]> 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
Merge branch 'fix' of git://github.com/ldzhong/mdadm
[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 (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0)
301 major = BITMAP_MAJOR_CLUSTERED;
302
303 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
304 if (errno == ENOMEM)
305 pr_err("Memory allocation failure.\n");
306 else
307 pr_err("bitmaps not supported by this kernel.\n");
308 return 1;
309 }
310 if (bmf.pathname[0]) {
311 if (strcmp(s->bitmap_file,"none")==0) {
312 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
313 pr_err("failed to remove bitmap %s\n",
314 bmf.pathname);
315 return 1;
316 }
317 return 0;
318 }
319 pr_err("%s already has a bitmap (%s)\n",
320 devname, bmf.pathname);
321 return 1;
322 }
323 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
324 pr_err("cannot get array status for %s\n", devname);
325 return 1;
326 }
327 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
328 if (strcmp(s->bitmap_file, "none")==0) {
329 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
330 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
331 pr_err("failed to remove internal bitmap.\n");
332 return 1;
333 }
334 return 0;
335 }
336 pr_err("%s bitmap already present on %s\n", s->bitmap_file, devname);
337 return 1;
338 }
339
340 if (strcmp(s->bitmap_file, "none") == 0) {
341 pr_err("no bitmap found on %s\n", devname);
342 return 1;
343 }
344 if (array.level <= 0) {
345 pr_err("Bitmaps not meaningful with level %s\n",
346 map_num(pers, array.level)?:"of this array");
347 return 1;
348 }
349 bitmapsize = array.size;
350 bitmapsize <<= 1;
351 if (get_dev_size(fd, NULL, &array_size) &&
352 array_size > (0x7fffffffULL<<9)) {
353 /* Array is big enough that we cannot trust array.size
354 * try other approaches
355 */
356 bitmapsize = get_component_size(fd);
357 }
358 if (bitmapsize == 0) {
359 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
360 return 1;
361 }
362
363 if (array.level == 10) {
364 int ncopies = (array.layout&255)*((array.layout>>8)&255);
365 bitmapsize = bitmapsize * array.raid_disks / ncopies;
366 }
367
368 st = super_by_fd(fd, &subarray);
369 if (!st) {
370 pr_err("Cannot understand version %d.%d\n",
371 array.major_version, array.minor_version);
372 return 1;
373 }
374 if (subarray) {
375 pr_err("Cannot add bitmaps to sub-arrays yet\n");
376 free(subarray);
377 free(st);
378 return 1;
379 }
380 if (strcmp(s->bitmap_file, "internal") == 0 ||
381 strcmp(s->bitmap_file, "clustered") == 0) {
382 int rv;
383 int d;
384 int offset_setable = 0;
385 struct mdinfo *mdi;
386 if (st->ss->add_internal_bitmap == NULL) {
387 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
388 return 1;
389 }
390 st->nodes = c->nodes;
391 st->cluster_name = c->homecluster;
392 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
393 if (mdi)
394 offset_setable = 1;
395 for (d=0; d< st->max_devs; d++) {
396 mdu_disk_info_t disk;
397 char *dv;
398 disk.number = d;
399 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
400 continue;
401 if (disk.major == 0 &&
402 disk.minor == 0)
403 continue;
404 if ((disk.state & (1<<MD_DISK_SYNC))==0)
405 continue;
406 dv = map_dev(disk.major, disk.minor, 1);
407 if (dv) {
408 int fd2 = dev_open(dv, O_RDWR);
409 if (fd2 < 0)
410 continue;
411 if (st->ss->load_super(st, fd2, NULL)==0) {
412 if (st->ss->add_internal_bitmap(
413 st,
414 &s->bitmap_chunk, c->delay, s->write_behind,
415 bitmapsize, offset_setable,
416 major)
417 )
418 st->ss->write_bitmap(st, fd2, NoUpdate);
419 else {
420 pr_err("failed to create internal bitmap - chunksize problem.\n");
421 close(fd2);
422 return 1;
423 }
424 }
425 close(fd2);
426 }
427 }
428 if (offset_setable) {
429 st->ss->getinfo_super(st, mdi, NULL);
430 sysfs_init(mdi, fd, NULL);
431 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
432 mdi->bitmap_offset);
433 } else {
434 if (strcmp(s->bitmap_file, "clustered") == 0)
435 array.state |= (1<<MD_SB_CLUSTERED);
436 array.state |= (1<<MD_SB_BITMAP_PRESENT);
437 rv = ioctl(fd, SET_ARRAY_INFO, &array);
438 }
439 if (rv < 0) {
440 if (errno == EBUSY)
441 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
442 pr_err("failed to set internal bitmap.\n");
443 return 1;
444 }
445 } else {
446 int uuid[4];
447 int bitmap_fd;
448 int d;
449 int max_devs = st->max_devs;
450
451 /* try to load a superblock */
452 for (d = 0; d < max_devs; d++) {
453 mdu_disk_info_t disk;
454 char *dv;
455 int fd2;
456 disk.number = d;
457 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
458 continue;
459 if ((disk.major==0 && disk.minor==0) ||
460 (disk.state & (1<<MD_DISK_REMOVED)))
461 continue;
462 dv = map_dev(disk.major, disk.minor, 1);
463 if (!dv)
464 continue;
465 fd2 = dev_open(dv, O_RDONLY);
466 if (fd2 >= 0) {
467 if (st->ss->load_super(st, fd2, NULL) == 0) {
468 close(fd2);
469 st->ss->uuid_from_super(st, uuid);
470 break;
471 }
472 close(fd2);
473 }
474 }
475 if (d == max_devs) {
476 pr_err("cannot find UUID for array!\n");
477 return 1;
478 }
479 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
480 c->delay, s->write_behind, bitmapsize, major)) {
481 return 1;
482 }
483 bitmap_fd = open(s->bitmap_file, O_RDWR);
484 if (bitmap_fd < 0) {
485 pr_err("weird: %s cannot be opened\n",
486 s->bitmap_file);
487 return 1;
488 }
489 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
490 int err = errno;
491 if (errno == EBUSY)
492 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
493 pr_err("Cannot set bitmap file for %s: %s\n",
494 devname, strerror(err));
495 return 1;
496 }
497 }
498
499 return 0;
500 }
501
502 /*
503 * When reshaping an array we might need to backup some data.
504 * This is written to all spares with a 'super_block' describing it.
505 * The superblock goes 4K from the end of the used space on the
506 * device.
507 * It if written after the backup is complete.
508 * It has the following structure.
509 */
510
511 static struct mdp_backup_super {
512 char magic[16]; /* md_backup_data-1 or -2 */
513 __u8 set_uuid[16];
514 __u64 mtime;
515 /* start/sizes in 512byte sectors */
516 __u64 devstart; /* address on backup device/file of data */
517 __u64 arraystart;
518 __u64 length;
519 __u32 sb_csum; /* csum of preceeding bytes. */
520 __u32 pad1;
521 __u64 devstart2; /* offset in to data of second section */
522 __u64 arraystart2;
523 __u64 length2;
524 __u32 sb_csum2; /* csum of preceeding bytes. */
525 __u8 pad[512-68-32];
526 } __attribute__((aligned(512))) bsb, bsb2;
527
528 static __u32 bsb_csum(char *buf, int len)
529 {
530 int i;
531 int csum = 0;
532 for (i = 0; i < len; i++)
533 csum = (csum<<3) + buf[0];
534 return __cpu_to_le32(csum);
535 }
536
537 static int check_idle(struct supertype *st)
538 {
539 /* Check that all member arrays for this container, or the
540 * container of this array, are idle
541 */
542 char *container = (st->container_devnm[0]
543 ? st->container_devnm : st->devnm);
544 struct mdstat_ent *ent, *e;
545 int is_idle = 1;
546
547 ent = mdstat_read(0, 0);
548 for (e = ent ; e; e = e->next) {
549 if (!is_container_member(e, container))
550 continue;
551 if (e->percent >= 0) {
552 is_idle = 0;
553 break;
554 }
555 }
556 free_mdstat(ent);
557 return is_idle;
558 }
559
560 static int freeze_container(struct supertype *st)
561 {
562 char *container = (st->container_devnm[0]
563 ? st->container_devnm : st->devnm);
564
565 if (!check_idle(st))
566 return -1;
567
568 if (block_monitor(container, 1)) {
569 pr_err("failed to freeze container\n");
570 return -2;
571 }
572
573 return 1;
574 }
575
576 static void unfreeze_container(struct supertype *st)
577 {
578 char *container = (st->container_devnm[0]
579 ? st->container_devnm : st->devnm);
580
581 unblock_monitor(container, 1);
582 }
583
584 static int freeze(struct supertype *st)
585 {
586 /* Try to freeze resync/rebuild on this array/container.
587 * Return -1 if the array is busy,
588 * return -2 container cannot be frozen,
589 * return 0 if this kernel doesn't support 'frozen'
590 * return 1 if it worked.
591 */
592 if (st->ss->external)
593 return freeze_container(st);
594 else {
595 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
596 int err;
597 char buf[20];
598
599 if (!sra)
600 return -1;
601 /* Need to clear any 'read-auto' status */
602 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
603 strncmp(buf, "read-auto", 9) == 0)
604 sysfs_set_str(sra, NULL, "array_state", "clean");
605
606 err = sysfs_freeze_array(sra);
607 sysfs_free(sra);
608 return err;
609 }
610 }
611
612 static void unfreeze(struct supertype *st)
613 {
614 if (st->ss->external)
615 return unfreeze_container(st);
616 else {
617 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
618 char buf[20];
619
620 if (sra &&
621 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
622 && strcmp(buf, "frozen\n") == 0)
623 sysfs_set_str(sra, NULL, "sync_action", "idle");
624 sysfs_free(sra);
625 }
626 }
627
628 static void wait_reshape(struct mdinfo *sra)
629 {
630 int fd = sysfs_get_fd(sra, NULL, "sync_action");
631 char action[20];
632
633 if (fd < 0)
634 return;
635
636 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
637 strncmp(action, "reshape", 7) == 0)
638 sysfs_wait(fd, NULL);
639 close(fd);
640 }
641
642 static int reshape_super(struct supertype *st, unsigned long long size,
643 int level, int layout, int chunksize, int raid_disks,
644 int delta_disks, char *backup_file, char *dev,
645 int direction, int verbose)
646 {
647 /* nothing extra to check in the native case */
648 if (!st->ss->external)
649 return 0;
650 if (!st->ss->reshape_super ||
651 !st->ss->manage_reshape) {
652 pr_err("%s metadata does not support reshape\n",
653 st->ss->name);
654 return 1;
655 }
656
657 return st->ss->reshape_super(st, size, level, layout, chunksize,
658 raid_disks, delta_disks, backup_file, dev,
659 direction, verbose);
660 }
661
662 static void sync_metadata(struct supertype *st)
663 {
664 if (st->ss->external) {
665 if (st->update_tail) {
666 flush_metadata_updates(st);
667 st->update_tail = &st->updates;
668 } else
669 st->ss->sync_metadata(st);
670 }
671 }
672
673 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
674 {
675 /* when dealing with external metadata subarrays we need to be
676 * prepared to handle EAGAIN. The kernel may need to wait for
677 * mdmon to mark the array active so the kernel can handle
678 * allocations/writeback when preparing the reshape action
679 * (md_allow_write()). We temporarily disable safe_mode_delay
680 * to close a race with the array_state going clean before the
681 * next write to raid_disks / stripe_cache_size
682 */
683 char safe[50];
684 int rc;
685
686 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
687 if (!container ||
688 (strcmp(name, "raid_disks") != 0 &&
689 strcmp(name, "stripe_cache_size") != 0))
690 return sysfs_set_num(sra, NULL, name, n);
691
692 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
693 if (rc <= 0)
694 return -1;
695 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
696 rc = sysfs_set_num(sra, NULL, name, n);
697 if (rc < 0 && errno == EAGAIN) {
698 ping_monitor(container);
699 /* if we get EAGAIN here then the monitor is not active
700 * so stop trying
701 */
702 rc = sysfs_set_num(sra, NULL, name, n);
703 }
704 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
705 return rc;
706 }
707
708 int start_reshape(struct mdinfo *sra, int already_running,
709 int before_data_disks, int data_disks)
710 {
711 int err;
712 unsigned long long sync_max_to_set;
713
714 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
715 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
716 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
717 sra->reshape_progress);
718 if (before_data_disks <= data_disks)
719 sync_max_to_set = sra->reshape_progress / data_disks;
720 else
721 sync_max_to_set = (sra->component_size * data_disks
722 - sra->reshape_progress) / data_disks;
723 if (!already_running)
724 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
725 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
726 if (!already_running && err == 0) {
727 int cnt = 5;
728 do {
729 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
730 if (err)
731 sleep(1);
732 } while (err && errno == EBUSY && cnt-- > 0);
733 }
734 return err;
735 }
736
737 void abort_reshape(struct mdinfo *sra)
738 {
739 sysfs_set_str(sra, NULL, "sync_action", "idle");
740 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
741 sysfs_set_num(sra, NULL, "suspend_hi", 0);
742 sysfs_set_num(sra, NULL, "suspend_lo", 0);
743 sysfs_set_num(sra, NULL, "sync_min", 0);
744 // It isn't safe to reset sync_max as we aren't monitoring.
745 // Array really should be stopped at this point.
746 }
747
748 int remove_disks_for_takeover(struct supertype *st,
749 struct mdinfo *sra,
750 int layout)
751 {
752 int nr_of_copies;
753 struct mdinfo *remaining;
754 int slot;
755
756 if (sra->array.level == 10)
757 nr_of_copies = layout & 0xff;
758 else if (sra->array.level == 1)
759 nr_of_copies = sra->array.raid_disks;
760 else
761 return 1;
762
763 remaining = sra->devs;
764 sra->devs = NULL;
765 /* for each 'copy', select one device and remove from the list. */
766 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
767 struct mdinfo **diskp;
768 int found = 0;
769
770 /* Find a working device to keep */
771 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
772 struct mdinfo *disk = *diskp;
773
774 if (disk->disk.raid_disk < slot)
775 continue;
776 if (disk->disk.raid_disk >= slot + nr_of_copies)
777 continue;
778 if (disk->disk.state & (1<<MD_DISK_REMOVED))
779 continue;
780 if (disk->disk.state & (1<<MD_DISK_FAULTY))
781 continue;
782 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
783 continue;
784
785 /* We have found a good disk to use! */
786 *diskp = disk->next;
787 disk->next = sra->devs;
788 sra->devs = disk;
789 found = 1;
790 break;
791 }
792 if (!found)
793 break;
794 }
795
796 if (slot < sra->array.raid_disks) {
797 /* didn't find all slots */
798 struct mdinfo **e;
799 e = &remaining;
800 while (*e)
801 e = &(*e)->next;
802 *e = sra->devs;
803 sra->devs = remaining;
804 return 1;
805 }
806
807 /* Remove all 'remaining' devices from the array */
808 while (remaining) {
809 struct mdinfo *sd = remaining;
810 remaining = sd->next;
811
812 sysfs_set_str(sra, sd, "state", "faulty");
813 sysfs_set_str(sra, sd, "slot", "none");
814 /* for external metadata disks should be removed in mdmon */
815 if (!st->ss->external)
816 sysfs_set_str(sra, sd, "state", "remove");
817 sd->disk.state |= (1<<MD_DISK_REMOVED);
818 sd->disk.state &= ~(1<<MD_DISK_SYNC);
819 sd->next = sra->devs;
820 sra->devs = sd;
821 }
822 return 0;
823 }
824
825 void reshape_free_fdlist(int *fdlist,
826 unsigned long long *offsets,
827 int size)
828 {
829 int i;
830
831 for (i = 0; i < size; i++)
832 if (fdlist[i] >= 0)
833 close(fdlist[i]);
834
835 free(fdlist);
836 free(offsets);
837 }
838
839 int reshape_prepare_fdlist(char *devname,
840 struct mdinfo *sra,
841 int raid_disks,
842 int nrdisks,
843 unsigned long blocks,
844 char *backup_file,
845 int *fdlist,
846 unsigned long long *offsets)
847 {
848 int d = 0;
849 struct mdinfo *sd;
850
851 enable_fds(nrdisks);
852 for (d = 0; d <= nrdisks; d++)
853 fdlist[d] = -1;
854 d = raid_disks;
855 for (sd = sra->devs; sd; sd = sd->next) {
856 if (sd->disk.state & (1<<MD_DISK_FAULTY))
857 continue;
858 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
859 sd->disk.raid_disk < raid_disks) {
860 char *dn = map_dev(sd->disk.major,
861 sd->disk.minor, 1);
862 fdlist[sd->disk.raid_disk]
863 = dev_open(dn, O_RDONLY);
864 offsets[sd->disk.raid_disk] = sd->data_offset*512;
865 if (fdlist[sd->disk.raid_disk] < 0) {
866 pr_err("%s: cannot open component %s\n",
867 devname, dn ? dn : "-unknown-");
868 d = -1;
869 goto release;
870 }
871 } else if (backup_file == NULL) {
872 /* spare */
873 char *dn = map_dev(sd->disk.major,
874 sd->disk.minor, 1);
875 fdlist[d] = dev_open(dn, O_RDWR);
876 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
877 if (fdlist[d] < 0) {
878 pr_err("%s: cannot open component %s\n",
879 devname, dn ? dn : "-unknown-");
880 d = -1;
881 goto release;
882 }
883 d++;
884 }
885 }
886 release:
887 return d;
888 }
889
890 int reshape_open_backup_file(char *backup_file,
891 int fd,
892 char *devname,
893 long blocks,
894 int *fdlist,
895 unsigned long long *offsets,
896 char *sys_name,
897 int restart)
898 {
899 /* Return 1 on success, 0 on any form of failure */
900 /* need to check backup file is large enough */
901 char buf[512];
902 struct stat stb;
903 unsigned int dev;
904 int i;
905
906 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
907 S_IRUSR | S_IWUSR);
908 *offsets = 8 * 512;
909 if (*fdlist < 0) {
910 pr_err("%s: cannot create backup file %s: %s\n",
911 devname, backup_file, strerror(errno));
912 return 0;
913 }
914 /* Guard against backup file being on array device.
915 * If array is partitioned or if LVM etc is in the
916 * way this will not notice, but it is better than
917 * nothing.
918 */
919 fstat(*fdlist, &stb);
920 dev = stb.st_dev;
921 fstat(fd, &stb);
922 if (stb.st_rdev == dev) {
923 pr_err("backup file must NOT be on the array being reshaped.\n");
924 close(*fdlist);
925 return 0;
926 }
927
928 memset(buf, 0, 512);
929 for (i=0; i < blocks + 8 ; i++) {
930 if (write(*fdlist, buf, 512) != 512) {
931 pr_err("%s: cannot create backup file %s: %s\n",
932 devname, backup_file, strerror(errno));
933 return 0;
934 }
935 }
936 if (fsync(*fdlist) != 0) {
937 pr_err("%s: cannot create backup file %s: %s\n",
938 devname, backup_file, strerror(errno));
939 return 0;
940 }
941
942 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
943 char *bu = make_backup(sys_name);
944 if (symlink(backup_file, bu))
945 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
946 strerror(errno));
947 free(bu);
948 }
949
950 return 1;
951 }
952
953 unsigned long compute_backup_blocks(int nchunk, int ochunk,
954 unsigned int ndata, unsigned int odata)
955 {
956 unsigned long a, b, blocks;
957 /* So how much do we need to backup.
958 * We need an amount of data which is both a whole number of
959 * old stripes and a whole number of new stripes.
960 * So LCM for (chunksize*datadisks).
961 */
962 a = (ochunk/512) * odata;
963 b = (nchunk/512) * ndata;
964 /* Find GCD */
965 a = GCD(a, b);
966 /* LCM == product / GCD */
967 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
968
969 return blocks;
970 }
971
972 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
973 {
974 /* Based on the current array state in info->array and
975 * the changes in info->new_* etc, determine:
976 * - whether the change is possible
977 * - Intermediate level/raid_disks/layout
978 * - whether a restriping reshape is needed
979 * - number of sectors in minimum change unit. This
980 * will cover a whole number of stripes in 'before' and
981 * 'after'.
982 *
983 * Return message if the change should be rejected
984 * NULL if the change can be achieved
985 *
986 * This can be called as part of starting a reshape, or
987 * when assembling an array that is undergoing reshape.
988 */
989 int near, far, offset, copies;
990 int new_disks;
991 int old_chunk, new_chunk;
992 /* delta_parity records change in number of devices
993 * caused by level change
994 */
995 int delta_parity = 0;
996
997 memset(re, 0, sizeof(*re));
998
999 /* If a new level not explicitly given, we assume no-change */
1000 if (info->new_level == UnSet)
1001 info->new_level = info->array.level;
1002
1003 if (info->new_chunk)
1004 switch (info->new_level) {
1005 case 0:
1006 case 4:
1007 case 5:
1008 case 6:
1009 case 10:
1010 /* chunk size is meaningful, must divide component_size
1011 * evenly
1012 */
1013 if (info->component_size % (info->new_chunk/512)) {
1014 unsigned long long shrink = info->component_size;
1015 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1016 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1017 info->new_chunk/1024, info->component_size/2);
1018 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1019 devname, shrink/2);
1020 pr_err("will shrink the array so the given chunk size would work.\n");
1021 return "";
1022 }
1023 break;
1024 default:
1025 return "chunk size not meaningful for this level";
1026 }
1027 else
1028 info->new_chunk = info->array.chunk_size;
1029
1030 switch (info->array.level) {
1031 default:
1032 return "No reshape is possibly for this RAID level";
1033 case LEVEL_LINEAR:
1034 if (info->delta_disks != UnSet)
1035 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1036 else
1037 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1038 case 1:
1039 /* RAID1 can convert to RAID1 with different disks, or
1040 * raid5 with 2 disks, or
1041 * raid0 with 1 disk
1042 */
1043 if (info->new_level > 1 &&
1044 (info->component_size & 7))
1045 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1046 if (info->new_level == 0) {
1047 if (info->delta_disks != UnSet &&
1048 info->delta_disks != 0)
1049 return "Cannot change number of disks with RAID1->RAID0 conversion";
1050 re->level = 0;
1051 re->before.data_disks = 1;
1052 re->after.data_disks = 1;
1053 return NULL;
1054 }
1055 if (info->new_level == 1) {
1056 if (info->delta_disks == UnSet)
1057 /* Don't know what to do */
1058 return "no change requested for Growing RAID1";
1059 re->level = 1;
1060 return NULL;
1061 }
1062 if (info->array.raid_disks == 2 &&
1063 info->new_level == 5) {
1064
1065 re->level = 5;
1066 re->before.data_disks = 1;
1067 if (info->delta_disks != UnSet &&
1068 info->delta_disks != 0)
1069 re->after.data_disks = 1 + info->delta_disks;
1070 else
1071 re->after.data_disks = 1;
1072 if (re->after.data_disks < 1)
1073 return "Number of disks too small for RAID5";
1074
1075 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1076 info->array.chunk_size = 65536;
1077 break;
1078 }
1079 /* Could do some multi-stage conversions, but leave that to
1080 * later.
1081 */
1082 return "Impossibly level change request for RAID1";
1083
1084 case 10:
1085 /* RAID10 can be converted from near mode to
1086 * RAID0 by removing some devices.
1087 * It can also be reshaped if the kernel supports
1088 * new_data_offset.
1089 */
1090 switch (info->new_level) {
1091 case 0:
1092 if ((info->array.layout & ~0xff) != 0x100)
1093 return "Cannot Grow RAID10 with far/offset layout";
1094 /* number of devices must be multiple of number of copies */
1095 if (info->array.raid_disks % (info->array.layout & 0xff))
1096 return "RAID10 layout too complex for Grow operation";
1097
1098 new_disks = (info->array.raid_disks
1099 / (info->array.layout & 0xff));
1100 if (info->delta_disks == UnSet)
1101 info->delta_disks = (new_disks
1102 - info->array.raid_disks);
1103
1104 if (info->delta_disks != new_disks - info->array.raid_disks)
1105 return "New number of raid-devices impossible for RAID10";
1106 if (info->new_chunk &&
1107 info->new_chunk != info->array.chunk_size)
1108 return "Cannot change chunk-size with RAID10 Grow";
1109
1110 /* looks good */
1111 re->level = 0;
1112 re->before.data_disks = new_disks;
1113 re->after.data_disks = re->before.data_disks;
1114 return NULL;
1115
1116 case 10:
1117 near = info->array.layout & 0xff;
1118 far = (info->array.layout >> 8) & 0xff;
1119 offset = info->array.layout & 0x10000;
1120 if (far > 1 && !offset)
1121 return "Cannot reshape RAID10 in far-mode";
1122 copies = near * far;
1123
1124 old_chunk = info->array.chunk_size * far;
1125
1126 if (info->new_layout == UnSet)
1127 info->new_layout = info->array.layout;
1128 else {
1129 near = info->new_layout & 0xff;
1130 far = (info->new_layout >> 8) & 0xff;
1131 offset = info->new_layout & 0x10000;
1132 if (far > 1 && !offset)
1133 return "Cannot reshape RAID10 to far-mode";
1134 if (near * far != copies)
1135 return "Cannot change number of copies when reshaping RAID10";
1136 }
1137 if (info->delta_disks == UnSet)
1138 info->delta_disks = 0;
1139 new_disks = (info->array.raid_disks +
1140 info->delta_disks);
1141
1142 new_chunk = info->new_chunk * far;
1143
1144 re->level = 10;
1145 re->before.layout = info->array.layout;
1146 re->before.data_disks = info->array.raid_disks;
1147 re->after.layout = info->new_layout;
1148 re->after.data_disks = new_disks;
1149 /* For RAID10 we don't do backup but do allow reshape,
1150 * so set backup_blocks to INVALID_SECTORS rather than
1151 * zero.
1152 * And there is no need to synchronise stripes on both
1153 * 'old' and 'new'. So the important
1154 * number is the minimum data_offset difference
1155 * which is the larger of (offset copies * chunk).
1156 */
1157 re->backup_blocks = INVALID_SECTORS;
1158 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1159 if (new_disks < re->before.data_disks &&
1160 info->space_after < re->min_offset_change)
1161 /* Reduce component size by one chunk */
1162 re->new_size = (info->component_size -
1163 re->min_offset_change);
1164 else
1165 re->new_size = info->component_size;
1166 re->new_size = re->new_size * new_disks / copies;
1167 return NULL;
1168
1169 default:
1170 return "RAID10 can only be changed to RAID0";
1171 }
1172 case 0:
1173 /* RAID0 can be converted to RAID10, or to RAID456 */
1174 if (info->new_level == 10) {
1175 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1176 /* Assume near=2 layout */
1177 info->new_layout = 0x102;
1178 info->delta_disks = info->array.raid_disks;
1179 }
1180 if (info->new_layout == UnSet) {
1181 int copies = 1 + (info->delta_disks
1182 / info->array.raid_disks);
1183 if (info->array.raid_disks * (copies-1)
1184 != info->delta_disks)
1185 return "Impossible number of devices for RAID0->RAID10";
1186 info->new_layout = 0x100 + copies;
1187 }
1188 if (info->delta_disks == UnSet) {
1189 int copies = info->new_layout & 0xff;
1190 if (info->new_layout != 0x100 + copies)
1191 return "New layout impossible for RAID0->RAID10";;
1192 info->delta_disks = (copies - 1) *
1193 info->array.raid_disks;
1194 }
1195 if (info->new_chunk &&
1196 info->new_chunk != info->array.chunk_size)
1197 return "Cannot change chunk-size with RAID0->RAID10";
1198 /* looks good */
1199 re->level = 10;
1200 re->before.data_disks = (info->array.raid_disks +
1201 info->delta_disks);
1202 re->after.data_disks = re->before.data_disks;
1203 re->before.layout = info->new_layout;
1204 return NULL;
1205 }
1206
1207 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1208 * a raid4 style layout of the final level.
1209 */
1210 switch (info->new_level) {
1211 case 4:
1212 delta_parity = 1;
1213 case 0:
1214 re->level = 4;
1215 re->before.layout = 0;
1216 break;
1217 case 5:
1218 delta_parity = 1;
1219 re->level = 5;
1220 re->before.layout = ALGORITHM_PARITY_N;
1221 if (info->new_layout == UnSet)
1222 info->new_layout = map_name(r5layout, "default");
1223 break;
1224 case 6:
1225 delta_parity = 2;
1226 re->level = 6;
1227 re->before.layout = ALGORITHM_PARITY_N;
1228 if (info->new_layout == UnSet)
1229 info->new_layout = map_name(r6layout, "default");
1230 break;
1231 default:
1232 return "Impossible level change requested";
1233 }
1234 re->before.data_disks = info->array.raid_disks;
1235 /* determining 'after' layout happens outside this 'switch' */
1236 break;
1237
1238 case 4:
1239 info->array.layout = ALGORITHM_PARITY_N;
1240 case 5:
1241 switch (info->new_level) {
1242 case 0:
1243 delta_parity = -1;
1244 case 4:
1245 re->level = info->array.level;
1246 re->before.data_disks = info->array.raid_disks - 1;
1247 re->before.layout = info->array.layout;
1248 break;
1249 case 5:
1250 re->level = 5;
1251 re->before.data_disks = info->array.raid_disks - 1;
1252 re->before.layout = info->array.layout;
1253 break;
1254 case 6:
1255 delta_parity = 1;
1256 re->level = 6;
1257 re->before.data_disks = info->array.raid_disks - 1;
1258 switch (info->array.layout) {
1259 case ALGORITHM_LEFT_ASYMMETRIC:
1260 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1261 break;
1262 case ALGORITHM_RIGHT_ASYMMETRIC:
1263 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1264 break;
1265 case ALGORITHM_LEFT_SYMMETRIC:
1266 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1267 break;
1268 case ALGORITHM_RIGHT_SYMMETRIC:
1269 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1270 break;
1271 case ALGORITHM_PARITY_0:
1272 re->before.layout = ALGORITHM_PARITY_0_6;
1273 break;
1274 case ALGORITHM_PARITY_N:
1275 re->before.layout = ALGORITHM_PARITY_N_6;
1276 break;
1277 default:
1278 return "Cannot convert an array with this layout";
1279 }
1280 break;
1281 case 1:
1282 if (info->array.raid_disks != 2)
1283 return "Can only convert a 2-device array to RAID1";
1284 if (info->delta_disks != UnSet &&
1285 info->delta_disks != 0)
1286 return "Cannot set raid_disk when converting RAID5->RAID1";
1287 re->level = 1;
1288 info->new_chunk = 0;
1289 return NULL;
1290 default:
1291 return "Impossible level change requested";
1292 }
1293 break;
1294 case 6:
1295 switch (info->new_level) {
1296 case 4:
1297 case 5:
1298 delta_parity = -1;
1299 case 6:
1300 re->level = 6;
1301 re->before.data_disks = info->array.raid_disks - 2;
1302 re->before.layout = info->array.layout;
1303 break;
1304 default:
1305 return "Impossible level change requested";
1306 }
1307 break;
1308 }
1309
1310 /* If we reached here then it looks like a re-stripe is
1311 * happening. We have determined the intermediate level
1312 * and initial raid_disks/layout and stored these in 're'.
1313 *
1314 * We need to deduce the final layout that can be atomically
1315 * converted to the end state.
1316 */
1317 switch (info->new_level) {
1318 case 0:
1319 /* We can only get to RAID0 from RAID4 or RAID5
1320 * with appropriate layout and one extra device
1321 */
1322 if (re->level != 4 && re->level != 5)
1323 return "Cannot covert to RAID0 from this level";
1324
1325 switch (re->level) {
1326 case 4:
1327 re->before.layout = 0;
1328 re->after.layout = 0;
1329 break;
1330 case 5:
1331 re->after.layout = ALGORITHM_PARITY_N;
1332 break;
1333 }
1334 break;
1335
1336 case 4:
1337 /* We can only get to RAID4 from RAID5 */
1338 if (re->level != 4 && re->level != 5)
1339 return "Cannot convert to RAID4 from this level";
1340
1341 switch (re->level) {
1342 case 4:
1343 re->after.layout = 0;
1344 break;
1345 case 5:
1346 re->after.layout = ALGORITHM_PARITY_N;
1347 break;
1348 }
1349 break;
1350
1351 case 5:
1352 /* We get to RAID5 from RAID5 or RAID6 */
1353 if (re->level != 5 && re->level != 6)
1354 return "Cannot convert to RAID5 from this level";
1355
1356 switch (re->level) {
1357 case 5:
1358 if (info->new_layout == UnSet)
1359 re->after.layout = re->before.layout;
1360 else
1361 re->after.layout = info->new_layout;
1362 break;
1363 case 6:
1364 if (info->new_layout == UnSet)
1365 info->new_layout = re->before.layout;
1366
1367 /* after.layout needs to be raid6 version of new_layout */
1368 if (info->new_layout == ALGORITHM_PARITY_N)
1369 re->after.layout = ALGORITHM_PARITY_N;
1370 else {
1371 char layout[40];
1372 char *ls = map_num(r5layout, info->new_layout);
1373 int l;
1374 if (ls) {
1375 /* Current RAID6 layout has a RAID5
1376 * equivalent - good
1377 */
1378 strcat(strcpy(layout, ls), "-6");
1379 l = map_name(r6layout, layout);
1380 if (l == UnSet)
1381 return "Cannot find RAID6 layout to convert to";
1382 } else {
1383 /* Current RAID6 has no equivalent.
1384 * If it is already a '-6' layout we
1385 * can leave it unchanged, else we must
1386 * fail
1387 */
1388 ls = map_num(r6layout, info->new_layout);
1389 if (!ls ||
1390 strcmp(ls+strlen(ls)-2, "-6") != 0)
1391 return "Please specify new layout";
1392 l = info->new_layout;
1393 }
1394 re->after.layout = l;
1395 }
1396 }
1397 break;
1398
1399 case 6:
1400 /* We must already be at level 6 */
1401 if (re->level != 6)
1402 return "Impossible level change";
1403 if (info->new_layout == UnSet)
1404 re->after.layout = info->array.layout;
1405 else
1406 re->after.layout = info->new_layout;
1407 break;
1408 default:
1409 return "Impossible level change requested";
1410 }
1411 if (info->delta_disks == UnSet)
1412 info->delta_disks = delta_parity;
1413
1414 re->after.data_disks = (re->before.data_disks
1415 + info->delta_disks
1416 - delta_parity);
1417 switch (re->level) {
1418 case 6: re->parity = 2;
1419 break;
1420 case 4:
1421 case 5: re->parity = 1;
1422 break;
1423 default: re->parity = 0;
1424 break;
1425 }
1426 /* So we have a restripe operation, we need to calculate the number
1427 * of blocks per reshape operation.
1428 */
1429 re->new_size = info->component_size * re->before.data_disks;
1430 if (info->new_chunk == 0)
1431 info->new_chunk = info->array.chunk_size;
1432 if (re->after.data_disks == re->before.data_disks &&
1433 re->after.layout == re->before.layout &&
1434 info->new_chunk == info->array.chunk_size) {
1435 /* Nothing to change, can change level immediately. */
1436 re->level = info->new_level;
1437 re->backup_blocks = 0;
1438 return NULL;
1439 }
1440 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1441 /* chunk and layout changes make no difference */
1442 re->level = info->new_level;
1443 re->backup_blocks = 0;
1444 return NULL;
1445 }
1446
1447 if (re->after.data_disks == re->before.data_disks &&
1448 get_linux_version() < 2006032)
1449 return "in-place reshape is not safe before 2.6.32 - sorry.";
1450
1451 if (re->after.data_disks < re->before.data_disks &&
1452 get_linux_version() < 2006030)
1453 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1454
1455 re->backup_blocks = compute_backup_blocks(
1456 info->new_chunk, info->array.chunk_size,
1457 re->after.data_disks,
1458 re->before.data_disks);
1459 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1460
1461 re->new_size = info->component_size * re->after.data_disks;
1462 return NULL;
1463 }
1464
1465 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1466 char *text_version)
1467 {
1468 struct mdinfo *info;
1469 char *subarray;
1470 int ret_val = -1;
1471
1472 if ((st == NULL) || (sra == NULL))
1473 return ret_val;
1474
1475 if (text_version == NULL)
1476 text_version = sra->text_version;
1477 subarray = strchr(text_version+1, '/')+1;
1478 info = st->ss->container_content(st, subarray);
1479 if (info) {
1480 unsigned long long current_size = 0;
1481 unsigned long long new_size =
1482 info->custom_array_size/2;
1483
1484 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1485 new_size > current_size) {
1486 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1487 < 0)
1488 dprintf("Error: Cannot set array size");
1489 else {
1490 ret_val = 0;
1491 dprintf("Array size changed");
1492 }
1493 dprintf_cont(" from %llu to %llu.\n",
1494 current_size, new_size);
1495 }
1496 sysfs_free(info);
1497 } else
1498 dprintf("Error: set_array_size(): info pointer in NULL\n");
1499
1500 return ret_val;
1501 }
1502
1503 static int reshape_array(char *container, int fd, char *devname,
1504 struct supertype *st, struct mdinfo *info,
1505 int force, struct mddev_dev *devlist,
1506 unsigned long long data_offset,
1507 char *backup_file, int verbose, int forked,
1508 int restart, int freeze_reshape);
1509 static int reshape_container(char *container, char *devname,
1510 int mdfd,
1511 struct supertype *st,
1512 struct mdinfo *info,
1513 int force,
1514 char *backup_file, int verbose,
1515 int forked, int restart, int freeze_reshape);
1516
1517 int Grow_reshape(char *devname, int fd,
1518 struct mddev_dev *devlist,
1519 unsigned long long data_offset,
1520 struct context *c, struct shape *s)
1521 {
1522 /* Make some changes in the shape of an array.
1523 * The kernel must support the change.
1524 *
1525 * There are three different changes. Each can trigger
1526 * a resync or recovery so we freeze that until we have
1527 * requested everything (if kernel supports freezing - 2.6.30).
1528 * The steps are:
1529 * - change size (i.e. component_size)
1530 * - change level
1531 * - change layout/chunksize/ndisks
1532 *
1533 * The last can require a reshape. It is different on different
1534 * levels so we need to check the level before actioning it.
1535 * Some times the level change needs to be requested after the
1536 * reshape (e.g. raid6->raid5, raid5->raid0)
1537 *
1538 */
1539 struct mdu_array_info_s array;
1540 int rv = 0;
1541 struct supertype *st;
1542 char *subarray = NULL;
1543
1544 int frozen;
1545 int changed = 0;
1546 char *container = NULL;
1547 int cfd = -1;
1548
1549 struct mddev_dev *dv;
1550 int added_disks;
1551
1552 struct mdinfo info;
1553 struct mdinfo *sra;
1554
1555 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1556 pr_err("%s is not an active md array - aborting\n",
1557 devname);
1558 return 1;
1559 }
1560 if (data_offset != INVALID_SECTORS && array.level != 10
1561 && (array.level < 4 || array.level > 6)) {
1562 pr_err("--grow --data-offset not yet supported\n");
1563 return 1;
1564 }
1565
1566 if (s->size > 0 &&
1567 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1568 pr_err("cannot change component size at the same time as other changes.\n"
1569 " Change size first, then check data is intact before making other changes.\n");
1570 return 1;
1571 }
1572
1573 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1574 get_linux_version() < 2006032 &&
1575 !check_env("MDADM_FORCE_FEWER")) {
1576 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1577 " Please use a newer kernel\n");
1578 return 1;
1579 }
1580
1581 st = super_by_fd(fd, &subarray);
1582 if (!st) {
1583 pr_err("Unable to determine metadata format for %s\n", devname);
1584 return 1;
1585 }
1586 if (s->raiddisks > st->max_devs) {
1587 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1588 return 1;
1589 }
1590
1591 /* in the external case we need to check that the requested reshape is
1592 * supported, and perform an initial check that the container holds the
1593 * pre-requisite spare devices (mdmon owns final validation)
1594 */
1595 if (st->ss->external) {
1596 int rv;
1597
1598 if (subarray) {
1599 container = st->container_devnm;
1600 cfd = open_dev_excl(st->container_devnm);
1601 } else {
1602 container = st->devnm;
1603 close(fd);
1604 cfd = open_dev_excl(st->devnm);
1605 fd = cfd;
1606 }
1607 if (cfd < 0) {
1608 pr_err("Unable to open container for %s\n",
1609 devname);
1610 free(subarray);
1611 return 1;
1612 }
1613
1614 rv = st->ss->load_container(st, cfd, NULL);
1615
1616 if (rv) {
1617 pr_err("Cannot read superblock for %s\n",
1618 devname);
1619 free(subarray);
1620 return 1;
1621 }
1622
1623 /* check if operation is supported for metadata handler */
1624 if (st->ss->container_content) {
1625 struct mdinfo *cc = NULL;
1626 struct mdinfo *content = NULL;
1627
1628 cc = st->ss->container_content(st, subarray);
1629 for (content = cc; content ; content = content->next) {
1630 int allow_reshape = 1;
1631
1632 /* check if reshape is allowed based on metadata
1633 * indications stored in content.array.status
1634 */
1635 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1636 allow_reshape = 0;
1637 if (content->array.state
1638 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1639 allow_reshape = 0;
1640 if (!allow_reshape) {
1641 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1642 devname, container);
1643 sysfs_free(cc);
1644 free(subarray);
1645 return 1;
1646 }
1647 }
1648 sysfs_free(cc);
1649 }
1650 if (mdmon_running(container))
1651 st->update_tail = &st->updates;
1652 }
1653
1654 added_disks = 0;
1655 for (dv = devlist; dv; dv = dv->next)
1656 added_disks++;
1657 if (s->raiddisks > array.raid_disks &&
1658 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1659 !c->force) {
1660 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1661 " Use --force to over-ride this check.\n",
1662 s->raiddisks - array.raid_disks,
1663 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1664 array.spare_disks + added_disks);
1665 return 1;
1666 }
1667
1668 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1669 | GET_STATE | GET_VERSION);
1670 if (sra) {
1671 if (st->ss->external && subarray == NULL) {
1672 array.level = LEVEL_CONTAINER;
1673 sra->array.level = LEVEL_CONTAINER;
1674 }
1675 } else {
1676 pr_err("failed to read sysfs parameters for %s\n",
1677 devname);
1678 return 1;
1679 }
1680 frozen = freeze(st);
1681 if (frozen < -1) {
1682 /* freeze() already spewed the reason */
1683 sysfs_free(sra);
1684 return 1;
1685 } else if (frozen < 0) {
1686 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1687 sysfs_free(sra);
1688 return 1;
1689 }
1690
1691 /* ========= set size =============== */
1692 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1693 unsigned long long orig_size = get_component_size(fd)/2;
1694 unsigned long long min_csize;
1695 struct mdinfo *mdi;
1696 int raid0_takeover = 0;
1697
1698 if (orig_size == 0)
1699 orig_size = (unsigned) array.size;
1700
1701 if (orig_size == 0) {
1702 pr_err("Cannot set device size in this type of array.\n");
1703 rv = 1;
1704 goto release;
1705 }
1706
1707 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1708 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1709 rv = 1;
1710 goto release;
1711 }
1712 sync_metadata(st);
1713 if (st->ss->external) {
1714 /* metadata can have size limitation
1715 * update size value according to metadata information
1716 */
1717 struct mdinfo *sizeinfo =
1718 st->ss->container_content(st, subarray);
1719 if (sizeinfo) {
1720 unsigned long long new_size =
1721 sizeinfo->custom_array_size/2;
1722 int data_disks = get_data_disks(
1723 sizeinfo->array.level,
1724 sizeinfo->array.layout,
1725 sizeinfo->array.raid_disks);
1726 new_size /= data_disks;
1727 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size, new_size,
1728 new_size * data_disks);
1729 s->size = new_size;
1730 sysfs_free(sizeinfo);
1731 }
1732 }
1733
1734 /* Update the size of each member device in case
1735 * they have been resized. This will never reduce
1736 * below the current used-size. The "size" attribute
1737 * understands '0' to mean 'max'.
1738 */
1739 min_csize = 0;
1740 rv = 0;
1741 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1742 if (sysfs_set_num(sra, mdi, "size",
1743 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1744 /* Probably kernel refusing to let us
1745 * reduce the size - not an error.
1746 */
1747 break;
1748 }
1749 if (array.not_persistent == 0 &&
1750 array.major_version == 0 &&
1751 get_linux_version() < 3001000) {
1752 /* Dangerous to allow size to exceed 2TB */
1753 unsigned long long csize;
1754 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1755 if (csize >= 2ULL*1024*1024*1024)
1756 csize = 2ULL*1024*1024*1024;
1757 if ((min_csize == 0 || (min_csize
1758 > csize)))
1759 min_csize = csize;
1760 }
1761 }
1762 }
1763 if (rv) {
1764 pr_err("Cannot set size on array members.\n");
1765 goto size_change_error;
1766 }
1767 if (min_csize && s->size > min_csize) {
1768 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1769 rv = 1;
1770 goto size_change_error;
1771 }
1772 if (min_csize && s->size == MAX_SIZE) {
1773 /* Don't let the kernel choose a size - it will get
1774 * it wrong
1775 */
1776 pr_err("Limited v0.90 array to 2TB per device\n");
1777 s->size = min_csize;
1778 }
1779 if (st->ss->external) {
1780 if (sra->array.level == 0) {
1781 rv = sysfs_set_str(sra, NULL, "level",
1782 "raid5");
1783 if (!rv) {
1784 raid0_takeover = 1;
1785 /* get array parameters after takeover
1786 * to change one parameter at time only
1787 */
1788 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1789 }
1790 }
1791 /* make sure mdmon is
1792 * aware of the new level */
1793 if (!mdmon_running(st->container_devnm))
1794 start_mdmon(st->container_devnm);
1795 ping_monitor(container);
1796 if (mdmon_running(st->container_devnm) &&
1797 st->update_tail == NULL)
1798 st->update_tail = &st->updates;
1799 }
1800
1801 if (s->size == MAX_SIZE)
1802 s->size = 0;
1803 array.size = s->size;
1804 if (s->size & ~INT32_MAX) {
1805 /* got truncated to 32bit, write to
1806 * component_size instead
1807 */
1808 if (sra)
1809 rv = sysfs_set_num(sra, NULL,
1810 "component_size", s->size);
1811 else
1812 rv = -1;
1813 } else {
1814 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1815
1816 /* manage array size when it is managed externally
1817 */
1818 if ((rv == 0) && st->ss->external)
1819 rv = set_array_size(st, sra, sra->text_version);
1820 }
1821
1822 if (raid0_takeover) {
1823 /* do not recync non-existing parity,
1824 * we will drop it anyway
1825 */
1826 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1827 /* go back to raid0, drop parity disk
1828 */
1829 sysfs_set_str(sra, NULL, "level", "raid0");
1830 ioctl(fd, GET_ARRAY_INFO, &array);
1831 }
1832
1833 size_change_error:
1834 if (rv != 0) {
1835 int err = errno;
1836
1837 /* restore metadata */
1838 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1839 UnSet, NULL, devname,
1840 ROLLBACK_METADATA_CHANGES,
1841 c->verbose) == 0)
1842 sync_metadata(st);
1843 pr_err("Cannot set device size for %s: %s\n",
1844 devname, strerror(err));
1845 if (err == EBUSY &&
1846 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1847 cont_err("Bitmap must be removed before size can be changed\n");
1848 rv = 1;
1849 goto release;
1850 }
1851 if (s->assume_clean) {
1852 /* This will fail on kernels older than 3.0 unless
1853 * a backport has been arranged.
1854 */
1855 if (sra == NULL ||
1856 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1857 pr_err("--assume-clean not supported with --grow on this kernel\n");
1858 }
1859 ioctl(fd, GET_ARRAY_INFO, &array);
1860 s->size = get_component_size(fd)/2;
1861 if (s->size == 0)
1862 s->size = array.size;
1863 if (c->verbose >= 0) {
1864 if (s->size == orig_size)
1865 pr_err("component size of %s unchanged at %lluK\n",
1866 devname, s->size);
1867 else
1868 pr_err("component size of %s has been set to %lluK\n",
1869 devname, s->size);
1870 }
1871 changed = 1;
1872 } else if (array.level != LEVEL_CONTAINER) {
1873 s->size = get_component_size(fd)/2;
1874 if (s->size == 0)
1875 s->size = array.size;
1876 }
1877
1878 /* See if there is anything else to do */
1879 if ((s->level == UnSet || s->level == array.level) &&
1880 (s->layout_str == NULL) &&
1881 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1882 data_offset == INVALID_SECTORS &&
1883 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1884 /* Nothing more to do */
1885 if (!changed && c->verbose >= 0)
1886 pr_err("%s: no change requested\n",
1887 devname);
1888 goto release;
1889 }
1890
1891 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1892 * current implementation assumes that following conditions must be met:
1893 * - RAID10:
1894 * - far_copies == 1
1895 * - near_copies == 2
1896 */
1897 if ((s->level == 0 && array.level == 10 && sra &&
1898 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1899 (s->level == 0 && array.level == 1 && sra)) {
1900 int err;
1901 err = remove_disks_for_takeover(st, sra, array.layout);
1902 if (err) {
1903 dprintf("Array cannot be reshaped\n");
1904 if (cfd > -1)
1905 close(cfd);
1906 rv = 1;
1907 goto release;
1908 }
1909 /* Make sure mdmon has seen the device removal
1910 * and updated metadata before we continue with
1911 * level change
1912 */
1913 if (container)
1914 ping_monitor(container);
1915 }
1916
1917 memset(&info, 0, sizeof(info));
1918 info.array = array;
1919 sysfs_init(&info, fd, NULL);
1920 strcpy(info.text_version, sra->text_version);
1921 info.component_size = s->size*2;
1922 info.new_level = s->level;
1923 info.new_chunk = s->chunk * 1024;
1924 if (info.array.level == LEVEL_CONTAINER) {
1925 info.delta_disks = UnSet;
1926 info.array.raid_disks = s->raiddisks;
1927 } else if (s->raiddisks)
1928 info.delta_disks = s->raiddisks - info.array.raid_disks;
1929 else
1930 info.delta_disks = UnSet;
1931 if (s->layout_str == NULL) {
1932 info.new_layout = UnSet;
1933 if (info.array.level == 6 &&
1934 (info.new_level == 6 || info.new_level == UnSet) &&
1935 info.array.layout >= 16) {
1936 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
1937 cont_err("during the reshape, please specify --layout=preserve\n");
1938 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1939 rv = 1;
1940 goto release;
1941 }
1942 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1943 strcmp(s->layout_str, "normalize") == 0) {
1944 /* If we have a -6 RAID6 layout, remove the '-6'. */
1945 info.new_layout = UnSet;
1946 if (info.array.level == 6 && info.new_level == UnSet) {
1947 char l[40], *h;
1948 strcpy(l, map_num(r6layout, info.array.layout));
1949 h = strrchr(l, '-');
1950 if (h && strcmp(h, "-6") == 0) {
1951 *h = 0;
1952 info.new_layout = map_name(r6layout, l);
1953 }
1954 } else {
1955 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
1956 rv = 1;
1957 goto release;
1958 }
1959 } else if (strcmp(s->layout_str, "preserve") == 0) {
1960 /* This means that a non-standard RAID6 layout
1961 * is OK.
1962 * In particular:
1963 * - When reshape a RAID6 (e.g. adding a device)
1964 * which is in a non-standard layout, it is OK
1965 * to preserve that layout.
1966 * - When converting a RAID5 to RAID6, leave it in
1967 * the XXX-6 layout, don't re-layout.
1968 */
1969 if (info.array.level == 6 && info.new_level == UnSet)
1970 info.new_layout = info.array.layout;
1971 else if (info.array.level == 5 && info.new_level == 6) {
1972 char l[40];
1973 strcpy(l, map_num(r5layout, info.array.layout));
1974 strcat(l, "-6");
1975 info.new_layout = map_name(r6layout, l);
1976 } else {
1977 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
1978 rv = 1;
1979 goto release;
1980 }
1981 } else {
1982 int l = info.new_level;
1983 if (l == UnSet)
1984 l = info.array.level;
1985 switch (l) {
1986 case 5:
1987 info.new_layout = map_name(r5layout, s->layout_str);
1988 break;
1989 case 6:
1990 info.new_layout = map_name(r6layout, s->layout_str);
1991 break;
1992 case 10:
1993 info.new_layout = parse_layout_10(s->layout_str);
1994 break;
1995 case LEVEL_FAULTY:
1996 info.new_layout = parse_layout_faulty(s->layout_str);
1997 break;
1998 default:
1999 pr_err("layout not meaningful with this level\n");
2000 rv = 1;
2001 goto release;
2002 }
2003 if (info.new_layout == UnSet) {
2004 pr_err("layout %s not understood for this level\n",
2005 s->layout_str);
2006 rv = 1;
2007 goto release;
2008 }
2009 }
2010
2011 if (array.level == LEVEL_FAULTY) {
2012 if (s->level != UnSet && s->level != array.level) {
2013 pr_err("cannot change level of Faulty device\n");
2014 rv =1 ;
2015 }
2016 if (s->chunk) {
2017 pr_err("cannot set chunksize of Faulty device\n");
2018 rv =1 ;
2019 }
2020 if (s->raiddisks && s->raiddisks != 1) {
2021 pr_err("cannot set raid_disks of Faulty device\n");
2022 rv =1 ;
2023 }
2024 if (s->layout_str) {
2025 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2026 dprintf("Cannot get array information.\n");
2027 goto release;
2028 }
2029 array.layout = info.new_layout;
2030 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2031 pr_err("failed to set new layout\n");
2032 rv = 1;
2033 } else if (c->verbose >= 0)
2034 printf("layout for %s set to %d\n",
2035 devname, array.layout);
2036 }
2037 } else if (array.level == LEVEL_CONTAINER) {
2038 /* This change is to be applied to every array in the
2039 * container. This is only needed when the metadata imposes
2040 * restraints of the various arrays in the container.
2041 * Currently we only know that IMSM requires all arrays
2042 * to have the same number of devices so changing the
2043 * number of devices (On-Line Capacity Expansion) must be
2044 * performed at the level of the container
2045 */
2046 rv = reshape_container(container, devname, -1, st, &info,
2047 c->force, c->backup_file, c->verbose, 0, 0, 0);
2048 frozen = 0;
2049 } else {
2050 /* get spare devices from external metadata
2051 */
2052 if (st->ss->external) {
2053 struct mdinfo *info2;
2054
2055 info2 = st->ss->container_content(st, subarray);
2056 if (info2) {
2057 info.array.spare_disks =
2058 info2->array.spare_disks;
2059 sysfs_free(info2);
2060 }
2061 }
2062
2063 /* Impose these changes on a single array. First
2064 * check that the metadata is OK with the change. */
2065
2066 if (reshape_super(st, 0, info.new_level,
2067 info.new_layout, info.new_chunk,
2068 info.array.raid_disks, info.delta_disks,
2069 c->backup_file, devname, APPLY_METADATA_CHANGES,
2070 c->verbose)) {
2071 rv = 1;
2072 goto release;
2073 }
2074 sync_metadata(st);
2075 rv = reshape_array(container, fd, devname, st, &info, c->force,
2076 devlist, data_offset, c->backup_file, c->verbose,
2077 0, 0, 0);
2078 frozen = 0;
2079 }
2080 release:
2081 sysfs_free(sra);
2082 if (frozen > 0)
2083 unfreeze(st);
2084 return rv;
2085 }
2086
2087 /* verify_reshape_position()
2088 * Function checks if reshape position in metadata is not farther
2089 * than position in md.
2090 * Return value:
2091 * 0 : not valid sysfs entry
2092 * it can be caused by not started reshape, it should be started
2093 * by reshape array or raid0 array is before takeover
2094 * -1 : error, reshape position is obviously wrong
2095 * 1 : success, reshape progress correct or updated
2096 */
2097 static int verify_reshape_position(struct mdinfo *info, int level)
2098 {
2099 int ret_val = 0;
2100 char buf[40];
2101 int rv;
2102
2103 /* read sync_max, failure can mean raid0 array */
2104 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2105
2106 if (rv > 0) {
2107 char *ep;
2108 unsigned long long position = strtoull(buf, &ep, 0);
2109
2110 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2111 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2112 position *= get_data_disks(level,
2113 info->new_layout,
2114 info->array.raid_disks);
2115 if (info->reshape_progress < position) {
2116 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2117 info->reshape_progress, position);
2118 info->reshape_progress = position;
2119 ret_val = 1;
2120 } else if (info->reshape_progress > position) {
2121 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2122 position, info->reshape_progress);
2123 ret_val = -1;
2124 } else {
2125 dprintf("Reshape position in md and metadata are the same;");
2126 ret_val = 1;
2127 }
2128 }
2129 } else if (rv == 0) {
2130 /* for valid sysfs entry, 0-length content
2131 * should be indicated as error
2132 */
2133 ret_val = -1;
2134 }
2135
2136 return ret_val;
2137 }
2138
2139 static unsigned long long choose_offset(unsigned long long lo,
2140 unsigned long long hi,
2141 unsigned long long min,
2142 unsigned long long max)
2143 {
2144 /* Choose a new offset between hi and lo.
2145 * It must be between min and max, but
2146 * we would prefer something near the middle of hi/lo, and also
2147 * prefer to be aligned to a big power of 2.
2148 *
2149 * So we start with the middle, then for each bit,
2150 * starting at '1' and increasing, if it is set, we either
2151 * add it or subtract it if possible, preferring the option
2152 * which is furthest from the boundary.
2153 *
2154 * We stop once we get a 1MB alignment. As units are in sectors,
2155 * 1MB = 2*1024 sectors.
2156 */
2157 unsigned long long choice = (lo + hi) / 2;
2158 unsigned long long bit = 1;
2159
2160 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2161 unsigned long long bigger, smaller;
2162 if (! (bit & choice))
2163 continue;
2164 bigger = choice + bit;
2165 smaller = choice - bit;
2166 if (bigger > max && smaller < min)
2167 break;
2168 if (bigger > max)
2169 choice = smaller;
2170 else if (smaller < min)
2171 choice = bigger;
2172 else if (hi - bigger > smaller - lo)
2173 choice = bigger;
2174 else
2175 choice = smaller;
2176 }
2177 return choice;
2178 }
2179
2180 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2181 char *devname, int delta_disks,
2182 unsigned long long data_offset,
2183 unsigned long long min,
2184 int can_fallback)
2185 {
2186 struct mdinfo *sd;
2187 int dir = 0;
2188 int err = 0;
2189 unsigned long long before, after;
2190
2191 /* Need to find min space before and after so same is used
2192 * on all devices
2193 */
2194 before = UINT64_MAX;
2195 after = UINT64_MAX;
2196 for (sd = sra->devs; sd; sd = sd->next) {
2197 char *dn;
2198 int dfd;
2199 int rv;
2200 struct supertype *st2;
2201 struct mdinfo info2;
2202
2203 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2204 continue;
2205 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2206 dfd = dev_open(dn, O_RDONLY);
2207 if (dfd < 0) {
2208 pr_err("%s: cannot open component %s\n",
2209 devname, dn ? dn : "-unknown-");
2210 goto release;
2211 }
2212 st2 = dup_super(st);
2213 rv = st2->ss->load_super(st2,dfd, NULL);
2214 close(dfd);
2215 if (rv) {
2216 free(st2);
2217 pr_err("%s: cannot get superblock from %s\n",
2218 devname, dn);
2219 goto release;
2220 }
2221 st2->ss->getinfo_super(st2, &info2, NULL);
2222 st2->ss->free_super(st2);
2223 free(st2);
2224 if (info2.space_before == 0 &&
2225 info2.space_after == 0) {
2226 /* Metadata doesn't support data_offset changes */
2227 if (!can_fallback)
2228 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2229 devname);
2230 goto fallback;
2231 }
2232 if (before > info2.space_before)
2233 before = info2.space_before;
2234 if (after > info2.space_after)
2235 after = info2.space_after;
2236
2237 if (data_offset != INVALID_SECTORS) {
2238 if (dir == 0) {
2239 if (info2.data_offset == data_offset) {
2240 pr_err("%s: already has that data_offset\n",
2241 dn);
2242 goto release;
2243 }
2244 if (data_offset < info2.data_offset)
2245 dir = -1;
2246 else
2247 dir = 1;
2248 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2249 (data_offset >= info2.data_offset && dir == -1)) {
2250 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2251 dn);
2252 goto release;
2253 }
2254 }
2255 }
2256 if (before == UINT64_MAX)
2257 /* impossible really, there must be no devices */
2258 return 1;
2259
2260 for (sd = sra->devs; sd; sd = sd->next) {
2261 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2262 unsigned long long new_data_offset;
2263
2264 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2265 continue;
2266 if (delta_disks < 0) {
2267 /* Don't need any space as array is shrinking
2268 * just move data_offset up by min
2269 */
2270 if (data_offset == INVALID_SECTORS)
2271 new_data_offset = sd->data_offset + min;
2272 else {
2273 if (data_offset < sd->data_offset + min) {
2274 pr_err("--data-offset too small for %s\n",
2275 dn);
2276 goto release;
2277 }
2278 new_data_offset = data_offset;
2279 }
2280 } else if (delta_disks > 0) {
2281 /* need space before */
2282 if (before < min) {
2283 if (can_fallback)
2284 goto fallback;
2285 pr_err("Insufficient head-space for reshape on %s\n",
2286 dn);
2287 goto release;
2288 }
2289 if (data_offset == INVALID_SECTORS)
2290 new_data_offset = sd->data_offset - min;
2291 else {
2292 if (data_offset > sd->data_offset - min) {
2293 pr_err("--data-offset too large for %s\n",
2294 dn);
2295 goto release;
2296 }
2297 new_data_offset = data_offset;
2298 }
2299 } else {
2300 if (dir == 0) {
2301 /* can move up or down. If 'data_offset'
2302 * was set we would have already decided,
2303 * so just choose direction with most space.
2304 */
2305 if (before > after)
2306 dir = -1;
2307 else
2308 dir = 1;
2309 }
2310 sysfs_set_str(sra, NULL, "reshape_direction",
2311 dir == 1 ? "backwards" : "forwards");
2312 if (dir > 0) {
2313 /* Increase data offset */
2314 if (after < min) {
2315 if (can_fallback)
2316 goto fallback;
2317 pr_err("Insufficient tail-space for reshape on %s\n",
2318 dn);
2319 goto release;
2320 }
2321 if (data_offset != INVALID_SECTORS &&
2322 data_offset < sd->data_offset + min) {
2323 pr_err("--data-offset too small on %s\n",
2324 dn);
2325 goto release;
2326 }
2327 if (data_offset != INVALID_SECTORS)
2328 new_data_offset = data_offset;
2329 else
2330 new_data_offset = choose_offset(sd->data_offset,
2331 sd->data_offset + after,
2332 sd->data_offset + min,
2333 sd->data_offset + after);
2334 } else {
2335 /* Decrease data offset */
2336 if (before < min) {
2337 if (can_fallback)
2338 goto fallback;
2339 pr_err("insufficient head-room on %s\n",
2340 dn);
2341 goto release;
2342 }
2343 if (data_offset != INVALID_SECTORS &&
2344 data_offset < sd->data_offset - min) {
2345 pr_err("--data-offset too small on %s\n",
2346 dn);
2347 goto release;
2348 }
2349 if (data_offset != INVALID_SECTORS)
2350 new_data_offset = data_offset;
2351 else
2352 new_data_offset = choose_offset(sd->data_offset - before,
2353 sd->data_offset,
2354 sd->data_offset - before,
2355 sd->data_offset - min);
2356 }
2357 }
2358 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2359 if (err < 0 && errno == E2BIG) {
2360 /* try again after increasing data size to max */
2361 err = sysfs_set_num(sra, sd, "size", 0);
2362 if (err < 0 && errno == EINVAL &&
2363 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2364 /* some kernels have a bug where you cannot
2365 * use '0' on spare devices. */
2366 sysfs_set_num(sra, sd, "size",
2367 (sra->component_size + after)/2);
2368 }
2369 err = sysfs_set_num(sra, sd, "new_offset",
2370 new_data_offset);
2371 }
2372 if (err < 0) {
2373 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2374 pr_err("data-offset is too big for %s\n",
2375 dn);
2376 goto release;
2377 }
2378 if (sd == sra->devs &&
2379 (errno == ENOENT || errno == E2BIG))
2380 /* Early kernel, no 'new_offset' file,
2381 * or kernel doesn't like us.
2382 * For RAID5/6 this is not fatal
2383 */
2384 return 1;
2385 pr_err("Cannot set new_offset for %s\n",
2386 dn);
2387 break;
2388 }
2389 }
2390 return err;
2391 release:
2392 return -1;
2393 fallback:
2394 /* Just use a backup file */
2395 return 1;
2396 }
2397
2398 static int raid10_reshape(char *container, int fd, char *devname,
2399 struct supertype *st, struct mdinfo *info,
2400 struct reshape *reshape,
2401 unsigned long long data_offset,
2402 int force, int verbose)
2403 {
2404 /* Changing raid_disks, layout, chunksize or possibly
2405 * just data_offset for a RAID10.
2406 * We must always change data_offset. We change by at least
2407 * ->min_offset_change which is the largest of the old and new
2408 * chunk sizes.
2409 * If raid_disks is increasing, then data_offset must decrease
2410 * by at least this copy size.
2411 * If raid_disks is unchanged, data_offset must increase or
2412 * decrease by at least min_offset_change but preferably by much more.
2413 * We choose half of the available space.
2414 * If raid_disks is decreasing, data_offset must increase by
2415 * at least min_offset_change. To allow of this, component_size
2416 * must be decreased by the same amount.
2417 *
2418 * So we calculate the required minimum and direction, possibly
2419 * reduce the component_size, then iterate through the devices
2420 * and set the new_data_offset.
2421 * If that all works, we set chunk_size, layout, raid_disks, and start
2422 * 'reshape'
2423 */
2424 struct mdinfo *sra;
2425 unsigned long long min;
2426 int err = 0;
2427
2428 sra = sysfs_read(fd, NULL,
2429 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2430 );
2431 if (!sra) {
2432 pr_err("%s: Cannot get array details from sysfs\n",
2433 devname);
2434 goto release;
2435 }
2436 min = reshape->min_offset_change;
2437
2438 if (info->delta_disks)
2439 sysfs_set_str(sra, NULL, "reshape_direction",
2440 info->delta_disks < 0 ? "backwards" : "forwards");
2441 if (info->delta_disks < 0 &&
2442 info->space_after < min) {
2443 int rv = sysfs_set_num(sra, NULL, "component_size",
2444 (sra->component_size -
2445 min)/2);
2446 if (rv) {
2447 pr_err("cannot reduce component size\n");
2448 goto release;
2449 }
2450 }
2451 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2452 min, 0);
2453 if (err == 1) {
2454 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2455 cont_err("supported on this kernel\n");
2456 err = -1;
2457 }
2458 if (err < 0)
2459 goto release;
2460
2461 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2462 err = errno;
2463 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2464 err = errno;
2465 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2466 info->array.raid_disks + info->delta_disks) < 0)
2467 err = errno;
2468 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2469 err = errno;
2470 if (err) {
2471 pr_err("Cannot set array shape for %s\n",
2472 devname);
2473 if (err == EBUSY &&
2474 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2475 cont_err(" Bitmap must be removed before shape can be changed\n");
2476 goto release;
2477 }
2478 sysfs_free(sra);
2479 return 0;
2480 release:
2481 sysfs_free(sra);
2482 return 1;
2483 }
2484
2485 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2486 {
2487 struct mdinfo *sra, *sd;
2488 /* Initialisation to silence compiler warning */
2489 unsigned long long min_space_before = 0, min_space_after = 0;
2490 int first = 1;
2491
2492 sra = sysfs_read(fd, NULL, GET_DEVS);
2493 if (!sra)
2494 return;
2495 for (sd = sra->devs; sd; sd = sd->next) {
2496 char *dn;
2497 int dfd;
2498 struct supertype *st2;
2499 struct mdinfo info2;
2500
2501 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2502 continue;
2503 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2504 dfd = dev_open(dn, O_RDONLY);
2505 if (dfd < 0)
2506 break;
2507 st2 = dup_super(st);
2508 if (st2->ss->load_super(st2,dfd, NULL)) {
2509 close(dfd);
2510 free(st2);
2511 break;
2512 }
2513 close(dfd);
2514 st2->ss->getinfo_super(st2, &info2, NULL);
2515 st2->ss->free_super(st2);
2516 free(st2);
2517 if (first ||
2518 min_space_before > info2.space_before)
2519 min_space_before = info2.space_before;
2520 if (first ||
2521 min_space_after > info2.space_after)
2522 min_space_after = info2.space_after;
2523 first = 0;
2524 }
2525 if (sd == NULL && !first) {
2526 info->space_after = min_space_after;
2527 info->space_before = min_space_before;
2528 }
2529 sysfs_free(sra);
2530 }
2531
2532 static void update_cache_size(char *container, struct mdinfo *sra,
2533 struct mdinfo *info,
2534 int disks, unsigned long long blocks)
2535 {
2536 /* Check that the internal stripe cache is
2537 * large enough, or it won't work.
2538 * It must hold at least 4 stripes of the larger
2539 * chunk size
2540 */
2541 unsigned long cache;
2542 cache = max(info->array.chunk_size, info->new_chunk);
2543 cache *= 4; /* 4 stripes minimum */
2544 cache /= 512; /* convert to sectors */
2545 /* make sure there is room for 'blocks' with a bit to spare */
2546 if (cache < 16 + blocks / disks)
2547 cache = 16 + blocks / disks;
2548 cache /= (4096/512); /* Convert from sectors to pages */
2549
2550 if (sra->cache_size < cache)
2551 subarray_set_num(container, sra, "stripe_cache_size",
2552 cache+1);
2553 }
2554
2555 static int impose_reshape(struct mdinfo *sra,
2556 struct mdinfo *info,
2557 struct supertype *st,
2558 int fd,
2559 int restart,
2560 char *devname, char *container,
2561 struct reshape *reshape)
2562 {
2563 struct mdu_array_info_s array;
2564
2565 sra->new_chunk = info->new_chunk;
2566
2567 if (restart) {
2568 /* for external metadata checkpoint saved by mdmon can be lost
2569 * or missed /due to e.g. crash/. Check if md is not during
2570 * restart farther than metadata points to.
2571 * If so, this means metadata information is obsolete.
2572 */
2573 if (st->ss->external)
2574 verify_reshape_position(info, reshape->level);
2575 sra->reshape_progress = info->reshape_progress;
2576 } else {
2577 sra->reshape_progress = 0;
2578 if (reshape->after.data_disks < reshape->before.data_disks)
2579 /* start from the end of the new array */
2580 sra->reshape_progress = (sra->component_size
2581 * reshape->after.data_disks);
2582 }
2583
2584 ioctl(fd, GET_ARRAY_INFO, &array);
2585 if (info->array.chunk_size == info->new_chunk &&
2586 reshape->before.layout == reshape->after.layout &&
2587 st->ss->external == 0) {
2588 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2589 array.raid_disks = reshape->after.data_disks + reshape->parity;
2590 if (!restart &&
2591 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2592 int err = errno;
2593
2594 pr_err("Cannot set device shape for %s: %s\n",
2595 devname, strerror(errno));
2596
2597 if (err == EBUSY &&
2598 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2599 cont_err("Bitmap must be removed before shape can be changed\n");
2600
2601 goto release;
2602 }
2603 } else if (!restart) {
2604 /* set them all just in case some old 'new_*' value
2605 * persists from some earlier problem.
2606 */
2607 int err = 0;
2608 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2609 err = errno;
2610 if (!err && sysfs_set_num(sra, NULL, "layout",
2611 reshape->after.layout) < 0)
2612 err = errno;
2613 if (!err && subarray_set_num(container, sra, "raid_disks",
2614 reshape->after.data_disks +
2615 reshape->parity) < 0)
2616 err = errno;
2617 if (err) {
2618 pr_err("Cannot set device shape for %s\n",
2619 devname);
2620
2621 if (err == EBUSY &&
2622 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2623 cont_err("Bitmap must be removed before shape can be changed\n");
2624 goto release;
2625 }
2626 }
2627 return 0;
2628 release:
2629 return -1;
2630 }
2631
2632 static int impose_level(int fd, int level, char *devname, int verbose)
2633 {
2634 char *c;
2635 struct mdu_array_info_s array;
2636 struct mdinfo info;
2637 sysfs_init(&info, fd, NULL);
2638
2639 ioctl(fd, GET_ARRAY_INFO, &array);
2640 if (level == 0 &&
2641 (array.level >= 4 && array.level <= 6)) {
2642 /* To convert to RAID0 we need to fail and
2643 * remove any non-data devices. */
2644 int found = 0;
2645 int d;
2646 int data_disks = array.raid_disks - 1;
2647 if (array.level == 6)
2648 data_disks -= 1;
2649 if (array.level == 5 &&
2650 array.layout != ALGORITHM_PARITY_N)
2651 return -1;
2652 if (array.level == 6 &&
2653 array.layout != ALGORITHM_PARITY_N_6)
2654 return -1;
2655 sysfs_set_str(&info, NULL,"sync_action", "idle");
2656 /* First remove any spares so no recovery starts */
2657 for (d = 0, found = 0;
2658 d < MAX_DISKS && found < array.nr_disks;
2659 d++) {
2660 mdu_disk_info_t disk;
2661 disk.number = d;
2662 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2663 continue;
2664 if (disk.major == 0 && disk.minor == 0)
2665 continue;
2666 found++;
2667 if ((disk.state & (1 << MD_DISK_ACTIVE))
2668 && disk.raid_disk < data_disks)
2669 /* keep this */
2670 continue;
2671 ioctl(fd, HOT_REMOVE_DISK,
2672 makedev(disk.major, disk.minor));
2673 }
2674 /* Now fail anything left */
2675 ioctl(fd, GET_ARRAY_INFO, &array);
2676 for (d = 0, found = 0;
2677 d < MAX_DISKS && found < array.nr_disks;
2678 d++) {
2679 int cnt;
2680 mdu_disk_info_t disk;
2681 disk.number = d;
2682 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2683 continue;
2684 if (disk.major == 0 && disk.minor == 0)
2685 continue;
2686 found++;
2687 if ((disk.state & (1 << MD_DISK_ACTIVE))
2688 && disk.raid_disk < data_disks)
2689 /* keep this */
2690 continue;
2691 ioctl(fd, SET_DISK_FAULTY,
2692 makedev(disk.major, disk.minor));
2693 cnt = 5;
2694 while (ioctl(fd, HOT_REMOVE_DISK,
2695 makedev(disk.major, disk.minor)) < 0
2696 && errno == EBUSY
2697 && cnt--) {
2698 usleep(10000);
2699 }
2700 }
2701 }
2702 c = map_num(pers, level);
2703 if (c) {
2704 int err = sysfs_set_str(&info, NULL, "level", c);
2705 if (err) {
2706 err = errno;
2707 pr_err("%s: could not set level to %s\n",
2708 devname, c);
2709 if (err == EBUSY &&
2710 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2711 cont_err("Bitmap must be removed before level can be changed\n");
2712 return err;
2713 }
2714 if (verbose >= 0)
2715 pr_err("level of %s changed to %s\n",
2716 devname, c);
2717 }
2718 return 0;
2719 }
2720
2721 int sigterm = 0;
2722 static void catch_term(int sig)
2723 {
2724 sigterm = 1;
2725 }
2726
2727 static int continue_via_systemd(char *devnm)
2728 {
2729 int skipped, i, pid, status;
2730 char pathbuf[1024];
2731 /* In a systemd/udev world, it is best to get systemd to
2732 * run "mdadm --grow --continue" rather than running in the
2733 * background.
2734 */
2735 switch(fork()) {
2736 case 0:
2737 /* FIXME yuk. CLOSE_EXEC?? */
2738 skipped = 0;
2739 for (i = 3; skipped < 20; i++)
2740 if (close(i) < 0)
2741 skipped++;
2742 else
2743 skipped = 0;
2744
2745 /* Don't want to see error messages from
2746 * systemctl. If the service doesn't exist,
2747 * we fork ourselves.
2748 */
2749 close(2);
2750 open("/dev/null", O_WRONLY);
2751 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2752 devnm);
2753 status = execl("/usr/bin/systemctl", "systemctl",
2754 "start",
2755 pathbuf, NULL);
2756 status = execl("/bin/systemctl", "systemctl", "start",
2757 pathbuf, NULL);
2758 exit(1);
2759 case -1: /* Just do it ourselves. */
2760 break;
2761 default: /* parent - good */
2762 pid = wait(&status);
2763 if (pid >= 0 && status == 0)
2764 return 1;
2765 }
2766 return 0;
2767 }
2768
2769 static int reshape_array(char *container, int fd, char *devname,
2770 struct supertype *st, struct mdinfo *info,
2771 int force, struct mddev_dev *devlist,
2772 unsigned long long data_offset,
2773 char *backup_file, int verbose, int forked,
2774 int restart, int freeze_reshape)
2775 {
2776 struct reshape reshape;
2777 int spares_needed;
2778 char *msg;
2779 int orig_level = UnSet;
2780 int odisks;
2781 int delayed;
2782
2783 struct mdu_array_info_s array;
2784 char *c;
2785
2786 struct mddev_dev *dv;
2787 int added_disks;
2788
2789 int *fdlist = NULL;
2790 unsigned long long *offsets = NULL;
2791 int d;
2792 int nrdisks;
2793 int err;
2794 unsigned long blocks;
2795 unsigned long long array_size;
2796 int done;
2797 struct mdinfo *sra = NULL;
2798 char buf[20];
2799
2800 /* when reshaping a RAID0, the component_size might be zero.
2801 * So try to fix that up.
2802 */
2803 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2804 dprintf("Cannot get array information.\n");
2805 goto release;
2806 }
2807 if (array.level == 0 && info->component_size == 0) {
2808 get_dev_size(fd, NULL, &array_size);
2809 info->component_size = array_size / array.raid_disks;
2810 }
2811
2812 if (array.level == 10)
2813 /* Need space_after info */
2814 get_space_after(fd, st, info);
2815
2816 if (info->reshape_active) {
2817 int new_level = info->new_level;
2818 info->new_level = UnSet;
2819 if (info->delta_disks > 0)
2820 info->array.raid_disks -= info->delta_disks;
2821 msg = analyse_change(devname, info, &reshape);
2822 info->new_level = new_level;
2823 if (info->delta_disks > 0)
2824 info->array.raid_disks += info->delta_disks;
2825 if (!restart)
2826 /* Make sure the array isn't read-only */
2827 ioctl(fd, RESTART_ARRAY_RW, 0);
2828 } else
2829 msg = analyse_change(devname, info, &reshape);
2830 if (msg) {
2831 /* if msg == "", error has already been printed */
2832 if (msg[0])
2833 pr_err("%s\n", msg);
2834 goto release;
2835 }
2836 if (restart &&
2837 (reshape.level != info->array.level ||
2838 reshape.before.layout != info->array.layout ||
2839 reshape.before.data_disks + reshape.parity
2840 != info->array.raid_disks - max(0, info->delta_disks))) {
2841 pr_err("reshape info is not in native format - cannot continue.\n");
2842 goto release;
2843 }
2844
2845 if (st->ss->external && restart && (info->reshape_progress == 0) &&
2846 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
2847 (strncmp(buf, "reshape", 7) == 0))) {
2848 /* When reshape is restarted from '0', very begin of array
2849 * it is possible that for external metadata reshape and array
2850 * configuration doesn't happen.
2851 * Check if md has the same opinion, and reshape is restarted
2852 * from 0. If so, this is regular reshape start after reshape
2853 * switch in metadata to next array only.
2854 */
2855 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2856 (info->reshape_progress == 0))
2857 restart = 0;
2858 }
2859 if (restart) {
2860 /* reshape already started. just skip to monitoring the reshape */
2861 if (reshape.backup_blocks == 0)
2862 return 0;
2863 if (restart & RESHAPE_NO_BACKUP)
2864 return 0;
2865
2866 /* Need 'sra' down at 'started:' */
2867 sra = sysfs_read(fd, NULL,
2868 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2869 GET_CACHE);
2870 if (!sra) {
2871 pr_err("%s: Cannot get array details from sysfs\n",
2872 devname);
2873 goto release;
2874 }
2875
2876 if (!backup_file)
2877 backup_file = locate_backup(sra->sys_name);
2878
2879 goto started;
2880 }
2881 /* The container is frozen but the array may not be.
2882 * So freeze the array so spares don't get put to the wrong use
2883 * FIXME there should probably be a cleaner separation between
2884 * freeze_array and freeze_container.
2885 */
2886 sysfs_freeze_array(info);
2887 /* Check we have enough spares to not be degraded */
2888 added_disks = 0;
2889 for (dv = devlist; dv ; dv=dv->next)
2890 added_disks++;
2891 spares_needed = max(reshape.before.data_disks,
2892 reshape.after.data_disks)
2893 + reshape.parity - array.raid_disks;
2894
2895 if (!force &&
2896 info->new_level > 1 && info->array.level > 1 &&
2897 spares_needed > info->array.spare_disks + added_disks) {
2898 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2899 " Use --force to over-ride this check.\n",
2900 spares_needed,
2901 spares_needed == 1 ? "" : "s",
2902 info->array.spare_disks + added_disks);
2903 goto release;
2904 }
2905 /* Check we have enough spares to not fail */
2906 spares_needed = max(reshape.before.data_disks,
2907 reshape.after.data_disks)
2908 - array.raid_disks;
2909 if ((info->new_level > 1 || info->new_level == 0) &&
2910 spares_needed > info->array.spare_disks +added_disks) {
2911 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2912 spares_needed,
2913 spares_needed == 1 ? "" : "s",
2914 info->array.spare_disks + added_disks);
2915 goto release;
2916 }
2917
2918 if (reshape.level != array.level) {
2919 int err = impose_level(fd, reshape.level, devname, verbose);
2920 if (err)
2921 goto release;
2922 info->new_layout = UnSet; /* after level change,
2923 * layout is meaningless */
2924 orig_level = array.level;
2925 sysfs_freeze_array(info);
2926
2927 if (reshape.level > 0 && st->ss->external) {
2928 /* make sure mdmon is aware of the new level */
2929 if (mdmon_running(container))
2930 flush_mdmon(container);
2931
2932 if (!mdmon_running(container))
2933 start_mdmon(container);
2934 ping_monitor(container);
2935 if (mdmon_running(container) &&
2936 st->update_tail == NULL)
2937 st->update_tail = &st->updates;
2938 }
2939 }
2940 /* ->reshape_super might have chosen some spares from the
2941 * container that it wants to be part of the new array.
2942 * We can collect them with ->container_content and give
2943 * them to the kernel.
2944 */
2945 if (st->ss->reshape_super && st->ss->container_content) {
2946 char *subarray = strchr(info->text_version+1, '/')+1;
2947 struct mdinfo *info2 =
2948 st->ss->container_content(st, subarray);
2949 struct mdinfo *d;
2950
2951 if (info2) {
2952 sysfs_init(info2, fd, st->devnm);
2953 /* When increasing number of devices, we need to set
2954 * new raid_disks before adding these, or they might
2955 * be rejected.
2956 */
2957 if (reshape.backup_blocks &&
2958 reshape.after.data_disks > reshape.before.data_disks)
2959 subarray_set_num(container, info2, "raid_disks",
2960 reshape.after.data_disks +
2961 reshape.parity);
2962 for (d = info2->devs; d; d = d->next) {
2963 if (d->disk.state == 0 &&
2964 d->disk.raid_disk >= 0) {
2965 /* This is a spare that wants to
2966 * be part of the array.
2967 */
2968 add_disk(fd, st, info2, d);
2969 }
2970 }
2971 sysfs_free(info2);
2972 }
2973 }
2974 /* We might have been given some devices to add to the
2975 * array. Now that the array has been changed to the right
2976 * level and frozen, we can safely add them.
2977 */
2978 if (devlist)
2979 Manage_subdevs(devname, fd, devlist, verbose,
2980 0,NULL, 0);
2981
2982 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
2983 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
2984 if (reshape.backup_blocks == 0) {
2985 /* No restriping needed, but we might need to impose
2986 * some more changes: layout, raid_disks, chunk_size
2987 */
2988 /* read current array info */
2989 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2990 dprintf("Cannot get array information.\n");
2991 goto release;
2992 }
2993 /* compare current array info with new values and if
2994 * it is different update them to new */
2995 if (info->new_layout != UnSet &&
2996 info->new_layout != array.layout) {
2997 array.layout = info->new_layout;
2998 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2999 pr_err("failed to set new layout\n");
3000 goto release;
3001 } else if (verbose >= 0)
3002 printf("layout for %s set to %d\n",
3003 devname, array.layout);
3004 }
3005 if (info->delta_disks != UnSet &&
3006 info->delta_disks != 0 &&
3007 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3008 array.raid_disks += info->delta_disks;
3009 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3010 pr_err("failed to set raid disks\n");
3011 goto release;
3012 } else if (verbose >= 0) {
3013 printf("raid_disks for %s set to %d\n",
3014 devname, array.raid_disks);
3015 }
3016 }
3017 if (info->new_chunk != 0 &&
3018 info->new_chunk != array.chunk_size) {
3019 if (sysfs_set_num(info, NULL,
3020 "chunk_size", info->new_chunk) != 0) {
3021 pr_err("failed to set chunk size\n");
3022 goto release;
3023 } else if (verbose >= 0)
3024 printf("chunk size for %s set to %d\n",
3025 devname, array.chunk_size);
3026 }
3027 unfreeze(st);
3028 return 0;
3029 }
3030
3031 /*
3032 * There are three possibilities.
3033 * 1/ The array will shrink.
3034 * We need to ensure the reshape will pause before reaching
3035 * the 'critical section'. We also need to fork and wait for
3036 * that to happen. When it does we
3037 * suspend/backup/complete/unfreeze
3038 *
3039 * 2/ The array will not change size.
3040 * This requires that we keep a backup of a sliding window
3041 * so that we can restore data after a crash. So we need
3042 * to fork and monitor progress.
3043 * In future we will allow the data_offset to change, so
3044 * a sliding backup becomes unnecessary.
3045 *
3046 * 3/ The array will grow. This is relatively easy.
3047 * However the kernel's restripe routines will cheerfully
3048 * overwrite some early data before it is safe. So we
3049 * need to make a backup of the early parts of the array
3050 * and be ready to restore it if rebuild aborts very early.
3051 * For externally managed metadata, we still need a forked
3052 * child to monitor the reshape and suspend IO over the region
3053 * that is being reshaped.
3054 *
3055 * We backup data by writing it to one spare, or to a
3056 * file which was given on command line.
3057 *
3058 * In each case, we first make sure that storage is available
3059 * for the required backup.
3060 * Then we:
3061 * - request the shape change.
3062 * - fork to handle backup etc.
3063 */
3064 /* Check that we can hold all the data */
3065 get_dev_size(fd, NULL, &array_size);
3066 if (reshape.new_size < (array_size/512)) {
3067 pr_err("this change will reduce the size of the array.\n"
3068 " use --grow --array-size first to truncate array.\n"
3069 " e.g. mdadm --grow %s --array-size %llu\n",
3070 devname, reshape.new_size/2);
3071 goto release;
3072 }
3073
3074 if (array.level == 10) {
3075 /* Reshaping RAID10 does not require any data backup by
3076 * user-space. Instead it requires that the data_offset
3077 * is changed to avoid the need for backup.
3078 * So this is handled very separately
3079 */
3080 if (restart)
3081 /* Nothing to do. */
3082 return 0;
3083 return raid10_reshape(container, fd, devname, st, info,
3084 &reshape, data_offset,
3085 force, verbose);
3086 }
3087 sra = sysfs_read(fd, NULL,
3088 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3089 GET_CACHE);
3090 if (!sra) {
3091 pr_err("%s: Cannot get array details from sysfs\n",
3092 devname);
3093 goto release;
3094 }
3095
3096 if (!backup_file)
3097 switch(set_new_data_offset(sra, st, devname,
3098 reshape.after.data_disks - reshape.before.data_disks,
3099 data_offset,
3100 reshape.min_offset_change, 1)) {
3101 case -1:
3102 goto release;
3103 case 0:
3104 /* Updated data_offset, so it's easy now */
3105 update_cache_size(container, sra, info,
3106 min(reshape.before.data_disks,
3107 reshape.after.data_disks),
3108 reshape.backup_blocks);
3109
3110 /* Right, everything seems fine. Let's kick things off.
3111 */
3112 sync_metadata(st);
3113
3114 if (impose_reshape(sra, info, st, fd, restart,
3115 devname, container, &reshape) < 0)
3116 goto release;
3117 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3118 struct mdinfo *sd;
3119 if (errno != EINVAL) {
3120 pr_err("Failed to initiate reshape!\n");
3121 goto release;
3122 }
3123 /* revert data_offset and try the old way */
3124 for (sd = sra->devs; sd; sd = sd->next) {
3125 sysfs_set_num(sra, sd, "new_offset",
3126 sd->data_offset);
3127 sysfs_set_str(sra, NULL, "reshape_direction",
3128 "forwards");
3129 }
3130 break;
3131 }
3132 if (info->new_level == reshape.level)
3133 return 0;
3134 /* need to adjust level when reshape completes */
3135 switch(fork()) {
3136 case -1: /* ignore error, but don't wait */
3137 return 0;
3138 default: /* parent */
3139 return 0;
3140 case 0:
3141 map_fork();
3142 break;
3143 }
3144 close(fd);
3145 wait_reshape(sra);
3146 fd = open_dev(sra->sys_name);
3147 if (fd >= 0)
3148 impose_level(fd, info->new_level, devname, verbose);
3149 return 0;
3150 case 1: /* Couldn't set data_offset, try the old way */
3151 if (data_offset != INVALID_SECTORS) {
3152 pr_err("Cannot update data_offset on this array\n");
3153 goto release;
3154 }
3155 break;
3156 }
3157
3158 started:
3159 /* Decide how many blocks (sectors) for a reshape
3160 * unit. The number we have so far is just a minimum
3161 */
3162 blocks = reshape.backup_blocks;
3163 if (reshape.before.data_disks ==
3164 reshape.after.data_disks) {
3165 /* Make 'blocks' bigger for better throughput, but
3166 * not so big that we reject it below.
3167 * Try for 16 megabytes
3168 */
3169 while (blocks * 32 < sra->component_size &&
3170 blocks < 16*1024*2)
3171 blocks *= 2;
3172 } else
3173 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3174
3175 if (blocks >= sra->component_size/2) {
3176 pr_err("%s: Something wrong - reshape aborted\n",
3177 devname);
3178 goto release;
3179 }
3180
3181 /* Now we need to open all these devices so we can read/write.
3182 */
3183 nrdisks = max(reshape.before.data_disks,
3184 reshape.after.data_disks) + reshape.parity
3185 + sra->array.spare_disks;
3186 fdlist = xcalloc((1+nrdisks), sizeof(int));
3187 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3188
3189 odisks = reshape.before.data_disks + reshape.parity;
3190 d = reshape_prepare_fdlist(devname, sra, odisks,
3191 nrdisks, blocks, backup_file,
3192 fdlist, offsets);
3193 if (d < odisks) {
3194 goto release;
3195 }
3196 if ((st->ss->manage_reshape == NULL) ||
3197 (st->ss->recover_backup == NULL)) {
3198 if (backup_file == NULL) {
3199 if (reshape.after.data_disks <=
3200 reshape.before.data_disks) {
3201 pr_err("%s: Cannot grow - need backup-file\n",
3202 devname);
3203 pr_err(" Please provide one with \"--backup=...\"\n");
3204 goto release;
3205 } else if (d == odisks) {
3206 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3207 goto release;
3208 }
3209 } else {
3210 if (!reshape_open_backup_file(backup_file, fd, devname,
3211 (signed)blocks,
3212 fdlist+d, offsets+d,
3213 sra->sys_name,
3214 restart)) {
3215 goto release;
3216 }
3217 d++;
3218 }
3219 }
3220
3221 update_cache_size(container, sra, info,
3222 min(reshape.before.data_disks, reshape.after.data_disks),
3223 blocks);
3224
3225 /* Right, everything seems fine. Let's kick things off.
3226 * If only changing raid_disks, use ioctl, else use
3227 * sysfs.
3228 */
3229 sync_metadata(st);
3230
3231 if (impose_reshape(sra, info, st, fd, restart,
3232 devname, container, &reshape) < 0)
3233 goto release;
3234
3235 err = start_reshape(sra, restart, reshape.before.data_disks,
3236 reshape.after.data_disks);
3237 if (err) {
3238 pr_err("Cannot %s reshape for %s\n",
3239 restart ? "continue" : "start",
3240 devname);
3241 goto release;
3242 }
3243 if (restart)
3244 sysfs_set_str(sra, NULL, "array_state", "active");
3245 if (freeze_reshape) {
3246 free(fdlist);
3247 free(offsets);
3248 sysfs_free(sra);
3249 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3250 sra->reshape_progress);
3251 return 1;
3252 }
3253
3254 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3255 if (continue_via_systemd(container ?: sra->sys_name)) {
3256 free(fdlist);
3257 free(offsets);
3258 sysfs_free(sra);
3259 return 0;
3260 }
3261
3262 /* Now we just need to kick off the reshape and watch, while
3263 * handling backups of the data...
3264 * This is all done by a forked background process.
3265 */
3266 switch(forked ? 0 : fork()) {
3267 case -1:
3268 pr_err("Cannot run child to monitor reshape: %s\n",
3269 strerror(errno));
3270 abort_reshape(sra);
3271 goto release;
3272 default:
3273 free(fdlist);
3274 free(offsets);
3275 sysfs_free(sra);
3276 return 0;
3277 case 0:
3278 map_fork();
3279 break;
3280 }
3281
3282 /* If another array on the same devices is busy, the
3283 * reshape will wait for them. This would mean that
3284 * the first section that we suspend will stay suspended
3285 * for a long time. So check on that possibility
3286 * by looking for "DELAYED" in /proc/mdstat, and if found,
3287 * wait a while
3288 */
3289 do {
3290 struct mdstat_ent *mds, *m;
3291 delayed = 0;
3292 mds = mdstat_read(1, 0);
3293 for (m = mds; m; m = m->next)
3294 if (strcmp(m->devnm, sra->sys_name) == 0) {
3295 if (m->resync &&
3296 m->percent == RESYNC_DELAYED)
3297 delayed = 1;
3298 if (m->resync == 0)
3299 /* Haven't started the reshape thread
3300 * yet, wait a bit
3301 */
3302 delayed = 2;
3303 break;
3304 }
3305 free_mdstat(mds);
3306 if (delayed == 1 && get_linux_version() < 3007000) {
3307 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3308 " You might experience problems until other reshapes complete.\n");
3309 delayed = 0;
3310 }
3311 if (delayed)
3312 mdstat_wait(30 - (delayed-1) * 25);
3313 } while (delayed);
3314 mdstat_close();
3315 close(fd);
3316 if (check_env("MDADM_GROW_VERIFY"))
3317 fd = open(devname, O_RDONLY | O_DIRECT);
3318 else
3319 fd = -1;
3320 mlockall(MCL_FUTURE);
3321
3322 signal(SIGTERM, catch_term);
3323
3324 if (st->ss->external) {
3325 /* metadata handler takes it from here */
3326 done = st->ss->manage_reshape(
3327 fd, sra, &reshape, st, blocks,
3328 fdlist, offsets,
3329 d - odisks, fdlist+odisks,
3330 offsets+odisks);
3331 } else
3332 done = child_monitor(
3333 fd, sra, &reshape, st, blocks,
3334 fdlist, offsets,
3335 d - odisks, fdlist+odisks,
3336 offsets+odisks);
3337
3338 free(fdlist);
3339 free(offsets);
3340
3341 if (backup_file && done) {
3342 char *bul;
3343 bul = make_backup(sra->sys_name);
3344 if (bul) {
3345 char buf[1024];
3346 int l = readlink(bul, buf, sizeof(buf) - 1);
3347 if (l > 0) {
3348 buf[l]=0;
3349 unlink(buf);
3350 }
3351 unlink(bul);
3352 free(bul);
3353 }
3354 unlink(backup_file);
3355 }
3356 if (!done) {
3357 abort_reshape(sra);
3358 goto out;
3359 }
3360
3361 if (!st->ss->external &&
3362 !(reshape.before.data_disks != reshape.after.data_disks
3363 && info->custom_array_size) &&
3364 info->new_level == reshape.level &&
3365 !forked) {
3366 /* no need to wait for the reshape to finish as
3367 * there is nothing more to do.
3368 */
3369 sysfs_free(sra);
3370 exit(0);
3371 }
3372 wait_reshape(sra);
3373
3374 if (st->ss->external) {
3375 /* Re-load the metadata as much could have changed */
3376 int cfd = open_dev(st->container_devnm);
3377 if (cfd >= 0) {
3378 flush_mdmon(container);
3379 st->ss->free_super(st);
3380 st->ss->load_container(st, cfd, container);
3381 close(cfd);
3382 }
3383 }
3384
3385 /* set new array size if required customer_array_size is used
3386 * by this metadata.
3387 */
3388 if (reshape.before.data_disks !=
3389 reshape.after.data_disks &&
3390 info->custom_array_size)
3391 set_array_size(st, info, info->text_version);
3392
3393 if (info->new_level != reshape.level) {
3394 if (fd < 0)
3395 fd = open(devname, O_RDONLY);
3396 impose_level(fd, info->new_level, devname, verbose);
3397 close(fd);
3398 if (info->new_level == 0)
3399 st->update_tail = NULL;
3400 }
3401 out:
3402 sysfs_free(sra);
3403 if (forked)
3404 return 0;
3405 unfreeze(st);
3406 exit(0);
3407
3408 release:
3409 free(fdlist);
3410 free(offsets);
3411 if (orig_level != UnSet && sra) {
3412 c = map_num(pers, orig_level);
3413 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3414 pr_err("aborting level change\n");
3415 }
3416 sysfs_free(sra);
3417 if (!forked)
3418 unfreeze(st);
3419 return 1;
3420 }
3421
3422 /* mdfd handle is passed to be closed in child process (after fork).
3423 */
3424 int reshape_container(char *container, char *devname,
3425 int mdfd,
3426 struct supertype *st,
3427 struct mdinfo *info,
3428 int force,
3429 char *backup_file, int verbose,
3430 int forked, int restart, int freeze_reshape)
3431 {
3432 struct mdinfo *cc = NULL;
3433 int rv = restart;
3434 char last_devnm[32] = "";
3435
3436 /* component_size is not meaningful for a container,
3437 * so pass '0' meaning 'no change'
3438 */
3439 if (!restart &&
3440 reshape_super(st, 0, info->new_level,
3441 info->new_layout, info->new_chunk,
3442 info->array.raid_disks, info->delta_disks,
3443 backup_file, devname, APPLY_METADATA_CHANGES,
3444 verbose)) {
3445 unfreeze(st);
3446 return 1;
3447 }
3448
3449 sync_metadata(st);
3450
3451 /* ping monitor to be sure that update is on disk
3452 */
3453 ping_monitor(container);
3454
3455 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3456 if (continue_via_systemd(container))
3457 return 0;
3458
3459 switch (forked ? 0 : fork()) {
3460 case -1: /* error */
3461 perror("Cannot fork to complete reshape\n");
3462 unfreeze(st);
3463 return 1;
3464 default: /* parent */
3465 if (!freeze_reshape)
3466 printf("%s: multi-array reshape continues in background\n", Name);
3467 return 0;
3468 case 0: /* child */
3469 map_fork();
3470 break;
3471 }
3472
3473 /* close unused handle in child process
3474 */
3475 if (mdfd > -1)
3476 close(mdfd);
3477
3478 while(1) {
3479 /* For each member array with reshape_active,
3480 * we need to perform the reshape.
3481 * We pick the first array that needs reshaping and
3482 * reshape it. reshape_array() will re-read the metadata
3483 * so the next time through a different array should be
3484 * ready for reshape.
3485 * It is possible that the 'different' array will not
3486 * be assembled yet. In that case we simple exit.
3487 * When it is assembled, the mdadm which assembles it
3488 * will take over the reshape.
3489 */
3490 struct mdinfo *content;
3491 int fd;
3492 struct mdstat_ent *mdstat;
3493 char *adev;
3494 int devid;
3495
3496 sysfs_free(cc);
3497
3498 cc = st->ss->container_content(st, NULL);
3499
3500 for (content = cc; content ; content = content->next) {
3501 char *subarray;
3502 if (!content->reshape_active)
3503 continue;
3504
3505 subarray = strchr(content->text_version+1, '/')+1;
3506 mdstat = mdstat_by_subdev(subarray, container);
3507 if (!mdstat)
3508 continue;
3509 if (mdstat->active == 0) {
3510 pr_err("Skipping inactive array %s.\n",
3511 mdstat->devnm);
3512 free_mdstat(mdstat);
3513 mdstat = NULL;
3514 continue;
3515 }
3516 break;
3517 }
3518 if (!content)
3519 break;
3520
3521 devid = devnm2devid(mdstat->devnm);
3522 adev = map_dev(major(devid), minor(devid), 0);
3523 if (!adev)
3524 adev = content->text_version;
3525
3526 fd = open_dev(mdstat->devnm);
3527 if (fd < 0) {
3528 pr_err("Device %s cannot be opened for reshape.\n", adev);
3529 break;
3530 }
3531
3532 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3533 /* Do not allow for multiple reshape_array() calls for
3534 * the same array.
3535 * It can happen when reshape_array() returns without
3536 * error, when reshape is not finished (wrong reshape
3537 * starting/continuation conditions). Mdmon doesn't
3538 * switch to next array in container and reentry
3539 * conditions for the same array occur.
3540 * This is possibly interim until the behaviour of
3541 * reshape_array is resolved().
3542 */
3543 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3544 close(fd);
3545 break;
3546 }
3547 strcpy(last_devnm, mdstat->devnm);
3548
3549 sysfs_init(content, fd, mdstat->devnm);
3550
3551 if (mdmon_running(container))
3552 flush_mdmon(container);
3553
3554 rv = reshape_array(container, fd, adev, st,
3555 content, force, NULL, INVALID_SECTORS,
3556 backup_file, verbose, 1, restart,
3557 freeze_reshape);
3558 close(fd);
3559
3560 if (freeze_reshape) {
3561 sysfs_free(cc);
3562 exit(0);
3563 }
3564
3565 restart = 0;
3566 if (rv)
3567 break;
3568
3569 if (mdmon_running(container))
3570 flush_mdmon(container);
3571 }
3572 if (!rv)
3573 unfreeze(st);
3574 sysfs_free(cc);
3575 exit(0);
3576 }
3577
3578 /*
3579 * We run a child process in the background which performs the following
3580 * steps:
3581 * - wait for resync to reach a certain point
3582 * - suspend io to the following section
3583 * - backup that section
3584 * - allow resync to proceed further
3585 * - resume io
3586 * - discard the backup.
3587 *
3588 * When are combined in slightly different ways in the three cases.
3589 * Grow:
3590 * - suspend/backup/allow/wait/resume/discard
3591 * Shrink:
3592 * - allow/wait/suspend/backup/allow/wait/resume/discard
3593 * same-size:
3594 * - wait/resume/discard/suspend/backup/allow
3595 *
3596 * suspend/backup/allow always come together
3597 * wait/resume/discard do too.
3598 * For the same-size case we have two backups to improve flow.
3599 *
3600 */
3601
3602 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3603 unsigned long long backup_point,
3604 unsigned long long wait_point,
3605 unsigned long long *suspend_point,
3606 unsigned long long *reshape_completed, int *frozen)
3607 {
3608 /* This function is called repeatedly by the reshape manager.
3609 * It determines how much progress can safely be made and allows
3610 * that progress.
3611 * - 'info' identifies the array and particularly records in
3612 * ->reshape_progress the metadata's knowledge of progress
3613 * This is a sector offset from the start of the array
3614 * of the next array block to be relocated. This number
3615 * may increase from 0 or decrease from array_size, depending
3616 * on the type of reshape that is happening.
3617 * Note that in contrast, 'sync_completed' is a block count of the
3618 * reshape so far. It gives the distance between the start point
3619 * (head or tail of device) and the next place that data will be
3620 * written. It always increases.
3621 * - 'reshape' is the structure created by analyse_change
3622 * - 'backup_point' shows how much the metadata manager has backed-up
3623 * data. For reshapes with increasing progress, it is the next address
3624 * to be backed up, previous addresses have been backed-up. For
3625 * decreasing progress, it is the earliest address that has been
3626 * backed up - later address are also backed up.
3627 * So addresses between reshape_progress and backup_point are
3628 * backed up providing those are in the 'correct' order.
3629 * - 'wait_point' is an array address. When reshape_completed
3630 * passes this point, progress_reshape should return. It might
3631 * return earlier if it determines that ->reshape_progress needs
3632 * to be updated or further backup is needed.
3633 * - suspend_point is maintained by progress_reshape and the caller
3634 * should not touch it except to initialise to zero.
3635 * It is an array address and it only increases in 2.6.37 and earlier.
3636 * This makes it difficult to handle reducing reshapes with
3637 * external metadata.
3638 * However: it is similar to backup_point in that it records the
3639 * other end of a suspended region from reshape_progress.
3640 * it is moved to extend the region that is safe to backup and/or
3641 * reshape
3642 * - reshape_completed is read from sysfs and returned. The caller
3643 * should copy this into ->reshape_progress when it has reason to
3644 * believe that the metadata knows this, and any backup outside this
3645 * has been erased.
3646 *
3647 * Return value is:
3648 * 1 if more data from backup_point - but only as far as suspend_point,
3649 * should be backed up
3650 * 0 if things are progressing smoothly
3651 * -1 if the reshape is finished because it is all done,
3652 * -2 if the reshape is finished due to an error.
3653 */
3654
3655 int advancing = (reshape->after.data_disks
3656 >= reshape->before.data_disks);
3657 unsigned long long need_backup; /* All data between start of array and
3658 * here will at some point need to
3659 * be backed up.
3660 */
3661 unsigned long long read_offset, write_offset;
3662 unsigned long long write_range;
3663 unsigned long long max_progress, target, completed;
3664 unsigned long long array_size = (info->component_size
3665 * reshape->before.data_disks);
3666 int fd;
3667 char buf[20];
3668
3669 /* First, we unsuspend any region that is now known to be safe.
3670 * If suspend_point is on the 'wrong' side of reshape_progress, then
3671 * we don't have or need suspension at the moment. This is true for
3672 * native metadata when we don't need to back-up.
3673 */
3674 if (advancing) {
3675 if (info->reshape_progress <= *suspend_point)
3676 sysfs_set_num(info, NULL, "suspend_lo",
3677 info->reshape_progress);
3678 } else {
3679 /* Note: this won't work in 2.6.37 and before.
3680 * Something somewhere should make sure we don't need it!
3681 */
3682 if (info->reshape_progress >= *suspend_point)
3683 sysfs_set_num(info, NULL, "suspend_hi",
3684 info->reshape_progress);
3685 }
3686
3687 /* Now work out how far it is safe to progress.
3688 * If the read_offset for ->reshape_progress is less than
3689 * 'blocks' beyond the write_offset, we can only progress as far
3690 * as a backup.
3691 * Otherwise we can progress until the write_offset for the new location
3692 * reaches (within 'blocks' of) the read_offset at the current location.
3693 * However that region must be suspended unless we are using native
3694 * metadata.
3695 * If we need to suspend more, we limit it to 128M per device, which is
3696 * rather arbitrary and should be some time-based calculation.
3697 */
3698 read_offset = info->reshape_progress / reshape->before.data_disks;
3699 write_offset = info->reshape_progress / reshape->after.data_disks;
3700 write_range = info->new_chunk/512;
3701 if (reshape->before.data_disks == reshape->after.data_disks)
3702 need_backup = array_size;
3703 else
3704 need_backup = reshape->backup_blocks;
3705 if (advancing) {
3706 if (read_offset < write_offset + write_range)
3707 max_progress = backup_point;
3708 else
3709 max_progress =
3710 read_offset *
3711 reshape->after.data_disks;
3712 } else {
3713 if (read_offset > write_offset - write_range)
3714 /* Can only progress as far as has been backed up,
3715 * which must be suspended */
3716 max_progress = backup_point;
3717 else if (info->reshape_progress <= need_backup)
3718 max_progress = backup_point;
3719 else {
3720 if (info->array.major_version >= 0)
3721 /* Can progress until backup is needed */
3722 max_progress = need_backup;
3723 else {
3724 /* Can progress until metadata update is required */
3725 max_progress =
3726 read_offset *
3727 reshape->after.data_disks;
3728 /* but data must be suspended */
3729 if (max_progress < *suspend_point)
3730 max_progress = *suspend_point;
3731 }
3732 }
3733 }
3734
3735 /* We know it is safe to progress to 'max_progress' providing
3736 * it is suspended or we are using native metadata.
3737 * Consider extending suspend_point 128M per device if it
3738 * is less than 64M per device beyond reshape_progress.
3739 * But always do a multiple of 'blocks'
3740 * FIXME this is too big - it takes to long to complete
3741 * this much.
3742 */
3743 target = 64*1024*2 * min(reshape->before.data_disks,
3744 reshape->after.data_disks);
3745 target /= reshape->backup_blocks;
3746 if (target < 2)
3747 target = 2;
3748 target *= reshape->backup_blocks;
3749
3750 /* For externally managed metadata we always need to suspend IO to
3751 * the area being reshaped so we regularly push suspend_point forward.
3752 * For native metadata we only need the suspend if we are going to do
3753 * a backup.
3754 */
3755 if (advancing) {
3756 if ((need_backup > info->reshape_progress
3757 || info->array.major_version < 0) &&
3758 *suspend_point < info->reshape_progress + target) {
3759 if (need_backup < *suspend_point + 2 * target)
3760 *suspend_point = need_backup;
3761 else if (*suspend_point + 2 * target < array_size)
3762 *suspend_point += 2 * target;
3763 else
3764 *suspend_point = array_size;
3765 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3766 if (max_progress > *suspend_point)
3767 max_progress = *suspend_point;
3768 }
3769 } else {
3770 if (info->array.major_version >= 0) {
3771 /* Only need to suspend when about to backup */
3772 if (info->reshape_progress < need_backup * 2 &&
3773 *suspend_point > 0) {
3774 *suspend_point = 0;
3775 sysfs_set_num(info, NULL, "suspend_lo", 0);
3776 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3777 }
3778 } else {
3779 /* Need to suspend continually */
3780 if (info->reshape_progress < *suspend_point)
3781 *suspend_point = info->reshape_progress;
3782 if (*suspend_point + target < info->reshape_progress)
3783 /* No need to move suspend region yet */;
3784 else {
3785 if (*suspend_point >= 2 * target)
3786 *suspend_point -= 2 * target;
3787 else
3788 *suspend_point = 0;
3789 sysfs_set_num(info, NULL, "suspend_lo",
3790 *suspend_point);
3791 }
3792 if (max_progress < *suspend_point)
3793 max_progress = *suspend_point;
3794 }
3795 }
3796
3797 /* now set sync_max to allow that progress. sync_max, like
3798 * sync_completed is a count of sectors written per device, so
3799 * we find the difference between max_progress and the start point,
3800 * and divide that by after.data_disks to get a sync_max
3801 * number.
3802 * At the same time we convert wait_point to a similar number
3803 * for comparing against sync_completed.
3804 */
3805 /* scale down max_progress to per_disk */
3806 max_progress /= reshape->after.data_disks;
3807 /* Round to chunk size as some kernels give an erroneously high number */
3808 max_progress /= info->new_chunk/512;
3809 max_progress *= info->new_chunk/512;
3810 /* And round to old chunk size as the kernel wants that */
3811 max_progress /= info->array.chunk_size/512;
3812 max_progress *= info->array.chunk_size/512;
3813 /* Limit progress to the whole device */
3814 if (max_progress > info->component_size)
3815 max_progress = info->component_size;
3816 wait_point /= reshape->after.data_disks;
3817 if (!advancing) {
3818 /* switch from 'device offset' to 'processed block count' */
3819 max_progress = info->component_size - max_progress;
3820 wait_point = info->component_size - wait_point;
3821 }
3822
3823 if (!*frozen)
3824 sysfs_set_num(info, NULL, "sync_max", max_progress);
3825
3826 /* Now wait. If we have already reached the point that we were
3827 * asked to wait to, don't wait at all, else wait for any change.
3828 * We need to select on 'sync_completed' as that is the place that
3829 * notifications happen, but we are really interested in
3830 * 'reshape_position'
3831 */
3832 fd = sysfs_get_fd(info, NULL, "sync_completed");
3833 if (fd < 0)
3834 goto check_progress;
3835
3836 if (sysfs_fd_get_ll(fd, &completed) < 0)
3837 goto check_progress;
3838
3839 while (completed < max_progress && completed < wait_point) {
3840 /* Check that sync_action is still 'reshape' to avoid
3841 * waiting forever on a dead array
3842 */
3843 char action[20];
3844 if (sysfs_get_str(info, NULL, "sync_action",
3845 action, 20) <= 0 ||
3846 strncmp(action, "reshape", 7) != 0)
3847 break;
3848 /* Some kernels reset 'sync_completed' to zero
3849 * before setting 'sync_action' to 'idle'.
3850 * So we need these extra tests.
3851 */
3852 if (completed == 0 && advancing
3853 && strncmp(action, "idle", 4) == 0
3854 && info->reshape_progress > 0)
3855 break;
3856 if (completed == 0 && !advancing
3857 && strncmp(action, "idle", 4) == 0
3858 && info->reshape_progress < (info->component_size
3859 * reshape->after.data_disks))
3860 break;
3861 sysfs_wait(fd, NULL);
3862 if (sysfs_fd_get_ll(fd, &completed) < 0)
3863 goto check_progress;
3864 }
3865 /* Some kernels reset 'sync_completed' to zero,
3866 * we need to have real point we are in md.
3867 * So in that case, read 'reshape_position' from sysfs.
3868 */
3869 if (completed == 0) {
3870 unsigned long long reshapep;
3871 char action[20];
3872 if (sysfs_get_str(info, NULL, "sync_action",
3873 action, 20) > 0 &&
3874 strncmp(action, "idle", 4) == 0 &&
3875 sysfs_get_ll(info, NULL,
3876 "reshape_position", &reshapep) == 0)
3877 *reshape_completed = reshapep;
3878 } else {
3879 /* some kernels can give an incorrectly high
3880 * 'completed' number, so round down */
3881 completed /= (info->new_chunk/512);
3882 completed *= (info->new_chunk/512);
3883 /* Convert 'completed' back in to a 'progress' number */
3884 completed *= reshape->after.data_disks;
3885 if (!advancing)
3886 completed = (info->component_size
3887 * reshape->after.data_disks
3888 - completed);
3889 *reshape_completed = completed;
3890 }
3891
3892 close(fd);
3893
3894 /* We return the need_backup flag. Caller will decide
3895 * how much - a multiple of ->backup_blocks up to *suspend_point
3896 */
3897 if (advancing)
3898 return need_backup > info->reshape_progress;
3899 else
3900 return need_backup >= info->reshape_progress;
3901
3902 check_progress:
3903 /* if we couldn't read a number from sync_completed, then
3904 * either the reshape did complete, or it aborted.
3905 * We can tell which by checking for 'none' in reshape_position.
3906 * If it did abort, then it might immediately restart if it
3907 * it was just a device failure that leaves us degraded but
3908 * functioning.
3909 */
3910 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3911 || strncmp(buf, "none", 4) != 0) {
3912 /* The abort might only be temporary. Wait up to 10
3913 * seconds for fd to contain a valid number again.
3914 */
3915 int wait = 10000;
3916 int rv = -2;
3917 unsigned long long new_sync_max;
3918 while (fd >= 0 && rv < 0 && wait > 0) {
3919 if (sysfs_wait(fd, &wait) != 1)
3920 break;
3921 switch (sysfs_fd_get_ll(fd, &completed)) {
3922 case 0:
3923 /* all good again */
3924 rv = 1;
3925 /* If "sync_max" is no longer max_progress
3926 * we need to freeze things
3927 */
3928 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3929 *frozen = (new_sync_max != max_progress);
3930 break;
3931 case -2: /* read error - abort */
3932 wait = 0;
3933 break;
3934 }
3935 }
3936 if (fd >= 0)
3937 close(fd);
3938 return rv; /* abort */
3939 } else {
3940 /* Maybe racing with array shutdown - check state */
3941 if (fd >= 0)
3942 close(fd);
3943 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3944 || strncmp(buf, "inactive", 8) == 0
3945 || strncmp(buf, "clear",5) == 0)
3946 return -2; /* abort */
3947 return -1; /* complete */
3948 }
3949 }
3950
3951 /* FIXME return status is never checked */
3952 static int grow_backup(struct mdinfo *sra,
3953 unsigned long long offset, /* per device */
3954 unsigned long stripes, /* per device, in old chunks */
3955 int *sources, unsigned long long *offsets,
3956 int disks, int chunk, int level, int layout,
3957 int dests, int *destfd, unsigned long long *destoffsets,
3958 int part, int *degraded,
3959 char *buf)
3960 {
3961 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3962 * to storage 'destfd' (offset 'destoffsets'), after first
3963 * suspending IO. Then allow resync to continue
3964 * over the suspended section.
3965 * Use part 'part' of the backup-super-block.
3966 */
3967 int odata = disks;
3968 int rv = 0;
3969 int i;
3970 unsigned long long ll;
3971 int new_degraded;
3972 //printf("offset %llu\n", offset);
3973 if (level >= 4)
3974 odata--;
3975 if (level == 6)
3976 odata--;
3977
3978 /* Check that array hasn't become degraded, else we might backup the wrong data */
3979 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3980 return -1; /* FIXME this error is ignored */
3981 new_degraded = (int)ll;
3982 if (new_degraded != *degraded) {
3983 /* check each device to ensure it is still working */
3984 struct mdinfo *sd;
3985 for (sd = sra->devs ; sd ; sd = sd->next) {
3986 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3987 continue;
3988 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3989 char sbuf[20];
3990 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3991 strstr(sbuf, "faulty") ||
3992 strstr(sbuf, "in_sync") == NULL) {
3993 /* this device is dead */
3994 sd->disk.state = (1<<MD_DISK_FAULTY);
3995 if (sd->disk.raid_disk >= 0 &&
3996 sources[sd->disk.raid_disk] >= 0) {
3997 close(sources[sd->disk.raid_disk]);
3998 sources[sd->disk.raid_disk] = -1;
3999 }
4000 }
4001 }
4002 }
4003 *degraded = new_degraded;
4004 }
4005 if (part) {
4006 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4007 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4008 } else {
4009 bsb.arraystart = __cpu_to_le64(offset * odata);
4010 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4011 }
4012 if (part)
4013 bsb.magic[15] = '2';
4014 for (i = 0; i < dests; i++)
4015 if (part)
4016 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4017 else
4018 lseek64(destfd[i], destoffsets[i], 0);
4019
4020 rv = save_stripes(sources, offsets,
4021 disks, chunk, level, layout,
4022 dests, destfd,
4023 offset*512*odata, stripes * chunk * odata,
4024 buf);
4025
4026 if (rv)
4027 return rv;
4028 bsb.mtime = __cpu_to_le64(time(0));
4029 for (i = 0; i < dests; i++) {
4030 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4031
4032 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4033 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4034 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4035 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4036
4037 rv = -1;
4038 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4039 != destoffsets[i] - 4096)
4040 break;
4041 if (write(destfd[i], &bsb, 512) != 512)
4042 break;
4043 if (destoffsets[i] > 4096) {
4044 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4045 destoffsets[i]+stripes*chunk*odata)
4046 break;
4047 if (write(destfd[i], &bsb, 512) != 512)
4048 break;
4049 }
4050 fsync(destfd[i]);
4051 rv = 0;
4052 }
4053
4054 return rv;
4055 }
4056
4057 /* in 2.6.30, the value reported by sync_completed can be
4058 * less that it should be by one stripe.
4059 * This only happens when reshape hits sync_max and pauses.
4060 * So allow wait_backup to either extent sync_max further
4061 * than strictly necessary, or return before the
4062 * sync has got quite as far as we would really like.
4063 * This is what 'blocks2' is for.
4064 * The various caller give appropriate values so that
4065 * every works.
4066 */
4067 /* FIXME return value is often ignored */
4068 static int forget_backup(int dests, int *destfd,
4069 unsigned long long *destoffsets,
4070 int part)
4071 {
4072 /*
4073 * Erase backup 'part' (which is 0 or 1)
4074 */
4075 int i;
4076 int rv;
4077
4078 if (part) {
4079 bsb.arraystart2 = __cpu_to_le64(0);
4080 bsb.length2 = __cpu_to_le64(0);
4081 } else {
4082 bsb.arraystart = __cpu_to_le64(0);
4083 bsb.length = __cpu_to_le64(0);
4084 }
4085 bsb.mtime = __cpu_to_le64(time(0));
4086 rv = 0;
4087 for (i = 0; i < dests; i++) {
4088 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4089 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4090 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4091 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4092 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4093 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4094 destoffsets[i]-4096)
4095 rv = -1;
4096 if (rv == 0 &&
4097 write(destfd[i], &bsb, 512) != 512)
4098 rv = -1;
4099 fsync(destfd[i]);
4100 }
4101 return rv;
4102 }
4103
4104 static void fail(char *msg)
4105 {
4106 int rv;
4107 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4108 rv |= (write(2, "\n", 1) != 1);
4109 exit(rv ? 1 : 2);
4110 }
4111
4112 static char *abuf, *bbuf;
4113 static unsigned long long abuflen;
4114 static void validate(int afd, int bfd, unsigned long long offset)
4115 {
4116 /* check that the data in the backup against the array.
4117 * This is only used for regression testing and should not
4118 * be used while the array is active
4119 */
4120 if (afd < 0)
4121 return;
4122 lseek64(bfd, offset - 4096, 0);
4123 if (read(bfd, &bsb2, 512) != 512)
4124 fail("cannot read bsb");
4125 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4126 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4127 fail("first csum bad");
4128 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4129 fail("magic is bad");
4130 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4131 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4132 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4133 fail("second csum bad");
4134
4135 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4136 fail("devstart is wrong");
4137
4138 if (bsb2.length) {
4139 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4140
4141 if (abuflen < len) {
4142 free(abuf);
4143 free(bbuf);
4144 abuflen = len;
4145 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4146 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4147 abuflen = 0;
4148 /* just stop validating on mem-alloc failure */
4149 return;
4150 }
4151 }
4152
4153 lseek64(bfd, offset, 0);
4154 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4155 //printf("len %llu\n", len);
4156 fail("read first backup failed");
4157 }
4158 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4159 if ((unsigned long long)read(afd, abuf, len) != len)
4160 fail("read first from array failed");
4161 if (memcmp(bbuf, abuf, len) != 0) {
4162 #if 0
4163 int i;
4164 printf("offset=%llu len=%llu\n",
4165 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4166 for (i=0; i<len; i++)
4167 if (bbuf[i] != abuf[i]) {
4168 printf("first diff byte %d\n", i);
4169 break;
4170 }
4171 #endif
4172 fail("data1 compare failed");
4173 }
4174 }
4175 if (bsb2.length2) {
4176 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4177
4178 if (abuflen < len) {
4179 free(abuf);
4180 free(bbuf);
4181 abuflen = len;
4182 abuf = xmalloc(abuflen);
4183 bbuf = xmalloc(abuflen);
4184 }
4185
4186 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4187 if ((unsigned long long)read(bfd, bbuf, len) != len)
4188 fail("read second backup failed");
4189 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4190 if ((unsigned long long)read(afd, abuf, len) != len)
4191 fail("read second from array failed");
4192 if (memcmp(bbuf, abuf, len) != 0)
4193 fail("data2 compare failed");
4194 }
4195 }
4196
4197 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4198 struct supertype *st, unsigned long blocks,
4199 int *fds, unsigned long long *offsets,
4200 int dests, int *destfd, unsigned long long *destoffsets)
4201 {
4202 /* Monitor a reshape where backup is being performed using
4203 * 'native' mechanism - either to a backup file, or
4204 * to some space in a spare.
4205 */
4206 char *buf;
4207 int degraded = -1;
4208 unsigned long long speed;
4209 unsigned long long suspend_point, array_size;
4210 unsigned long long backup_point, wait_point;
4211 unsigned long long reshape_completed;
4212 int done = 0;
4213 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4214 int part = 0; /* The next part of the backup area to fill. It may already
4215 * be full, so we need to check */
4216 int level = reshape->level;
4217 int layout = reshape->before.layout;
4218 int data = reshape->before.data_disks;
4219 int disks = reshape->before.data_disks + reshape->parity;
4220 int chunk = sra->array.chunk_size;
4221 struct mdinfo *sd;
4222 unsigned long stripes;
4223 int uuid[4];
4224 int frozen = 0;
4225
4226 /* set up the backup-super-block. This requires the
4227 * uuid from the array.
4228 */
4229 /* Find a superblock */
4230 for (sd = sra->devs; sd; sd = sd->next) {
4231 char *dn;
4232 int devfd;
4233 int ok;
4234 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4235 continue;
4236 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4237 devfd = dev_open(dn, O_RDONLY);
4238 if (devfd < 0)
4239 continue;
4240 ok = st->ss->load_super(st, devfd, NULL);
4241 close(devfd);
4242 if (ok == 0)
4243 break;
4244 }
4245 if (!sd) {
4246 pr_err("Cannot find a superblock\n");
4247 return 0;
4248 }
4249
4250 memset(&bsb, 0, 512);
4251 memcpy(bsb.magic, "md_backup_data-1", 16);
4252 st->ss->uuid_from_super(st, uuid);
4253 memcpy(bsb.set_uuid, uuid, 16);
4254 bsb.mtime = __cpu_to_le64(time(0));
4255 bsb.devstart2 = blocks;
4256
4257 stripes = blocks / (sra->array.chunk_size/512) /
4258 reshape->before.data_disks;
4259
4260 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4261 /* Don't start the 'reshape' */
4262 return 0;
4263 if (reshape->before.data_disks == reshape->after.data_disks) {
4264 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4265 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4266 }
4267
4268 if (increasing) {
4269 array_size = sra->component_size * reshape->after.data_disks;
4270 backup_point = sra->reshape_progress;
4271 suspend_point = 0;
4272 } else {
4273 array_size = sra->component_size * reshape->before.data_disks;
4274 backup_point = reshape->backup_blocks;
4275 suspend_point = array_size;
4276 }
4277
4278 while (!done) {
4279 int rv;
4280
4281 /* Want to return as soon the oldest backup slot can
4282 * be released as that allows us to start backing up
4283 * some more, providing suspend_point has been
4284 * advanced, which it should have.
4285 */
4286 if (increasing) {
4287 wait_point = array_size;
4288 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4289 wait_point = (__le64_to_cpu(bsb.arraystart) +
4290 __le64_to_cpu(bsb.length));
4291 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4292 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4293 __le64_to_cpu(bsb.length2));
4294 } else {
4295 wait_point = 0;
4296 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4297 wait_point = __le64_to_cpu(bsb.arraystart);
4298 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4299 wait_point = __le64_to_cpu(bsb.arraystart2);
4300 }
4301
4302 reshape_completed = sra->reshape_progress;
4303 rv = progress_reshape(sra, reshape,
4304 backup_point, wait_point,
4305 &suspend_point, &reshape_completed,
4306 &frozen);
4307 /* external metadata would need to ping_monitor here */
4308 sra->reshape_progress = reshape_completed;
4309
4310 /* Clear any backup region that is before 'here' */
4311 if (increasing) {
4312 if (__le64_to_cpu(bsb.length) > 0 &&
4313 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4314 __le64_to_cpu(bsb.length)))
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 __le64_to_cpu(bsb.length2)))
4320 forget_backup(dests, destfd,
4321 destoffsets, 1);
4322 } else {
4323 if (__le64_to_cpu(bsb.length) > 0 &&
4324 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4325 forget_backup(dests, destfd,
4326 destoffsets, 0);
4327 if (__le64_to_cpu(bsb.length2) > 0 &&
4328 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4329 forget_backup(dests, destfd,
4330 destoffsets, 1);
4331 }
4332 if (sigterm)
4333 rv = -2;
4334 if (rv < 0) {
4335 if (rv == -1)
4336 done = 1;
4337 break;
4338 }
4339 if (rv == 0 && increasing && !st->ss->external) {
4340 /* No longer need to monitor this reshape */
4341 sysfs_set_str(sra, NULL, "sync_max", "max");
4342 done = 1;
4343 break;
4344 }
4345
4346 while (rv) {
4347 unsigned long long offset;
4348 unsigned long actual_stripes;
4349 /* Need to backup some data.
4350 * If 'part' is not used and the desired
4351 * backup size is suspended, do a backup,
4352 * then consider the next part.
4353 */
4354 /* Check that 'part' is unused */
4355 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4356 break;
4357 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4358 break;
4359
4360 offset = backup_point / data;
4361 actual_stripes = stripes;
4362 if (increasing) {
4363 if (offset + actual_stripes * (chunk/512) >
4364 sra->component_size)
4365 actual_stripes = ((sra->component_size - offset)
4366 / (chunk/512));
4367 if (offset + actual_stripes * (chunk/512) >
4368 suspend_point/data)
4369 break;
4370 } else {
4371 if (offset < actual_stripes * (chunk/512))
4372 actual_stripes = offset / (chunk/512);
4373 offset -= actual_stripes * (chunk/512);
4374 if (offset < suspend_point/data)
4375 break;
4376 }
4377 if (actual_stripes == 0)
4378 break;
4379 grow_backup(sra, offset, actual_stripes,
4380 fds, offsets,
4381 disks, chunk, level, layout,
4382 dests, destfd, destoffsets,
4383 part, &degraded, buf);
4384 validate(afd, destfd[0], destoffsets[0]);
4385 /* record where 'part' is up to */
4386 part = !part;
4387 if (increasing)
4388 backup_point += actual_stripes * (chunk/512) * data;
4389 else
4390 backup_point -= actual_stripes * (chunk/512) * data;
4391 }
4392 }
4393
4394 /* FIXME maybe call progress_reshape one more time instead */
4395 /* remove any remaining suspension */
4396 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4397 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4398 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4399 sysfs_set_num(sra, NULL, "sync_min", 0);
4400
4401 if (reshape->before.data_disks == reshape->after.data_disks)
4402 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4403 free(buf);
4404 return done;
4405 }
4406
4407 /*
4408 * If any spare contains md_back_data-1 which is recent wrt mtime,
4409 * write that data into the array and update the super blocks with
4410 * the new reshape_progress
4411 */
4412 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4413 char *backup_file, int verbose)
4414 {
4415 int i, j;
4416 int old_disks;
4417 unsigned long long *offsets;
4418 unsigned long long nstripe, ostripe;
4419 int ndata, odata;
4420
4421 odata = info->array.raid_disks - info->delta_disks - 1;
4422 if (info->array.level == 6) odata--; /* number of data disks */
4423 ndata = info->array.raid_disks - 1;
4424 if (info->new_level == 6) ndata--;
4425
4426 old_disks = info->array.raid_disks - info->delta_disks;
4427
4428 if (info->delta_disks <= 0)
4429 /* Didn't grow, so the backup file must have
4430 * been used
4431 */
4432 old_disks = cnt;
4433 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4434 struct mdinfo dinfo;
4435 int fd;
4436 int bsbsize;
4437 char *devname, namebuf[20];
4438 unsigned long long lo, hi;
4439
4440 /* This was a spare and may have some saved data on it.
4441 * Load the superblock, find and load the
4442 * backup_super_block.
4443 * If either fail, go on to next device.
4444 * If the backup contains no new info, just return
4445 * else restore data and update all superblocks
4446 */
4447 if (i == old_disks-1) {
4448 fd = open(backup_file, O_RDONLY);
4449 if (fd<0) {
4450 pr_err("backup file %s inaccessible: %s\n",
4451 backup_file, strerror(errno));
4452 continue;
4453 }
4454 devname = backup_file;
4455 } else {
4456 fd = fdlist[i];
4457 if (fd < 0)
4458 continue;
4459 if (st->ss->load_super(st, fd, NULL))
4460 continue;
4461
4462 st->ss->getinfo_super(st, &dinfo, NULL);
4463 st->ss->free_super(st);
4464
4465 if (lseek64(fd,
4466 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4467 0) < 0) {
4468 pr_err("Cannot seek on device %d\n", i);
4469 continue; /* Cannot seek */
4470 }
4471 sprintf(namebuf, "device-%d", i);
4472 devname = namebuf;
4473 }
4474 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4475 if (verbose)
4476 pr_err("Cannot read from %s\n", devname);
4477 continue; /* Cannot read */
4478 }
4479 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4480 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4481 if (verbose)
4482 pr_err("No backup metadata on %s\n", devname);
4483 continue;
4484 }
4485 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4486 if (verbose)
4487 pr_err("Bad backup-metadata checksum on %s\n", devname);
4488 continue; /* bad checksum */
4489 }
4490 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4491 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4492 if (verbose)
4493 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4494 continue; /* Bad second checksum */
4495 }
4496 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4497 if (verbose)
4498 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4499 continue; /* Wrong uuid */
4500 }
4501
4502 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4503 * sometimes they aren't... So allow considerable flexability in matching, and allow
4504 * this test to be overridden by an environment variable.
4505 */
4506 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4507 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4508 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4509 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4510 (unsigned long)__le64_to_cpu(bsb.mtime),
4511 (unsigned long)info->array.utime);
4512 } else {
4513 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4514 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4515 continue; /* time stamp is too bad */
4516 }
4517 }
4518
4519 if (bsb.magic[15] == '1') {
4520 if (bsb.length == 0)
4521 continue;
4522 if (info->delta_disks >= 0) {
4523 /* reshape_progress is increasing */
4524 if (__le64_to_cpu(bsb.arraystart)
4525 + __le64_to_cpu(bsb.length)
4526 < info->reshape_progress) {
4527 nonew:
4528 if (verbose)
4529 pr_err("backup-metadata found on %s but is not needed\n", devname);
4530 continue; /* No new data here */
4531 }
4532 } else {
4533 /* reshape_progress is decreasing */
4534 if (__le64_to_cpu(bsb.arraystart) >=
4535 info->reshape_progress)
4536 goto nonew; /* No new data here */
4537 }
4538 } else {
4539 if (bsb.length == 0 && bsb.length2 == 0)
4540 continue;
4541 if (info->delta_disks >= 0) {
4542 /* reshape_progress is increasing */
4543 if ((__le64_to_cpu(bsb.arraystart)
4544 + __le64_to_cpu(bsb.length)
4545 < info->reshape_progress)
4546 &&
4547 (__le64_to_cpu(bsb.arraystart2)
4548 + __le64_to_cpu(bsb.length2)
4549 < info->reshape_progress))
4550 goto nonew; /* No new data here */
4551 } else {
4552 /* reshape_progress is decreasing */
4553 if (__le64_to_cpu(bsb.arraystart) >=
4554 info->reshape_progress &&
4555 __le64_to_cpu(bsb.arraystart2) >=
4556 info->reshape_progress)
4557 goto nonew; /* No new data here */
4558 }
4559 }
4560 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4561 second_fail:
4562 if (verbose)
4563 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4564 devname);
4565 continue; /* Cannot seek */
4566 }
4567 /* There should be a duplicate backup superblock 4k before here */
4568 if (lseek64(fd, -4096, 1) < 0 ||
4569 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4570 goto second_fail; /* Cannot find leading superblock */
4571 if (bsb.magic[15] == '1')
4572 bsbsize = offsetof(struct mdp_backup_super, pad1);
4573 else
4574 bsbsize = offsetof(struct mdp_backup_super, pad);
4575 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4576 goto second_fail; /* Cannot find leading superblock */
4577
4578 /* Now need the data offsets for all devices. */
4579 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4580 for(j=0; j<info->array.raid_disks; j++) {
4581 if (fdlist[j] < 0)
4582 continue;
4583 if (st->ss->load_super(st, fdlist[j], NULL))
4584 /* FIXME should be this be an error */
4585 continue;
4586 st->ss->getinfo_super(st, &dinfo, NULL);
4587 st->ss->free_super(st);
4588 offsets[j] = dinfo.data_offset * 512;
4589 }
4590 printf("%s: restoring critical section\n", Name);
4591
4592 if (restore_stripes(fdlist, offsets,
4593 info->array.raid_disks,
4594 info->new_chunk,
4595 info->new_level,
4596 info->new_layout,
4597 fd, __le64_to_cpu(bsb.devstart)*512,
4598 __le64_to_cpu(bsb.arraystart)*512,
4599 __le64_to_cpu(bsb.length)*512, NULL)) {
4600 /* didn't succeed, so giveup */
4601 if (verbose)
4602 pr_err("Error restoring backup from %s\n",
4603 devname);
4604 free(offsets);
4605 return 1;
4606 }
4607
4608 if (bsb.magic[15] == '2' &&
4609 restore_stripes(fdlist, offsets,
4610 info->array.raid_disks,
4611 info->new_chunk,
4612 info->new_level,
4613 info->new_layout,
4614 fd, __le64_to_cpu(bsb.devstart)*512 +
4615 __le64_to_cpu(bsb.devstart2)*512,
4616 __le64_to_cpu(bsb.arraystart2)*512,
4617 __le64_to_cpu(bsb.length2)*512, NULL)) {
4618 /* didn't succeed, so giveup */
4619 if (verbose)
4620 pr_err("Error restoring second backup from %s\n",
4621 devname);
4622 free(offsets);
4623 return 1;
4624 }
4625
4626 free(offsets);
4627
4628 /* Ok, so the data is restored. Let's update those superblocks. */
4629
4630 lo = hi = 0;
4631 if (bsb.length) {
4632 lo = __le64_to_cpu(bsb.arraystart);
4633 hi = lo + __le64_to_cpu(bsb.length);
4634 }
4635 if (bsb.magic[15] == '2' && bsb.length2) {
4636 unsigned long long lo1, hi1;
4637 lo1 = __le64_to_cpu(bsb.arraystart2);
4638 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4639 if (lo == hi) {
4640 lo = lo1;
4641 hi = hi1;
4642 } else if (lo < lo1)
4643 hi = hi1;
4644 else
4645 lo = lo1;
4646 }
4647 if (lo < hi &&
4648 (info->reshape_progress < lo ||
4649 info->reshape_progress > hi))
4650 /* backup does not affect reshape_progress*/ ;
4651 else if (info->delta_disks >= 0) {
4652 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4653 __le64_to_cpu(bsb.length);
4654 if (bsb.magic[15] == '2') {
4655 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4656 __le64_to_cpu(bsb.length2);
4657 if (p2 > info->reshape_progress)
4658 info->reshape_progress = p2;
4659 }
4660 } else {
4661 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4662 if (bsb.magic[15] == '2') {
4663 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4664 if (p2 < info->reshape_progress)
4665 info->reshape_progress = p2;
4666 }
4667 }
4668 for (j=0; j<info->array.raid_disks; j++) {
4669 if (fdlist[j] < 0)
4670 continue;
4671 if (st->ss->load_super(st, fdlist[j], NULL))
4672 continue;
4673 st->ss->getinfo_super(st, &dinfo, NULL);
4674 dinfo.reshape_progress = info->reshape_progress;
4675 st->ss->update_super(st, &dinfo,
4676 "_reshape_progress",
4677 NULL,0, 0, NULL);
4678 st->ss->store_super(st, fdlist[j]);
4679 st->ss->free_super(st);
4680 }
4681 return 0;
4682 }
4683 /* Didn't find any backup data, try to see if any
4684 * was needed.
4685 */
4686 if (info->delta_disks < 0) {
4687 /* When shrinking, the critical section is at the end.
4688 * So see if we are before the critical section.
4689 */
4690 unsigned long long first_block;
4691 nstripe = ostripe = 0;
4692 first_block = 0;
4693 while (ostripe >= nstripe) {
4694 ostripe += info->array.chunk_size / 512;
4695 first_block = ostripe * odata;
4696 nstripe = first_block / ndata / (info->new_chunk/512) *
4697 (info->new_chunk/512);
4698 }
4699
4700 if (info->reshape_progress >= first_block)
4701 return 0;
4702 }
4703 if (info->delta_disks > 0) {
4704 /* See if we are beyond the critical section. */
4705 unsigned long long last_block;
4706 nstripe = ostripe = 0;
4707 last_block = 0;
4708 while (nstripe >= ostripe) {
4709 nstripe += info->new_chunk / 512;
4710 last_block = nstripe * ndata;
4711 ostripe = last_block / odata / (info->array.chunk_size/512) *
4712 (info->array.chunk_size/512);
4713 }
4714
4715 if (info->reshape_progress >= last_block)
4716 return 0;
4717 }
4718 /* needed to recover critical section! */
4719 if (verbose)
4720 pr_err("Failed to find backup of critical section\n");
4721 return 1;
4722 }
4723
4724 int Grow_continue_command(char *devname, int fd,
4725 char *backup_file, int verbose)
4726 {
4727 int ret_val = 0;
4728 struct supertype *st = NULL;
4729 struct mdinfo *content = NULL;
4730 struct mdinfo array;
4731 char *subarray = NULL;
4732 struct mdinfo *cc = NULL;
4733 struct mdstat_ent *mdstat = NULL;
4734 int cfd = -1;
4735 int fd2 = -1;
4736
4737 dprintf("Grow continue from command line called for %s\n",
4738 devname);
4739
4740 st = super_by_fd(fd, &subarray);
4741 if (!st || !st->ss) {
4742 pr_err("Unable to determine metadata format for %s\n",
4743 devname);
4744 return 1;
4745 }
4746 dprintf("Grow continue is run for ");
4747 if (st->ss->external == 0) {
4748 int d;
4749 dprintf_cont("native array (%s)\n", devname);
4750 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4751 pr_err("%s is not an active md array - aborting\n", devname);
4752 ret_val = 1;
4753 goto Grow_continue_command_exit;
4754 }
4755 content = &array;
4756 /* Need to load a superblock.
4757 * FIXME we should really get what we need from
4758 * sysfs
4759 */
4760 for (d = 0; d < MAX_DISKS; d++) {
4761 mdu_disk_info_t disk;
4762 char *dv;
4763 int err;
4764 disk.number = d;
4765 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4766 continue;
4767 if (disk.major == 0 && disk.minor == 0)
4768 continue;
4769 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4770 continue;
4771 dv = map_dev(disk.major, disk.minor, 1);
4772 if (!dv)
4773 continue;
4774 fd2 = dev_open(dv, O_RDONLY);
4775 if (fd2 < 0)
4776 continue;
4777 err = st->ss->load_super(st, fd2, NULL);
4778 close(fd2);
4779 /* invalidate fd2 to avoid possible double close() */
4780 fd2 = -1;
4781 if (err)
4782 continue;
4783 break;
4784 }
4785 if (d == MAX_DISKS) {
4786 pr_err("Unable to load metadata for %s\n",
4787 devname);
4788 ret_val = 1;
4789 goto Grow_continue_command_exit;
4790 }
4791 st->ss->getinfo_super(st, content, NULL);
4792 } else {
4793 char *container;
4794
4795 if (subarray) {
4796 dprintf_cont("subarray (%s)\n", subarray);
4797 container = st->container_devnm;
4798 cfd = open_dev_excl(st->container_devnm);
4799 } else {
4800 container = st->devnm;
4801 close(fd);
4802 cfd = open_dev_excl(st->devnm);
4803 dprintf_cont("container (%s)\n", container);
4804 fd = cfd;
4805 }
4806 if (cfd < 0) {
4807 pr_err("Unable to open container for %s\n", devname);
4808 ret_val = 1;
4809 goto Grow_continue_command_exit;
4810 }
4811
4812 /* find in container array under reshape
4813 */
4814 ret_val = st->ss->load_container(st, cfd, NULL);
4815 if (ret_val) {
4816 pr_err("Cannot read superblock for %s\n",
4817 devname);
4818 ret_val = 1;
4819 goto Grow_continue_command_exit;
4820 }
4821
4822 cc = st->ss->container_content(st, subarray);
4823 for (content = cc; content ; content = content->next) {
4824 char *array;
4825 int allow_reshape = 1;
4826
4827 if (content->reshape_active == 0)
4828 continue;
4829 /* The decision about array or container wide
4830 * reshape is taken in Grow_continue based
4831 * content->reshape_active state, therefore we
4832 * need to check_reshape based on
4833 * reshape_active and subarray name
4834 */
4835 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4836 allow_reshape = 0;
4837 if (content->reshape_active == CONTAINER_RESHAPE &&
4838 (content->array.state
4839 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4840 allow_reshape = 0;
4841
4842 if (!allow_reshape) {
4843 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4844 devname, container);
4845 ret_val = 1;
4846 goto Grow_continue_command_exit;
4847 }
4848
4849 array = strchr(content->text_version+1, '/')+1;
4850 mdstat = mdstat_by_subdev(array, container);
4851 if (!mdstat)
4852 continue;
4853 if (mdstat->active == 0) {
4854 pr_err("Skipping inactive array %s.\n",
4855 mdstat->devnm);
4856 free_mdstat(mdstat);
4857 mdstat = NULL;
4858 continue;
4859 }
4860 break;
4861 }
4862 if (!content) {
4863 pr_err("Unable to determine reshaped array for %s\n", devname);
4864 ret_val = 1;
4865 goto Grow_continue_command_exit;
4866 }
4867 fd2 = open_dev(mdstat->devnm);
4868 if (fd2 < 0) {
4869 pr_err("cannot open (%s)\n", mdstat->devnm);
4870 ret_val = 1;
4871 goto Grow_continue_command_exit;
4872 }
4873
4874 sysfs_init(content, fd2, mdstat->devnm);
4875
4876 /* start mdmon in case it is not running
4877 */
4878 if (!mdmon_running(container))
4879 start_mdmon(container);
4880 ping_monitor(container);
4881
4882 if (mdmon_running(container))
4883 st->update_tail = &st->updates;
4884 else {
4885 pr_err("No mdmon found. Grow cannot continue.\n");
4886 ret_val = 1;
4887 goto Grow_continue_command_exit;
4888 }
4889 }
4890
4891 /* verify that array under reshape is started from
4892 * correct position
4893 */
4894 if (verify_reshape_position(content, content->array.level) < 0) {
4895 ret_val = 1;
4896 goto Grow_continue_command_exit;
4897 }
4898
4899 /* continue reshape
4900 */
4901 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4902
4903 Grow_continue_command_exit:
4904 if (fd2 > -1)
4905 close(fd2);
4906 if (cfd > -1)
4907 close(cfd);
4908 st->ss->free_super(st);
4909 free_mdstat(mdstat);
4910 sysfs_free(cc);
4911 free(subarray);
4912
4913 return ret_val;
4914 }
4915
4916 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4917 char *backup_file, int forked, int freeze_reshape)
4918 {
4919 int ret_val = 2;
4920
4921 if (!info->reshape_active)
4922 return ret_val;
4923
4924 if (st->ss->external) {
4925 int cfd = open_dev(st->container_devnm);
4926
4927 if (cfd < 0)
4928 return 1;
4929
4930 st->ss->load_container(st, cfd, st->container_devnm);
4931 close(cfd);
4932 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4933 st, info, 0, backup_file,
4934 0, forked,
4935 1 | info->reshape_active,
4936 freeze_reshape);
4937 } else
4938 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4939 NULL, INVALID_SECTORS,
4940 backup_file, 0, forked,
4941 1 | info->reshape_active,
4942 freeze_reshape);
4943
4944 return ret_val;
4945 }
4946
4947 char *make_backup(char *name)
4948 {
4949 char *base = "backup_file-";
4950 int len;
4951 char *fname;
4952
4953 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4954 fname = xmalloc(len);
4955 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4956 return fname;
4957 }
4958
4959 char *locate_backup(char *name)
4960 {
4961 char *fl = make_backup(name);
4962 struct stat stb;
4963
4964 if (stat(fl, &stb) == 0 &&
4965 S_ISREG(stb.st_mode))
4966 return fl;
4967
4968 free(fl);
4969 return NULL;
4970 }
Unnamed repository; edit this file 'description' to name the repository.