]> 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
config: new option to suppress adding bad block lists.
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27 #include <stddef.h>
28 #include <stdint.h>
29 #include <signal.h>
30 #include <sys/wait.h>
31
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
34 #endif
35 #include "md_u.h"
36 #include "md_p.h"
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char **backup_filep,
43 int verbose)
44 {
45 int i;
46 int *fdlist;
47 struct mdinfo *dev;
48 int err;
49 int disk_count = next_spare + working_disks;
50 char *backup_file = *backup_filep;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61 sprintf(buf, "%d:%d",
62 dev->disk.major,
63 dev->disk.minor);
64 fd = dev_open(buf, O_RDWR);
65
66 if (dev->disk.raid_disk >= 0)
67 fdlist[dev->disk.raid_disk] = fd;
68 else
69 fdlist[next_spare++] = fd;
70 }
71
72 if (!backup_file) {
73 backup_file = locate_backup(content->sys_name);
74 *backup_filep = backup_file;
75 }
76
77 if (st->ss->external && st->ss->recover_backup)
78 err = st->ss->recover_backup(st, content);
79 else
80 err = Grow_restart(st, content, fdlist, next_spare,
81 backup_file, verbose > 0);
82
83 while (next_spare > 0) {
84 next_spare--;
85 if (fdlist[next_spare] >= 0)
86 close(fdlist[next_spare]);
87 }
88 free(fdlist);
89 if (err) {
90 pr_err("Failed to restore critical"
91 " section for reshape - sorry.\n");
92 if (!backup_file)
93 pr_err("Possibly you need"
94 " to specify a --backup-file\n");
95 return 1;
96 }
97
98 dprintf("restore_backup() returns status OK.\n");
99 return 0;
100 }
101
102 int Grow_Add_device(char *devname, int fd, char *newdev)
103 {
104 /* Add a device to an active array.
105 * Currently, just extend a linear array.
106 * This requires writing a new superblock on the
107 * new device, calling the kernel to add the device,
108 * and if that succeeds, update the superblock on
109 * all other devices.
110 * This means that we need to *find* all other devices.
111 */
112 struct mdinfo info;
113
114 struct stat stb;
115 int nfd, fd2;
116 int d, nd;
117 struct supertype *st = NULL;
118 char *subarray = NULL;
119
120 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
121 pr_err("cannot get array info for %s\n", devname);
122 return 1;
123 }
124
125 if (info.array.level != -1) {
126 pr_err("can only add devices to linear arrays\n");
127 return 1;
128 }
129
130 st = super_by_fd(fd, &subarray);
131 if (!st) {
132 pr_err("cannot handle arrays with superblock version %d\n",
133 info.array.major_version);
134 return 1;
135 }
136
137 if (subarray) {
138 pr_err("Cannot grow linear sub-arrays yet\n");
139 free(subarray);
140 free(st);
141 return 1;
142 }
143
144 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
145 if (nfd < 0) {
146 pr_err("cannot open %s\n", newdev);
147 free(st);
148 return 1;
149 }
150 fstat(nfd, &stb);
151 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
152 pr_err("%s is not a block device!\n", newdev);
153 close(nfd);
154 free(st);
155 return 1;
156 }
157 /* now check out all the devices and make sure we can read the
158 * superblock */
159 for (d=0 ; d < info.array.raid_disks ; d++) {
160 mdu_disk_info_t disk;
161 char *dv;
162
163 st->ss->free_super(st);
164
165 disk.number = d;
166 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
167 pr_err("cannot get device detail for device %d\n",
168 d);
169 close(nfd);
170 free(st);
171 return 1;
172 }
173 dv = map_dev(disk.major, disk.minor, 1);
174 if (!dv) {
175 pr_err("cannot find device file for device %d\n",
176 d);
177 close(nfd);
178 free(st);
179 return 1;
180 }
181 fd2 = dev_open(dv, O_RDWR);
182 if (fd2 < 0) {
183 pr_err("cannot open device file %s\n", dv);
184 close(nfd);
185 free(st);
186 return 1;
187 }
188
189 if (st->ss->load_super(st, fd2, NULL)) {
190 pr_err("cannot find super block on %s\n", dv);
191 close(nfd);
192 close(fd2);
193 free(st);
194 return 1;
195 }
196 close(fd2);
197 }
198 /* Ok, looks good. Lets update the superblock and write it out to
199 * newdev.
200 */
201
202 info.disk.number = d;
203 info.disk.major = major(stb.st_rdev);
204 info.disk.minor = minor(stb.st_rdev);
205 info.disk.raid_disk = d;
206 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
207 st->ss->update_super(st, &info, "linear-grow-new", newdev,
208 0, 0, NULL);
209
210 if (st->ss->store_super(st, nfd)) {
211 pr_err("Cannot store new superblock on %s\n",
212 newdev);
213 close(nfd);
214 return 1;
215 }
216 close(nfd);
217
218 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
219 pr_err("Cannot add new disk to this array\n");
220 return 1;
221 }
222 /* Well, that seems to have worked.
223 * Now go through and update all superblocks
224 */
225
226 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
227 pr_err("cannot get array info for %s\n", devname);
228 return 1;
229 }
230
231 nd = d;
232 for (d=0 ; d < info.array.raid_disks ; d++) {
233 mdu_disk_info_t disk;
234 char *dv;
235
236 disk.number = d;
237 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
238 pr_err("cannot get device detail for device %d\n",
239 d);
240 return 1;
241 }
242 dv = map_dev(disk.major, disk.minor, 1);
243 if (!dv) {
244 pr_err("cannot find device file for device %d\n",
245 d);
246 return 1;
247 }
248 fd2 = dev_open(dv, O_RDWR);
249 if (fd2 < 0) {
250 pr_err("cannot open device file %s\n", dv);
251 return 1;
252 }
253 if (st->ss->load_super(st, fd2, NULL)) {
254 pr_err("cannot find super block on %s\n", dv);
255 close(fd);
256 return 1;
257 }
258 info.array.raid_disks = nd+1;
259 info.array.nr_disks = nd+1;
260 info.array.active_disks = nd+1;
261 info.array.working_disks = nd+1;
262
263 st->ss->update_super(st, &info, "linear-grow-update", dv,
264 0, 0, NULL);
265
266 if (st->ss->store_super(st, fd2)) {
267 pr_err("Cannot store new superblock on %s\n", dv);
268 close(fd2);
269 return 1;
270 }
271 close(fd2);
272 }
273
274 return 0;
275 }
276
277 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
278 {
279 /*
280 * First check that array doesn't have a bitmap
281 * Then create the bitmap
282 * Then add it
283 *
284 * For internal bitmaps, we need to check the version,
285 * find all the active devices, and write the bitmap block
286 * to all devices
287 */
288 mdu_bitmap_file_t bmf;
289 mdu_array_info_t array;
290 struct supertype *st;
291 char *subarray = NULL;
292 int major = BITMAP_MAJOR_HI;
293 int vers = md_get_version(fd);
294 unsigned long long bitmapsize, array_size;
295
296 if (vers < 9003) {
297 major = BITMAP_MAJOR_HOSTENDIAN;
298 pr_err("Warning - bitmaps created on this kernel"
299 " are not portable\n"
300 " between different architectures. Consider upgrading"
301 " the Linux kernel.\n");
302 }
303
304 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
305 if (errno == ENOMEM)
306 pr_err("Memory allocation failure.\n");
307 else
308 pr_err("bitmaps not supported by this kernel.\n");
309 return 1;
310 }
311 if (bmf.pathname[0]) {
312 if (strcmp(s->bitmap_file,"none")==0) {
313 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
314 pr_err("failed to remove bitmap %s\n",
315 bmf.pathname);
316 return 1;
317 }
318 return 0;
319 }
320 pr_err("%s already has a bitmap (%s)\n",
321 devname, bmf.pathname);
322 return 1;
323 }
324 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
325 pr_err("cannot get array status for %s\n", devname);
326 return 1;
327 }
328 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
329 if (strcmp(s->bitmap_file, "none")==0) {
330 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
331 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
332 pr_err("failed to remove internal bitmap.\n");
333 return 1;
334 }
335 return 0;
336 }
337 pr_err("Internal bitmap already present on %s\n",
338 devname);
339 return 1;
340 }
341
342 if (strcmp(s->bitmap_file, "none") == 0) {
343 pr_err("no bitmap found on %s\n", devname);
344 return 1;
345 }
346 if (array.level <= 0) {
347 pr_err("Bitmaps not meaningful with level %s\n",
348 map_num(pers, array.level)?:"of this array");
349 return 1;
350 }
351 bitmapsize = array.size;
352 bitmapsize <<= 1;
353 if (get_dev_size(fd, NULL, &array_size) &&
354 array_size > (0x7fffffffULL<<9)) {
355 /* Array is big enough that we cannot trust array.size
356 * try other approaches
357 */
358 bitmapsize = get_component_size(fd);
359 }
360 if (bitmapsize == 0) {
361 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
362 return 1;
363 }
364
365 if (array.level == 10) {
366 int ncopies = (array.layout&255)*((array.layout>>8)&255);
367 bitmapsize = bitmapsize * array.raid_disks / ncopies;
368 }
369
370 st = super_by_fd(fd, &subarray);
371 if (!st) {
372 pr_err("Cannot understand version %d.%d\n",
373 array.major_version, array.minor_version);
374 return 1;
375 }
376 if (subarray) {
377 pr_err("Cannot add bitmaps to sub-arrays yet\n");
378 free(subarray);
379 free(st);
380 return 1;
381 }
382 if (strcmp(s->bitmap_file, "internal") == 0) {
383 int rv;
384 int d;
385 int offset_setable = 0;
386 struct mdinfo *mdi;
387 if (st->ss->add_internal_bitmap == NULL) {
388 pr_err("Internal bitmaps not supported "
389 "with %s metadata\n", st->ss->name);
390 return 1;
391 }
392 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
393 if (mdi)
394 offset_setable = 1;
395 for (d=0; d< st->max_devs; d++) {
396 mdu_disk_info_t disk;
397 char *dv;
398 disk.number = d;
399 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
400 continue;
401 if (disk.major == 0 &&
402 disk.minor == 0)
403 continue;
404 if ((disk.state & (1<<MD_DISK_SYNC))==0)
405 continue;
406 dv = map_dev(disk.major, disk.minor, 1);
407 if (dv) {
408 int fd2 = dev_open(dv, O_RDWR);
409 if (fd2 < 0)
410 continue;
411 if (st->ss->load_super(st, fd2, NULL)==0) {
412 if (st->ss->add_internal_bitmap(
413 st,
414 &s->bitmap_chunk, c->delay, s->write_behind,
415 bitmapsize, offset_setable,
416 major)
417 )
418 st->ss->write_bitmap(st, fd2);
419 else {
420 pr_err("failed to create internal bitmap"
421 " - chunksize problem.\n");
422 close(fd2);
423 return 1;
424 }
425 }
426 close(fd2);
427 }
428 }
429 if (offset_setable) {
430 st->ss->getinfo_super(st, mdi, NULL);
431 sysfs_init(mdi, fd, NULL);
432 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
433 mdi->bitmap_offset);
434 } else {
435 array.state |= (1<<MD_SB_BITMAP_PRESENT);
436 rv = ioctl(fd, SET_ARRAY_INFO, &array);
437 }
438 if (rv < 0) {
439 if (errno == EBUSY)
440 pr_err("Cannot add bitmap while array is"
441 " resyncing or reshaping etc.\n");
442 pr_err("failed to set internal bitmap.\n");
443 return 1;
444 }
445 } else {
446 int uuid[4];
447 int bitmap_fd;
448 int d;
449 int max_devs = st->max_devs;
450
451 /* try to load a superblock */
452 for (d = 0; d < max_devs; d++) {
453 mdu_disk_info_t disk;
454 char *dv;
455 int fd2;
456 disk.number = d;
457 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
458 continue;
459 if ((disk.major==0 && disk.minor==0) ||
460 (disk.state & (1<<MD_DISK_REMOVED)))
461 continue;
462 dv = map_dev(disk.major, disk.minor, 1);
463 if (!dv)
464 continue;
465 fd2 = dev_open(dv, O_RDONLY);
466 if (fd2 >= 0) {
467 if (st->ss->load_super(st, fd2, NULL) == 0) {
468 close(fd2);
469 st->ss->uuid_from_super(st, uuid);
470 break;
471 }
472 close(fd2);
473 }
474 }
475 if (d == max_devs) {
476 pr_err("cannot find UUID for array!\n");
477 return 1;
478 }
479 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
480 c->delay, s->write_behind, bitmapsize, major)) {
481 return 1;
482 }
483 bitmap_fd = open(s->bitmap_file, O_RDWR);
484 if (bitmap_fd < 0) {
485 pr_err("weird: %s cannot be opened\n",
486 s->bitmap_file);
487 return 1;
488 }
489 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
490 int err = errno;
491 if (errno == EBUSY)
492 pr_err("Cannot add bitmap while array is"
493 " resyncing or reshaping etc.\n");
494 pr_err("Cannot set bitmap file for %s: %s\n",
495 devname, strerror(err));
496 return 1;
497 }
498 }
499
500 return 0;
501 }
502
503 /*
504 * When reshaping an array we might need to backup some data.
505 * This is written to all spares with a 'super_block' describing it.
506 * The superblock goes 4K from the end of the used space on the
507 * device.
508 * It if written after the backup is complete.
509 * It has the following structure.
510 */
511
512 static struct mdp_backup_super {
513 char magic[16]; /* md_backup_data-1 or -2 */
514 __u8 set_uuid[16];
515 __u64 mtime;
516 /* start/sizes in 512byte sectors */
517 __u64 devstart; /* address on backup device/file of data */
518 __u64 arraystart;
519 __u64 length;
520 __u32 sb_csum; /* csum of preceeding bytes. */
521 __u32 pad1;
522 __u64 devstart2; /* offset in to data of second section */
523 __u64 arraystart2;
524 __u64 length2;
525 __u32 sb_csum2; /* csum of preceeding bytes. */
526 __u8 pad[512-68-32];
527 } __attribute__((aligned(512))) bsb, bsb2;
528
529 static __u32 bsb_csum(char *buf, int len)
530 {
531 int i;
532 int csum = 0;
533 for (i = 0; i < len; i++)
534 csum = (csum<<3) + buf[0];
535 return __cpu_to_le32(csum);
536 }
537
538 static int check_idle(struct supertype *st)
539 {
540 /* Check that all member arrays for this container, or the
541 * container of this array, are idle
542 */
543 char *container = (st->container_devnm[0]
544 ? st->container_devnm : st->devnm);
545 struct mdstat_ent *ent, *e;
546 int is_idle = 1;
547
548 ent = mdstat_read(0, 0);
549 for (e = ent ; e; e = e->next) {
550 if (!is_container_member(e, container))
551 continue;
552 if (e->percent >= 0) {
553 is_idle = 0;
554 break;
555 }
556 }
557 free_mdstat(ent);
558 return is_idle;
559 }
560
561 static int freeze_container(struct supertype *st)
562 {
563 char *container = (st->container_devnm[0]
564 ? st->container_devnm : st->devnm);
565
566 if (!check_idle(st))
567 return -1;
568
569 if (block_monitor(container, 1)) {
570 pr_err("failed to freeze container\n");
571 return -2;
572 }
573
574 return 1;
575 }
576
577 static void unfreeze_container(struct supertype *st)
578 {
579 char *container = (st->container_devnm[0]
580 ? st->container_devnm : st->devnm);
581
582 unblock_monitor(container, 1);
583 }
584
585 static int freeze(struct supertype *st)
586 {
587 /* Try to freeze resync/rebuild on this array/container.
588 * Return -1 if the array is busy,
589 * return -2 container cannot be frozen,
590 * return 0 if this kernel doesn't support 'frozen'
591 * return 1 if it worked.
592 */
593 if (st->ss->external)
594 return freeze_container(st);
595 else {
596 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
597 int err;
598 char buf[20];
599
600 if (!sra)
601 return -1;
602 /* Need to clear any 'read-auto' status */
603 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
604 strncmp(buf, "read-auto", 9) == 0)
605 sysfs_set_str(sra, NULL, "array_state", "clean");
606
607 err = sysfs_freeze_array(sra);
608 sysfs_free(sra);
609 return err;
610 }
611 }
612
613 static void unfreeze(struct supertype *st)
614 {
615 if (st->ss->external)
616 return unfreeze_container(st);
617 else {
618 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
619 char buf[20];
620
621 if (sra &&
622 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
623 && strcmp(buf, "frozen\n") == 0) {
624 printf("unfreeze\n");
625 sysfs_set_str(sra, NULL, "sync_action", "idle");
626 }
627 sysfs_free(sra);
628 }
629 }
630
631 static void wait_reshape(struct mdinfo *sra)
632 {
633 int fd = sysfs_get_fd(sra, NULL, "sync_action");
634 char action[20];
635
636 if (fd < 0)
637 return;
638
639 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
640 strncmp(action, "reshape", 7) == 0)
641 sysfs_wait(fd, NULL);
642 close(fd);
643 }
644
645 static int reshape_super(struct supertype *st, unsigned long long size,
646 int level, int layout, int chunksize, int raid_disks,
647 int delta_disks, char *backup_file, char *dev,
648 int direction, int verbose)
649 {
650 /* nothing extra to check in the native case */
651 if (!st->ss->external)
652 return 0;
653 if (!st->ss->reshape_super ||
654 !st->ss->manage_reshape) {
655 pr_err("%s metadata does not support reshape\n",
656 st->ss->name);
657 return 1;
658 }
659
660 return st->ss->reshape_super(st, size, level, layout, chunksize,
661 raid_disks, delta_disks, backup_file, dev,
662 direction, verbose);
663 }
664
665 static void sync_metadata(struct supertype *st)
666 {
667 if (st->ss->external) {
668 if (st->update_tail) {
669 flush_metadata_updates(st);
670 st->update_tail = &st->updates;
671 } else
672 st->ss->sync_metadata(st);
673 }
674 }
675
676 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
677 {
678 /* when dealing with external metadata subarrays we need to be
679 * prepared to handle EAGAIN. The kernel may need to wait for
680 * mdmon to mark the array active so the kernel can handle
681 * allocations/writeback when preparing the reshape action
682 * (md_allow_write()). We temporarily disable safe_mode_delay
683 * to close a race with the array_state going clean before the
684 * next write to raid_disks / stripe_cache_size
685 */
686 char safe[50];
687 int rc;
688
689 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
690 if (!container ||
691 (strcmp(name, "raid_disks") != 0 &&
692 strcmp(name, "stripe_cache_size") != 0))
693 return sysfs_set_num(sra, NULL, name, n);
694
695 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
696 if (rc <= 0)
697 return -1;
698 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
699 rc = sysfs_set_num(sra, NULL, name, n);
700 if (rc < 0 && errno == EAGAIN) {
701 ping_monitor(container);
702 /* if we get EAGAIN here then the monitor is not active
703 * so stop trying
704 */
705 rc = sysfs_set_num(sra, NULL, name, n);
706 }
707 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
708 return rc;
709 }
710
711 int start_reshape(struct mdinfo *sra, int already_running,
712 int before_data_disks, int data_disks)
713 {
714 int err;
715 unsigned long long sync_max_to_set;
716
717 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
718 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
719 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
720 sra->reshape_progress);
721 if (before_data_disks <= data_disks)
722 sync_max_to_set = sra->reshape_progress / data_disks;
723 else
724 sync_max_to_set = (sra->component_size * data_disks
725 - sra->reshape_progress) / data_disks;
726 if (!already_running)
727 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
728 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
729 if (!already_running)
730 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
731
732 return err;
733 }
734
735 void abort_reshape(struct mdinfo *sra)
736 {
737 sysfs_set_str(sra, NULL, "sync_action", "idle");
738 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
739 sysfs_set_num(sra, NULL, "suspend_hi", 0);
740 sysfs_set_num(sra, NULL, "suspend_lo", 0);
741 sysfs_set_num(sra, NULL, "sync_min", 0);
742 // It isn't safe to reset sync_max as we aren't monitoring.
743 // Array really should be stopped at this point.
744 }
745
746 int remove_disks_for_takeover(struct supertype *st,
747 struct mdinfo *sra,
748 int layout)
749 {
750 int nr_of_copies;
751 struct mdinfo *remaining;
752 int slot;
753
754 if (sra->array.level == 10)
755 nr_of_copies = layout & 0xff;
756 else if (sra->array.level == 1)
757 nr_of_copies = sra->array.raid_disks;
758 else
759 return 1;
760
761 remaining = sra->devs;
762 sra->devs = NULL;
763 /* for each 'copy', select one device and remove from the list. */
764 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
765 struct mdinfo **diskp;
766 int found = 0;
767
768 /* Find a working device to keep */
769 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
770 struct mdinfo *disk = *diskp;
771
772 if (disk->disk.raid_disk < slot)
773 continue;
774 if (disk->disk.raid_disk >= slot + nr_of_copies)
775 continue;
776 if (disk->disk.state & (1<<MD_DISK_REMOVED))
777 continue;
778 if (disk->disk.state & (1<<MD_DISK_FAULTY))
779 continue;
780 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
781 continue;
782
783 /* We have found a good disk to use! */
784 *diskp = disk->next;
785 disk->next = sra->devs;
786 sra->devs = disk;
787 found = 1;
788 break;
789 }
790 if (!found)
791 break;
792 }
793
794 if (slot < sra->array.raid_disks) {
795 /* didn't find all slots */
796 struct mdinfo **e;
797 e = &remaining;
798 while (*e)
799 e = &(*e)->next;
800 *e = sra->devs;
801 sra->devs = remaining;
802 return 1;
803 }
804
805 /* Remove all 'remaining' devices from the array */
806 while (remaining) {
807 struct mdinfo *sd = remaining;
808 remaining = sd->next;
809
810 sysfs_set_str(sra, sd, "state", "faulty");
811 sysfs_set_str(sra, sd, "slot", "none");
812 /* for external metadata disks should be removed in mdmon */
813 if (!st->ss->external)
814 sysfs_set_str(sra, sd, "state", "remove");
815 sd->disk.state |= (1<<MD_DISK_REMOVED);
816 sd->disk.state &= ~(1<<MD_DISK_SYNC);
817 sd->next = sra->devs;
818 sra->devs = sd;
819 }
820 return 0;
821 }
822
823 void reshape_free_fdlist(int *fdlist,
824 unsigned long long *offsets,
825 int size)
826 {
827 int i;
828
829 for (i = 0; i < size; i++)
830 if (fdlist[i] >= 0)
831 close(fdlist[i]);
832
833 free(fdlist);
834 free(offsets);
835 }
836
837 int reshape_prepare_fdlist(char *devname,
838 struct mdinfo *sra,
839 int raid_disks,
840 int nrdisks,
841 unsigned long blocks,
842 char *backup_file,
843 int *fdlist,
844 unsigned long long *offsets)
845 {
846 int d = 0;
847 struct mdinfo *sd;
848
849 enable_fds(nrdisks);
850 for (d = 0; d <= nrdisks; d++)
851 fdlist[d] = -1;
852 d = raid_disks;
853 for (sd = sra->devs; sd; sd = sd->next) {
854 if (sd->disk.state & (1<<MD_DISK_FAULTY))
855 continue;
856 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
857 char *dn = map_dev(sd->disk.major,
858 sd->disk.minor, 1);
859 fdlist[sd->disk.raid_disk]
860 = dev_open(dn, O_RDONLY);
861 offsets[sd->disk.raid_disk] = sd->data_offset*512;
862 if (fdlist[sd->disk.raid_disk] < 0) {
863 pr_err("%s: cannot open component %s\n",
864 devname, dn ? dn : "-unknown-");
865 d = -1;
866 goto release;
867 }
868 } else if (backup_file == NULL) {
869 /* spare */
870 char *dn = map_dev(sd->disk.major,
871 sd->disk.minor, 1);
872 fdlist[d] = dev_open(dn, O_RDWR);
873 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
874 if (fdlist[d] < 0) {
875 pr_err("%s: cannot open component %s\n",
876 devname, dn ? dn : "-unknown-");
877 d = -1;
878 goto release;
879 }
880 d++;
881 }
882 }
883 release:
884 return d;
885 }
886
887 int reshape_open_backup_file(char *backup_file,
888 int fd,
889 char *devname,
890 long blocks,
891 int *fdlist,
892 unsigned long long *offsets,
893 char *sys_name,
894 int restart)
895 {
896 /* Return 1 on success, 0 on any form of failure */
897 /* need to check backup file is large enough */
898 char buf[512];
899 struct stat stb;
900 unsigned int dev;
901 int i;
902
903 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
904 S_IRUSR | S_IWUSR);
905 *offsets = 8 * 512;
906 if (*fdlist < 0) {
907 pr_err("%s: cannot create backup file %s: %s\n",
908 devname, backup_file, strerror(errno));
909 return 0;
910 }
911 /* Guard against backup file being on array device.
912 * If array is partitioned or if LVM etc is in the
913 * way this will not notice, but it is better than
914 * nothing.
915 */
916 fstat(*fdlist, &stb);
917 dev = stb.st_dev;
918 fstat(fd, &stb);
919 if (stb.st_rdev == dev) {
920 pr_err("backup file must NOT be"
921 " on the array being reshaped.\n");
922 close(*fdlist);
923 return 0;
924 }
925
926 memset(buf, 0, 512);
927 for (i=0; i < blocks + 8 ; i++) {
928 if (write(*fdlist, buf, 512) != 512) {
929 pr_err("%s: cannot create"
930 " backup file %s: %s\n",
931 devname, backup_file, strerror(errno));
932 return 0;
933 }
934 }
935 if (fsync(*fdlist) != 0) {
936 pr_err("%s: cannot create backup file %s: %s\n",
937 devname, backup_file, strerror(errno));
938 return 0;
939 }
940
941 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
942 char *bu = make_backup(sys_name);
943 if (symlink(backup_file, bu))
944 pr_err("Recording backup file in " MAP_DIR "failed: %s\n",
945 strerror(errno));
946 free(bu);
947 }
948
949 return 1;
950 }
951
952 unsigned long compute_backup_blocks(int nchunk, int ochunk,
953 unsigned int ndata, unsigned int odata)
954 {
955 unsigned long a, b, blocks;
956 /* So how much do we need to backup.
957 * We need an amount of data which is both a whole number of
958 * old stripes and a whole number of new stripes.
959 * So LCM for (chunksize*datadisks).
960 */
961 a = (ochunk/512) * odata;
962 b = (nchunk/512) * ndata;
963 /* Find GCD */
964 a = GCD(a, b);
965 /* LCM == product / GCD */
966 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
967
968 return blocks;
969 }
970
971 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
972 {
973 /* Based on the current array state in info->array and
974 * the changes in info->new_* etc, determine:
975 * - whether the change is possible
976 * - Intermediate level/raid_disks/layout
977 * - whether a restriping reshape is needed
978 * - number of sectors in minimum change unit. This
979 * will cover a whole number of stripes in 'before' and
980 * 'after'.
981 *
982 * Return message if the change should be rejected
983 * NULL if the change can be achieved
984 *
985 * This can be called as part of starting a reshape, or
986 * when assembling an array that is undergoing reshape.
987 */
988 int near, far, offset, copies;
989 int new_disks;
990 int old_chunk, new_chunk;
991 /* delta_parity records change in number of devices
992 * caused by level change
993 */
994 int delta_parity = 0;
995
996 memset(re, 0, sizeof(*re));
997
998 /* If a new level not explicitly given, we assume no-change */
999 if (info->new_level == UnSet)
1000 info->new_level = info->array.level;
1001
1002 if (info->new_chunk)
1003 switch (info->new_level) {
1004 case 0:
1005 case 4:
1006 case 5:
1007 case 6:
1008 case 10:
1009 /* chunk size is meaningful, must divide component_size
1010 * evenly
1011 */
1012 if (info->component_size % (info->new_chunk/512)) {
1013 unsigned long long shrink = info->component_size;
1014 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1015 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1016 info->new_chunk/1024, info->component_size/2);
1017 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1018 devname, shrink/2);
1019 pr_err("will shrink the array so the given chunk size would work.\n");
1020 return "";
1021 }
1022 break;
1023 default:
1024 return "chunk size not meaningful for this level";
1025 }
1026 else
1027 info->new_chunk = info->array.chunk_size;
1028
1029 switch (info->array.level) {
1030 default:
1031 return "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 - "
1045 "reduce size to multiple of 4K first.";
1046 if (info->new_level == 0) {
1047 if (info->delta_disks != UnSet &&
1048 info->delta_disks != 0)
1049 return "Cannot change number of disks "
1050 "with RAID1->RAID0 conversion";
1051 re->level = 0;
1052 re->before.data_disks = 1;
1053 re->after.data_disks = 1;
1054 return NULL;
1055 }
1056 if (info->new_level == 1) {
1057 if (info->delta_disks == UnSet)
1058 /* Don't know what to do */
1059 return "no change requested for Growing RAID1";
1060 re->level = 1;
1061 return NULL;
1062 }
1063 if (info->array.raid_disks == 2 &&
1064 info->new_level == 5) {
1065
1066 re->level = 5;
1067 re->before.data_disks = 1;
1068 if (info->delta_disks != UnSet &&
1069 info->delta_disks != 0)
1070 re->after.data_disks = 1 + info->delta_disks;
1071 else
1072 re->after.data_disks = 1;
1073 if (re->after.data_disks < 1)
1074 return "Number of disks too small for RAID5";
1075
1076 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1077 info->array.chunk_size = 65536;
1078 break;
1079 }
1080 /* Could do some multi-stage conversions, but leave that to
1081 * later.
1082 */
1083 return "Impossibly level change request for RAID1";
1084
1085 case 10:
1086 /* RAID10 can be converted from near mode to
1087 * RAID0 by removing some devices.
1088 * It can also be reshaped if the kernel supports
1089 * new_data_offset.
1090 */
1091 switch (info->new_level) {
1092 case 0:
1093 if ((info->array.layout & ~0xff) != 0x100)
1094 return "Cannot Grow RAID10 with far/offset layout";
1095 /* number of devices must be multiple of number of copies */
1096 if (info->array.raid_disks % (info->array.layout & 0xff))
1097 return "RAID10 layout too complex for Grow operation";
1098
1099 new_disks = (info->array.raid_disks
1100 / (info->array.layout & 0xff));
1101 if (info->delta_disks == UnSet)
1102 info->delta_disks = (new_disks
1103 - info->array.raid_disks);
1104
1105 if (info->delta_disks != new_disks - info->array.raid_disks)
1106 return "New number of raid-devices impossible for RAID10";
1107 if (info->new_chunk &&
1108 info->new_chunk != info->array.chunk_size)
1109 return "Cannot change chunk-size with RAID10 Grow";
1110
1111 /* looks good */
1112 re->level = 0;
1113 re->before.data_disks = new_disks;
1114 re->after.data_disks = re->before.data_disks;
1115 return NULL;
1116
1117 case 10:
1118 near = info->array.layout & 0xff;
1119 far = (info->array.layout >> 8) & 0xff;
1120 offset = info->array.layout & 0x10000;
1121 if (far > 1 && !offset)
1122 return "Cannot reshape RAID10 in far-mode";
1123 copies = near * far;
1124
1125 old_chunk = info->array.chunk_size * far;
1126
1127 if (info->new_layout == UnSet)
1128 info->new_layout = info->array.layout;
1129 else {
1130 near = info->new_layout & 0xff;
1131 far = (info->new_layout >> 8) & 0xff;
1132 offset = info->new_layout & 0x10000;
1133 if (far > 1 && !offset)
1134 return "Cannot reshape RAID10 to far-mode";
1135 if (near * far != copies)
1136 return "Cannot change number of copies"
1137 " when reshaping RAID10";
1138 }
1139 if (info->delta_disks == UnSet)
1140 info->delta_disks = 0;
1141 new_disks = (info->array.raid_disks +
1142 info->delta_disks);
1143
1144 new_chunk = info->new_chunk * far;
1145
1146 re->level = 10;
1147 re->before.layout = info->array.layout;
1148 re->before.data_disks = info->array.raid_disks;
1149 re->after.layout = info->new_layout;
1150 re->after.data_disks = new_disks;
1151 /* For RAID10 we don't do backup but do allow reshape,
1152 * so set backup_blocks to INVALID_SECTORS rather than
1153 * zero.
1154 * And there is no need to synchronise stripes on both
1155 * 'old' and 'new'. So the important
1156 * number is the minimum data_offset difference
1157 * which is the larger of (offset copies * chunk).
1158 */
1159 re->backup_blocks = INVALID_SECTORS;
1160 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1161 if (new_disks < re->before.data_disks &&
1162 info->space_after < re->min_offset_change)
1163 /* Reduce component size by one chunk */
1164 re->new_size = (info->component_size -
1165 re->min_offset_change);
1166 else
1167 re->new_size = info->component_size;
1168 re->new_size = re->new_size * new_disks / copies;
1169 return NULL;
1170
1171 default:
1172 return "RAID10 can only be changed to RAID0";
1173 }
1174 case 0:
1175 /* RAID0 can be converted to RAID10, or to RAID456 */
1176 if (info->new_level == 10) {
1177 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1178 /* Assume near=2 layout */
1179 info->new_layout = 0x102;
1180 info->delta_disks = info->array.raid_disks;
1181 }
1182 if (info->new_layout == UnSet) {
1183 int copies = 1 + (info->delta_disks
1184 / info->array.raid_disks);
1185 if (info->array.raid_disks * (copies-1)
1186 != info->delta_disks)
1187 return "Impossible number of devices"
1188 " for RAID0->RAID10";
1189 info->new_layout = 0x100 + copies;
1190 }
1191 if (info->delta_disks == UnSet) {
1192 int copies = info->new_layout & 0xff;
1193 if (info->new_layout != 0x100 + copies)
1194 return "New layout impossible"
1195 " for RAID0->RAID10";;
1196 info->delta_disks = (copies - 1) *
1197 info->array.raid_disks;
1198 }
1199 if (info->new_chunk &&
1200 info->new_chunk != info->array.chunk_size)
1201 return "Cannot change chunk-size with RAID0->RAID10";
1202 /* looks good */
1203 re->level = 10;
1204 re->before.data_disks = (info->array.raid_disks +
1205 info->delta_disks);
1206 re->after.data_disks = re->before.data_disks;
1207 re->before.layout = info->new_layout;
1208 return NULL;
1209 }
1210
1211 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1212 * a raid4 style layout of the final level.
1213 */
1214 switch (info->new_level) {
1215 case 4:
1216 delta_parity = 1;
1217 case 0:
1218 re->level = 4;
1219 re->before.layout = 0;
1220 break;
1221 case 5:
1222 delta_parity = 1;
1223 re->level = 5;
1224 re->before.layout = ALGORITHM_PARITY_N;
1225 if (info->new_layout == UnSet)
1226 info->new_layout = map_name(r5layout, "default");
1227 break;
1228 case 6:
1229 delta_parity = 2;
1230 re->level = 6;
1231 re->before.layout = ALGORITHM_PARITY_N;
1232 if (info->new_layout == UnSet)
1233 info->new_layout = map_name(r6layout, "default");
1234 break;
1235 default:
1236 return "Impossible level change requested";
1237 }
1238 re->before.data_disks = info->array.raid_disks;
1239 /* determining 'after' layout happens outside this 'switch' */
1240 break;
1241
1242 case 4:
1243 info->array.layout = ALGORITHM_PARITY_N;
1244 case 5:
1245 switch (info->new_level) {
1246 case 0:
1247 delta_parity = -1;
1248 case 4:
1249 re->level = info->array.level;
1250 re->before.data_disks = info->array.raid_disks - 1;
1251 re->before.layout = info->array.layout;
1252 break;
1253 case 5:
1254 re->level = 5;
1255 re->before.data_disks = info->array.raid_disks - 1;
1256 re->before.layout = info->array.layout;
1257 break;
1258 case 6:
1259 delta_parity = 1;
1260 re->level = 6;
1261 re->before.data_disks = info->array.raid_disks - 1;
1262 switch (info->array.layout) {
1263 case ALGORITHM_LEFT_ASYMMETRIC:
1264 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1265 break;
1266 case ALGORITHM_RIGHT_ASYMMETRIC:
1267 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1268 break;
1269 case ALGORITHM_LEFT_SYMMETRIC:
1270 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1271 break;
1272 case ALGORITHM_RIGHT_SYMMETRIC:
1273 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1274 break;
1275 case ALGORITHM_PARITY_0:
1276 re->before.layout = ALGORITHM_PARITY_0_6;
1277 break;
1278 case ALGORITHM_PARITY_N:
1279 re->before.layout = ALGORITHM_PARITY_N_6;
1280 break;
1281 default:
1282 return "Cannot convert an array with this layout";
1283 }
1284 break;
1285 case 1:
1286 if (info->array.raid_disks != 2)
1287 return "Can only convert a 2-device array to RAID1";
1288 if (info->delta_disks != UnSet &&
1289 info->delta_disks != 0)
1290 return "Cannot set raid_disk when "
1291 "converting RAID5->RAID1";
1292 re->level = 1;
1293 info->new_chunk = 0;
1294 return NULL;
1295 default:
1296 return "Impossible level change requested";
1297 }
1298 break;
1299 case 6:
1300 switch (info->new_level) {
1301 case 4:
1302 case 5:
1303 delta_parity = -1;
1304 case 6:
1305 re->level = 6;
1306 re->before.data_disks = info->array.raid_disks - 2;
1307 re->before.layout = info->array.layout;
1308 break;
1309 default:
1310 return "Impossible level change requested";
1311 }
1312 break;
1313 }
1314
1315 /* If we reached here then it looks like a re-stripe is
1316 * happening. We have determined the intermediate level
1317 * and initial raid_disks/layout and stored these in 're'.
1318 *
1319 * We need to deduce the final layout that can be atomically
1320 * converted to the end state.
1321 */
1322 switch (info->new_level) {
1323 case 0:
1324 /* We can only get to RAID0 from RAID4 or RAID5
1325 * with appropriate layout and one extra device
1326 */
1327 if (re->level != 4 && re->level != 5)
1328 return "Cannot covert to RAID0 from this level";
1329
1330 switch (re->level) {
1331 case 4:
1332 re->before.layout = 0;
1333 re->after.layout = 0;
1334 break;
1335 case 5:
1336 re->after.layout = ALGORITHM_PARITY_N;
1337 break;
1338 }
1339 break;
1340
1341 case 4:
1342 /* We can only get to RAID4 from RAID5 */
1343 if (re->level != 4 && re->level != 5)
1344 return "Cannot convert to RAID4 from this level";
1345
1346 switch (re->level) {
1347 case 4:
1348 re->after.layout = 0;
1349 break;
1350 case 5:
1351 re->after.layout = ALGORITHM_PARITY_N;
1352 break;
1353 }
1354 break;
1355
1356 case 5:
1357 /* We get to RAID5 from RAID5 or RAID6 */
1358 if (re->level != 5 && re->level != 6)
1359 return "Cannot convert to RAID5 from this level";
1360
1361 switch (re->level) {
1362 case 5:
1363 if (info->new_layout == UnSet)
1364 re->after.layout = re->before.layout;
1365 else
1366 re->after.layout = info->new_layout;
1367 break;
1368 case 6:
1369 if (info->new_layout == UnSet)
1370 info->new_layout = re->before.layout;
1371
1372 /* after.layout needs to be raid6 version of new_layout */
1373 if (info->new_layout == ALGORITHM_PARITY_N)
1374 re->after.layout = ALGORITHM_PARITY_N;
1375 else {
1376 char layout[40];
1377 char *ls = map_num(r5layout, info->new_layout);
1378 int l;
1379 if (ls) {
1380 /* Current RAID6 layout has a RAID5
1381 * equivalent - good
1382 */
1383 strcat(strcpy(layout, ls), "-6");
1384 l = map_name(r6layout, layout);
1385 if (l == UnSet)
1386 return "Cannot find RAID6 layout"
1387 " to convert to";
1388 } else {
1389 /* Current RAID6 has no equivalent.
1390 * If it is already a '-6' layout we
1391 * can leave it unchanged, else we must
1392 * fail
1393 */
1394 ls = map_num(r6layout, info->new_layout);
1395 if (!ls ||
1396 strcmp(ls+strlen(ls)-2, "-6") != 0)
1397 return "Please specify new layout";
1398 l = info->new_layout;
1399 }
1400 re->after.layout = l;
1401 }
1402 }
1403 break;
1404
1405 case 6:
1406 /* We must already be at level 6 */
1407 if (re->level != 6)
1408 return "Impossible level change";
1409 if (info->new_layout == UnSet)
1410 re->after.layout = info->array.layout;
1411 else
1412 re->after.layout = info->new_layout;
1413 break;
1414 default:
1415 return "Impossible level change requested";
1416 }
1417 if (info->delta_disks == UnSet)
1418 info->delta_disks = delta_parity;
1419
1420 re->after.data_disks = (re->before.data_disks
1421 + info->delta_disks
1422 - delta_parity);
1423 switch (re->level) {
1424 case 6: re->parity = 2;
1425 break;
1426 case 4:
1427 case 5: re->parity = 1;
1428 break;
1429 default: re->parity = 0;
1430 break;
1431 }
1432 /* So we have a restripe operation, we need to calculate the number
1433 * of blocks per reshape operation.
1434 */
1435 re->new_size = info->component_size * re->before.data_disks;
1436 if (info->new_chunk == 0)
1437 info->new_chunk = info->array.chunk_size;
1438 if (re->after.data_disks == re->before.data_disks &&
1439 re->after.layout == re->before.layout &&
1440 info->new_chunk == info->array.chunk_size) {
1441 /* Nothing to change, can change level immediately. */
1442 re->level = info->new_level;
1443 re->backup_blocks = 0;
1444 return NULL;
1445 }
1446 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1447 /* chunk and layout changes make no difference */
1448 re->level = info->new_level;
1449 re->backup_blocks = 0;
1450 return NULL;
1451 }
1452
1453 if (re->after.data_disks == re->before.data_disks &&
1454 get_linux_version() < 2006032)
1455 return "in-place reshape is not safe before 2.6.32 - sorry.";
1456
1457 if (re->after.data_disks < re->before.data_disks &&
1458 get_linux_version() < 2006030)
1459 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1460
1461 re->backup_blocks = compute_backup_blocks(
1462 info->new_chunk, info->array.chunk_size,
1463 re->after.data_disks,
1464 re->before.data_disks);
1465 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1466
1467 re->new_size = info->component_size * re->after.data_disks;
1468 return NULL;
1469 }
1470
1471 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1472 char *text_version)
1473 {
1474 struct mdinfo *info;
1475 char *subarray;
1476 int ret_val = -1;
1477
1478 if ((st == NULL) || (sra == NULL))
1479 return ret_val;
1480
1481 if (text_version == NULL)
1482 text_version = sra->text_version;
1483 subarray = strchr(text_version+1, '/')+1;
1484 info = st->ss->container_content(st, subarray);
1485 if (info) {
1486 unsigned long long current_size = 0;
1487 unsigned long long new_size =
1488 info->custom_array_size/2;
1489
1490 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1491 new_size > current_size) {
1492 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1493 < 0)
1494 dprintf("Error: Cannot set array size");
1495 else {
1496 ret_val = 0;
1497 dprintf("Array size changed");
1498 }
1499 dprintf(" from %llu to %llu.\n",
1500 current_size, new_size);
1501 }
1502 sysfs_free(info);
1503 } else
1504 dprintf("Error: set_array_size(): info pointer in NULL\n");
1505
1506 return ret_val;
1507 }
1508
1509 static int reshape_array(char *container, int fd, char *devname,
1510 struct supertype *st, struct mdinfo *info,
1511 int force, struct mddev_dev *devlist,
1512 unsigned long long data_offset,
1513 char *backup_file, int verbose, int forked,
1514 int restart, int freeze_reshape);
1515 static int reshape_container(char *container, char *devname,
1516 int mdfd,
1517 struct supertype *st,
1518 struct mdinfo *info,
1519 int force,
1520 char *backup_file, int verbose,
1521 int forked, int restart, int freeze_reshape);
1522
1523 int Grow_reshape(char *devname, int fd,
1524 struct mddev_dev *devlist,
1525 unsigned long long data_offset,
1526 struct context *c, struct shape *s)
1527 {
1528 /* Make some changes in the shape of an array.
1529 * The kernel must support the change.
1530 *
1531 * There are three different changes. Each can trigger
1532 * a resync or recovery so we freeze that until we have
1533 * requested everything (if kernel supports freezing - 2.6.30).
1534 * The steps are:
1535 * - change size (i.e. component_size)
1536 * - change level
1537 * - change layout/chunksize/ndisks
1538 *
1539 * The last can require a reshape. It is different on different
1540 * levels so we need to check the level before actioning it.
1541 * Some times the level change needs to be requested after the
1542 * reshape (e.g. raid6->raid5, raid5->raid0)
1543 *
1544 */
1545 struct mdu_array_info_s array;
1546 int rv = 0;
1547 struct supertype *st;
1548 char *subarray = NULL;
1549
1550 int frozen;
1551 int changed = 0;
1552 char *container = NULL;
1553 int cfd = -1;
1554
1555 struct mddev_dev *dv;
1556 int added_disks;
1557
1558 struct mdinfo info;
1559 struct mdinfo *sra;
1560
1561 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1562 pr_err("%s is not an active md array - aborting\n",
1563 devname);
1564 return 1;
1565 }
1566 if (data_offset != INVALID_SECTORS && array.level != 10
1567 && (array.level < 4 || array.level > 6)) {
1568 pr_err("--grow --data-offset not yet supported\n");
1569 return 1;
1570 }
1571
1572 if (s->size > 0 &&
1573 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1574 pr_err("cannot change component size at the same time "
1575 "as other changes.\n"
1576 " Change size first, then check data is intact before "
1577 "making other changes.\n");
1578 return 1;
1579 }
1580
1581 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1582 get_linux_version() < 2006032 &&
1583 !check_env("MDADM_FORCE_FEWER")) {
1584 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1585 " Please use a newer kernel\n");
1586 return 1;
1587 }
1588
1589 st = super_by_fd(fd, &subarray);
1590 if (!st) {
1591 pr_err("Unable to determine metadata format for %s\n", devname);
1592 return 1;
1593 }
1594 if (s->raiddisks > st->max_devs) {
1595 pr_err("Cannot increase raid-disks on this array"
1596 " beyond %d\n", st->max_devs);
1597 return 1;
1598 }
1599
1600 /* in the external case we need to check that the requested reshape is
1601 * supported, and perform an initial check that the container holds the
1602 * pre-requisite spare devices (mdmon owns final validation)
1603 */
1604 if (st->ss->external) {
1605 int rv;
1606
1607 if (subarray) {
1608 container = st->container_devnm;
1609 cfd = open_dev_excl(st->container_devnm);
1610 } else {
1611 container = st->devnm;
1612 close(fd);
1613 cfd = open_dev_excl(st->devnm);
1614 fd = cfd;
1615 }
1616 if (cfd < 0) {
1617 pr_err("Unable to open container for %s\n",
1618 devname);
1619 free(subarray);
1620 return 1;
1621 }
1622
1623 rv = st->ss->load_container(st, cfd, NULL);
1624
1625 if (rv) {
1626 pr_err("Cannot read superblock for %s\n",
1627 devname);
1628 free(subarray);
1629 return 1;
1630 }
1631
1632 /* check if operation is supported for metadata handler */
1633 if (st->ss->container_content) {
1634 struct mdinfo *cc = NULL;
1635 struct mdinfo *content = NULL;
1636
1637 cc = st->ss->container_content(st, subarray);
1638 for (content = cc; content ; content = content->next) {
1639 int allow_reshape = 1;
1640
1641 /* check if reshape is allowed based on metadata
1642 * indications stored in content.array.status
1643 */
1644 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1645 allow_reshape = 0;
1646 if (content->array.state
1647 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1648 allow_reshape = 0;
1649 if (!allow_reshape) {
1650 pr_err("cannot reshape arrays in"
1651 " container with unsupported"
1652 " metadata: %s(%s)\n",
1653 devname, container);
1654 sysfs_free(cc);
1655 free(subarray);
1656 return 1;
1657 }
1658 }
1659 sysfs_free(cc);
1660 }
1661 if (mdmon_running(container))
1662 st->update_tail = &st->updates;
1663 }
1664
1665 added_disks = 0;
1666 for (dv = devlist; dv; dv = dv->next)
1667 added_disks++;
1668 if (s->raiddisks > array.raid_disks &&
1669 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1670 !c->force) {
1671 pr_err("Need %d spare%s to avoid degraded array,"
1672 " and only have %d.\n"
1673 " Use --force to over-ride this check.\n",
1674 s->raiddisks - array.raid_disks,
1675 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1676 array.spare_disks + added_disks);
1677 return 1;
1678 }
1679
1680 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1681 | GET_STATE | GET_VERSION);
1682 if (sra) {
1683 if (st->ss->external && subarray == NULL) {
1684 array.level = LEVEL_CONTAINER;
1685 sra->array.level = LEVEL_CONTAINER;
1686 }
1687 } else {
1688 pr_err("failed to read sysfs parameters for %s\n",
1689 devname);
1690 return 1;
1691 }
1692 frozen = freeze(st);
1693 if (frozen < -1) {
1694 /* freeze() already spewed the reason */
1695 sysfs_free(sra);
1696 return 1;
1697 } else if (frozen < 0) {
1698 pr_err("%s is performing resync/recovery and cannot"
1699 " be reshaped\n", devname);
1700 sysfs_free(sra);
1701 return 1;
1702 }
1703
1704 /* ========= set size =============== */
1705 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1706 unsigned long long orig_size = get_component_size(fd)/2;
1707 unsigned long long min_csize;
1708 struct mdinfo *mdi;
1709 int raid0_takeover = 0;
1710
1711 if (orig_size == 0)
1712 orig_size = (unsigned) array.size;
1713
1714 if (orig_size == 0) {
1715 pr_err("Cannot set device size in this type of array.\n");
1716 rv = 1;
1717 goto release;
1718 }
1719
1720 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1721 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1722 rv = 1;
1723 goto release;
1724 }
1725 sync_metadata(st);
1726 if (st->ss->external) {
1727 /* metadata can have size limitation
1728 * update size value according to metadata information
1729 */
1730 struct mdinfo *sizeinfo =
1731 st->ss->container_content(st, subarray);
1732 if (sizeinfo) {
1733 unsigned long long new_size =
1734 sizeinfo->custom_array_size/2;
1735 int data_disks = get_data_disks(
1736 sizeinfo->array.level,
1737 sizeinfo->array.layout,
1738 sizeinfo->array.raid_disks);
1739 new_size /= data_disks;
1740 dprintf("Metadata size correction from %llu to "
1741 "%llu (%llu)\n", orig_size, new_size,
1742 new_size * data_disks);
1743 s->size = new_size;
1744 sysfs_free(sizeinfo);
1745 }
1746 }
1747
1748 /* Update the size of each member device in case
1749 * they have been resized. This will never reduce
1750 * below the current used-size. The "size" attribute
1751 * understands '0' to mean 'max'.
1752 */
1753 min_csize = 0;
1754 rv = 0;
1755 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1756 if (sysfs_set_num(sra, mdi, "size",
1757 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1758 /* Probably kernel refusing to let us
1759 * reduce the size - not an error.
1760 */
1761 break;
1762 }
1763 if (array.not_persistent == 0 &&
1764 array.major_version == 0 &&
1765 get_linux_version() < 3001000) {
1766 /* Dangerous to allow size to exceed 2TB */
1767 unsigned long long csize;
1768 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1769 if (csize >= 2ULL*1024*1024*1024)
1770 csize = 2ULL*1024*1024*1024;
1771 if ((min_csize == 0 || (min_csize
1772 > csize)))
1773 min_csize = csize;
1774 }
1775 }
1776 }
1777 if (rv) {
1778 pr_err("Cannot set size on "
1779 "array members.\n");
1780 goto size_change_error;
1781 }
1782 if (min_csize && s->size > min_csize) {
1783 pr_err("Cannot safely make this array "
1784 "use more than 2TB per device on this kernel.\n");
1785 rv = 1;
1786 goto size_change_error;
1787 }
1788 if (min_csize && s->size == MAX_SIZE) {
1789 /* Don't let the kernel choose a size - it will get
1790 * it wrong
1791 */
1792 pr_err("Limited v0.90 array to "
1793 "2TB per device\n");
1794 s->size = min_csize;
1795 }
1796 if (st->ss->external) {
1797 if (sra->array.level == 0) {
1798 rv = sysfs_set_str(sra, NULL, "level",
1799 "raid5");
1800 if (!rv) {
1801 raid0_takeover = 1;
1802 /* get array parametes after takeover
1803 * to chane one parameter at time only
1804 */
1805 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1806 }
1807 }
1808 /* make sure mdmon is
1809 * aware of the new level */
1810 if (!mdmon_running(st->container_devnm))
1811 start_mdmon(st->container_devnm);
1812 ping_monitor(container);
1813 if (mdmon_running(st->container_devnm) &&
1814 st->update_tail == NULL)
1815 st->update_tail = &st->updates;
1816 }
1817
1818 if (s->size == MAX_SIZE)
1819 s->size = 0;
1820 array.size = s->size;
1821 if (array.size != (signed)s->size) {
1822 /* got truncated to 32bit, write to
1823 * component_size instead
1824 */
1825 if (sra)
1826 rv = sysfs_set_num(sra, NULL,
1827 "component_size", s->size);
1828 else
1829 rv = -1;
1830 } else {
1831 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1832
1833 /* manage array size when it is managed externally
1834 */
1835 if ((rv == 0) && st->ss->external)
1836 rv = set_array_size(st, sra, sra->text_version);
1837 }
1838
1839 if (raid0_takeover) {
1840 /* do not recync non-existing parity,
1841 * we will drop it anyway
1842 */
1843 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1844 /* go back to raid0, drop parity disk
1845 */
1846 sysfs_set_str(sra, NULL, "level", "raid0");
1847 ioctl(fd, GET_ARRAY_INFO, &array);
1848 }
1849
1850 size_change_error:
1851 if (rv != 0) {
1852 int err = errno;
1853
1854 /* restore metadata */
1855 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1856 UnSet, NULL, devname,
1857 ROLLBACK_METADATA_CHANGES,
1858 c->verbose) == 0)
1859 sync_metadata(st);
1860 pr_err("Cannot set device size for %s: %s\n",
1861 devname, strerror(err));
1862 if (err == EBUSY &&
1863 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1864 cont_err("Bitmap must be removed before size can be changed\n");
1865 rv = 1;
1866 goto release;
1867 }
1868 if (s->assume_clean) {
1869 /* This will fail on kernels older than 3.0 unless
1870 * a backport has been arranged.
1871 */
1872 if (sra == NULL ||
1873 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1874 pr_err("--assume-clean not supported with --grow on this kernel\n");
1875 }
1876 ioctl(fd, GET_ARRAY_INFO, &array);
1877 s->size = get_component_size(fd)/2;
1878 if (s->size == 0)
1879 s->size = array.size;
1880 if (c->verbose >= 0) {
1881 if (s->size == orig_size)
1882 pr_err("component size of %s "
1883 "unchanged at %lluK\n",
1884 devname, s->size);
1885 else
1886 pr_err("component size of %s "
1887 "has been set to %lluK\n",
1888 devname, s->size);
1889 }
1890 changed = 1;
1891 } else if (array.level != LEVEL_CONTAINER) {
1892 s->size = get_component_size(fd)/2;
1893 if (s->size == 0)
1894 s->size = array.size;
1895 }
1896
1897 /* See if there is anything else to do */
1898 if ((s->level == UnSet || s->level == array.level) &&
1899 (s->layout_str == NULL) &&
1900 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1901 data_offset == INVALID_SECTORS &&
1902 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1903 /* Nothing more to do */
1904 if (!changed && c->verbose >= 0)
1905 pr_err("%s: no change requested\n",
1906 devname);
1907 goto release;
1908 }
1909
1910 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1911 * current implementation assumes that following conditions must be met:
1912 * - RAID10:
1913 * - far_copies == 1
1914 * - near_copies == 2
1915 */
1916 if ((s->level == 0 && array.level == 10 && sra &&
1917 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1918 (s->level == 0 && array.level == 1 && sra)) {
1919 int err;
1920 err = remove_disks_for_takeover(st, sra, array.layout);
1921 if (err) {
1922 dprintf(Name": Array cannot be reshaped\n");
1923 if (cfd > -1)
1924 close(cfd);
1925 rv = 1;
1926 goto release;
1927 }
1928 /* Make sure mdmon has seen the device removal
1929 * and updated metadata before we continue with
1930 * level change
1931 */
1932 if (container)
1933 ping_monitor(container);
1934 }
1935
1936 memset(&info, 0, sizeof(info));
1937 info.array = array;
1938 sysfs_init(&info, fd, NULL);
1939 strcpy(info.text_version, sra->text_version);
1940 info.component_size = s->size*2;
1941 info.new_level = s->level;
1942 info.new_chunk = s->chunk * 1024;
1943 if (info.array.level == LEVEL_CONTAINER) {
1944 info.delta_disks = UnSet;
1945 info.array.raid_disks = s->raiddisks;
1946 } else if (s->raiddisks)
1947 info.delta_disks = s->raiddisks - info.array.raid_disks;
1948 else
1949 info.delta_disks = UnSet;
1950 if (s->layout_str == NULL) {
1951 info.new_layout = UnSet;
1952 if (info.array.level == 6 &&
1953 (info.new_level == 6 || info.new_level == UnSet) &&
1954 info.array.layout >= 16) {
1955 pr_err("%s has a non-standard layout. If you"
1956 " wish to preserve this\n", devname);
1957 cont_err("during the reshape, please specify"
1958 " --layout=preserve\n");
1959 cont_err("If you want to change it, specify a"
1960 " layout or use --layout=normalise\n");
1961 rv = 1;
1962 goto release;
1963 }
1964 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1965 strcmp(s->layout_str, "normalize") == 0) {
1966 /* If we have a -6 RAID6 layout, remove the '-6'. */
1967 info.new_layout = UnSet;
1968 if (info.array.level == 6 && info.new_level == UnSet) {
1969 char l[40], *h;
1970 strcpy(l, map_num(r6layout, info.array.layout));
1971 h = strrchr(l, '-');
1972 if (h && strcmp(h, "-6") == 0) {
1973 *h = 0;
1974 info.new_layout = map_name(r6layout, l);
1975 }
1976 } else {
1977 pr_err("%s is only meaningful when reshaping"
1978 " a RAID6 array.\n", s->layout_str);
1979 rv = 1;
1980 goto release;
1981 }
1982 } else if (strcmp(s->layout_str, "preserve") == 0) {
1983 /* This means that a non-standard RAID6 layout
1984 * is OK.
1985 * In particular:
1986 * - When reshape a RAID6 (e.g. adding a device)
1987 * which is in a non-standard layout, it is OK
1988 * to preserve that layout.
1989 * - When converting a RAID5 to RAID6, leave it in
1990 * the XXX-6 layout, don't re-layout.
1991 */
1992 if (info.array.level == 6 && info.new_level == UnSet)
1993 info.new_layout = info.array.layout;
1994 else if (info.array.level == 5 && info.new_level == 6) {
1995 char l[40];
1996 strcpy(l, map_num(r5layout, info.array.layout));
1997 strcat(l, "-6");
1998 info.new_layout = map_name(r6layout, l);
1999 } else {
2000 pr_err("%s in only meaningful when reshaping"
2001 " to RAID6\n", s->layout_str);
2002 rv = 1;
2003 goto release;
2004 }
2005 } else {
2006 int l = info.new_level;
2007 if (l == UnSet)
2008 l = info.array.level;
2009 switch (l) {
2010 case 5:
2011 info.new_layout = map_name(r5layout, s->layout_str);
2012 break;
2013 case 6:
2014 info.new_layout = map_name(r6layout, s->layout_str);
2015 break;
2016 case 10:
2017 info.new_layout = parse_layout_10(s->layout_str);
2018 break;
2019 case LEVEL_FAULTY:
2020 info.new_layout = parse_layout_faulty(s->layout_str);
2021 break;
2022 default:
2023 pr_err("layout not meaningful"
2024 " with this level\n");
2025 rv = 1;
2026 goto release;
2027 }
2028 if (info.new_layout == UnSet) {
2029 pr_err("layout %s not understood"
2030 " for this level\n",
2031 s->layout_str);
2032 rv = 1;
2033 goto release;
2034 }
2035 }
2036
2037 if (array.level == LEVEL_FAULTY) {
2038 if (s->level != UnSet && s->level != array.level) {
2039 pr_err("cannot change level of Faulty device\n");
2040 rv =1 ;
2041 }
2042 if (s->chunk) {
2043 pr_err("cannot set chunksize of Faulty device\n");
2044 rv =1 ;
2045 }
2046 if (s->raiddisks && s->raiddisks != 1) {
2047 pr_err("cannot set raid_disks of Faulty device\n");
2048 rv =1 ;
2049 }
2050 if (s->layout_str) {
2051 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2052 dprintf("Cannot get array information.\n");
2053 goto release;
2054 }
2055 array.layout = info.new_layout;
2056 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2057 pr_err("failed to set new layout\n");
2058 rv = 1;
2059 } else if (c->verbose >= 0)
2060 printf("layout for %s set to %d\n",
2061 devname, array.layout);
2062 }
2063 } else if (array.level == LEVEL_CONTAINER) {
2064 /* This change is to be applied to every array in the
2065 * container. This is only needed when the metadata imposes
2066 * restraints of the various arrays in the container.
2067 * Currently we only know that IMSM requires all arrays
2068 * to have the same number of devices so changing the
2069 * number of devices (On-Line Capacity Expansion) must be
2070 * performed at the level of the container
2071 */
2072 rv = reshape_container(container, devname, -1, st, &info,
2073 c->force, c->backup_file, c->verbose, 0, 0, 0);
2074 frozen = 0;
2075 } else {
2076 /* get spare devices from external metadata
2077 */
2078 if (st->ss->external) {
2079 struct mdinfo *info2;
2080
2081 info2 = st->ss->container_content(st, subarray);
2082 if (info2) {
2083 info.array.spare_disks =
2084 info2->array.spare_disks;
2085 sysfs_free(info2);
2086 }
2087 }
2088
2089 /* Impose these changes on a single array. First
2090 * check that the metadata is OK with the change. */
2091
2092 if (reshape_super(st, 0, info.new_level,
2093 info.new_layout, info.new_chunk,
2094 info.array.raid_disks, info.delta_disks,
2095 c->backup_file, devname, APPLY_METADATA_CHANGES,
2096 c->verbose)) {
2097 rv = 1;
2098 goto release;
2099 }
2100 sync_metadata(st);
2101 rv = reshape_array(container, fd, devname, st, &info, c->force,
2102 devlist, data_offset, c->backup_file, c->verbose,
2103 0, 0, 0);
2104 frozen = 0;
2105 }
2106 release:
2107 sysfs_free(sra);
2108 if (frozen > 0)
2109 unfreeze(st);
2110 return rv;
2111 }
2112
2113 /* verify_reshape_position()
2114 * Function checks if reshape position in metadata is not farther
2115 * than position in md.
2116 * Return value:
2117 * 0 : not valid sysfs entry
2118 * it can be caused by not started reshape, it should be started
2119 * by reshape array or raid0 array is before takeover
2120 * -1 : error, reshape position is obviously wrong
2121 * 1 : success, reshape progress correct or updated
2122 */
2123 static int verify_reshape_position(struct mdinfo *info, int level)
2124 {
2125 int ret_val = 0;
2126 char buf[40];
2127 int rv;
2128
2129 /* read sync_max, failure can mean raid0 array */
2130 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2131
2132 if (rv > 0) {
2133 char *ep;
2134 unsigned long long position = strtoull(buf, &ep, 0);
2135
2136 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2137 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2138 position *= get_data_disks(level,
2139 info->new_layout,
2140 info->array.raid_disks);
2141 if (info->reshape_progress < position) {
2142 dprintf("Corrected reshape progress (%llu) to "
2143 "md position (%llu)\n",
2144 info->reshape_progress, position);
2145 info->reshape_progress = position;
2146 ret_val = 1;
2147 } else if (info->reshape_progress > position) {
2148 pr_err("Fatal error: array "
2149 "reshape was not properly frozen "
2150 "(expected reshape position is %llu, "
2151 "but reshape progress is %llu.\n",
2152 position, info->reshape_progress);
2153 ret_val = -1;
2154 } else {
2155 dprintf("Reshape position in md and metadata "
2156 "are the same;");
2157 ret_val = 1;
2158 }
2159 }
2160 } else if (rv == 0) {
2161 /* for valid sysfs entry, 0-length content
2162 * should be indicated as error
2163 */
2164 ret_val = -1;
2165 }
2166
2167 return ret_val;
2168 }
2169
2170 static unsigned long long choose_offset(unsigned long long lo,
2171 unsigned long long hi,
2172 unsigned long long min,
2173 unsigned long long max)
2174 {
2175 /* Choose a new offset between hi and lo.
2176 * It must be between min and max, but
2177 * we would prefer something near the middle of hi/lo, and also
2178 * prefer to be aligned to a big power of 2.
2179 *
2180 * So we start with the middle, then for each bit,
2181 * starting at '1' and increasing, if it is set, we either
2182 * add it or subtract it if possible, preferring the option
2183 * which is furthest from the boundary.
2184 *
2185 * We stop once we get a 1MB alignment. As units are in sectors,
2186 * 1MB = 2*1024 sectors.
2187 */
2188 unsigned long long choice = (lo + hi) / 2;
2189 unsigned long long bit = 1;
2190
2191 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2192 unsigned long long bigger, smaller;
2193 if (! (bit & choice))
2194 continue;
2195 bigger = choice + bit;
2196 smaller = choice - bit;
2197 if (bigger > max && smaller < min)
2198 break;
2199 if (bigger > max)
2200 choice = smaller;
2201 else if (smaller < min)
2202 choice = bigger;
2203 else if (hi - bigger > smaller - lo)
2204 choice = bigger;
2205 else
2206 choice = smaller;
2207 }
2208 return choice;
2209 }
2210
2211 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2212 char *devname, int delta_disks,
2213 unsigned long long data_offset,
2214 unsigned long long min,
2215 int can_fallback)
2216 {
2217 struct mdinfo *sd;
2218 int dir = 0;
2219 int err = 0;
2220 unsigned long long before, after;
2221
2222 /* Need to find min space before and after so same is used
2223 * on all devices
2224 */
2225 before = UINT64_MAX;
2226 after = UINT64_MAX;
2227 for (sd = sra->devs; sd; sd = sd->next) {
2228 char *dn;
2229 int dfd;
2230 int rv;
2231 struct supertype *st2;
2232 struct mdinfo info2;
2233
2234 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2235 continue;
2236 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2237 dfd = dev_open(dn, O_RDONLY);
2238 if (dfd < 0) {
2239 pr_err("%s: cannot open component %s\n",
2240 devname, dn ? dn : "-unknown-");
2241 goto release;
2242 }
2243 st2 = dup_super(st);
2244 rv = st2->ss->load_super(st2,dfd, NULL);
2245 close(dfd);
2246 if (rv) {
2247 free(st2);
2248 pr_err("%s: cannot get superblock from %s\n",
2249 devname, dn);
2250 goto release;
2251 }
2252 st2->ss->getinfo_super(st2, &info2, NULL);
2253 st2->ss->free_super(st2);
2254 free(st2);
2255 if (info2.space_before == 0 &&
2256 info2.space_after == 0) {
2257 /* Metadata doesn't support data_offset changes */
2258 return 1;
2259 }
2260 if (before > info2.space_before)
2261 before = info2.space_before;
2262 if (after > info2.space_after)
2263 after = info2.space_after;
2264
2265 if (data_offset != INVALID_SECTORS) {
2266 if (dir == 0) {
2267 if (info2.data_offset == data_offset) {
2268 pr_err("%s: already has that data_offset\n",
2269 dn);
2270 goto release;
2271 }
2272 if (data_offset < info2.data_offset)
2273 dir = -1;
2274 else
2275 dir = 1;
2276 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2277 (data_offset >= info2.data_offset && dir == -1)) {
2278 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2279 dn);
2280 goto release;
2281 }
2282 }
2283 }
2284 if (before == UINT64_MAX)
2285 /* impossible really, there must be no devices */
2286 return 1;
2287
2288 for (sd = sra->devs; sd; sd = sd->next) {
2289 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2290 unsigned long long new_data_offset;
2291
2292 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2293 continue;
2294 if (delta_disks < 0) {
2295 /* Don't need any space as array is shrinking
2296 * just move data_offset up by min
2297 */
2298 if (data_offset == INVALID_SECTORS)
2299 new_data_offset = sd->data_offset + min;
2300 else {
2301 if (data_offset < sd->data_offset + min) {
2302 pr_err("--data-offset too small for %s\n",
2303 dn);
2304 goto release;
2305 }
2306 new_data_offset = data_offset;
2307 }
2308 } else if (delta_disks > 0) {
2309 /* need space before */
2310 if (before < min) {
2311 if (can_fallback)
2312 goto fallback;
2313 pr_err("Insufficient head-space for reshape on %s\n",
2314 dn);
2315 goto release;
2316 }
2317 if (data_offset == INVALID_SECTORS)
2318 new_data_offset = sd->data_offset - min;
2319 else {
2320 if (data_offset > sd->data_offset - min) {
2321 pr_err("--data-offset too large for %s\n",
2322 dn);
2323 goto release;
2324 }
2325 new_data_offset = data_offset;
2326 }
2327 } else {
2328 if (dir == 0) {
2329 /* can move up or down. If 'data_offset'
2330 * was set we would have already decided,
2331 * so just choose direction with most space.
2332 */
2333 if (before > after)
2334 dir = -1;
2335 else
2336 dir = 1;
2337 }
2338 sysfs_set_str(sra, NULL, "reshape_direction",
2339 dir == 1 ? "backwards" : "forwards");
2340 if (dir > 0) {
2341 /* Increase data offset */
2342 if (after < min) {
2343 if (can_fallback)
2344 goto fallback;
2345 pr_err("Insufficient tail-space for reshape on %s\n",
2346 dn);
2347 goto release;
2348 }
2349 if (data_offset != INVALID_SECTORS &&
2350 data_offset < sd->data_offset + min) {
2351 pr_err("--data-offset too small on %s\n",
2352 dn);
2353 goto release;
2354 }
2355 if (data_offset != INVALID_SECTORS)
2356 new_data_offset = data_offset;
2357 else
2358 new_data_offset = choose_offset(sd->data_offset,
2359 sd->data_offset + after,
2360 sd->data_offset + min,
2361 sd->data_offset + after);
2362 } else {
2363 /* Decrease data offset */
2364 if (before < min) {
2365 if (can_fallback)
2366 goto fallback;
2367 pr_err("insufficient head-room on %s\n",
2368 dn);
2369 goto release;
2370 }
2371 if (data_offset != INVALID_SECTORS &&
2372 data_offset < sd->data_offset - min) {
2373 pr_err("--data-offset too small on %s\n",
2374 dn);
2375 goto release;
2376 }
2377 if (data_offset != INVALID_SECTORS)
2378 new_data_offset = data_offset;
2379 else
2380 new_data_offset = choose_offset(sd->data_offset - before,
2381 sd->data_offset,
2382 sd->data_offset - before,
2383 sd->data_offset - min);
2384 }
2385 }
2386 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2387 if (err < 0 && errno == E2BIG) {
2388 /* try again after increasing data size to max */
2389 err = sysfs_set_num(sra, sd, "size", 0);
2390 if (err < 0 && errno == EINVAL &&
2391 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2392 /* some kernels have a bug where you cannot
2393 * use '0' on spare devices. */
2394 sysfs_set_num(sra, sd, "size",
2395 (sra->component_size + after)/2);
2396 }
2397 err = sysfs_set_num(sra, sd, "new_offset",
2398 new_data_offset);
2399 }
2400 if (err < 0) {
2401 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2402 pr_err("data-offset is too big for %s\n",
2403 dn);
2404 goto release;
2405 }
2406 if (sd == sra->devs &&
2407 (errno == ENOENT || errno == E2BIG))
2408 /* Early kernel, no 'new_offset' file,
2409 * or kernel doesn't like us.
2410 * For RAID5/6 this is not fatal
2411 */
2412 return 1;
2413 pr_err("Cannot set new_offset for %s\n",
2414 dn);
2415 break;
2416 }
2417 }
2418 return err;
2419 release:
2420 return -1;
2421 fallback:
2422 /* Just use a backup file */
2423 return 1;
2424 }
2425
2426 static int raid10_reshape(char *container, int fd, char *devname,
2427 struct supertype *st, struct mdinfo *info,
2428 struct reshape *reshape,
2429 unsigned long long data_offset,
2430 int force, int verbose)
2431 {
2432 /* Changing raid_disks, layout, chunksize or possibly
2433 * just data_offset for a RAID10.
2434 * We must always change data_offset. We change by at least
2435 * ->min_offset_change which is the largest of the old and new
2436 * chunk sizes.
2437 * If raid_disks is increasing, then data_offset must decrease
2438 * by at least this copy size.
2439 * If raid_disks is unchanged, data_offset must increase or
2440 * decrease by at least min_offset_change but preferably by much more.
2441 * We choose half of the available space.
2442 * If raid_disks is decreasing, data_offset must increase by
2443 * at least min_offset_change. To allow of this, component_size
2444 * must be decreased by the same amount.
2445 *
2446 * So we calculate the required minimum and direction, possibly
2447 * reduce the component_size, then iterate through the devices
2448 * and set the new_data_offset.
2449 * If that all works, we set chunk_size, layout, raid_disks, and start
2450 * 'reshape'
2451 */
2452 struct mdinfo *sra;
2453 unsigned long long min;
2454 int err = 0;
2455
2456 sra = sysfs_read(fd, NULL,
2457 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2458 );
2459 if (!sra) {
2460 pr_err("%s: Cannot get array details from sysfs\n",
2461 devname);
2462 goto release;
2463 }
2464 min = reshape->min_offset_change;
2465
2466 if (info->delta_disks)
2467 sysfs_set_str(sra, NULL, "reshape_direction",
2468 info->delta_disks < 0 ? "backwards" : "forwards");
2469 if (info->delta_disks < 0 &&
2470 info->space_after < min) {
2471 int rv = sysfs_set_num(sra, NULL, "component_size",
2472 (sra->component_size -
2473 min)/2);
2474 if (rv) {
2475 pr_err("cannot reduce component size\n");
2476 goto release;
2477 }
2478 }
2479 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2480 min, 0);
2481 if (err == 1) {
2482 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2483 cont_err("supported on this kernel\n");
2484 err = -1;
2485 }
2486 if (err < 0)
2487 goto release;
2488
2489 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2490 err = errno;
2491 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2492 err = errno;
2493 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2494 info->array.raid_disks + info->delta_disks) < 0)
2495 err = errno;
2496 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2497 err = errno;
2498 if (err) {
2499 pr_err("Cannot set array shape for %s\n",
2500 devname);
2501 if (err == EBUSY &&
2502 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2503 cont_err(" Bitmap must be removed before"
2504 " shape can be changed\n");
2505 goto release;
2506 }
2507 sysfs_free(sra);
2508 return 0;
2509 release:
2510 sysfs_free(sra);
2511 return 1;
2512 }
2513
2514 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2515 {
2516 struct mdinfo *sra, *sd;
2517 /* Initialisation to silence compiler warning */
2518 unsigned long long min_space_before = 0, min_space_after = 0;
2519 int first = 1;
2520
2521 sra = sysfs_read(fd, NULL, GET_DEVS);
2522 if (!sra)
2523 return;
2524 for (sd = sra->devs; sd; sd = sd->next) {
2525 char *dn;
2526 int dfd;
2527 struct supertype *st2;
2528 struct mdinfo info2;
2529
2530 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2531 continue;
2532 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2533 dfd = dev_open(dn, O_RDONLY);
2534 if (dfd < 0)
2535 break;
2536 st2 = dup_super(st);
2537 if (st2->ss->load_super(st2,dfd, NULL)) {
2538 close(dfd);
2539 free(st2);
2540 break;
2541 }
2542 close(dfd);
2543 st2->ss->getinfo_super(st2, &info2, NULL);
2544 st2->ss->free_super(st2);
2545 free(st2);
2546 if (first ||
2547 min_space_before > info2.space_before)
2548 min_space_before = info2.space_before;
2549 if (first ||
2550 min_space_after > info2.space_after)
2551 min_space_after = info2.space_after;
2552 first = 0;
2553 }
2554 if (sd == NULL && !first) {
2555 info->space_after = min_space_after;
2556 info->space_before = min_space_before;
2557 }
2558 sysfs_free(sra);
2559 }
2560
2561 static void update_cache_size(char *container, struct mdinfo *sra,
2562 struct mdinfo *info,
2563 int disks, unsigned long long blocks)
2564 {
2565 /* Check that the internal stripe cache is
2566 * large enough, or it won't work.
2567 * It must hold at least 4 stripes of the larger
2568 * chunk size
2569 */
2570 unsigned long cache;
2571 cache = max(info->array.chunk_size, info->new_chunk);
2572 cache *= 4; /* 4 stripes minimum */
2573 cache /= 512; /* convert to sectors */
2574 /* make sure there is room for 'blocks' with a bit to spare */
2575 if (cache < 16 + blocks / disks)
2576 cache = 16 + blocks / disks;
2577 cache /= (4096/512); /* Covert from sectors to pages */
2578
2579 if (sra->cache_size < cache)
2580 subarray_set_num(container, sra, "stripe_cache_size",
2581 cache+1);
2582 }
2583
2584 static int impose_reshape(struct mdinfo *sra,
2585 struct mdinfo *info,
2586 struct supertype *st,
2587 int fd,
2588 int restart,
2589 char *devname, char *container,
2590 struct reshape *reshape)
2591 {
2592 struct mdu_array_info_s array;
2593
2594 sra->new_chunk = info->new_chunk;
2595
2596 if (restart) {
2597 /* for external metadata checkpoint saved by mdmon can be lost
2598 * or missed /due to e.g. crash/. Check if md is not during
2599 * restart farther than metadata points to.
2600 * If so, this means metadata information is obsolete.
2601 */
2602 if (st->ss->external)
2603 verify_reshape_position(info, reshape->level);
2604 sra->reshape_progress = info->reshape_progress;
2605 } else {
2606 sra->reshape_progress = 0;
2607 if (reshape->after.data_disks < reshape->before.data_disks)
2608 /* start from the end of the new array */
2609 sra->reshape_progress = (sra->component_size
2610 * reshape->after.data_disks);
2611 }
2612
2613 ioctl(fd, GET_ARRAY_INFO, &array);
2614 if (info->array.chunk_size == info->new_chunk &&
2615 reshape->before.layout == reshape->after.layout &&
2616 st->ss->external == 0) {
2617 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2618 array.raid_disks = reshape->after.data_disks + reshape->parity;
2619 if (!restart &&
2620 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2621 int err = errno;
2622
2623 pr_err("Cannot set device shape for %s: %s\n",
2624 devname, strerror(errno));
2625
2626 if (err == EBUSY &&
2627 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2628 cont_err("Bitmap must be removed before"
2629 " shape can be changed\n");
2630
2631 goto release;
2632 }
2633 } else if (!restart) {
2634 /* set them all just in case some old 'new_*' value
2635 * persists from some earlier problem.
2636 */
2637 int err = 0;
2638 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2639 err = errno;
2640 if (!err && sysfs_set_num(sra, NULL, "layout",
2641 reshape->after.layout) < 0)
2642 err = errno;
2643 if (!err && subarray_set_num(container, sra, "raid_disks",
2644 reshape->after.data_disks +
2645 reshape->parity) < 0)
2646 err = errno;
2647 if (err) {
2648 pr_err("Cannot set device shape for %s\n",
2649 devname);
2650
2651 if (err == EBUSY &&
2652 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2653 cont_err("Bitmap must be removed before"
2654 " shape can be changed\n");
2655 goto release;
2656 }
2657 }
2658 return 0;
2659 release:
2660 return -1;
2661 }
2662
2663 static int impose_level(int fd, int level, char *devname, int verbose)
2664 {
2665 char *c;
2666 struct mdu_array_info_s array;
2667 struct mdinfo info;
2668 sysfs_init(&info, fd, NULL);
2669
2670 ioctl(fd, GET_ARRAY_INFO, &array);
2671 if (level == 0 &&
2672 (array.level >= 4 && array.level <= 6)) {
2673 /* To convert to RAID0 we need to fail and
2674 * remove any non-data devices. */
2675 int found = 0;
2676 int d;
2677 int data_disks = array.raid_disks - 1;
2678 if (array.level == 6)
2679 data_disks -= 1;
2680 if (array.level == 5 &&
2681 array.layout != ALGORITHM_PARITY_N)
2682 return -1;
2683 if (array.level == 6 &&
2684 array.layout != ALGORITHM_PARITY_N_6)
2685 return -1;
2686 sysfs_set_str(&info, NULL,"sync_action", "idle");
2687 /* First remove any spares so no recovery starts */
2688 for (d = 0, found = 0;
2689 d < MAX_DISKS && found < array.nr_disks;
2690 d++) {
2691 mdu_disk_info_t disk;
2692 disk.number = d;
2693 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2694 continue;
2695 if (disk.major == 0 && disk.minor == 0)
2696 continue;
2697 found++;
2698 if ((disk.state & (1 << MD_DISK_ACTIVE))
2699 && disk.raid_disk < data_disks)
2700 /* keep this */
2701 continue;
2702 ioctl(fd, HOT_REMOVE_DISK,
2703 makedev(disk.major, disk.minor));
2704 }
2705 /* Now fail anything left */
2706 ioctl(fd, GET_ARRAY_INFO, &array);
2707 for (d = 0, found = 0;
2708 d < MAX_DISKS && found < array.nr_disks;
2709 d++) {
2710 int cnt;
2711 mdu_disk_info_t disk;
2712 disk.number = d;
2713 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2714 continue;
2715 if (disk.major == 0 && disk.minor == 0)
2716 continue;
2717 found++;
2718 if ((disk.state & (1 << MD_DISK_ACTIVE))
2719 && disk.raid_disk < data_disks)
2720 /* keep this */
2721 continue;
2722 ioctl(fd, SET_DISK_FAULTY,
2723 makedev(disk.major, disk.minor));
2724 cnt = 5;
2725 while (ioctl(fd, HOT_REMOVE_DISK,
2726 makedev(disk.major, disk.minor)) < 0
2727 && errno == EBUSY
2728 && cnt--) {
2729 usleep(10000);
2730 }
2731 }
2732 }
2733 c = map_num(pers, level);
2734 if (c) {
2735 int err = sysfs_set_str(&info, NULL, "level", c);
2736 if (err) {
2737 err = errno;
2738 pr_err("%s: could not set level to %s\n",
2739 devname, c);
2740 if (err == EBUSY &&
2741 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2742 cont_err("Bitmap must be removed"
2743 " before level can be changed\n");
2744 return err;
2745 }
2746 if (verbose >= 0)
2747 pr_err("level of %s changed to %s\n",
2748 devname, c);
2749 }
2750 return 0;
2751 }
2752
2753 int sigterm = 0;
2754 static void catch_term(int sig)
2755 {
2756 sigterm = 1;
2757 }
2758
2759 static int continue_via_systemd(char *devnm)
2760 {
2761 int skipped, i, pid, status;
2762 char pathbuf[1024];
2763 /* In a systemd/udev world, it is best to get systemd to
2764 * run "mdadm --grow --continue" rather than running in the
2765 * background.
2766 */
2767 switch(fork()) {
2768 case 0:
2769 /* FIXME yuk. CLOSE_EXEC?? */
2770 skipped = 0;
2771 for (i = 3; skipped < 20; i++)
2772 if (close(i) < 0)
2773 skipped++;
2774 else
2775 skipped = 0;
2776
2777 /* Don't want to see error messages from
2778 * systemctl. If the service doesn't exist,
2779 * we fork ourselves.
2780 */
2781 close(2);
2782 open("/dev/null", O_WRONLY);
2783 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2784 devnm);
2785 status = execl("/usr/bin/systemctl", "systemctl",
2786 "start",
2787 pathbuf, NULL);
2788 status = execl("/bin/systemctl", "systemctl", "start",
2789 pathbuf, NULL);
2790 exit(1);
2791 case -1: /* Just do it ourselves. */
2792 break;
2793 default: /* parent - good */
2794 pid = wait(&status);
2795 if (pid >= 0 && status == 0)
2796 return 1;
2797 }
2798 return 0;
2799 }
2800
2801 static int reshape_array(char *container, int fd, char *devname,
2802 struct supertype *st, struct mdinfo *info,
2803 int force, struct mddev_dev *devlist,
2804 unsigned long long data_offset,
2805 char *backup_file, int verbose, int forked,
2806 int restart, int freeze_reshape)
2807 {
2808 struct reshape reshape;
2809 int spares_needed;
2810 char *msg;
2811 int orig_level = UnSet;
2812 int odisks;
2813 int delayed;
2814
2815 struct mdu_array_info_s array;
2816 char *c;
2817
2818 struct mddev_dev *dv;
2819 int added_disks;
2820
2821 int *fdlist = NULL;
2822 unsigned long long *offsets = NULL;
2823 int d;
2824 int nrdisks;
2825 int err;
2826 unsigned long blocks;
2827 unsigned long long array_size;
2828 int done;
2829 struct mdinfo *sra = NULL;
2830 char buf[20];
2831
2832 /* when reshaping a RAID0, the component_size might be zero.
2833 * So try to fix that up.
2834 */
2835 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2836 dprintf("Cannot get array information.\n");
2837 goto release;
2838 }
2839 if (array.level == 0 && info->component_size == 0) {
2840 get_dev_size(fd, NULL, &array_size);
2841 info->component_size = array_size / array.raid_disks;
2842 }
2843
2844 if (array.level == 10)
2845 /* Need space_after info */
2846 get_space_after(fd, st, info);
2847
2848 if (info->reshape_active) {
2849 int new_level = info->new_level;
2850 info->new_level = UnSet;
2851 if (info->delta_disks > 0)
2852 info->array.raid_disks -= info->delta_disks;
2853 msg = analyse_change(devname, info, &reshape);
2854 info->new_level = new_level;
2855 if (info->delta_disks > 0)
2856 info->array.raid_disks += info->delta_disks;
2857 if (!restart)
2858 /* Make sure the array isn't read-only */
2859 ioctl(fd, RESTART_ARRAY_RW, 0);
2860 } else
2861 msg = analyse_change(devname, info, &reshape);
2862 if (msg) {
2863 /* if msg == "", error has already been printed */
2864 if (msg[0])
2865 pr_err("%s\n", msg);
2866 goto release;
2867 }
2868 if (restart &&
2869 (reshape.level != info->array.level ||
2870 reshape.before.layout != info->array.layout ||
2871 reshape.before.data_disks + reshape.parity
2872 != info->array.raid_disks - max(0, info->delta_disks))) {
2873 pr_err("reshape info is not in native format -"
2874 " cannot continue.\n");
2875 goto release;
2876 }
2877
2878 if (st->ss->external && restart && (info->reshape_progress == 0) &&
2879 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
2880 (strncmp(buf, "reshape", 7) == 0))) {
2881 /* When reshape is restarted from '0', very begin of array
2882 * it is possible that for external metadata reshape and array
2883 * configuration doesn't happen.
2884 * Check if md has the same opinion, and reshape is restarted
2885 * from 0. If so, this is regular reshape start after reshape
2886 * switch in metadata to next array only.
2887 */
2888 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2889 (info->reshape_progress == 0))
2890 restart = 0;
2891 }
2892 if (restart) {
2893 /* reshape already started. just skip to monitoring the reshape */
2894 if (reshape.backup_blocks == 0)
2895 return 0;
2896 if (restart & RESHAPE_NO_BACKUP)
2897 return 0;
2898
2899 /* Need 'sra' down at 'started:' */
2900 sra = sysfs_read(fd, NULL,
2901 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2902 GET_CACHE);
2903 if (!sra) {
2904 pr_err("%s: Cannot get array details from sysfs\n",
2905 devname);
2906 goto release;
2907 }
2908
2909 if (!backup_file)
2910 backup_file = locate_backup(sra->sys_name);
2911
2912 goto started;
2913 }
2914 /* The container is frozen but the array may not be.
2915 * So freeze the array so spares don't get put to the wrong use
2916 * FIXME there should probably be a cleaner separation between
2917 * freeze_array and freeze_container.
2918 */
2919 sysfs_freeze_array(info);
2920 /* Check we have enough spares to not be degraded */
2921 added_disks = 0;
2922 for (dv = devlist; dv ; dv=dv->next)
2923 added_disks++;
2924 spares_needed = max(reshape.before.data_disks,
2925 reshape.after.data_disks)
2926 + reshape.parity - array.raid_disks;
2927
2928 if (!force &&
2929 info->new_level > 1 && info->array.level > 1 &&
2930 spares_needed > info->array.spare_disks + added_disks) {
2931 pr_err("Need %d spare%s to avoid degraded array,"
2932 " and only have %d.\n"
2933 " Use --force to over-ride this check.\n",
2934 spares_needed,
2935 spares_needed == 1 ? "" : "s",
2936 info->array.spare_disks + added_disks);
2937 goto release;
2938 }
2939 /* Check we have enough spares to not fail */
2940 spares_needed = max(reshape.before.data_disks,
2941 reshape.after.data_disks)
2942 - array.raid_disks;
2943 if ((info->new_level > 1 || info->new_level == 0) &&
2944 spares_needed > info->array.spare_disks +added_disks) {
2945 pr_err("Need %d spare%s to create working array,"
2946 " and only have %d.\n",
2947 spares_needed,
2948 spares_needed == 1 ? "" : "s",
2949 info->array.spare_disks + added_disks);
2950 goto release;
2951 }
2952
2953 if (reshape.level != array.level) {
2954 int err = impose_level(fd, reshape.level, devname, verbose);
2955 if (err)
2956 goto release;
2957 info->new_layout = UnSet; /* after level change,
2958 * layout is meaningless */
2959 orig_level = array.level;
2960 sysfs_freeze_array(info);
2961
2962 if (reshape.level > 0 && st->ss->external) {
2963 /* make sure mdmon is aware of the new level */
2964 if (mdmon_running(container))
2965 flush_mdmon(container);
2966
2967 if (!mdmon_running(container))
2968 start_mdmon(container);
2969 ping_monitor(container);
2970 if (mdmon_running(container) &&
2971 st->update_tail == NULL)
2972 st->update_tail = &st->updates;
2973 }
2974 }
2975 /* ->reshape_super might have chosen some spares from the
2976 * container that it wants to be part of the new array.
2977 * We can collect them with ->container_content and give
2978 * them to the kernel.
2979 */
2980 if (st->ss->reshape_super && st->ss->container_content) {
2981 char *subarray = strchr(info->text_version+1, '/')+1;
2982 struct mdinfo *info2 =
2983 st->ss->container_content(st, subarray);
2984 struct mdinfo *d;
2985
2986 if (info2) {
2987 sysfs_init(info2, fd, st->devnm);
2988 /* When increasing number of devices, we need to set
2989 * new raid_disks before adding these, or they might
2990 * be rejected.
2991 */
2992 if (reshape.backup_blocks &&
2993 reshape.after.data_disks > reshape.before.data_disks)
2994 subarray_set_num(container, info2, "raid_disks",
2995 reshape.after.data_disks +
2996 reshape.parity);
2997 for (d = info2->devs; d; d = d->next) {
2998 if (d->disk.state == 0 &&
2999 d->disk.raid_disk >= 0) {
3000 /* This is a spare that wants to
3001 * be part of the array.
3002 */
3003 add_disk(fd, st, info2, d);
3004 }
3005 }
3006 sysfs_free(info2);
3007 }
3008 }
3009 /* We might have been given some devices to add to the
3010 * array. Now that the array has been changed to the right
3011 * level and frozen, we can safely add them.
3012 */
3013 if (devlist)
3014 Manage_subdevs(devname, fd, devlist, verbose,
3015 0,NULL, 0);
3016
3017 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3018 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3019 if (reshape.backup_blocks == 0) {
3020 /* No restriping needed, but we might need to impose
3021 * some more changes: layout, raid_disks, chunk_size
3022 */
3023 /* read current array info */
3024 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
3025 dprintf("Cannot get array information.\n");
3026 goto release;
3027 }
3028 /* compare current array info with new values and if
3029 * it is different update them to new */
3030 if (info->new_layout != UnSet &&
3031 info->new_layout != array.layout) {
3032 array.layout = info->new_layout;
3033 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3034 pr_err("failed to set new layout\n");
3035 goto release;
3036 } else if (verbose >= 0)
3037 printf("layout for %s set to %d\n",
3038 devname, array.layout);
3039 }
3040 if (info->delta_disks != UnSet &&
3041 info->delta_disks != 0 &&
3042 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3043 array.raid_disks += info->delta_disks;
3044 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3045 pr_err("failed to set raid disks\n");
3046 goto release;
3047 } else if (verbose >= 0) {
3048 printf("raid_disks for %s set to %d\n",
3049 devname, array.raid_disks);
3050 }
3051 }
3052 if (info->new_chunk != 0 &&
3053 info->new_chunk != array.chunk_size) {
3054 if (sysfs_set_num(info, NULL,
3055 "chunk_size", info->new_chunk) != 0) {
3056 pr_err("failed to set chunk size\n");
3057 goto release;
3058 } else if (verbose >= 0)
3059 printf("chunk size for %s set to %d\n",
3060 devname, array.chunk_size);
3061 }
3062 unfreeze(st);
3063 return 0;
3064 }
3065
3066 /*
3067 * There are three possibilities.
3068 * 1/ The array will shrink.
3069 * We need to ensure the reshape will pause before reaching
3070 * the 'critical section'. We also need to fork and wait for
3071 * that to happen. When it does we
3072 * suspend/backup/complete/unfreeze
3073 *
3074 * 2/ The array will not change size.
3075 * This requires that we keep a backup of a sliding window
3076 * so that we can restore data after a crash. So we need
3077 * to fork and monitor progress.
3078 * In future we will allow the data_offset to change, so
3079 * a sliding backup becomes unnecessary.
3080 *
3081 * 3/ The array will grow. This is relatively easy.
3082 * However the kernel's restripe routines will cheerfully
3083 * overwrite some early data before it is safe. So we
3084 * need to make a backup of the early parts of the array
3085 * and be ready to restore it if rebuild aborts very early.
3086 * For externally managed metadata, we still need a forked
3087 * child to monitor the reshape and suspend IO over the region
3088 * that is being reshaped.
3089 *
3090 * We backup data by writing it to one spare, or to a
3091 * file which was given on command line.
3092 *
3093 * In each case, we first make sure that storage is available
3094 * for the required backup.
3095 * Then we:
3096 * - request the shape change.
3097 * - fork to handle backup etc.
3098 */
3099 /* Check that we can hold all the data */
3100 get_dev_size(fd, NULL, &array_size);
3101 if (reshape.new_size < (array_size/512)) {
3102 pr_err("this change will reduce the size of the array.\n"
3103 " use --grow --array-size first to truncate array.\n"
3104 " e.g. mdadm --grow %s --array-size %llu\n",
3105 devname, reshape.new_size/2);
3106 goto release;
3107 }
3108
3109 if (array.level == 10) {
3110 /* Reshaping RAID10 does not require any data backup by
3111 * user-space. Instead it requires that the data_offset
3112 * is changed to avoid the need for backup.
3113 * So this is handled very separately
3114 */
3115 if (restart)
3116 /* Nothing to do. */
3117 return 0;
3118 return raid10_reshape(container, fd, devname, st, info,
3119 &reshape, data_offset,
3120 force, verbose);
3121 }
3122 sra = sysfs_read(fd, NULL,
3123 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3124 GET_CACHE);
3125 if (!sra) {
3126 pr_err("%s: Cannot get array details from sysfs\n",
3127 devname);
3128 goto release;
3129 }
3130
3131 if (!backup_file)
3132 switch(set_new_data_offset(sra, st, devname,
3133 reshape.after.data_disks - reshape.before.data_disks,
3134 data_offset,
3135 reshape.min_offset_change, 1)) {
3136 case -1:
3137 goto release;
3138 case 0:
3139 /* Updated data_offset, so it's easy now */
3140 update_cache_size(container, sra, info,
3141 min(reshape.before.data_disks,
3142 reshape.after.data_disks),
3143 reshape.backup_blocks);
3144
3145 /* Right, everything seems fine. Let's kick things off.
3146 */
3147 sync_metadata(st);
3148
3149 if (impose_reshape(sra, info, st, fd, restart,
3150 devname, container, &reshape) < 0)
3151 goto release;
3152 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3153 pr_err("Failed to initiate reshape!\n");
3154 goto release;
3155 }
3156 if (info->new_level == reshape.level)
3157 return 0;
3158 /* need to adjust level when reshape completes */
3159 switch(fork()) {
3160 case -1: /* ignore error, but don't wait */
3161 return 0;
3162 default: /* parent */
3163 return 0;
3164 case 0:
3165 map_fork();
3166 break;
3167 }
3168 close(fd);
3169 wait_reshape(sra);
3170 fd = open_dev(sra->sys_name);
3171 if (fd >= 0)
3172 impose_level(fd, info->new_level, devname, verbose);
3173 return 0;
3174 case 1: /* Couldn't set data_offset, try the old way */
3175 if (data_offset != INVALID_SECTORS) {
3176 pr_err("Cannot update data_offset on this array\n");
3177 goto release;
3178 }
3179 break;
3180 }
3181
3182 started:
3183 /* Decide how many blocks (sectors) for a reshape
3184 * unit. The number we have so far is just a minimum
3185 */
3186 blocks = reshape.backup_blocks;
3187 if (reshape.before.data_disks ==
3188 reshape.after.data_disks) {
3189 /* Make 'blocks' bigger for better throughput, but
3190 * not so big that we reject it below.
3191 * Try for 16 megabytes
3192 */
3193 while (blocks * 32 < sra->component_size &&
3194 blocks < 16*1024*2)
3195 blocks *= 2;
3196 } else
3197 pr_err("Need to backup %luK of critical "
3198 "section..\n", blocks/2);
3199
3200 if (blocks >= sra->component_size/2) {
3201 pr_err("%s: Something wrong"
3202 " - reshape aborted\n",
3203 devname);
3204 goto release;
3205 }
3206
3207 /* Now we need to open all these devices so we can read/write.
3208 */
3209 nrdisks = max(reshape.before.data_disks,
3210 reshape.after.data_disks) + reshape.parity
3211 + sra->array.spare_disks;
3212 fdlist = xcalloc((1+nrdisks), sizeof(int));
3213 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3214
3215 odisks = reshape.before.data_disks + reshape.parity;
3216 d = reshape_prepare_fdlist(devname, sra, odisks,
3217 nrdisks, blocks, backup_file,
3218 fdlist, offsets);
3219 if (d < 0) {
3220 goto release;
3221 }
3222 if ((st->ss->manage_reshape == NULL) ||
3223 (st->ss->recover_backup == NULL)) {
3224 if (backup_file == NULL) {
3225 if (reshape.after.data_disks <=
3226 reshape.before.data_disks) {
3227 pr_err("%s: Cannot grow - need backup-file\n",
3228 devname);
3229 pr_err(" Please provide one with \"--backup=...\"\n");
3230 goto release;
3231 } else if (sra->array.spare_disks == 0) {
3232 pr_err("%s: Cannot grow - "
3233 "need a spare or backup-file to backup "
3234 "critical section\n", devname);
3235 goto release;
3236 }
3237 } else {
3238 if (!reshape_open_backup_file(backup_file, fd, devname,
3239 (signed)blocks,
3240 fdlist+d, offsets+d,
3241 sra->sys_name,
3242 restart)) {
3243 goto release;
3244 }
3245 d++;
3246 }
3247 }
3248
3249 update_cache_size(container, sra, info,
3250 min(reshape.before.data_disks, reshape.after.data_disks),
3251 blocks);
3252
3253 /* Right, everything seems fine. Let's kick things off.
3254 * If only changing raid_disks, use ioctl, else use
3255 * sysfs.
3256 */
3257 sync_metadata(st);
3258
3259 if (impose_reshape(sra, info, st, fd, restart,
3260 devname, container, &reshape) < 0)
3261 goto release;
3262
3263 err = start_reshape(sra, restart, reshape.before.data_disks,
3264 reshape.after.data_disks);
3265 if (err) {
3266 pr_err("Cannot %s reshape for %s\n",
3267 restart ? "continue" : "start",
3268 devname);
3269 goto release;
3270 }
3271 if (restart)
3272 sysfs_set_str(sra, NULL, "array_state", "active");
3273 if (freeze_reshape) {
3274 free(fdlist);
3275 free(offsets);
3276 sysfs_free(sra);
3277 pr_err("Reshape has to be continued from"
3278 " location %llu when root filesystem has been mounted.\n",
3279 sra->reshape_progress);
3280 return 1;
3281 }
3282
3283 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3284 if (continue_via_systemd(container ?: sra->sys_name)) {
3285 free(fdlist);
3286 free(offsets);
3287 sysfs_free(sra);
3288 return 0;
3289 }
3290
3291 /* Now we just need to kick off the reshape and watch, while
3292 * handling backups of the data...
3293 * This is all done by a forked background process.
3294 */
3295 switch(forked ? 0 : fork()) {
3296 case -1:
3297 pr_err("Cannot run child to monitor reshape: %s\n",
3298 strerror(errno));
3299 abort_reshape(sra);
3300 goto release;
3301 default:
3302 free(fdlist);
3303 free(offsets);
3304 sysfs_free(sra);
3305 return 0;
3306 case 0:
3307 map_fork();
3308 break;
3309 }
3310
3311 /* If another array on the same devices is busy, the
3312 * reshape will wait for them. This would mean that
3313 * the first section that we suspend will stay suspended
3314 * for a long time. So check on that possibility
3315 * by looking for "DELAYED" in /proc/mdstat, and if found,
3316 * wait a while
3317 */
3318 do {
3319 struct mdstat_ent *mds, *m;
3320 delayed = 0;
3321 mds = mdstat_read(1, 0);
3322 for (m = mds; m; m = m->next)
3323 if (strcmp(m->devnm, sra->sys_name) == 0) {
3324 if (m->resync &&
3325 m->percent == RESYNC_DELAYED)
3326 delayed = 1;
3327 if (m->resync == 0)
3328 /* Haven't started the reshape thread
3329 * yet, wait a bit
3330 */
3331 delayed = 2;
3332 break;
3333 }
3334 free_mdstat(mds);
3335 if (delayed == 1 && get_linux_version() < 3007000) {
3336 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3337 " You might experience problems until other reshapes complete.\n");
3338 delayed = 0;
3339 }
3340 if (delayed)
3341 mdstat_wait(30 - (delayed-1) * 25);
3342 } while (delayed);
3343 mdstat_close();
3344 close(fd);
3345 if (check_env("MDADM_GROW_VERIFY"))
3346 fd = open(devname, O_RDONLY | O_DIRECT);
3347 else
3348 fd = -1;
3349 mlockall(MCL_FUTURE);
3350
3351 signal(SIGTERM, catch_term);
3352
3353 if (st->ss->external) {
3354 /* metadata handler takes it from here */
3355 done = st->ss->manage_reshape(
3356 fd, sra, &reshape, st, blocks,
3357 fdlist, offsets,
3358 d - odisks, fdlist+odisks,
3359 offsets+odisks);
3360 } else
3361 done = child_monitor(
3362 fd, sra, &reshape, st, blocks,
3363 fdlist, offsets,
3364 d - odisks, fdlist+odisks,
3365 offsets+odisks);
3366
3367 free(fdlist);
3368 free(offsets);
3369
3370 if (backup_file && done) {
3371 char *bul;
3372 bul = make_backup(sra->sys_name);
3373 if (bul) {
3374 char buf[1024];
3375 int l = readlink(bul, buf, sizeof(buf));
3376 if (l > 0) {
3377 buf[l]=0;
3378 unlink(buf);
3379 }
3380 unlink(bul);
3381 free(bul);
3382 }
3383 unlink(backup_file);
3384 }
3385 if (!done) {
3386 abort_reshape(sra);
3387 goto out;
3388 }
3389
3390 if (!st->ss->external &&
3391 !(reshape.before.data_disks != reshape.after.data_disks
3392 && info->custom_array_size) &&
3393 info->new_level == reshape.level &&
3394 !forked) {
3395 /* no need to wait for the reshape to finish as
3396 * there is nothing more to do.
3397 */
3398 sysfs_free(sra);
3399 exit(0);
3400 }
3401 wait_reshape(sra);
3402
3403 if (st->ss->external) {
3404 /* Re-load the metadata as much could have changed */
3405 int cfd = open_dev(st->container_devnm);
3406 if (cfd >= 0) {
3407 flush_mdmon(container);
3408 st->ss->free_super(st);
3409 st->ss->load_container(st, cfd, container);
3410 close(cfd);
3411 }
3412 }
3413
3414 /* set new array size if required customer_array_size is used
3415 * by this metadata.
3416 */
3417 if (reshape.before.data_disks !=
3418 reshape.after.data_disks &&
3419 info->custom_array_size)
3420 set_array_size(st, info, info->text_version);
3421
3422 if (info->new_level != reshape.level) {
3423 if (fd < 0)
3424 fd = open(devname, O_RDONLY);
3425 impose_level(fd, info->new_level, devname, verbose);
3426 close(fd);
3427 if (info->new_level == 0)
3428 st->update_tail = NULL;
3429 }
3430 out:
3431 sysfs_free(sra);
3432 if (forked)
3433 return 0;
3434 unfreeze(st);
3435 exit(0);
3436
3437 release:
3438 free(fdlist);
3439 free(offsets);
3440 if (orig_level != UnSet && sra) {
3441 c = map_num(pers, orig_level);
3442 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3443 pr_err("aborting level change\n");
3444 }
3445 sysfs_free(sra);
3446 if (!forked)
3447 unfreeze(st);
3448 return 1;
3449 }
3450
3451 /* mdfd handle is passed to be closed in child process (after fork).
3452 */
3453 int reshape_container(char *container, char *devname,
3454 int mdfd,
3455 struct supertype *st,
3456 struct mdinfo *info,
3457 int force,
3458 char *backup_file, int verbose,
3459 int forked, int restart, int freeze_reshape)
3460 {
3461 struct mdinfo *cc = NULL;
3462 int rv = restart;
3463 char last_devnm[32] = "";
3464
3465 /* component_size is not meaningful for a container,
3466 * so pass '0' meaning 'no change'
3467 */
3468 if (!restart &&
3469 reshape_super(st, 0, info->new_level,
3470 info->new_layout, info->new_chunk,
3471 info->array.raid_disks, info->delta_disks,
3472 backup_file, devname, APPLY_METADATA_CHANGES,
3473 verbose)) {
3474 unfreeze(st);
3475 return 1;
3476 }
3477
3478 sync_metadata(st);
3479
3480 /* ping monitor to be sure that update is on disk
3481 */
3482 ping_monitor(container);
3483
3484 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3485 if (continue_via_systemd(container))
3486 return 0;
3487
3488 switch (forked ? 0 : fork()) {
3489 case -1: /* error */
3490 perror("Cannot fork to complete reshape\n");
3491 unfreeze(st);
3492 return 1;
3493 default: /* parent */
3494 if (!freeze_reshape)
3495 printf(Name ": multi-array reshape continues"
3496 " in background\n");
3497 return 0;
3498 case 0: /* child */
3499 map_fork();
3500 break;
3501 }
3502
3503 /* close unused handle in child process
3504 */
3505 if (mdfd > -1)
3506 close(mdfd);
3507
3508 while(1) {
3509 /* For each member array with reshape_active,
3510 * we need to perform the reshape.
3511 * We pick the first array that needs reshaping and
3512 * reshape it. reshape_array() will re-read the metadata
3513 * so the next time through a different array should be
3514 * ready for reshape.
3515 * It is possible that the 'different' array will not
3516 * be assembled yet. In that case we simple exit.
3517 * When it is assembled, the mdadm which assembles it
3518 * will take over the reshape.
3519 */
3520 struct mdinfo *content;
3521 int fd;
3522 struct mdstat_ent *mdstat;
3523 char *adev;
3524 int devid;
3525
3526 sysfs_free(cc);
3527
3528 cc = st->ss->container_content(st, NULL);
3529
3530 for (content = cc; content ; content = content->next) {
3531 char *subarray;
3532 if (!content->reshape_active)
3533 continue;
3534
3535 subarray = strchr(content->text_version+1, '/')+1;
3536 mdstat = mdstat_by_subdev(subarray, container);
3537 if (!mdstat)
3538 continue;
3539 if (mdstat->active == 0) {
3540 pr_err("Skipping inactive array %s.\n",
3541 mdstat->devnm);
3542 free_mdstat(mdstat);
3543 mdstat = NULL;
3544 continue;
3545 }
3546 break;
3547 }
3548 if (!content)
3549 break;
3550
3551 devid = devnm2devid(mdstat->devnm);
3552 adev = map_dev(major(devid), minor(devid), 0);
3553 if (!adev)
3554 adev = content->text_version;
3555
3556 fd = open_dev(mdstat->devnm);
3557 if (fd < 0) {
3558 printf(Name ": Device %s cannot be opened for reshape.",
3559 adev);
3560 break;
3561 }
3562
3563 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3564 /* Do not allow for multiple reshape_array() calls for
3565 * the same array.
3566 * It can happen when reshape_array() returns without
3567 * error, when reshape is not finished (wrong reshape
3568 * starting/continuation conditions). Mdmon doesn't
3569 * switch to next array in container and reentry
3570 * conditions for the same array occur.
3571 * This is possibly interim until the behaviour of
3572 * reshape_array is resolved().
3573 */
3574 printf(Name ": Multiple reshape execution detected for "
3575 "device %s.", adev);
3576 close(fd);
3577 break;
3578 }
3579 strcpy(last_devnm, mdstat->devnm);
3580
3581 sysfs_init(content, fd, mdstat->devnm);
3582
3583 if (mdmon_running(container))
3584 flush_mdmon(container);
3585
3586 rv = reshape_array(container, fd, adev, st,
3587 content, force, NULL, INVALID_SECTORS,
3588 backup_file, verbose, 1, restart,
3589 freeze_reshape);
3590 close(fd);
3591
3592 if (freeze_reshape) {
3593 sysfs_free(cc);
3594 exit(0);
3595 }
3596
3597 restart = 0;
3598 if (rv)
3599 break;
3600
3601 if (mdmon_running(container))
3602 flush_mdmon(container);
3603 }
3604 if (!rv)
3605 unfreeze(st);
3606 sysfs_free(cc);
3607 exit(0);
3608 }
3609
3610 /*
3611 * We run a child process in the background which performs the following
3612 * steps:
3613 * - wait for resync to reach a certain point
3614 * - suspend io to the following section
3615 * - backup that section
3616 * - allow resync to proceed further
3617 * - resume io
3618 * - discard the backup.
3619 *
3620 * When are combined in slightly different ways in the three cases.
3621 * Grow:
3622 * - suspend/backup/allow/wait/resume/discard
3623 * Shrink:
3624 * - allow/wait/suspend/backup/allow/wait/resume/discard
3625 * same-size:
3626 * - wait/resume/discard/suspend/backup/allow
3627 *
3628 * suspend/backup/allow always come together
3629 * wait/resume/discard do too.
3630 * For the same-size case we have two backups to improve flow.
3631 *
3632 */
3633
3634 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3635 unsigned long long backup_point,
3636 unsigned long long wait_point,
3637 unsigned long long *suspend_point,
3638 unsigned long long *reshape_completed, int *frozen)
3639 {
3640 /* This function is called repeatedly by the reshape manager.
3641 * It determines how much progress can safely be made and allows
3642 * that progress.
3643 * - 'info' identifies the array and particularly records in
3644 * ->reshape_progress the metadata's knowledge of progress
3645 * This is a sector offset from the start of the array
3646 * of the next array block to be relocated. This number
3647 * may increase from 0 or decrease from array_size, depending
3648 * on the type of reshape that is happening.
3649 * Note that in contrast, 'sync_completed' is a block count of the
3650 * reshape so far. It gives the distance between the start point
3651 * (head or tail of device) and the next place that data will be
3652 * written. It always increases.
3653 * - 'reshape' is the structure created by analyse_change
3654 * - 'backup_point' shows how much the metadata manager has backed-up
3655 * data. For reshapes with increasing progress, it is the next address
3656 * to be backed up, previous addresses have been backed-up. For
3657 * decreasing progress, it is the earliest address that has been
3658 * backed up - later address are also backed up.
3659 * So addresses between reshape_progress and backup_point are
3660 * backed up providing those are in the 'correct' order.
3661 * - 'wait_point' is an array address. When reshape_completed
3662 * passes this point, progress_reshape should return. It might
3663 * return earlier if it determines that ->reshape_progress needs
3664 * to be updated or further backup is needed.
3665 * - suspend_point is maintained by progress_reshape and the caller
3666 * should not touch it except to initialise to zero.
3667 * It is an array address and it only increases in 2.6.37 and earlier.
3668 * This makes it difficult to handle reducing reshapes with
3669 * external metadata.
3670 * However: it is similar to backup_point in that it records the
3671 * other end of a suspended region from reshape_progress.
3672 * it is moved to extend the region that is safe to backup and/or
3673 * reshape
3674 * - reshape_completed is read from sysfs and returned. The caller
3675 * should copy this into ->reshape_progress when it has reason to
3676 * believe that the metadata knows this, and any backup outside this
3677 * has been erased.
3678 *
3679 * Return value is:
3680 * 1 if more data from backup_point - but only as far as suspend_point,
3681 * should be backed up
3682 * 0 if things are progressing smoothly
3683 * -1 if the reshape is finished because it is all done,
3684 * -2 if the reshape is finished due to an error.
3685 */
3686
3687 int advancing = (reshape->after.data_disks
3688 >= reshape->before.data_disks);
3689 unsigned long long need_backup; /* All data between start of array and
3690 * here will at some point need to
3691 * be backed up.
3692 */
3693 unsigned long long read_offset, write_offset;
3694 unsigned long long write_range;
3695 unsigned long long max_progress, target, completed;
3696 unsigned long long array_size = (info->component_size
3697 * reshape->before.data_disks);
3698 int fd;
3699 char buf[20];
3700
3701 /* First, we unsuspend any region that is now known to be safe.
3702 * If suspend_point is on the 'wrong' side of reshape_progress, then
3703 * we don't have or need suspension at the moment. This is true for
3704 * native metadata when we don't need to back-up.
3705 */
3706 if (advancing) {
3707 if (info->reshape_progress <= *suspend_point)
3708 sysfs_set_num(info, NULL, "suspend_lo",
3709 info->reshape_progress);
3710 } else {
3711 /* Note: this won't work in 2.6.37 and before.
3712 * Something somewhere should make sure we don't need it!
3713 */
3714 if (info->reshape_progress >= *suspend_point)
3715 sysfs_set_num(info, NULL, "suspend_hi",
3716 info->reshape_progress);
3717 }
3718
3719 /* Now work out how far it is safe to progress.
3720 * If the read_offset for ->reshape_progress is less than
3721 * 'blocks' beyond the write_offset, we can only progress as far
3722 * as a backup.
3723 * Otherwise we can progress until the write_offset for the new location
3724 * reaches (within 'blocks' of) the read_offset at the current location.
3725 * However that region must be suspended unless we are using native
3726 * metadata.
3727 * If we need to suspend more, we limit it to 128M per device, which is
3728 * rather arbitrary and should be some time-based calculation.
3729 */
3730 read_offset = info->reshape_progress / reshape->before.data_disks;
3731 write_offset = info->reshape_progress / reshape->after.data_disks;
3732 write_range = info->new_chunk/512;
3733 if (reshape->before.data_disks == reshape->after.data_disks)
3734 need_backup = array_size;
3735 else
3736 need_backup = reshape->backup_blocks;
3737 if (advancing) {
3738 if (read_offset < write_offset + write_range)
3739 max_progress = backup_point;
3740 else
3741 max_progress =
3742 read_offset *
3743 reshape->after.data_disks;
3744 } else {
3745 if (read_offset > write_offset - write_range)
3746 /* Can only progress as far as has been backed up,
3747 * which must be suspended */
3748 max_progress = backup_point;
3749 else if (info->reshape_progress <= need_backup)
3750 max_progress = backup_point;
3751 else {
3752 if (info->array.major_version >= 0)
3753 /* Can progress until backup is needed */
3754 max_progress = need_backup;
3755 else {
3756 /* Can progress until metadata update is required */
3757 max_progress =
3758 read_offset *
3759 reshape->after.data_disks;
3760 /* but data must be suspended */
3761 if (max_progress < *suspend_point)
3762 max_progress = *suspend_point;
3763 }
3764 }
3765 }
3766
3767 /* We know it is safe to progress to 'max_progress' providing
3768 * it is suspended or we are using native metadata.
3769 * Consider extending suspend_point 128M per device if it
3770 * is less than 64M per device beyond reshape_progress.
3771 * But always do a multiple of 'blocks'
3772 * FIXME this is too big - it takes to long to complete
3773 * this much.
3774 */
3775 target = 64*1024*2 * min(reshape->before.data_disks,
3776 reshape->after.data_disks);
3777 target /= reshape->backup_blocks;
3778 if (target < 2)
3779 target = 2;
3780 target *= reshape->backup_blocks;
3781
3782 /* For externally managed metadata we always need to suspend IO to
3783 * the area being reshaped so we regularly push suspend_point forward.
3784 * For native metadata we only need the suspend if we are going to do
3785 * a backup.
3786 */
3787 if (advancing) {
3788 if ((need_backup > info->reshape_progress
3789 || info->array.major_version < 0) &&
3790 *suspend_point < info->reshape_progress + target) {
3791 if (need_backup < *suspend_point + 2 * target)
3792 *suspend_point = need_backup;
3793 else if (*suspend_point + 2 * target < array_size)
3794 *suspend_point += 2 * target;
3795 else
3796 *suspend_point = array_size;
3797 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3798 if (max_progress > *suspend_point)
3799 max_progress = *suspend_point;
3800 }
3801 } else {
3802 if (info->array.major_version >= 0) {
3803 /* Only need to suspend when about to backup */
3804 if (info->reshape_progress < need_backup * 2 &&
3805 *suspend_point > 0) {
3806 *suspend_point = 0;
3807 sysfs_set_num(info, NULL, "suspend_lo", 0);
3808 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3809 }
3810 } else {
3811 /* Need to suspend continually */
3812 if (info->reshape_progress < *suspend_point)
3813 *suspend_point = info->reshape_progress;
3814 if (*suspend_point + target < info->reshape_progress)
3815 /* No need to move suspend region yet */;
3816 else {
3817 if (*suspend_point >= 2 * target)
3818 *suspend_point -= 2 * target;
3819 else
3820 *suspend_point = 0;
3821 sysfs_set_num(info, NULL, "suspend_lo",
3822 *suspend_point);
3823 }
3824 if (max_progress < *suspend_point)
3825 max_progress = *suspend_point;
3826 }
3827 }
3828
3829 /* now set sync_max to allow that progress. sync_max, like
3830 * sync_completed is a count of sectors written per device, so
3831 * we find the difference between max_progress and the start point,
3832 * and divide that by after.data_disks to get a sync_max
3833 * number.
3834 * At the same time we convert wait_point to a similar number
3835 * for comparing against sync_completed.
3836 */
3837 /* scale down max_progress to per_disk */
3838 max_progress /= reshape->after.data_disks;
3839 /* Round to chunk size as some kernels give an erroneously high number */
3840 max_progress /= info->new_chunk/512;
3841 max_progress *= info->new_chunk/512;
3842 /* And round to old chunk size as the kernel wants that */
3843 max_progress /= info->array.chunk_size/512;
3844 max_progress *= info->array.chunk_size/512;
3845 /* Limit progress to the whole device */
3846 if (max_progress > info->component_size)
3847 max_progress = info->component_size;
3848 wait_point /= reshape->after.data_disks;
3849 if (!advancing) {
3850 /* switch from 'device offset' to 'processed block count' */
3851 max_progress = info->component_size - max_progress;
3852 wait_point = info->component_size - wait_point;
3853 }
3854
3855 if (!*frozen)
3856 sysfs_set_num(info, NULL, "sync_max", max_progress);
3857
3858 /* Now wait. If we have already reached the point that we were
3859 * asked to wait to, don't wait at all, else wait for any change.
3860 * We need to select on 'sync_completed' as that is the place that
3861 * notifications happen, but we are really interested in
3862 * 'reshape_position'
3863 */
3864 fd = sysfs_get_fd(info, NULL, "sync_completed");
3865 if (fd < 0)
3866 goto check_progress;
3867
3868 if (sysfs_fd_get_ll(fd, &completed) < 0)
3869 goto check_progress;
3870
3871 while (completed < max_progress && completed < wait_point) {
3872 /* Check that sync_action is still 'reshape' to avoid
3873 * waiting forever on a dead array
3874 */
3875 char action[20];
3876 if (sysfs_get_str(info, NULL, "sync_action",
3877 action, 20) <= 0 ||
3878 strncmp(action, "reshape", 7) != 0)
3879 break;
3880 /* Some kernels reset 'sync_completed' to zero
3881 * before setting 'sync_action' to 'idle'.
3882 * So we need these extra tests.
3883 */
3884 if (completed == 0 && advancing
3885 && info->reshape_progress > 0)
3886 break;
3887 if (completed == 0 && !advancing
3888 && info->reshape_progress < (info->component_size
3889 * reshape->after.data_disks))
3890 break;
3891 sysfs_wait(fd, NULL);
3892 if (sysfs_fd_get_ll(fd, &completed) < 0)
3893 goto check_progress;
3894 }
3895 /* Some kernels reset 'sync_completed' to zero,
3896 * we need to have real point we are in md
3897 */
3898 if (completed == 0)
3899 completed = max_progress;
3900
3901 /* some kernels can give an incorrectly high 'completed' number */
3902 completed /= (info->new_chunk/512);
3903 completed *= (info->new_chunk/512);
3904 /* Convert 'completed' back in to a 'progress' number */
3905 completed *= reshape->after.data_disks;
3906 if (!advancing) {
3907 completed = info->component_size * reshape->after.data_disks
3908 - completed;
3909 }
3910 *reshape_completed = completed;
3911
3912 close(fd);
3913
3914 /* We return the need_backup flag. Caller will decide
3915 * how much - a multiple of ->backup_blocks up to *suspend_point
3916 */
3917 if (advancing)
3918 return need_backup > info->reshape_progress;
3919 else
3920 return need_backup >= info->reshape_progress;
3921
3922 check_progress:
3923 /* if we couldn't read a number from sync_completed, then
3924 * either the reshape did complete, or it aborted.
3925 * We can tell which by checking for 'none' in reshape_position.
3926 * If it did abort, then it might immediately restart if it
3927 * it was just a device failure that leaves us degraded but
3928 * functioning.
3929 */
3930 strcpy(buf, "hi");
3931 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3932 || strncmp(buf, "none", 4) != 0) {
3933 /* The abort might only be temporary. Wait up to 10
3934 * seconds for fd to contain a valid number again.
3935 */
3936 int wait = 10000;
3937 int rv = -2;
3938 unsigned long long new_sync_max;
3939 while (fd >= 0 && rv < 0 && wait > 0) {
3940 if (sysfs_wait(fd, &wait) != 1)
3941 break;
3942 switch (sysfs_fd_get_ll(fd, &completed)) {
3943 case 0:
3944 /* all good again */
3945 rv = 1;
3946 /* If "sync_max" is no longer max_progress
3947 * we need to freeze things
3948 */
3949 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3950 *frozen = (new_sync_max != max_progress);
3951 break;
3952 case -2: /* read error - abort */
3953 wait = 0;
3954 break;
3955 }
3956 }
3957 if (fd >= 0)
3958 close(fd);
3959 return rv; /* abort */
3960 } else {
3961 /* Maybe racing with array shutdown - check state */
3962 if (fd >= 0)
3963 close(fd);
3964 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3965 || strncmp(buf, "inactive", 8) == 0
3966 || strncmp(buf, "clear",5) == 0)
3967 return -2; /* abort */
3968 return -1; /* complete */
3969 }
3970 }
3971
3972 /* FIXME return status is never checked */
3973 static int grow_backup(struct mdinfo *sra,
3974 unsigned long long offset, /* per device */
3975 unsigned long stripes, /* per device, in old chunks */
3976 int *sources, unsigned long long *offsets,
3977 int disks, int chunk, int level, int layout,
3978 int dests, int *destfd, unsigned long long *destoffsets,
3979 int part, int *degraded,
3980 char *buf)
3981 {
3982 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3983 * to storage 'destfd' (offset 'destoffsets'), after first
3984 * suspending IO. Then allow resync to continue
3985 * over the suspended section.
3986 * Use part 'part' of the backup-super-block.
3987 */
3988 int odata = disks;
3989 int rv = 0;
3990 int i;
3991 unsigned long long ll;
3992 int new_degraded;
3993 //printf("offset %llu\n", offset);
3994 if (level >= 4)
3995 odata--;
3996 if (level == 6)
3997 odata--;
3998
3999 /* Check that array hasn't become degraded, else we might backup the wrong data */
4000 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4001 return -1; /* FIXME this error is ignored */
4002 new_degraded = (int)ll;
4003 if (new_degraded != *degraded) {
4004 /* check each device to ensure it is still working */
4005 struct mdinfo *sd;
4006 for (sd = sra->devs ; sd ; sd = sd->next) {
4007 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4008 continue;
4009 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4010 char sbuf[20];
4011 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
4012 strstr(sbuf, "faulty") ||
4013 strstr(sbuf, "in_sync") == NULL) {
4014 /* this device is dead */
4015 sd->disk.state = (1<<MD_DISK_FAULTY);
4016 if (sd->disk.raid_disk >= 0 &&
4017 sources[sd->disk.raid_disk] >= 0) {
4018 close(sources[sd->disk.raid_disk]);
4019 sources[sd->disk.raid_disk] = -1;
4020 }
4021 }
4022 }
4023 }
4024 *degraded = new_degraded;
4025 }
4026 if (part) {
4027 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4028 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4029 } else {
4030 bsb.arraystart = __cpu_to_le64(offset * odata);
4031 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4032 }
4033 if (part)
4034 bsb.magic[15] = '2';
4035 for (i = 0; i < dests; i++)
4036 if (part)
4037 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4038 else
4039 lseek64(destfd[i], destoffsets[i], 0);
4040
4041 rv = save_stripes(sources, offsets,
4042 disks, chunk, level, layout,
4043 dests, destfd,
4044 offset*512*odata, stripes * chunk * odata,
4045 buf);
4046
4047 if (rv)
4048 return rv;
4049 bsb.mtime = __cpu_to_le64(time(0));
4050 for (i = 0; i < dests; i++) {
4051 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4052
4053 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4054 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4055 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4056 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4057
4058 rv = -1;
4059 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4060 != destoffsets[i] - 4096)
4061 break;
4062 if (write(destfd[i], &bsb, 512) != 512)
4063 break;
4064 if (destoffsets[i] > 4096) {
4065 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4066 destoffsets[i]+stripes*chunk*odata)
4067 break;
4068 if (write(destfd[i], &bsb, 512) != 512)
4069 break;
4070 }
4071 fsync(destfd[i]);
4072 rv = 0;
4073 }
4074
4075 return rv;
4076 }
4077
4078 /* in 2.6.30, the value reported by sync_completed can be
4079 * less that it should be by one stripe.
4080 * This only happens when reshape hits sync_max and pauses.
4081 * So allow wait_backup to either extent sync_max further
4082 * than strictly necessary, or return before the
4083 * sync has got quite as far as we would really like.
4084 * This is what 'blocks2' is for.
4085 * The various caller give appropriate values so that
4086 * every works.
4087 */
4088 /* FIXME return value is often ignored */
4089 static int forget_backup(int dests, int *destfd,
4090 unsigned long long *destoffsets,
4091 int part)
4092 {
4093 /*
4094 * Erase backup 'part' (which is 0 or 1)
4095 */
4096 int i;
4097 int rv;
4098
4099 if (part) {
4100 bsb.arraystart2 = __cpu_to_le64(0);
4101 bsb.length2 = __cpu_to_le64(0);
4102 } else {
4103 bsb.arraystart = __cpu_to_le64(0);
4104 bsb.length = __cpu_to_le64(0);
4105 }
4106 bsb.mtime = __cpu_to_le64(time(0));
4107 rv = 0;
4108 for (i = 0; i < dests; i++) {
4109 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4110 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4111 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4112 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4113 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4114 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4115 destoffsets[i]-4096)
4116 rv = -1;
4117 if (rv == 0 &&
4118 write(destfd[i], &bsb, 512) != 512)
4119 rv = -1;
4120 fsync(destfd[i]);
4121 }
4122 return rv;
4123 }
4124
4125 static void fail(char *msg)
4126 {
4127 int rv;
4128 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4129 rv |= (write(2, "\n", 1) != 1);
4130 exit(rv ? 1 : 2);
4131 }
4132
4133 static char *abuf, *bbuf;
4134 static unsigned long long abuflen;
4135 static void validate(int afd, int bfd, unsigned long long offset)
4136 {
4137 /* check that the data in the backup against the array.
4138 * This is only used for regression testing and should not
4139 * be used while the array is active
4140 */
4141 if (afd < 0)
4142 return;
4143 lseek64(bfd, offset - 4096, 0);
4144 if (read(bfd, &bsb2, 512) != 512)
4145 fail("cannot read bsb");
4146 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4147 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4148 fail("first csum bad");
4149 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4150 fail("magic is bad");
4151 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4152 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4153 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4154 fail("second csum bad");
4155
4156 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4157 fail("devstart is wrong");
4158
4159 if (bsb2.length) {
4160 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4161
4162 if (abuflen < len) {
4163 free(abuf);
4164 free(bbuf);
4165 abuflen = len;
4166 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4167 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4168 abuflen = 0;
4169 /* just stop validating on mem-alloc failure */
4170 return;
4171 }
4172 }
4173
4174 lseek64(bfd, offset, 0);
4175 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4176 //printf("len %llu\n", len);
4177 fail("read first backup failed");
4178 }
4179 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4180 if ((unsigned long long)read(afd, abuf, len) != len)
4181 fail("read first from array failed");
4182 if (memcmp(bbuf, abuf, len) != 0) {
4183 #if 0
4184 int i;
4185 printf("offset=%llu len=%llu\n",
4186 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4187 for (i=0; i<len; i++)
4188 if (bbuf[i] != abuf[i]) {
4189 printf("first diff byte %d\n", i);
4190 break;
4191 }
4192 #endif
4193 fail("data1 compare failed");
4194 }
4195 }
4196 if (bsb2.length2) {
4197 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4198
4199 if (abuflen < len) {
4200 free(abuf);
4201 free(bbuf);
4202 abuflen = len;
4203 abuf = xmalloc(abuflen);
4204 bbuf = xmalloc(abuflen);
4205 }
4206
4207 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4208 if ((unsigned long long)read(bfd, bbuf, len) != len)
4209 fail("read second backup failed");
4210 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4211 if ((unsigned long long)read(afd, abuf, len) != len)
4212 fail("read second from array failed");
4213 if (memcmp(bbuf, abuf, len) != 0)
4214 fail("data2 compare failed");
4215 }
4216 }
4217
4218 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4219 struct supertype *st, unsigned long blocks,
4220 int *fds, unsigned long long *offsets,
4221 int dests, int *destfd, unsigned long long *destoffsets)
4222 {
4223 /* Monitor a reshape where backup is being performed using
4224 * 'native' mechanism - either to a backup file, or
4225 * to some space in a spare.
4226 */
4227 char *buf;
4228 int degraded = -1;
4229 unsigned long long speed;
4230 unsigned long long suspend_point, array_size;
4231 unsigned long long backup_point, wait_point;
4232 unsigned long long reshape_completed;
4233 int done = 0;
4234 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4235 int part = 0; /* The next part of the backup area to fill. It may already
4236 * be full, so we need to check */
4237 int level = reshape->level;
4238 int layout = reshape->before.layout;
4239 int data = reshape->before.data_disks;
4240 int disks = reshape->before.data_disks + reshape->parity;
4241 int chunk = sra->array.chunk_size;
4242 struct mdinfo *sd;
4243 unsigned long stripes;
4244 int uuid[4];
4245 int frozen = 0;
4246
4247 /* set up the backup-super-block. This requires the
4248 * uuid from the array.
4249 */
4250 /* Find a superblock */
4251 for (sd = sra->devs; sd; sd = sd->next) {
4252 char *dn;
4253 int devfd;
4254 int ok;
4255 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4256 continue;
4257 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4258 devfd = dev_open(dn, O_RDONLY);
4259 if (devfd < 0)
4260 continue;
4261 ok = st->ss->load_super(st, devfd, NULL);
4262 close(devfd);
4263 if (ok == 0)
4264 break;
4265 }
4266 if (!sd) {
4267 pr_err("Cannot find a superblock\n");
4268 return 0;
4269 }
4270
4271 memset(&bsb, 0, 512);
4272 memcpy(bsb.magic, "md_backup_data-1", 16);
4273 st->ss->uuid_from_super(st, uuid);
4274 memcpy(bsb.set_uuid, uuid, 16);
4275 bsb.mtime = __cpu_to_le64(time(0));
4276 bsb.devstart2 = blocks;
4277
4278 stripes = blocks / (sra->array.chunk_size/512) /
4279 reshape->before.data_disks;
4280
4281 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4282 /* Don't start the 'reshape' */
4283 return 0;
4284 if (reshape->before.data_disks == reshape->after.data_disks) {
4285 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4286 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4287 }
4288
4289 if (increasing) {
4290 array_size = sra->component_size * reshape->after.data_disks;
4291 backup_point = sra->reshape_progress;
4292 suspend_point = 0;
4293 } else {
4294 array_size = sra->component_size * reshape->before.data_disks;
4295 backup_point = reshape->backup_blocks;
4296 suspend_point = array_size;
4297 }
4298
4299 while (!done) {
4300 int rv;
4301
4302 /* Want to return as soon the oldest backup slot can
4303 * be released as that allows us to start backing up
4304 * some more, providing suspend_point has been
4305 * advanced, which it should have.
4306 */
4307 if (increasing) {
4308 wait_point = array_size;
4309 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4310 wait_point = (__le64_to_cpu(bsb.arraystart) +
4311 __le64_to_cpu(bsb.length));
4312 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4313 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4314 __le64_to_cpu(bsb.length2));
4315 } else {
4316 wait_point = 0;
4317 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4318 wait_point = __le64_to_cpu(bsb.arraystart);
4319 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4320 wait_point = __le64_to_cpu(bsb.arraystart2);
4321 }
4322
4323 reshape_completed = sra->reshape_progress;
4324 rv = progress_reshape(sra, reshape,
4325 backup_point, wait_point,
4326 &suspend_point, &reshape_completed,
4327 &frozen);
4328 /* external metadata would need to ping_monitor here */
4329 sra->reshape_progress = reshape_completed;
4330
4331 /* Clear any backup region that is before 'here' */
4332 if (increasing) {
4333 if (__le64_to_cpu(bsb.length) > 0 &&
4334 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4335 __le64_to_cpu(bsb.length)))
4336 forget_backup(dests, destfd,
4337 destoffsets, 0);
4338 if (__le64_to_cpu(bsb.length2) > 0 &&
4339 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4340 __le64_to_cpu(bsb.length2)))
4341 forget_backup(dests, destfd,
4342 destoffsets, 1);
4343 } else {
4344 if (__le64_to_cpu(bsb.length) > 0 &&
4345 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4346 forget_backup(dests, destfd,
4347 destoffsets, 0);
4348 if (__le64_to_cpu(bsb.length2) > 0 &&
4349 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4350 forget_backup(dests, destfd,
4351 destoffsets, 1);
4352 }
4353 if (sigterm)
4354 rv = -2;
4355 if (rv < 0) {
4356 if (rv == -1)
4357 done = 1;
4358 break;
4359 }
4360 if (rv == 0 && increasing && !st->ss->external) {
4361 /* No longer need to monitor this reshape */
4362 sysfs_set_str(sra, NULL, "sync_max", "max");
4363 done = 1;
4364 break;
4365 }
4366
4367 while (rv) {
4368 unsigned long long offset;
4369 unsigned long actual_stripes;
4370 /* Need to backup some data.
4371 * If 'part' is not used and the desired
4372 * backup size is suspended, do a backup,
4373 * then consider the next part.
4374 */
4375 /* Check that 'part' is unused */
4376 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4377 break;
4378 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4379 break;
4380
4381 offset = backup_point / data;
4382 actual_stripes = stripes;
4383 if (increasing) {
4384 if (offset + actual_stripes * (chunk/512) >
4385 sra->component_size)
4386 actual_stripes = ((sra->component_size - offset)
4387 / (chunk/512));
4388 if (offset + actual_stripes * (chunk/512) >
4389 suspend_point/data)
4390 break;
4391 } else {
4392 if (offset < actual_stripes * (chunk/512))
4393 actual_stripes = offset / (chunk/512);
4394 offset -= actual_stripes * (chunk/512);
4395 if (offset < suspend_point/data)
4396 break;
4397 }
4398 if (actual_stripes == 0)
4399 break;
4400 grow_backup(sra, offset, actual_stripes,
4401 fds, offsets,
4402 disks, chunk, level, layout,
4403 dests, destfd, destoffsets,
4404 part, &degraded, buf);
4405 validate(afd, destfd[0], destoffsets[0]);
4406 /* record where 'part' is up to */
4407 part = !part;
4408 if (increasing)
4409 backup_point += actual_stripes * (chunk/512) * data;
4410 else
4411 backup_point -= actual_stripes * (chunk/512) * data;
4412 }
4413 }
4414
4415 /* FIXME maybe call progress_reshape one more time instead */
4416 /* remove any remaining suspension */
4417 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4418 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4419 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4420 sysfs_set_num(sra, NULL, "sync_min", 0);
4421
4422 if (reshape->before.data_disks == reshape->after.data_disks)
4423 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4424 free(buf);
4425 return done;
4426 }
4427
4428 /*
4429 * If any spare contains md_back_data-1 which is recent wrt mtime,
4430 * write that data into the array and update the super blocks with
4431 * the new reshape_progress
4432 */
4433 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4434 char *backup_file, int verbose)
4435 {
4436 int i, j;
4437 int old_disks;
4438 unsigned long long *offsets;
4439 unsigned long long nstripe, ostripe;
4440 int ndata, odata;
4441
4442 odata = info->array.raid_disks - info->delta_disks - 1;
4443 if (info->array.level == 6) odata--; /* number of data disks */
4444 ndata = info->array.raid_disks - 1;
4445 if (info->new_level == 6) ndata--;
4446
4447 old_disks = info->array.raid_disks - info->delta_disks;
4448
4449 if (info->delta_disks <= 0)
4450 /* Didn't grow, so the backup file must have
4451 * been used
4452 */
4453 old_disks = cnt;
4454 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4455 struct mdinfo dinfo;
4456 int fd;
4457 int bsbsize;
4458 char *devname, namebuf[20];
4459 unsigned long long lo, hi;
4460
4461 /* This was a spare and may have some saved data on it.
4462 * Load the superblock, find and load the
4463 * backup_super_block.
4464 * If either fail, go on to next device.
4465 * If the backup contains no new info, just return
4466 * else restore data and update all superblocks
4467 */
4468 if (i == old_disks-1) {
4469 fd = open(backup_file, O_RDONLY);
4470 if (fd<0) {
4471 pr_err("backup file %s inaccessible: %s\n",
4472 backup_file, strerror(errno));
4473 continue;
4474 }
4475 devname = backup_file;
4476 } else {
4477 fd = fdlist[i];
4478 if (fd < 0)
4479 continue;
4480 if (st->ss->load_super(st, fd, NULL))
4481 continue;
4482
4483 st->ss->getinfo_super(st, &dinfo, NULL);
4484 st->ss->free_super(st);
4485
4486 if (lseek64(fd,
4487 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4488 0) < 0) {
4489 pr_err("Cannot seek on device %d\n", i);
4490 continue; /* Cannot seek */
4491 }
4492 sprintf(namebuf, "device-%d", i);
4493 devname = namebuf;
4494 }
4495 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4496 if (verbose)
4497 pr_err("Cannot read from %s\n", devname);
4498 continue; /* Cannot read */
4499 }
4500 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4501 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4502 if (verbose)
4503 pr_err("No backup metadata on %s\n", devname);
4504 continue;
4505 }
4506 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4507 if (verbose)
4508 pr_err("Bad backup-metadata checksum on %s\n", devname);
4509 continue; /* bad checksum */
4510 }
4511 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4512 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4513 if (verbose)
4514 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4515 continue; /* Bad second checksum */
4516 }
4517 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4518 if (verbose)
4519 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4520 continue; /* Wrong uuid */
4521 }
4522
4523 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4524 * sometimes they aren't... So allow considerable flexability in matching, and allow
4525 * this test to be overridden by an environment variable.
4526 */
4527 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4528 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4529 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4530 pr_err("accepting backup with timestamp %lu "
4531 "for array with timestamp %lu\n",
4532 (unsigned long)__le64_to_cpu(bsb.mtime),
4533 (unsigned long)info->array.utime);
4534 } else {
4535 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4536 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4537 continue; /* time stamp is too bad */
4538 }
4539 }
4540
4541 if (bsb.magic[15] == '1') {
4542 if (bsb.length == 0)
4543 continue;
4544 if (info->delta_disks >= 0) {
4545 /* reshape_progress is increasing */
4546 if (__le64_to_cpu(bsb.arraystart)
4547 + __le64_to_cpu(bsb.length)
4548 < info->reshape_progress) {
4549 nonew:
4550 if (verbose)
4551 pr_err("backup-metadata found on %s but is not needed\n", devname);
4552 continue; /* No new data here */
4553 }
4554 } else {
4555 /* reshape_progress is decreasing */
4556 if (__le64_to_cpu(bsb.arraystart) >=
4557 info->reshape_progress)
4558 goto nonew; /* No new data here */
4559 }
4560 } else {
4561 if (bsb.length == 0 && bsb.length2 == 0)
4562 continue;
4563 if (info->delta_disks >= 0) {
4564 /* reshape_progress is increasing */
4565 if ((__le64_to_cpu(bsb.arraystart)
4566 + __le64_to_cpu(bsb.length)
4567 < info->reshape_progress)
4568 &&
4569 (__le64_to_cpu(bsb.arraystart2)
4570 + __le64_to_cpu(bsb.length2)
4571 < info->reshape_progress))
4572 goto nonew; /* No new data here */
4573 } else {
4574 /* reshape_progress is decreasing */
4575 if (__le64_to_cpu(bsb.arraystart) >=
4576 info->reshape_progress &&
4577 __le64_to_cpu(bsb.arraystart2) >=
4578 info->reshape_progress)
4579 goto nonew; /* No new data here */
4580 }
4581 }
4582 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4583 second_fail:
4584 if (verbose)
4585 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4586 devname);
4587 continue; /* Cannot seek */
4588 }
4589 /* There should be a duplicate backup superblock 4k before here */
4590 if (lseek64(fd, -4096, 1) < 0 ||
4591 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4592 goto second_fail; /* Cannot find leading superblock */
4593 if (bsb.magic[15] == '1')
4594 bsbsize = offsetof(struct mdp_backup_super, pad1);
4595 else
4596 bsbsize = offsetof(struct mdp_backup_super, pad);
4597 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4598 goto second_fail; /* Cannot find leading superblock */
4599
4600 /* Now need the data offsets for all devices. */
4601 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4602 for(j=0; j<info->array.raid_disks; j++) {
4603 if (fdlist[j] < 0)
4604 continue;
4605 if (st->ss->load_super(st, fdlist[j], NULL))
4606 /* FIXME should be this be an error */
4607 continue;
4608 st->ss->getinfo_super(st, &dinfo, NULL);
4609 st->ss->free_super(st);
4610 offsets[j] = dinfo.data_offset * 512;
4611 }
4612 printf(Name ": restoring critical section\n");
4613
4614 if (restore_stripes(fdlist, offsets,
4615 info->array.raid_disks,
4616 info->new_chunk,
4617 info->new_level,
4618 info->new_layout,
4619 fd, __le64_to_cpu(bsb.devstart)*512,
4620 __le64_to_cpu(bsb.arraystart)*512,
4621 __le64_to_cpu(bsb.length)*512, NULL)) {
4622 /* didn't succeed, so giveup */
4623 if (verbose)
4624 pr_err("Error restoring backup from %s\n",
4625 devname);
4626 free(offsets);
4627 return 1;
4628 }
4629
4630 if (bsb.magic[15] == '2' &&
4631 restore_stripes(fdlist, offsets,
4632 info->array.raid_disks,
4633 info->new_chunk,
4634 info->new_level,
4635 info->new_layout,
4636 fd, __le64_to_cpu(bsb.devstart)*512 +
4637 __le64_to_cpu(bsb.devstart2)*512,
4638 __le64_to_cpu(bsb.arraystart2)*512,
4639 __le64_to_cpu(bsb.length2)*512, NULL)) {
4640 /* didn't succeed, so giveup */
4641 if (verbose)
4642 pr_err("Error restoring second backup from %s\n",
4643 devname);
4644 free(offsets);
4645 return 1;
4646 }
4647
4648 free(offsets);
4649
4650 /* Ok, so the data is restored. Let's update those superblocks. */
4651
4652 lo = hi = 0;
4653 if (bsb.length) {
4654 lo = __le64_to_cpu(bsb.arraystart);
4655 hi = lo + __le64_to_cpu(bsb.length);
4656 }
4657 if (bsb.magic[15] == '2' && bsb.length2) {
4658 unsigned long long lo1, hi1;
4659 lo1 = __le64_to_cpu(bsb.arraystart2);
4660 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4661 if (lo == hi) {
4662 lo = lo1;
4663 hi = hi1;
4664 } else if (lo < lo1)
4665 hi = hi1;
4666 else
4667 lo = lo1;
4668 }
4669 if (lo < hi &&
4670 (info->reshape_progress < lo ||
4671 info->reshape_progress > hi))
4672 /* backup does not affect reshape_progress*/ ;
4673 else if (info->delta_disks >= 0) {
4674 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4675 __le64_to_cpu(bsb.length);
4676 if (bsb.magic[15] == '2') {
4677 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4678 __le64_to_cpu(bsb.length2);
4679 if (p2 > info->reshape_progress)
4680 info->reshape_progress = p2;
4681 }
4682 } else {
4683 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4684 if (bsb.magic[15] == '2') {
4685 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4686 if (p2 < info->reshape_progress)
4687 info->reshape_progress = p2;
4688 }
4689 }
4690 for (j=0; j<info->array.raid_disks; j++) {
4691 if (fdlist[j] < 0)
4692 continue;
4693 if (st->ss->load_super(st, fdlist[j], NULL))
4694 continue;
4695 st->ss->getinfo_super(st, &dinfo, NULL);
4696 dinfo.reshape_progress = info->reshape_progress;
4697 st->ss->update_super(st, &dinfo,
4698 "_reshape_progress",
4699 NULL,0, 0, NULL);
4700 st->ss->store_super(st, fdlist[j]);
4701 st->ss->free_super(st);
4702 }
4703 return 0;
4704 }
4705 /* Didn't find any backup data, try to see if any
4706 * was needed.
4707 */
4708 if (info->delta_disks < 0) {
4709 /* When shrinking, the critical section is at the end.
4710 * So see if we are before the critical section.
4711 */
4712 unsigned long long first_block;
4713 nstripe = ostripe = 0;
4714 first_block = 0;
4715 while (ostripe >= nstripe) {
4716 ostripe += info->array.chunk_size / 512;
4717 first_block = ostripe * odata;
4718 nstripe = first_block / ndata / (info->new_chunk/512) *
4719 (info->new_chunk/512);
4720 }
4721
4722 if (info->reshape_progress >= first_block)
4723 return 0;
4724 }
4725 if (info->delta_disks > 0) {
4726 /* See if we are beyond the critical section. */
4727 unsigned long long last_block;
4728 nstripe = ostripe = 0;
4729 last_block = 0;
4730 while (nstripe >= ostripe) {
4731 nstripe += info->new_chunk / 512;
4732 last_block = nstripe * ndata;
4733 ostripe = last_block / odata / (info->array.chunk_size/512) *
4734 (info->array.chunk_size/512);
4735 }
4736
4737 if (info->reshape_progress >= last_block)
4738 return 0;
4739 }
4740 /* needed to recover critical section! */
4741 if (verbose)
4742 pr_err("Failed to find backup of critical section\n");
4743 return 1;
4744 }
4745
4746 int Grow_continue_command(char *devname, int fd,
4747 char *backup_file, int verbose)
4748 {
4749 int ret_val = 0;
4750 struct supertype *st = NULL;
4751 struct mdinfo *content = NULL;
4752 struct mdinfo array;
4753 char *subarray = NULL;
4754 struct mdinfo *cc = NULL;
4755 struct mdstat_ent *mdstat = NULL;
4756 int cfd = -1;
4757 int fd2 = -1;
4758
4759 dprintf("Grow continue from command line called for %s\n",
4760 devname);
4761
4762 st = super_by_fd(fd, &subarray);
4763 if (!st || !st->ss) {
4764 pr_err("Unable to determine metadata format for %s\n",
4765 devname);
4766 return 1;
4767 }
4768 dprintf("Grow continue is run for ");
4769 if (st->ss->external == 0) {
4770 int d;
4771 dprintf("native array (%s)\n", devname);
4772 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4773 pr_err("%s is not an active md array -"
4774 " aborting\n", devname);
4775 ret_val = 1;
4776 goto Grow_continue_command_exit;
4777 }
4778 content = &array;
4779 /* Need to load a superblock.
4780 * FIXME we should really get what we need from
4781 * sysfs
4782 */
4783 for (d = 0; d < MAX_DISKS; d++) {
4784 mdu_disk_info_t disk;
4785 char *dv;
4786 int err;
4787 disk.number = d;
4788 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4789 continue;
4790 if (disk.major == 0 && disk.minor == 0)
4791 continue;
4792 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4793 continue;
4794 dv = map_dev(disk.major, disk.minor, 1);
4795 if (!dv)
4796 continue;
4797 fd2 = dev_open(dv, O_RDONLY);
4798 if (fd2 < 0)
4799 continue;
4800 err = st->ss->load_super(st, fd2, NULL);
4801 close(fd2);
4802 /* invalidate fd2 to avoid possible double close() */
4803 fd2 = -1;
4804 if (err)
4805 continue;
4806 break;
4807 }
4808 if (d == MAX_DISKS) {
4809 pr_err("Unable to load metadata for %s\n",
4810 devname);
4811 ret_val = 1;
4812 goto Grow_continue_command_exit;
4813 }
4814 st->ss->getinfo_super(st, content, NULL);
4815 } else {
4816 char *container;
4817
4818 if (subarray) {
4819 dprintf("subarray (%s)\n", subarray);
4820 container = st->container_devnm;
4821 cfd = open_dev_excl(st->container_devnm);
4822 } else {
4823 container = st->devnm;
4824 close(fd);
4825 cfd = open_dev_excl(st->devnm);
4826 dprintf("container (%s)\n", container);
4827 fd = cfd;
4828 }
4829 if (cfd < 0) {
4830 pr_err("Unable to open container "
4831 "for %s\n", devname);
4832 ret_val = 1;
4833 goto Grow_continue_command_exit;
4834 }
4835
4836 /* find in container array under reshape
4837 */
4838 ret_val = st->ss->load_container(st, cfd, NULL);
4839 if (ret_val) {
4840 pr_err("Cannot read superblock for %s\n",
4841 devname);
4842 ret_val = 1;
4843 goto Grow_continue_command_exit;
4844 }
4845
4846 cc = st->ss->container_content(st, subarray);
4847 for (content = cc; content ; content = content->next) {
4848 char *array;
4849 int allow_reshape = 1;
4850
4851 if (content->reshape_active == 0)
4852 continue;
4853 /* The decision about array or container wide
4854 * reshape is taken in Grow_continue based
4855 * content->reshape_active state, therefore we
4856 * need to check_reshape based on
4857 * reshape_active and subarray name
4858 */
4859 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4860 allow_reshape = 0;
4861 if (content->reshape_active == CONTAINER_RESHAPE &&
4862 (content->array.state
4863 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4864 allow_reshape = 0;
4865
4866 if (!allow_reshape) {
4867 pr_err("cannot continue reshape of an array"
4868 " in container with unsupported"
4869 " metadata: %s(%s)\n",
4870 devname, container);
4871 ret_val = 1;
4872 goto Grow_continue_command_exit;
4873 }
4874
4875 array = strchr(content->text_version+1, '/')+1;
4876 mdstat = mdstat_by_subdev(array, container);
4877 if (!mdstat)
4878 continue;
4879 if (mdstat->active == 0) {
4880 pr_err("Skipping inactive array %s.\n",
4881 mdstat->devnm);
4882 free_mdstat(mdstat);
4883 mdstat = NULL;
4884 continue;
4885 }
4886 break;
4887 }
4888 if (!content) {
4889 pr_err("Unable to determine reshaped "
4890 "array for %s\n", devname);
4891 ret_val = 1;
4892 goto Grow_continue_command_exit;
4893 }
4894 fd2 = open_dev(mdstat->devnm);
4895 if (fd2 < 0) {
4896 pr_err("cannot open (%s)\n", mdstat->devnm);
4897 ret_val = 1;
4898 goto Grow_continue_command_exit;
4899 }
4900
4901 sysfs_init(content, fd2, mdstat->devnm);
4902
4903 /* start mdmon in case it is not running
4904 */
4905 if (!mdmon_running(container))
4906 start_mdmon(container);
4907 ping_monitor(container);
4908
4909 if (mdmon_running(container))
4910 st->update_tail = &st->updates;
4911 else {
4912 pr_err("No mdmon found. "
4913 "Grow cannot continue.\n");
4914 ret_val = 1;
4915 goto Grow_continue_command_exit;
4916 }
4917 }
4918
4919 /* verify that array under reshape is started from
4920 * correct position
4921 */
4922 if (verify_reshape_position(content, content->array.level) < 0) {
4923 ret_val = 1;
4924 goto Grow_continue_command_exit;
4925 }
4926
4927 /* continue reshape
4928 */
4929 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4930
4931 Grow_continue_command_exit:
4932 if (fd2 > -1)
4933 close(fd2);
4934 if (cfd > -1)
4935 close(cfd);
4936 st->ss->free_super(st);
4937 free_mdstat(mdstat);
4938 sysfs_free(cc);
4939 free(subarray);
4940
4941 return ret_val;
4942 }
4943
4944 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4945 char *backup_file, int forked, int freeze_reshape)
4946 {
4947 int ret_val = 2;
4948
4949 if (!info->reshape_active)
4950 return ret_val;
4951
4952 if (st->ss->external) {
4953 int cfd = open_dev(st->container_devnm);
4954
4955 if (cfd < 0)
4956 return 1;
4957
4958 st->ss->load_container(st, cfd, st->container_devnm);
4959 close(cfd);
4960 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4961 st, info, 0, backup_file,
4962 0, forked,
4963 1 | info->reshape_active,
4964 freeze_reshape);
4965 } else
4966 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4967 NULL, INVALID_SECTORS,
4968 backup_file, 0, forked,
4969 1 | info->reshape_active,
4970 freeze_reshape);
4971
4972 return ret_val;
4973 }
4974
4975 char *make_backup(char *name)
4976 {
4977 char *base = "backup_file-";
4978 int len;
4979 char *fname;
4980
4981 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4982 fname = xmalloc(len);
4983 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4984 return fname;
4985 }
4986
4987 char *locate_backup(char *name)
4988 {
4989 char *fl = make_backup(name);
4990 struct stat stb;
4991
4992 if (stat(fl, &stb) == 0 &&
4993 S_ISREG(stb.st_mode))
4994 return fl;
4995
4996 free(fl);
4997 return NULL;
4998 }
Unnamed repository; edit this file 'description' to name the repository.