]> 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
mdadm: add the ability to change cluster name
[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 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
301 if (errno == ENOMEM)
302 pr_err("Memory allocation failure.\n");
303 else
304 pr_err("bitmaps not supported by this kernel.\n");
305 return 1;
306 }
307 if (bmf.pathname[0]) {
308 if (strcmp(s->bitmap_file,"none")==0) {
309 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
310 pr_err("failed to remove bitmap %s\n",
311 bmf.pathname);
312 return 1;
313 }
314 return 0;
315 }
316 pr_err("%s already has a bitmap (%s)\n",
317 devname, bmf.pathname);
318 return 1;
319 }
320 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
321 pr_err("cannot get array status for %s\n", devname);
322 return 1;
323 }
324 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
325 if (strcmp(s->bitmap_file, "none")==0) {
326 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
327 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
328 pr_err("failed to remove internal bitmap.\n");
329 return 1;
330 }
331 return 0;
332 }
333 pr_err("%s bitmap already present on %s\n", s->bitmap_file, devname);
334 return 1;
335 }
336
337 if (strcmp(s->bitmap_file, "none") == 0) {
338 pr_err("no bitmap found on %s\n", devname);
339 return 1;
340 }
341 if (array.level <= 0) {
342 pr_err("Bitmaps not meaningful with level %s\n",
343 map_num(pers, array.level)?:"of this array");
344 return 1;
345 }
346 bitmapsize = array.size;
347 bitmapsize <<= 1;
348 if (get_dev_size(fd, NULL, &array_size) &&
349 array_size > (0x7fffffffULL<<9)) {
350 /* Array is big enough that we cannot trust array.size
351 * try other approaches
352 */
353 bitmapsize = get_component_size(fd);
354 }
355 if (bitmapsize == 0) {
356 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
357 return 1;
358 }
359
360 if (array.level == 10) {
361 int ncopies = (array.layout&255)*((array.layout>>8)&255);
362 bitmapsize = bitmapsize * array.raid_disks / ncopies;
363 }
364
365 st = super_by_fd(fd, &subarray);
366 if (!st) {
367 pr_err("Cannot understand version %d.%d\n",
368 array.major_version, array.minor_version);
369 return 1;
370 }
371 if (subarray) {
372 pr_err("Cannot add bitmaps to sub-arrays yet\n");
373 free(subarray);
374 free(st);
375 return 1;
376 }
377 if (strcmp(s->bitmap_file, "internal") == 0 ||
378 strcmp(s->bitmap_file, "clustered") == 0) {
379 int rv;
380 int d;
381 int offset_setable = 0;
382 struct mdinfo *mdi;
383 if (st->ss->add_internal_bitmap == NULL) {
384 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
385 return 1;
386 }
387 st->nodes = c->nodes;
388 st->cluster_name = c->homecluster;
389 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
390 if (mdi)
391 offset_setable = 1;
392 for (d=0; d< st->max_devs; d++) {
393 mdu_disk_info_t disk;
394 char *dv;
395 disk.number = d;
396 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
397 continue;
398 if (disk.major == 0 &&
399 disk.minor == 0)
400 continue;
401 if ((disk.state & (1<<MD_DISK_SYNC))==0)
402 continue;
403 dv = map_dev(disk.major, disk.minor, 1);
404 if (dv) {
405 int fd2 = dev_open(dv, O_RDWR);
406 if (fd2 < 0)
407 continue;
408 if (st->ss->load_super(st, fd2, NULL)==0) {
409 if (st->ss->add_internal_bitmap(
410 st,
411 &s->bitmap_chunk, c->delay, s->write_behind,
412 bitmapsize, offset_setable,
413 major)
414 )
415 st->ss->write_bitmap(st, fd2, NoUpdate);
416 else {
417 pr_err("failed to create internal bitmap - chunksize problem.\n");
418 close(fd2);
419 return 1;
420 }
421 }
422 close(fd2);
423 }
424 }
425 if (offset_setable) {
426 st->ss->getinfo_super(st, mdi, NULL);
427 sysfs_init(mdi, fd, NULL);
428 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
429 mdi->bitmap_offset);
430 } else {
431 if (strcmp(s->bitmap_file, "clustered") == 0)
432 array.state |= (1<<MD_SB_CLUSTERED);
433 array.state |= (1<<MD_SB_BITMAP_PRESENT);
434 rv = ioctl(fd, SET_ARRAY_INFO, &array);
435 }
436 if (rv < 0) {
437 if (errno == EBUSY)
438 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
439 pr_err("failed to set internal bitmap.\n");
440 return 1;
441 }
442 } else {
443 int uuid[4];
444 int bitmap_fd;
445 int d;
446 int max_devs = st->max_devs;
447
448 /* try to load a superblock */
449 for (d = 0; d < max_devs; d++) {
450 mdu_disk_info_t disk;
451 char *dv;
452 int fd2;
453 disk.number = d;
454 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
455 continue;
456 if ((disk.major==0 && disk.minor==0) ||
457 (disk.state & (1<<MD_DISK_REMOVED)))
458 continue;
459 dv = map_dev(disk.major, disk.minor, 1);
460 if (!dv)
461 continue;
462 fd2 = dev_open(dv, O_RDONLY);
463 if (fd2 >= 0) {
464 if (st->ss->load_super(st, fd2, NULL) == 0) {
465 close(fd2);
466 st->ss->uuid_from_super(st, uuid);
467 break;
468 }
469 close(fd2);
470 }
471 }
472 if (d == max_devs) {
473 pr_err("cannot find UUID for array!\n");
474 return 1;
475 }
476 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
477 c->delay, s->write_behind, bitmapsize, major)) {
478 return 1;
479 }
480 bitmap_fd = open(s->bitmap_file, O_RDWR);
481 if (bitmap_fd < 0) {
482 pr_err("weird: %s cannot be opened\n",
483 s->bitmap_file);
484 return 1;
485 }
486 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
487 int err = errno;
488 if (errno == EBUSY)
489 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
490 pr_err("Cannot set bitmap file for %s: %s\n",
491 devname, strerror(err));
492 return 1;
493 }
494 }
495
496 return 0;
497 }
498
499 /*
500 * When reshaping an array we might need to backup some data.
501 * This is written to all spares with a 'super_block' describing it.
502 * The superblock goes 4K from the end of the used space on the
503 * device.
504 * It if written after the backup is complete.
505 * It has the following structure.
506 */
507
508 static struct mdp_backup_super {
509 char magic[16]; /* md_backup_data-1 or -2 */
510 __u8 set_uuid[16];
511 __u64 mtime;
512 /* start/sizes in 512byte sectors */
513 __u64 devstart; /* address on backup device/file of data */
514 __u64 arraystart;
515 __u64 length;
516 __u32 sb_csum; /* csum of preceeding bytes. */
517 __u32 pad1;
518 __u64 devstart2; /* offset in to data of second section */
519 __u64 arraystart2;
520 __u64 length2;
521 __u32 sb_csum2; /* csum of preceeding bytes. */
522 __u8 pad[512-68-32];
523 } __attribute__((aligned(512))) bsb, bsb2;
524
525 static __u32 bsb_csum(char *buf, int len)
526 {
527 int i;
528 int csum = 0;
529 for (i = 0; i < len; i++)
530 csum = (csum<<3) + buf[0];
531 return __cpu_to_le32(csum);
532 }
533
534 static int check_idle(struct supertype *st)
535 {
536 /* Check that all member arrays for this container, or the
537 * container of this array, are idle
538 */
539 char *container = (st->container_devnm[0]
540 ? st->container_devnm : st->devnm);
541 struct mdstat_ent *ent, *e;
542 int is_idle = 1;
543
544 ent = mdstat_read(0, 0);
545 for (e = ent ; e; e = e->next) {
546 if (!is_container_member(e, container))
547 continue;
548 if (e->percent >= 0) {
549 is_idle = 0;
550 break;
551 }
552 }
553 free_mdstat(ent);
554 return is_idle;
555 }
556
557 static int freeze_container(struct supertype *st)
558 {
559 char *container = (st->container_devnm[0]
560 ? st->container_devnm : st->devnm);
561
562 if (!check_idle(st))
563 return -1;
564
565 if (block_monitor(container, 1)) {
566 pr_err("failed to freeze container\n");
567 return -2;
568 }
569
570 return 1;
571 }
572
573 static void unfreeze_container(struct supertype *st)
574 {
575 char *container = (st->container_devnm[0]
576 ? st->container_devnm : st->devnm);
577
578 unblock_monitor(container, 1);
579 }
580
581 static int freeze(struct supertype *st)
582 {
583 /* Try to freeze resync/rebuild on this array/container.
584 * Return -1 if the array is busy,
585 * return -2 container cannot be frozen,
586 * return 0 if this kernel doesn't support 'frozen'
587 * return 1 if it worked.
588 */
589 if (st->ss->external)
590 return freeze_container(st);
591 else {
592 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
593 int err;
594 char buf[20];
595
596 if (!sra)
597 return -1;
598 /* Need to clear any 'read-auto' status */
599 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
600 strncmp(buf, "read-auto", 9) == 0)
601 sysfs_set_str(sra, NULL, "array_state", "clean");
602
603 err = sysfs_freeze_array(sra);
604 sysfs_free(sra);
605 return err;
606 }
607 }
608
609 static void unfreeze(struct supertype *st)
610 {
611 if (st->ss->external)
612 return unfreeze_container(st);
613 else {
614 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
615 char buf[20];
616
617 if (sra &&
618 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
619 && strcmp(buf, "frozen\n") == 0) {
620 printf("unfreeze\n");
621 sysfs_set_str(sra, NULL, "sync_action", "idle");
622 }
623 sysfs_free(sra);
624 }
625 }
626
627 static void wait_reshape(struct mdinfo *sra)
628 {
629 int fd = sysfs_get_fd(sra, NULL, "sync_action");
630 char action[20];
631
632 if (fd < 0)
633 return;
634
635 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
636 strncmp(action, "reshape", 7) == 0)
637 sysfs_wait(fd, NULL);
638 close(fd);
639 }
640
641 static int reshape_super(struct supertype *st, unsigned long long size,
642 int level, int layout, int chunksize, int raid_disks,
643 int delta_disks, char *backup_file, char *dev,
644 int direction, int verbose)
645 {
646 /* nothing extra to check in the native case */
647 if (!st->ss->external)
648 return 0;
649 if (!st->ss->reshape_super ||
650 !st->ss->manage_reshape) {
651 pr_err("%s metadata does not support reshape\n",
652 st->ss->name);
653 return 1;
654 }
655
656 return st->ss->reshape_super(st, size, level, layout, chunksize,
657 raid_disks, delta_disks, backup_file, dev,
658 direction, verbose);
659 }
660
661 static void sync_metadata(struct supertype *st)
662 {
663 if (st->ss->external) {
664 if (st->update_tail) {
665 flush_metadata_updates(st);
666 st->update_tail = &st->updates;
667 } else
668 st->ss->sync_metadata(st);
669 }
670 }
671
672 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
673 {
674 /* when dealing with external metadata subarrays we need to be
675 * prepared to handle EAGAIN. The kernel may need to wait for
676 * mdmon to mark the array active so the kernel can handle
677 * allocations/writeback when preparing the reshape action
678 * (md_allow_write()). We temporarily disable safe_mode_delay
679 * to close a race with the array_state going clean before the
680 * next write to raid_disks / stripe_cache_size
681 */
682 char safe[50];
683 int rc;
684
685 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
686 if (!container ||
687 (strcmp(name, "raid_disks") != 0 &&
688 strcmp(name, "stripe_cache_size") != 0))
689 return sysfs_set_num(sra, NULL, name, n);
690
691 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
692 if (rc <= 0)
693 return -1;
694 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
695 rc = sysfs_set_num(sra, NULL, name, n);
696 if (rc < 0 && errno == EAGAIN) {
697 ping_monitor(container);
698 /* if we get EAGAIN here then the monitor is not active
699 * so stop trying
700 */
701 rc = sysfs_set_num(sra, NULL, name, n);
702 }
703 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
704 return rc;
705 }
706
707 int start_reshape(struct mdinfo *sra, int already_running,
708 int before_data_disks, int data_disks)
709 {
710 int err;
711 unsigned long long sync_max_to_set;
712
713 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
714 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
715 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
716 sra->reshape_progress);
717 if (before_data_disks <= data_disks)
718 sync_max_to_set = sra->reshape_progress / data_disks;
719 else
720 sync_max_to_set = (sra->component_size * data_disks
721 - sra->reshape_progress) / data_disks;
722 if (!already_running)
723 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
724 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
725 if (!already_running && err == 0) {
726 int cnt = 5;
727 do {
728 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
729 if (err)
730 sleep(1);
731 } while (err && errno == EBUSY && cnt-- > 0);
732 }
733 return err;
734 }
735
736 void abort_reshape(struct mdinfo *sra)
737 {
738 sysfs_set_str(sra, NULL, "sync_action", "idle");
739 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
740 sysfs_set_num(sra, NULL, "suspend_hi", 0);
741 sysfs_set_num(sra, NULL, "suspend_lo", 0);
742 sysfs_set_num(sra, NULL, "sync_min", 0);
743 // It isn't safe to reset sync_max as we aren't monitoring.
744 // Array really should be stopped at this point.
745 }
746
747 int remove_disks_for_takeover(struct supertype *st,
748 struct mdinfo *sra,
749 int layout)
750 {
751 int nr_of_copies;
752 struct mdinfo *remaining;
753 int slot;
754
755 if (sra->array.level == 10)
756 nr_of_copies = layout & 0xff;
757 else if (sra->array.level == 1)
758 nr_of_copies = sra->array.raid_disks;
759 else
760 return 1;
761
762 remaining = sra->devs;
763 sra->devs = NULL;
764 /* for each 'copy', select one device and remove from the list. */
765 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
766 struct mdinfo **diskp;
767 int found = 0;
768
769 /* Find a working device to keep */
770 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
771 struct mdinfo *disk = *diskp;
772
773 if (disk->disk.raid_disk < slot)
774 continue;
775 if (disk->disk.raid_disk >= slot + nr_of_copies)
776 continue;
777 if (disk->disk.state & (1<<MD_DISK_REMOVED))
778 continue;
779 if (disk->disk.state & (1<<MD_DISK_FAULTY))
780 continue;
781 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
782 continue;
783
784 /* We have found a good disk to use! */
785 *diskp = disk->next;
786 disk->next = sra->devs;
787 sra->devs = disk;
788 found = 1;
789 break;
790 }
791 if (!found)
792 break;
793 }
794
795 if (slot < sra->array.raid_disks) {
796 /* didn't find all slots */
797 struct mdinfo **e;
798 e = &remaining;
799 while (*e)
800 e = &(*e)->next;
801 *e = sra->devs;
802 sra->devs = remaining;
803 return 1;
804 }
805
806 /* Remove all 'remaining' devices from the array */
807 while (remaining) {
808 struct mdinfo *sd = remaining;
809 remaining = sd->next;
810
811 sysfs_set_str(sra, sd, "state", "faulty");
812 sysfs_set_str(sra, sd, "slot", "none");
813 /* for external metadata disks should be removed in mdmon */
814 if (!st->ss->external)
815 sysfs_set_str(sra, sd, "state", "remove");
816 sd->disk.state |= (1<<MD_DISK_REMOVED);
817 sd->disk.state &= ~(1<<MD_DISK_SYNC);
818 sd->next = sra->devs;
819 sra->devs = sd;
820 }
821 return 0;
822 }
823
824 void reshape_free_fdlist(int *fdlist,
825 unsigned long long *offsets,
826 int size)
827 {
828 int i;
829
830 for (i = 0; i < size; i++)
831 if (fdlist[i] >= 0)
832 close(fdlist[i]);
833
834 free(fdlist);
835 free(offsets);
836 }
837
838 int reshape_prepare_fdlist(char *devname,
839 struct mdinfo *sra,
840 int raid_disks,
841 int nrdisks,
842 unsigned long blocks,
843 char *backup_file,
844 int *fdlist,
845 unsigned long long *offsets)
846 {
847 int d = 0;
848 struct mdinfo *sd;
849
850 enable_fds(nrdisks);
851 for (d = 0; d <= nrdisks; d++)
852 fdlist[d] = -1;
853 d = raid_disks;
854 for (sd = sra->devs; sd; sd = sd->next) {
855 if (sd->disk.state & (1<<MD_DISK_FAULTY))
856 continue;
857 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
858 sd->disk.raid_disk < raid_disks) {
859 char *dn = map_dev(sd->disk.major,
860 sd->disk.minor, 1);
861 fdlist[sd->disk.raid_disk]
862 = dev_open(dn, O_RDONLY);
863 offsets[sd->disk.raid_disk] = sd->data_offset*512;
864 if (fdlist[sd->disk.raid_disk] < 0) {
865 pr_err("%s: cannot open component %s\n",
866 devname, dn ? dn : "-unknown-");
867 d = -1;
868 goto release;
869 }
870 } else if (backup_file == NULL) {
871 /* spare */
872 char *dn = map_dev(sd->disk.major,
873 sd->disk.minor, 1);
874 fdlist[d] = dev_open(dn, O_RDWR);
875 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
876 if (fdlist[d] < 0) {
877 pr_err("%s: cannot open component %s\n",
878 devname, dn ? dn : "-unknown-");
879 d = -1;
880 goto release;
881 }
882 d++;
883 }
884 }
885 release:
886 return d;
887 }
888
889 int reshape_open_backup_file(char *backup_file,
890 int fd,
891 char *devname,
892 long blocks,
893 int *fdlist,
894 unsigned long long *offsets,
895 char *sys_name,
896 int restart)
897 {
898 /* Return 1 on success, 0 on any form of failure */
899 /* need to check backup file is large enough */
900 char buf[512];
901 struct stat stb;
902 unsigned int dev;
903 int i;
904
905 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
906 S_IRUSR | S_IWUSR);
907 *offsets = 8 * 512;
908 if (*fdlist < 0) {
909 pr_err("%s: cannot create backup file %s: %s\n",
910 devname, backup_file, strerror(errno));
911 return 0;
912 }
913 /* Guard against backup file being on array device.
914 * If array is partitioned or if LVM etc is in the
915 * way this will not notice, but it is better than
916 * nothing.
917 */
918 fstat(*fdlist, &stb);
919 dev = stb.st_dev;
920 fstat(fd, &stb);
921 if (stb.st_rdev == dev) {
922 pr_err("backup file must NOT be on the array being reshaped.\n");
923 close(*fdlist);
924 return 0;
925 }
926
927 memset(buf, 0, 512);
928 for (i=0; i < blocks + 8 ; i++) {
929 if (write(*fdlist, buf, 512) != 512) {
930 pr_err("%s: cannot create backup file %s: %s\n",
931 devname, backup_file, strerror(errno));
932 return 0;
933 }
934 }
935 if (fsync(*fdlist) != 0) {
936 pr_err("%s: cannot create backup file %s: %s\n",
937 devname, backup_file, strerror(errno));
938 return 0;
939 }
940
941 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
942 char *bu = make_backup(sys_name);
943 if (symlink(backup_file, bu))
944 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
945 strerror(errno));
946 free(bu);
947 }
948
949 return 1;
950 }
951
952 unsigned long compute_backup_blocks(int nchunk, int ochunk,
953 unsigned int ndata, unsigned int odata)
954 {
955 unsigned long a, b, blocks;
956 /* So how much do we need to backup.
957 * We need an amount of data which is both a whole number of
958 * old stripes and a whole number of new stripes.
959 * So LCM for (chunksize*datadisks).
960 */
961 a = (ochunk/512) * odata;
962 b = (nchunk/512) * ndata;
963 /* Find GCD */
964 a = GCD(a, b);
965 /* LCM == product / GCD */
966 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
967
968 return blocks;
969 }
970
971 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
972 {
973 /* Based on the current array state in info->array and
974 * the changes in info->new_* etc, determine:
975 * - whether the change is possible
976 * - Intermediate level/raid_disks/layout
977 * - whether a restriping reshape is needed
978 * - number of sectors in minimum change unit. This
979 * will cover a whole number of stripes in 'before' and
980 * 'after'.
981 *
982 * Return message if the change should be rejected
983 * NULL if the change can be achieved
984 *
985 * This can be called as part of starting a reshape, or
986 * when assembling an array that is undergoing reshape.
987 */
988 int near, far, offset, copies;
989 int new_disks;
990 int old_chunk, new_chunk;
991 /* delta_parity records change in number of devices
992 * caused by level change
993 */
994 int delta_parity = 0;
995
996 memset(re, 0, sizeof(*re));
997
998 /* If a new level not explicitly given, we assume no-change */
999 if (info->new_level == UnSet)
1000 info->new_level = info->array.level;
1001
1002 if (info->new_chunk)
1003 switch (info->new_level) {
1004 case 0:
1005 case 4:
1006 case 5:
1007 case 6:
1008 case 10:
1009 /* chunk size is meaningful, must divide component_size
1010 * evenly
1011 */
1012 if (info->component_size % (info->new_chunk/512)) {
1013 unsigned long long shrink = info->component_size;
1014 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1015 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1016 info->new_chunk/1024, info->component_size/2);
1017 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1018 devname, shrink/2);
1019 pr_err("will shrink the array so the given chunk size would work.\n");
1020 return "";
1021 }
1022 break;
1023 default:
1024 return "chunk size not meaningful for this level";
1025 }
1026 else
1027 info->new_chunk = info->array.chunk_size;
1028
1029 switch (info->array.level) {
1030 default:
1031 return "No reshape is possibly for this RAID level";
1032 case LEVEL_LINEAR:
1033 if (info->delta_disks != UnSet)
1034 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1035 else
1036 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1037 case 1:
1038 /* RAID1 can convert to RAID1 with different disks, or
1039 * raid5 with 2 disks, or
1040 * raid0 with 1 disk
1041 */
1042 if (info->new_level > 1 &&
1043 (info->component_size & 7))
1044 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1045 if (info->new_level == 0) {
1046 if (info->delta_disks != UnSet &&
1047 info->delta_disks != 0)
1048 return "Cannot change number of disks with RAID1->RAID0 conversion";
1049 re->level = 0;
1050 re->before.data_disks = 1;
1051 re->after.data_disks = 1;
1052 return NULL;
1053 }
1054 if (info->new_level == 1) {
1055 if (info->delta_disks == UnSet)
1056 /* Don't know what to do */
1057 return "no change requested for Growing RAID1";
1058 re->level = 1;
1059 return NULL;
1060 }
1061 if (info->array.raid_disks == 2 &&
1062 info->new_level == 5) {
1063
1064 re->level = 5;
1065 re->before.data_disks = 1;
1066 if (info->delta_disks != UnSet &&
1067 info->delta_disks != 0)
1068 re->after.data_disks = 1 + info->delta_disks;
1069 else
1070 re->after.data_disks = 1;
1071 if (re->after.data_disks < 1)
1072 return "Number of disks too small for RAID5";
1073
1074 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1075 info->array.chunk_size = 65536;
1076 break;
1077 }
1078 /* Could do some multi-stage conversions, but leave that to
1079 * later.
1080 */
1081 return "Impossibly level change request for RAID1";
1082
1083 case 10:
1084 /* RAID10 can be converted from near mode to
1085 * RAID0 by removing some devices.
1086 * It can also be reshaped if the kernel supports
1087 * new_data_offset.
1088 */
1089 switch (info->new_level) {
1090 case 0:
1091 if ((info->array.layout & ~0xff) != 0x100)
1092 return "Cannot Grow RAID10 with far/offset layout";
1093 /* number of devices must be multiple of number of copies */
1094 if (info->array.raid_disks % (info->array.layout & 0xff))
1095 return "RAID10 layout too complex for Grow operation";
1096
1097 new_disks = (info->array.raid_disks
1098 / (info->array.layout & 0xff));
1099 if (info->delta_disks == UnSet)
1100 info->delta_disks = (new_disks
1101 - info->array.raid_disks);
1102
1103 if (info->delta_disks != new_disks - info->array.raid_disks)
1104 return "New number of raid-devices impossible for RAID10";
1105 if (info->new_chunk &&
1106 info->new_chunk != info->array.chunk_size)
1107 return "Cannot change chunk-size with RAID10 Grow";
1108
1109 /* looks good */
1110 re->level = 0;
1111 re->before.data_disks = new_disks;
1112 re->after.data_disks = re->before.data_disks;
1113 return NULL;
1114
1115 case 10:
1116 near = info->array.layout & 0xff;
1117 far = (info->array.layout >> 8) & 0xff;
1118 offset = info->array.layout & 0x10000;
1119 if (far > 1 && !offset)
1120 return "Cannot reshape RAID10 in far-mode";
1121 copies = near * far;
1122
1123 old_chunk = info->array.chunk_size * far;
1124
1125 if (info->new_layout == UnSet)
1126 info->new_layout = info->array.layout;
1127 else {
1128 near = info->new_layout & 0xff;
1129 far = (info->new_layout >> 8) & 0xff;
1130 offset = info->new_layout & 0x10000;
1131 if (far > 1 && !offset)
1132 return "Cannot reshape RAID10 to far-mode";
1133 if (near * far != copies)
1134 return "Cannot change number of copies when reshaping RAID10";
1135 }
1136 if (info->delta_disks == UnSet)
1137 info->delta_disks = 0;
1138 new_disks = (info->array.raid_disks +
1139 info->delta_disks);
1140
1141 new_chunk = info->new_chunk * far;
1142
1143 re->level = 10;
1144 re->before.layout = info->array.layout;
1145 re->before.data_disks = info->array.raid_disks;
1146 re->after.layout = info->new_layout;
1147 re->after.data_disks = new_disks;
1148 /* For RAID10 we don't do backup but do allow reshape,
1149 * so set backup_blocks to INVALID_SECTORS rather than
1150 * zero.
1151 * And there is no need to synchronise stripes on both
1152 * 'old' and 'new'. So the important
1153 * number is the minimum data_offset difference
1154 * which is the larger of (offset copies * chunk).
1155 */
1156 re->backup_blocks = INVALID_SECTORS;
1157 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1158 if (new_disks < re->before.data_disks &&
1159 info->space_after < re->min_offset_change)
1160 /* Reduce component size by one chunk */
1161 re->new_size = (info->component_size -
1162 re->min_offset_change);
1163 else
1164 re->new_size = info->component_size;
1165 re->new_size = re->new_size * new_disks / copies;
1166 return NULL;
1167
1168 default:
1169 return "RAID10 can only be changed to RAID0";
1170 }
1171 case 0:
1172 /* RAID0 can be converted to RAID10, or to RAID456 */
1173 if (info->new_level == 10) {
1174 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1175 /* Assume near=2 layout */
1176 info->new_layout = 0x102;
1177 info->delta_disks = info->array.raid_disks;
1178 }
1179 if (info->new_layout == UnSet) {
1180 int copies = 1 + (info->delta_disks
1181 / info->array.raid_disks);
1182 if (info->array.raid_disks * (copies-1)
1183 != info->delta_disks)
1184 return "Impossible number of devices for RAID0->RAID10";
1185 info->new_layout = 0x100 + copies;
1186 }
1187 if (info->delta_disks == UnSet) {
1188 int copies = info->new_layout & 0xff;
1189 if (info->new_layout != 0x100 + copies)
1190 return "New layout impossible for RAID0->RAID10";;
1191 info->delta_disks = (copies - 1) *
1192 info->array.raid_disks;
1193 }
1194 if (info->new_chunk &&
1195 info->new_chunk != info->array.chunk_size)
1196 return "Cannot change chunk-size with RAID0->RAID10";
1197 /* looks good */
1198 re->level = 10;
1199 re->before.data_disks = (info->array.raid_disks +
1200 info->delta_disks);
1201 re->after.data_disks = re->before.data_disks;
1202 re->before.layout = info->new_layout;
1203 return NULL;
1204 }
1205
1206 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1207 * a raid4 style layout of the final level.
1208 */
1209 switch (info->new_level) {
1210 case 4:
1211 delta_parity = 1;
1212 case 0:
1213 re->level = 4;
1214 re->before.layout = 0;
1215 break;
1216 case 5:
1217 delta_parity = 1;
1218 re->level = 5;
1219 re->before.layout = ALGORITHM_PARITY_N;
1220 if (info->new_layout == UnSet)
1221 info->new_layout = map_name(r5layout, "default");
1222 break;
1223 case 6:
1224 delta_parity = 2;
1225 re->level = 6;
1226 re->before.layout = ALGORITHM_PARITY_N;
1227 if (info->new_layout == UnSet)
1228 info->new_layout = map_name(r6layout, "default");
1229 break;
1230 default:
1231 return "Impossible level change requested";
1232 }
1233 re->before.data_disks = info->array.raid_disks;
1234 /* determining 'after' layout happens outside this 'switch' */
1235 break;
1236
1237 case 4:
1238 info->array.layout = ALGORITHM_PARITY_N;
1239 case 5:
1240 switch (info->new_level) {
1241 case 0:
1242 delta_parity = -1;
1243 case 4:
1244 re->level = info->array.level;
1245 re->before.data_disks = info->array.raid_disks - 1;
1246 re->before.layout = info->array.layout;
1247 break;
1248 case 5:
1249 re->level = 5;
1250 re->before.data_disks = info->array.raid_disks - 1;
1251 re->before.layout = info->array.layout;
1252 break;
1253 case 6:
1254 delta_parity = 1;
1255 re->level = 6;
1256 re->before.data_disks = info->array.raid_disks - 1;
1257 switch (info->array.layout) {
1258 case ALGORITHM_LEFT_ASYMMETRIC:
1259 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1260 break;
1261 case ALGORITHM_RIGHT_ASYMMETRIC:
1262 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1263 break;
1264 case ALGORITHM_LEFT_SYMMETRIC:
1265 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1266 break;
1267 case ALGORITHM_RIGHT_SYMMETRIC:
1268 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1269 break;
1270 case ALGORITHM_PARITY_0:
1271 re->before.layout = ALGORITHM_PARITY_0_6;
1272 break;
1273 case ALGORITHM_PARITY_N:
1274 re->before.layout = ALGORITHM_PARITY_N_6;
1275 break;
1276 default:
1277 return "Cannot convert an array with this layout";
1278 }
1279 break;
1280 case 1:
1281 if (info->array.raid_disks != 2)
1282 return "Can only convert a 2-device array to RAID1";
1283 if (info->delta_disks != UnSet &&
1284 info->delta_disks != 0)
1285 return "Cannot set raid_disk when converting RAID5->RAID1";
1286 re->level = 1;
1287 info->new_chunk = 0;
1288 return NULL;
1289 default:
1290 return "Impossible level change requested";
1291 }
1292 break;
1293 case 6:
1294 switch (info->new_level) {
1295 case 4:
1296 case 5:
1297 delta_parity = -1;
1298 case 6:
1299 re->level = 6;
1300 re->before.data_disks = info->array.raid_disks - 2;
1301 re->before.layout = info->array.layout;
1302 break;
1303 default:
1304 return "Impossible level change requested";
1305 }
1306 break;
1307 }
1308
1309 /* If we reached here then it looks like a re-stripe is
1310 * happening. We have determined the intermediate level
1311 * and initial raid_disks/layout and stored these in 're'.
1312 *
1313 * We need to deduce the final layout that can be atomically
1314 * converted to the end state.
1315 */
1316 switch (info->new_level) {
1317 case 0:
1318 /* We can only get to RAID0 from RAID4 or RAID5
1319 * with appropriate layout and one extra device
1320 */
1321 if (re->level != 4 && re->level != 5)
1322 return "Cannot covert to RAID0 from this level";
1323
1324 switch (re->level) {
1325 case 4:
1326 re->before.layout = 0;
1327 re->after.layout = 0;
1328 break;
1329 case 5:
1330 re->after.layout = ALGORITHM_PARITY_N;
1331 break;
1332 }
1333 break;
1334
1335 case 4:
1336 /* We can only get to RAID4 from RAID5 */
1337 if (re->level != 4 && re->level != 5)
1338 return "Cannot convert to RAID4 from this level";
1339
1340 switch (re->level) {
1341 case 4:
1342 re->after.layout = 0;
1343 break;
1344 case 5:
1345 re->after.layout = ALGORITHM_PARITY_N;
1346 break;
1347 }
1348 break;
1349
1350 case 5:
1351 /* We get to RAID5 from RAID5 or RAID6 */
1352 if (re->level != 5 && re->level != 6)
1353 return "Cannot convert to RAID5 from this level";
1354
1355 switch (re->level) {
1356 case 5:
1357 if (info->new_layout == UnSet)
1358 re->after.layout = re->before.layout;
1359 else
1360 re->after.layout = info->new_layout;
1361 break;
1362 case 6:
1363 if (info->new_layout == UnSet)
1364 info->new_layout = re->before.layout;
1365
1366 /* after.layout needs to be raid6 version of new_layout */
1367 if (info->new_layout == ALGORITHM_PARITY_N)
1368 re->after.layout = ALGORITHM_PARITY_N;
1369 else {
1370 char layout[40];
1371 char *ls = map_num(r5layout, info->new_layout);
1372 int l;
1373 if (ls) {
1374 /* Current RAID6 layout has a RAID5
1375 * equivalent - good
1376 */
1377 strcat(strcpy(layout, ls), "-6");
1378 l = map_name(r6layout, layout);
1379 if (l == UnSet)
1380 return "Cannot find RAID6 layout to convert to";
1381 } else {
1382 /* Current RAID6 has no equivalent.
1383 * If it is already a '-6' layout we
1384 * can leave it unchanged, else we must
1385 * fail
1386 */
1387 ls = map_num(r6layout, info->new_layout);
1388 if (!ls ||
1389 strcmp(ls+strlen(ls)-2, "-6") != 0)
1390 return "Please specify new layout";
1391 l = info->new_layout;
1392 }
1393 re->after.layout = l;
1394 }
1395 }
1396 break;
1397
1398 case 6:
1399 /* We must already be at level 6 */
1400 if (re->level != 6)
1401 return "Impossible level change";
1402 if (info->new_layout == UnSet)
1403 re->after.layout = info->array.layout;
1404 else
1405 re->after.layout = info->new_layout;
1406 break;
1407 default:
1408 return "Impossible level change requested";
1409 }
1410 if (info->delta_disks == UnSet)
1411 info->delta_disks = delta_parity;
1412
1413 re->after.data_disks = (re->before.data_disks
1414 + info->delta_disks
1415 - delta_parity);
1416 switch (re->level) {
1417 case 6: re->parity = 2;
1418 break;
1419 case 4:
1420 case 5: re->parity = 1;
1421 break;
1422 default: re->parity = 0;
1423 break;
1424 }
1425 /* So we have a restripe operation, we need to calculate the number
1426 * of blocks per reshape operation.
1427 */
1428 re->new_size = info->component_size * re->before.data_disks;
1429 if (info->new_chunk == 0)
1430 info->new_chunk = info->array.chunk_size;
1431 if (re->after.data_disks == re->before.data_disks &&
1432 re->after.layout == re->before.layout &&
1433 info->new_chunk == info->array.chunk_size) {
1434 /* Nothing to change, can change level immediately. */
1435 re->level = info->new_level;
1436 re->backup_blocks = 0;
1437 return NULL;
1438 }
1439 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1440 /* chunk and layout changes make no difference */
1441 re->level = info->new_level;
1442 re->backup_blocks = 0;
1443 return NULL;
1444 }
1445
1446 if (re->after.data_disks == re->before.data_disks &&
1447 get_linux_version() < 2006032)
1448 return "in-place reshape is not safe before 2.6.32 - sorry.";
1449
1450 if (re->after.data_disks < re->before.data_disks &&
1451 get_linux_version() < 2006030)
1452 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1453
1454 re->backup_blocks = compute_backup_blocks(
1455 info->new_chunk, info->array.chunk_size,
1456 re->after.data_disks,
1457 re->before.data_disks);
1458 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1459
1460 re->new_size = info->component_size * re->after.data_disks;
1461 return NULL;
1462 }
1463
1464 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1465 char *text_version)
1466 {
1467 struct mdinfo *info;
1468 char *subarray;
1469 int ret_val = -1;
1470
1471 if ((st == NULL) || (sra == NULL))
1472 return ret_val;
1473
1474 if (text_version == NULL)
1475 text_version = sra->text_version;
1476 subarray = strchr(text_version+1, '/')+1;
1477 info = st->ss->container_content(st, subarray);
1478 if (info) {
1479 unsigned long long current_size = 0;
1480 unsigned long long new_size =
1481 info->custom_array_size/2;
1482
1483 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1484 new_size > current_size) {
1485 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1486 < 0)
1487 dprintf("Error: Cannot set array size");
1488 else {
1489 ret_val = 0;
1490 dprintf("Array size changed");
1491 }
1492 dprintf_cont(" from %llu to %llu.\n",
1493 current_size, new_size);
1494 }
1495 sysfs_free(info);
1496 } else
1497 dprintf("Error: set_array_size(): info pointer in NULL\n");
1498
1499 return ret_val;
1500 }
1501
1502 static int reshape_array(char *container, int fd, char *devname,
1503 struct supertype *st, struct mdinfo *info,
1504 int force, struct mddev_dev *devlist,
1505 unsigned long long data_offset,
1506 char *backup_file, int verbose, int forked,
1507 int restart, int freeze_reshape);
1508 static int reshape_container(char *container, char *devname,
1509 int mdfd,
1510 struct supertype *st,
1511 struct mdinfo *info,
1512 int force,
1513 char *backup_file, int verbose,
1514 int forked, int restart, int freeze_reshape);
1515
1516 int Grow_reshape(char *devname, int fd,
1517 struct mddev_dev *devlist,
1518 unsigned long long data_offset,
1519 struct context *c, struct shape *s)
1520 {
1521 /* Make some changes in the shape of an array.
1522 * The kernel must support the change.
1523 *
1524 * There are three different changes. Each can trigger
1525 * a resync or recovery so we freeze that until we have
1526 * requested everything (if kernel supports freezing - 2.6.30).
1527 * The steps are:
1528 * - change size (i.e. component_size)
1529 * - change level
1530 * - change layout/chunksize/ndisks
1531 *
1532 * The last can require a reshape. It is different on different
1533 * levels so we need to check the level before actioning it.
1534 * Some times the level change needs to be requested after the
1535 * reshape (e.g. raid6->raid5, raid5->raid0)
1536 *
1537 */
1538 struct mdu_array_info_s array;
1539 int rv = 0;
1540 struct supertype *st;
1541 char *subarray = NULL;
1542
1543 int frozen;
1544 int changed = 0;
1545 char *container = NULL;
1546 int cfd = -1;
1547
1548 struct mddev_dev *dv;
1549 int added_disks;
1550
1551 struct mdinfo info;
1552 struct mdinfo *sra;
1553
1554 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1555 pr_err("%s is not an active md array - aborting\n",
1556 devname);
1557 return 1;
1558 }
1559 if (data_offset != INVALID_SECTORS && array.level != 10
1560 && (array.level < 4 || array.level > 6)) {
1561 pr_err("--grow --data-offset not yet supported\n");
1562 return 1;
1563 }
1564
1565 if (s->size > 0 &&
1566 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1567 pr_err("cannot change component size at the same time as other changes.\n"
1568 " Change size first, then check data is intact before making other changes.\n");
1569 return 1;
1570 }
1571
1572 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1573 get_linux_version() < 2006032 &&
1574 !check_env("MDADM_FORCE_FEWER")) {
1575 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1576 " Please use a newer kernel\n");
1577 return 1;
1578 }
1579
1580 st = super_by_fd(fd, &subarray);
1581 if (!st) {
1582 pr_err("Unable to determine metadata format for %s\n", devname);
1583 return 1;
1584 }
1585 if (s->raiddisks > st->max_devs) {
1586 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1587 return 1;
1588 }
1589
1590 /* in the external case we need to check that the requested reshape is
1591 * supported, and perform an initial check that the container holds the
1592 * pre-requisite spare devices (mdmon owns final validation)
1593 */
1594 if (st->ss->external) {
1595 int rv;
1596
1597 if (subarray) {
1598 container = st->container_devnm;
1599 cfd = open_dev_excl(st->container_devnm);
1600 } else {
1601 container = st->devnm;
1602 close(fd);
1603 cfd = open_dev_excl(st->devnm);
1604 fd = cfd;
1605 }
1606 if (cfd < 0) {
1607 pr_err("Unable to open container for %s\n",
1608 devname);
1609 free(subarray);
1610 return 1;
1611 }
1612
1613 rv = st->ss->load_container(st, cfd, NULL);
1614
1615 if (rv) {
1616 pr_err("Cannot read superblock for %s\n",
1617 devname);
1618 free(subarray);
1619 return 1;
1620 }
1621
1622 /* check if operation is supported for metadata handler */
1623 if (st->ss->container_content) {
1624 struct mdinfo *cc = NULL;
1625 struct mdinfo *content = NULL;
1626
1627 cc = st->ss->container_content(st, subarray);
1628 for (content = cc; content ; content = content->next) {
1629 int allow_reshape = 1;
1630
1631 /* check if reshape is allowed based on metadata
1632 * indications stored in content.array.status
1633 */
1634 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1635 allow_reshape = 0;
1636 if (content->array.state
1637 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1638 allow_reshape = 0;
1639 if (!allow_reshape) {
1640 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1641 devname, container);
1642 sysfs_free(cc);
1643 free(subarray);
1644 return 1;
1645 }
1646 }
1647 sysfs_free(cc);
1648 }
1649 if (mdmon_running(container))
1650 st->update_tail = &st->updates;
1651 }
1652
1653 added_disks = 0;
1654 for (dv = devlist; dv; dv = dv->next)
1655 added_disks++;
1656 if (s->raiddisks > array.raid_disks &&
1657 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1658 !c->force) {
1659 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1660 " Use --force to over-ride this check.\n",
1661 s->raiddisks - array.raid_disks,
1662 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1663 array.spare_disks + added_disks);
1664 return 1;
1665 }
1666
1667 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1668 | GET_STATE | GET_VERSION);
1669 if (sra) {
1670 if (st->ss->external && subarray == NULL) {
1671 array.level = LEVEL_CONTAINER;
1672 sra->array.level = LEVEL_CONTAINER;
1673 }
1674 } else {
1675 pr_err("failed to read sysfs parameters for %s\n",
1676 devname);
1677 return 1;
1678 }
1679 frozen = freeze(st);
1680 if (frozen < -1) {
1681 /* freeze() already spewed the reason */
1682 sysfs_free(sra);
1683 return 1;
1684 } else if (frozen < 0) {
1685 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1686 sysfs_free(sra);
1687 return 1;
1688 }
1689
1690 /* ========= set size =============== */
1691 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1692 unsigned long long orig_size = get_component_size(fd)/2;
1693 unsigned long long min_csize;
1694 struct mdinfo *mdi;
1695 int raid0_takeover = 0;
1696
1697 if (orig_size == 0)
1698 orig_size = (unsigned) array.size;
1699
1700 if (orig_size == 0) {
1701 pr_err("Cannot set device size in this type of array.\n");
1702 rv = 1;
1703 goto release;
1704 }
1705
1706 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1707 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1708 rv = 1;
1709 goto release;
1710 }
1711 sync_metadata(st);
1712 if (st->ss->external) {
1713 /* metadata can have size limitation
1714 * update size value according to metadata information
1715 */
1716 struct mdinfo *sizeinfo =
1717 st->ss->container_content(st, subarray);
1718 if (sizeinfo) {
1719 unsigned long long new_size =
1720 sizeinfo->custom_array_size/2;
1721 int data_disks = get_data_disks(
1722 sizeinfo->array.level,
1723 sizeinfo->array.layout,
1724 sizeinfo->array.raid_disks);
1725 new_size /= data_disks;
1726 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size, new_size,
1727 new_size * data_disks);
1728 s->size = new_size;
1729 sysfs_free(sizeinfo);
1730 }
1731 }
1732
1733 /* Update the size of each member device in case
1734 * they have been resized. This will never reduce
1735 * below the current used-size. The "size" attribute
1736 * understands '0' to mean 'max'.
1737 */
1738 min_csize = 0;
1739 rv = 0;
1740 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1741 if (sysfs_set_num(sra, mdi, "size",
1742 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1743 /* Probably kernel refusing to let us
1744 * reduce the size - not an error.
1745 */
1746 break;
1747 }
1748 if (array.not_persistent == 0 &&
1749 array.major_version == 0 &&
1750 get_linux_version() < 3001000) {
1751 /* Dangerous to allow size to exceed 2TB */
1752 unsigned long long csize;
1753 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1754 if (csize >= 2ULL*1024*1024*1024)
1755 csize = 2ULL*1024*1024*1024;
1756 if ((min_csize == 0 || (min_csize
1757 > csize)))
1758 min_csize = csize;
1759 }
1760 }
1761 }
1762 if (rv) {
1763 pr_err("Cannot set size on array members.\n");
1764 goto size_change_error;
1765 }
1766 if (min_csize && s->size > min_csize) {
1767 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1768 rv = 1;
1769 goto size_change_error;
1770 }
1771 if (min_csize && s->size == MAX_SIZE) {
1772 /* Don't let the kernel choose a size - it will get
1773 * it wrong
1774 */
1775 pr_err("Limited v0.90 array to 2TB per device\n");
1776 s->size = min_csize;
1777 }
1778 if (st->ss->external) {
1779 if (sra->array.level == 0) {
1780 rv = sysfs_set_str(sra, NULL, "level",
1781 "raid5");
1782 if (!rv) {
1783 raid0_takeover = 1;
1784 /* get array parameters after takeover
1785 * to change one parameter at time only
1786 */
1787 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1788 }
1789 }
1790 /* make sure mdmon is
1791 * aware of the new level */
1792 if (!mdmon_running(st->container_devnm))
1793 start_mdmon(st->container_devnm);
1794 ping_monitor(container);
1795 if (mdmon_running(st->container_devnm) &&
1796 st->update_tail == NULL)
1797 st->update_tail = &st->updates;
1798 }
1799
1800 if (s->size == MAX_SIZE)
1801 s->size = 0;
1802 array.size = s->size;
1803 if (s->size & ~INT32_MAX) {
1804 /* got truncated to 32bit, write to
1805 * component_size instead
1806 */
1807 if (sra)
1808 rv = sysfs_set_num(sra, NULL,
1809 "component_size", s->size);
1810 else
1811 rv = -1;
1812 } else {
1813 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1814
1815 /* manage array size when it is managed externally
1816 */
1817 if ((rv == 0) && st->ss->external)
1818 rv = set_array_size(st, sra, sra->text_version);
1819 }
1820
1821 if (raid0_takeover) {
1822 /* do not recync non-existing parity,
1823 * we will drop it anyway
1824 */
1825 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1826 /* go back to raid0, drop parity disk
1827 */
1828 sysfs_set_str(sra, NULL, "level", "raid0");
1829 ioctl(fd, GET_ARRAY_INFO, &array);
1830 }
1831
1832 size_change_error:
1833 if (rv != 0) {
1834 int err = errno;
1835
1836 /* restore metadata */
1837 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1838 UnSet, NULL, devname,
1839 ROLLBACK_METADATA_CHANGES,
1840 c->verbose) == 0)
1841 sync_metadata(st);
1842 pr_err("Cannot set device size for %s: %s\n",
1843 devname, strerror(err));
1844 if (err == EBUSY &&
1845 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1846 cont_err("Bitmap must be removed before size can be changed\n");
1847 rv = 1;
1848 goto release;
1849 }
1850 if (s->assume_clean) {
1851 /* This will fail on kernels older than 3.0 unless
1852 * a backport has been arranged.
1853 */
1854 if (sra == NULL ||
1855 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1856 pr_err("--assume-clean not supported with --grow on this kernel\n");
1857 }
1858 ioctl(fd, GET_ARRAY_INFO, &array);
1859 s->size = get_component_size(fd)/2;
1860 if (s->size == 0)
1861 s->size = array.size;
1862 if (c->verbose >= 0) {
1863 if (s->size == orig_size)
1864 pr_err("component size of %s unchanged at %lluK\n",
1865 devname, s->size);
1866 else
1867 pr_err("component size of %s has been set to %lluK\n",
1868 devname, s->size);
1869 }
1870 changed = 1;
1871 } else if (array.level != LEVEL_CONTAINER) {
1872 s->size = get_component_size(fd)/2;
1873 if (s->size == 0)
1874 s->size = array.size;
1875 }
1876
1877 /* See if there is anything else to do */
1878 if ((s->level == UnSet || s->level == array.level) &&
1879 (s->layout_str == NULL) &&
1880 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1881 data_offset == INVALID_SECTORS &&
1882 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1883 /* Nothing more to do */
1884 if (!changed && c->verbose >= 0)
1885 pr_err("%s: no change requested\n",
1886 devname);
1887 goto release;
1888 }
1889
1890 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1891 * current implementation assumes that following conditions must be met:
1892 * - RAID10:
1893 * - far_copies == 1
1894 * - near_copies == 2
1895 */
1896 if ((s->level == 0 && array.level == 10 && sra &&
1897 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1898 (s->level == 0 && array.level == 1 && sra)) {
1899 int err;
1900 err = remove_disks_for_takeover(st, sra, array.layout);
1901 if (err) {
1902 dprintf("Array cannot be reshaped\n");
1903 if (cfd > -1)
1904 close(cfd);
1905 rv = 1;
1906 goto release;
1907 }
1908 /* Make sure mdmon has seen the device removal
1909 * and updated metadata before we continue with
1910 * level change
1911 */
1912 if (container)
1913 ping_monitor(container);
1914 }
1915
1916 memset(&info, 0, sizeof(info));
1917 info.array = array;
1918 sysfs_init(&info, fd, NULL);
1919 strcpy(info.text_version, sra->text_version);
1920 info.component_size = s->size*2;
1921 info.new_level = s->level;
1922 info.new_chunk = s->chunk * 1024;
1923 if (info.array.level == LEVEL_CONTAINER) {
1924 info.delta_disks = UnSet;
1925 info.array.raid_disks = s->raiddisks;
1926 } else if (s->raiddisks)
1927 info.delta_disks = s->raiddisks - info.array.raid_disks;
1928 else
1929 info.delta_disks = UnSet;
1930 if (s->layout_str == NULL) {
1931 info.new_layout = UnSet;
1932 if (info.array.level == 6 &&
1933 (info.new_level == 6 || info.new_level == UnSet) &&
1934 info.array.layout >= 16) {
1935 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
1936 cont_err("during the reshape, please specify --layout=preserve\n");
1937 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1938 rv = 1;
1939 goto release;
1940 }
1941 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1942 strcmp(s->layout_str, "normalize") == 0) {
1943 /* If we have a -6 RAID6 layout, remove the '-6'. */
1944 info.new_layout = UnSet;
1945 if (info.array.level == 6 && info.new_level == UnSet) {
1946 char l[40], *h;
1947 strcpy(l, map_num(r6layout, info.array.layout));
1948 h = strrchr(l, '-');
1949 if (h && strcmp(h, "-6") == 0) {
1950 *h = 0;
1951 info.new_layout = map_name(r6layout, l);
1952 }
1953 } else {
1954 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
1955 rv = 1;
1956 goto release;
1957 }
1958 } else if (strcmp(s->layout_str, "preserve") == 0) {
1959 /* This means that a non-standard RAID6 layout
1960 * is OK.
1961 * In particular:
1962 * - When reshape a RAID6 (e.g. adding a device)
1963 * which is in a non-standard layout, it is OK
1964 * to preserve that layout.
1965 * - When converting a RAID5 to RAID6, leave it in
1966 * the XXX-6 layout, don't re-layout.
1967 */
1968 if (info.array.level == 6 && info.new_level == UnSet)
1969 info.new_layout = info.array.layout;
1970 else if (info.array.level == 5 && info.new_level == 6) {
1971 char l[40];
1972 strcpy(l, map_num(r5layout, info.array.layout));
1973 strcat(l, "-6");
1974 info.new_layout = map_name(r6layout, l);
1975 } else {
1976 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
1977 rv = 1;
1978 goto release;
1979 }
1980 } else {
1981 int l = info.new_level;
1982 if (l == UnSet)
1983 l = info.array.level;
1984 switch (l) {
1985 case 5:
1986 info.new_layout = map_name(r5layout, s->layout_str);
1987 break;
1988 case 6:
1989 info.new_layout = map_name(r6layout, s->layout_str);
1990 break;
1991 case 10:
1992 info.new_layout = parse_layout_10(s->layout_str);
1993 break;
1994 case LEVEL_FAULTY:
1995 info.new_layout = parse_layout_faulty(s->layout_str);
1996 break;
1997 default:
1998 pr_err("layout not meaningful with this level\n");
1999 rv = 1;
2000 goto release;
2001 }
2002 if (info.new_layout == UnSet) {
2003 pr_err("layout %s not understood for this level\n",
2004 s->layout_str);
2005 rv = 1;
2006 goto release;
2007 }
2008 }
2009
2010 if (array.level == LEVEL_FAULTY) {
2011 if (s->level != UnSet && s->level != array.level) {
2012 pr_err("cannot change level of Faulty device\n");
2013 rv =1 ;
2014 }
2015 if (s->chunk) {
2016 pr_err("cannot set chunksize of Faulty device\n");
2017 rv =1 ;
2018 }
2019 if (s->raiddisks && s->raiddisks != 1) {
2020 pr_err("cannot set raid_disks of Faulty device\n");
2021 rv =1 ;
2022 }
2023 if (s->layout_str) {
2024 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2025 dprintf("Cannot get array information.\n");
2026 goto release;
2027 }
2028 array.layout = info.new_layout;
2029 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2030 pr_err("failed to set new layout\n");
2031 rv = 1;
2032 } else if (c->verbose >= 0)
2033 printf("layout for %s set to %d\n",
2034 devname, array.layout);
2035 }
2036 } else if (array.level == LEVEL_CONTAINER) {
2037 /* This change is to be applied to every array in the
2038 * container. This is only needed when the metadata imposes
2039 * restraints of the various arrays in the container.
2040 * Currently we only know that IMSM requires all arrays
2041 * to have the same number of devices so changing the
2042 * number of devices (On-Line Capacity Expansion) must be
2043 * performed at the level of the container
2044 */
2045 rv = reshape_container(container, devname, -1, st, &info,
2046 c->force, c->backup_file, c->verbose, 0, 0, 0);
2047 frozen = 0;
2048 } else {
2049 /* get spare devices from external metadata
2050 */
2051 if (st->ss->external) {
2052 struct mdinfo *info2;
2053
2054 info2 = st->ss->container_content(st, subarray);
2055 if (info2) {
2056 info.array.spare_disks =
2057 info2->array.spare_disks;
2058 sysfs_free(info2);
2059 }
2060 }
2061
2062 /* Impose these changes on a single array. First
2063 * check that the metadata is OK with the change. */
2064
2065 if (reshape_super(st, 0, info.new_level,
2066 info.new_layout, info.new_chunk,
2067 info.array.raid_disks, info.delta_disks,
2068 c->backup_file, devname, APPLY_METADATA_CHANGES,
2069 c->verbose)) {
2070 rv = 1;
2071 goto release;
2072 }
2073 sync_metadata(st);
2074 rv = reshape_array(container, fd, devname, st, &info, c->force,
2075 devlist, data_offset, c->backup_file, c->verbose,
2076 0, 0, 0);
2077 frozen = 0;
2078 }
2079 release:
2080 sysfs_free(sra);
2081 if (frozen > 0)
2082 unfreeze(st);
2083 return rv;
2084 }
2085
2086 /* verify_reshape_position()
2087 * Function checks if reshape position in metadata is not farther
2088 * than position in md.
2089 * Return value:
2090 * 0 : not valid sysfs entry
2091 * it can be caused by not started reshape, it should be started
2092 * by reshape array or raid0 array is before takeover
2093 * -1 : error, reshape position is obviously wrong
2094 * 1 : success, reshape progress correct or updated
2095 */
2096 static int verify_reshape_position(struct mdinfo *info, int level)
2097 {
2098 int ret_val = 0;
2099 char buf[40];
2100 int rv;
2101
2102 /* read sync_max, failure can mean raid0 array */
2103 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2104
2105 if (rv > 0) {
2106 char *ep;
2107 unsigned long long position = strtoull(buf, &ep, 0);
2108
2109 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2110 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2111 position *= get_data_disks(level,
2112 info->new_layout,
2113 info->array.raid_disks);
2114 if (info->reshape_progress < position) {
2115 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2116 info->reshape_progress, position);
2117 info->reshape_progress = position;
2118 ret_val = 1;
2119 } else if (info->reshape_progress > position) {
2120 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2121 position, info->reshape_progress);
2122 ret_val = -1;
2123 } else {
2124 dprintf("Reshape position in md and metadata are the same;");
2125 ret_val = 1;
2126 }
2127 }
2128 } else if (rv == 0) {
2129 /* for valid sysfs entry, 0-length content
2130 * should be indicated as error
2131 */
2132 ret_val = -1;
2133 }
2134
2135 return ret_val;
2136 }
2137
2138 static unsigned long long choose_offset(unsigned long long lo,
2139 unsigned long long hi,
2140 unsigned long long min,
2141 unsigned long long max)
2142 {
2143 /* Choose a new offset between hi and lo.
2144 * It must be between min and max, but
2145 * we would prefer something near the middle of hi/lo, and also
2146 * prefer to be aligned to a big power of 2.
2147 *
2148 * So we start with the middle, then for each bit,
2149 * starting at '1' and increasing, if it is set, we either
2150 * add it or subtract it if possible, preferring the option
2151 * which is furthest from the boundary.
2152 *
2153 * We stop once we get a 1MB alignment. As units are in sectors,
2154 * 1MB = 2*1024 sectors.
2155 */
2156 unsigned long long choice = (lo + hi) / 2;
2157 unsigned long long bit = 1;
2158
2159 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2160 unsigned long long bigger, smaller;
2161 if (! (bit & choice))
2162 continue;
2163 bigger = choice + bit;
2164 smaller = choice - bit;
2165 if (bigger > max && smaller < min)
2166 break;
2167 if (bigger > max)
2168 choice = smaller;
2169 else if (smaller < min)
2170 choice = bigger;
2171 else if (hi - bigger > smaller - lo)
2172 choice = bigger;
2173 else
2174 choice = smaller;
2175 }
2176 return choice;
2177 }
2178
2179 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2180 char *devname, int delta_disks,
2181 unsigned long long data_offset,
2182 unsigned long long min,
2183 int can_fallback)
2184 {
2185 struct mdinfo *sd;
2186 int dir = 0;
2187 int err = 0;
2188 unsigned long long before, after;
2189
2190 /* Need to find min space before and after so same is used
2191 * on all devices
2192 */
2193 before = UINT64_MAX;
2194 after = UINT64_MAX;
2195 for (sd = sra->devs; sd; sd = sd->next) {
2196 char *dn;
2197 int dfd;
2198 int rv;
2199 struct supertype *st2;
2200 struct mdinfo info2;
2201
2202 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2203 continue;
2204 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2205 dfd = dev_open(dn, O_RDONLY);
2206 if (dfd < 0) {
2207 pr_err("%s: cannot open component %s\n",
2208 devname, dn ? dn : "-unknown-");
2209 goto release;
2210 }
2211 st2 = dup_super(st);
2212 rv = st2->ss->load_super(st2,dfd, NULL);
2213 close(dfd);
2214 if (rv) {
2215 free(st2);
2216 pr_err("%s: cannot get superblock from %s\n",
2217 devname, dn);
2218 goto release;
2219 }
2220 st2->ss->getinfo_super(st2, &info2, NULL);
2221 st2->ss->free_super(st2);
2222 free(st2);
2223 if (info2.space_before == 0 &&
2224 info2.space_after == 0) {
2225 /* Metadata doesn't support data_offset changes */
2226 if (!can_fallback)
2227 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2228 devname);
2229 goto fallback;
2230 }
2231 if (before > info2.space_before)
2232 before = info2.space_before;
2233 if (after > info2.space_after)
2234 after = info2.space_after;
2235
2236 if (data_offset != INVALID_SECTORS) {
2237 if (dir == 0) {
2238 if (info2.data_offset == data_offset) {
2239 pr_err("%s: already has that data_offset\n",
2240 dn);
2241 goto release;
2242 }
2243 if (data_offset < info2.data_offset)
2244 dir = -1;
2245 else
2246 dir = 1;
2247 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2248 (data_offset >= info2.data_offset && dir == -1)) {
2249 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2250 dn);
2251 goto release;
2252 }
2253 }
2254 }
2255 if (before == UINT64_MAX)
2256 /* impossible really, there must be no devices */
2257 return 1;
2258
2259 for (sd = sra->devs; sd; sd = sd->next) {
2260 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2261 unsigned long long new_data_offset;
2262
2263 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2264 continue;
2265 if (delta_disks < 0) {
2266 /* Don't need any space as array is shrinking
2267 * just move data_offset up by min
2268 */
2269 if (data_offset == INVALID_SECTORS)
2270 new_data_offset = sd->data_offset + min;
2271 else {
2272 if (data_offset < sd->data_offset + min) {
2273 pr_err("--data-offset too small for %s\n",
2274 dn);
2275 goto release;
2276 }
2277 new_data_offset = data_offset;
2278 }
2279 } else if (delta_disks > 0) {
2280 /* need space before */
2281 if (before < min) {
2282 if (can_fallback)
2283 goto fallback;
2284 pr_err("Insufficient head-space for reshape on %s\n",
2285 dn);
2286 goto release;
2287 }
2288 if (data_offset == INVALID_SECTORS)
2289 new_data_offset = sd->data_offset - min;
2290 else {
2291 if (data_offset > sd->data_offset - min) {
2292 pr_err("--data-offset too large for %s\n",
2293 dn);
2294 goto release;
2295 }
2296 new_data_offset = data_offset;
2297 }
2298 } else {
2299 if (dir == 0) {
2300 /* can move up or down. If 'data_offset'
2301 * was set we would have already decided,
2302 * so just choose direction with most space.
2303 */
2304 if (before > after)
2305 dir = -1;
2306 else
2307 dir = 1;
2308 }
2309 sysfs_set_str(sra, NULL, "reshape_direction",
2310 dir == 1 ? "backwards" : "forwards");
2311 if (dir > 0) {
2312 /* Increase data offset */
2313 if (after < min) {
2314 if (can_fallback)
2315 goto fallback;
2316 pr_err("Insufficient tail-space for reshape on %s\n",
2317 dn);
2318 goto release;
2319 }
2320 if (data_offset != INVALID_SECTORS &&
2321 data_offset < sd->data_offset + min) {
2322 pr_err("--data-offset too small on %s\n",
2323 dn);
2324 goto release;
2325 }
2326 if (data_offset != INVALID_SECTORS)
2327 new_data_offset = data_offset;
2328 else
2329 new_data_offset = choose_offset(sd->data_offset,
2330 sd->data_offset + after,
2331 sd->data_offset + min,
2332 sd->data_offset + after);
2333 } else {
2334 /* Decrease data offset */
2335 if (before < min) {
2336 if (can_fallback)
2337 goto fallback;
2338 pr_err("insufficient head-room on %s\n",
2339 dn);
2340 goto release;
2341 }
2342 if (data_offset != INVALID_SECTORS &&
2343 data_offset < sd->data_offset - min) {
2344 pr_err("--data-offset too small on %s\n",
2345 dn);
2346 goto release;
2347 }
2348 if (data_offset != INVALID_SECTORS)
2349 new_data_offset = data_offset;
2350 else
2351 new_data_offset = choose_offset(sd->data_offset - before,
2352 sd->data_offset,
2353 sd->data_offset - before,
2354 sd->data_offset - min);
2355 }
2356 }
2357 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2358 if (err < 0 && errno == E2BIG) {
2359 /* try again after increasing data size to max */
2360 err = sysfs_set_num(sra, sd, "size", 0);
2361 if (err < 0 && errno == EINVAL &&
2362 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2363 /* some kernels have a bug where you cannot
2364 * use '0' on spare devices. */
2365 sysfs_set_num(sra, sd, "size",
2366 (sra->component_size + after)/2);
2367 }
2368 err = sysfs_set_num(sra, sd, "new_offset",
2369 new_data_offset);
2370 }
2371 if (err < 0) {
2372 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2373 pr_err("data-offset is too big for %s\n",
2374 dn);
2375 goto release;
2376 }
2377 if (sd == sra->devs &&
2378 (errno == ENOENT || errno == E2BIG))
2379 /* Early kernel, no 'new_offset' file,
2380 * or kernel doesn't like us.
2381 * For RAID5/6 this is not fatal
2382 */
2383 return 1;
2384 pr_err("Cannot set new_offset for %s\n",
2385 dn);
2386 break;
2387 }
2388 }
2389 return err;
2390 release:
2391 return -1;
2392 fallback:
2393 /* Just use a backup file */
2394 return 1;
2395 }
2396
2397 static int raid10_reshape(char *container, int fd, char *devname,
2398 struct supertype *st, struct mdinfo *info,
2399 struct reshape *reshape,
2400 unsigned long long data_offset,
2401 int force, int verbose)
2402 {
2403 /* Changing raid_disks, layout, chunksize or possibly
2404 * just data_offset for a RAID10.
2405 * We must always change data_offset. We change by at least
2406 * ->min_offset_change which is the largest of the old and new
2407 * chunk sizes.
2408 * If raid_disks is increasing, then data_offset must decrease
2409 * by at least this copy size.
2410 * If raid_disks is unchanged, data_offset must increase or
2411 * decrease by at least min_offset_change but preferably by much more.
2412 * We choose half of the available space.
2413 * If raid_disks is decreasing, data_offset must increase by
2414 * at least min_offset_change. To allow of this, component_size
2415 * must be decreased by the same amount.
2416 *
2417 * So we calculate the required minimum and direction, possibly
2418 * reduce the component_size, then iterate through the devices
2419 * and set the new_data_offset.
2420 * If that all works, we set chunk_size, layout, raid_disks, and start
2421 * 'reshape'
2422 */
2423 struct mdinfo *sra;
2424 unsigned long long min;
2425 int err = 0;
2426
2427 sra = sysfs_read(fd, NULL,
2428 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2429 );
2430 if (!sra) {
2431 pr_err("%s: Cannot get array details from sysfs\n",
2432 devname);
2433 goto release;
2434 }
2435 min = reshape->min_offset_change;
2436
2437 if (info->delta_disks)
2438 sysfs_set_str(sra, NULL, "reshape_direction",
2439 info->delta_disks < 0 ? "backwards" : "forwards");
2440 if (info->delta_disks < 0 &&
2441 info->space_after < min) {
2442 int rv = sysfs_set_num(sra, NULL, "component_size",
2443 (sra->component_size -
2444 min)/2);
2445 if (rv) {
2446 pr_err("cannot reduce component size\n");
2447 goto release;
2448 }
2449 }
2450 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2451 min, 0);
2452 if (err == 1) {
2453 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2454 cont_err("supported on this kernel\n");
2455 err = -1;
2456 }
2457 if (err < 0)
2458 goto release;
2459
2460 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2461 err = errno;
2462 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2463 err = errno;
2464 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2465 info->array.raid_disks + info->delta_disks) < 0)
2466 err = errno;
2467 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2468 err = errno;
2469 if (err) {
2470 pr_err("Cannot set array shape for %s\n",
2471 devname);
2472 if (err == EBUSY &&
2473 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2474 cont_err(" Bitmap must be removed before shape can be changed\n");
2475 goto release;
2476 }
2477 sysfs_free(sra);
2478 return 0;
2479 release:
2480 sysfs_free(sra);
2481 return 1;
2482 }
2483
2484 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2485 {
2486 struct mdinfo *sra, *sd;
2487 /* Initialisation to silence compiler warning */
2488 unsigned long long min_space_before = 0, min_space_after = 0;
2489 int first = 1;
2490
2491 sra = sysfs_read(fd, NULL, GET_DEVS);
2492 if (!sra)
2493 return;
2494 for (sd = sra->devs; sd; sd = sd->next) {
2495 char *dn;
2496 int dfd;
2497 struct supertype *st2;
2498 struct mdinfo info2;
2499
2500 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2501 continue;
2502 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2503 dfd = dev_open(dn, O_RDONLY);
2504 if (dfd < 0)
2505 break;
2506 st2 = dup_super(st);
2507 if (st2->ss->load_super(st2,dfd, NULL)) {
2508 close(dfd);
2509 free(st2);
2510 break;
2511 }
2512 close(dfd);
2513 st2->ss->getinfo_super(st2, &info2, NULL);
2514 st2->ss->free_super(st2);
2515 free(st2);
2516 if (first ||
2517 min_space_before > info2.space_before)
2518 min_space_before = info2.space_before;
2519 if (first ||
2520 min_space_after > info2.space_after)
2521 min_space_after = info2.space_after;
2522 first = 0;
2523 }
2524 if (sd == NULL && !first) {
2525 info->space_after = min_space_after;
2526 info->space_before = min_space_before;
2527 }
2528 sysfs_free(sra);
2529 }
2530
2531 static void update_cache_size(char *container, struct mdinfo *sra,
2532 struct mdinfo *info,
2533 int disks, unsigned long long blocks)
2534 {
2535 /* Check that the internal stripe cache is
2536 * large enough, or it won't work.
2537 * It must hold at least 4 stripes of the larger
2538 * chunk size
2539 */
2540 unsigned long cache;
2541 cache = max(info->array.chunk_size, info->new_chunk);
2542 cache *= 4; /* 4 stripes minimum */
2543 cache /= 512; /* convert to sectors */
2544 /* make sure there is room for 'blocks' with a bit to spare */
2545 if (cache < 16 + blocks / disks)
2546 cache = 16 + blocks / disks;
2547 cache /= (4096/512); /* Covert from sectors to pages */
2548
2549 if (sra->cache_size < cache)
2550 subarray_set_num(container, sra, "stripe_cache_size",
2551 cache+1);
2552 }
2553
2554 static int impose_reshape(struct mdinfo *sra,
2555 struct mdinfo *info,
2556 struct supertype *st,
2557 int fd,
2558 int restart,
2559 char *devname, char *container,
2560 struct reshape *reshape)
2561 {
2562 struct mdu_array_info_s array;
2563
2564 sra->new_chunk = info->new_chunk;
2565
2566 if (restart) {
2567 /* for external metadata checkpoint saved by mdmon can be lost
2568 * or missed /due to e.g. crash/. Check if md is not during
2569 * restart farther than metadata points to.
2570 * If so, this means metadata information is obsolete.
2571 */
2572 if (st->ss->external)
2573 verify_reshape_position(info, reshape->level);
2574 sra->reshape_progress = info->reshape_progress;
2575 } else {
2576 sra->reshape_progress = 0;
2577 if (reshape->after.data_disks < reshape->before.data_disks)
2578 /* start from the end of the new array */
2579 sra->reshape_progress = (sra->component_size
2580 * reshape->after.data_disks);
2581 }
2582
2583 ioctl(fd, GET_ARRAY_INFO, &array);
2584 if (info->array.chunk_size == info->new_chunk &&
2585 reshape->before.layout == reshape->after.layout &&
2586 st->ss->external == 0) {
2587 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2588 array.raid_disks = reshape->after.data_disks + reshape->parity;
2589 if (!restart &&
2590 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2591 int err = errno;
2592
2593 pr_err("Cannot set device shape for %s: %s\n",
2594 devname, strerror(errno));
2595
2596 if (err == EBUSY &&
2597 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2598 cont_err("Bitmap must be removed before shape can be changed\n");
2599
2600 goto release;
2601 }
2602 } else if (!restart) {
2603 /* set them all just in case some old 'new_*' value
2604 * persists from some earlier problem.
2605 */
2606 int err = 0;
2607 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2608 err = errno;
2609 if (!err && sysfs_set_num(sra, NULL, "layout",
2610 reshape->after.layout) < 0)
2611 err = errno;
2612 if (!err && subarray_set_num(container, sra, "raid_disks",
2613 reshape->after.data_disks +
2614 reshape->parity) < 0)
2615 err = errno;
2616 if (err) {
2617 pr_err("Cannot set device shape for %s\n",
2618 devname);
2619
2620 if (err == EBUSY &&
2621 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2622 cont_err("Bitmap must be removed before shape can be changed\n");
2623 goto release;
2624 }
2625 }
2626 return 0;
2627 release:
2628 return -1;
2629 }
2630
2631 static int impose_level(int fd, int level, char *devname, int verbose)
2632 {
2633 char *c;
2634 struct mdu_array_info_s array;
2635 struct mdinfo info;
2636 sysfs_init(&info, fd, NULL);
2637
2638 ioctl(fd, GET_ARRAY_INFO, &array);
2639 if (level == 0 &&
2640 (array.level >= 4 && array.level <= 6)) {
2641 /* To convert to RAID0 we need to fail and
2642 * remove any non-data devices. */
2643 int found = 0;
2644 int d;
2645 int data_disks = array.raid_disks - 1;
2646 if (array.level == 6)
2647 data_disks -= 1;
2648 if (array.level == 5 &&
2649 array.layout != ALGORITHM_PARITY_N)
2650 return -1;
2651 if (array.level == 6 &&
2652 array.layout != ALGORITHM_PARITY_N_6)
2653 return -1;
2654 sysfs_set_str(&info, NULL,"sync_action", "idle");
2655 /* First remove any spares so no recovery starts */
2656 for (d = 0, found = 0;
2657 d < MAX_DISKS && found < array.nr_disks;
2658 d++) {
2659 mdu_disk_info_t disk;
2660 disk.number = d;
2661 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2662 continue;
2663 if (disk.major == 0 && disk.minor == 0)
2664 continue;
2665 found++;
2666 if ((disk.state & (1 << MD_DISK_ACTIVE))
2667 && disk.raid_disk < data_disks)
2668 /* keep this */
2669 continue;
2670 ioctl(fd, HOT_REMOVE_DISK,
2671 makedev(disk.major, disk.minor));
2672 }
2673 /* Now fail anything left */
2674 ioctl(fd, GET_ARRAY_INFO, &array);
2675 for (d = 0, found = 0;
2676 d < MAX_DISKS && found < array.nr_disks;
2677 d++) {
2678 int cnt;
2679 mdu_disk_info_t disk;
2680 disk.number = d;
2681 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2682 continue;
2683 if (disk.major == 0 && disk.minor == 0)
2684 continue;
2685 found++;
2686 if ((disk.state & (1 << MD_DISK_ACTIVE))
2687 && disk.raid_disk < data_disks)
2688 /* keep this */
2689 continue;
2690 ioctl(fd, SET_DISK_FAULTY,
2691 makedev(disk.major, disk.minor));
2692 cnt = 5;
2693 while (ioctl(fd, HOT_REMOVE_DISK,
2694 makedev(disk.major, disk.minor)) < 0
2695 && errno == EBUSY
2696 && cnt--) {
2697 usleep(10000);
2698 }
2699 }
2700 }
2701 c = map_num(pers, level);
2702 if (c) {
2703 int err = sysfs_set_str(&info, NULL, "level", c);
2704 if (err) {
2705 err = errno;
2706 pr_err("%s: could not set level to %s\n",
2707 devname, c);
2708 if (err == EBUSY &&
2709 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2710 cont_err("Bitmap must be removed before level can be changed\n");
2711 return err;
2712 }
2713 if (verbose >= 0)
2714 pr_err("level of %s changed to %s\n",
2715 devname, c);
2716 }
2717 return 0;
2718 }
2719
2720 int sigterm = 0;
2721 static void catch_term(int sig)
2722 {
2723 sigterm = 1;
2724 }
2725
2726 static int continue_via_systemd(char *devnm)
2727 {
2728 int skipped, i, pid, status;
2729 char pathbuf[1024];
2730 /* In a systemd/udev world, it is best to get systemd to
2731 * run "mdadm --grow --continue" rather than running in the
2732 * background.
2733 */
2734 switch(fork()) {
2735 case 0:
2736 /* FIXME yuk. CLOSE_EXEC?? */
2737 skipped = 0;
2738 for (i = 3; skipped < 20; i++)
2739 if (close(i) < 0)
2740 skipped++;
2741 else
2742 skipped = 0;
2743
2744 /* Don't want to see error messages from
2745 * systemctl. If the service doesn't exist,
2746 * we fork ourselves.
2747 */
2748 close(2);
2749 open("/dev/null", O_WRONLY);
2750 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2751 devnm);
2752 status = execl("/usr/bin/systemctl", "systemctl",
2753 "start",
2754 pathbuf, NULL);
2755 status = execl("/bin/systemctl", "systemctl", "start",
2756 pathbuf, NULL);
2757 exit(1);
2758 case -1: /* Just do it ourselves. */
2759 break;
2760 default: /* parent - good */
2761 pid = wait(&status);
2762 if (pid >= 0 && status == 0)
2763 return 1;
2764 }
2765 return 0;
2766 }
2767
2768 static int reshape_array(char *container, int fd, char *devname,
2769 struct supertype *st, struct mdinfo *info,
2770 int force, struct mddev_dev *devlist,
2771 unsigned long long data_offset,
2772 char *backup_file, int verbose, int forked,
2773 int restart, int freeze_reshape)
2774 {
2775 struct reshape reshape;
2776 int spares_needed;
2777 char *msg;
2778 int orig_level = UnSet;
2779 int odisks;
2780 int delayed;
2781
2782 struct mdu_array_info_s array;
2783 char *c;
2784
2785 struct mddev_dev *dv;
2786 int added_disks;
2787
2788 int *fdlist = NULL;
2789 unsigned long long *offsets = NULL;
2790 int d;
2791 int nrdisks;
2792 int err;
2793 unsigned long blocks;
2794 unsigned long long array_size;
2795 int done;
2796 struct mdinfo *sra = NULL;
2797 char buf[20];
2798
2799 /* when reshaping a RAID0, the component_size might be zero.
2800 * So try to fix that up.
2801 */
2802 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2803 dprintf("Cannot get array information.\n");
2804 goto release;
2805 }
2806 if (array.level == 0 && info->component_size == 0) {
2807 get_dev_size(fd, NULL, &array_size);
2808 info->component_size = array_size / array.raid_disks;
2809 }
2810
2811 if (array.level == 10)
2812 /* Need space_after info */
2813 get_space_after(fd, st, info);
2814
2815 if (info->reshape_active) {
2816 int new_level = info->new_level;
2817 info->new_level = UnSet;
2818 if (info->delta_disks > 0)
2819 info->array.raid_disks -= info->delta_disks;
2820 msg = analyse_change(devname, info, &reshape);
2821 info->new_level = new_level;
2822 if (info->delta_disks > 0)
2823 info->array.raid_disks += info->delta_disks;
2824 if (!restart)
2825 /* Make sure the array isn't read-only */
2826 ioctl(fd, RESTART_ARRAY_RW, 0);
2827 } else
2828 msg = analyse_change(devname, info, &reshape);
2829 if (msg) {
2830 /* if msg == "", error has already been printed */
2831 if (msg[0])
2832 pr_err("%s\n", msg);
2833 goto release;
2834 }
2835 if (restart &&
2836 (reshape.level != info->array.level ||
2837 reshape.before.layout != info->array.layout ||
2838 reshape.before.data_disks + reshape.parity
2839 != info->array.raid_disks - max(0, info->delta_disks))) {
2840 pr_err("reshape info is not in native format - cannot continue.\n");
2841 goto release;
2842 }
2843
2844 if (st->ss->external && restart && (info->reshape_progress == 0) &&
2845 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
2846 (strncmp(buf, "reshape", 7) == 0))) {
2847 /* When reshape is restarted from '0', very begin of array
2848 * it is possible that for external metadata reshape and array
2849 * configuration doesn't happen.
2850 * Check if md has the same opinion, and reshape is restarted
2851 * from 0. If so, this is regular reshape start after reshape
2852 * switch in metadata to next array only.
2853 */
2854 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2855 (info->reshape_progress == 0))
2856 restart = 0;
2857 }
2858 if (restart) {
2859 /* reshape already started. just skip to monitoring the reshape */
2860 if (reshape.backup_blocks == 0)
2861 return 0;
2862 if (restart & RESHAPE_NO_BACKUP)
2863 return 0;
2864
2865 /* Need 'sra' down at 'started:' */
2866 sra = sysfs_read(fd, NULL,
2867 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2868 GET_CACHE);
2869 if (!sra) {
2870 pr_err("%s: Cannot get array details from sysfs\n",
2871 devname);
2872 goto release;
2873 }
2874
2875 if (!backup_file)
2876 backup_file = locate_backup(sra->sys_name);
2877
2878 goto started;
2879 }
2880 /* The container is frozen but the array may not be.
2881 * So freeze the array so spares don't get put to the wrong use
2882 * FIXME there should probably be a cleaner separation between
2883 * freeze_array and freeze_container.
2884 */
2885 sysfs_freeze_array(info);
2886 /* Check we have enough spares to not be degraded */
2887 added_disks = 0;
2888 for (dv = devlist; dv ; dv=dv->next)
2889 added_disks++;
2890 spares_needed = max(reshape.before.data_disks,
2891 reshape.after.data_disks)
2892 + reshape.parity - array.raid_disks;
2893
2894 if (!force &&
2895 info->new_level > 1 && info->array.level > 1 &&
2896 spares_needed > info->array.spare_disks + added_disks) {
2897 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2898 " Use --force to over-ride this check.\n",
2899 spares_needed,
2900 spares_needed == 1 ? "" : "s",
2901 info->array.spare_disks + added_disks);
2902 goto release;
2903 }
2904 /* Check we have enough spares to not fail */
2905 spares_needed = max(reshape.before.data_disks,
2906 reshape.after.data_disks)
2907 - array.raid_disks;
2908 if ((info->new_level > 1 || info->new_level == 0) &&
2909 spares_needed > info->array.spare_disks +added_disks) {
2910 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2911 spares_needed,
2912 spares_needed == 1 ? "" : "s",
2913 info->array.spare_disks + added_disks);
2914 goto release;
2915 }
2916
2917 if (reshape.level != array.level) {
2918 int err = impose_level(fd, reshape.level, devname, verbose);
2919 if (err)
2920 goto release;
2921 info->new_layout = UnSet; /* after level change,
2922 * layout is meaningless */
2923 orig_level = array.level;
2924 sysfs_freeze_array(info);
2925
2926 if (reshape.level > 0 && st->ss->external) {
2927 /* make sure mdmon is aware of the new level */
2928 if (mdmon_running(container))
2929 flush_mdmon(container);
2930
2931 if (!mdmon_running(container))
2932 start_mdmon(container);
2933 ping_monitor(container);
2934 if (mdmon_running(container) &&
2935 st->update_tail == NULL)
2936 st->update_tail = &st->updates;
2937 }
2938 }
2939 /* ->reshape_super might have chosen some spares from the
2940 * container that it wants to be part of the new array.
2941 * We can collect them with ->container_content and give
2942 * them to the kernel.
2943 */
2944 if (st->ss->reshape_super && st->ss->container_content) {
2945 char *subarray = strchr(info->text_version+1, '/')+1;
2946 struct mdinfo *info2 =
2947 st->ss->container_content(st, subarray);
2948 struct mdinfo *d;
2949
2950 if (info2) {
2951 sysfs_init(info2, fd, st->devnm);
2952 /* When increasing number of devices, we need to set
2953 * new raid_disks before adding these, or they might
2954 * be rejected.
2955 */
2956 if (reshape.backup_blocks &&
2957 reshape.after.data_disks > reshape.before.data_disks)
2958 subarray_set_num(container, info2, "raid_disks",
2959 reshape.after.data_disks +
2960 reshape.parity);
2961 for (d = info2->devs; d; d = d->next) {
2962 if (d->disk.state == 0 &&
2963 d->disk.raid_disk >= 0) {
2964 /* This is a spare that wants to
2965 * be part of the array.
2966 */
2967 add_disk(fd, st, info2, d);
2968 }
2969 }
2970 sysfs_free(info2);
2971 }
2972 }
2973 /* We might have been given some devices to add to the
2974 * array. Now that the array has been changed to the right
2975 * level and frozen, we can safely add them.
2976 */
2977 if (devlist)
2978 Manage_subdevs(devname, fd, devlist, verbose,
2979 0,NULL, 0);
2980
2981 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
2982 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
2983 if (reshape.backup_blocks == 0) {
2984 /* No restriping needed, but we might need to impose
2985 * some more changes: layout, raid_disks, chunk_size
2986 */
2987 /* read current array info */
2988 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2989 dprintf("Cannot get array information.\n");
2990 goto release;
2991 }
2992 /* compare current array info with new values and if
2993 * it is different update them to new */
2994 if (info->new_layout != UnSet &&
2995 info->new_layout != array.layout) {
2996 array.layout = info->new_layout;
2997 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2998 pr_err("failed to set new layout\n");
2999 goto release;
3000 } else if (verbose >= 0)
3001 printf("layout for %s set to %d\n",
3002 devname, array.layout);
3003 }
3004 if (info->delta_disks != UnSet &&
3005 info->delta_disks != 0 &&
3006 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3007 array.raid_disks += info->delta_disks;
3008 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3009 pr_err("failed to set raid disks\n");
3010 goto release;
3011 } else if (verbose >= 0) {
3012 printf("raid_disks for %s set to %d\n",
3013 devname, array.raid_disks);
3014 }
3015 }
3016 if (info->new_chunk != 0 &&
3017 info->new_chunk != array.chunk_size) {
3018 if (sysfs_set_num(info, NULL,
3019 "chunk_size", info->new_chunk) != 0) {
3020 pr_err("failed to set chunk size\n");
3021 goto release;
3022 } else if (verbose >= 0)
3023 printf("chunk size for %s set to %d\n",
3024 devname, array.chunk_size);
3025 }
3026 unfreeze(st);
3027 return 0;
3028 }
3029
3030 /*
3031 * There are three possibilities.
3032 * 1/ The array will shrink.
3033 * We need to ensure the reshape will pause before reaching
3034 * the 'critical section'. We also need to fork and wait for
3035 * that to happen. When it does we
3036 * suspend/backup/complete/unfreeze
3037 *
3038 * 2/ The array will not change size.
3039 * This requires that we keep a backup of a sliding window
3040 * so that we can restore data after a crash. So we need
3041 * to fork and monitor progress.
3042 * In future we will allow the data_offset to change, so
3043 * a sliding backup becomes unnecessary.
3044 *
3045 * 3/ The array will grow. This is relatively easy.
3046 * However the kernel's restripe routines will cheerfully
3047 * overwrite some early data before it is safe. So we
3048 * need to make a backup of the early parts of the array
3049 * and be ready to restore it if rebuild aborts very early.
3050 * For externally managed metadata, we still need a forked
3051 * child to monitor the reshape and suspend IO over the region
3052 * that is being reshaped.
3053 *
3054 * We backup data by writing it to one spare, or to a
3055 * file which was given on command line.
3056 *
3057 * In each case, we first make sure that storage is available
3058 * for the required backup.
3059 * Then we:
3060 * - request the shape change.
3061 * - fork to handle backup etc.
3062 */
3063 /* Check that we can hold all the data */
3064 get_dev_size(fd, NULL, &array_size);
3065 if (reshape.new_size < (array_size/512)) {
3066 pr_err("this change will reduce the size of the array.\n"
3067 " use --grow --array-size first to truncate array.\n"
3068 " e.g. mdadm --grow %s --array-size %llu\n",
3069 devname, reshape.new_size/2);
3070 goto release;
3071 }
3072
3073 if (array.level == 10) {
3074 /* Reshaping RAID10 does not require any data backup by
3075 * user-space. Instead it requires that the data_offset
3076 * is changed to avoid the need for backup.
3077 * So this is handled very separately
3078 */
3079 if (restart)
3080 /* Nothing to do. */
3081 return 0;
3082 return raid10_reshape(container, fd, devname, st, info,
3083 &reshape, data_offset,
3084 force, verbose);
3085 }
3086 sra = sysfs_read(fd, NULL,
3087 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3088 GET_CACHE);
3089 if (!sra) {
3090 pr_err("%s: Cannot get array details from sysfs\n",
3091 devname);
3092 goto release;
3093 }
3094
3095 if (!backup_file)
3096 switch(set_new_data_offset(sra, st, devname,
3097 reshape.after.data_disks - reshape.before.data_disks,
3098 data_offset,
3099 reshape.min_offset_change, 1)) {
3100 case -1:
3101 goto release;
3102 case 0:
3103 /* Updated data_offset, so it's easy now */
3104 update_cache_size(container, sra, info,
3105 min(reshape.before.data_disks,
3106 reshape.after.data_disks),
3107 reshape.backup_blocks);
3108
3109 /* Right, everything seems fine. Let's kick things off.
3110 */
3111 sync_metadata(st);
3112
3113 if (impose_reshape(sra, info, st, fd, restart,
3114 devname, container, &reshape) < 0)
3115 goto release;
3116 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3117 struct mdinfo *sd;
3118 if (errno != EINVAL) {
3119 pr_err("Failed to initiate reshape!\n");
3120 goto release;
3121 }
3122 /* revert data_offset and try the old way */
3123 for (sd = sra->devs; sd; sd = sd->next) {
3124 sysfs_set_num(sra, sd, "new_offset",
3125 sd->data_offset);
3126 sysfs_set_str(sra, NULL, "reshape_direction",
3127 "forwards");
3128 }
3129 break;
3130 }
3131 if (info->new_level == reshape.level)
3132 return 0;
3133 /* need to adjust level when reshape completes */
3134 switch(fork()) {
3135 case -1: /* ignore error, but don't wait */
3136 return 0;
3137 default: /* parent */
3138 return 0;
3139 case 0:
3140 map_fork();
3141 break;
3142 }
3143 close(fd);
3144 wait_reshape(sra);
3145 fd = open_dev(sra->sys_name);
3146 if (fd >= 0)
3147 impose_level(fd, info->new_level, devname, verbose);
3148 return 0;
3149 case 1: /* Couldn't set data_offset, try the old way */
3150 if (data_offset != INVALID_SECTORS) {
3151 pr_err("Cannot update data_offset on this array\n");
3152 goto release;
3153 }
3154 break;
3155 }
3156
3157 started:
3158 /* Decide how many blocks (sectors) for a reshape
3159 * unit. The number we have so far is just a minimum
3160 */
3161 blocks = reshape.backup_blocks;
3162 if (reshape.before.data_disks ==
3163 reshape.after.data_disks) {
3164 /* Make 'blocks' bigger for better throughput, but
3165 * not so big that we reject it below.
3166 * Try for 16 megabytes
3167 */
3168 while (blocks * 32 < sra->component_size &&
3169 blocks < 16*1024*2)
3170 blocks *= 2;
3171 } else
3172 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3173
3174 if (blocks >= sra->component_size/2) {
3175 pr_err("%s: Something wrong - reshape aborted\n",
3176 devname);
3177 goto release;
3178 }
3179
3180 /* Now we need to open all these devices so we can read/write.
3181 */
3182 nrdisks = max(reshape.before.data_disks,
3183 reshape.after.data_disks) + reshape.parity
3184 + sra->array.spare_disks;
3185 fdlist = xcalloc((1+nrdisks), sizeof(int));
3186 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3187
3188 odisks = reshape.before.data_disks + reshape.parity;
3189 d = reshape_prepare_fdlist(devname, sra, odisks,
3190 nrdisks, blocks, backup_file,
3191 fdlist, offsets);
3192 if (d < odisks) {
3193 goto release;
3194 }
3195 if ((st->ss->manage_reshape == NULL) ||
3196 (st->ss->recover_backup == NULL)) {
3197 if (backup_file == NULL) {
3198 if (reshape.after.data_disks <=
3199 reshape.before.data_disks) {
3200 pr_err("%s: Cannot grow - need backup-file\n",
3201 devname);
3202 pr_err(" Please provide one with \"--backup=...\"\n");
3203 goto release;
3204 } else if (d == odisks) {
3205 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3206 goto release;
3207 }
3208 } else {
3209 if (!reshape_open_backup_file(backup_file, fd, devname,
3210 (signed)blocks,
3211 fdlist+d, offsets+d,
3212 sra->sys_name,
3213 restart)) {
3214 goto release;
3215 }
3216 d++;
3217 }
3218 }
3219
3220 update_cache_size(container, sra, info,
3221 min(reshape.before.data_disks, reshape.after.data_disks),
3222 blocks);
3223
3224 /* Right, everything seems fine. Let's kick things off.
3225 * If only changing raid_disks, use ioctl, else use
3226 * sysfs.
3227 */
3228 sync_metadata(st);
3229
3230 if (impose_reshape(sra, info, st, fd, restart,
3231 devname, container, &reshape) < 0)
3232 goto release;
3233
3234 err = start_reshape(sra, restart, reshape.before.data_disks,
3235 reshape.after.data_disks);
3236 if (err) {
3237 pr_err("Cannot %s reshape for %s\n",
3238 restart ? "continue" : "start",
3239 devname);
3240 goto release;
3241 }
3242 if (restart)
3243 sysfs_set_str(sra, NULL, "array_state", "active");
3244 if (freeze_reshape) {
3245 free(fdlist);
3246 free(offsets);
3247 sysfs_free(sra);
3248 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3249 sra->reshape_progress);
3250 return 1;
3251 }
3252
3253 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3254 if (continue_via_systemd(container ?: sra->sys_name)) {
3255 free(fdlist);
3256 free(offsets);
3257 sysfs_free(sra);
3258 return 0;
3259 }
3260
3261 /* Now we just need to kick off the reshape and watch, while
3262 * handling backups of the data...
3263 * This is all done by a forked background process.
3264 */
3265 switch(forked ? 0 : fork()) {
3266 case -1:
3267 pr_err("Cannot run child to monitor reshape: %s\n",
3268 strerror(errno));
3269 abort_reshape(sra);
3270 goto release;
3271 default:
3272 free(fdlist);
3273 free(offsets);
3274 sysfs_free(sra);
3275 return 0;
3276 case 0:
3277 map_fork();
3278 break;
3279 }
3280
3281 /* If another array on the same devices is busy, the
3282 * reshape will wait for them. This would mean that
3283 * the first section that we suspend will stay suspended
3284 * for a long time. So check on that possibility
3285 * by looking for "DELAYED" in /proc/mdstat, and if found,
3286 * wait a while
3287 */
3288 do {
3289 struct mdstat_ent *mds, *m;
3290 delayed = 0;
3291 mds = mdstat_read(1, 0);
3292 for (m = mds; m; m = m->next)
3293 if (strcmp(m->devnm, sra->sys_name) == 0) {
3294 if (m->resync &&
3295 m->percent == RESYNC_DELAYED)
3296 delayed = 1;
3297 if (m->resync == 0)
3298 /* Haven't started the reshape thread
3299 * yet, wait a bit
3300 */
3301 delayed = 2;
3302 break;
3303 }
3304 free_mdstat(mds);
3305 if (delayed == 1 && get_linux_version() < 3007000) {
3306 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3307 " You might experience problems until other reshapes complete.\n");
3308 delayed = 0;
3309 }
3310 if (delayed)
3311 mdstat_wait(30 - (delayed-1) * 25);
3312 } while (delayed);
3313 mdstat_close();
3314 close(fd);
3315 if (check_env("MDADM_GROW_VERIFY"))
3316 fd = open(devname, O_RDONLY | O_DIRECT);
3317 else
3318 fd = -1;
3319 mlockall(MCL_FUTURE);
3320
3321 signal(SIGTERM, catch_term);
3322
3323 if (st->ss->external) {
3324 /* metadata handler takes it from here */
3325 done = st->ss->manage_reshape(
3326 fd, sra, &reshape, st, blocks,
3327 fdlist, offsets,
3328 d - odisks, fdlist+odisks,
3329 offsets+odisks);
3330 } else
3331 done = child_monitor(
3332 fd, sra, &reshape, st, blocks,
3333 fdlist, offsets,
3334 d - odisks, fdlist+odisks,
3335 offsets+odisks);
3336
3337 free(fdlist);
3338 free(offsets);
3339
3340 if (backup_file && done) {
3341 char *bul;
3342 bul = make_backup(sra->sys_name);
3343 if (bul) {
3344 char buf[1024];
3345 int l = readlink(bul, buf, sizeof(buf) - 1);
3346 if (l > 0) {
3347 buf[l]=0;
3348 unlink(buf);
3349 }
3350 unlink(bul);
3351 free(bul);
3352 }
3353 unlink(backup_file);
3354 }
3355 if (!done) {
3356 abort_reshape(sra);
3357 goto out;
3358 }
3359
3360 if (!st->ss->external &&
3361 !(reshape.before.data_disks != reshape.after.data_disks
3362 && info->custom_array_size) &&
3363 info->new_level == reshape.level &&
3364 !forked) {
3365 /* no need to wait for the reshape to finish as
3366 * there is nothing more to do.
3367 */
3368 sysfs_free(sra);
3369 exit(0);
3370 }
3371 wait_reshape(sra);
3372
3373 if (st->ss->external) {
3374 /* Re-load the metadata as much could have changed */
3375 int cfd = open_dev(st->container_devnm);
3376 if (cfd >= 0) {
3377 flush_mdmon(container);
3378 st->ss->free_super(st);
3379 st->ss->load_container(st, cfd, container);
3380 close(cfd);
3381 }
3382 }
3383
3384 /* set new array size if required customer_array_size is used
3385 * by this metadata.
3386 */
3387 if (reshape.before.data_disks !=
3388 reshape.after.data_disks &&
3389 info->custom_array_size)
3390 set_array_size(st, info, info->text_version);
3391
3392 if (info->new_level != reshape.level) {
3393 if (fd < 0)
3394 fd = open(devname, O_RDONLY);
3395 impose_level(fd, info->new_level, devname, verbose);
3396 close(fd);
3397 if (info->new_level == 0)
3398 st->update_tail = NULL;
3399 }
3400 out:
3401 sysfs_free(sra);
3402 if (forked)
3403 return 0;
3404 unfreeze(st);
3405 exit(0);
3406
3407 release:
3408 free(fdlist);
3409 free(offsets);
3410 if (orig_level != UnSet && sra) {
3411 c = map_num(pers, orig_level);
3412 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3413 pr_err("aborting level change\n");
3414 }
3415 sysfs_free(sra);
3416 if (!forked)
3417 unfreeze(st);
3418 return 1;
3419 }
3420
3421 /* mdfd handle is passed to be closed in child process (after fork).
3422 */
3423 int reshape_container(char *container, char *devname,
3424 int mdfd,
3425 struct supertype *st,
3426 struct mdinfo *info,
3427 int force,
3428 char *backup_file, int verbose,
3429 int forked, int restart, int freeze_reshape)
3430 {
3431 struct mdinfo *cc = NULL;
3432 int rv = restart;
3433 char last_devnm[32] = "";
3434
3435 /* component_size is not meaningful for a container,
3436 * so pass '0' meaning 'no change'
3437 */
3438 if (!restart &&
3439 reshape_super(st, 0, info->new_level,
3440 info->new_layout, info->new_chunk,
3441 info->array.raid_disks, info->delta_disks,
3442 backup_file, devname, APPLY_METADATA_CHANGES,
3443 verbose)) {
3444 unfreeze(st);
3445 return 1;
3446 }
3447
3448 sync_metadata(st);
3449
3450 /* ping monitor to be sure that update is on disk
3451 */
3452 ping_monitor(container);
3453
3454 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3455 if (continue_via_systemd(container))
3456 return 0;
3457
3458 switch (forked ? 0 : fork()) {
3459 case -1: /* error */
3460 perror("Cannot fork to complete reshape\n");
3461 unfreeze(st);
3462 return 1;
3463 default: /* parent */
3464 if (!freeze_reshape)
3465 printf("%s: multi-array reshape continues in background\n", Name);
3466 return 0;
3467 case 0: /* child */
3468 map_fork();
3469 break;
3470 }
3471
3472 /* close unused handle in child process
3473 */
3474 if (mdfd > -1)
3475 close(mdfd);
3476
3477 while(1) {
3478 /* For each member array with reshape_active,
3479 * we need to perform the reshape.
3480 * We pick the first array that needs reshaping and
3481 * reshape it. reshape_array() will re-read the metadata
3482 * so the next time through a different array should be
3483 * ready for reshape.
3484 * It is possible that the 'different' array will not
3485 * be assembled yet. In that case we simple exit.
3486 * When it is assembled, the mdadm which assembles it
3487 * will take over the reshape.
3488 */
3489 struct mdinfo *content;
3490 int fd;
3491 struct mdstat_ent *mdstat;
3492 char *adev;
3493 int devid;
3494
3495 sysfs_free(cc);
3496
3497 cc = st->ss->container_content(st, NULL);
3498
3499 for (content = cc; content ; content = content->next) {
3500 char *subarray;
3501 if (!content->reshape_active)
3502 continue;
3503
3504 subarray = strchr(content->text_version+1, '/')+1;
3505 mdstat = mdstat_by_subdev(subarray, container);
3506 if (!mdstat)
3507 continue;
3508 if (mdstat->active == 0) {
3509 pr_err("Skipping inactive array %s.\n",
3510 mdstat->devnm);
3511 free_mdstat(mdstat);
3512 mdstat = NULL;
3513 continue;
3514 }
3515 break;
3516 }
3517 if (!content)
3518 break;
3519
3520 devid = devnm2devid(mdstat->devnm);
3521 adev = map_dev(major(devid), minor(devid), 0);
3522 if (!adev)
3523 adev = content->text_version;
3524
3525 fd = open_dev(mdstat->devnm);
3526 if (fd < 0) {
3527 pr_err("Device %s cannot be opened for reshape.\n", adev);
3528 break;
3529 }
3530
3531 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3532 /* Do not allow for multiple reshape_array() calls for
3533 * the same array.
3534 * It can happen when reshape_array() returns without
3535 * error, when reshape is not finished (wrong reshape
3536 * starting/continuation conditions). Mdmon doesn't
3537 * switch to next array in container and reentry
3538 * conditions for the same array occur.
3539 * This is possibly interim until the behaviour of
3540 * reshape_array is resolved().
3541 */
3542 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3543 close(fd);
3544 break;
3545 }
3546 strcpy(last_devnm, mdstat->devnm);
3547
3548 sysfs_init(content, fd, mdstat->devnm);
3549
3550 if (mdmon_running(container))
3551 flush_mdmon(container);
3552
3553 rv = reshape_array(container, fd, adev, st,
3554 content, force, NULL, INVALID_SECTORS,
3555 backup_file, verbose, 1, restart,
3556 freeze_reshape);
3557 close(fd);
3558
3559 if (freeze_reshape) {
3560 sysfs_free(cc);
3561 exit(0);
3562 }
3563
3564 restart = 0;
3565 if (rv)
3566 break;
3567
3568 if (mdmon_running(container))
3569 flush_mdmon(container);
3570 }
3571 if (!rv)
3572 unfreeze(st);
3573 sysfs_free(cc);
3574 exit(0);
3575 }
3576
3577 /*
3578 * We run a child process in the background which performs the following
3579 * steps:
3580 * - wait for resync to reach a certain point
3581 * - suspend io to the following section
3582 * - backup that section
3583 * - allow resync to proceed further
3584 * - resume io
3585 * - discard the backup.
3586 *
3587 * When are combined in slightly different ways in the three cases.
3588 * Grow:
3589 * - suspend/backup/allow/wait/resume/discard
3590 * Shrink:
3591 * - allow/wait/suspend/backup/allow/wait/resume/discard
3592 * same-size:
3593 * - wait/resume/discard/suspend/backup/allow
3594 *
3595 * suspend/backup/allow always come together
3596 * wait/resume/discard do too.
3597 * For the same-size case we have two backups to improve flow.
3598 *
3599 */
3600
3601 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3602 unsigned long long backup_point,
3603 unsigned long long wait_point,
3604 unsigned long long *suspend_point,
3605 unsigned long long *reshape_completed, int *frozen)
3606 {
3607 /* This function is called repeatedly by the reshape manager.
3608 * It determines how much progress can safely be made and allows
3609 * that progress.
3610 * - 'info' identifies the array and particularly records in
3611 * ->reshape_progress the metadata's knowledge of progress
3612 * This is a sector offset from the start of the array
3613 * of the next array block to be relocated. This number
3614 * may increase from 0 or decrease from array_size, depending
3615 * on the type of reshape that is happening.
3616 * Note that in contrast, 'sync_completed' is a block count of the
3617 * reshape so far. It gives the distance between the start point
3618 * (head or tail of device) and the next place that data will be
3619 * written. It always increases.
3620 * - 'reshape' is the structure created by analyse_change
3621 * - 'backup_point' shows how much the metadata manager has backed-up
3622 * data. For reshapes with increasing progress, it is the next address
3623 * to be backed up, previous addresses have been backed-up. For
3624 * decreasing progress, it is the earliest address that has been
3625 * backed up - later address are also backed up.
3626 * So addresses between reshape_progress and backup_point are
3627 * backed up providing those are in the 'correct' order.
3628 * - 'wait_point' is an array address. When reshape_completed
3629 * passes this point, progress_reshape should return. It might
3630 * return earlier if it determines that ->reshape_progress needs
3631 * to be updated or further backup is needed.
3632 * - suspend_point is maintained by progress_reshape and the caller
3633 * should not touch it except to initialise to zero.
3634 * It is an array address and it only increases in 2.6.37 and earlier.
3635 * This makes it difficult to handle reducing reshapes with
3636 * external metadata.
3637 * However: it is similar to backup_point in that it records the
3638 * other end of a suspended region from reshape_progress.
3639 * it is moved to extend the region that is safe to backup and/or
3640 * reshape
3641 * - reshape_completed is read from sysfs and returned. The caller
3642 * should copy this into ->reshape_progress when it has reason to
3643 * believe that the metadata knows this, and any backup outside this
3644 * has been erased.
3645 *
3646 * Return value is:
3647 * 1 if more data from backup_point - but only as far as suspend_point,
3648 * should be backed up
3649 * 0 if things are progressing smoothly
3650 * -1 if the reshape is finished because it is all done,
3651 * -2 if the reshape is finished due to an error.
3652 */
3653
3654 int advancing = (reshape->after.data_disks
3655 >= reshape->before.data_disks);
3656 unsigned long long need_backup; /* All data between start of array and
3657 * here will at some point need to
3658 * be backed up.
3659 */
3660 unsigned long long read_offset, write_offset;
3661 unsigned long long write_range;
3662 unsigned long long max_progress, target, completed;
3663 unsigned long long array_size = (info->component_size
3664 * reshape->before.data_disks);
3665 int fd;
3666 char buf[20];
3667
3668 /* First, we unsuspend any region that is now known to be safe.
3669 * If suspend_point is on the 'wrong' side of reshape_progress, then
3670 * we don't have or need suspension at the moment. This is true for
3671 * native metadata when we don't need to back-up.
3672 */
3673 if (advancing) {
3674 if (info->reshape_progress <= *suspend_point)
3675 sysfs_set_num(info, NULL, "suspend_lo",
3676 info->reshape_progress);
3677 } else {
3678 /* Note: this won't work in 2.6.37 and before.
3679 * Something somewhere should make sure we don't need it!
3680 */
3681 if (info->reshape_progress >= *suspend_point)
3682 sysfs_set_num(info, NULL, "suspend_hi",
3683 info->reshape_progress);
3684 }
3685
3686 /* Now work out how far it is safe to progress.
3687 * If the read_offset for ->reshape_progress is less than
3688 * 'blocks' beyond the write_offset, we can only progress as far
3689 * as a backup.
3690 * Otherwise we can progress until the write_offset for the new location
3691 * reaches (within 'blocks' of) the read_offset at the current location.
3692 * However that region must be suspended unless we are using native
3693 * metadata.
3694 * If we need to suspend more, we limit it to 128M per device, which is
3695 * rather arbitrary and should be some time-based calculation.
3696 */
3697 read_offset = info->reshape_progress / reshape->before.data_disks;
3698 write_offset = info->reshape_progress / reshape->after.data_disks;
3699 write_range = info->new_chunk/512;
3700 if (reshape->before.data_disks == reshape->after.data_disks)
3701 need_backup = array_size;
3702 else
3703 need_backup = reshape->backup_blocks;
3704 if (advancing) {
3705 if (read_offset < write_offset + write_range)
3706 max_progress = backup_point;
3707 else
3708 max_progress =
3709 read_offset *
3710 reshape->after.data_disks;
3711 } else {
3712 if (read_offset > write_offset - write_range)
3713 /* Can only progress as far as has been backed up,
3714 * which must be suspended */
3715 max_progress = backup_point;
3716 else if (info->reshape_progress <= need_backup)
3717 max_progress = backup_point;
3718 else {
3719 if (info->array.major_version >= 0)
3720 /* Can progress until backup is needed */
3721 max_progress = need_backup;
3722 else {
3723 /* Can progress until metadata update is required */
3724 max_progress =
3725 read_offset *
3726 reshape->after.data_disks;
3727 /* but data must be suspended */
3728 if (max_progress < *suspend_point)
3729 max_progress = *suspend_point;
3730 }
3731 }
3732 }
3733
3734 /* We know it is safe to progress to 'max_progress' providing
3735 * it is suspended or we are using native metadata.
3736 * Consider extending suspend_point 128M per device if it
3737 * is less than 64M per device beyond reshape_progress.
3738 * But always do a multiple of 'blocks'
3739 * FIXME this is too big - it takes to long to complete
3740 * this much.
3741 */
3742 target = 64*1024*2 * min(reshape->before.data_disks,
3743 reshape->after.data_disks);
3744 target /= reshape->backup_blocks;
3745 if (target < 2)
3746 target = 2;
3747 target *= reshape->backup_blocks;
3748
3749 /* For externally managed metadata we always need to suspend IO to
3750 * the area being reshaped so we regularly push suspend_point forward.
3751 * For native metadata we only need the suspend if we are going to do
3752 * a backup.
3753 */
3754 if (advancing) {
3755 if ((need_backup > info->reshape_progress
3756 || info->array.major_version < 0) &&
3757 *suspend_point < info->reshape_progress + target) {
3758 if (need_backup < *suspend_point + 2 * target)
3759 *suspend_point = need_backup;
3760 else if (*suspend_point + 2 * target < array_size)
3761 *suspend_point += 2 * target;
3762 else
3763 *suspend_point = array_size;
3764 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3765 if (max_progress > *suspend_point)
3766 max_progress = *suspend_point;
3767 }
3768 } else {
3769 if (info->array.major_version >= 0) {
3770 /* Only need to suspend when about to backup */
3771 if (info->reshape_progress < need_backup * 2 &&
3772 *suspend_point > 0) {
3773 *suspend_point = 0;
3774 sysfs_set_num(info, NULL, "suspend_lo", 0);
3775 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3776 }
3777 } else {
3778 /* Need to suspend continually */
3779 if (info->reshape_progress < *suspend_point)
3780 *suspend_point = info->reshape_progress;
3781 if (*suspend_point + target < info->reshape_progress)
3782 /* No need to move suspend region yet */;
3783 else {
3784 if (*suspend_point >= 2 * target)
3785 *suspend_point -= 2 * target;
3786 else
3787 *suspend_point = 0;
3788 sysfs_set_num(info, NULL, "suspend_lo",
3789 *suspend_point);
3790 }
3791 if (max_progress < *suspend_point)
3792 max_progress = *suspend_point;
3793 }
3794 }
3795
3796 /* now set sync_max to allow that progress. sync_max, like
3797 * sync_completed is a count of sectors written per device, so
3798 * we find the difference between max_progress and the start point,
3799 * and divide that by after.data_disks to get a sync_max
3800 * number.
3801 * At the same time we convert wait_point to a similar number
3802 * for comparing against sync_completed.
3803 */
3804 /* scale down max_progress to per_disk */
3805 max_progress /= reshape->after.data_disks;
3806 /* Round to chunk size as some kernels give an erroneously high number */
3807 max_progress /= info->new_chunk/512;
3808 max_progress *= info->new_chunk/512;
3809 /* And round to old chunk size as the kernel wants that */
3810 max_progress /= info->array.chunk_size/512;
3811 max_progress *= info->array.chunk_size/512;
3812 /* Limit progress to the whole device */
3813 if (max_progress > info->component_size)
3814 max_progress = info->component_size;
3815 wait_point /= reshape->after.data_disks;
3816 if (!advancing) {
3817 /* switch from 'device offset' to 'processed block count' */
3818 max_progress = info->component_size - max_progress;
3819 wait_point = info->component_size - wait_point;
3820 }
3821
3822 if (!*frozen)
3823 sysfs_set_num(info, NULL, "sync_max", max_progress);
3824
3825 /* Now wait. If we have already reached the point that we were
3826 * asked to wait to, don't wait at all, else wait for any change.
3827 * We need to select on 'sync_completed' as that is the place that
3828 * notifications happen, but we are really interested in
3829 * 'reshape_position'
3830 */
3831 fd = sysfs_get_fd(info, NULL, "sync_completed");
3832 if (fd < 0)
3833 goto check_progress;
3834
3835 if (sysfs_fd_get_ll(fd, &completed) < 0)
3836 goto check_progress;
3837
3838 while (completed < max_progress && completed < wait_point) {
3839 /* Check that sync_action is still 'reshape' to avoid
3840 * waiting forever on a dead array
3841 */
3842 char action[20];
3843 if (sysfs_get_str(info, NULL, "sync_action",
3844 action, 20) <= 0 ||
3845 strncmp(action, "reshape", 7) != 0)
3846 break;
3847 /* Some kernels reset 'sync_completed' to zero
3848 * before setting 'sync_action' to 'idle'.
3849 * So we need these extra tests.
3850 */
3851 if (completed == 0 && advancing
3852 && strncmp(action, "idle", 4) == 0
3853 && info->reshape_progress > 0)
3854 break;
3855 if (completed == 0 && !advancing
3856 && strncmp(action, "idle", 4) == 0
3857 && info->reshape_progress < (info->component_size
3858 * reshape->after.data_disks))
3859 break;
3860 sysfs_wait(fd, NULL);
3861 if (sysfs_fd_get_ll(fd, &completed) < 0)
3862 goto check_progress;
3863 }
3864 /* Some kernels reset 'sync_completed' to zero,
3865 * we need to have real point we are in md.
3866 * So in that case, read 'reshape_position' from sysfs.
3867 */
3868 if (completed == 0) {
3869 unsigned long long reshapep;
3870 char action[20];
3871 if (sysfs_get_str(info, NULL, "sync_action",
3872 action, 20) > 0 &&
3873 strncmp(action, "idle", 4) == 0 &&
3874 sysfs_get_ll(info, NULL,
3875 "reshape_position", &reshapep) == 0)
3876 *reshape_completed = reshapep;
3877 } else {
3878 /* some kernels can give an incorrectly high
3879 * 'completed' number, so round down */
3880 completed /= (info->new_chunk/512);
3881 completed *= (info->new_chunk/512);
3882 /* Convert 'completed' back in to a 'progress' number */
3883 completed *= reshape->after.data_disks;
3884 if (!advancing)
3885 completed = (info->component_size
3886 * reshape->after.data_disks
3887 - completed);
3888 *reshape_completed = completed;
3889 }
3890
3891 close(fd);
3892
3893 /* We return the need_backup flag. Caller will decide
3894 * how much - a multiple of ->backup_blocks up to *suspend_point
3895 */
3896 if (advancing)
3897 return need_backup > info->reshape_progress;
3898 else
3899 return need_backup >= info->reshape_progress;
3900
3901 check_progress:
3902 /* if we couldn't read a number from sync_completed, then
3903 * either the reshape did complete, or it aborted.
3904 * We can tell which by checking for 'none' in reshape_position.
3905 * If it did abort, then it might immediately restart if it
3906 * it was just a device failure that leaves us degraded but
3907 * functioning.
3908 */
3909 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3910 || strncmp(buf, "none", 4) != 0) {
3911 /* The abort might only be temporary. Wait up to 10
3912 * seconds for fd to contain a valid number again.
3913 */
3914 int wait = 10000;
3915 int rv = -2;
3916 unsigned long long new_sync_max;
3917 while (fd >= 0 && rv < 0 && wait > 0) {
3918 if (sysfs_wait(fd, &wait) != 1)
3919 break;
3920 switch (sysfs_fd_get_ll(fd, &completed)) {
3921 case 0:
3922 /* all good again */
3923 rv = 1;
3924 /* If "sync_max" is no longer max_progress
3925 * we need to freeze things
3926 */
3927 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3928 *frozen = (new_sync_max != max_progress);
3929 break;
3930 case -2: /* read error - abort */
3931 wait = 0;
3932 break;
3933 }
3934 }
3935 if (fd >= 0)
3936 close(fd);
3937 return rv; /* abort */
3938 } else {
3939 /* Maybe racing with array shutdown - check state */
3940 if (fd >= 0)
3941 close(fd);
3942 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3943 || strncmp(buf, "inactive", 8) == 0
3944 || strncmp(buf, "clear",5) == 0)
3945 return -2; /* abort */
3946 return -1; /* complete */
3947 }
3948 }
3949
3950 /* FIXME return status is never checked */
3951 static int grow_backup(struct mdinfo *sra,
3952 unsigned long long offset, /* per device */
3953 unsigned long stripes, /* per device, in old chunks */
3954 int *sources, unsigned long long *offsets,
3955 int disks, int chunk, int level, int layout,
3956 int dests, int *destfd, unsigned long long *destoffsets,
3957 int part, int *degraded,
3958 char *buf)
3959 {
3960 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3961 * to storage 'destfd' (offset 'destoffsets'), after first
3962 * suspending IO. Then allow resync to continue
3963 * over the suspended section.
3964 * Use part 'part' of the backup-super-block.
3965 */
3966 int odata = disks;
3967 int rv = 0;
3968 int i;
3969 unsigned long long ll;
3970 int new_degraded;
3971 //printf("offset %llu\n", offset);
3972 if (level >= 4)
3973 odata--;
3974 if (level == 6)
3975 odata--;
3976
3977 /* Check that array hasn't become degraded, else we might backup the wrong data */
3978 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3979 return -1; /* FIXME this error is ignored */
3980 new_degraded = (int)ll;
3981 if (new_degraded != *degraded) {
3982 /* check each device to ensure it is still working */
3983 struct mdinfo *sd;
3984 for (sd = sra->devs ; sd ; sd = sd->next) {
3985 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3986 continue;
3987 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3988 char sbuf[20];
3989 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3990 strstr(sbuf, "faulty") ||
3991 strstr(sbuf, "in_sync") == NULL) {
3992 /* this device is dead */
3993 sd->disk.state = (1<<MD_DISK_FAULTY);
3994 if (sd->disk.raid_disk >= 0 &&
3995 sources[sd->disk.raid_disk] >= 0) {
3996 close(sources[sd->disk.raid_disk]);
3997 sources[sd->disk.raid_disk] = -1;
3998 }
3999 }
4000 }
4001 }
4002 *degraded = new_degraded;
4003 }
4004 if (part) {
4005 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4006 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4007 } else {
4008 bsb.arraystart = __cpu_to_le64(offset * odata);
4009 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4010 }
4011 if (part)
4012 bsb.magic[15] = '2';
4013 for (i = 0; i < dests; i++)
4014 if (part)
4015 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4016 else
4017 lseek64(destfd[i], destoffsets[i], 0);
4018
4019 rv = save_stripes(sources, offsets,
4020 disks, chunk, level, layout,
4021 dests, destfd,
4022 offset*512*odata, stripes * chunk * odata,
4023 buf);
4024
4025 if (rv)
4026 return rv;
4027 bsb.mtime = __cpu_to_le64(time(0));
4028 for (i = 0; i < dests; i++) {
4029 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4030
4031 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4032 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4033 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4034 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4035
4036 rv = -1;
4037 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4038 != destoffsets[i] - 4096)
4039 break;
4040 if (write(destfd[i], &bsb, 512) != 512)
4041 break;
4042 if (destoffsets[i] > 4096) {
4043 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4044 destoffsets[i]+stripes*chunk*odata)
4045 break;
4046 if (write(destfd[i], &bsb, 512) != 512)
4047 break;
4048 }
4049 fsync(destfd[i]);
4050 rv = 0;
4051 }
4052
4053 return rv;
4054 }
4055
4056 /* in 2.6.30, the value reported by sync_completed can be
4057 * less that it should be by one stripe.
4058 * This only happens when reshape hits sync_max and pauses.
4059 * So allow wait_backup to either extent sync_max further
4060 * than strictly necessary, or return before the
4061 * sync has got quite as far as we would really like.
4062 * This is what 'blocks2' is for.
4063 * The various caller give appropriate values so that
4064 * every works.
4065 */
4066 /* FIXME return value is often ignored */
4067 static int forget_backup(int dests, int *destfd,
4068 unsigned long long *destoffsets,
4069 int part)
4070 {
4071 /*
4072 * Erase backup 'part' (which is 0 or 1)
4073 */
4074 int i;
4075 int rv;
4076
4077 if (part) {
4078 bsb.arraystart2 = __cpu_to_le64(0);
4079 bsb.length2 = __cpu_to_le64(0);
4080 } else {
4081 bsb.arraystart = __cpu_to_le64(0);
4082 bsb.length = __cpu_to_le64(0);
4083 }
4084 bsb.mtime = __cpu_to_le64(time(0));
4085 rv = 0;
4086 for (i = 0; i < dests; i++) {
4087 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4088 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4089 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4090 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4091 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4092 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4093 destoffsets[i]-4096)
4094 rv = -1;
4095 if (rv == 0 &&
4096 write(destfd[i], &bsb, 512) != 512)
4097 rv = -1;
4098 fsync(destfd[i]);
4099 }
4100 return rv;
4101 }
4102
4103 static void fail(char *msg)
4104 {
4105 int rv;
4106 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4107 rv |= (write(2, "\n", 1) != 1);
4108 exit(rv ? 1 : 2);
4109 }
4110
4111 static char *abuf, *bbuf;
4112 static unsigned long long abuflen;
4113 static void validate(int afd, int bfd, unsigned long long offset)
4114 {
4115 /* check that the data in the backup against the array.
4116 * This is only used for regression testing and should not
4117 * be used while the array is active
4118 */
4119 if (afd < 0)
4120 return;
4121 lseek64(bfd, offset - 4096, 0);
4122 if (read(bfd, &bsb2, 512) != 512)
4123 fail("cannot read bsb");
4124 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4125 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4126 fail("first csum bad");
4127 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4128 fail("magic is bad");
4129 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4130 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4131 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4132 fail("second csum bad");
4133
4134 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4135 fail("devstart is wrong");
4136
4137 if (bsb2.length) {
4138 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4139
4140 if (abuflen < len) {
4141 free(abuf);
4142 free(bbuf);
4143 abuflen = len;
4144 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4145 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4146 abuflen = 0;
4147 /* just stop validating on mem-alloc failure */
4148 return;
4149 }
4150 }
4151
4152 lseek64(bfd, offset, 0);
4153 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4154 //printf("len %llu\n", len);
4155 fail("read first backup failed");
4156 }
4157 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4158 if ((unsigned long long)read(afd, abuf, len) != len)
4159 fail("read first from array failed");
4160 if (memcmp(bbuf, abuf, len) != 0) {
4161 #if 0
4162 int i;
4163 printf("offset=%llu len=%llu\n",
4164 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4165 for (i=0; i<len; i++)
4166 if (bbuf[i] != abuf[i]) {
4167 printf("first diff byte %d\n", i);
4168 break;
4169 }
4170 #endif
4171 fail("data1 compare failed");
4172 }
4173 }
4174 if (bsb2.length2) {
4175 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4176
4177 if (abuflen < len) {
4178 free(abuf);
4179 free(bbuf);
4180 abuflen = len;
4181 abuf = xmalloc(abuflen);
4182 bbuf = xmalloc(abuflen);
4183 }
4184
4185 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4186 if ((unsigned long long)read(bfd, bbuf, len) != len)
4187 fail("read second backup failed");
4188 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4189 if ((unsigned long long)read(afd, abuf, len) != len)
4190 fail("read second from array failed");
4191 if (memcmp(bbuf, abuf, len) != 0)
4192 fail("data2 compare failed");
4193 }
4194 }
4195
4196 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4197 struct supertype *st, unsigned long blocks,
4198 int *fds, unsigned long long *offsets,
4199 int dests, int *destfd, unsigned long long *destoffsets)
4200 {
4201 /* Monitor a reshape where backup is being performed using
4202 * 'native' mechanism - either to a backup file, or
4203 * to some space in a spare.
4204 */
4205 char *buf;
4206 int degraded = -1;
4207 unsigned long long speed;
4208 unsigned long long suspend_point, array_size;
4209 unsigned long long backup_point, wait_point;
4210 unsigned long long reshape_completed;
4211 int done = 0;
4212 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4213 int part = 0; /* The next part of the backup area to fill. It may already
4214 * be full, so we need to check */
4215 int level = reshape->level;
4216 int layout = reshape->before.layout;
4217 int data = reshape->before.data_disks;
4218 int disks = reshape->before.data_disks + reshape->parity;
4219 int chunk = sra->array.chunk_size;
4220 struct mdinfo *sd;
4221 unsigned long stripes;
4222 int uuid[4];
4223 int frozen = 0;
4224
4225 /* set up the backup-super-block. This requires the
4226 * uuid from the array.
4227 */
4228 /* Find a superblock */
4229 for (sd = sra->devs; sd; sd = sd->next) {
4230 char *dn;
4231 int devfd;
4232 int ok;
4233 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4234 continue;
4235 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4236 devfd = dev_open(dn, O_RDONLY);
4237 if (devfd < 0)
4238 continue;
4239 ok = st->ss->load_super(st, devfd, NULL);
4240 close(devfd);
4241 if (ok == 0)
4242 break;
4243 }
4244 if (!sd) {
4245 pr_err("Cannot find a superblock\n");
4246 return 0;
4247 }
4248
4249 memset(&bsb, 0, 512);
4250 memcpy(bsb.magic, "md_backup_data-1", 16);
4251 st->ss->uuid_from_super(st, uuid);
4252 memcpy(bsb.set_uuid, uuid, 16);
4253 bsb.mtime = __cpu_to_le64(time(0));
4254 bsb.devstart2 = blocks;
4255
4256 stripes = blocks / (sra->array.chunk_size/512) /
4257 reshape->before.data_disks;
4258
4259 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4260 /* Don't start the 'reshape' */
4261 return 0;
4262 if (reshape->before.data_disks == reshape->after.data_disks) {
4263 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4264 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4265 }
4266
4267 if (increasing) {
4268 array_size = sra->component_size * reshape->after.data_disks;
4269 backup_point = sra->reshape_progress;
4270 suspend_point = 0;
4271 } else {
4272 array_size = sra->component_size * reshape->before.data_disks;
4273 backup_point = reshape->backup_blocks;
4274 suspend_point = array_size;
4275 }
4276
4277 while (!done) {
4278 int rv;
4279
4280 /* Want to return as soon the oldest backup slot can
4281 * be released as that allows us to start backing up
4282 * some more, providing suspend_point has been
4283 * advanced, which it should have.
4284 */
4285 if (increasing) {
4286 wait_point = array_size;
4287 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4288 wait_point = (__le64_to_cpu(bsb.arraystart) +
4289 __le64_to_cpu(bsb.length));
4290 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4291 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4292 __le64_to_cpu(bsb.length2));
4293 } else {
4294 wait_point = 0;
4295 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4296 wait_point = __le64_to_cpu(bsb.arraystart);
4297 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4298 wait_point = __le64_to_cpu(bsb.arraystart2);
4299 }
4300
4301 reshape_completed = sra->reshape_progress;
4302 rv = progress_reshape(sra, reshape,
4303 backup_point, wait_point,
4304 &suspend_point, &reshape_completed,
4305 &frozen);
4306 /* external metadata would need to ping_monitor here */
4307 sra->reshape_progress = reshape_completed;
4308
4309 /* Clear any backup region that is before 'here' */
4310 if (increasing) {
4311 if (__le64_to_cpu(bsb.length) > 0 &&
4312 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4313 __le64_to_cpu(bsb.length)))
4314 forget_backup(dests, destfd,
4315 destoffsets, 0);
4316 if (__le64_to_cpu(bsb.length2) > 0 &&
4317 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4318 __le64_to_cpu(bsb.length2)))
4319 forget_backup(dests, destfd,
4320 destoffsets, 1);
4321 } else {
4322 if (__le64_to_cpu(bsb.length) > 0 &&
4323 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4324 forget_backup(dests, destfd,
4325 destoffsets, 0);
4326 if (__le64_to_cpu(bsb.length2) > 0 &&
4327 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4328 forget_backup(dests, destfd,
4329 destoffsets, 1);
4330 }
4331 if (sigterm)
4332 rv = -2;
4333 if (rv < 0) {
4334 if (rv == -1)
4335 done = 1;
4336 break;
4337 }
4338 if (rv == 0 && increasing && !st->ss->external) {
4339 /* No longer need to monitor this reshape */
4340 sysfs_set_str(sra, NULL, "sync_max", "max");
4341 done = 1;
4342 break;
4343 }
4344
4345 while (rv) {
4346 unsigned long long offset;
4347 unsigned long actual_stripes;
4348 /* Need to backup some data.
4349 * If 'part' is not used and the desired
4350 * backup size is suspended, do a backup,
4351 * then consider the next part.
4352 */
4353 /* Check that 'part' is unused */
4354 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4355 break;
4356 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4357 break;
4358
4359 offset = backup_point / data;
4360 actual_stripes = stripes;
4361 if (increasing) {
4362 if (offset + actual_stripes * (chunk/512) >
4363 sra->component_size)
4364 actual_stripes = ((sra->component_size - offset)
4365 / (chunk/512));
4366 if (offset + actual_stripes * (chunk/512) >
4367 suspend_point/data)
4368 break;
4369 } else {
4370 if (offset < actual_stripes * (chunk/512))
4371 actual_stripes = offset / (chunk/512);
4372 offset -= actual_stripes * (chunk/512);
4373 if (offset < suspend_point/data)
4374 break;
4375 }
4376 if (actual_stripes == 0)
4377 break;
4378 grow_backup(sra, offset, actual_stripes,
4379 fds, offsets,
4380 disks, chunk, level, layout,
4381 dests, destfd, destoffsets,
4382 part, &degraded, buf);
4383 validate(afd, destfd[0], destoffsets[0]);
4384 /* record where 'part' is up to */
4385 part = !part;
4386 if (increasing)
4387 backup_point += actual_stripes * (chunk/512) * data;
4388 else
4389 backup_point -= actual_stripes * (chunk/512) * data;
4390 }
4391 }
4392
4393 /* FIXME maybe call progress_reshape one more time instead */
4394 /* remove any remaining suspension */
4395 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4396 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4397 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4398 sysfs_set_num(sra, NULL, "sync_min", 0);
4399
4400 if (reshape->before.data_disks == reshape->after.data_disks)
4401 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4402 free(buf);
4403 return done;
4404 }
4405
4406 /*
4407 * If any spare contains md_back_data-1 which is recent wrt mtime,
4408 * write that data into the array and update the super blocks with
4409 * the new reshape_progress
4410 */
4411 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4412 char *backup_file, int verbose)
4413 {
4414 int i, j;
4415 int old_disks;
4416 unsigned long long *offsets;
4417 unsigned long long nstripe, ostripe;
4418 int ndata, odata;
4419
4420 odata = info->array.raid_disks - info->delta_disks - 1;
4421 if (info->array.level == 6) odata--; /* number of data disks */
4422 ndata = info->array.raid_disks - 1;
4423 if (info->new_level == 6) ndata--;
4424
4425 old_disks = info->array.raid_disks - info->delta_disks;
4426
4427 if (info->delta_disks <= 0)
4428 /* Didn't grow, so the backup file must have
4429 * been used
4430 */
4431 old_disks = cnt;
4432 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4433 struct mdinfo dinfo;
4434 int fd;
4435 int bsbsize;
4436 char *devname, namebuf[20];
4437 unsigned long long lo, hi;
4438
4439 /* This was a spare and may have some saved data on it.
4440 * Load the superblock, find and load the
4441 * backup_super_block.
4442 * If either fail, go on to next device.
4443 * If the backup contains no new info, just return
4444 * else restore data and update all superblocks
4445 */
4446 if (i == old_disks-1) {
4447 fd = open(backup_file, O_RDONLY);
4448 if (fd<0) {
4449 pr_err("backup file %s inaccessible: %s\n",
4450 backup_file, strerror(errno));
4451 continue;
4452 }
4453 devname = backup_file;
4454 } else {
4455 fd = fdlist[i];
4456 if (fd < 0)
4457 continue;
4458 if (st->ss->load_super(st, fd, NULL))
4459 continue;
4460
4461 st->ss->getinfo_super(st, &dinfo, NULL);
4462 st->ss->free_super(st);
4463
4464 if (lseek64(fd,
4465 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4466 0) < 0) {
4467 pr_err("Cannot seek on device %d\n", i);
4468 continue; /* Cannot seek */
4469 }
4470 sprintf(namebuf, "device-%d", i);
4471 devname = namebuf;
4472 }
4473 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4474 if (verbose)
4475 pr_err("Cannot read from %s\n", devname);
4476 continue; /* Cannot read */
4477 }
4478 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4479 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4480 if (verbose)
4481 pr_err("No backup metadata on %s\n", devname);
4482 continue;
4483 }
4484 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4485 if (verbose)
4486 pr_err("Bad backup-metadata checksum on %s\n", devname);
4487 continue; /* bad checksum */
4488 }
4489 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4490 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4491 if (verbose)
4492 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4493 continue; /* Bad second checksum */
4494 }
4495 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4496 if (verbose)
4497 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4498 continue; /* Wrong uuid */
4499 }
4500
4501 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4502 * sometimes they aren't... So allow considerable flexability in matching, and allow
4503 * this test to be overridden by an environment variable.
4504 */
4505 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4506 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4507 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4508 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4509 (unsigned long)__le64_to_cpu(bsb.mtime),
4510 (unsigned long)info->array.utime);
4511 } else {
4512 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4513 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4514 continue; /* time stamp is too bad */
4515 }
4516 }
4517
4518 if (bsb.magic[15] == '1') {
4519 if (bsb.length == 0)
4520 continue;
4521 if (info->delta_disks >= 0) {
4522 /* reshape_progress is increasing */
4523 if (__le64_to_cpu(bsb.arraystart)
4524 + __le64_to_cpu(bsb.length)
4525 < info->reshape_progress) {
4526 nonew:
4527 if (verbose)
4528 pr_err("backup-metadata found on %s but is not needed\n", devname);
4529 continue; /* No new data here */
4530 }
4531 } else {
4532 /* reshape_progress is decreasing */
4533 if (__le64_to_cpu(bsb.arraystart) >=
4534 info->reshape_progress)
4535 goto nonew; /* No new data here */
4536 }
4537 } else {
4538 if (bsb.length == 0 && bsb.length2 == 0)
4539 continue;
4540 if (info->delta_disks >= 0) {
4541 /* reshape_progress is increasing */
4542 if ((__le64_to_cpu(bsb.arraystart)
4543 + __le64_to_cpu(bsb.length)
4544 < info->reshape_progress)
4545 &&
4546 (__le64_to_cpu(bsb.arraystart2)
4547 + __le64_to_cpu(bsb.length2)
4548 < info->reshape_progress))
4549 goto nonew; /* No new data here */
4550 } else {
4551 /* reshape_progress is decreasing */
4552 if (__le64_to_cpu(bsb.arraystart) >=
4553 info->reshape_progress &&
4554 __le64_to_cpu(bsb.arraystart2) >=
4555 info->reshape_progress)
4556 goto nonew; /* No new data here */
4557 }
4558 }
4559 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4560 second_fail:
4561 if (verbose)
4562 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4563 devname);
4564 continue; /* Cannot seek */
4565 }
4566 /* There should be a duplicate backup superblock 4k before here */
4567 if (lseek64(fd, -4096, 1) < 0 ||
4568 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4569 goto second_fail; /* Cannot find leading superblock */
4570 if (bsb.magic[15] == '1')
4571 bsbsize = offsetof(struct mdp_backup_super, pad1);
4572 else
4573 bsbsize = offsetof(struct mdp_backup_super, pad);
4574 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4575 goto second_fail; /* Cannot find leading superblock */
4576
4577 /* Now need the data offsets for all devices. */
4578 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4579 for(j=0; j<info->array.raid_disks; j++) {
4580 if (fdlist[j] < 0)
4581 continue;
4582 if (st->ss->load_super(st, fdlist[j], NULL))
4583 /* FIXME should be this be an error */
4584 continue;
4585 st->ss->getinfo_super(st, &dinfo, NULL);
4586 st->ss->free_super(st);
4587 offsets[j] = dinfo.data_offset * 512;
4588 }
4589 printf("%s: restoring critical section\n", Name);
4590
4591 if (restore_stripes(fdlist, offsets,
4592 info->array.raid_disks,
4593 info->new_chunk,
4594 info->new_level,
4595 info->new_layout,
4596 fd, __le64_to_cpu(bsb.devstart)*512,
4597 __le64_to_cpu(bsb.arraystart)*512,
4598 __le64_to_cpu(bsb.length)*512, NULL)) {
4599 /* didn't succeed, so giveup */
4600 if (verbose)
4601 pr_err("Error restoring backup from %s\n",
4602 devname);
4603 free(offsets);
4604 return 1;
4605 }
4606
4607 if (bsb.magic[15] == '2' &&
4608 restore_stripes(fdlist, offsets,
4609 info->array.raid_disks,
4610 info->new_chunk,
4611 info->new_level,
4612 info->new_layout,
4613 fd, __le64_to_cpu(bsb.devstart)*512 +
4614 __le64_to_cpu(bsb.devstart2)*512,
4615 __le64_to_cpu(bsb.arraystart2)*512,
4616 __le64_to_cpu(bsb.length2)*512, NULL)) {
4617 /* didn't succeed, so giveup */
4618 if (verbose)
4619 pr_err("Error restoring second backup from %s\n",
4620 devname);
4621 free(offsets);
4622 return 1;
4623 }
4624
4625 free(offsets);
4626
4627 /* Ok, so the data is restored. Let's update those superblocks. */
4628
4629 lo = hi = 0;
4630 if (bsb.length) {
4631 lo = __le64_to_cpu(bsb.arraystart);
4632 hi = lo + __le64_to_cpu(bsb.length);
4633 }
4634 if (bsb.magic[15] == '2' && bsb.length2) {
4635 unsigned long long lo1, hi1;
4636 lo1 = __le64_to_cpu(bsb.arraystart2);
4637 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4638 if (lo == hi) {
4639 lo = lo1;
4640 hi = hi1;
4641 } else if (lo < lo1)
4642 hi = hi1;
4643 else
4644 lo = lo1;
4645 }
4646 if (lo < hi &&
4647 (info->reshape_progress < lo ||
4648 info->reshape_progress > hi))
4649 /* backup does not affect reshape_progress*/ ;
4650 else if (info->delta_disks >= 0) {
4651 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4652 __le64_to_cpu(bsb.length);
4653 if (bsb.magic[15] == '2') {
4654 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4655 __le64_to_cpu(bsb.length2);
4656 if (p2 > info->reshape_progress)
4657 info->reshape_progress = p2;
4658 }
4659 } else {
4660 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4661 if (bsb.magic[15] == '2') {
4662 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4663 if (p2 < info->reshape_progress)
4664 info->reshape_progress = p2;
4665 }
4666 }
4667 for (j=0; j<info->array.raid_disks; j++) {
4668 if (fdlist[j] < 0)
4669 continue;
4670 if (st->ss->load_super(st, fdlist[j], NULL))
4671 continue;
4672 st->ss->getinfo_super(st, &dinfo, NULL);
4673 dinfo.reshape_progress = info->reshape_progress;
4674 st->ss->update_super(st, &dinfo,
4675 "_reshape_progress",
4676 NULL,0, 0, NULL);
4677 st->ss->store_super(st, fdlist[j]);
4678 st->ss->free_super(st);
4679 }
4680 return 0;
4681 }
4682 /* Didn't find any backup data, try to see if any
4683 * was needed.
4684 */
4685 if (info->delta_disks < 0) {
4686 /* When shrinking, the critical section is at the end.
4687 * So see if we are before the critical section.
4688 */
4689 unsigned long long first_block;
4690 nstripe = ostripe = 0;
4691 first_block = 0;
4692 while (ostripe >= nstripe) {
4693 ostripe += info->array.chunk_size / 512;
4694 first_block = ostripe * odata;
4695 nstripe = first_block / ndata / (info->new_chunk/512) *
4696 (info->new_chunk/512);
4697 }
4698
4699 if (info->reshape_progress >= first_block)
4700 return 0;
4701 }
4702 if (info->delta_disks > 0) {
4703 /* See if we are beyond the critical section. */
4704 unsigned long long last_block;
4705 nstripe = ostripe = 0;
4706 last_block = 0;
4707 while (nstripe >= ostripe) {
4708 nstripe += info->new_chunk / 512;
4709 last_block = nstripe * ndata;
4710 ostripe = last_block / odata / (info->array.chunk_size/512) *
4711 (info->array.chunk_size/512);
4712 }
4713
4714 if (info->reshape_progress >= last_block)
4715 return 0;
4716 }
4717 /* needed to recover critical section! */
4718 if (verbose)
4719 pr_err("Failed to find backup of critical section\n");
4720 return 1;
4721 }
4722
4723 int Grow_continue_command(char *devname, int fd,
4724 char *backup_file, int verbose)
4725 {
4726 int ret_val = 0;
4727 struct supertype *st = NULL;
4728 struct mdinfo *content = NULL;
4729 struct mdinfo array;
4730 char *subarray = NULL;
4731 struct mdinfo *cc = NULL;
4732 struct mdstat_ent *mdstat = NULL;
4733 int cfd = -1;
4734 int fd2 = -1;
4735
4736 dprintf("Grow continue from command line called for %s\n",
4737 devname);
4738
4739 st = super_by_fd(fd, &subarray);
4740 if (!st || !st->ss) {
4741 pr_err("Unable to determine metadata format for %s\n",
4742 devname);
4743 return 1;
4744 }
4745 dprintf("Grow continue is run for ");
4746 if (st->ss->external == 0) {
4747 int d;
4748 dprintf_cont("native array (%s)\n", devname);
4749 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4750 pr_err("%s is not an active md array - aborting\n", devname);
4751 ret_val = 1;
4752 goto Grow_continue_command_exit;
4753 }
4754 content = &array;
4755 /* Need to load a superblock.
4756 * FIXME we should really get what we need from
4757 * sysfs
4758 */
4759 for (d = 0; d < MAX_DISKS; d++) {
4760 mdu_disk_info_t disk;
4761 char *dv;
4762 int err;
4763 disk.number = d;
4764 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4765 continue;
4766 if (disk.major == 0 && disk.minor == 0)
4767 continue;
4768 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4769 continue;
4770 dv = map_dev(disk.major, disk.minor, 1);
4771 if (!dv)
4772 continue;
4773 fd2 = dev_open(dv, O_RDONLY);
4774 if (fd2 < 0)
4775 continue;
4776 err = st->ss->load_super(st, fd2, NULL);
4777 close(fd2);
4778 /* invalidate fd2 to avoid possible double close() */
4779 fd2 = -1;
4780 if (err)
4781 continue;
4782 break;
4783 }
4784 if (d == MAX_DISKS) {
4785 pr_err("Unable to load metadata for %s\n",
4786 devname);
4787 ret_val = 1;
4788 goto Grow_continue_command_exit;
4789 }
4790 st->ss->getinfo_super(st, content, NULL);
4791 } else {
4792 char *container;
4793
4794 if (subarray) {
4795 dprintf_cont("subarray (%s)\n", subarray);
4796 container = st->container_devnm;
4797 cfd = open_dev_excl(st->container_devnm);
4798 } else {
4799 container = st->devnm;
4800 close(fd);
4801 cfd = open_dev_excl(st->devnm);
4802 dprintf_cont("container (%s)\n", container);
4803 fd = cfd;
4804 }
4805 if (cfd < 0) {
4806 pr_err("Unable to open container for %s\n", devname);
4807 ret_val = 1;
4808 goto Grow_continue_command_exit;
4809 }
4810
4811 /* find in container array under reshape
4812 */
4813 ret_val = st->ss->load_container(st, cfd, NULL);
4814 if (ret_val) {
4815 pr_err("Cannot read superblock for %s\n",
4816 devname);
4817 ret_val = 1;
4818 goto Grow_continue_command_exit;
4819 }
4820
4821 cc = st->ss->container_content(st, subarray);
4822 for (content = cc; content ; content = content->next) {
4823 char *array;
4824 int allow_reshape = 1;
4825
4826 if (content->reshape_active == 0)
4827 continue;
4828 /* The decision about array or container wide
4829 * reshape is taken in Grow_continue based
4830 * content->reshape_active state, therefore we
4831 * need to check_reshape based on
4832 * reshape_active and subarray name
4833 */
4834 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4835 allow_reshape = 0;
4836 if (content->reshape_active == CONTAINER_RESHAPE &&
4837 (content->array.state
4838 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4839 allow_reshape = 0;
4840
4841 if (!allow_reshape) {
4842 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4843 devname, container);
4844 ret_val = 1;
4845 goto Grow_continue_command_exit;
4846 }
4847
4848 array = strchr(content->text_version+1, '/')+1;
4849 mdstat = mdstat_by_subdev(array, container);
4850 if (!mdstat)
4851 continue;
4852 if (mdstat->active == 0) {
4853 pr_err("Skipping inactive array %s.\n",
4854 mdstat->devnm);
4855 free_mdstat(mdstat);
4856 mdstat = NULL;
4857 continue;
4858 }
4859 break;
4860 }
4861 if (!content) {
4862 pr_err("Unable to determine reshaped array for %s\n", devname);
4863 ret_val = 1;
4864 goto Grow_continue_command_exit;
4865 }
4866 fd2 = open_dev(mdstat->devnm);
4867 if (fd2 < 0) {
4868 pr_err("cannot open (%s)\n", mdstat->devnm);
4869 ret_val = 1;
4870 goto Grow_continue_command_exit;
4871 }
4872
4873 sysfs_init(content, fd2, mdstat->devnm);
4874
4875 /* start mdmon in case it is not running
4876 */
4877 if (!mdmon_running(container))
4878 start_mdmon(container);
4879 ping_monitor(container);
4880
4881 if (mdmon_running(container))
4882 st->update_tail = &st->updates;
4883 else {
4884 pr_err("No mdmon found. Grow cannot continue.\n");
4885 ret_val = 1;
4886 goto Grow_continue_command_exit;
4887 }
4888 }
4889
4890 /* verify that array under reshape is started from
4891 * correct position
4892 */
4893 if (verify_reshape_position(content, content->array.level) < 0) {
4894 ret_val = 1;
4895 goto Grow_continue_command_exit;
4896 }
4897
4898 /* continue reshape
4899 */
4900 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4901
4902 Grow_continue_command_exit:
4903 if (fd2 > -1)
4904 close(fd2);
4905 if (cfd > -1)
4906 close(cfd);
4907 st->ss->free_super(st);
4908 free_mdstat(mdstat);
4909 sysfs_free(cc);
4910 free(subarray);
4911
4912 return ret_val;
4913 }
4914
4915 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4916 char *backup_file, int forked, int freeze_reshape)
4917 {
4918 int ret_val = 2;
4919
4920 if (!info->reshape_active)
4921 return ret_val;
4922
4923 if (st->ss->external) {
4924 int cfd = open_dev(st->container_devnm);
4925
4926 if (cfd < 0)
4927 return 1;
4928
4929 st->ss->load_container(st, cfd, st->container_devnm);
4930 close(cfd);
4931 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4932 st, info, 0, backup_file,
4933 0, forked,
4934 1 | info->reshape_active,
4935 freeze_reshape);
4936 } else
4937 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4938 NULL, INVALID_SECTORS,
4939 backup_file, 0, forked,
4940 1 | info->reshape_active,
4941 freeze_reshape);
4942
4943 return ret_val;
4944 }
4945
4946 char *make_backup(char *name)
4947 {
4948 char *base = "backup_file-";
4949 int len;
4950 char *fname;
4951
4952 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4953 fname = xmalloc(len);
4954 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4955 return fname;
4956 }
4957
4958 char *locate_backup(char *name)
4959 {
4960 char *fl = make_backup(name);
4961 struct stat stb;
4962
4963 if (stat(fl, &stb) == 0 &&
4964 S_ISREG(stb.st_mode))
4965 return fl;
4966
4967 free(fl);
4968 return NULL;
4969 }
Unnamed repository; edit this file 'description' to name the repository.