking, the critical section is at the end.
4715 * So see if we are before the critical section.
4716 */
4717 unsigned long long first_block;
4718 nstripe = ostripe = 0;
4719 first_block = 0;
4720 while (ostripe >= nstripe) {
4721 ostripe += info->array.chunk_size / 512;
4722 first_block = ostripe * odata;
4723 nstripe = first_block / ndata / (info->new_chunk/512) *
4724 (info->new_chunk/512);
4725 }
4726
4727 if (info->reshape_progress >= first_block)
4728 return 0;
4729 }
4730 if (info->delta_disks > 0) {
4731 /* See if we are beyond the critical section. */
4732 unsigned long long last_block;
4733 nstripe = ostripe = 0;
4734 last_block = 0;
4735 while (nstripe >= ostripe) {
4736 nstripe += info->new_chunk / 512;
4737 last_block = nstripe * ndata;
4738 ostripe = last_block / odata / (info->array.chunk_size/512) *
4739 (info->array.chunk_size/512);
4740 }
4741
4742 if (info->reshape_progress >= last_block)
4743 return 0;
4744 }
4745 /* needed to recover critical section! */
4746 if (verbose)
4747 pr_err("Failed to find backup of critical section\n");
4748 return 1;
4749 }
4750
4751 int Grow_continue_command(char *devname, int fd,
4752 char *backup_file, int verbose)
4753 {
4754 int ret_val = 0;
4755 struct supertype *st = NULL;
4756 struct mdinfo *content = NULL;
4757 struct mdinfo array;
4758 char *subarray = NULL;
4759 struct mdinfo *cc = NULL;
4760 struct mdstat_ent *mdstat = NULL;
4761 int cfd = -1;
4762 int fd2;
4763
4764 dprintf("Grow continue from command line called for %s\n",
4765 devname);
4766
4767 st = super_by_fd(fd, &subarray);
4768 if (!st || !st->ss) {
4769 pr_err("Unable to determine metadata format for %s\n",
4770 devname);
4771 return 1;
4772 }
4773 dprintf("Grow continue is run for ");
4774 if (st->ss->external == 0) {
4775 int d;
4776 dprintf_cont("native array (%s)\n", devname);
4777 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4778 pr_err("%s is not an active md array - aborting\n", devname);
4779 ret_val = 1;
4780 goto Grow_continue_command_exit;
4781 }
4782 content = &array;
4783 /* Need to load a superblock.
4784 * FIXME we should really get what we need from
4785 * sysfs
4786 */
4787 for (d = 0; d < MAX_DISKS; d++) {
4788 mdu_disk_info_t disk;
4789 char *dv;
4790 int err;
4791 disk.number = d;
4792 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4793 continue;
4794 if (disk.major == 0 && disk.minor == 0)
4795 continue;
4796 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4797 continue;
4798 dv = map_dev(disk.major, disk.minor, 1);
4799 if (!dv)
4800 continue;
4801 fd2 = dev_open(dv, O_RDONLY);
4802 if (fd2 < 0)
4803 continue;
4804 err = st->ss->load_super(st, fd2, NULL);
4805 close(fd2);
4806 if (err)
4807 continue;
4808 break;
4809 }
4810 if (d == MAX_DISKS) {
4811 pr_err("Unable to load metadata for %s\n",
4812 devname);
4813 ret_val = 1;
4814 goto Grow_continue_command_exit;
4815 }
4816 st->ss->getinfo_super(st, content, NULL);
4817 } else {
4818 char *container;
4819
4820 if (subarray) {
4821 dprintf_cont("subarray (%s)\n", subarray);
4822 container = st->container_devnm;
4823 cfd = open_dev_excl(st->container_devnm);
4824 } else {
4825 container = st->devnm;
4826 close(fd);
4827 cfd = open_dev_excl(st->devnm);
4828 dprintf_cont("container (%s)\n", container);
4829 fd = cfd;
4830 }
4831 if (cfd < 0) {
4832 pr_err("Unable to open container for %s\n", devname);
4833 ret_val = 1;
4834 goto Grow_continue_command_exit;
4835 }
4836
4837 /* find in container array under reshape
4838 */
4839 ret_val = st->ss->load_container(st, cfd, NULL);
4840 if (ret_val) {
4841 pr_err("Cannot read superblock for %s\n",
4842 devname);
4843 ret_val = 1;
4844 goto Grow_continue_command_exit;
4845 }
4846
4847 cc = st->ss->container_content(st, subarray);
4848 for (content = cc; content ; content = content->next) {
4849 char *array;
4850 int allow_reshape = 1;
4851
4852 if (content->reshape_active == 0)
4853 continue;
4854 /* The decision about array or container wide
4855 * reshape is taken in Grow_continue based
4856 * content->reshape_active state, therefore we
4857 * need to check_reshape based on
4858 * reshape_active and subarray name
4859 */
4860 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4861 allow_reshape = 0;
4862 if (content->reshape_active == CONTAINER_RESHAPE &&
4863 (content->array.state
4864 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4865 allow_reshape = 0;
4866
4867 if (!allow_reshape) {
4868 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4869 devname, container);
4870 ret_val = 1;
4871 goto Grow_continue_command_exit;
4872 }
4873
4874 array = strchr(content->text_version+1, '/')+1;
4875 mdstat = mdstat_by_subdev(array, container);
4876 if (!mdstat)
4877 continue;
4878 if (mdstat->active == 0) {
4879 pr_err("Skipping inactive array %s.\n",
4880 mdstat->devnm);
4881 free_mdstat(mdstat);
4882 mdstat = NULL;
4883 continue;
4884 }
4885 break;
4886 }
4887 if (!content) {
4888 pr_err("Unable to determine reshaped array for %s\n", devname);
4889 ret_val = 1;
4890 goto Grow_continue_command_exit;
4891 }
4892 fd2 = open_dev(mdstat->devnm);
4893 if (fd2 < 0) {
4894 pr_err("cannot open (%s)\n", mdstat->devnm);
4895 ret_val = 1;
4896 goto Grow_continue_command_exit;
4897 }
4898
4899 sysfs_init(content, fd2, mdstat->devnm);
4900
4901 close(fd2);
4902
4903 /* start mdmon in case it is not running
4904 */
4905 if (!mdmon_running(container))
4906 start_mdmon(container);
4907 ping_monitor(container);
4908
4909 if (mdmon_running(container))
4910 st->update_tail = &st->updates;
4911 else {
4912 pr_err("No mdmon found. Grow cannot continue.\n");
4913 ret_val = 1;
4914 goto Grow_continue_command_exit;
4915 }
4916 }
4917
4918 /* verify that array under reshape is started from
4919 * correct position
4920 */
4921 if (verify_reshape_position(content, content->array.level) < 0) {
4922 ret_val = 1;
4923 goto Grow_continue_command_exit;
4924 }
4925
4926 /* continue reshape
4927 */
4928 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4929
4930 Grow_continue_command_exit:
4931 if (cfd > -1)
4932 close(cfd);
4933 st->ss->free_super(st);
4934 free_mdstat(mdstat);
4935 sysfs_free(cc);
4936 free(subarray);
4937
4938 return ret_val;
4939 }
4940
4941 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4942 char *backup_file, int forked, int freeze_reshape)
4943 {
4944 int ret_val = 2;
4945
4946 if (!info->reshape_active)
4947 return ret_val;
4948
4949 if (st->ss->external) {
4950 int cfd = open_dev(st->container_devnm);
4951
4952 if (cfd < 0)
4953 return 1;
4954
4955 st->ss->load_container(st, cfd, st->container_devnm);
4956 close(cfd);
4957 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4958 st, info, 0, backup_file,
4959 0, forked,
4960 1 | info->reshape_active,
4961 freeze_reshape);
4962 } else
4963 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4964 NULL, INVALID_SECTORS,
4965 backup_file, 0, forked,
4966 1 | info->reshape_active,
4967 freeze_reshape);
4968
4969 return ret_val;
4970 }
4971
4972 char *make_backup(char *name)
4973 {
4974 char *base = "backup_file-";
4975 int len;
4976 char *fname;
4977
4978 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4979 fname = xmalloc(len);
4980 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4981 return fname;
4982 }
4983
4984 char *locate_backup(char *name)
4985 {
4986 char *fl = make_backup(name);
4987 struct stat stb;
4988
4989 if (stat(fl, &stb) == 0 &&
4990 S_ISREG(stb.st_mode))
4991 return fl;
4992
4993 free(fl);
4994 return NULL;
4995 }
Unnamed repository; edit this file 'description' to name the repository.