]> 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
Grow: retry when writing 'reshape' to 'sync_action' is EBUSY.
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27 #include <stddef.h>
28 #include <stdint.h>
29 #include <signal.h>
30 #include <sys/wait.h>
31
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
34 #endif
35 #include "md_u.h"
36 #include "md_p.h"
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char **backup_filep,
43 int verbose)
44 {
45 int i;
46 int *fdlist;
47 struct mdinfo *dev;
48 int err;
49 int disk_count = next_spare + working_disks;
50 char *backup_file = *backup_filep;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61 sprintf(buf, "%d:%d",
62 dev->disk.major,
63 dev->disk.minor);
64 fd = dev_open(buf, O_RDWR);
65
66 if (dev->disk.raid_disk >= 0)
67 fdlist[dev->disk.raid_disk] = fd;
68 else
69 fdlist[next_spare++] = fd;
70 }
71
72 if (!backup_file) {
73 backup_file = locate_backup(content->sys_name);
74 *backup_filep = backup_file;
75 }
76
77 if (st->ss->external && st->ss->recover_backup)
78 err = st->ss->recover_backup(st, content);
79 else
80 err = Grow_restart(st, content, fdlist, next_spare,
81 backup_file, verbose > 0);
82
83 while (next_spare > 0) {
84 next_spare--;
85 if (fdlist[next_spare] >= 0)
86 close(fdlist[next_spare]);
87 }
88 free(fdlist);
89 if (err) {
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
91 if (!backup_file)
92 pr_err("Possibly you need to specify a --backup-file\n");
93 return 1;
94 }
95
96 dprintf("restore_backup() returns status OK.\n");
97 return 0;
98 }
99
100 int Grow_Add_device(char *devname, int fd, char *newdev)
101 {
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
107 * all other devices.
108 * This means that we need to *find* all other devices.
109 */
110 struct mdinfo info;
111
112 struct stat stb;
113 int nfd, fd2;
114 int d, nd;
115 struct supertype *st = NULL;
116 char *subarray = NULL;
117
118 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
119 pr_err("cannot get array info for %s\n", devname);
120 return 1;
121 }
122
123 if (info.array.level != -1) {
124 pr_err("can only add devices to linear arrays\n");
125 return 1;
126 }
127
128 st = super_by_fd(fd, &subarray);
129 if (!st) {
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info.array.major_version);
132 return 1;
133 }
134
135 if (subarray) {
136 pr_err("Cannot grow linear sub-arrays yet\n");
137 free(subarray);
138 free(st);
139 return 1;
140 }
141
142 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
143 if (nfd < 0) {
144 pr_err("cannot open %s\n", newdev);
145 free(st);
146 return 1;
147 }
148 fstat(nfd, &stb);
149 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
150 pr_err("%s is not a block device!\n", newdev);
151 close(nfd);
152 free(st);
153 return 1;
154 }
155 /* now check out all the devices and make sure we can read the
156 * superblock */
157 for (d=0 ; d < info.array.raid_disks ; d++) {
158 mdu_disk_info_t disk;
159 char *dv;
160
161 st->ss->free_super(st);
162
163 disk.number = d;
164 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
165 pr_err("cannot get device detail for device %d\n",
166 d);
167 close(nfd);
168 free(st);
169 return 1;
170 }
171 dv = map_dev(disk.major, disk.minor, 1);
172 if (!dv) {
173 pr_err("cannot find device file for device %d\n",
174 d);
175 close(nfd);
176 free(st);
177 return 1;
178 }
179 fd2 = dev_open(dv, O_RDWR);
180 if (fd2 < 0) {
181 pr_err("cannot open device file %s\n", dv);
182 close(nfd);
183 free(st);
184 return 1;
185 }
186
187 if (st->ss->load_super(st, fd2, NULL)) {
188 pr_err("cannot find super block on %s\n", dv);
189 close(nfd);
190 close(fd2);
191 free(st);
192 return 1;
193 }
194 close(fd2);
195 }
196 /* Ok, looks good. Lets update the superblock and write it out to
197 * newdev.
198 */
199
200 info.disk.number = d;
201 info.disk.major = major(stb.st_rdev);
202 info.disk.minor = minor(stb.st_rdev);
203 info.disk.raid_disk = d;
204 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
205 st->ss->update_super(st, &info, "linear-grow-new", newdev,
206 0, 0, NULL);
207
208 if (st->ss->store_super(st, nfd)) {
209 pr_err("Cannot store new superblock on %s\n",
210 newdev);
211 close(nfd);
212 return 1;
213 }
214 close(nfd);
215
216 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
217 pr_err("Cannot add new disk to this array\n");
218 return 1;
219 }
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
222 */
223
224 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
225 pr_err("cannot get array info for %s\n", devname);
226 return 1;
227 }
228
229 nd = d;
230 for (d=0 ; d < info.array.raid_disks ; d++) {
231 mdu_disk_info_t disk;
232 char *dv;
233
234 disk.number = d;
235 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
236 pr_err("cannot get device detail for device %d\n",
237 d);
238 return 1;
239 }
240 dv = map_dev(disk.major, disk.minor, 1);
241 if (!dv) {
242 pr_err("cannot find device file for device %d\n",
243 d);
244 return 1;
245 }
246 fd2 = dev_open(dv, O_RDWR);
247 if (fd2 < 0) {
248 pr_err("cannot open device file %s\n", dv);
249 return 1;
250 }
251 if (st->ss->load_super(st, fd2, NULL)) {
252 pr_err("cannot find super block on %s\n", dv);
253 close(fd);
254 return 1;
255 }
256 info.array.raid_disks = nd+1;
257 info.array.nr_disks = nd+1;
258 info.array.active_disks = nd+1;
259 info.array.working_disks = nd+1;
260
261 st->ss->update_super(st, &info, "linear-grow-update", dv,
262 0, 0, NULL);
263
264 if (st->ss->store_super(st, fd2)) {
265 pr_err("Cannot store new superblock on %s\n", dv);
266 close(fd2);
267 return 1;
268 }
269 close(fd2);
270 }
271
272 return 0;
273 }
274
275 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
276 {
277 /*
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
280 * Then add it
281 *
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
284 * to all devices
285 */
286 mdu_bitmap_file_t bmf;
287 mdu_array_info_t array;
288 struct supertype *st;
289 char *subarray = NULL;
290 int major = BITMAP_MAJOR_HI;
291 int vers = md_get_version(fd);
292 unsigned long long bitmapsize, array_size;
293
294 if (vers < 9003) {
295 major = BITMAP_MAJOR_HOSTENDIAN;
296 pr_err("Warning - bitmaps created on this kernel are not portable\n"
297 " between different architectures. Consider upgrading the Linux kernel.\n");
298 }
299
300 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
301 if (errno == ENOMEM)
302 pr_err("Memory allocation failure.\n");
303 else
304 pr_err("bitmaps not supported by this kernel.\n");
305 return 1;
306 }
307 if (bmf.pathname[0]) {
308 if (strcmp(s->bitmap_file,"none")==0) {
309 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
310 pr_err("failed to remove bitmap %s\n",
311 bmf.pathname);
312 return 1;
313 }
314 return 0;
315 }
316 pr_err("%s already has a bitmap (%s)\n",
317 devname, bmf.pathname);
318 return 1;
319 }
320 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
321 pr_err("cannot get array status for %s\n", devname);
322 return 1;
323 }
324 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
325 if (strcmp(s->bitmap_file, "none")==0) {
326 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
327 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
328 pr_err("failed to remove internal bitmap.\n");
329 return 1;
330 }
331 return 0;
332 }
333 pr_err("Internal bitmap already present on %s\n",
334 devname);
335 return 1;
336 }
337
338 if (strcmp(s->bitmap_file, "none") == 0) {
339 pr_err("no bitmap found on %s\n", devname);
340 return 1;
341 }
342 if (array.level <= 0) {
343 pr_err("Bitmaps not meaningful with level %s\n",
344 map_num(pers, array.level)?:"of this array");
345 return 1;
346 }
347 bitmapsize = array.size;
348 bitmapsize <<= 1;
349 if (get_dev_size(fd, NULL, &array_size) &&
350 array_size > (0x7fffffffULL<<9)) {
351 /* Array is big enough that we cannot trust array.size
352 * try other approaches
353 */
354 bitmapsize = get_component_size(fd);
355 }
356 if (bitmapsize == 0) {
357 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
358 return 1;
359 }
360
361 if (array.level == 10) {
362 int ncopies = (array.layout&255)*((array.layout>>8)&255);
363 bitmapsize = bitmapsize * array.raid_disks / ncopies;
364 }
365
366 st = super_by_fd(fd, &subarray);
367 if (!st) {
368 pr_err("Cannot understand version %d.%d\n",
369 array.major_version, array.minor_version);
370 return 1;
371 }
372 if (subarray) {
373 pr_err("Cannot add bitmaps to sub-arrays yet\n");
374 free(subarray);
375 free(st);
376 return 1;
377 }
378 if (strcmp(s->bitmap_file, "internal") == 0) {
379 int rv;
380 int d;
381 int offset_setable = 0;
382 struct mdinfo *mdi;
383 if (st->ss->add_internal_bitmap == NULL) {
384 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
385 return 1;
386 }
387 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
388 if (mdi)
389 offset_setable = 1;
390 for (d=0; d< st->max_devs; d++) {
391 mdu_disk_info_t disk;
392 char *dv;
393 disk.number = d;
394 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
395 continue;
396 if (disk.major == 0 &&
397 disk.minor == 0)
398 continue;
399 if ((disk.state & (1<<MD_DISK_SYNC))==0)
400 continue;
401 dv = map_dev(disk.major, disk.minor, 1);
402 if (dv) {
403 int fd2 = dev_open(dv, O_RDWR);
404 if (fd2 < 0)
405 continue;
406 if (st->ss->load_super(st, fd2, NULL)==0) {
407 if (st->ss->add_internal_bitmap(
408 st,
409 &s->bitmap_chunk, c->delay, s->write_behind,
410 bitmapsize, offset_setable,
411 major)
412 )
413 st->ss->write_bitmap(st, fd2);
414 else {
415 pr_err("failed to create internal bitmap - chunksize problem.\n");
416 close(fd2);
417 return 1;
418 }
419 }
420 close(fd2);
421 }
422 }
423 if (offset_setable) {
424 st->ss->getinfo_super(st, mdi, NULL);
425 sysfs_init(mdi, fd, NULL);
426 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
427 mdi->bitmap_offset);
428 } else {
429 array.state |= (1<<MD_SB_BITMAP_PRESENT);
430 rv = ioctl(fd, SET_ARRAY_INFO, &array);
431 }
432 if (rv < 0) {
433 if (errno == EBUSY)
434 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
435 pr_err("failed to set internal bitmap.\n");
436 return 1;
437 }
438 } else {
439 int uuid[4];
440 int bitmap_fd;
441 int d;
442 int max_devs = st->max_devs;
443
444 /* try to load a superblock */
445 for (d = 0; d < max_devs; d++) {
446 mdu_disk_info_t disk;
447 char *dv;
448 int fd2;
449 disk.number = d;
450 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
451 continue;
452 if ((disk.major==0 && disk.minor==0) ||
453 (disk.state & (1<<MD_DISK_REMOVED)))
454 continue;
455 dv = map_dev(disk.major, disk.minor, 1);
456 if (!dv)
457 continue;
458 fd2 = dev_open(dv, O_RDONLY);
459 if (fd2 >= 0) {
460 if (st->ss->load_super(st, fd2, NULL) == 0) {
461 close(fd2);
462 st->ss->uuid_from_super(st, uuid);
463 break;
464 }
465 close(fd2);
466 }
467 }
468 if (d == max_devs) {
469 pr_err("cannot find UUID for array!\n");
470 return 1;
471 }
472 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
473 c->delay, s->write_behind, bitmapsize, major)) {
474 return 1;
475 }
476 bitmap_fd = open(s->bitmap_file, O_RDWR);
477 if (bitmap_fd < 0) {
478 pr_err("weird: %s cannot be opened\n",
479 s->bitmap_file);
480 return 1;
481 }
482 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
483 int err = errno;
484 if (errno == EBUSY)
485 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
486 pr_err("Cannot set bitmap file for %s: %s\n",
487 devname, strerror(err));
488 return 1;
489 }
490 }
491
492 return 0;
493 }
494
495 /*
496 * When reshaping an array we might need to backup some data.
497 * This is written to all spares with a 'super_block' describing it.
498 * The superblock goes 4K from the end of the used space on the
499 * device.
500 * It if written after the backup is complete.
501 * It has the following structure.
502 */
503
504 static struct mdp_backup_super {
505 char magic[16]; /* md_backup_data-1 or -2 */
506 __u8 set_uuid[16];
507 __u64 mtime;
508 /* start/sizes in 512byte sectors */
509 __u64 devstart; /* address on backup device/file of data */
510 __u64 arraystart;
511 __u64 length;
512 __u32 sb_csum; /* csum of preceeding bytes. */
513 __u32 pad1;
514 __u64 devstart2; /* offset in to data of second section */
515 __u64 arraystart2;
516 __u64 length2;
517 __u32 sb_csum2; /* csum of preceeding bytes. */
518 __u8 pad[512-68-32];
519 } __attribute__((aligned(512))) bsb, bsb2;
520
521 static __u32 bsb_csum(char *buf, int len)
522 {
523 int i;
524 int csum = 0;
525 for (i = 0; i < len; i++)
526 csum = (csum<<3) + buf[0];
527 return __cpu_to_le32(csum);
528 }
529
530 static int check_idle(struct supertype *st)
531 {
532 /* Check that all member arrays for this container, or the
533 * container of this array, are idle
534 */
535 char *container = (st->container_devnm[0]
536 ? st->container_devnm : st->devnm);
537 struct mdstat_ent *ent, *e;
538 int is_idle = 1;
539
540 ent = mdstat_read(0, 0);
541 for (e = ent ; e; e = e->next) {
542 if (!is_container_member(e, container))
543 continue;
544 if (e->percent >= 0) {
545 is_idle = 0;
546 break;
547 }
548 }
549 free_mdstat(ent);
550 return is_idle;
551 }
552
553 static int freeze_container(struct supertype *st)
554 {
555 char *container = (st->container_devnm[0]
556 ? st->container_devnm : st->devnm);
557
558 if (!check_idle(st))
559 return -1;
560
561 if (block_monitor(container, 1)) {
562 pr_err("failed to freeze container\n");
563 return -2;
564 }
565
566 return 1;
567 }
568
569 static void unfreeze_container(struct supertype *st)
570 {
571 char *container = (st->container_devnm[0]
572 ? st->container_devnm : st->devnm);
573
574 unblock_monitor(container, 1);
575 }
576
577 static int freeze(struct supertype *st)
578 {
579 /* Try to freeze resync/rebuild on this array/container.
580 * Return -1 if the array is busy,
581 * return -2 container cannot be frozen,
582 * return 0 if this kernel doesn't support 'frozen'
583 * return 1 if it worked.
584 */
585 if (st->ss->external)
586 return freeze_container(st);
587 else {
588 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
589 int err;
590 char buf[20];
591
592 if (!sra)
593 return -1;
594 /* Need to clear any 'read-auto' status */
595 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
596 strncmp(buf, "read-auto", 9) == 0)
597 sysfs_set_str(sra, NULL, "array_state", "clean");
598
599 err = sysfs_freeze_array(sra);
600 sysfs_free(sra);
601 return err;
602 }
603 }
604
605 static void unfreeze(struct supertype *st)
606 {
607 if (st->ss->external)
608 return unfreeze_container(st);
609 else {
610 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
611 char buf[20];
612
613 if (sra &&
614 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
615 && strcmp(buf, "frozen\n") == 0) {
616 printf("unfreeze\n");
617 sysfs_set_str(sra, NULL, "sync_action", "idle");
618 }
619 sysfs_free(sra);
620 }
621 }
622
623 static void wait_reshape(struct mdinfo *sra)
624 {
625 int fd = sysfs_get_fd(sra, NULL, "sync_action");
626 char action[20];
627
628 if (fd < 0)
629 return;
630
631 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
632 strncmp(action, "reshape", 7) == 0)
633 sysfs_wait(fd, NULL);
634 close(fd);
635 }
636
637 static int reshape_super(struct supertype *st, unsigned long long size,
638 int level, int layout, int chunksize, int raid_disks,
639 int delta_disks, char *backup_file, char *dev,
640 int direction, int verbose)
641 {
642 /* nothing extra to check in the native case */
643 if (!st->ss->external)
644 return 0;
645 if (!st->ss->reshape_super ||
646 !st->ss->manage_reshape) {
647 pr_err("%s metadata does not support reshape\n",
648 st->ss->name);
649 return 1;
650 }
651
652 return st->ss->reshape_super(st, size, level, layout, chunksize,
653 raid_disks, delta_disks, backup_file, dev,
654 direction, verbose);
655 }
656
657 static void sync_metadata(struct supertype *st)
658 {
659 if (st->ss->external) {
660 if (st->update_tail) {
661 flush_metadata_updates(st);
662 st->update_tail = &st->updates;
663 } else
664 st->ss->sync_metadata(st);
665 }
666 }
667
668 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
669 {
670 /* when dealing with external metadata subarrays we need to be
671 * prepared to handle EAGAIN. The kernel may need to wait for
672 * mdmon to mark the array active so the kernel can handle
673 * allocations/writeback when preparing the reshape action
674 * (md_allow_write()). We temporarily disable safe_mode_delay
675 * to close a race with the array_state going clean before the
676 * next write to raid_disks / stripe_cache_size
677 */
678 char safe[50];
679 int rc;
680
681 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
682 if (!container ||
683 (strcmp(name, "raid_disks") != 0 &&
684 strcmp(name, "stripe_cache_size") != 0))
685 return sysfs_set_num(sra, NULL, name, n);
686
687 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
688 if (rc <= 0)
689 return -1;
690 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
691 rc = sysfs_set_num(sra, NULL, name, n);
692 if (rc < 0 && errno == EAGAIN) {
693 ping_monitor(container);
694 /* if we get EAGAIN here then the monitor is not active
695 * so stop trying
696 */
697 rc = sysfs_set_num(sra, NULL, name, n);
698 }
699 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
700 return rc;
701 }
702
703 int start_reshape(struct mdinfo *sra, int already_running,
704 int before_data_disks, int data_disks)
705 {
706 int err;
707 unsigned long long sync_max_to_set;
708
709 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
710 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
711 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
712 sra->reshape_progress);
713 if (before_data_disks <= data_disks)
714 sync_max_to_set = sra->reshape_progress / data_disks;
715 else
716 sync_max_to_set = (sra->component_size * data_disks
717 - sra->reshape_progress) / data_disks;
718 if (!already_running)
719 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
720 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
721 if (!already_running && err == 0) {
722 int cnt = 5;
723 do {
724 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
725 if (err)
726 sleep(1);
727 } while (err && errno == EBUSY && cnt-- > 0);
728 }
729 return err;
730 }
731
732 void abort_reshape(struct mdinfo *sra)
733 {
734 sysfs_set_str(sra, NULL, "sync_action", "idle");
735 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
736 sysfs_set_num(sra, NULL, "suspend_hi", 0);
737 sysfs_set_num(sra, NULL, "suspend_lo", 0);
738 sysfs_set_num(sra, NULL, "sync_min", 0);
739 // It isn't safe to reset sync_max as we aren't monitoring.
740 // Array really should be stopped at this point.
741 }
742
743 int remove_disks_for_takeover(struct supertype *st,
744 struct mdinfo *sra,
745 int layout)
746 {
747 int nr_of_copies;
748 struct mdinfo *remaining;
749 int slot;
750
751 if (sra->array.level == 10)
752 nr_of_copies = layout & 0xff;
753 else if (sra->array.level == 1)
754 nr_of_copies = sra->array.raid_disks;
755 else
756 return 1;
757
758 remaining = sra->devs;
759 sra->devs = NULL;
760 /* for each 'copy', select one device and remove from the list. */
761 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
762 struct mdinfo **diskp;
763 int found = 0;
764
765 /* Find a working device to keep */
766 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
767 struct mdinfo *disk = *diskp;
768
769 if (disk->disk.raid_disk < slot)
770 continue;
771 if (disk->disk.raid_disk >= slot + nr_of_copies)
772 continue;
773 if (disk->disk.state & (1<<MD_DISK_REMOVED))
774 continue;
775 if (disk->disk.state & (1<<MD_DISK_FAULTY))
776 continue;
777 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
778 continue;
779
780 /* We have found a good disk to use! */
781 *diskp = disk->next;
782 disk->next = sra->devs;
783 sra->devs = disk;
784 found = 1;
785 break;
786 }
787 if (!found)
788 break;
789 }
790
791 if (slot < sra->array.raid_disks) {
792 /* didn't find all slots */
793 struct mdinfo **e;
794 e = &remaining;
795 while (*e)
796 e = &(*e)->next;
797 *e = sra->devs;
798 sra->devs = remaining;
799 return 1;
800 }
801
802 /* Remove all 'remaining' devices from the array */
803 while (remaining) {
804 struct mdinfo *sd = remaining;
805 remaining = sd->next;
806
807 sysfs_set_str(sra, sd, "state", "faulty");
808 sysfs_set_str(sra, sd, "slot", "none");
809 /* for external metadata disks should be removed in mdmon */
810 if (!st->ss->external)
811 sysfs_set_str(sra, sd, "state", "remove");
812 sd->disk.state |= (1<<MD_DISK_REMOVED);
813 sd->disk.state &= ~(1<<MD_DISK_SYNC);
814 sd->next = sra->devs;
815 sra->devs = sd;
816 }
817 return 0;
818 }
819
820 void reshape_free_fdlist(int *fdlist,
821 unsigned long long *offsets,
822 int size)
823 {
824 int i;
825
826 for (i = 0; i < size; i++)
827 if (fdlist[i] >= 0)
828 close(fdlist[i]);
829
830 free(fdlist);
831 free(offsets);
832 }
833
834 int reshape_prepare_fdlist(char *devname,
835 struct mdinfo *sra,
836 int raid_disks,
837 int nrdisks,
838 unsigned long blocks,
839 char *backup_file,
840 int *fdlist,
841 unsigned long long *offsets)
842 {
843 int d = 0;
844 struct mdinfo *sd;
845
846 enable_fds(nrdisks);
847 for (d = 0; d <= nrdisks; d++)
848 fdlist[d] = -1;
849 d = raid_disks;
850 for (sd = sra->devs; sd; sd = sd->next) {
851 if (sd->disk.state & (1<<MD_DISK_FAULTY))
852 continue;
853 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
854 char *dn = map_dev(sd->disk.major,
855 sd->disk.minor, 1);
856 fdlist[sd->disk.raid_disk]
857 = dev_open(dn, O_RDONLY);
858 offsets[sd->disk.raid_disk] = sd->data_offset*512;
859 if (fdlist[sd->disk.raid_disk] < 0) {
860 pr_err("%s: cannot open component %s\n",
861 devname, dn ? dn : "-unknown-");
862 d = -1;
863 goto release;
864 }
865 } else if (backup_file == NULL) {
866 /* spare */
867 char *dn = map_dev(sd->disk.major,
868 sd->disk.minor, 1);
869 fdlist[d] = dev_open(dn, O_RDWR);
870 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
871 if (fdlist[d] < 0) {
872 pr_err("%s: cannot open component %s\n",
873 devname, dn ? dn : "-unknown-");
874 d = -1;
875 goto release;
876 }
877 d++;
878 }
879 }
880 release:
881 return d;
882 }
883
884 int reshape_open_backup_file(char *backup_file,
885 int fd,
886 char *devname,
887 long blocks,
888 int *fdlist,
889 unsigned long long *offsets,
890 char *sys_name,
891 int restart)
892 {
893 /* Return 1 on success, 0 on any form of failure */
894 /* need to check backup file is large enough */
895 char buf[512];
896 struct stat stb;
897 unsigned int dev;
898 int i;
899
900 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
901 S_IRUSR | S_IWUSR);
902 *offsets = 8 * 512;
903 if (*fdlist < 0) {
904 pr_err("%s: cannot create backup file %s: %s\n",
905 devname, backup_file, strerror(errno));
906 return 0;
907 }
908 /* Guard against backup file being on array device.
909 * If array is partitioned or if LVM etc is in the
910 * way this will not notice, but it is better than
911 * nothing.
912 */
913 fstat(*fdlist, &stb);
914 dev = stb.st_dev;
915 fstat(fd, &stb);
916 if (stb.st_rdev == dev) {
917 pr_err("backup file must NOT be on the array being reshaped.\n");
918 close(*fdlist);
919 return 0;
920 }
921
922 memset(buf, 0, 512);
923 for (i=0; i < blocks + 8 ; i++) {
924 if (write(*fdlist, buf, 512) != 512) {
925 pr_err("%s: cannot create backup file %s: %s\n",
926 devname, backup_file, strerror(errno));
927 return 0;
928 }
929 }
930 if (fsync(*fdlist) != 0) {
931 pr_err("%s: cannot create backup file %s: %s\n",
932 devname, backup_file, strerror(errno));
933 return 0;
934 }
935
936 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
937 char *bu = make_backup(sys_name);
938 if (symlink(backup_file, bu))
939 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
940 strerror(errno));
941 free(bu);
942 }
943
944 return 1;
945 }
946
947 unsigned long compute_backup_blocks(int nchunk, int ochunk,
948 unsigned int ndata, unsigned int odata)
949 {
950 unsigned long a, b, blocks;
951 /* So how much do we need to backup.
952 * We need an amount of data which is both a whole number of
953 * old stripes and a whole number of new stripes.
954 * So LCM for (chunksize*datadisks).
955 */
956 a = (ochunk/512) * odata;
957 b = (nchunk/512) * ndata;
958 /* Find GCD */
959 a = GCD(a, b);
960 /* LCM == product / GCD */
961 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
962
963 return blocks;
964 }
965
966 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
967 {
968 /* Based on the current array state in info->array and
969 * the changes in info->new_* etc, determine:
970 * - whether the change is possible
971 * - Intermediate level/raid_disks/layout
972 * - whether a restriping reshape is needed
973 * - number of sectors in minimum change unit. This
974 * will cover a whole number of stripes in 'before' and
975 * 'after'.
976 *
977 * Return message if the change should be rejected
978 * NULL if the change can be achieved
979 *
980 * This can be called as part of starting a reshape, or
981 * when assembling an array that is undergoing reshape.
982 */
983 int near, far, offset, copies;
984 int new_disks;
985 int old_chunk, new_chunk;
986 /* delta_parity records change in number of devices
987 * caused by level change
988 */
989 int delta_parity = 0;
990
991 memset(re, 0, sizeof(*re));
992
993 /* If a new level not explicitly given, we assume no-change */
994 if (info->new_level == UnSet)
995 info->new_level = info->array.level;
996
997 if (info->new_chunk)
998 switch (info->new_level) {
999 case 0:
1000 case 4:
1001 case 5:
1002 case 6:
1003 case 10:
1004 /* chunk size is meaningful, must divide component_size
1005 * evenly
1006 */
1007 if (info->component_size % (info->new_chunk/512)) {
1008 unsigned long long shrink = info->component_size;
1009 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1010 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1011 info->new_chunk/1024, info->component_size/2);
1012 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1013 devname, shrink/2);
1014 pr_err("will shrink the array so the given chunk size would work.\n");
1015 return "";
1016 }
1017 break;
1018 default:
1019 return "chunk size not meaningful for this level";
1020 }
1021 else
1022 info->new_chunk = info->array.chunk_size;
1023
1024 switch (info->array.level) {
1025 default:
1026 return "No reshape is possibly for this RAID level";
1027 case LEVEL_LINEAR:
1028 if (info->delta_disks != UnSet)
1029 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1030 else
1031 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1032 case 1:
1033 /* RAID1 can convert to RAID1 with different disks, or
1034 * raid5 with 2 disks, or
1035 * raid0 with 1 disk
1036 */
1037 if (info->new_level > 1 &&
1038 (info->component_size & 7))
1039 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1040 if (info->new_level == 0) {
1041 if (info->delta_disks != UnSet &&
1042 info->delta_disks != 0)
1043 return "Cannot change number of disks with RAID1->RAID0 conversion";
1044 re->level = 0;
1045 re->before.data_disks = 1;
1046 re->after.data_disks = 1;
1047 return NULL;
1048 }
1049 if (info->new_level == 1) {
1050 if (info->delta_disks == UnSet)
1051 /* Don't know what to do */
1052 return "no change requested for Growing RAID1";
1053 re->level = 1;
1054 return NULL;
1055 }
1056 if (info->array.raid_disks == 2 &&
1057 info->new_level == 5) {
1058
1059 re->level = 5;
1060 re->before.data_disks = 1;
1061 if (info->delta_disks != UnSet &&
1062 info->delta_disks != 0)
1063 re->after.data_disks = 1 + info->delta_disks;
1064 else
1065 re->after.data_disks = 1;
1066 if (re->after.data_disks < 1)
1067 return "Number of disks too small for RAID5";
1068
1069 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1070 info->array.chunk_size = 65536;
1071 break;
1072 }
1073 /* Could do some multi-stage conversions, but leave that to
1074 * later.
1075 */
1076 return "Impossibly level change request for RAID1";
1077
1078 case 10:
1079 /* RAID10 can be converted from near mode to
1080 * RAID0 by removing some devices.
1081 * It can also be reshaped if the kernel supports
1082 * new_data_offset.
1083 */
1084 switch (info->new_level) {
1085 case 0:
1086 if ((info->array.layout & ~0xff) != 0x100)
1087 return "Cannot Grow RAID10 with far/offset layout";
1088 /* number of devices must be multiple of number of copies */
1089 if (info->array.raid_disks % (info->array.layout & 0xff))
1090 return "RAID10 layout too complex for Grow operation";
1091
1092 new_disks = (info->array.raid_disks
1093 / (info->array.layout & 0xff));
1094 if (info->delta_disks == UnSet)
1095 info->delta_disks = (new_disks
1096 - info->array.raid_disks);
1097
1098 if (info->delta_disks != new_disks - info->array.raid_disks)
1099 return "New number of raid-devices impossible for RAID10";
1100 if (info->new_chunk &&
1101 info->new_chunk != info->array.chunk_size)
1102 return "Cannot change chunk-size with RAID10 Grow";
1103
1104 /* looks good */
1105 re->level = 0;
1106 re->before.data_disks = new_disks;
1107 re->after.data_disks = re->before.data_disks;
1108 return NULL;
1109
1110 case 10:
1111 near = info->array.layout & 0xff;
1112 far = (info->array.layout >> 8) & 0xff;
1113 offset = info->array.layout & 0x10000;
1114 if (far > 1 && !offset)
1115 return "Cannot reshape RAID10 in far-mode";
1116 copies = near * far;
1117
1118 old_chunk = info->array.chunk_size * far;
1119
1120 if (info->new_layout == UnSet)
1121 info->new_layout = info->array.layout;
1122 else {
1123 near = info->new_layout & 0xff;
1124 far = (info->new_layout >> 8) & 0xff;
1125 offset = info->new_layout & 0x10000;
1126 if (far > 1 && !offset)
1127 return "Cannot reshape RAID10 to far-mode";
1128 if (near * far != copies)
1129 return "Cannot change number of copies when reshaping RAID10";
1130 }
1131 if (info->delta_disks == UnSet)
1132 info->delta_disks = 0;
1133 new_disks = (info->array.raid_disks +
1134 info->delta_disks);
1135
1136 new_chunk = info->new_chunk * far;
1137
1138 re->level = 10;
1139 re->before.layout = info->array.layout;
1140 re->before.data_disks = info->array.raid_disks;
1141 re->after.layout = info->new_layout;
1142 re->after.data_disks = new_disks;
1143 /* For RAID10 we don't do backup but do allow reshape,
1144 * so set backup_blocks to INVALID_SECTORS rather than
1145 * zero.
1146 * And there is no need to synchronise stripes on both
1147 * 'old' and 'new'. So the important
1148 * number is the minimum data_offset difference
1149 * which is the larger of (offset copies * chunk).
1150 */
1151 re->backup_blocks = INVALID_SECTORS;
1152 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1153 if (new_disks < re->before.data_disks &&
1154 info->space_after < re->min_offset_change)
1155 /* Reduce component size by one chunk */
1156 re->new_size = (info->component_size -
1157 re->min_offset_change);
1158 else
1159 re->new_size = info->component_size;
1160 re->new_size = re->new_size * new_disks / copies;
1161 return NULL;
1162
1163 default:
1164 return "RAID10 can only be changed to RAID0";
1165 }
1166 case 0:
1167 /* RAID0 can be converted to RAID10, or to RAID456 */
1168 if (info->new_level == 10) {
1169 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1170 /* Assume near=2 layout */
1171 info->new_layout = 0x102;
1172 info->delta_disks = info->array.raid_disks;
1173 }
1174 if (info->new_layout == UnSet) {
1175 int copies = 1 + (info->delta_disks
1176 / info->array.raid_disks);
1177 if (info->array.raid_disks * (copies-1)
1178 != info->delta_disks)
1179 return "Impossible number of devices for RAID0->RAID10";
1180 info->new_layout = 0x100 + copies;
1181 }
1182 if (info->delta_disks == UnSet) {
1183 int copies = info->new_layout & 0xff;
1184 if (info->new_layout != 0x100 + copies)
1185 return "New layout impossible for RAID0->RAID10";;
1186 info->delta_disks = (copies - 1) *
1187 info->array.raid_disks;
1188 }
1189 if (info->new_chunk &&
1190 info->new_chunk != info->array.chunk_size)
1191 return "Cannot change chunk-size with RAID0->RAID10";
1192 /* looks good */
1193 re->level = 10;
1194 re->before.data_disks = (info->array.raid_disks +
1195 info->delta_disks);
1196 re->after.data_disks = re->before.data_disks;
1197 re->before.layout = info->new_layout;
1198 return NULL;
1199 }
1200
1201 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1202 * a raid4 style layout of the final level.
1203 */
1204 switch (info->new_level) {
1205 case 4:
1206 delta_parity = 1;
1207 case 0:
1208 re->level = 4;
1209 re->before.layout = 0;
1210 break;
1211 case 5:
1212 delta_parity = 1;
1213 re->level = 5;
1214 re->before.layout = ALGORITHM_PARITY_N;
1215 if (info->new_layout == UnSet)
1216 info->new_layout = map_name(r5layout, "default");
1217 break;
1218 case 6:
1219 delta_parity = 2;
1220 re->level = 6;
1221 re->before.layout = ALGORITHM_PARITY_N;
1222 if (info->new_layout == UnSet)
1223 info->new_layout = map_name(r6layout, "default");
1224 break;
1225 default:
1226 return "Impossible level change requested";
1227 }
1228 re->before.data_disks = info->array.raid_disks;
1229 /* determining 'after' layout happens outside this 'switch' */
1230 break;
1231
1232 case 4:
1233 info->array.layout = ALGORITHM_PARITY_N;
1234 case 5:
1235 switch (info->new_level) {
1236 case 0:
1237 delta_parity = -1;
1238 case 4:
1239 re->level = info->array.level;
1240 re->before.data_disks = info->array.raid_disks - 1;
1241 re->before.layout = info->array.layout;
1242 break;
1243 case 5:
1244 re->level = 5;
1245 re->before.data_disks = info->array.raid_disks - 1;
1246 re->before.layout = info->array.layout;
1247 break;
1248 case 6:
1249 delta_parity = 1;
1250 re->level = 6;
1251 re->before.data_disks = info->array.raid_disks - 1;
1252 switch (info->array.layout) {
1253 case ALGORITHM_LEFT_ASYMMETRIC:
1254 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1255 break;
1256 case ALGORITHM_RIGHT_ASYMMETRIC:
1257 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1258 break;
1259 case ALGORITHM_LEFT_SYMMETRIC:
1260 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1261 break;
1262 case ALGORITHM_RIGHT_SYMMETRIC:
1263 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1264 break;
1265 case ALGORITHM_PARITY_0:
1266 re->before.layout = ALGORITHM_PARITY_0_6;
1267 break;
1268 case ALGORITHM_PARITY_N:
1269 re->before.layout = ALGORITHM_PARITY_N_6;
1270 break;
1271 default:
1272 return "Cannot convert an array with this layout";
1273 }
1274 break;
1275 case 1:
1276 if (info->array.raid_disks != 2)
1277 return "Can only convert a 2-device array to RAID1";
1278 if (info->delta_disks != UnSet &&
1279 info->delta_disks != 0)
1280 return "Cannot set raid_disk when converting RAID5->RAID1";
1281 re->level = 1;
1282 info->new_chunk = 0;
1283 return NULL;
1284 default:
1285 return "Impossible level change requested";
1286 }
1287 break;
1288 case 6:
1289 switch (info->new_level) {
1290 case 4:
1291 case 5:
1292 delta_parity = -1;
1293 case 6:
1294 re->level = 6;
1295 re->before.data_disks = info->array.raid_disks - 2;
1296 re->before.layout = info->array.layout;
1297 break;
1298 default:
1299 return "Impossible level change requested";
1300 }
1301 break;
1302 }
1303
1304 /* If we reached here then it looks like a re-stripe is
1305 * happening. We have determined the intermediate level
1306 * and initial raid_disks/layout and stored these in 're'.
1307 *
1308 * We need to deduce the final layout that can be atomically
1309 * converted to the end state.
1310 */
1311 switch (info->new_level) {
1312 case 0:
1313 /* We can only get to RAID0 from RAID4 or RAID5
1314 * with appropriate layout and one extra device
1315 */
1316 if (re->level != 4 && re->level != 5)
1317 return "Cannot covert to RAID0 from this level";
1318
1319 switch (re->level) {
1320 case 4:
1321 re->before.layout = 0;
1322 re->after.layout = 0;
1323 break;
1324 case 5:
1325 re->after.layout = ALGORITHM_PARITY_N;
1326 break;
1327 }
1328 break;
1329
1330 case 4:
1331 /* We can only get to RAID4 from RAID5 */
1332 if (re->level != 4 && re->level != 5)
1333 return "Cannot convert to RAID4 from this level";
1334
1335 switch (re->level) {
1336 case 4:
1337 re->after.layout = 0;
1338 break;
1339 case 5:
1340 re->after.layout = ALGORITHM_PARITY_N;
1341 break;
1342 }
1343 break;
1344
1345 case 5:
1346 /* We get to RAID5 from RAID5 or RAID6 */
1347 if (re->level != 5 && re->level != 6)
1348 return "Cannot convert to RAID5 from this level";
1349
1350 switch (re->level) {
1351 case 5:
1352 if (info->new_layout == UnSet)
1353 re->after.layout = re->before.layout;
1354 else
1355 re->after.layout = info->new_layout;
1356 break;
1357 case 6:
1358 if (info->new_layout == UnSet)
1359 info->new_layout = re->before.layout;
1360
1361 /* after.layout needs to be raid6 version of new_layout */
1362 if (info->new_layout == ALGORITHM_PARITY_N)
1363 re->after.layout = ALGORITHM_PARITY_N;
1364 else {
1365 char layout[40];
1366 char *ls = map_num(r5layout, info->new_layout);
1367 int l;
1368 if (ls) {
1369 /* Current RAID6 layout has a RAID5
1370 * equivalent - good
1371 */
1372 strcat(strcpy(layout, ls), "-6");
1373 l = map_name(r6layout, layout);
1374 if (l == UnSet)
1375 return "Cannot find RAID6 layout to convert to";
1376 } else {
1377 /* Current RAID6 has no equivalent.
1378 * If it is already a '-6' layout we
1379 * can leave it unchanged, else we must
1380 * fail
1381 */
1382 ls = map_num(r6layout, info->new_layout);
1383 if (!ls ||
1384 strcmp(ls+strlen(ls)-2, "-6") != 0)
1385 return "Please specify new layout";
1386 l = info->new_layout;
1387 }
1388 re->after.layout = l;
1389 }
1390 }
1391 break;
1392
1393 case 6:
1394 /* We must already be at level 6 */
1395 if (re->level != 6)
1396 return "Impossible level change";
1397 if (info->new_layout == UnSet)
1398 re->after.layout = info->array.layout;
1399 else
1400 re->after.layout = info->new_layout;
1401 break;
1402 default:
1403 return "Impossible level change requested";
1404 }
1405 if (info->delta_disks == UnSet)
1406 info->delta_disks = delta_parity;
1407
1408 re->after.data_disks = (re->before.data_disks
1409 + info->delta_disks
1410 - delta_parity);
1411 switch (re->level) {
1412 case 6: re->parity = 2;
1413 break;
1414 case 4:
1415 case 5: re->parity = 1;
1416 break;
1417 default: re->parity = 0;
1418 break;
1419 }
1420 /* So we have a restripe operation, we need to calculate the number
1421 * of blocks per reshape operation.
1422 */
1423 re->new_size = info->component_size * re->before.data_disks;
1424 if (info->new_chunk == 0)
1425 info->new_chunk = info->array.chunk_size;
1426 if (re->after.data_disks == re->before.data_disks &&
1427 re->after.layout == re->before.layout &&
1428 info->new_chunk == info->array.chunk_size) {
1429 /* Nothing to change, can change level immediately. */
1430 re->level = info->new_level;
1431 re->backup_blocks = 0;
1432 return NULL;
1433 }
1434 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1435 /* chunk and layout changes make no difference */
1436 re->level = info->new_level;
1437 re->backup_blocks = 0;
1438 return NULL;
1439 }
1440
1441 if (re->after.data_disks == re->before.data_disks &&
1442 get_linux_version() < 2006032)
1443 return "in-place reshape is not safe before 2.6.32 - sorry.";
1444
1445 if (re->after.data_disks < re->before.data_disks &&
1446 get_linux_version() < 2006030)
1447 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1448
1449 re->backup_blocks = compute_backup_blocks(
1450 info->new_chunk, info->array.chunk_size,
1451 re->after.data_disks,
1452 re->before.data_disks);
1453 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1454
1455 re->new_size = info->component_size * re->after.data_disks;
1456 return NULL;
1457 }
1458
1459 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1460 char *text_version)
1461 {
1462 struct mdinfo *info;
1463 char *subarray;
1464 int ret_val = -1;
1465
1466 if ((st == NULL) || (sra == NULL))
1467 return ret_val;
1468
1469 if (text_version == NULL)
1470 text_version = sra->text_version;
1471 subarray = strchr(text_version+1, '/')+1;
1472 info = st->ss->container_content(st, subarray);
1473 if (info) {
1474 unsigned long long current_size = 0;
1475 unsigned long long new_size =
1476 info->custom_array_size/2;
1477
1478 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1479 new_size > current_size) {
1480 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1481 < 0)
1482 dprintf("Error: Cannot set array size");
1483 else {
1484 ret_val = 0;
1485 dprintf("Array size changed");
1486 }
1487 dprintf_cont(" from %llu to %llu.\n",
1488 current_size, new_size);
1489 }
1490 sysfs_free(info);
1491 } else
1492 dprintf("Error: set_array_size(): info pointer in NULL\n");
1493
1494 return ret_val;
1495 }
1496
1497 static int reshape_array(char *container, int fd, char *devname,
1498 struct supertype *st, struct mdinfo *info,
1499 int force, struct mddev_dev *devlist,
1500 unsigned long long data_offset,
1501 char *backup_file, int verbose, int forked,
1502 int restart, int freeze_reshape);
1503 static int reshape_container(char *container, char *devname,
1504 int mdfd,
1505 struct supertype *st,
1506 struct mdinfo *info,
1507 int force,
1508 char *backup_file, int verbose,
1509 int forked, int restart, int freeze_reshape);
1510
1511 int Grow_reshape(char *devname, int fd,
1512 struct mddev_dev *devlist,
1513 unsigned long long data_offset,
1514 struct context *c, struct shape *s)
1515 {
1516 /* Make some changes in the shape of an array.
1517 * The kernel must support the change.
1518 *
1519 * There are three different changes. Each can trigger
1520 * a resync or recovery so we freeze that until we have
1521 * requested everything (if kernel supports freezing - 2.6.30).
1522 * The steps are:
1523 * - change size (i.e. component_size)
1524 * - change level
1525 * - change layout/chunksize/ndisks
1526 *
1527 * The last can require a reshape. It is different on different
1528 * levels so we need to check the level before actioning it.
1529 * Some times the level change needs to be requested after the
1530 * reshape (e.g. raid6->raid5, raid5->raid0)
1531 *
1532 */
1533 struct mdu_array_info_s array;
1534 int rv = 0;
1535 struct supertype *st;
1536 char *subarray = NULL;
1537
1538 int frozen;
1539 int changed = 0;
1540 char *container = NULL;
1541 int cfd = -1;
1542
1543 struct mddev_dev *dv;
1544 int added_disks;
1545
1546 struct mdinfo info;
1547 struct mdinfo *sra;
1548
1549 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1550 pr_err("%s is not an active md array - aborting\n",
1551 devname);
1552 return 1;
1553 }
1554 if (data_offset != INVALID_SECTORS && array.level != 10
1555 && (array.level < 4 || array.level > 6)) {
1556 pr_err("--grow --data-offset not yet supported\n");
1557 return 1;
1558 }
1559
1560 if (s->size > 0 &&
1561 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1562 pr_err("cannot change component size at the same time as other changes.\n"
1563 " Change size first, then check data is intact before making other changes.\n");
1564 return 1;
1565 }
1566
1567 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1568 get_linux_version() < 2006032 &&
1569 !check_env("MDADM_FORCE_FEWER")) {
1570 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1571 " Please use a newer kernel\n");
1572 return 1;
1573 }
1574
1575 st = super_by_fd(fd, &subarray);
1576 if (!st) {
1577 pr_err("Unable to determine metadata format for %s\n", devname);
1578 return 1;
1579 }
1580 if (s->raiddisks > st->max_devs) {
1581 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1582 return 1;
1583 }
1584
1585 /* in the external case we need to check that the requested reshape is
1586 * supported, and perform an initial check that the container holds the
1587 * pre-requisite spare devices (mdmon owns final validation)
1588 */
1589 if (st->ss->external) {
1590 int rv;
1591
1592 if (subarray) {
1593 container = st->container_devnm;
1594 cfd = open_dev_excl(st->container_devnm);
1595 } else {
1596 container = st->devnm;
1597 close(fd);
1598 cfd = open_dev_excl(st->devnm);
1599 fd = cfd;
1600 }
1601 if (cfd < 0) {
1602 pr_err("Unable to open container for %s\n",
1603 devname);
1604 free(subarray);
1605 return 1;
1606 }
1607
1608 rv = st->ss->load_container(st, cfd, NULL);
1609
1610 if (rv) {
1611 pr_err("Cannot read superblock for %s\n",
1612 devname);
1613 free(subarray);
1614 return 1;
1615 }
1616
1617 /* check if operation is supported for metadata handler */
1618 if (st->ss->container_content) {
1619 struct mdinfo *cc = NULL;
1620 struct mdinfo *content = NULL;
1621
1622 cc = st->ss->container_content(st, subarray);
1623 for (content = cc; content ; content = content->next) {
1624 int allow_reshape = 1;
1625
1626 /* check if reshape is allowed based on metadata
1627 * indications stored in content.array.status
1628 */
1629 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1630 allow_reshape = 0;
1631 if (content->array.state
1632 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1633 allow_reshape = 0;
1634 if (!allow_reshape) {
1635 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1636 devname, container);
1637 sysfs_free(cc);
1638 free(subarray);
1639 return 1;
1640 }
1641 }
1642 sysfs_free(cc);
1643 }
1644 if (mdmon_running(container))
1645 st->update_tail = &st->updates;
1646 }
1647
1648 added_disks = 0;
1649 for (dv = devlist; dv; dv = dv->next)
1650 added_disks++;
1651 if (s->raiddisks > array.raid_disks &&
1652 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1653 !c->force) {
1654 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1655 " Use --force to over-ride this check.\n",
1656 s->raiddisks - array.raid_disks,
1657 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1658 array.spare_disks + added_disks);
1659 return 1;
1660 }
1661
1662 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1663 | GET_STATE | GET_VERSION);
1664 if (sra) {
1665 if (st->ss->external && subarray == NULL) {
1666 array.level = LEVEL_CONTAINER;
1667 sra->array.level = LEVEL_CONTAINER;
1668 }
1669 } else {
1670 pr_err("failed to read sysfs parameters for %s\n",
1671 devname);
1672 return 1;
1673 }
1674 frozen = freeze(st);
1675 if (frozen < -1) {
1676 /* freeze() already spewed the reason */
1677 sysfs_free(sra);
1678 return 1;
1679 } else if (frozen < 0) {
1680 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1681 sysfs_free(sra);
1682 return 1;
1683 }
1684
1685 /* ========= set size =============== */
1686 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1687 unsigned long long orig_size = get_component_size(fd)/2;
1688 unsigned long long min_csize;
1689 struct mdinfo *mdi;
1690 int raid0_takeover = 0;
1691
1692 if (orig_size == 0)
1693 orig_size = (unsigned) array.size;
1694
1695 if (orig_size == 0) {
1696 pr_err("Cannot set device size in this type of array.\n");
1697 rv = 1;
1698 goto release;
1699 }
1700
1701 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1702 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1703 rv = 1;
1704 goto release;
1705 }
1706 sync_metadata(st);
1707 if (st->ss->external) {
1708 /* metadata can have size limitation
1709 * update size value according to metadata information
1710 */
1711 struct mdinfo *sizeinfo =
1712 st->ss->container_content(st, subarray);
1713 if (sizeinfo) {
1714 unsigned long long new_size =
1715 sizeinfo->custom_array_size/2;
1716 int data_disks = get_data_disks(
1717 sizeinfo->array.level,
1718 sizeinfo->array.layout,
1719 sizeinfo->array.raid_disks);
1720 new_size /= data_disks;
1721 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size, new_size,
1722 new_size * data_disks);
1723 s->size = new_size;
1724 sysfs_free(sizeinfo);
1725 }
1726 }
1727
1728 /* Update the size of each member device in case
1729 * they have been resized. This will never reduce
1730 * below the current used-size. The "size" attribute
1731 * understands '0' to mean 'max'.
1732 */
1733 min_csize = 0;
1734 rv = 0;
1735 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1736 if (sysfs_set_num(sra, mdi, "size",
1737 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1738 /* Probably kernel refusing to let us
1739 * reduce the size - not an error.
1740 */
1741 break;
1742 }
1743 if (array.not_persistent == 0 &&
1744 array.major_version == 0 &&
1745 get_linux_version() < 3001000) {
1746 /* Dangerous to allow size to exceed 2TB */
1747 unsigned long long csize;
1748 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1749 if (csize >= 2ULL*1024*1024*1024)
1750 csize = 2ULL*1024*1024*1024;
1751 if ((min_csize == 0 || (min_csize
1752 > csize)))
1753 min_csize = csize;
1754 }
1755 }
1756 }
1757 if (rv) {
1758 pr_err("Cannot set size on array members.\n");
1759 goto size_change_error;
1760 }
1761 if (min_csize && s->size > min_csize) {
1762 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1763 rv = 1;
1764 goto size_change_error;
1765 }
1766 if (min_csize && s->size == MAX_SIZE) {
1767 /* Don't let the kernel choose a size - it will get
1768 * it wrong
1769 */
1770 pr_err("Limited v0.90 array to 2TB per device\n");
1771 s->size = min_csize;
1772 }
1773 if (st->ss->external) {
1774 if (sra->array.level == 0) {
1775 rv = sysfs_set_str(sra, NULL, "level",
1776 "raid5");
1777 if (!rv) {
1778 raid0_takeover = 1;
1779 /* get array parametes after takeover
1780 * to chane one parameter at time only
1781 */
1782 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1783 }
1784 }
1785 /* make sure mdmon is
1786 * aware of the new level */
1787 if (!mdmon_running(st->container_devnm))
1788 start_mdmon(st->container_devnm);
1789 ping_monitor(container);
1790 if (mdmon_running(st->container_devnm) &&
1791 st->update_tail == NULL)
1792 st->update_tail = &st->updates;
1793 }
1794
1795 if (s->size == MAX_SIZE)
1796 s->size = 0;
1797 array.size = s->size;
1798 if (s->size & ~INT32_MAX) {
1799 /* got truncated to 32bit, write to
1800 * component_size instead
1801 */
1802 if (sra)
1803 rv = sysfs_set_num(sra, NULL,
1804 "component_size", s->size);
1805 else
1806 rv = -1;
1807 } else {
1808 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1809
1810 /* manage array size when it is managed externally
1811 */
1812 if ((rv == 0) && st->ss->external)
1813 rv = set_array_size(st, sra, sra->text_version);
1814 }
1815
1816 if (raid0_takeover) {
1817 /* do not recync non-existing parity,
1818 * we will drop it anyway
1819 */
1820 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1821 /* go back to raid0, drop parity disk
1822 */
1823 sysfs_set_str(sra, NULL, "level", "raid0");
1824 ioctl(fd, GET_ARRAY_INFO, &array);
1825 }
1826
1827 size_change_error:
1828 if (rv != 0) {
1829 int err = errno;
1830
1831 /* restore metadata */
1832 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1833 UnSet, NULL, devname,
1834 ROLLBACK_METADATA_CHANGES,
1835 c->verbose) == 0)
1836 sync_metadata(st);
1837 pr_err("Cannot set device size for %s: %s\n",
1838 devname, strerror(err));
1839 if (err == EBUSY &&
1840 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1841 cont_err("Bitmap must be removed before size can be changed\n");
1842 rv = 1;
1843 goto release;
1844 }
1845 if (s->assume_clean) {
1846 /* This will fail on kernels older than 3.0 unless
1847 * a backport has been arranged.
1848 */
1849 if (sra == NULL ||
1850 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1851 pr_err("--assume-clean not supported with --grow on this kernel\n");
1852 }
1853 ioctl(fd, GET_ARRAY_INFO, &array);
1854 s->size = get_component_size(fd)/2;
1855 if (s->size == 0)
1856 s->size = array.size;
1857 if (c->verbose >= 0) {
1858 if (s->size == orig_size)
1859 pr_err("component size of %s unchanged at %lluK\n",
1860 devname, s->size);
1861 else
1862 pr_err("component size of %s has been set to %lluK\n",
1863 devname, s->size);
1864 }
1865 changed = 1;
1866 } else if (array.level != LEVEL_CONTAINER) {
1867 s->size = get_component_size(fd)/2;
1868 if (s->size == 0)
1869 s->size = array.size;
1870 }
1871
1872 /* See if there is anything else to do */
1873 if ((s->level == UnSet || s->level == array.level) &&
1874 (s->layout_str == NULL) &&
1875 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1876 data_offset == INVALID_SECTORS &&
1877 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1878 /* Nothing more to do */
1879 if (!changed && c->verbose >= 0)
1880 pr_err("%s: no change requested\n",
1881 devname);
1882 goto release;
1883 }
1884
1885 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1886 * current implementation assumes that following conditions must be met:
1887 * - RAID10:
1888 * - far_copies == 1
1889 * - near_copies == 2
1890 */
1891 if ((s->level == 0 && array.level == 10 && sra &&
1892 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1893 (s->level == 0 && array.level == 1 && sra)) {
1894 int err;
1895 err = remove_disks_for_takeover(st, sra, array.layout);
1896 if (err) {
1897 dprintf("Array cannot be reshaped\n");
1898 if (cfd > -1)
1899 close(cfd);
1900 rv = 1;
1901 goto release;
1902 }
1903 /* Make sure mdmon has seen the device removal
1904 * and updated metadata before we continue with
1905 * level change
1906 */
1907 if (container)
1908 ping_monitor(container);
1909 }
1910
1911 memset(&info, 0, sizeof(info));
1912 info.array = array;
1913 sysfs_init(&info, fd, NULL);
1914 strcpy(info.text_version, sra->text_version);
1915 info.component_size = s->size*2;
1916 info.new_level = s->level;
1917 info.new_chunk = s->chunk * 1024;
1918 if (info.array.level == LEVEL_CONTAINER) {
1919 info.delta_disks = UnSet;
1920 info.array.raid_disks = s->raiddisks;
1921 } else if (s->raiddisks)
1922 info.delta_disks = s->raiddisks - info.array.raid_disks;
1923 else
1924 info.delta_disks = UnSet;
1925 if (s->layout_str == NULL) {
1926 info.new_layout = UnSet;
1927 if (info.array.level == 6 &&
1928 (info.new_level == 6 || info.new_level == UnSet) &&
1929 info.array.layout >= 16) {
1930 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
1931 cont_err("during the reshape, please specify --layout=preserve\n");
1932 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1933 rv = 1;
1934 goto release;
1935 }
1936 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1937 strcmp(s->layout_str, "normalize") == 0) {
1938 /* If we have a -6 RAID6 layout, remove the '-6'. */
1939 info.new_layout = UnSet;
1940 if (info.array.level == 6 && info.new_level == UnSet) {
1941 char l[40], *h;
1942 strcpy(l, map_num(r6layout, info.array.layout));
1943 h = strrchr(l, '-');
1944 if (h && strcmp(h, "-6") == 0) {
1945 *h = 0;
1946 info.new_layout = map_name(r6layout, l);
1947 }
1948 } else {
1949 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
1950 rv = 1;
1951 goto release;
1952 }
1953 } else if (strcmp(s->layout_str, "preserve") == 0) {
1954 /* This means that a non-standard RAID6 layout
1955 * is OK.
1956 * In particular:
1957 * - When reshape a RAID6 (e.g. adding a device)
1958 * which is in a non-standard layout, it is OK
1959 * to preserve that layout.
1960 * - When converting a RAID5 to RAID6, leave it in
1961 * the XXX-6 layout, don't re-layout.
1962 */
1963 if (info.array.level == 6 && info.new_level == UnSet)
1964 info.new_layout = info.array.layout;
1965 else if (info.array.level == 5 && info.new_level == 6) {
1966 char l[40];
1967 strcpy(l, map_num(r5layout, info.array.layout));
1968 strcat(l, "-6");
1969 info.new_layout = map_name(r6layout, l);
1970 } else {
1971 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
1972 rv = 1;
1973 goto release;
1974 }
1975 } else {
1976 int l = info.new_level;
1977 if (l == UnSet)
1978 l = info.array.level;
1979 switch (l) {
1980 case 5:
1981 info.new_layout = map_name(r5layout, s->layout_str);
1982 break;
1983 case 6:
1984 info.new_layout = map_name(r6layout, s->layout_str);
1985 break;
1986 case 10:
1987 info.new_layout = parse_layout_10(s->layout_str);
1988 break;
1989 case LEVEL_FAULTY:
1990 info.new_layout = parse_layout_faulty(s->layout_str);
1991 break;
1992 default:
1993 pr_err("layout not meaningful with this level\n");
1994 rv = 1;
1995 goto release;
1996 }
1997 if (info.new_layout == UnSet) {
1998 pr_err("layout %s not understood for this level\n",
1999 s->layout_str);
2000 rv = 1;
2001 goto release;
2002 }
2003 }
2004
2005 if (array.level == LEVEL_FAULTY) {
2006 if (s->level != UnSet && s->level != array.level) {
2007 pr_err("cannot change level of Faulty device\n");
2008 rv =1 ;
2009 }
2010 if (s->chunk) {
2011 pr_err("cannot set chunksize of Faulty device\n");
2012 rv =1 ;
2013 }
2014 if (s->raiddisks && s->raiddisks != 1) {
2015 pr_err("cannot set raid_disks of Faulty device\n");
2016 rv =1 ;
2017 }
2018 if (s->layout_str) {
2019 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2020 dprintf("Cannot get array information.\n");
2021 goto release;
2022 }
2023 array.layout = info.new_layout;
2024 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2025 pr_err("failed to set new layout\n");
2026 rv = 1;
2027 } else if (c->verbose >= 0)
2028 printf("layout for %s set to %d\n",
2029 devname, array.layout);
2030 }
2031 } else if (array.level == LEVEL_CONTAINER) {
2032 /* This change is to be applied to every array in the
2033 * container. This is only needed when the metadata imposes
2034 * restraints of the various arrays in the container.
2035 * Currently we only know that IMSM requires all arrays
2036 * to have the same number of devices so changing the
2037 * number of devices (On-Line Capacity Expansion) must be
2038 * performed at the level of the container
2039 */
2040 rv = reshape_container(container, devname, -1, st, &info,
2041 c->force, c->backup_file, c->verbose, 0, 0, 0);
2042 frozen = 0;
2043 } else {
2044 /* get spare devices from external metadata
2045 */
2046 if (st->ss->external) {
2047 struct mdinfo *info2;
2048
2049 info2 = st->ss->container_content(st, subarray);
2050 if (info2) {
2051 info.array.spare_disks =
2052 info2->array.spare_disks;
2053 sysfs_free(info2);
2054 }
2055 }
2056
2057 /* Impose these changes on a single array. First
2058 * check that the metadata is OK with the change. */
2059
2060 if (reshape_super(st, 0, info.new_level,
2061 info.new_layout, info.new_chunk,
2062 info.array.raid_disks, info.delta_disks,
2063 c->backup_file, devname, APPLY_METADATA_CHANGES,
2064 c->verbose)) {
2065 rv = 1;
2066 goto release;
2067 }
2068 sync_metadata(st);
2069 rv = reshape_array(container, fd, devname, st, &info, c->force,
2070 devlist, data_offset, c->backup_file, c->verbose,
2071 0, 0, 0);
2072 frozen = 0;
2073 }
2074 release:
2075 sysfs_free(sra);
2076 if (frozen > 0)
2077 unfreeze(st);
2078 return rv;
2079 }
2080
2081 /* verify_reshape_position()
2082 * Function checks if reshape position in metadata is not farther
2083 * than position in md.
2084 * Return value:
2085 * 0 : not valid sysfs entry
2086 * it can be caused by not started reshape, it should be started
2087 * by reshape array or raid0 array is before takeover
2088 * -1 : error, reshape position is obviously wrong
2089 * 1 : success, reshape progress correct or updated
2090 */
2091 static int verify_reshape_position(struct mdinfo *info, int level)
2092 {
2093 int ret_val = 0;
2094 char buf[40];
2095 int rv;
2096
2097 /* read sync_max, failure can mean raid0 array */
2098 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2099
2100 if (rv > 0) {
2101 char *ep;
2102 unsigned long long position = strtoull(buf, &ep, 0);
2103
2104 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2105 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2106 position *= get_data_disks(level,
2107 info->new_layout,
2108 info->array.raid_disks);
2109 if (info->reshape_progress < position) {
2110 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2111 info->reshape_progress, position);
2112 info->reshape_progress = position;
2113 ret_val = 1;
2114 } else if (info->reshape_progress > position) {
2115 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2116 position, info->reshape_progress);
2117 ret_val = -1;
2118 } else {
2119 dprintf("Reshape position in md and metadata are the same;");
2120 ret_val = 1;
2121 }
2122 }
2123 } else if (rv == 0) {
2124 /* for valid sysfs entry, 0-length content
2125 * should be indicated as error
2126 */
2127 ret_val = -1;
2128 }
2129
2130 return ret_val;
2131 }
2132
2133 static unsigned long long choose_offset(unsigned long long lo,
2134 unsigned long long hi,
2135 unsigned long long min,
2136 unsigned long long max)
2137 {
2138 /* Choose a new offset between hi and lo.
2139 * It must be between min and max, but
2140 * we would prefer something near the middle of hi/lo, and also
2141 * prefer to be aligned to a big power of 2.
2142 *
2143 * So we start with the middle, then for each bit,
2144 * starting at '1' and increasing, if it is set, we either
2145 * add it or subtract it if possible, preferring the option
2146 * which is furthest from the boundary.
2147 *
2148 * We stop once we get a 1MB alignment. As units are in sectors,
2149 * 1MB = 2*1024 sectors.
2150 */
2151 unsigned long long choice = (lo + hi) / 2;
2152 unsigned long long bit = 1;
2153
2154 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2155 unsigned long long bigger, smaller;
2156 if (! (bit & choice))
2157 continue;
2158 bigger = choice + bit;
2159 smaller = choice - bit;
2160 if (bigger > max && smaller < min)
2161 break;
2162 if (bigger > max)
2163 choice = smaller;
2164 else if (smaller < min)
2165 choice = bigger;
2166 else if (hi - bigger > smaller - lo)
2167 choice = bigger;
2168 else
2169 choice = smaller;
2170 }
2171 return choice;
2172 }
2173
2174 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2175 char *devname, int delta_disks,
2176 unsigned long long data_offset,
2177 unsigned long long min,
2178 int can_fallback)
2179 {
2180 struct mdinfo *sd;
2181 int dir = 0;
2182 int err = 0;
2183 unsigned long long before, after;
2184
2185 /* Need to find min space before and after so same is used
2186 * on all devices
2187 */
2188 before = UINT64_MAX;
2189 after = UINT64_MAX;
2190 for (sd = sra->devs; sd; sd = sd->next) {
2191 char *dn;
2192 int dfd;
2193 int rv;
2194 struct supertype *st2;
2195 struct mdinfo info2;
2196
2197 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2198 continue;
2199 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2200 dfd = dev_open(dn, O_RDONLY);
2201 if (dfd < 0) {
2202 pr_err("%s: cannot open component %s\n",
2203 devname, dn ? dn : "-unknown-");
2204 goto release;
2205 }
2206 st2 = dup_super(st);
2207 rv = st2->ss->load_super(st2,dfd, NULL);
2208 close(dfd);
2209 if (rv) {
2210 free(st2);
2211 pr_err("%s: cannot get superblock from %s\n",
2212 devname, dn);
2213 goto release;
2214 }
2215 st2->ss->getinfo_super(st2, &info2, NULL);
2216 st2->ss->free_super(st2);
2217 free(st2);
2218 if (info2.space_before == 0 &&
2219 info2.space_after == 0) {
2220 /* Metadata doesn't support data_offset changes */
2221 if (!can_fallback)
2222 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2223 devname);
2224 goto fallback;
2225 }
2226 if (before > info2.space_before)
2227 before = info2.space_before;
2228 if (after > info2.space_after)
2229 after = info2.space_after;
2230
2231 if (data_offset != INVALID_SECTORS) {
2232 if (dir == 0) {
2233 if (info2.data_offset == data_offset) {
2234 pr_err("%s: already has that data_offset\n",
2235 dn);
2236 goto release;
2237 }
2238 if (data_offset < info2.data_offset)
2239 dir = -1;
2240 else
2241 dir = 1;
2242 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2243 (data_offset >= info2.data_offset && dir == -1)) {
2244 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2245 dn);
2246 goto release;
2247 }
2248 }
2249 }
2250 if (before == UINT64_MAX)
2251 /* impossible really, there must be no devices */
2252 return 1;
2253
2254 for (sd = sra->devs; sd; sd = sd->next) {
2255 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2256 unsigned long long new_data_offset;
2257
2258 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2259 continue;
2260 if (delta_disks < 0) {
2261 /* Don't need any space as array is shrinking
2262 * just move data_offset up by min
2263 */
2264 if (data_offset == INVALID_SECTORS)
2265 new_data_offset = sd->data_offset + min;
2266 else {
2267 if (data_offset < sd->data_offset + min) {
2268 pr_err("--data-offset too small for %s\n",
2269 dn);
2270 goto release;
2271 }
2272 new_data_offset = data_offset;
2273 }
2274 } else if (delta_disks > 0) {
2275 /* need space before */
2276 if (before < min) {
2277 if (can_fallback)
2278 goto fallback;
2279 pr_err("Insufficient head-space for reshape on %s\n",
2280 dn);
2281 goto release;
2282 }
2283 if (data_offset == INVALID_SECTORS)
2284 new_data_offset = sd->data_offset - min;
2285 else {
2286 if (data_offset > sd->data_offset - min) {
2287 pr_err("--data-offset too large for %s\n",
2288 dn);
2289 goto release;
2290 }
2291 new_data_offset = data_offset;
2292 }
2293 } else {
2294 if (dir == 0) {
2295 /* can move up or down. If 'data_offset'
2296 * was set we would have already decided,
2297 * so just choose direction with most space.
2298 */
2299 if (before > after)
2300 dir = -1;
2301 else
2302 dir = 1;
2303 }
2304 sysfs_set_str(sra, NULL, "reshape_direction",
2305 dir == 1 ? "backwards" : "forwards");
2306 if (dir > 0) {
2307 /* Increase data offset */
2308 if (after < min) {
2309 if (can_fallback)
2310 goto fallback;
2311 pr_err("Insufficient tail-space for reshape on %s\n",
2312 dn);
2313 goto release;
2314 }
2315 if (data_offset != INVALID_SECTORS &&
2316 data_offset < sd->data_offset + min) {
2317 pr_err("--data-offset too small on %s\n",
2318 dn);
2319 goto release;
2320 }
2321 if (data_offset != INVALID_SECTORS)
2322 new_data_offset = data_offset;
2323 else
2324 new_data_offset = choose_offset(sd->data_offset,
2325 sd->data_offset + after,
2326 sd->data_offset + min,
2327 sd->data_offset + after);
2328 } else {
2329 /* Decrease data offset */
2330 if (before < min) {
2331 if (can_fallback)
2332 goto fallback;
2333 pr_err("insufficient head-room on %s\n",
2334 dn);
2335 goto release;
2336 }
2337 if (data_offset != INVALID_SECTORS &&
2338 data_offset < sd->data_offset - min) {
2339 pr_err("--data-offset too small on %s\n",
2340 dn);
2341 goto release;
2342 }
2343 if (data_offset != INVALID_SECTORS)
2344 new_data_offset = data_offset;
2345 else
2346 new_data_offset = choose_offset(sd->data_offset - before,
2347 sd->data_offset,
2348 sd->data_offset - before,
2349 sd->data_offset - min);
2350 }
2351 }
2352 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2353 if (err < 0 && errno == E2BIG) {
2354 /* try again after increasing data size to max */
2355 err = sysfs_set_num(sra, sd, "size", 0);
2356 if (err < 0 && errno == EINVAL &&
2357 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2358 /* some kernels have a bug where you cannot
2359 * use '0' on spare devices. */
2360 sysfs_set_num(sra, sd, "size",
2361 (sra->component_size + after)/2);
2362 }
2363 err = sysfs_set_num(sra, sd, "new_offset",
2364 new_data_offset);
2365 }
2366 if (err < 0) {
2367 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2368 pr_err("data-offset is too big for %s\n",
2369 dn);
2370 goto release;
2371 }
2372 if (sd == sra->devs &&
2373 (errno == ENOENT || errno == E2BIG))
2374 /* Early kernel, no 'new_offset' file,
2375 * or kernel doesn't like us.
2376 * For RAID5/6 this is not fatal
2377 */
2378 return 1;
2379 pr_err("Cannot set new_offset for %s\n",
2380 dn);
2381 break;
2382 }
2383 }
2384 return err;
2385 release:
2386 return -1;
2387 fallback:
2388 /* Just use a backup file */
2389 return 1;
2390 }
2391
2392 static int raid10_reshape(char *container, int fd, char *devname,
2393 struct supertype *st, struct mdinfo *info,
2394 struct reshape *reshape,
2395 unsigned long long data_offset,
2396 int force, int verbose)
2397 {
2398 /* Changing raid_disks, layout, chunksize or possibly
2399 * just data_offset for a RAID10.
2400 * We must always change data_offset. We change by at least
2401 * ->min_offset_change which is the largest of the old and new
2402 * chunk sizes.
2403 * If raid_disks is increasing, then data_offset must decrease
2404 * by at least this copy size.
2405 * If raid_disks is unchanged, data_offset must increase or
2406 * decrease by at least min_offset_change but preferably by much more.
2407 * We choose half of the available space.
2408 * If raid_disks is decreasing, data_offset must increase by
2409 * at least min_offset_change. To allow of this, component_size
2410 * must be decreased by the same amount.
2411 *
2412 * So we calculate the required minimum and direction, possibly
2413 * reduce the component_size, then iterate through the devices
2414 * and set the new_data_offset.
2415 * If that all works, we set chunk_size, layout, raid_disks, and start
2416 * 'reshape'
2417 */
2418 struct mdinfo *sra;
2419 unsigned long long min;
2420 int err = 0;
2421
2422 sra = sysfs_read(fd, NULL,
2423 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2424 );
2425 if (!sra) {
2426 pr_err("%s: Cannot get array details from sysfs\n",
2427 devname);
2428 goto release;
2429 }
2430 min = reshape->min_offset_change;
2431
2432 if (info->delta_disks)
2433 sysfs_set_str(sra, NULL, "reshape_direction",
2434 info->delta_disks < 0 ? "backwards" : "forwards");
2435 if (info->delta_disks < 0 &&
2436 info->space_after < min) {
2437 int rv = sysfs_set_num(sra, NULL, "component_size",
2438 (sra->component_size -
2439 min)/2);
2440 if (rv) {
2441 pr_err("cannot reduce component size\n");
2442 goto release;
2443 }
2444 }
2445 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2446 min, 0);
2447 if (err == 1) {
2448 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2449 cont_err("supported on this kernel\n");
2450 err = -1;
2451 }
2452 if (err < 0)
2453 goto release;
2454
2455 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2456 err = errno;
2457 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2458 err = errno;
2459 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2460 info->array.raid_disks + info->delta_disks) < 0)
2461 err = errno;
2462 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2463 err = errno;
2464 if (err) {
2465 pr_err("Cannot set array shape for %s\n",
2466 devname);
2467 if (err == EBUSY &&
2468 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2469 cont_err(" Bitmap must be removed before shape can be changed\n");
2470 goto release;
2471 }
2472 sysfs_free(sra);
2473 return 0;
2474 release:
2475 sysfs_free(sra);
2476 return 1;
2477 }
2478
2479 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2480 {
2481 struct mdinfo *sra, *sd;
2482 /* Initialisation to silence compiler warning */
2483 unsigned long long min_space_before = 0, min_space_after = 0;
2484 int first = 1;
2485
2486 sra = sysfs_read(fd, NULL, GET_DEVS);
2487 if (!sra)
2488 return;
2489 for (sd = sra->devs; sd; sd = sd->next) {
2490 char *dn;
2491 int dfd;
2492 struct supertype *st2;
2493 struct mdinfo info2;
2494
2495 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2496 continue;
2497 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2498 dfd = dev_open(dn, O_RDONLY);
2499 if (dfd < 0)
2500 break;
2501 st2 = dup_super(st);
2502 if (st2->ss->load_super(st2,dfd, NULL)) {
2503 close(dfd);
2504 free(st2);
2505 break;
2506 }
2507 close(dfd);
2508 st2->ss->getinfo_super(st2, &info2, NULL);
2509 st2->ss->free_super(st2);
2510 free(st2);
2511 if (first ||
2512 min_space_before > info2.space_before)
2513 min_space_before = info2.space_before;
2514 if (first ||
2515 min_space_after > info2.space_after)
2516 min_space_after = info2.space_after;
2517 first = 0;
2518 }
2519 if (sd == NULL && !first) {
2520 info->space_after = min_space_after;
2521 info->space_before = min_space_before;
2522 }
2523 sysfs_free(sra);
2524 }
2525
2526 static void update_cache_size(char *container, struct mdinfo *sra,
2527 struct mdinfo *info,
2528 int disks, unsigned long long blocks)
2529 {
2530 /* Check that the internal stripe cache is
2531 * large enough, or it won't work.
2532 * It must hold at least 4 stripes of the larger
2533 * chunk size
2534 */
2535 unsigned long cache;
2536 cache = max(info->array.chunk_size, info->new_chunk);
2537 cache *= 4; /* 4 stripes minimum */
2538 cache /= 512; /* convert to sectors */
2539 /* make sure there is room for 'blocks' with a bit to spare */
2540 if (cache < 16 + blocks / disks)
2541 cache = 16 + blocks / disks;
2542 cache /= (4096/512); /* Covert from sectors to pages */
2543
2544 if (sra->cache_size < cache)
2545 subarray_set_num(container, sra, "stripe_cache_size",
2546 cache+1);
2547 }
2548
2549 static int impose_reshape(struct mdinfo *sra,
2550 struct mdinfo *info,
2551 struct supertype *st,
2552 int fd,
2553 int restart,
2554 char *devname, char *container,
2555 struct reshape *reshape)
2556 {
2557 struct mdu_array_info_s array;
2558
2559 sra->new_chunk = info->new_chunk;
2560
2561 if (restart) {
2562 /* for external metadata checkpoint saved by mdmon can be lost
2563 * or missed /due to e.g. crash/. Check if md is not during
2564 * restart farther than metadata points to.
2565 * If so, this means metadata information is obsolete.
2566 */
2567 if (st->ss->external)
2568 verify_reshape_position(info, reshape->level);
2569 sra->reshape_progress = info->reshape_progress;
2570 } else {
2571 sra->reshape_progress = 0;
2572 if (reshape->after.data_disks < reshape->before.data_disks)
2573 /* start from the end of the new array */
2574 sra->reshape_progress = (sra->component_size
2575 * reshape->after.data_disks);
2576 }
2577
2578 ioctl(fd, GET_ARRAY_INFO, &array);
2579 if (info->array.chunk_size == info->new_chunk &&
2580 reshape->before.layout == reshape->after.layout &&
2581 st->ss->external == 0) {
2582 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2583 array.raid_disks = reshape->after.data_disks + reshape->parity;
2584 if (!restart &&
2585 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2586 int err = errno;
2587
2588 pr_err("Cannot set device shape for %s: %s\n",
2589 devname, strerror(errno));
2590
2591 if (err == EBUSY &&
2592 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2593 cont_err("Bitmap must be removed before shape can be changed\n");
2594
2595 goto release;
2596 }
2597 } else if (!restart) {
2598 /* set them all just in case some old 'new_*' value
2599 * persists from some earlier problem.
2600 */
2601 int err = 0;
2602 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2603 err = errno;
2604 if (!err && sysfs_set_num(sra, NULL, "layout",
2605 reshape->after.layout) < 0)
2606 err = errno;
2607 if (!err && subarray_set_num(container, sra, "raid_disks",
2608 reshape->after.data_disks +
2609 reshape->parity) < 0)
2610 err = errno;
2611 if (err) {
2612 pr_err("Cannot set device shape for %s\n",
2613 devname);
2614
2615 if (err == EBUSY &&
2616 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2617 cont_err("Bitmap must be removed before shape can be changed\n");
2618 goto release;
2619 }
2620 }
2621 return 0;
2622 release:
2623 return -1;
2624 }
2625
2626 static int impose_level(int fd, int level, char *devname, int verbose)
2627 {
2628 char *c;
2629 struct mdu_array_info_s array;
2630 struct mdinfo info;
2631 sysfs_init(&info, fd, NULL);
2632
2633 ioctl(fd, GET_ARRAY_INFO, &array);
2634 if (level == 0 &&
2635 (array.level >= 4 && array.level <= 6)) {
2636 /* To convert to RAID0 we need to fail and
2637 * remove any non-data devices. */
2638 int found = 0;
2639 int d;
2640 int data_disks = array.raid_disks - 1;
2641 if (array.level == 6)
2642 data_disks -= 1;
2643 if (array.level == 5 &&
2644 array.layout != ALGORITHM_PARITY_N)
2645 return -1;
2646 if (array.level == 6 &&
2647 array.layout != ALGORITHM_PARITY_N_6)
2648 return -1;
2649 sysfs_set_str(&info, NULL,"sync_action", "idle");
2650 /* First remove any spares so no recovery starts */
2651 for (d = 0, found = 0;
2652 d < MAX_DISKS && found < array.nr_disks;
2653 d++) {
2654 mdu_disk_info_t disk;
2655 disk.number = d;
2656 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2657 continue;
2658 if (disk.major == 0 && disk.minor == 0)
2659 continue;
2660 found++;
2661 if ((disk.state & (1 << MD_DISK_ACTIVE))
2662 && disk.raid_disk < data_disks)
2663 /* keep this */
2664 continue;
2665 ioctl(fd, HOT_REMOVE_DISK,
2666 makedev(disk.major, disk.minor));
2667 }
2668 /* Now fail anything left */
2669 ioctl(fd, GET_ARRAY_INFO, &array);
2670 for (d = 0, found = 0;
2671 d < MAX_DISKS && found < array.nr_disks;
2672 d++) {
2673 int cnt;
2674 mdu_disk_info_t disk;
2675 disk.number = d;
2676 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2677 continue;
2678 if (disk.major == 0 && disk.minor == 0)
2679 continue;
2680 found++;
2681 if ((disk.state & (1 << MD_DISK_ACTIVE))
2682 && disk.raid_disk < data_disks)
2683 /* keep this */
2684 continue;
2685 ioctl(fd, SET_DISK_FAULTY,
2686 makedev(disk.major, disk.minor));
2687 cnt = 5;
2688 while (ioctl(fd, HOT_REMOVE_DISK,
2689 makedev(disk.major, disk.minor)) < 0
2690 && errno == EBUSY
2691 && cnt--) {
2692 usleep(10000);
2693 }
2694 }
2695 }
2696 c = map_num(pers, level);
2697 if (c) {
2698 int err = sysfs_set_str(&info, NULL, "level", c);
2699 if (err) {
2700 err = errno;
2701 pr_err("%s: could not set level to %s\n",
2702 devname, c);
2703 if (err == EBUSY &&
2704 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2705 cont_err("Bitmap must be removed before level can be changed\n");
2706 return err;
2707 }
2708 if (verbose >= 0)
2709 pr_err("level of %s changed to %s\n",
2710 devname, c);
2711 }
2712 return 0;
2713 }
2714
2715 int sigterm = 0;
2716 static void catch_term(int sig)
2717 {
2718 sigterm = 1;
2719 }
2720
2721 static int continue_via_systemd(char *devnm)
2722 {
2723 int skipped, i, pid, status;
2724 char pathbuf[1024];
2725 /* In a systemd/udev world, it is best to get systemd to
2726 * run "mdadm --grow --continue" rather than running in the
2727 * background.
2728 */
2729 switch(fork()) {
2730 case 0:
2731 /* FIXME yuk. CLOSE_EXEC?? */
2732 skipped = 0;
2733 for (i = 3; skipped < 20; i++)
2734 if (close(i) < 0)
2735 skipped++;
2736 else
2737 skipped = 0;
2738
2739 /* Don't want to see error messages from
2740 * systemctl. If the service doesn't exist,
2741 * we fork ourselves.
2742 */
2743 close(2);
2744 open("/dev/null", O_WRONLY);
2745 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2746 devnm);
2747 status = execl("/usr/bin/systemctl", "systemctl",
2748 "start",
2749 pathbuf, NULL);
2750 status = execl("/bin/systemctl", "systemctl", "start",
2751 pathbuf, NULL);
2752 exit(1);
2753 case -1: /* Just do it ourselves. */
2754 break;
2755 default: /* parent - good */
2756 pid = wait(&status);
2757 if (pid >= 0 && status == 0)
2758 return 1;
2759 }
2760 return 0;
2761 }
2762
2763 static int reshape_array(char *container, int fd, char *devname,
2764 struct supertype *st, struct mdinfo *info,
2765 int force, struct mddev_dev *devlist,
2766 unsigned long long data_offset,
2767 char *backup_file, int verbose, int forked,
2768 int restart, int freeze_reshape)
2769 {
2770 struct reshape reshape;
2771 int spares_needed;
2772 char *msg;
2773 int orig_level = UnSet;
2774 int odisks;
2775 int delayed;
2776
2777 struct mdu_array_info_s array;
2778 char *c;
2779
2780 struct mddev_dev *dv;
2781 int added_disks;
2782
2783 int *fdlist = NULL;
2784 unsigned long long *offsets = NULL;
2785 int d;
2786 int nrdisks;
2787 int err;
2788 unsigned long blocks;
2789 unsigned long long array_size;
2790 int done;
2791 struct mdinfo *sra = NULL;
2792 char buf[20];
2793
2794 /* when reshaping a RAID0, the component_size might be zero.
2795 * So try to fix that up.
2796 */
2797 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2798 dprintf("Cannot get array information.\n");
2799 goto release;
2800 }
2801 if (array.level == 0 && info->component_size == 0) {
2802 get_dev_size(fd, NULL, &array_size);
2803 info->component_size = array_size / array.raid_disks;
2804 }
2805
2806 if (array.level == 10)
2807 /* Need space_after info */
2808 get_space_after(fd, st, info);
2809
2810 if (info->reshape_active) {
2811 int new_level = info->new_level;
2812 info->new_level = UnSet;
2813 if (info->delta_disks > 0)
2814 info->array.raid_disks -= info->delta_disks;
2815 msg = analyse_change(devname, info, &reshape);
2816 info->new_level = new_level;
2817 if (info->delta_disks > 0)
2818 info->array.raid_disks += info->delta_disks;
2819 if (!restart)
2820 /* Make sure the array isn't read-only */
2821 ioctl(fd, RESTART_ARRAY_RW, 0);
2822 } else
2823 msg = analyse_change(devname, info, &reshape);
2824 if (msg) {
2825 /* if msg == "", error has already been printed */
2826 if (msg[0])
2827 pr_err("%s\n", msg);
2828 goto release;
2829 }
2830 if (restart &&
2831 (reshape.level != info->array.level ||
2832 reshape.before.layout != info->array.layout ||
2833 reshape.before.data_disks + reshape.parity
2834 != info->array.raid_disks - max(0, info->delta_disks))) {
2835 pr_err("reshape info is not in native format - cannot continue.\n");
2836 goto release;
2837 }
2838
2839 if (st->ss->external && restart && (info->reshape_progress == 0) &&
2840 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
2841 (strncmp(buf, "reshape", 7) == 0))) {
2842 /* When reshape is restarted from '0', very begin of array
2843 * it is possible that for external metadata reshape and array
2844 * configuration doesn't happen.
2845 * Check if md has the same opinion, and reshape is restarted
2846 * from 0. If so, this is regular reshape start after reshape
2847 * switch in metadata to next array only.
2848 */
2849 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2850 (info->reshape_progress == 0))
2851 restart = 0;
2852 }
2853 if (restart) {
2854 /* reshape already started. just skip to monitoring the reshape */
2855 if (reshape.backup_blocks == 0)
2856 return 0;
2857 if (restart & RESHAPE_NO_BACKUP)
2858 return 0;
2859
2860 /* Need 'sra' down at 'started:' */
2861 sra = sysfs_read(fd, NULL,
2862 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2863 GET_CACHE);
2864 if (!sra) {
2865 pr_err("%s: Cannot get array details from sysfs\n",
2866 devname);
2867 goto release;
2868 }
2869
2870 if (!backup_file)
2871 backup_file = locate_backup(sra->sys_name);
2872
2873 goto started;
2874 }
2875 /* The container is frozen but the array may not be.
2876 * So freeze the array so spares don't get put to the wrong use
2877 * FIXME there should probably be a cleaner separation between
2878 * freeze_array and freeze_container.
2879 */
2880 sysfs_freeze_array(info);
2881 /* Check we have enough spares to not be degraded */
2882 added_disks = 0;
2883 for (dv = devlist; dv ; dv=dv->next)
2884 added_disks++;
2885 spares_needed = max(reshape.before.data_disks,
2886 reshape.after.data_disks)
2887 + reshape.parity - array.raid_disks;
2888
2889 if (!force &&
2890 info->new_level > 1 && info->array.level > 1 &&
2891 spares_needed > info->array.spare_disks + added_disks) {
2892 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2893 " Use --force to over-ride this check.\n",
2894 spares_needed,
2895 spares_needed == 1 ? "" : "s",
2896 info->array.spare_disks + added_disks);
2897 goto release;
2898 }
2899 /* Check we have enough spares to not fail */
2900 spares_needed = max(reshape.before.data_disks,
2901 reshape.after.data_disks)
2902 - array.raid_disks;
2903 if ((info->new_level > 1 || info->new_level == 0) &&
2904 spares_needed > info->array.spare_disks +added_disks) {
2905 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2906 spares_needed,
2907 spares_needed == 1 ? "" : "s",
2908 info->array.spare_disks + added_disks);
2909 goto release;
2910 }
2911
2912 if (reshape.level != array.level) {
2913 int err = impose_level(fd, reshape.level, devname, verbose);
2914 if (err)
2915 goto release;
2916 info->new_layout = UnSet; /* after level change,
2917 * layout is meaningless */
2918 orig_level = array.level;
2919 sysfs_freeze_array(info);
2920
2921 if (reshape.level > 0 && st->ss->external) {
2922 /* make sure mdmon is aware of the new level */
2923 if (mdmon_running(container))
2924 flush_mdmon(container);
2925
2926 if (!mdmon_running(container))
2927 start_mdmon(container);
2928 ping_monitor(container);
2929 if (mdmon_running(container) &&
2930 st->update_tail == NULL)
2931 st->update_tail = &st->updates;
2932 }
2933 }
2934 /* ->reshape_super might have chosen some spares from the
2935 * container that it wants to be part of the new array.
2936 * We can collect them with ->container_content and give
2937 * them to the kernel.
2938 */
2939 if (st->ss->reshape_super && st->ss->container_content) {
2940 char *subarray = strchr(info->text_version+1, '/')+1;
2941 struct mdinfo *info2 =
2942 st->ss->container_content(st, subarray);
2943 struct mdinfo *d;
2944
2945 if (info2) {
2946 sysfs_init(info2, fd, st->devnm);
2947 /* When increasing number of devices, we need to set
2948 * new raid_disks before adding these, or they might
2949 * be rejected.
2950 */
2951 if (reshape.backup_blocks &&
2952 reshape.after.data_disks > reshape.before.data_disks)
2953 subarray_set_num(container, info2, "raid_disks",
2954 reshape.after.data_disks +
2955 reshape.parity);
2956 for (d = info2->devs; d; d = d->next) {
2957 if (d->disk.state == 0 &&
2958 d->disk.raid_disk >= 0) {
2959 /* This is a spare that wants to
2960 * be part of the array.
2961 */
2962 add_disk(fd, st, info2, d);
2963 }
2964 }
2965 sysfs_free(info2);
2966 }
2967 }
2968 /* We might have been given some devices to add to the
2969 * array. Now that the array has been changed to the right
2970 * level and frozen, we can safely add them.
2971 */
2972 if (devlist)
2973 Manage_subdevs(devname, fd, devlist, verbose,
2974 0,NULL, 0);
2975
2976 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
2977 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
2978 if (reshape.backup_blocks == 0) {
2979 /* No restriping needed, but we might need to impose
2980 * some more changes: layout, raid_disks, chunk_size
2981 */
2982 /* read current array info */
2983 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2984 dprintf("Cannot get array information.\n");
2985 goto release;
2986 }
2987 /* compare current array info with new values and if
2988 * it is different update them to new */
2989 if (info->new_layout != UnSet &&
2990 info->new_layout != array.layout) {
2991 array.layout = info->new_layout;
2992 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2993 pr_err("failed to set new layout\n");
2994 goto release;
2995 } else if (verbose >= 0)
2996 printf("layout for %s set to %d\n",
2997 devname, array.layout);
2998 }
2999 if (info->delta_disks != UnSet &&
3000 info->delta_disks != 0 &&
3001 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3002 array.raid_disks += info->delta_disks;
3003 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3004 pr_err("failed to set raid disks\n");
3005 goto release;
3006 } else if (verbose >= 0) {
3007 printf("raid_disks for %s set to %d\n",
3008 devname, array.raid_disks);
3009 }
3010 }
3011 if (info->new_chunk != 0 &&
3012 info->new_chunk != array.chunk_size) {
3013 if (sysfs_set_num(info, NULL,
3014 "chunk_size", info->new_chunk) != 0) {
3015 pr_err("failed to set chunk size\n");
3016 goto release;
3017 } else if (verbose >= 0)
3018 printf("chunk size for %s set to %d\n",
3019 devname, array.chunk_size);
3020 }
3021 unfreeze(st);
3022 return 0;
3023 }
3024
3025 /*
3026 * There are three possibilities.
3027 * 1/ The array will shrink.
3028 * We need to ensure the reshape will pause before reaching
3029 * the 'critical section'. We also need to fork and wait for
3030 * that to happen. When it does we
3031 * suspend/backup/complete/unfreeze
3032 *
3033 * 2/ The array will not change size.
3034 * This requires that we keep a backup of a sliding window
3035 * so that we can restore data after a crash. So we need
3036 * to fork and monitor progress.
3037 * In future we will allow the data_offset to change, so
3038 * a sliding backup becomes unnecessary.
3039 *
3040 * 3/ The array will grow. This is relatively easy.
3041 * However the kernel's restripe routines will cheerfully
3042 * overwrite some early data before it is safe. So we
3043 * need to make a backup of the early parts of the array
3044 * and be ready to restore it if rebuild aborts very early.
3045 * For externally managed metadata, we still need a forked
3046 * child to monitor the reshape and suspend IO over the region
3047 * that is being reshaped.
3048 *
3049 * We backup data by writing it to one spare, or to a
3050 * file which was given on command line.
3051 *
3052 * In each case, we first make sure that storage is available
3053 * for the required backup.
3054 * Then we:
3055 * - request the shape change.
3056 * - fork to handle backup etc.
3057 */
3058 /* Check that we can hold all the data */
3059 get_dev_size(fd, NULL, &array_size);
3060 if (reshape.new_size < (array_size/512)) {
3061 pr_err("this change will reduce the size of the array.\n"
3062 " use --grow --array-size first to truncate array.\n"
3063 " e.g. mdadm --grow %s --array-size %llu\n",
3064 devname, reshape.new_size/2);
3065 goto release;
3066 }
3067
3068 if (array.level == 10) {
3069 /* Reshaping RAID10 does not require any data backup by
3070 * user-space. Instead it requires that the data_offset
3071 * is changed to avoid the need for backup.
3072 * So this is handled very separately
3073 */
3074 if (restart)
3075 /* Nothing to do. */
3076 return 0;
3077 return raid10_reshape(container, fd, devname, st, info,
3078 &reshape, data_offset,
3079 force, verbose);
3080 }
3081 sra = sysfs_read(fd, NULL,
3082 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3083 GET_CACHE);
3084 if (!sra) {
3085 pr_err("%s: Cannot get array details from sysfs\n",
3086 devname);
3087 goto release;
3088 }
3089
3090 if (!backup_file)
3091 switch(set_new_data_offset(sra, st, devname,
3092 reshape.after.data_disks - reshape.before.data_disks,
3093 data_offset,
3094 reshape.min_offset_change, 1)) {
3095 case -1:
3096 goto release;
3097 case 0:
3098 /* Updated data_offset, so it's easy now */
3099 update_cache_size(container, sra, info,
3100 min(reshape.before.data_disks,
3101 reshape.after.data_disks),
3102 reshape.backup_blocks);
3103
3104 /* Right, everything seems fine. Let's kick things off.
3105 */
3106 sync_metadata(st);
3107
3108 if (impose_reshape(sra, info, st, fd, restart,
3109 devname, container, &reshape) < 0)
3110 goto release;
3111 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3112 struct mdinfo *sd;
3113 if (errno != EINVAL) {
3114 pr_err("Failed to initiate reshape!\n");
3115 goto release;
3116 }
3117 /* revert data_offset and try the old way */
3118 for (sd = sra->devs; sd; sd = sd->next) {
3119 sysfs_set_num(sra, sd, "new_offset",
3120 sd->data_offset);
3121 sysfs_set_str(sra, NULL, "reshape_direction",
3122 "forwards");
3123 }
3124 break;
3125 }
3126 if (info->new_level == reshape.level)
3127 return 0;
3128 /* need to adjust level when reshape completes */
3129 switch(fork()) {
3130 case -1: /* ignore error, but don't wait */
3131 return 0;
3132 default: /* parent */
3133 return 0;
3134 case 0:
3135 map_fork();
3136 break;
3137 }
3138 close(fd);
3139 wait_reshape(sra);
3140 fd = open_dev(sra->sys_name);
3141 if (fd >= 0)
3142 impose_level(fd, info->new_level, devname, verbose);
3143 return 0;
3144 case 1: /* Couldn't set data_offset, try the old way */
3145 if (data_offset != INVALID_SECTORS) {
3146 pr_err("Cannot update data_offset on this array\n");
3147 goto release;
3148 }
3149 break;
3150 }
3151
3152 started:
3153 /* Decide how many blocks (sectors) for a reshape
3154 * unit. The number we have so far is just a minimum
3155 */
3156 blocks = reshape.backup_blocks;
3157 if (reshape.before.data_disks ==
3158 reshape.after.data_disks) {
3159 /* Make 'blocks' bigger for better throughput, but
3160 * not so big that we reject it below.
3161 * Try for 16 megabytes
3162 */
3163 while (blocks * 32 < sra->component_size &&
3164 blocks < 16*1024*2)
3165 blocks *= 2;
3166 } else
3167 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3168
3169 if (blocks >= sra->component_size/2) {
3170 pr_err("%s: Something wrong - reshape aborted\n",
3171 devname);
3172 goto release;
3173 }
3174
3175 /* Now we need to open all these devices so we can read/write.
3176 */
3177 nrdisks = max(reshape.before.data_disks,
3178 reshape.after.data_disks) + reshape.parity
3179 + sra->array.spare_disks;
3180 fdlist = xcalloc((1+nrdisks), sizeof(int));
3181 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3182
3183 odisks = reshape.before.data_disks + reshape.parity;
3184 d = reshape_prepare_fdlist(devname, sra, odisks,
3185 nrdisks, blocks, backup_file,
3186 fdlist, offsets);
3187 if (d < 0) {
3188 goto release;
3189 }
3190 if ((st->ss->manage_reshape == NULL) ||
3191 (st->ss->recover_backup == NULL)) {
3192 if (backup_file == NULL) {
3193 if (reshape.after.data_disks <=
3194 reshape.before.data_disks) {
3195 pr_err("%s: Cannot grow - need backup-file\n",
3196 devname);
3197 pr_err(" Please provide one with \"--backup=...\"\n");
3198 goto release;
3199 } else if (sra->array.spare_disks == 0) {
3200 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3201 goto release;
3202 }
3203 } else {
3204 if (!reshape_open_backup_file(backup_file, fd, devname,
3205 (signed)blocks,
3206 fdlist+d, offsets+d,
3207 sra->sys_name,
3208 restart)) {
3209 goto release;
3210 }
3211 d++;
3212 }
3213 }
3214
3215 update_cache_size(container, sra, info,
3216 min(reshape.before.data_disks, reshape.after.data_disks),
3217 blocks);
3218
3219 /* Right, everything seems fine. Let's kick things off.
3220 * If only changing raid_disks, use ioctl, else use
3221 * sysfs.
3222 */
3223 sync_metadata(st);
3224
3225 if (impose_reshape(sra, info, st, fd, restart,
3226 devname, container, &reshape) < 0)
3227 goto release;
3228
3229 err = start_reshape(sra, restart, reshape.before.data_disks,
3230 reshape.after.data_disks);
3231 if (err) {
3232 pr_err("Cannot %s reshape for %s\n",
3233 restart ? "continue" : "start",
3234 devname);
3235 goto release;
3236 }
3237 if (restart)
3238 sysfs_set_str(sra, NULL, "array_state", "active");
3239 if (freeze_reshape) {
3240 free(fdlist);
3241 free(offsets);
3242 sysfs_free(sra);
3243 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3244 sra->reshape_progress);
3245 return 1;
3246 }
3247
3248 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3249 if (continue_via_systemd(container ?: sra->sys_name)) {
3250 free(fdlist);
3251 free(offsets);
3252 sysfs_free(sra);
3253 return 0;
3254 }
3255
3256 /* Now we just need to kick off the reshape and watch, while
3257 * handling backups of the data...
3258 * This is all done by a forked background process.
3259 */
3260 switch(forked ? 0 : fork()) {
3261 case -1:
3262 pr_err("Cannot run child to monitor reshape: %s\n",
3263 strerror(errno));
3264 abort_reshape(sra);
3265 goto release;
3266 default:
3267 free(fdlist);
3268 free(offsets);
3269 sysfs_free(sra);
3270 return 0;
3271 case 0:
3272 map_fork();
3273 break;
3274 }
3275
3276 /* If another array on the same devices is busy, the
3277 * reshape will wait for them. This would mean that
3278 * the first section that we suspend will stay suspended
3279 * for a long time. So check on that possibility
3280 * by looking for "DELAYED" in /proc/mdstat, and if found,
3281 * wait a while
3282 */
3283 do {
3284 struct mdstat_ent *mds, *m;
3285 delayed = 0;
3286 mds = mdstat_read(1, 0);
3287 for (m = mds; m; m = m->next)
3288 if (strcmp(m->devnm, sra->sys_name) == 0) {
3289 if (m->resync &&
3290 m->percent == RESYNC_DELAYED)
3291 delayed = 1;
3292 if (m->resync == 0)
3293 /* Haven't started the reshape thread
3294 * yet, wait a bit
3295 */
3296 delayed = 2;
3297 break;
3298 }
3299 free_mdstat(mds);
3300 if (delayed == 1 && get_linux_version() < 3007000) {
3301 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3302 " You might experience problems until other reshapes complete.\n");
3303 delayed = 0;
3304 }
3305 if (delayed)
3306 mdstat_wait(30 - (delayed-1) * 25);
3307 } while (delayed);
3308 mdstat_close();
3309 close(fd);
3310 if (check_env("MDADM_GROW_VERIFY"))
3311 fd = open(devname, O_RDONLY | O_DIRECT);
3312 else
3313 fd = -1;
3314 mlockall(MCL_FUTURE);
3315
3316 signal(SIGTERM, catch_term);
3317
3318 if (st->ss->external) {
3319 /* metadata handler takes it from here */
3320 done = st->ss->manage_reshape(
3321 fd, sra, &reshape, st, blocks,
3322 fdlist, offsets,
3323 d - odisks, fdlist+odisks,
3324 offsets+odisks);
3325 } else
3326 done = child_monitor(
3327 fd, sra, &reshape, st, blocks,
3328 fdlist, offsets,
3329 d - odisks, fdlist+odisks,
3330 offsets+odisks);
3331
3332 free(fdlist);
3333 free(offsets);
3334
3335 if (backup_file && done) {
3336 char *bul;
3337 bul = make_backup(sra->sys_name);
3338 if (bul) {
3339 char buf[1024];
3340 int l = readlink(bul, buf, sizeof(buf) - 1);
3341 if (l > 0) {
3342 buf[l]=0;
3343 unlink(buf);
3344 }
3345 unlink(bul);
3346 free(bul);
3347 }
3348 unlink(backup_file);
3349 }
3350 if (!done) {
3351 abort_reshape(sra);
3352 goto out;
3353 }
3354
3355 if (!st->ss->external &&
3356 !(reshape.before.data_disks != reshape.after.data_disks
3357 && info->custom_array_size) &&
3358 info->new_level == reshape.level &&
3359 !forked) {
3360 /* no need to wait for the reshape to finish as
3361 * there is nothing more to do.
3362 */
3363 sysfs_free(sra);
3364 exit(0);
3365 }
3366 wait_reshape(sra);
3367
3368 if (st->ss->external) {
3369 /* Re-load the metadata as much could have changed */
3370 int cfd = open_dev(st->container_devnm);
3371 if (cfd >= 0) {
3372 flush_mdmon(container);
3373 st->ss->free_super(st);
3374 st->ss->load_container(st, cfd, container);
3375 close(cfd);
3376 }
3377 }
3378
3379 /* set new array size if required customer_array_size is used
3380 * by this metadata.
3381 */
3382 if (reshape.before.data_disks !=
3383 reshape.after.data_disks &&
3384 info->custom_array_size)
3385 set_array_size(st, info, info->text_version);
3386
3387 if (info->new_level != reshape.level) {
3388 if (fd < 0)
3389 fd = open(devname, O_RDONLY);
3390 impose_level(fd, info->new_level, devname, verbose);
3391 close(fd);
3392 if (info->new_level == 0)
3393 st->update_tail = NULL;
3394 }
3395 out:
3396 sysfs_free(sra);
3397 if (forked)
3398 return 0;
3399 unfreeze(st);
3400 exit(0);
3401
3402 release:
3403 free(fdlist);
3404 free(offsets);
3405 if (orig_level != UnSet && sra) {
3406 c = map_num(pers, orig_level);
3407 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3408 pr_err("aborting level change\n");
3409 }
3410 sysfs_free(sra);
3411 if (!forked)
3412 unfreeze(st);
3413 return 1;
3414 }
3415
3416 /* mdfd handle is passed to be closed in child process (after fork).
3417 */
3418 int reshape_container(char *container, char *devname,
3419 int mdfd,
3420 struct supertype *st,
3421 struct mdinfo *info,
3422 int force,
3423 char *backup_file, int verbose,
3424 int forked, int restart, int freeze_reshape)
3425 {
3426 struct mdinfo *cc = NULL;
3427 int rv = restart;
3428 char last_devnm[32] = "";
3429
3430 /* component_size is not meaningful for a container,
3431 * so pass '0' meaning 'no change'
3432 */
3433 if (!restart &&
3434 reshape_super(st, 0, info->new_level,
3435 info->new_layout, info->new_chunk,
3436 info->array.raid_disks, info->delta_disks,
3437 backup_file, devname, APPLY_METADATA_CHANGES,
3438 verbose)) {
3439 unfreeze(st);
3440 return 1;
3441 }
3442
3443 sync_metadata(st);
3444
3445 /* ping monitor to be sure that update is on disk
3446 */
3447 ping_monitor(container);
3448
3449 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3450 if (continue_via_systemd(container))
3451 return 0;
3452
3453 switch (forked ? 0 : fork()) {
3454 case -1: /* error */
3455 perror("Cannot fork to complete reshape\n");
3456 unfreeze(st);
3457 return 1;
3458 default: /* parent */
3459 if (!freeze_reshape)
3460 printf("%s: multi-array reshape continues in background\n", Name);
3461 return 0;
3462 case 0: /* child */
3463 map_fork();
3464 break;
3465 }
3466
3467 /* close unused handle in child process
3468 */
3469 if (mdfd > -1)
3470 close(mdfd);
3471
3472 while(1) {
3473 /* For each member array with reshape_active,
3474 * we need to perform the reshape.
3475 * We pick the first array that needs reshaping and
3476 * reshape it. reshape_array() will re-read the metadata
3477 * so the next time through a different array should be
3478 * ready for reshape.
3479 * It is possible that the 'different' array will not
3480 * be assembled yet. In that case we simple exit.
3481 * When it is assembled, the mdadm which assembles it
3482 * will take over the reshape.
3483 */
3484 struct mdinfo *content;
3485 int fd;
3486 struct mdstat_ent *mdstat;
3487 char *adev;
3488 int devid;
3489
3490 sysfs_free(cc);
3491
3492 cc = st->ss->container_content(st, NULL);
3493
3494 for (content = cc; content ; content = content->next) {
3495 char *subarray;
3496 if (!content->reshape_active)
3497 continue;
3498
3499 subarray = strchr(content->text_version+1, '/')+1;
3500 mdstat = mdstat_by_subdev(subarray, container);
3501 if (!mdstat)
3502 continue;
3503 if (mdstat->active == 0) {
3504 pr_err("Skipping inactive array %s.\n",
3505 mdstat->devnm);
3506 free_mdstat(mdstat);
3507 mdstat = NULL;
3508 continue;
3509 }
3510 break;
3511 }
3512 if (!content)
3513 break;
3514
3515 devid = devnm2devid(mdstat->devnm);
3516 adev = map_dev(major(devid), minor(devid), 0);
3517 if (!adev)
3518 adev = content->text_version;
3519
3520 fd = open_dev(mdstat->devnm);
3521 if (fd < 0) {
3522 pr_err("Device %s cannot be opened for reshape.\n", adev);
3523 break;
3524 }
3525
3526 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3527 /* Do not allow for multiple reshape_array() calls for
3528 * the same array.
3529 * It can happen when reshape_array() returns without
3530 * error, when reshape is not finished (wrong reshape
3531 * starting/continuation conditions). Mdmon doesn't
3532 * switch to next array in container and reentry
3533 * conditions for the same array occur.
3534 * This is possibly interim until the behaviour of
3535 * reshape_array is resolved().
3536 */
3537 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3538 close(fd);
3539 break;
3540 }
3541 strcpy(last_devnm, mdstat->devnm);
3542
3543 sysfs_init(content, fd, mdstat->devnm);
3544
3545 if (mdmon_running(container))
3546 flush_mdmon(container);
3547
3548 rv = reshape_array(container, fd, adev, st,
3549 content, force, NULL, INVALID_SECTORS,
3550 backup_file, verbose, 1, restart,
3551 freeze_reshape);
3552 close(fd);
3553
3554 if (freeze_reshape) {
3555 sysfs_free(cc);
3556 exit(0);
3557 }
3558
3559 restart = 0;
3560 if (rv)
3561 break;
3562
3563 if (mdmon_running(container))
3564 flush_mdmon(container);
3565 }
3566 if (!rv)
3567 unfreeze(st);
3568 sysfs_free(cc);
3569 exit(0);
3570 }
3571
3572 /*
3573 * We run a child process in the background which performs the following
3574 * steps:
3575 * - wait for resync to reach a certain point
3576 * - suspend io to the following section
3577 * - backup that section
3578 * - allow resync to proceed further
3579 * - resume io
3580 * - discard the backup.
3581 *
3582 * When are combined in slightly different ways in the three cases.
3583 * Grow:
3584 * - suspend/backup/allow/wait/resume/discard
3585 * Shrink:
3586 * - allow/wait/suspend/backup/allow/wait/resume/discard
3587 * same-size:
3588 * - wait/resume/discard/suspend/backup/allow
3589 *
3590 * suspend/backup/allow always come together
3591 * wait/resume/discard do too.
3592 * For the same-size case we have two backups to improve flow.
3593 *
3594 */
3595
3596 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3597 unsigned long long backup_point,
3598 unsigned long long wait_point,
3599 unsigned long long *suspend_point,
3600 unsigned long long *reshape_completed, int *frozen)
3601 {
3602 /* This function is called repeatedly by the reshape manager.
3603 * It determines how much progress can safely be made and allows
3604 * that progress.
3605 * - 'info' identifies the array and particularly records in
3606 * ->reshape_progress the metadata's knowledge of progress
3607 * This is a sector offset from the start of the array
3608 * of the next array block to be relocated. This number
3609 * may increase from 0 or decrease from array_size, depending
3610 * on the type of reshape that is happening.
3611 * Note that in contrast, 'sync_completed' is a block count of the
3612 * reshape so far. It gives the distance between the start point
3613 * (head or tail of device) and the next place that data will be
3614 * written. It always increases.
3615 * - 'reshape' is the structure created by analyse_change
3616 * - 'backup_point' shows how much the metadata manager has backed-up
3617 * data. For reshapes with increasing progress, it is the next address
3618 * to be backed up, previous addresses have been backed-up. For
3619 * decreasing progress, it is the earliest address that has been
3620 * backed up - later address are also backed up.
3621 * So addresses between reshape_progress and backup_point are
3622 * backed up providing those are in the 'correct' order.
3623 * - 'wait_point' is an array address. When reshape_completed
3624 * passes this point, progress_reshape should return. It might
3625 * return earlier if it determines that ->reshape_progress needs
3626 * to be updated or further backup is needed.
3627 * - suspend_point is maintained by progress_reshape and the caller
3628 * should not touch it except to initialise to zero.
3629 * It is an array address and it only increases in 2.6.37 and earlier.
3630 * This makes it difficult to handle reducing reshapes with
3631 * external metadata.
3632 * However: it is similar to backup_point in that it records the
3633 * other end of a suspended region from reshape_progress.
3634 * it is moved to extend the region that is safe to backup and/or
3635 * reshape
3636 * - reshape_completed is read from sysfs and returned. The caller
3637 * should copy this into ->reshape_progress when it has reason to
3638 * believe that the metadata knows this, and any backup outside this
3639 * has been erased.
3640 *
3641 * Return value is:
3642 * 1 if more data from backup_point - but only as far as suspend_point,
3643 * should be backed up
3644 * 0 if things are progressing smoothly
3645 * -1 if the reshape is finished because it is all done,
3646 * -2 if the reshape is finished due to an error.
3647 */
3648
3649 int advancing = (reshape->after.data_disks
3650 >= reshape->before.data_disks);
3651 unsigned long long need_backup; /* All data between start of array and
3652 * here will at some point need to
3653 * be backed up.
3654 */
3655 unsigned long long read_offset, write_offset;
3656 unsigned long long write_range;
3657 unsigned long long max_progress, target, completed;
3658 unsigned long long array_size = (info->component_size
3659 * reshape->before.data_disks);
3660 int fd;
3661 char buf[20];
3662
3663 /* First, we unsuspend any region that is now known to be safe.
3664 * If suspend_point is on the 'wrong' side of reshape_progress, then
3665 * we don't have or need suspension at the moment. This is true for
3666 * native metadata when we don't need to back-up.
3667 */
3668 if (advancing) {
3669 if (info->reshape_progress <= *suspend_point)
3670 sysfs_set_num(info, NULL, "suspend_lo",
3671 info->reshape_progress);
3672 } else {
3673 /* Note: this won't work in 2.6.37 and before.
3674 * Something somewhere should make sure we don't need it!
3675 */
3676 if (info->reshape_progress >= *suspend_point)
3677 sysfs_set_num(info, NULL, "suspend_hi",
3678 info->reshape_progress);
3679 }
3680
3681 /* Now work out how far it is safe to progress.
3682 * If the read_offset for ->reshape_progress is less than
3683 * 'blocks' beyond the write_offset, we can only progress as far
3684 * as a backup.
3685 * Otherwise we can progress until the write_offset for the new location
3686 * reaches (within 'blocks' of) the read_offset at the current location.
3687 * However that region must be suspended unless we are using native
3688 * metadata.
3689 * If we need to suspend more, we limit it to 128M per device, which is
3690 * rather arbitrary and should be some time-based calculation.
3691 */
3692 read_offset = info->reshape_progress / reshape->before.data_disks;
3693 write_offset = info->reshape_progress / reshape->after.data_disks;
3694 write_range = info->new_chunk/512;
3695 if (reshape->before.data_disks == reshape->after.data_disks)
3696 need_backup = array_size;
3697 else
3698 need_backup = reshape->backup_blocks;
3699 if (advancing) {
3700 if (read_offset < write_offset + write_range)
3701 max_progress = backup_point;
3702 else
3703 max_progress =
3704 read_offset *
3705 reshape->after.data_disks;
3706 } else {
3707 if (read_offset > write_offset - write_range)
3708 /* Can only progress as far as has been backed up,
3709 * which must be suspended */
3710 max_progress = backup_point;
3711 else if (info->reshape_progress <= need_backup)
3712 max_progress = backup_point;
3713 else {
3714 if (info->array.major_version >= 0)
3715 /* Can progress until backup is needed */
3716 max_progress = need_backup;
3717 else {
3718 /* Can progress until metadata update is required */
3719 max_progress =
3720 read_offset *
3721 reshape->after.data_disks;
3722 /* but data must be suspended */
3723 if (max_progress < *suspend_point)
3724 max_progress = *suspend_point;
3725 }
3726 }
3727 }
3728
3729 /* We know it is safe to progress to 'max_progress' providing
3730 * it is suspended or we are using native metadata.
3731 * Consider extending suspend_point 128M per device if it
3732 * is less than 64M per device beyond reshape_progress.
3733 * But always do a multiple of 'blocks'
3734 * FIXME this is too big - it takes to long to complete
3735 * this much.
3736 */
3737 target = 64*1024*2 * min(reshape->before.data_disks,
3738 reshape->after.data_disks);
3739 target /= reshape->backup_blocks;
3740 if (target < 2)
3741 target = 2;
3742 target *= reshape->backup_blocks;
3743
3744 /* For externally managed metadata we always need to suspend IO to
3745 * the area being reshaped so we regularly push suspend_point forward.
3746 * For native metadata we only need the suspend if we are going to do
3747 * a backup.
3748 */
3749 if (advancing) {
3750 if ((need_backup > info->reshape_progress
3751 || info->array.major_version < 0) &&
3752 *suspend_point < info->reshape_progress + target) {
3753 if (need_backup < *suspend_point + 2 * target)
3754 *suspend_point = need_backup;
3755 else if (*suspend_point + 2 * target < array_size)
3756 *suspend_point += 2 * target;
3757 else
3758 *suspend_point = array_size;
3759 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3760 if (max_progress > *suspend_point)
3761 max_progress = *suspend_point;
3762 }
3763 } else {
3764 if (info->array.major_version >= 0) {
3765 /* Only need to suspend when about to backup */
3766 if (info->reshape_progress < need_backup * 2 &&
3767 *suspend_point > 0) {
3768 *suspend_point = 0;
3769 sysfs_set_num(info, NULL, "suspend_lo", 0);
3770 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3771 }
3772 } else {
3773 /* Need to suspend continually */
3774 if (info->reshape_progress < *suspend_point)
3775 *suspend_point = info->reshape_progress;
3776 if (*suspend_point + target < info->reshape_progress)
3777 /* No need to move suspend region yet */;
3778 else {
3779 if (*suspend_point >= 2 * target)
3780 *suspend_point -= 2 * target;
3781 else
3782 *suspend_point = 0;
3783 sysfs_set_num(info, NULL, "suspend_lo",
3784 *suspend_point);
3785 }
3786 if (max_progress < *suspend_point)
3787 max_progress = *suspend_point;
3788 }
3789 }
3790
3791 /* now set sync_max to allow that progress. sync_max, like
3792 * sync_completed is a count of sectors written per device, so
3793 * we find the difference between max_progress and the start point,
3794 * and divide that by after.data_disks to get a sync_max
3795 * number.
3796 * At the same time we convert wait_point to a similar number
3797 * for comparing against sync_completed.
3798 */
3799 /* scale down max_progress to per_disk */
3800 max_progress /= reshape->after.data_disks;
3801 /* Round to chunk size as some kernels give an erroneously high number */
3802 max_progress /= info->new_chunk/512;
3803 max_progress *= info->new_chunk/512;
3804 /* And round to old chunk size as the kernel wants that */
3805 max_progress /= info->array.chunk_size/512;
3806 max_progress *= info->array.chunk_size/512;
3807 /* Limit progress to the whole device */
3808 if (max_progress > info->component_size)
3809 max_progress = info->component_size;
3810 wait_point /= reshape->after.data_disks;
3811 if (!advancing) {
3812 /* switch from 'device offset' to 'processed block count' */
3813 max_progress = info->component_size - max_progress;
3814 wait_point = info->component_size - wait_point;
3815 }
3816
3817 if (!*frozen)
3818 sysfs_set_num(info, NULL, "sync_max", max_progress);
3819
3820 /* Now wait. If we have already reached the point that we were
3821 * asked to wait to, don't wait at all, else wait for any change.
3822 * We need to select on 'sync_completed' as that is the place that
3823 * notifications happen, but we are really interested in
3824 * 'reshape_position'
3825 */
3826 fd = sysfs_get_fd(info, NULL, "sync_completed");
3827 if (fd < 0)
3828 goto check_progress;
3829
3830 if (sysfs_fd_get_ll(fd, &completed) < 0)
3831 goto check_progress;
3832
3833 while (completed < max_progress && completed < wait_point) {
3834 /* Check that sync_action is still 'reshape' to avoid
3835 * waiting forever on a dead array
3836 */
3837 char action[20];
3838 if (sysfs_get_str(info, NULL, "sync_action",
3839 action, 20) <= 0 ||
3840 strncmp(action, "reshape", 7) != 0)
3841 break;
3842 /* Some kernels reset 'sync_completed' to zero
3843 * before setting 'sync_action' to 'idle'.
3844 * So we need these extra tests.
3845 */
3846 if (completed == 0 && advancing
3847 && info->reshape_progress > 0)
3848 break;
3849 if (completed == 0 && !advancing
3850 && info->reshape_progress < (info->component_size
3851 * reshape->after.data_disks))
3852 break;
3853 sysfs_wait(fd, NULL);
3854 if (sysfs_fd_get_ll(fd, &completed) < 0)
3855 goto check_progress;
3856 }
3857 /* Some kernels reset 'sync_completed' to zero,
3858 * we need to have real point we are in md.
3859 * But only if array is actually still reshaping,
3860 * not stopped.
3861 */
3862 if (completed == 0) {
3863 char action[20];
3864 if (sysfs_get_str(info, NULL, "sync_action",
3865 action, 20) > 0 &&
3866 strncmp(action, "reshape", 7) == 0)
3867 completed = max_progress;
3868 }
3869
3870 /* some kernels can give an incorrectly high 'completed' number */
3871 completed /= (info->new_chunk/512);
3872 completed *= (info->new_chunk/512);
3873 /* Convert 'completed' back in to a 'progress' number */
3874 completed *= reshape->after.data_disks;
3875 if (!advancing) {
3876 completed = info->component_size * reshape->after.data_disks
3877 - completed;
3878 }
3879 *reshape_completed = completed;
3880
3881 close(fd);
3882
3883 /* We return the need_backup flag. Caller will decide
3884 * how much - a multiple of ->backup_blocks up to *suspend_point
3885 */
3886 if (advancing)
3887 return need_backup > info->reshape_progress;
3888 else
3889 return need_backup >= info->reshape_progress;
3890
3891 check_progress:
3892 /* if we couldn't read a number from sync_completed, then
3893 * either the reshape did complete, or it aborted.
3894 * We can tell which by checking for 'none' in reshape_position.
3895 * If it did abort, then it might immediately restart if it
3896 * it was just a device failure that leaves us degraded but
3897 * functioning.
3898 */
3899 strcpy(buf, "hi");
3900 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3901 || strncmp(buf, "none", 4) != 0) {
3902 /* The abort might only be temporary. Wait up to 10
3903 * seconds for fd to contain a valid number again.
3904 */
3905 int wait = 10000;
3906 int rv = -2;
3907 unsigned long long new_sync_max;
3908 while (fd >= 0 && rv < 0 && wait > 0) {
3909 if (sysfs_wait(fd, &wait) != 1)
3910 break;
3911 switch (sysfs_fd_get_ll(fd, &completed)) {
3912 case 0:
3913 /* all good again */
3914 rv = 1;
3915 /* If "sync_max" is no longer max_progress
3916 * we need to freeze things
3917 */
3918 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3919 *frozen = (new_sync_max != max_progress);
3920 break;
3921 case -2: /* read error - abort */
3922 wait = 0;
3923 break;
3924 }
3925 }
3926 if (fd >= 0)
3927 close(fd);
3928 return rv; /* abort */
3929 } else {
3930 /* Maybe racing with array shutdown - check state */
3931 if (fd >= 0)
3932 close(fd);
3933 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3934 || strncmp(buf, "inactive", 8) == 0
3935 || strncmp(buf, "clear",5) == 0)
3936 return -2; /* abort */
3937 return -1; /* complete */
3938 }
3939 }
3940
3941 /* FIXME return status is never checked */
3942 static int grow_backup(struct mdinfo *sra,
3943 unsigned long long offset, /* per device */
3944 unsigned long stripes, /* per device, in old chunks */
3945 int *sources, unsigned long long *offsets,
3946 int disks, int chunk, int level, int layout,
3947 int dests, int *destfd, unsigned long long *destoffsets,
3948 int part, int *degraded,
3949 char *buf)
3950 {
3951 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3952 * to storage 'destfd' (offset 'destoffsets'), after first
3953 * suspending IO. Then allow resync to continue
3954 * over the suspended section.
3955 * Use part 'part' of the backup-super-block.
3956 */
3957 int odata = disks;
3958 int rv = 0;
3959 int i;
3960 unsigned long long ll;
3961 int new_degraded;
3962 //printf("offset %llu\n", offset);
3963 if (level >= 4)
3964 odata--;
3965 if (level == 6)
3966 odata--;
3967
3968 /* Check that array hasn't become degraded, else we might backup the wrong data */
3969 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3970 return -1; /* FIXME this error is ignored */
3971 new_degraded = (int)ll;
3972 if (new_degraded != *degraded) {
3973 /* check each device to ensure it is still working */
3974 struct mdinfo *sd;
3975 for (sd = sra->devs ; sd ; sd = sd->next) {
3976 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3977 continue;
3978 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3979 char sbuf[20];
3980 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3981 strstr(sbuf, "faulty") ||
3982 strstr(sbuf, "in_sync") == NULL) {
3983 /* this device is dead */
3984 sd->disk.state = (1<<MD_DISK_FAULTY);
3985 if (sd->disk.raid_disk >= 0 &&
3986 sources[sd->disk.raid_disk] >= 0) {
3987 close(sources[sd->disk.raid_disk]);
3988 sources[sd->disk.raid_disk] = -1;
3989 }
3990 }
3991 }
3992 }
3993 *degraded = new_degraded;
3994 }
3995 if (part) {
3996 bsb.arraystart2 = __cpu_to_le64(offset * odata);
3997 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
3998 } else {
3999 bsb.arraystart = __cpu_to_le64(offset * odata);
4000 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4001 }
4002 if (part)
4003 bsb.magic[15] = '2';
4004 for (i = 0; i < dests; i++)
4005 if (part)
4006 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4007 else
4008 lseek64(destfd[i], destoffsets[i], 0);
4009
4010 rv = save_stripes(sources, offsets,
4011 disks, chunk, level, layout,
4012 dests, destfd,
4013 offset*512*odata, stripes * chunk * odata,
4014 buf);
4015
4016 if (rv)
4017 return rv;
4018 bsb.mtime = __cpu_to_le64(time(0));
4019 for (i = 0; i < dests; i++) {
4020 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4021
4022 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4023 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4024 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4025 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4026
4027 rv = -1;
4028 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4029 != destoffsets[i] - 4096)
4030 break;
4031 if (write(destfd[i], &bsb, 512) != 512)
4032 break;
4033 if (destoffsets[i] > 4096) {
4034 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4035 destoffsets[i]+stripes*chunk*odata)
4036 break;
4037 if (write(destfd[i], &bsb, 512) != 512)
4038 break;
4039 }
4040 fsync(destfd[i]);
4041 rv = 0;
4042 }
4043
4044 return rv;
4045 }
4046
4047 /* in 2.6.30, the value reported by sync_completed can be
4048 * less that it should be by one stripe.
4049 * This only happens when reshape hits sync_max and pauses.
4050 * So allow wait_backup to either extent sync_max further
4051 * than strictly necessary, or return before the
4052 * sync has got quite as far as we would really like.
4053 * This is what 'blocks2' is for.
4054 * The various caller give appropriate values so that
4055 * every works.
4056 */
4057 /* FIXME return value is often ignored */
4058 static int forget_backup(int dests, int *destfd,
4059 unsigned long long *destoffsets,
4060 int part)
4061 {
4062 /*
4063 * Erase backup 'part' (which is 0 or 1)
4064 */
4065 int i;
4066 int rv;
4067
4068 if (part) {
4069 bsb.arraystart2 = __cpu_to_le64(0);
4070 bsb.length2 = __cpu_to_le64(0);
4071 } else {
4072 bsb.arraystart = __cpu_to_le64(0);
4073 bsb.length = __cpu_to_le64(0);
4074 }
4075 bsb.mtime = __cpu_to_le64(time(0));
4076 rv = 0;
4077 for (i = 0; i < dests; i++) {
4078 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4079 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4080 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4081 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4082 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4083 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4084 destoffsets[i]-4096)
4085 rv = -1;
4086 if (rv == 0 &&
4087 write(destfd[i], &bsb, 512) != 512)
4088 rv = -1;
4089 fsync(destfd[i]);
4090 }
4091 return rv;
4092 }
4093
4094 static void fail(char *msg)
4095 {
4096 int rv;
4097 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4098 rv |= (write(2, "\n", 1) != 1);
4099 exit(rv ? 1 : 2);
4100 }
4101
4102 static char *abuf, *bbuf;
4103 static unsigned long long abuflen;
4104 static void validate(int afd, int bfd, unsigned long long offset)
4105 {
4106 /* check that the data in the backup against the array.
4107 * This is only used for regression testing and should not
4108 * be used while the array is active
4109 */
4110 if (afd < 0)
4111 return;
4112 lseek64(bfd, offset - 4096, 0);
4113 if (read(bfd, &bsb2, 512) != 512)
4114 fail("cannot read bsb");
4115 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4116 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4117 fail("first csum bad");
4118 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4119 fail("magic is bad");
4120 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4121 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4122 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4123 fail("second csum bad");
4124
4125 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4126 fail("devstart is wrong");
4127
4128 if (bsb2.length) {
4129 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4130
4131 if (abuflen < len) {
4132 free(abuf);
4133 free(bbuf);
4134 abuflen = len;
4135 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4136 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4137 abuflen = 0;
4138 /* just stop validating on mem-alloc failure */
4139 return;
4140 }
4141 }
4142
4143 lseek64(bfd, offset, 0);
4144 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4145 //printf("len %llu\n", len);
4146 fail("read first backup failed");
4147 }
4148 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4149 if ((unsigned long long)read(afd, abuf, len) != len)
4150 fail("read first from array failed");
4151 if (memcmp(bbuf, abuf, len) != 0) {
4152 #if 0
4153 int i;
4154 printf("offset=%llu len=%llu\n",
4155 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4156 for (i=0; i<len; i++)
4157 if (bbuf[i] != abuf[i]) {
4158 printf("first diff byte %d\n", i);
4159 break;
4160 }
4161 #endif
4162 fail("data1 compare failed");
4163 }
4164 }
4165 if (bsb2.length2) {
4166 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4167
4168 if (abuflen < len) {
4169 free(abuf);
4170 free(bbuf);
4171 abuflen = len;
4172 abuf = xmalloc(abuflen);
4173 bbuf = xmalloc(abuflen);
4174 }
4175
4176 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4177 if ((unsigned long long)read(bfd, bbuf, len) != len)
4178 fail("read second backup failed");
4179 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4180 if ((unsigned long long)read(afd, abuf, len) != len)
4181 fail("read second from array failed");
4182 if (memcmp(bbuf, abuf, len) != 0)
4183 fail("data2 compare failed");
4184 }
4185 }
4186
4187 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4188 struct supertype *st, unsigned long blocks,
4189 int *fds, unsigned long long *offsets,
4190 int dests, int *destfd, unsigned long long *destoffsets)
4191 {
4192 /* Monitor a reshape where backup is being performed using
4193 * 'native' mechanism - either to a backup file, or
4194 * to some space in a spare.
4195 */
4196 char *buf;
4197 int degraded = -1;
4198 unsigned long long speed;
4199 unsigned long long suspend_point, array_size;
4200 unsigned long long backup_point, wait_point;
4201 unsigned long long reshape_completed;
4202 int done = 0;
4203 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4204 int part = 0; /* The next part of the backup area to fill. It may already
4205 * be full, so we need to check */
4206 int level = reshape->level;
4207 int layout = reshape->before.layout;
4208 int data = reshape->before.data_disks;
4209 int disks = reshape->before.data_disks + reshape->parity;
4210 int chunk = sra->array.chunk_size;
4211 struct mdinfo *sd;
4212 unsigned long stripes;
4213 int uuid[4];
4214 int frozen = 0;
4215
4216 /* set up the backup-super-block. This requires the
4217 * uuid from the array.
4218 */
4219 /* Find a superblock */
4220 for (sd = sra->devs; sd; sd = sd->next) {
4221 char *dn;
4222 int devfd;
4223 int ok;
4224 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4225 continue;
4226 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4227 devfd = dev_open(dn, O_RDONLY);
4228 if (devfd < 0)
4229 continue;
4230 ok = st->ss->load_super(st, devfd, NULL);
4231 close(devfd);
4232 if (ok == 0)
4233 break;
4234 }
4235 if (!sd) {
4236 pr_err("Cannot find a superblock\n");
4237 return 0;
4238 }
4239
4240 memset(&bsb, 0, 512);
4241 memcpy(bsb.magic, "md_backup_data-1", 16);
4242 st->ss->uuid_from_super(st, uuid);
4243 memcpy(bsb.set_uuid, uuid, 16);
4244 bsb.mtime = __cpu_to_le64(time(0));
4245 bsb.devstart2 = blocks;
4246
4247 stripes = blocks / (sra->array.chunk_size/512) /
4248 reshape->before.data_disks;
4249
4250 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4251 /* Don't start the 'reshape' */
4252 return 0;
4253 if (reshape->before.data_disks == reshape->after.data_disks) {
4254 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4255 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4256 }
4257
4258 if (increasing) {
4259 array_size = sra->component_size * reshape->after.data_disks;
4260 backup_point = sra->reshape_progress;
4261 suspend_point = 0;
4262 } else {
4263 array_size = sra->component_size * reshape->before.data_disks;
4264 backup_point = reshape->backup_blocks;
4265 suspend_point = array_size;
4266 }
4267
4268 while (!done) {
4269 int rv;
4270
4271 /* Want to return as soon the oldest backup slot can
4272 * be released as that allows us to start backing up
4273 * some more, providing suspend_point has been
4274 * advanced, which it should have.
4275 */
4276 if (increasing) {
4277 wait_point = array_size;
4278 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4279 wait_point = (__le64_to_cpu(bsb.arraystart) +
4280 __le64_to_cpu(bsb.length));
4281 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4282 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4283 __le64_to_cpu(bsb.length2));
4284 } else {
4285 wait_point = 0;
4286 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4287 wait_point = __le64_to_cpu(bsb.arraystart);
4288 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4289 wait_point = __le64_to_cpu(bsb.arraystart2);
4290 }
4291
4292 reshape_completed = sra->reshape_progress;
4293 rv = progress_reshape(sra, reshape,
4294 backup_point, wait_point,
4295 &suspend_point, &reshape_completed,
4296 &frozen);
4297 /* external metadata would need to ping_monitor here */
4298 sra->reshape_progress = reshape_completed;
4299
4300 /* Clear any backup region that is before 'here' */
4301 if (increasing) {
4302 if (__le64_to_cpu(bsb.length) > 0 &&
4303 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4304 __le64_to_cpu(bsb.length)))
4305 forget_backup(dests, destfd,
4306 destoffsets, 0);
4307 if (__le64_to_cpu(bsb.length2) > 0 &&
4308 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4309 __le64_to_cpu(bsb.length2)))
4310 forget_backup(dests, destfd,
4311 destoffsets, 1);
4312 } else {
4313 if (__le64_to_cpu(bsb.length) > 0 &&
4314 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4315 forget_backup(dests, destfd,
4316 destoffsets, 0);
4317 if (__le64_to_cpu(bsb.length2) > 0 &&
4318 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4319 forget_backup(dests, destfd,
4320 destoffsets, 1);
4321 }
4322 if (sigterm)
4323 rv = -2;
4324 if (rv < 0) {
4325 if (rv == -1)
4326 done = 1;
4327 break;
4328 }
4329 if (rv == 0 && increasing && !st->ss->external) {
4330 /* No longer need to monitor this reshape */
4331 sysfs_set_str(sra, NULL, "sync_max", "max");
4332 done = 1;
4333 break;
4334 }
4335
4336 while (rv) {
4337 unsigned long long offset;
4338 unsigned long actual_stripes;
4339 /* Need to backup some data.
4340 * If 'part' is not used and the desired
4341 * backup size is suspended, do a backup,
4342 * then consider the next part.
4343 */
4344 /* Check that 'part' is unused */
4345 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4346 break;
4347 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4348 break;
4349
4350 offset = backup_point / data;
4351 actual_stripes = stripes;
4352 if (increasing) {
4353 if (offset + actual_stripes * (chunk/512) >
4354 sra->component_size)
4355 actual_stripes = ((sra->component_size - offset)
4356 / (chunk/512));
4357 if (offset + actual_stripes * (chunk/512) >
4358 suspend_point/data)
4359 break;
4360 } else {
4361 if (offset < actual_stripes * (chunk/512))
4362 actual_stripes = offset / (chunk/512);
4363 offset -= actual_stripes * (chunk/512);
4364 if (offset < suspend_point/data)
4365 break;
4366 }
4367 if (actual_stripes == 0)
4368 break;
4369 grow_backup(sra, offset, actual_stripes,
4370 fds, offsets,
4371 disks, chunk, level, layout,
4372 dests, destfd, destoffsets,
4373 part, &degraded, buf);
4374 validate(afd, destfd[0], destoffsets[0]);
4375 /* record where 'part' is up to */
4376 part = !part;
4377 if (increasing)
4378 backup_point += actual_stripes * (chunk/512) * data;
4379 else
4380 backup_point -= actual_stripes * (chunk/512) * data;
4381 }
4382 }
4383
4384 /* FIXME maybe call progress_reshape one more time instead */
4385 /* remove any remaining suspension */
4386 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4387 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4388 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4389 sysfs_set_num(sra, NULL, "sync_min", 0);
4390
4391 if (reshape->before.data_disks == reshape->after.data_disks)
4392 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4393 free(buf);
4394 return done;
4395 }
4396
4397 /*
4398 * If any spare contains md_back_data-1 which is recent wrt mtime,
4399 * write that data into the array and update the super blocks with
4400 * the new reshape_progress
4401 */
4402 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4403 char *backup_file, int verbose)
4404 {
4405 int i, j;
4406 int old_disks;
4407 unsigned long long *offsets;
4408 unsigned long long nstripe, ostripe;
4409 int ndata, odata;
4410
4411 odata = info->array.raid_disks - info->delta_disks - 1;
4412 if (info->array.level == 6) odata--; /* number of data disks */
4413 ndata = info->array.raid_disks - 1;
4414 if (info->new_level == 6) ndata--;
4415
4416 old_disks = info->array.raid_disks - info->delta_disks;
4417
4418 if (info->delta_disks <= 0)
4419 /* Didn't grow, so the backup file must have
4420 * been used
4421 */
4422 old_disks = cnt;
4423 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4424 struct mdinfo dinfo;
4425 int fd;
4426 int bsbsize;
4427 char *devname, namebuf[20];
4428 unsigned long long lo, hi;
4429
4430 /* This was a spare and may have some saved data on it.
4431 * Load the superblock, find and load the
4432 * backup_super_block.
4433 * If either fail, go on to next device.
4434 * If the backup contains no new info, just return
4435 * else restore data and update all superblocks
4436 */
4437 if (i == old_disks-1) {
4438 fd = open(backup_file, O_RDONLY);
4439 if (fd<0) {
4440 pr_err("backup file %s inaccessible: %s\n",
4441 backup_file, strerror(errno));
4442 continue;
4443 }
4444 devname = backup_file;
4445 } else {
4446 fd = fdlist[i];
4447 if (fd < 0)
4448 continue;
4449 if (st->ss->load_super(st, fd, NULL))
4450 continue;
4451
4452 st->ss->getinfo_super(st, &dinfo, NULL);
4453 st->ss->free_super(st);
4454
4455 if (lseek64(fd,
4456 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4457 0) < 0) {
4458 pr_err("Cannot seek on device %d\n", i);
4459 continue; /* Cannot seek */
4460 }
4461 sprintf(namebuf, "device-%d", i);
4462 devname = namebuf;
4463 }
4464 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4465 if (verbose)
4466 pr_err("Cannot read from %s\n", devname);
4467 continue; /* Cannot read */
4468 }
4469 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4470 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4471 if (verbose)
4472 pr_err("No backup metadata on %s\n", devname);
4473 continue;
4474 }
4475 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4476 if (verbose)
4477 pr_err("Bad backup-metadata checksum on %s\n", devname);
4478 continue; /* bad checksum */
4479 }
4480 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4481 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4482 if (verbose)
4483 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4484 continue; /* Bad second checksum */
4485 }
4486 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4487 if (verbose)
4488 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4489 continue; /* Wrong uuid */
4490 }
4491
4492 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4493 * sometimes they aren't... So allow considerable flexability in matching, and allow
4494 * this test to be overridden by an environment variable.
4495 */
4496 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4497 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4498 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4499 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4500 (unsigned long)__le64_to_cpu(bsb.mtime),
4501 (unsigned long)info->array.utime);
4502 } else {
4503 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4504 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4505 continue; /* time stamp is too bad */
4506 }
4507 }
4508
4509 if (bsb.magic[15] == '1') {
4510 if (bsb.length == 0)
4511 continue;
4512 if (info->delta_disks >= 0) {
4513 /* reshape_progress is increasing */
4514 if (__le64_to_cpu(bsb.arraystart)
4515 + __le64_to_cpu(bsb.length)
4516 < info->reshape_progress) {
4517 nonew:
4518 if (verbose)
4519 pr_err("backup-metadata found on %s but is not needed\n", devname);
4520 continue; /* No new data here */
4521 }
4522 } else {
4523 /* reshape_progress is decreasing */
4524 if (__le64_to_cpu(bsb.arraystart) >=
4525 info->reshape_progress)
4526 goto nonew; /* No new data here */
4527 }
4528 } else {
4529 if (bsb.length == 0 && bsb.length2 == 0)
4530 continue;
4531 if (info->delta_disks >= 0) {
4532 /* reshape_progress is increasing */
4533 if ((__le64_to_cpu(bsb.arraystart)
4534 + __le64_to_cpu(bsb.length)
4535 < info->reshape_progress)
4536 &&
4537 (__le64_to_cpu(bsb.arraystart2)
4538 + __le64_to_cpu(bsb.length2)
4539 < info->reshape_progress))
4540 goto nonew; /* No new data here */
4541 } else {
4542 /* reshape_progress is decreasing */
4543 if (__le64_to_cpu(bsb.arraystart) >=
4544 info->reshape_progress &&
4545 __le64_to_cpu(bsb.arraystart2) >=
4546 info->reshape_progress)
4547 goto nonew; /* No new data here */
4548 }
4549 }
4550 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4551 second_fail:
4552 if (verbose)
4553 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4554 devname);
4555 continue; /* Cannot seek */
4556 }
4557 /* There should be a duplicate backup superblock 4k before here */
4558 if (lseek64(fd, -4096, 1) < 0 ||
4559 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4560 goto second_fail; /* Cannot find leading superblock */
4561 if (bsb.magic[15] == '1')
4562 bsbsize = offsetof(struct mdp_backup_super, pad1);
4563 else
4564 bsbsize = offsetof(struct mdp_backup_super, pad);
4565 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4566 goto second_fail; /* Cannot find leading superblock */
4567
4568 /* Now need the data offsets for all devices. */
4569 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4570 for(j=0; j<info->array.raid_disks; j++) {
4571 if (fdlist[j] < 0)
4572 continue;
4573 if (st->ss->load_super(st, fdlist[j], NULL))
4574 /* FIXME should be this be an error */
4575 continue;
4576 st->ss->getinfo_super(st, &dinfo, NULL);
4577 st->ss->free_super(st);
4578 offsets[j] = dinfo.data_offset * 512;
4579 }
4580 printf("%s: restoring critical section\n", Name);
4581
4582 if (restore_stripes(fdlist, offsets,
4583 info->array.raid_disks,
4584 info->new_chunk,
4585 info->new_level,
4586 info->new_layout,
4587 fd, __le64_to_cpu(bsb.devstart)*512,
4588 __le64_to_cpu(bsb.arraystart)*512,
4589 __le64_to_cpu(bsb.length)*512, NULL)) {
4590 /* didn't succeed, so giveup */
4591 if (verbose)
4592 pr_err("Error restoring backup from %s\n",
4593 devname);
4594 free(offsets);
4595 return 1;
4596 }
4597
4598 if (bsb.magic[15] == '2' &&
4599 restore_stripes(fdlist, offsets,
4600 info->array.raid_disks,
4601 info->new_chunk,
4602 info->new_level,
4603 info->new_layout,
4604 fd, __le64_to_cpu(bsb.devstart)*512 +
4605 __le64_to_cpu(bsb.devstart2)*512,
4606 __le64_to_cpu(bsb.arraystart2)*512,
4607 __le64_to_cpu(bsb.length2)*512, NULL)) {
4608 /* didn't succeed, so giveup */
4609 if (verbose)
4610 pr_err("Error restoring second backup from %s\n",
4611 devname);
4612 free(offsets);
4613 return 1;
4614 }
4615
4616 free(offsets);
4617
4618 /* Ok, so the data is restored. Let's update those superblocks. */
4619
4620 lo = hi = 0;
4621 if (bsb.length) {
4622 lo = __le64_to_cpu(bsb.arraystart);
4623 hi = lo + __le64_to_cpu(bsb.length);
4624 }
4625 if (bsb.magic[15] == '2' && bsb.length2) {
4626 unsigned long long lo1, hi1;
4627 lo1 = __le64_to_cpu(bsb.arraystart2);
4628 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4629 if (lo == hi) {
4630 lo = lo1;
4631 hi = hi1;
4632 } else if (lo < lo1)
4633 hi = hi1;
4634 else
4635 lo = lo1;
4636 }
4637 if (lo < hi &&
4638 (info->reshape_progress < lo ||
4639 info->reshape_progress > hi))
4640 /* backup does not affect reshape_progress*/ ;
4641 else if (info->delta_disks >= 0) {
4642 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4643 __le64_to_cpu(bsb.length);
4644 if (bsb.magic[15] == '2') {
4645 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4646 __le64_to_cpu(bsb.length2);
4647 if (p2 > info->reshape_progress)
4648 info->reshape_progress = p2;
4649 }
4650 } else {
4651 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4652 if (bsb.magic[15] == '2') {
4653 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4654 if (p2 < info->reshape_progress)
4655 info->reshape_progress = p2;
4656 }
4657 }
4658 for (j=0; j<info->array.raid_disks; j++) {
4659 if (fdlist[j] < 0)
4660 continue;
4661 if (st->ss->load_super(st, fdlist[j], NULL))
4662 continue;
4663 st->ss->getinfo_super(st, &dinfo, NULL);
4664 dinfo.reshape_progress = info->reshape_progress;
4665 st->ss->update_super(st, &dinfo,
4666 "_reshape_progress",
4667 NULL,0, 0, NULL);
4668 st->ss->store_super(st, fdlist[j]);
4669 st->ss->free_super(st);
4670 }
4671 return 0;
4672 }
4673 /* Didn't find any backup data, try to see if any
4674 * was needed.
4675 */
4676 if (info->delta_disks < 0) {
4677 /* When shrinking, the critical section is at the end.
4678 * So see if we are before the critical section.
4679 */
4680 unsigned long long first_block;
4681 nstripe = ostripe = 0;
4682 first_block = 0;
4683 while (ostripe >= nstripe) {
4684 ostripe += info->array.chunk_size / 512;
4685 first_block = ostripe * odata;
4686 nstripe = first_block / ndata / (info->new_chunk/512) *
4687 (info->new_chunk/512);
4688 }
4689
4690 if (info->reshape_progress >= first_block)
4691 return 0;
4692 }
4693 if (info->delta_disks > 0) {
4694 /* See if we are beyond the critical section. */
4695 unsigned long long last_block;
4696 nstripe = ostripe = 0;
4697 last_block = 0;
4698 while (nstripe >= ostripe) {
4699 nstripe += info->new_chunk / 512;
4700 last_block = nstripe * ndata;
4701 ostripe = last_block / odata / (info->array.chunk_size/512) *
4702 (info->array.chunk_size/512);
4703 }
4704
4705 if (info->reshape_progress >= last_block)
4706 return 0;
4707 }
4708 /* needed to recover critical section! */
4709 if (verbose)
4710 pr_err("Failed to find backup of critical section\n");
4711 return 1;
4712 }
4713
4714 int Grow_continue_command(char *devname, int fd,
4715 char *backup_file, int verbose)
4716 {
4717 int ret_val = 0;
4718 struct supertype *st = NULL;
4719 struct mdinfo *content = NULL;
4720 struct mdinfo array;
4721 char *subarray = NULL;
4722 struct mdinfo *cc = NULL;
4723 struct mdstat_ent *mdstat = NULL;
4724 int cfd = -1;
4725 int fd2 = -1;
4726
4727 dprintf("Grow continue from command line called for %s\n",
4728 devname);
4729
4730 st = super_by_fd(fd, &subarray);
4731 if (!st || !st->ss) {
4732 pr_err("Unable to determine metadata format for %s\n",
4733 devname);
4734 return 1;
4735 }
4736 dprintf("Grow continue is run for ");
4737 if (st->ss->external == 0) {
4738 int d;
4739 dprintf_cont("native array (%s)\n", devname);
4740 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4741 pr_err("%s is not an active md array - aborting\n", devname);
4742 ret_val = 1;
4743 goto Grow_continue_command_exit;
4744 }
4745 content = &array;
4746 /* Need to load a superblock.
4747 * FIXME we should really get what we need from
4748 * sysfs
4749 */
4750 for (d = 0; d < MAX_DISKS; d++) {
4751 mdu_disk_info_t disk;
4752 char *dv;
4753 int err;
4754 disk.number = d;
4755 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4756 continue;
4757 if (disk.major == 0 && disk.minor == 0)
4758 continue;
4759 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4760 continue;
4761 dv = map_dev(disk.major, disk.minor, 1);
4762 if (!dv)
4763 continue;
4764 fd2 = dev_open(dv, O_RDONLY);
4765 if (fd2 < 0)
4766 continue;
4767 err = st->ss->load_super(st, fd2, NULL);
4768 close(fd2);
4769 /* invalidate fd2 to avoid possible double close() */
4770 fd2 = -1;
4771 if (err)
4772 continue;
4773 break;
4774 }
4775 if (d == MAX_DISKS) {
4776 pr_err("Unable to load metadata for %s\n",
4777 devname);
4778 ret_val = 1;
4779 goto Grow_continue_command_exit;
4780 }
4781 st->ss->getinfo_super(st, content, NULL);
4782 } else {
4783 char *container;
4784
4785 if (subarray) {
4786 dprintf_cont("subarray (%s)\n", subarray);
4787 container = st->container_devnm;
4788 cfd = open_dev_excl(st->container_devnm);
4789 } else {
4790 container = st->devnm;
4791 close(fd);
4792 cfd = open_dev_excl(st->devnm);
4793 dprintf_cont("container (%s)\n", container);
4794 fd = cfd;
4795 }
4796 if (cfd < 0) {
4797 pr_err("Unable to open container for %s\n", devname);
4798 ret_val = 1;
4799 goto Grow_continue_command_exit;
4800 }
4801
4802 /* find in container array under reshape
4803 */
4804 ret_val = st->ss->load_container(st, cfd, NULL);
4805 if (ret_val) {
4806 pr_err("Cannot read superblock for %s\n",
4807 devname);
4808 ret_val = 1;
4809 goto Grow_continue_command_exit;
4810 }
4811
4812 cc = st->ss->container_content(st, subarray);
4813 for (content = cc; content ; content = content->next) {
4814 char *array;
4815 int allow_reshape = 1;
4816
4817 if (content->reshape_active == 0)
4818 continue;
4819 /* The decision about array or container wide
4820 * reshape is taken in Grow_continue based
4821 * content->reshape_active state, therefore we
4822 * need to check_reshape based on
4823 * reshape_active and subarray name
4824 */
4825 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4826 allow_reshape = 0;
4827 if (content->reshape_active == CONTAINER_RESHAPE &&
4828 (content->array.state
4829 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4830 allow_reshape = 0;
4831
4832 if (!allow_reshape) {
4833 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4834 devname, container);
4835 ret_val = 1;
4836 goto Grow_continue_command_exit;
4837 }
4838
4839 array = strchr(content->text_version+1, '/')+1;
4840 mdstat = mdstat_by_subdev(array, container);
4841 if (!mdstat)
4842 continue;
4843 if (mdstat->active == 0) {
4844 pr_err("Skipping inactive array %s.\n",
4845 mdstat->devnm);
4846 free_mdstat(mdstat);
4847 mdstat = NULL;
4848 continue;
4849 }
4850 break;
4851 }
4852 if (!content) {
4853 pr_err("Unable to determine reshaped array for %s\n", devname);
4854 ret_val = 1;
4855 goto Grow_continue_command_exit;
4856 }
4857 fd2 = open_dev(mdstat->devnm);
4858 if (fd2 < 0) {
4859 pr_err("cannot open (%s)\n", mdstat->devnm);
4860 ret_val = 1;
4861 goto Grow_continue_command_exit;
4862 }
4863
4864 sysfs_init(content, fd2, mdstat->devnm);
4865
4866 /* start mdmon in case it is not running
4867 */
4868 if (!mdmon_running(container))
4869 start_mdmon(container);
4870 ping_monitor(container);
4871
4872 if (mdmon_running(container))
4873 st->update_tail = &st->updates;
4874 else {
4875 pr_err("No mdmon found. Grow cannot continue.\n");
4876 ret_val = 1;
4877 goto Grow_continue_command_exit;
4878 }
4879 }
4880
4881 /* verify that array under reshape is started from
4882 * correct position
4883 */
4884 if (verify_reshape_position(content, content->array.level) < 0) {
4885 ret_val = 1;
4886 goto Grow_continue_command_exit;
4887 }
4888
4889 /* continue reshape
4890 */
4891 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4892
4893 Grow_continue_command_exit:
4894 if (fd2 > -1)
4895 close(fd2);
4896 if (cfd > -1)
4897 close(cfd);
4898 st->ss->free_super(st);
4899 free_mdstat(mdstat);
4900 sysfs_free(cc);
4901 free(subarray);
4902
4903 return ret_val;
4904 }
4905
4906 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4907 char *backup_file, int forked, int freeze_reshape)
4908 {
4909 int ret_val = 2;
4910
4911 if (!info->reshape_active)
4912 return ret_val;
4913
4914 if (st->ss->external) {
4915 int cfd = open_dev(st->container_devnm);
4916
4917 if (cfd < 0)
4918 return 1;
4919
4920 st->ss->load_container(st, cfd, st->container_devnm);
4921 close(cfd);
4922 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4923 st, info, 0, backup_file,
4924 0, forked,
4925 1 | info->reshape_active,
4926 freeze_reshape);
4927 } else
4928 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4929 NULL, INVALID_SECTORS,
4930 backup_file, 0, forked,
4931 1 | info->reshape_active,
4932 freeze_reshape);
4933
4934 return ret_val;
4935 }
4936
4937 char *make_backup(char *name)
4938 {
4939 char *base = "backup_file-";
4940 int len;
4941 char *fname;
4942
4943 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4944 fname = xmalloc(len);
4945 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4946 return fname;
4947 }
4948
4949 char *locate_backup(char *name)
4950 {
4951 char *fl = make_backup(name);
4952 struct stat stb;
4953
4954 if (stat(fl, &stb) == 0 &&
4955 S_ISREG(stb.st_mode))
4956 return fl;
4957
4958 free(fl);
4959 return NULL;
4960 }
Unnamed repository; edit this file 'description' to name the repository.