]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
projects / thirdparty / mdadm.git / blob
? search:


a3c8083c60d158549164d585688631448f6d5c9c
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2012 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 <stdint.h>
28
29 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
30 #error no endian defined
31 #endif
32 #include "md_u.h"
33 #include "md_p.h"
34
35 #ifndef offsetof
36 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
37 #endif
38
39 int restore_backup(struct supertype *st,
40 struct mdinfo *content,
41 int working_disks,
42 int next_spare,
43 char *backup_file,
44 int verbose)
45 {
46 int i;
47 int *fdlist;
48 struct mdinfo *dev;
49 int err;
50 int disk_count = next_spare + working_disks;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 for (i = 0; i < next_spare; i++)
56 fdlist[i] = -1;
57 for (dev = content->devs; dev; dev = dev->next) {
58 char buf[22];
59 int fd;
60 sprintf(buf, "%d:%d",
61 dev->disk.major,
62 dev->disk.minor);
63 fd = dev_open(buf, O_RDWR);
64
65 if (dev->disk.raid_disk >= 0)
66 fdlist[dev->disk.raid_disk] = fd;
67 else
68 fdlist[next_spare++] = fd;
69 }
70
71 if (st->ss->external && st->ss->recover_backup)
72 err = st->ss->recover_backup(st, content);
73 else
74 err = Grow_restart(st, content, fdlist, next_spare,
75 backup_file, verbose > 0);
76
77 while (next_spare > 0) {
78 next_spare--;
79 if (fdlist[next_spare] >= 0)
80 close(fdlist[next_spare]);
81 }
82 free(fdlist);
83 if (err) {
84 pr_err("Failed to restore critical"
85 " section for reshape - sorry.\n");
86 if (!backup_file)
87 pr_err("Possibly you need"
88 " to specify a --backup-file\n");
89 return 1;
90 }
91
92 dprintf("restore_backup() returns status OK.\n");
93 return 0;
94 }
95
96 int Grow_Add_device(char *devname, int fd, char *newdev)
97 {
98 /* Add a device to an active array.
99 * Currently, just extend a linear array.
100 * This requires writing a new superblock on the
101 * new device, calling the kernel to add the device,
102 * and if that succeeds, update the superblock on
103 * all other devices.
104 * This means that we need to *find* all other devices.
105 */
106 struct mdinfo info;
107
108 struct stat stb;
109 int nfd, fd2;
110 int d, nd;
111 struct supertype *st = NULL;
112 char *subarray = NULL;
113
114 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
115 pr_err("cannot get array info for %s\n", devname);
116 return 1;
117 }
118
119 if (info.array.level != -1) {
120 pr_err("can only add devices to linear arrays\n");
121 return 1;
122 }
123
124 st = super_by_fd(fd, &subarray);
125 if (!st) {
126 pr_err("cannot handle arrays with superblock version %d\n",
127 info.array.major_version);
128 return 1;
129 }
130
131 if (subarray) {
132 pr_err("Cannot grow linear sub-arrays yet\n");
133 free(subarray);
134 free(st);
135 return 1;
136 }
137
138 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
139 if (nfd < 0) {
140 pr_err("cannot open %s\n", newdev);
141 free(st);
142 return 1;
143 }
144 fstat(nfd, &stb);
145 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
146 pr_err("%s is not a block device!\n", newdev);
147 close(nfd);
148 free(st);
149 return 1;
150 }
151 /* now check out all the devices and make sure we can read the
152 * superblock */
153 for (d=0 ; d < info.array.raid_disks ; d++) {
154 mdu_disk_info_t disk;
155 char *dv;
156
157 st->ss->free_super(st);
158
159 disk.number = d;
160 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
161 pr_err("cannot get device detail for device %d\n",
162 d);
163 close(nfd);
164 free(st);
165 return 1;
166 }
167 dv = map_dev(disk.major, disk.minor, 1);
168 if (!dv) {
169 pr_err("cannot find device file for device %d\n",
170 d);
171 close(nfd);
172 free(st);
173 return 1;
174 }
175 fd2 = dev_open(dv, O_RDWR);
176 if (fd2 < 0) {
177 pr_err("cannot open device file %s\n", dv);
178 close(nfd);
179 free(st);
180 return 1;
181 }
182
183 if (st->ss->load_super(st, fd2, NULL)) {
184 pr_err("cannot find super block on %s\n", dv);
185 close(nfd);
186 close(fd2);
187 free(st);
188 return 1;
189 }
190 close(fd2);
191 }
192 /* Ok, looks good. Lets update the superblock and write it out to
193 * newdev.
194 */
195
196 info.disk.number = d;
197 info.disk.major = major(stb.st_rdev);
198 info.disk.minor = minor(stb.st_rdev);
199 info.disk.raid_disk = d;
200 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
201 st->ss->update_super(st, &info, "linear-grow-new", newdev,
202 0, 0, NULL);
203
204 if (st->ss->store_super(st, nfd)) {
205 pr_err("Cannot store new superblock on %s\n",
206 newdev);
207 close(nfd);
208 return 1;
209 }
210 close(nfd);
211
212 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
213 pr_err("Cannot add new disk to this array\n");
214 return 1;
215 }
216 /* Well, that seems to have worked.
217 * Now go through and update all superblocks
218 */
219
220 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
221 pr_err("cannot get array info for %s\n", devname);
222 return 1;
223 }
224
225 nd = d;
226 for (d=0 ; d < info.array.raid_disks ; d++) {
227 mdu_disk_info_t disk;
228 char *dv;
229
230 disk.number = d;
231 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
232 pr_err("cannot get device detail for device %d\n",
233 d);
234 return 1;
235 }
236 dv = map_dev(disk.major, disk.minor, 1);
237 if (!dv) {
238 pr_err("cannot find device file for device %d\n",
239 d);
240 return 1;
241 }
242 fd2 = dev_open(dv, O_RDWR);
243 if (fd2 < 0) {
244 pr_err("cannot open device file %s\n", dv);
245 return 1;
246 }
247 if (st->ss->load_super(st, fd2, NULL)) {
248 pr_err("cannot find super block on %s\n", dv);
249 close(fd);
250 return 1;
251 }
252 info.array.raid_disks = nd+1;
253 info.array.nr_disks = nd+1;
254 info.array.active_disks = nd+1;
255 info.array.working_disks = nd+1;
256
257 st->ss->update_super(st, &info, "linear-grow-update", dv,
258 0, 0, NULL);
259
260 if (st->ss->store_super(st, fd2)) {
261 pr_err("Cannot store new superblock on %s\n", dv);
262 close(fd2);
263 return 1;
264 }
265 close(fd2);
266 }
267
268 return 0;
269 }
270
271 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
272 {
273 /*
274 * First check that array doesn't have a bitmap
275 * Then create the bitmap
276 * Then add it
277 *
278 * For internal bitmaps, we need to check the version,
279 * find all the active devices, and write the bitmap block
280 * to all devices
281 */
282 mdu_bitmap_file_t bmf;
283 mdu_array_info_t array;
284 struct supertype *st;
285 char *subarray = NULL;
286 int major = BITMAP_MAJOR_HI;
287 int vers = md_get_version(fd);
288 unsigned long long bitmapsize, array_size;
289
290 if (vers < 9003) {
291 major = BITMAP_MAJOR_HOSTENDIAN;
292 pr_err("Warning - bitmaps created on this kernel"
293 " are not portable\n"
294 " between different architectures. Consider upgrading"
295 " the Linux kernel.\n");
296 }
297
298 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
299 if (errno == ENOMEM)
300 pr_err("Memory allocation failure.\n");
301 else
302 pr_err("bitmaps not supported by this kernel.\n");
303 return 1;
304 }
305 if (bmf.pathname[0]) {
306 if (strcmp(s->bitmap_file,"none")==0) {
307 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
308 pr_err("failed to remove bitmap %s\n",
309 bmf.pathname);
310 return 1;
311 }
312 return 0;
313 }
314 pr_err("%s already has a bitmap (%s)\n",
315 devname, bmf.pathname);
316 return 1;
317 }
318 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
319 pr_err("cannot get array status for %s\n", devname);
320 return 1;
321 }
322 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
323 if (strcmp(s->bitmap_file, "none")==0) {
324 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
325 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
326 pr_err("failed to remove internal bitmap.\n");
327 return 1;
328 }
329 return 0;
330 }
331 pr_err("Internal bitmap already present on %s\n",
332 devname);
333 return 1;
334 }
335
336 if (strcmp(s->bitmap_file, "none") == 0) {
337 pr_err("no bitmap found on %s\n", devname);
338 return 1;
339 }
340 if (array.level <= 0) {
341 pr_err("Bitmaps not meaningful with level %s\n",
342 map_num(pers, array.level)?:"of this array");
343 return 1;
344 }
345 bitmapsize = array.size;
346 bitmapsize <<= 1;
347 if (get_dev_size(fd, NULL, &array_size) &&
348 array_size > (0x7fffffffULL<<9)) {
349 /* Array is big enough that we cannot trust array.size
350 * try other approaches
351 */
352 bitmapsize = get_component_size(fd);
353 }
354 if (bitmapsize == 0) {
355 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
356 return 1;
357 }
358
359 if (array.level == 10) {
360 int ncopies = (array.layout&255)*((array.layout>>8)&255);
361 bitmapsize = bitmapsize * array.raid_disks / ncopies;
362 }
363
364 st = super_by_fd(fd, &subarray);
365 if (!st) {
366 pr_err("Cannot understand version %d.%d\n",
367 array.major_version, array.minor_version);
368 return 1;
369 }
370 if (subarray) {
371 pr_err("Cannot add bitmaps to sub-arrays yet\n");
372 free(subarray);
373 free(st);
374 return 1;
375 }
376 if (strcmp(s->bitmap_file, "internal") == 0) {
377 int rv;
378 int d;
379 int offset_setable = 0;
380 struct mdinfo *mdi;
381 if (st->ss->add_internal_bitmap == NULL) {
382 pr_err("Internal bitmaps not supported "
383 "with %s metadata\n", st->ss->name);
384 return 1;
385 }
386 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
387 if (mdi)
388 offset_setable = 1;
389 for (d=0; d< st->max_devs; d++) {
390 mdu_disk_info_t disk;
391 char *dv;
392 disk.number = d;
393 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
394 continue;
395 if (disk.major == 0 &&
396 disk.minor == 0)
397 continue;
398 if ((disk.state & (1<<MD_DISK_SYNC))==0)
399 continue;
400 dv = map_dev(disk.major, disk.minor, 1);
401 if (dv) {
402 int fd2 = dev_open(dv, O_RDWR);
403 if (fd2 < 0)
404 continue;
405 if (st->ss->load_super(st, fd2, NULL)==0) {
406 if (st->ss->add_internal_bitmap(
407 st,
408 &s->bitmap_chunk, c->delay, s->write_behind,
409 bitmapsize, offset_setable,
410 major)
411 )
412 st->ss->write_bitmap(st, fd2);
413 else {
414 pr_err("failed to create internal bitmap"
415 " - chunksize problem.\n");
416 close(fd2);
417 return 1;
418 }
419 }
420 close(fd2);
421 }
422 }
423 if (offset_setable) {
424 st->ss->getinfo_super(st, mdi, NULL);
425 sysfs_init(mdi, fd, NULL);
426 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
427 mdi->bitmap_offset);
428 } else {
429 array.state |= (1<<MD_SB_BITMAP_PRESENT);
430 rv = ioctl(fd, SET_ARRAY_INFO, &array);
431 }
432 if (rv < 0) {
433 if (errno == EBUSY)
434 pr_err("Cannot add bitmap while array is"
435 " resyncing or reshaping etc.\n");
436 pr_err("failed to set internal bitmap.\n");
437 return 1;
438 }
439 } else {
440 int uuid[4];
441 int bitmap_fd;
442 int d;
443 int max_devs = st->max_devs;
444
445 /* try to load a superblock */
446 for (d = 0; d < max_devs; d++) {
447 mdu_disk_info_t disk;
448 char *dv;
449 int fd2;
450 disk.number = d;
451 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
452 continue;
453 if ((disk.major==0 && disk.minor==0) ||
454 (disk.state & (1<<MD_DISK_REMOVED)))
455 continue;
456 dv = map_dev(disk.major, disk.minor, 1);
457 if (!dv)
458 continue;
459 fd2 = dev_open(dv, O_RDONLY);
460 if (fd2 >= 0) {
461 if (st->ss->load_super(st, fd2, NULL) == 0) {
462 close(fd2);
463 st->ss->uuid_from_super(st, uuid);
464 break;
465 }
466 close(fd2);
467 }
468 }
469 if (d == max_devs) {
470 pr_err("cannot find UUID for array!\n");
471 return 1;
472 }
473 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
474 c->delay, s->write_behind, bitmapsize, major)) {
475 return 1;
476 }
477 bitmap_fd = open(s->bitmap_file, O_RDWR);
478 if (bitmap_fd < 0) {
479 pr_err("weird: %s cannot be opened\n",
480 s->bitmap_file);
481 return 1;
482 }
483 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
484 int err = errno;
485 if (errno == EBUSY)
486 pr_err("Cannot add bitmap while array is"
487 " resyncing or reshaping etc.\n");
488 pr_err("Cannot set bitmap file for %s: %s\n",
489 devname, strerror(err));
490 return 1;
491 }
492 }
493
494 return 0;
495 }
496
497 /*
498 * When reshaping an array we might need to backup some data.
499 * This is written to all spares with a 'super_block' describing it.
500 * The superblock goes 4K from the end of the used space on the
501 * device.
502 * It if written after the backup is complete.
503 * It has the following structure.
504 */
505
506 static struct mdp_backup_super {
507 char magic[16]; /* md_backup_data-1 or -2 */
508 __u8 set_uuid[16];
509 __u64 mtime;
510 /* start/sizes in 512byte sectors */
511 __u64 devstart; /* address on backup device/file of data */
512 __u64 arraystart;
513 __u64 length;
514 __u32 sb_csum; /* csum of preceeding bytes. */
515 __u32 pad1;
516 __u64 devstart2; /* offset in to data of second section */
517 __u64 arraystart2;
518 __u64 length2;
519 __u32 sb_csum2; /* csum of preceeding bytes. */
520 __u8 pad[512-68-32];
521 } __attribute__((aligned(512))) bsb, bsb2;
522
523 static __u32 bsb_csum(char *buf, int len)
524 {
525 int i;
526 int csum = 0;
527 for (i = 0; i < len; i++)
528 csum = (csum<<3) + buf[0];
529 return __cpu_to_le32(csum);
530 }
531
532 static int check_idle(struct supertype *st)
533 {
534 /* Check that all member arrays for this container, or the
535 * container of this array, are idle
536 */
537 char *container = (st->container_devnm[0]
538 ? st->container_devnm : st->devnm);
539 struct mdstat_ent *ent, *e;
540 int is_idle = 1;
541
542 ent = mdstat_read(0, 0);
543 for (e = ent ; e; e = e->next) {
544 if (!is_container_member(e, container))
545 continue;
546 if (e->percent >= 0) {
547 is_idle = 0;
548 break;
549 }
550 }
551 free_mdstat(ent);
552 return is_idle;
553 }
554
555 static int freeze_container(struct supertype *st)
556 {
557 char *container = (st->container_devnm[0]
558 ? st->container_devnm : st->devnm);
559
560 if (!check_idle(st))
561 return -1;
562
563 if (block_monitor(container, 1)) {
564 pr_err("failed to freeze container\n");
565 return -2;
566 }
567
568 return 1;
569 }
570
571 static void unfreeze_container(struct supertype *st)
572 {
573 char *container = (st->container_devnm[0]
574 ? st->container_devnm : st->devnm);
575
576 unblock_monitor(container, 1);
577 }
578
579 static int freeze(struct supertype *st)
580 {
581 /* Try to freeze resync/rebuild on this array/container.
582 * Return -1 if the array is busy,
583 * return -2 container cannot be frozen,
584 * return 0 if this kernel doesn't support 'frozen'
585 * return 1 if it worked.
586 */
587 if (st->ss->external)
588 return freeze_container(st);
589 else {
590 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
591 int err;
592 char buf[20];
593
594 if (!sra)
595 return -1;
596 /* Need to clear any 'read-auto' status */
597 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
598 strncmp(buf, "read-auto", 9) == 0)
599 sysfs_set_str(sra, NULL, "array_state", "clean");
600
601 err = sysfs_freeze_array(sra);
602 sysfs_free(sra);
603 return err;
604 }
605 }
606
607 static void unfreeze(struct supertype *st)
608 {
609 if (st->ss->external)
610 return unfreeze_container(st);
611 else {
612 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
613
614 if (sra)
615 sysfs_set_str(sra, NULL, "sync_action", "idle");
616 sysfs_free(sra);
617 }
618 }
619
620 static void wait_reshape(struct mdinfo *sra)
621 {
622 int fd = sysfs_get_fd(sra, NULL, "sync_action");
623 char action[20];
624
625 if (fd < 0)
626 return;
627
628 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
629 strncmp(action, "reshape", 7) == 0) {
630 fd_set rfds;
631 FD_ZERO(&rfds);
632 FD_SET(fd, &rfds);
633 select(fd+1, NULL, NULL, &rfds, NULL);
634 }
635 close(fd);
636 }
637
638 static int reshape_super(struct supertype *st, unsigned long long size,
639 int level, int layout, int chunksize, int raid_disks,
640 int delta_disks, char *backup_file, char *dev,
641 int direction, int verbose)
642 {
643 /* nothing extra to check in the native case */
644 if (!st->ss->external)
645 return 0;
646 if (!st->ss->reshape_super ||
647 !st->ss->manage_reshape) {
648 pr_err("%s metadata does not support reshape\n",
649 st->ss->name);
650 return 1;
651 }
652
653 return st->ss->reshape_super(st, size, level, layout, chunksize,
654 raid_disks, delta_disks, backup_file, dev,
655 direction, verbose);
656 }
657
658 static void sync_metadata(struct supertype *st)
659 {
660 if (st->ss->external) {
661 if (st->update_tail) {
662 flush_metadata_updates(st);
663 st->update_tail = &st->updates;
664 } else
665 st->ss->sync_metadata(st);
666 }
667 }
668
669 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
670 {
671 /* when dealing with external metadata subarrays we need to be
672 * prepared to handle EAGAIN. The kernel may need to wait for
673 * mdmon to mark the array active so the kernel can handle
674 * allocations/writeback when preparing the reshape action
675 * (md_allow_write()). We temporarily disable safe_mode_delay
676 * to close a race with the array_state going clean before the
677 * next write to raid_disks / stripe_cache_size
678 */
679 char safe[50];
680 int rc;
681
682 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
683 if (!container ||
684 (strcmp(name, "raid_disks") != 0 &&
685 strcmp(name, "stripe_cache_size") != 0))
686 return sysfs_set_num(sra, NULL, name, n);
687
688 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
689 if (rc <= 0)
690 return -1;
691 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
692 rc = sysfs_set_num(sra, NULL, name, n);
693 if (rc < 0 && errno == EAGAIN) {
694 ping_monitor(container);
695 /* if we get EAGAIN here then the monitor is not active
696 * so stop trying
697 */
698 rc = sysfs_set_num(sra, NULL, name, n);
699 }
700 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
701 return rc;
702 }
703
704 int start_reshape(struct mdinfo *sra, int already_running,
705 int before_data_disks, int data_disks)
706 {
707 int err;
708 unsigned long long sync_max_to_set;
709
710 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
711 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
712 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
713 sra->reshape_progress);
714 if (before_data_disks <= data_disks)
715 sync_max_to_set = sra->reshape_progress / data_disks;
716 else
717 sync_max_to_set = (sra->component_size * data_disks
718 - sra->reshape_progress) / data_disks;
719 if (!already_running)
720 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
721 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
722 if (!already_running)
723 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
724
725 return err;
726 }
727
728 void abort_reshape(struct mdinfo *sra)
729 {
730 sysfs_set_str(sra, NULL, "sync_action", "idle");
731 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
732 sysfs_set_num(sra, NULL, "suspend_hi", 0);
733 sysfs_set_num(sra, NULL, "suspend_lo", 0);
734 sysfs_set_num(sra, NULL, "sync_min", 0);
735 sysfs_set_str(sra, NULL, "sync_max", "max");
736 }
737
738 int remove_disks_for_takeover(struct supertype *st,
739 struct mdinfo *sra,
740 int layout)
741 {
742 int nr_of_copies;
743 struct mdinfo *remaining;
744 int slot;
745
746 if (sra->array.level == 10)
747 nr_of_copies = layout & 0xff;
748 else if (sra->array.level == 1)
749 nr_of_copies = sra->array.raid_disks;
750 else
751 return 1;
752
753 remaining = sra->devs;
754 sra->devs = NULL;
755 /* for each 'copy', select one device and remove from the list. */
756 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
757 struct mdinfo **diskp;
758 int found = 0;
759
760 /* Find a working device to keep */
761 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
762 struct mdinfo *disk = *diskp;
763
764 if (disk->disk.raid_disk < slot)
765 continue;
766 if (disk->disk.raid_disk >= slot + nr_of_copies)
767 continue;
768 if (disk->disk.state & (1<<MD_DISK_REMOVED))
769 continue;
770 if (disk->disk.state & (1<<MD_DISK_FAULTY))
771 continue;
772 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
773 continue;
774
775 /* We have found a good disk to use! */
776 *diskp = disk->next;
777 disk->next = sra->devs;
778 sra->devs = disk;
779 found = 1;
780 break;
781 }
782 if (!found)
783 break;
784 }
785
786 if (slot < sra->array.raid_disks) {
787 /* didn't find all slots */
788 struct mdinfo **e;
789 e = &remaining;
790 while (*e)
791 e = &(*e)->next;
792 *e = sra->devs;
793 sra->devs = remaining;
794 return 1;
795 }
796
797 /* Remove all 'remaining' devices from the array */
798 while (remaining) {
799 struct mdinfo *sd = remaining;
800 remaining = sd->next;
801
802 sysfs_set_str(sra, sd, "state", "faulty");
803 sysfs_set_str(sra, sd, "slot", "none");
804 /* for external metadata disks should be removed in mdmon */
805 if (!st->ss->external)
806 sysfs_set_str(sra, sd, "state", "remove");
807 sd->disk.state |= (1<<MD_DISK_REMOVED);
808 sd->disk.state &= ~(1<<MD_DISK_SYNC);
809 sd->next = sra->devs;
810 sra->devs = sd;
811 }
812 return 0;
813 }
814
815 void reshape_free_fdlist(int *fdlist,
816 unsigned long long *offsets,
817 int size)
818 {
819 int i;
820
821 for (i = 0; i < size; i++)
822 if (fdlist[i] >= 0)
823 close(fdlist[i]);
824
825 free(fdlist);
826 free(offsets);
827 }
828
829 int reshape_prepare_fdlist(char *devname,
830 struct mdinfo *sra,
831 int raid_disks,
832 int nrdisks,
833 unsigned long blocks,
834 char *backup_file,
835 int *fdlist,
836 unsigned long long *offsets)
837 {
838 int d = 0;
839 struct mdinfo *sd;
840
841 for (d = 0; d <= nrdisks; d++)
842 fdlist[d] = -1;
843 d = raid_disks;
844 for (sd = sra->devs; sd; sd = sd->next) {
845 if (sd->disk.state & (1<<MD_DISK_FAULTY))
846 continue;
847 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
848 char *dn = map_dev(sd->disk.major,
849 sd->disk.minor, 1);
850 fdlist[sd->disk.raid_disk]
851 = dev_open(dn, O_RDONLY);
852 offsets[sd->disk.raid_disk] = sd->data_offset*512;
853 if (fdlist[sd->disk.raid_disk] < 0) {
854 pr_err("%s: cannot open component %s\n",
855 devname, dn ? dn : "-unknown-");
856 d = -1;
857 goto release;
858 }
859 } else if (backup_file == NULL) {
860 /* spare */
861 char *dn = map_dev(sd->disk.major,
862 sd->disk.minor, 1);
863 fdlist[d] = dev_open(dn, O_RDWR);
864 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
865 if (fdlist[d] < 0) {
866 pr_err("%s: cannot open component %s\n",
867 devname, dn ? dn : "-unknown-");
868 d = -1;
869 goto release;
870 }
871 d++;
872 }
873 }
874 release:
875 return d;
876 }
877
878 int reshape_open_backup_file(char *backup_file,
879 int fd,
880 char *devname,
881 long blocks,
882 int *fdlist,
883 unsigned long long *offsets,
884 int restart)
885 {
886 /* Return 1 on success, 0 on any form of failure */
887 /* need to check backup file is large enough */
888 char buf[512];
889 struct stat stb;
890 unsigned int dev;
891 int i;
892
893 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
894 S_IRUSR | S_IWUSR);
895 *offsets = 8 * 512;
896 if (*fdlist < 0) {
897 pr_err("%s: cannot create backup file %s: %s\n",
898 devname, backup_file, strerror(errno));
899 return 0;
900 }
901 /* Guard against backup file being on array device.
902 * If array is partitioned or if LVM etc is in the
903 * way this will not notice, but it is better than
904 * nothing.
905 */
906 fstat(*fdlist, &stb);
907 dev = stb.st_dev;
908 fstat(fd, &stb);
909 if (stb.st_rdev == dev) {
910 pr_err("backup file must NOT be"
911 " on the array being reshaped.\n");
912 close(*fdlist);
913 return 0;
914 }
915
916 memset(buf, 0, 512);
917 for (i=0; i < blocks + 8 ; i++) {
918 if (write(*fdlist, buf, 512) != 512) {
919 pr_err("%s: cannot create"
920 " backup file %s: %s\n",
921 devname, backup_file, strerror(errno));
922 return 0;
923 }
924 }
925 if (fsync(*fdlist) != 0) {
926 pr_err("%s: cannot create backup file %s: %s\n",
927 devname, backup_file, strerror(errno));
928 return 0;
929 }
930
931 return 1;
932 }
933
934 unsigned long GCD(unsigned long a, unsigned long b)
935 {
936 while (a != b) {
937 if (a < b)
938 b -= a;
939 if (b < a)
940 a -= b;
941 }
942 return a;
943 }
944
945 unsigned long compute_backup_blocks(int nchunk, int ochunk,
946 unsigned int ndata, unsigned int odata)
947 {
948 unsigned long a, b, blocks;
949 /* So how much do we need to backup.
950 * We need an amount of data which is both a whole number of
951 * old stripes and a whole number of new stripes.
952 * So LCM for (chunksize*datadisks).
953 */
954 a = (ochunk/512) * odata;
955 b = (nchunk/512) * ndata;
956 /* Find GCD */
957 a = GCD(a, b);
958 /* LCM == product / GCD */
959 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
960
961 return blocks;
962 }
963
964 char *analyse_change(struct mdinfo *info, struct reshape *re)
965 {
966 /* Based on the current array state in info->array and
967 * the changes in info->new_* etc, determine:
968 * - whether the change is possible
969 * - Intermediate level/raid_disks/layout
970 * - whether a restriping reshape is needed
971 * - number of sectors in minimum change unit. This
972 * will cover a whole number of stripes in 'before' and
973 * 'after'.
974 *
975 * Return message if the change should be rejected
976 * NULL if the change can be achieved
977 *
978 * This can be called as part of starting a reshape, or
979 * when assembling an array that is undergoing reshape.
980 */
981 int near, far, offset, copies;
982 int new_disks;
983 int old_chunk, new_chunk;
984 /* delta_parity records change in number of devices
985 * caused by level change
986 */
987 int delta_parity = 0;
988
989 memset(re, 0, sizeof(*re));
990
991 /* If a new level not explicitly given, we assume no-change */
992 if (info->new_level == UnSet)
993 info->new_level = info->array.level;
994
995 if (info->new_chunk)
996 switch (info->new_level) {
997 case 0:
998 case 4:
999 case 5:
1000 case 6:
1001 case 10:
1002 /* chunk size is meaningful, must divide component_size
1003 * evenly
1004 */
1005 if (info->component_size % (info->new_chunk/512))
1006 return "New chunk size does not"
1007 " divide component size";
1008 break;
1009 default:
1010 return "chunk size not meaningful for this level";
1011 }
1012 else
1013 info->new_chunk = info->array.chunk_size;
1014
1015 switch (info->array.level) {
1016 default:
1017 return "Cannot understand this RAID level";
1018 case 1:
1019 /* RAID1 can convert to RAID1 with different disks, or
1020 * raid5 with 2 disks, or
1021 * raid0 with 1 disk
1022 */
1023 if (info->new_level > 1 &&
1024 (info->component_size & 7))
1025 return "Cannot convert RAID1 of this size - "
1026 "reduce size to multiple of 4K first.";
1027 if (info->new_level == 0) {
1028 if (info->delta_disks != UnSet &&
1029 info->delta_disks != 0)
1030 return "Cannot change number of disks "
1031 "with RAID1->RAID0 conversion";
1032 re->level = 0;
1033 re->before.data_disks = 1;
1034 re->after.data_disks = 1;
1035 return NULL;
1036 }
1037 if (info->new_level == 1) {
1038 if (info->delta_disks == UnSet)
1039 /* Don't know what to do */
1040 return "no change requested for Growing RAID1";
1041 re->level = 1;
1042 return NULL;
1043 }
1044 if (info->array.raid_disks == 2 &&
1045 info->new_level == 5) {
1046
1047 re->level = 5;
1048 re->before.data_disks = 1;
1049 if (info->delta_disks != UnSet &&
1050 info->delta_disks != 0)
1051 re->after.data_disks = 1 + info->delta_disks;
1052 else
1053 re->after.data_disks = 1;
1054 if (re->after.data_disks < 1)
1055 return "Number of disks too small for RAID5";
1056
1057 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1058 info->array.chunk_size = 65536;
1059 break;
1060 }
1061 /* Could do some multi-stage conversions, but leave that to
1062 * later.
1063 */
1064 return "Impossibly level change request for RAID1";
1065
1066 case 10:
1067 /* RAID10 can be converted from near mode to
1068 * RAID0 by removing some devices.
1069 * It can also be reshaped if the kernel supports
1070 * new_data_offset.
1071 */
1072 switch (info->new_level) {
1073 case 0:
1074 if ((info->array.layout & ~0xff) != 0x100)
1075 return "Cannot Grow RAID10 with far/offset layout";
1076 /* number of devices must be multiple of number of copies */
1077 if (info->array.raid_disks % (info->array.layout & 0xff))
1078 return "RAID10 layout too complex for Grow operation";
1079
1080 new_disks = (info->array.raid_disks
1081 / (info->array.layout & 0xff));
1082 if (info->delta_disks == UnSet)
1083 info->delta_disks = (new_disks
1084 - info->array.raid_disks);
1085
1086 if (info->delta_disks != new_disks - info->array.raid_disks)
1087 return "New number of raid-devices impossible for RAID10";
1088 if (info->new_chunk &&
1089 info->new_chunk != info->array.chunk_size)
1090 return "Cannot change chunk-size with RAID10 Grow";
1091
1092 /* looks good */
1093 re->level = 0;
1094 re->before.data_disks = new_disks;
1095 re->after.data_disks = re->before.data_disks;
1096 return NULL;
1097
1098 case 10:
1099 near = info->array.layout & 0xff;
1100 far = (info->array.layout >> 8) & 0xff;
1101 offset = info->array.layout & 0x10000;
1102 if (far > 1 && !offset)
1103 return "Cannot reshape RAID10 in far-mode";
1104 copies = near * far;
1105
1106 old_chunk = info->array.chunk_size * far;
1107
1108 if (info->new_layout == UnSet)
1109 info->new_layout = info->array.layout;
1110 else {
1111 near = info->new_layout & 0xff;
1112 far = (info->new_layout >> 8) & 0xff;
1113 offset = info->new_layout & 0x10000;
1114 if (far > 1 && !offset)
1115 return "Cannot reshape RAID10 to far-mode";
1116 if (near * far != copies)
1117 return "Cannot change number of copies"
1118 " when reshaping RAID10";
1119 }
1120 if (info->delta_disks == UnSet)
1121 info->delta_disks = 0;
1122 new_disks = (info->array.raid_disks +
1123 info->delta_disks);
1124
1125 new_chunk = info->new_chunk * far;
1126
1127 re->level = 10;
1128 re->before.layout = info->array.layout;
1129 re->before.data_disks = info->array.raid_disks;
1130 re->after.layout = info->new_layout;
1131 re->after.data_disks = new_disks;
1132 /* For RAID10 we don't do backup but do allow reshape,
1133 * so set backup_blocks to INVALID_SECTORS rather than
1134 * zero.
1135 * And there is no need to synchronise stripes on both
1136 * 'old' and 'new'. So the important
1137 * number is the minimum data_offset difference
1138 * which is the larger of (offset copies * chunk).
1139 */
1140 re->backup_blocks = INVALID_SECTORS;
1141 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1142 if (new_disks < re->before.data_disks &&
1143 info->space_after < re->min_offset_change)
1144 /* Reduce component size by one chunk */
1145 re->new_size = (info->component_size -
1146 re->min_offset_change);
1147 else
1148 re->new_size = info->component_size;
1149 re->new_size = re->new_size * new_disks / copies;
1150 return NULL;
1151
1152 default:
1153 return "RAID10 can only be changed to RAID0";
1154 }
1155 case 0:
1156 /* RAID0 can be converted to RAID10, or to RAID456 */
1157 if (info->new_level == 10) {
1158 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1159 /* Assume near=2 layout */
1160 info->new_layout = 0x102;
1161 info->delta_disks = info->array.raid_disks;
1162 }
1163 if (info->new_layout == UnSet) {
1164 int copies = 1 + (info->delta_disks
1165 / info->array.raid_disks);
1166 if (info->array.raid_disks * (copies-1)
1167 != info->delta_disks)
1168 return "Impossible number of devices"
1169 " for RAID0->RAID10";
1170 info->new_layout = 0x100 + copies;
1171 }
1172 if (info->delta_disks == UnSet) {
1173 int copies = info->new_layout & 0xff;
1174 if (info->new_layout != 0x100 + copies)
1175 return "New layout impossible"
1176 " for RAID0->RAID10";;
1177 info->delta_disks = (copies - 1) *
1178 info->array.raid_disks;
1179 }
1180 if (info->new_chunk &&
1181 info->new_chunk != info->array.chunk_size)
1182 return "Cannot change chunk-size with RAID0->RAID10";
1183 /* looks good */
1184 re->level = 10;
1185 re->before.data_disks = (info->array.raid_disks +
1186 info->delta_disks);
1187 re->after.data_disks = re->before.data_disks;
1188 re->before.layout = info->new_layout;
1189 return NULL;
1190 }
1191
1192 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1193 * a raid4 style layout of the final level.
1194 */
1195 switch (info->new_level) {
1196 case 4:
1197 delta_parity = 1;
1198 case 0:
1199 re->level = 4;
1200 re->before.layout = 0;
1201 break;
1202 case 5:
1203 delta_parity = 1;
1204 re->level = 5;
1205 re->before.layout = ALGORITHM_PARITY_N;
1206 break;
1207 case 6:
1208 delta_parity = 2;
1209 re->level = 6;
1210 re->before.layout = ALGORITHM_PARITY_N;
1211 break;
1212 default:
1213 return "Impossible level change requested";
1214 }
1215 re->before.data_disks = info->array.raid_disks;
1216 /* determining 'after' layout happens outside this 'switch' */
1217 break;
1218
1219 case 4:
1220 info->array.layout = ALGORITHM_PARITY_N;
1221 case 5:
1222 switch (info->new_level) {
1223 case 0:
1224 delta_parity = -1;
1225 case 4:
1226 re->level = info->array.level;
1227 re->before.data_disks = info->array.raid_disks - 1;
1228 re->before.layout = info->array.layout;
1229 break;
1230 case 5:
1231 re->level = 5;
1232 re->before.data_disks = info->array.raid_disks - 1;
1233 re->before.layout = info->array.layout;
1234 break;
1235 case 6:
1236 delta_parity = 1;
1237 re->level = 6;
1238 re->before.data_disks = info->array.raid_disks - 1;
1239 switch (info->array.layout) {
1240 case ALGORITHM_LEFT_ASYMMETRIC:
1241 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1242 break;
1243 case ALGORITHM_RIGHT_ASYMMETRIC:
1244 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1245 break;
1246 case ALGORITHM_LEFT_SYMMETRIC:
1247 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1248 break;
1249 case ALGORITHM_RIGHT_SYMMETRIC:
1250 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1251 break;
1252 case ALGORITHM_PARITY_0:
1253 re->before.layout = ALGORITHM_PARITY_0_6;
1254 break;
1255 case ALGORITHM_PARITY_N:
1256 re->before.layout = ALGORITHM_PARITY_N_6;
1257 break;
1258 default:
1259 return "Cannot convert an array with this layout";
1260 }
1261 break;
1262 case 1:
1263 if (info->array.raid_disks != 2)
1264 return "Can only convert a 2-device array to RAID1";
1265 if (info->delta_disks != UnSet &&
1266 info->delta_disks != 0)
1267 return "Cannot set raid_disk when "
1268 "converting RAID5->RAID1";
1269 re->level = 1;
1270 info->new_chunk = 0;
1271 return NULL;
1272 default:
1273 return "Impossible level change requested";
1274 }
1275 break;
1276 case 6:
1277 switch (info->new_level) {
1278 case 4:
1279 case 5:
1280 delta_parity = -1;
1281 case 6:
1282 re->level = 6;
1283 re->before.data_disks = info->array.raid_disks - 2;
1284 re->before.layout = info->array.layout;
1285 break;
1286 default:
1287 return "Impossible level change requested";
1288 }
1289 break;
1290 }
1291
1292 /* If we reached here then it looks like a re-stripe is
1293 * happening. We have determined the intermediate level
1294 * and initial raid_disks/layout and stored these in 're'.
1295 *
1296 * We need to deduce the final layout that can be atomically
1297 * converted to the end state.
1298 */
1299 switch (info->new_level) {
1300 case 0:
1301 /* We can only get to RAID0 from RAID4 or RAID5
1302 * with appropriate layout and one extra device
1303 */
1304 if (re->level != 4 && re->level != 5)
1305 return "Cannot covert to RAID0 from this level";
1306
1307 switch (re->level) {
1308 case 4:
1309 re->before.layout = 0;
1310 re->after.layout = 0;
1311 break;
1312 case 5:
1313 re->after.layout = ALGORITHM_PARITY_N;
1314 break;
1315 }
1316 break;
1317
1318 case 4:
1319 /* We can only get to RAID4 from RAID5 */
1320 if (re->level != 4 && re->level != 5)
1321 return "Cannot convert to RAID4 from this level";
1322
1323 switch (re->level) {
1324 case 4:
1325 re->before.layout = 0;
1326 re->after.layout = 0;
1327 break;
1328 case 5:
1329 re->after.layout = ALGORITHM_PARITY_N;
1330 break;
1331 }
1332 break;
1333
1334 case 5:
1335 /* We get to RAID5 from RAID5 or RAID6 */
1336 if (re->level != 5 && re->level != 6)
1337 return "Cannot convert to RAID5 from this level";
1338
1339 switch (re->level) {
1340 case 5:
1341 if (info->new_layout == UnSet)
1342 re->after.layout = re->before.layout;
1343 else
1344 re->after.layout = info->new_layout;
1345 break;
1346 case 6:
1347 if (info->new_layout == UnSet)
1348 info->new_layout = re->before.layout;
1349
1350 /* after.layout needs to be raid6 version of new_layout */
1351 if (info->new_layout == ALGORITHM_PARITY_N)
1352 re->after.layout = ALGORITHM_PARITY_N;
1353 else {
1354 char layout[40];
1355 char *ls = map_num(r5layout, info->new_layout);
1356 int l;
1357 if (ls) {
1358 /* Current RAID6 layout has a RAID5
1359 * equivalent - good
1360 */
1361 strcat(strcpy(layout, ls), "-6");
1362 l = map_name(r6layout, layout);
1363 if (l == UnSet)
1364 return "Cannot find RAID6 layout"
1365 " to convert to";
1366 } else {
1367 /* Current RAID6 has no equivalent.
1368 * If it is already a '-6' layout we
1369 * can leave it unchanged, else we must
1370 * fail
1371 */
1372 ls = map_num(r6layout, info->new_layout);
1373 if (!ls ||
1374 strcmp(ls+strlen(ls)-2, "-6") != 0)
1375 return "Please specify new layout";
1376 l = info->new_layout;
1377 }
1378 re->after.layout = l;
1379 }
1380 }
1381 break;
1382
1383 case 6:
1384 /* We must already be at level 6 */
1385 if (re->level != 6)
1386 return "Impossible level change";
1387 if (info->new_layout == UnSet)
1388 re->after.layout = info->array.layout;
1389 else
1390 re->after.layout = info->new_layout;
1391 break;
1392 default:
1393 return "Impossible level change requested";
1394 }
1395 if (info->delta_disks == UnSet)
1396 info->delta_disks = delta_parity;
1397
1398 re->after.data_disks = (re->before.data_disks
1399 + info->delta_disks
1400 - delta_parity);
1401 switch (re->level) {
1402 case 6: re->parity = 2;
1403 break;
1404 case 4:
1405 case 5: re->parity = 1;
1406 break;
1407 default: re->parity = 0;
1408 break;
1409 }
1410 /* So we have a restripe operation, we need to calculate the number
1411 * of blocks per reshape operation.
1412 */
1413 if (info->new_chunk == 0)
1414 info->new_chunk = info->array.chunk_size;
1415 if (re->after.data_disks == re->before.data_disks &&
1416 re->after.layout == re->before.layout &&
1417 info->new_chunk == info->array.chunk_size) {
1418 /* Nothing to change, can change level immediately. */
1419 re->level = info->new_level;
1420 re->backup_blocks = 0;
1421 return NULL;
1422 }
1423 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1424 /* chunk and layout changes make no difference */
1425 re->level = info->new_level;
1426 re->backup_blocks = 0;
1427 return NULL;
1428 }
1429
1430 if (re->after.data_disks == re->before.data_disks &&
1431 get_linux_version() < 2006032)
1432 return "in-place reshape is not safe before 2.6.32 - sorry.";
1433
1434 if (re->after.data_disks < re->before.data_disks &&
1435 get_linux_version() < 2006030)
1436 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1437
1438 re->backup_blocks = compute_backup_blocks(
1439 info->new_chunk, info->array.chunk_size,
1440 re->after.data_disks,
1441 re->before.data_disks);
1442 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1443
1444 re->new_size = info->component_size * re->after.data_disks;
1445 return NULL;
1446 }
1447
1448 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1449 char *text_version)
1450 {
1451 struct mdinfo *info;
1452 char *subarray;
1453 int ret_val = -1;
1454
1455 if ((st == NULL) || (sra == NULL))
1456 return ret_val;
1457
1458 if (text_version == NULL)
1459 text_version = sra->text_version;
1460 subarray = strchr(text_version+1, '/')+1;
1461 info = st->ss->container_content(st, subarray);
1462 if (info) {
1463 unsigned long long current_size = 0;
1464 unsigned long long new_size =
1465 info->custom_array_size/2;
1466
1467 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1468 new_size > current_size) {
1469 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1470 < 0)
1471 dprintf("Error: Cannot set array size");
1472 else {
1473 ret_val = 0;
1474 dprintf("Array size changed");
1475 }
1476 dprintf(" from %llu to %llu.\n",
1477 current_size, new_size);
1478 }
1479 sysfs_free(info);
1480 } else
1481 dprintf("Error: set_array_size(): info pointer in NULL\n");
1482
1483 return ret_val;
1484 }
1485
1486 static int reshape_array(char *container, int fd, char *devname,
1487 struct supertype *st, struct mdinfo *info,
1488 int force, struct mddev_dev *devlist,
1489 unsigned long long data_offset,
1490 char *backup_file, int verbose, int forked,
1491 int restart, int freeze_reshape);
1492 static int reshape_container(char *container, char *devname,
1493 int mdfd,
1494 struct supertype *st,
1495 struct mdinfo *info,
1496 int force,
1497 char *backup_file,
1498 int verbose, int restart, int freeze_reshape);
1499
1500 int Grow_reshape(char *devname, int fd,
1501 struct mddev_dev *devlist,
1502 unsigned long long data_offset,
1503 struct context *c, struct shape *s)
1504 {
1505 /* Make some changes in the shape of an array.
1506 * The kernel must support the change.
1507 *
1508 * There are three different changes. Each can trigger
1509 * a resync or recovery so we freeze that until we have
1510 * requested everything (if kernel supports freezing - 2.6.30).
1511 * The steps are:
1512 * - change size (i.e. component_size)
1513 * - change level
1514 * - change layout/chunksize/ndisks
1515 *
1516 * The last can require a reshape. It is different on different
1517 * levels so we need to check the level before actioning it.
1518 * Some times the level change needs to be requested after the
1519 * reshape (e.g. raid6->raid5, raid5->raid0)
1520 *
1521 */
1522 struct mdu_array_info_s array;
1523 int rv = 0;
1524 struct supertype *st;
1525 char *subarray = NULL;
1526
1527 int frozen;
1528 int changed = 0;
1529 char *container = NULL;
1530 int cfd = -1;
1531
1532 struct mddev_dev *dv;
1533 int added_disks;
1534
1535 struct mdinfo info;
1536 struct mdinfo *sra;
1537
1538
1539 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1540 pr_err("%s is not an active md array - aborting\n",
1541 devname);
1542 return 1;
1543 }
1544 if (data_offset != INVALID_SECTORS && array.level != 10
1545 && (array.level < 4 || array.level > 6)) {
1546 pr_err("--grow --data-offset not yet supported\n");
1547 return 1;
1548 }
1549
1550 if (s->size > 0 &&
1551 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1552 pr_err("cannot change component size at the same time "
1553 "as other changes.\n"
1554 " Change size first, then check data is intact before "
1555 "making other changes.\n");
1556 return 1;
1557 }
1558
1559 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1560 get_linux_version() < 2006032 &&
1561 !check_env("MDADM_FORCE_FEWER")) {
1562 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1563 " Please use a newer kernel\n");
1564 return 1;
1565 }
1566
1567 st = super_by_fd(fd, &subarray);
1568 if (!st) {
1569 pr_err("Unable to determine metadata format for %s\n", devname);
1570 return 1;
1571 }
1572 if (s->raiddisks > st->max_devs) {
1573 pr_err("Cannot increase raid-disks on this array"
1574 " beyond %d\n", st->max_devs);
1575 return 1;
1576 }
1577
1578 /* in the external case we need to check that the requested reshape is
1579 * supported, and perform an initial check that the container holds the
1580 * pre-requisite spare devices (mdmon owns final validation)
1581 */
1582 if (st->ss->external) {
1583 int rv;
1584
1585 if (subarray) {
1586 container = st->container_devnm;
1587 cfd = open_dev_excl(st->container_devnm);
1588 } else {
1589 container = st->devnm;
1590 close(fd);
1591 cfd = open_dev_excl(st->devnm);
1592 fd = cfd;
1593 }
1594 if (cfd < 0) {
1595 pr_err("Unable to open container for %s\n",
1596 devname);
1597 free(subarray);
1598 return 1;
1599 }
1600
1601 rv = st->ss->load_container(st, cfd, NULL);
1602
1603 if (rv) {
1604 pr_err("Cannot read superblock for %s\n",
1605 devname);
1606 free(subarray);
1607 return 1;
1608 }
1609
1610 /* check if operation is supported for metadata handler */
1611 if (st->ss->container_content) {
1612 struct mdinfo *cc = NULL;
1613 struct mdinfo *content = NULL;
1614
1615 cc = st->ss->container_content(st, subarray);
1616 for (content = cc; content ; content = content->next) {
1617 int allow_reshape = 1;
1618
1619 /* check if reshape is allowed based on metadata
1620 * indications stored in content.array.status
1621 */
1622 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1623 allow_reshape = 0;
1624 if (content->array.state
1625 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1626 allow_reshape = 0;
1627 if (!allow_reshape) {
1628 pr_err("cannot reshape arrays in"
1629 " container with unsupported"
1630 " metadata: %s(%s)\n",
1631 devname, container);
1632 sysfs_free(cc);
1633 free(subarray);
1634 return 1;
1635 }
1636 }
1637 sysfs_free(cc);
1638 }
1639 if (mdmon_running(container))
1640 st->update_tail = &st->updates;
1641 }
1642
1643 added_disks = 0;
1644 for (dv = devlist; dv; dv = dv->next)
1645 added_disks++;
1646 if (s->raiddisks > array.raid_disks &&
1647 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1648 !c->force) {
1649 pr_err("Need %d spare%s to avoid degraded array,"
1650 " and only have %d.\n"
1651 " Use --force to over-ride this check.\n",
1652 s->raiddisks - array.raid_disks,
1653 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1654 array.spare_disks + added_disks);
1655 return 1;
1656 }
1657
1658 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1659 | GET_STATE | GET_VERSION);
1660 if (sra) {
1661 if (st->ss->external && subarray == NULL) {
1662 array.level = LEVEL_CONTAINER;
1663 sra->array.level = LEVEL_CONTAINER;
1664 }
1665 } else {
1666 pr_err("failed to read sysfs parameters for %s\n",
1667 devname);
1668 return 1;
1669 }
1670 frozen = freeze(st);
1671 if (frozen < -1) {
1672 /* freeze() already spewed the reason */
1673 sysfs_free(sra);
1674 return 1;
1675 } else if (frozen < 0) {
1676 pr_err("%s is performing resync/recovery and cannot"
1677 " be reshaped\n", devname);
1678 sysfs_free(sra);
1679 return 1;
1680 }
1681
1682 /* ========= set size =============== */
1683 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1684 unsigned long long orig_size = get_component_size(fd)/2;
1685 unsigned long long min_csize;
1686 struct mdinfo *mdi;
1687 int raid0_takeover = 0;
1688
1689 if (orig_size == 0)
1690 orig_size = (unsigned) array.size;
1691
1692 if (orig_size == 0) {
1693 pr_err("Cannot set device size in this type of array.\n");
1694 rv = 1;
1695 goto release;
1696 }
1697
1698 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1699 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1700 rv = 1;
1701 goto release;
1702 }
1703 sync_metadata(st);
1704 if (st->ss->external) {
1705 /* metadata can have size limitation
1706 * update size value according to metadata information
1707 */
1708 struct mdinfo *sizeinfo =
1709 st->ss->container_content(st, subarray);
1710 if (sizeinfo) {
1711 unsigned long long new_size =
1712 sizeinfo->custom_array_size/2;
1713 int data_disks = get_data_disks(
1714 sizeinfo->array.level,
1715 sizeinfo->array.layout,
1716 sizeinfo->array.raid_disks);
1717 new_size /= data_disks;
1718 dprintf("Metadata size correction from %llu to "
1719 "%llu (%llu)\n", orig_size, new_size,
1720 new_size * data_disks);
1721 s->size = new_size;
1722 sysfs_free(sizeinfo);
1723 }
1724 }
1725
1726 /* Update the size of each member device in case
1727 * they have been resized. This will never reduce
1728 * below the current used-size. The "size" attribute
1729 * understands '0' to mean 'max'.
1730 */
1731 min_csize = 0;
1732 rv = 0;
1733 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1734 if (sysfs_set_num(sra, mdi, "size",
1735 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1736 /* Probably kernel refusing to let us
1737 * reduce the size - not an error.
1738 */
1739 break;
1740 }
1741 if (array.not_persistent == 0 &&
1742 array.major_version == 0 &&
1743 get_linux_version() < 3001000) {
1744 /* Dangerous to allow size to exceed 2TB */
1745 unsigned long long csize;
1746 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1747 if (csize >= 2ULL*1024*1024*1024)
1748 csize = 2ULL*1024*1024*1024;
1749 if ((min_csize == 0 || (min_csize
1750 > csize)))
1751 min_csize = csize;
1752 }
1753 }
1754 }
1755 if (rv) {
1756 pr_err("Cannot set size on "
1757 "array members.\n");
1758 goto size_change_error;
1759 }
1760 if (min_csize && s->size > min_csize) {
1761 pr_err("Cannot safely make this array "
1762 "use more than 2TB per device on this kernel.\n");
1763 rv = 1;
1764 goto size_change_error;
1765 }
1766 if (min_csize && s->size == MAX_SIZE) {
1767 /* Don't let the kernel choose a size - it will get
1768 * it wrong
1769 */
1770 pr_err("Limited v0.90 array to "
1771 "2TB per device\n");
1772 s->size = min_csize;
1773 }
1774 if (st->ss->external) {
1775 if (sra->array.level == 0) {
1776 rv = sysfs_set_str(sra, NULL, "level",
1777 "raid5");
1778 if (!rv) {
1779 raid0_takeover = 1;
1780 /* get array parametes after takeover
1781 * to chane one parameter at time only
1782 */
1783 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1784 }
1785 }
1786 /* make sure mdmon is
1787 * aware of the new level */
1788 if (!mdmon_running(st->container_devnm))
1789 start_mdmon(st->container_devnm);
1790 ping_monitor(container);
1791 if (mdmon_running(st->container_devnm) &&
1792 st->update_tail == NULL)
1793 st->update_tail = &st->updates;
1794 }
1795
1796 if (s->size == MAX_SIZE)
1797 s->size = 0;
1798 array.size = s->size;
1799 if ((unsigned)array.size != s->size) {
1800 /* got truncated to 32bit, write to
1801 * component_size instead
1802 */
1803 if (sra)
1804 rv = sysfs_set_num(sra, NULL,
1805 "component_size", s->size);
1806 else
1807 rv = -1;
1808 } else {
1809 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1810
1811 /* manage array size when it is managed externally
1812 */
1813 if ((rv == 0) && st->ss->external)
1814 rv = set_array_size(st, sra, sra->text_version);
1815 }
1816
1817 if (raid0_takeover) {
1818 /* do not recync non-existing parity,
1819 * we will drop it anyway
1820 */
1821 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1822 /* go back to raid0, drop parity disk
1823 */
1824 sysfs_set_str(sra, NULL, "level", "raid0");
1825 ioctl(fd, GET_ARRAY_INFO, &array);
1826 }
1827
1828 size_change_error:
1829 if (rv != 0) {
1830 int err = errno;
1831
1832 /* restore metadata */
1833 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1834 UnSet, NULL, devname,
1835 ROLLBACK_METADATA_CHANGES,
1836 c->verbose) == 0)
1837 sync_metadata(st);
1838 pr_err("Cannot set device size for %s: %s\n",
1839 devname, strerror(err));
1840 if (err == EBUSY &&
1841 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1842 cont_err("Bitmap must be removed before size can be changed\n");
1843 rv = 1;
1844 goto release;
1845 }
1846 if (s->assume_clean) {
1847 /* This will fail on kernels older than 3.0 unless
1848 * a backport has been arranged.
1849 */
1850 if (sra == NULL ||
1851 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1852 pr_err("--assume-clean not supported with --grow on this kernel\n");
1853 }
1854 ioctl(fd, GET_ARRAY_INFO, &array);
1855 s->size = get_component_size(fd)/2;
1856 if (s->size == 0)
1857 s->size = array.size;
1858 if (c->verbose >= 0) {
1859 if (s->size == orig_size)
1860 pr_err("component size of %s "
1861 "unchanged at %lluK\n",
1862 devname, s->size);
1863 else
1864 pr_err("component size of %s "
1865 "has been set to %lluK\n",
1866 devname, s->size);
1867 }
1868 changed = 1;
1869 } else if (array.level != LEVEL_CONTAINER) {
1870 s->size = get_component_size(fd)/2;
1871 if (s->size == 0)
1872 s->size = array.size;
1873 }
1874
1875 /* See if there is anything else to do */
1876 if ((s->level == UnSet || s->level == array.level) &&
1877 (s->layout_str == NULL) &&
1878 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1879 data_offset == INVALID_SECTORS &&
1880 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1881 /* Nothing more to do */
1882 if (!changed && c->verbose >= 0)
1883 pr_err("%s: no change requested\n",
1884 devname);
1885 goto release;
1886 }
1887
1888 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1889 * current implementation assumes that following conditions must be met:
1890 * - RAID10:
1891 * - far_copies == 1
1892 * - near_copies == 2
1893 */
1894 if ((s->level == 0 && array.level == 10 && sra &&
1895 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1896 (s->level == 0 && array.level == 1 && sra)) {
1897 int err;
1898 err = remove_disks_for_takeover(st, sra, array.layout);
1899 if (err) {
1900 dprintf(Name": Array cannot be reshaped\n");
1901 if (cfd > -1)
1902 close(cfd);
1903 rv = 1;
1904 goto release;
1905 }
1906 /* Make sure mdmon has seen the device removal
1907 * and updated metadata before we continue with
1908 * level change
1909 */
1910 if (container)
1911 ping_monitor(container);
1912 }
1913
1914 memset(&info, 0, sizeof(info));
1915 info.array = array;
1916 sysfs_init(&info, fd, NULL);
1917 strcpy(info.text_version, sra->text_version);
1918 info.component_size = s->size*2;
1919 info.new_level = s->level;
1920 info.new_chunk = s->chunk * 1024;
1921 if (info.array.level == LEVEL_CONTAINER) {
1922 info.delta_disks = UnSet;
1923 info.array.raid_disks = s->raiddisks;
1924 } else if (s->raiddisks)
1925 info.delta_disks = s->raiddisks - info.array.raid_disks;
1926 else
1927 info.delta_disks = UnSet;
1928 if (s->layout_str == NULL) {
1929 info.new_layout = UnSet;
1930 if (info.array.level == 6 &&
1931 (info.new_level == 6 || info.new_level == UnSet) &&
1932 info.array.layout >= 16) {
1933 pr_err("%s has a non-standard layout. If you"
1934 " wish to preserve this\n", devname);
1935 cont_err("during the reshape, please specify"
1936 " --layout=preserve\n");
1937 cont_err("If you want to change it, specify a"
1938 " 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"
1956 " a RAID6 array.\n", s->layout_str);
1957 rv = 1;
1958 goto release;
1959 }
1960 } else if (strcmp(s->layout_str, "preserve") == 0) {
1961 /* This means that a non-standard RAID6 layout
1962 * is OK.
1963 * In particular:
1964 * - When reshape a RAID6 (e.g. adding a device)
1965 * which is in a non-standard layout, it is OK
1966 * to preserve that layout.
1967 * - When converting a RAID5 to RAID6, leave it in
1968 * the XXX-6 layout, don't re-layout.
1969 */
1970 if (info.array.level == 6 && info.new_level == UnSet)
1971 info.new_layout = info.array.layout;
1972 else if (info.array.level == 5 && info.new_level == 6) {
1973 char l[40];
1974 strcpy(l, map_num(r5layout, info.array.layout));
1975 strcat(l, "-6");
1976 info.new_layout = map_name(r6layout, l);
1977 } else {
1978 pr_err("%s in only meaningful when reshaping"
1979 " to RAID6\n", s->layout_str);
1980 rv = 1;
1981 goto release;
1982 }
1983 } else {
1984 int l = info.new_level;
1985 if (l == UnSet)
1986 l = info.array.level;
1987 switch (l) {
1988 case 5:
1989 info.new_layout = map_name(r5layout, s->layout_str);
1990 break;
1991 case 6:
1992 info.new_layout = map_name(r6layout, s->layout_str);
1993 break;
1994 case 10:
1995 info.new_layout = parse_layout_10(s->layout_str);
1996 break;
1997 case LEVEL_FAULTY:
1998 info.new_layout = parse_layout_faulty(s->layout_str);
1999 break;
2000 default:
2001 pr_err("layout not meaningful"
2002 " with this level\n");
2003 rv = 1;
2004 goto release;
2005 }
2006 if (info.new_layout == UnSet) {
2007 pr_err("layout %s not understood"
2008 " for this level\n",
2009 s->layout_str);
2010 rv = 1;
2011 goto release;
2012 }
2013 }
2014
2015 if (array.level == LEVEL_FAULTY) {
2016 if (s->level != UnSet && s->level != array.level) {
2017 pr_err("cannot change level of Faulty device\n");
2018 rv =1 ;
2019 }
2020 if (s->chunk) {
2021 pr_err("cannot set chunksize of Faulty device\n");
2022 rv =1 ;
2023 }
2024 if (s->raiddisks && s->raiddisks != 1) {
2025 pr_err("cannot set raid_disks of Faulty device\n");
2026 rv =1 ;
2027 }
2028 if (s->layout_str) {
2029 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2030 dprintf("Cannot get array information.\n");
2031 goto release;
2032 }
2033 array.layout = info.new_layout;
2034 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2035 pr_err("failed to set new layout\n");
2036 rv = 1;
2037 } else if (c->verbose >= 0)
2038 printf("layout for %s set to %d\n",
2039 devname, array.layout);
2040 }
2041 } else if (array.level == LEVEL_CONTAINER) {
2042 /* This change is to be applied to every array in the
2043 * container. This is only needed when the metadata imposes
2044 * restraints of the various arrays in the container.
2045 * Currently we only know that IMSM requires all arrays
2046 * to have the same number of devices so changing the
2047 * number of devices (On-Line Capacity Expansion) must be
2048 * performed at the level of the container
2049 */
2050 rv = reshape_container(container, devname, -1, st, &info,
2051 c->force, c->backup_file, c->verbose, 0, 0);
2052 frozen = 0;
2053 } else {
2054 /* get spare devices from external metadata
2055 */
2056 if (st->ss->external) {
2057 struct mdinfo *info2;
2058
2059 info2 = st->ss->container_content(st, subarray);
2060 if (info2) {
2061 info.array.spare_disks =
2062 info2->array.spare_disks;
2063 sysfs_free(info2);
2064 }
2065 }
2066
2067 /* Impose these changes on a single array. First
2068 * check that the metadata is OK with the change. */
2069
2070 if (reshape_super(st, 0, info.new_level,
2071 info.new_layout, info.new_chunk,
2072 info.array.raid_disks, info.delta_disks,
2073 c->backup_file, devname, APPLY_METADATA_CHANGES,
2074 c->verbose)) {
2075 rv = 1;
2076 goto release;
2077 }
2078 sync_metadata(st);
2079 rv = reshape_array(container, fd, devname, st, &info, c->force,
2080 devlist, data_offset, c->backup_file, c->verbose,
2081 0, 0, 0);
2082 frozen = 0;
2083 }
2084 release:
2085 sysfs_free(sra);
2086 if (frozen > 0)
2087 unfreeze(st);
2088 return rv;
2089 }
2090
2091 /* verify_reshape_position()
2092 * Function checks if reshape position in metadata is not farther
2093 * than position in md.
2094 * Return value:
2095 * 0 : not valid sysfs entry
2096 * it can be caused by not started reshape, it should be started
2097 * by reshape array or raid0 array is before takeover
2098 * -1 : error, reshape position is obviously wrong
2099 * 1 : success, reshape progress correct or updated
2100 */
2101 static int verify_reshape_position(struct mdinfo *info, int level)
2102 {
2103 int ret_val = 0;
2104 char buf[40];
2105 int rv;
2106
2107 /* read sync_max, failure can mean raid0 array */
2108 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2109
2110 if (rv > 0) {
2111 char *ep;
2112 unsigned long long position = strtoull(buf, &ep, 0);
2113
2114 dprintf(Name": Read sync_max sysfs entry is: %s\n", buf);
2115 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2116 position *= get_data_disks(level,
2117 info->new_layout,
2118 info->array.raid_disks);
2119 if (info->reshape_progress < position) {
2120 dprintf("Corrected reshape progress (%llu) to "
2121 "md position (%llu)\n",
2122 info->reshape_progress, position);
2123 info->reshape_progress = position;
2124 ret_val = 1;
2125 } else if (info->reshape_progress > position) {
2126 pr_err("Fatal error: array "
2127 "reshape was not properly frozen "
2128 "(expected reshape position is %llu, "
2129 "but reshape progress is %llu.\n",
2130 position, info->reshape_progress);
2131 ret_val = -1;
2132 } else {
2133 dprintf("Reshape position in md and metadata "
2134 "are the same;");
2135 ret_val = 1;
2136 }
2137 }
2138 } else if (rv == 0) {
2139 /* for valid sysfs entry, 0-length content
2140 * should be indicated as error
2141 */
2142 ret_val = -1;
2143 }
2144
2145 return ret_val;
2146 }
2147
2148 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2149 char *devname, int delta_disks,
2150 unsigned long long data_offset,
2151 unsigned long long min)
2152 {
2153 struct mdinfo *sd;
2154 int dir = 0;
2155 int err = 0;
2156 unsigned long long before, after;
2157
2158 /* Need to find min space before and after so same is used
2159 * on all devices
2160 */
2161 before = UINT64_MAX;
2162 after = UINT64_MAX;
2163 for (sd = sra->devs; sd; sd = sd->next) {
2164 char *dn;
2165 int dfd;
2166 int rv;
2167 struct supertype *st2;
2168 struct mdinfo info2;
2169
2170 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2171 continue;
2172 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2173 dfd = dev_open(dn, O_RDONLY);
2174 if (dfd < 0) {
2175 pr_err("%s: cannot open component %s\n",
2176 devname, dn ? dn : "-unknown-");
2177 goto release;
2178 }
2179 st2 = dup_super(st);
2180 rv = st2->ss->load_super(st2,dfd, NULL);
2181 close(dfd);
2182 if (rv) {
2183 free(st2);
2184 pr_err("%s: cannot get superblock from %s\n",
2185 devname, dn);
2186 goto release;
2187 }
2188 st2->ss->getinfo_super(st2, &info2, NULL);
2189 st2->ss->free_super(st2);
2190 free(st2);
2191 if (info2.space_before == 0 &&
2192 info2.space_after == 0) {
2193 /* Metadata doesn't support data_offset changes */
2194 return 1;
2195 }
2196 if (before > info2.space_before)
2197 before = info2.space_before;
2198 if (after > info2.space_after)
2199 after = info2.space_after;
2200
2201 if (data_offset != INVALID_SECTORS) {
2202 if (dir == 0) {
2203 if (info2.data_offset == data_offset) {
2204 pr_err("%s: already has that data_offset\n",
2205 dn);
2206 goto release;
2207 }
2208 if (data_offset < info2.data_offset)
2209 dir = -1;
2210 else
2211 dir = 1;
2212 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2213 (data_offset >= info2.data_offset && dir == -1)) {
2214 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2215 dn);
2216 goto release;
2217 }
2218 }
2219 }
2220 if (before == UINT64_MAX)
2221 /* impossible really, there must be no devices */
2222 return 1;
2223
2224 for (sd = sra->devs; sd; sd = sd->next) {
2225 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2226
2227 struct mdinfo info2;
2228 if (delta_disks < 0) {
2229 /* Don't need any space as array is shrinking
2230 * just move data_offset up by min
2231 */
2232 if (data_offset == INVALID_SECTORS)
2233 info2.new_data_offset = info2.data_offset + min;
2234 else {
2235 if (data_offset < info2.data_offset + min) {
2236 pr_err("--data-offset too small for %s\n",
2237 dn);
2238 goto release;
2239 }
2240 info2.new_data_offset = data_offset;
2241 }
2242 } else if (delta_disks > 0) {
2243 /* need space before */
2244 if (before < min) {
2245 pr_err("Insufficient head-space for reshape on %s\n",
2246 dn);
2247 goto release;
2248 }
2249 if (data_offset == INVALID_SECTORS)
2250 info2.new_data_offset = info2.data_offset - min;
2251 else {
2252 if (data_offset > info2.data_offset - min) {
2253 pr_err("--data-offset too large for %s\n",
2254 dn);
2255 goto release;
2256 }
2257 info2.new_data_offset = data_offset;
2258 }
2259 } else {
2260 if (dir == 0) {
2261 /* can move up or down. If 'data_offset'
2262 * was set we would have already decided,
2263 * so just choose direction with most space.
2264 */
2265 if (info2.space_before > info2.space_after)
2266 dir = -1;
2267 else
2268 dir = 1;
2269 sysfs_set_str(sra, NULL, "reshape_direction",
2270 dir == 1 ? "backwards" : "forwards");
2271 }
2272
2273 switch (dir) {
2274 case 1: /* Increase data offset */
2275 if (after < min) {
2276 pr_err("Insufficient tail-space for reshape on %s\n",
2277 dn);
2278 goto release;
2279 }
2280 if (data_offset != INVALID_SECTORS &&
2281 data_offset < info2.data_offset + min) {
2282 pr_err("--data-offset too small on %s\n",
2283 dn);
2284 goto release;
2285 }
2286 if (data_offset != INVALID_SECTORS)
2287 info2.new_data_offset = data_offset;
2288 else {
2289 unsigned long long off = after / 2;
2290 off &= ~7ULL;
2291 if (off < min)
2292 off = min;
2293 info2.new_data_offset =
2294 info2.data_offset + off;
2295 }
2296 break;
2297 case -1: /* Decrease data offset */
2298 if (before < min) {
2299 pr_err("insufficient head-room on %s\n",
2300 dn);
2301 goto release;
2302 }
2303 if (data_offset != INVALID_SECTORS &&
2304 data_offset < info2.data_offset - min) {
2305 pr_err("--data-offset too small on %s\n",
2306 dn);
2307 goto release;
2308 }
2309 if (data_offset != INVALID_SECTORS)
2310 info2.new_data_offset = data_offset;
2311 else {
2312 unsigned long long off = before / 2;
2313 off &= ~7ULL;
2314 if (off < min)
2315 off = min;
2316 info2.new_data_offset =
2317 info2.data_offset - off;
2318 }
2319 break;
2320 }
2321 }
2322 if (sysfs_set_num(sra, sd, "new_offset",
2323 info2.new_data_offset) < 0) {
2324 err = errno;
2325 err = -1;
2326 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2327 pr_err("data-offset is too big for %s\n",
2328 dn);
2329 goto release;
2330 }
2331 if (sd == sra->devs &&
2332 (errno == ENOENT || errno == E2BIG))
2333 /* Early kernel, no 'new_offset' file,
2334 * or kernel doesn't like us.
2335 * For RAID5/6 this is not fatal
2336 */
2337 return 1;
2338 pr_err("Cannot set new_offset for %s\n",
2339 dn);
2340 break;
2341 }
2342 }
2343 return err;
2344 release:
2345 return -1;
2346 }
2347
2348 static int raid10_reshape(char *container, int fd, char *devname,
2349 struct supertype *st, struct mdinfo *info,
2350 struct reshape *reshape,
2351 unsigned long long data_offset,
2352 int force, int verbose)
2353 {
2354 /* Changing raid_disks, layout, chunksize or possibly
2355 * just data_offset for a RAID10.
2356 * We must always change data_offset. We change by at least
2357 * ->min_offset_change which is the largest of the old and new
2358 * chunk sizes.
2359 * If raid_disks is increasing, then data_offset must decrease
2360 * by at least this copy size.
2361 * If raid_disks is unchanged, data_offset must increase or
2362 * decrease by at least min_offset_change but preferably by much more.
2363 * We choose half of the available space.
2364 * If raid_disks is decreasing, data_offset must increase by
2365 * at least min_offset_change. To allow of this, component_size
2366 * must be decreased by the same amount.
2367 *
2368 * So we calculate the required minimum and direction, possibly
2369 * reduce the component_size, then iterate through the devices
2370 * and set the new_data_offset.
2371 * If that all works, we set chunk_size, layout, raid_disks, and start
2372 * 'reshape'
2373 */
2374 struct mdinfo *sra;
2375 unsigned long long min;
2376 int err = 0;
2377
2378 sra = sysfs_read(fd, NULL,
2379 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2380 );
2381 if (!sra) {
2382 pr_err("%s: Cannot get array details from sysfs\n",
2383 devname);
2384 goto release;
2385 }
2386 min = reshape->min_offset_change;
2387
2388 if (info->delta_disks)
2389 sysfs_set_str(sra, NULL, "reshape_direction",
2390 info->delta_disks < 0 ? "backwards" : "forwards");
2391 if (info->delta_disks < 0 &&
2392 info->space_after < min) {
2393 int rv = sysfs_set_num(sra, NULL, "component_size",
2394 (sra->component_size -
2395 min)/2);
2396 if (rv) {
2397 pr_err("cannot reduce component size\n");
2398 goto release;
2399 }
2400 }
2401 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2402 min);
2403 if (err == 1) {
2404 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2405 cont_err("supported on this kernel\n");
2406 err = -1;
2407 }
2408 if (err < 0)
2409 goto release;
2410
2411 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2412 err = errno;
2413 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2414 err = errno;
2415 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2416 info->array.raid_disks + info->delta_disks) < 0)
2417 err = errno;
2418 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2419 err = errno;
2420 if (err) {
2421 pr_err("Cannot set array shape for %s\n",
2422 devname);
2423 if (err == EBUSY &&
2424 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2425 cont_err(" Bitmap must be removed before"
2426 " shape can be changed\n");
2427 goto release;
2428 }
2429 sysfs_free(sra);
2430 return 0;
2431 release:
2432 sysfs_free(sra);
2433 return 1;
2434 }
2435
2436 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2437 {
2438 struct mdinfo *sra, *sd;
2439 /* Initialisation to silence compiler warning */
2440 unsigned long long min_space_before = 0, min_space_after = 0;
2441 int first = 1;
2442
2443 sra = sysfs_read(fd, NULL, GET_DEVS);
2444 if (!sra)
2445 return;
2446 for (sd = sra->devs; sd; sd = sd->next) {
2447 char *dn;
2448 int dfd;
2449 struct supertype *st2;
2450 struct mdinfo info2;
2451
2452 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2453 continue;
2454 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2455 dfd = dev_open(dn, O_RDONLY);
2456 if (dfd < 0)
2457 break;
2458 st2 = dup_super(st);
2459 if (st2->ss->load_super(st2,dfd, NULL)) {
2460 close(dfd);
2461 free(st2);
2462 break;
2463 }
2464 close(dfd);
2465 st2->ss->getinfo_super(st2, &info2, NULL);
2466 st2->ss->free_super(st2);
2467 free(st2);
2468 if (first ||
2469 min_space_before > info2.space_before)
2470 min_space_before = info2.space_before;
2471 if (first ||
2472 min_space_after > info2.space_after)
2473 min_space_after = info2.space_after;
2474 first = 0;
2475 }
2476 if (sd == NULL && !first) {
2477 info->space_after = min_space_after;
2478 info->space_before = min_space_before;
2479 }
2480 sysfs_free(sra);
2481 }
2482
2483 static void update_cache_size(char *container, struct mdinfo *sra,
2484 struct mdinfo *info,
2485 int disks, unsigned long long blocks)
2486 {
2487 /* Check that the internal stripe cache is
2488 * large enough, or it won't work.
2489 * It must hold at least 4 stripes of the larger
2490 * chunk size
2491 */
2492 unsigned long cache;
2493 cache = max(info->array.chunk_size, info->new_chunk);
2494 cache *= 4; /* 4 stripes minimum */
2495 cache /= 512; /* convert to sectors */
2496 /* make sure there is room for 'blocks' with a bit to spare */
2497 if (cache < 16 + blocks / disks)
2498 cache = 16 + blocks / disks;
2499 cache /= (4096/512); /* Covert from sectors to pages */
2500
2501 if (sra->cache_size < cache)
2502 subarray_set_num(container, sra, "stripe_cache_size",
2503 cache+1);
2504 }
2505
2506 static int impose_reshape(struct mdinfo *sra,
2507 struct mdinfo *info,
2508 struct supertype *st,
2509 int fd,
2510 int restart,
2511 char *devname, char *container,
2512 struct reshape *reshape)
2513 {
2514 struct mdu_array_info_s array;
2515
2516 sra->new_chunk = info->new_chunk;
2517
2518 if (restart) {
2519 /* for external metadata checkpoint saved by mdmon can be lost
2520 * or missed /due to e.g. crash/. Check if md is not during
2521 * restart farther than metadata points to.
2522 * If so, this means metadata information is obsolete.
2523 */
2524 if (st->ss->external)
2525 verify_reshape_position(info, reshape->level);
2526 sra->reshape_progress = info->reshape_progress;
2527 } else {
2528 sra->reshape_progress = 0;
2529 if (reshape->after.data_disks < reshape->before.data_disks)
2530 /* start from the end of the new array */
2531 sra->reshape_progress = (sra->component_size
2532 * reshape->after.data_disks);
2533 }
2534
2535 ioctl(fd, GET_ARRAY_INFO, &array);
2536 if (info->array.chunk_size == info->new_chunk &&
2537 reshape->before.layout == reshape->after.layout &&
2538 st->ss->external == 0) {
2539 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2540 array.raid_disks = reshape->after.data_disks + reshape->parity;
2541 if (!restart &&
2542 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2543 int err = errno;
2544
2545 pr_err("Cannot set device shape for %s: %s\n",
2546 devname, strerror(errno));
2547
2548 if (err == EBUSY &&
2549 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2550 cont_err("Bitmap must be removed before"
2551 " shape can be changed\n");
2552
2553 goto release;
2554 }
2555 } else if (!restart) {
2556 /* set them all just in case some old 'new_*' value
2557 * persists from some earlier problem.
2558 */
2559 int err = 0;
2560 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2561 err = errno;
2562 if (!err && sysfs_set_num(sra, NULL, "layout",
2563 reshape->after.layout) < 0)
2564 err = errno;
2565 if (!err && subarray_set_num(container, sra, "raid_disks",
2566 reshape->after.data_disks +
2567 reshape->parity) < 0)
2568 err = errno;
2569 if (err) {
2570 pr_err("Cannot set device shape for %s\n",
2571 devname);
2572
2573 if (err == EBUSY &&
2574 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2575 cont_err("Bitmap must be removed before"
2576 " shape can be changed\n");
2577 goto release;
2578 }
2579 }
2580 return 0;
2581 release:
2582 return -1;
2583 }
2584
2585 static int reshape_array(char *container, int fd, char *devname,
2586 struct supertype *st, struct mdinfo *info,
2587 int force, struct mddev_dev *devlist,
2588 unsigned long long data_offset,
2589 char *backup_file, int verbose, int forked,
2590 int restart, int freeze_reshape)
2591 {
2592 struct reshape reshape;
2593 int spares_needed;
2594 char *msg;
2595 int orig_level = UnSet;
2596 int odisks;
2597 int delayed;
2598
2599 struct mdu_array_info_s array;
2600 char *c;
2601
2602 struct mddev_dev *dv;
2603 int added_disks;
2604
2605 int *fdlist = NULL;
2606 unsigned long long *offsets = NULL;
2607 int d;
2608 int nrdisks;
2609 int err;
2610 unsigned long blocks;
2611 unsigned long long array_size;
2612 int done;
2613 struct mdinfo *sra = NULL;
2614
2615 /* when reshaping a RAID0, the component_size might be zero.
2616 * So try to fix that up.
2617 */
2618 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2619 dprintf("Cannot get array information.\n");
2620 goto release;
2621 }
2622 if (array.level == 0 && info->component_size == 0) {
2623 get_dev_size(fd, NULL, &array_size);
2624 info->component_size = array_size / array.raid_disks;
2625 }
2626
2627 if (array.level == 10)
2628 /* Need space_after info */
2629 get_space_after(fd, st, info);
2630
2631 if (info->reshape_active) {
2632 int new_level = info->new_level;
2633 info->new_level = UnSet;
2634 if (info->delta_disks > 0)
2635 info->array.raid_disks -= info->delta_disks;
2636 msg = analyse_change(info, &reshape);
2637 info->new_level = new_level;
2638 if (info->delta_disks > 0)
2639 info->array.raid_disks += info->delta_disks;
2640 if (!restart)
2641 /* Make sure the array isn't read-only */
2642 ioctl(fd, RESTART_ARRAY_RW, 0);
2643 } else
2644 msg = analyse_change(info, &reshape);
2645 if (msg) {
2646 pr_err("%s\n", msg);
2647 goto release;
2648 }
2649 if (restart &&
2650 (reshape.level != info->array.level ||
2651 reshape.before.layout != info->array.layout ||
2652 reshape.before.data_disks + reshape.parity
2653 != info->array.raid_disks - max(0, info->delta_disks))) {
2654 pr_err("reshape info is not in native format -"
2655 " cannot continue.\n");
2656 goto release;
2657 }
2658
2659 if (st->ss->external && restart && (info->reshape_progress == 0)) {
2660 /* When reshape is restarted from '0', very begin of array
2661 * it is possible that for external metadata reshape and array
2662 * configuration doesn't happen.
2663 * Check if md has the same opinion, and reshape is restarted
2664 * from 0. If so, this is regular reshape start after reshape
2665 * switch in metadata to next array only.
2666 */
2667 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2668 (info->reshape_progress == 0))
2669 restart = 0;
2670 }
2671 if (restart) {
2672 /* reshape already started. just skip to monitoring the reshape */
2673 if (reshape.backup_blocks == 0)
2674 return 0;
2675 if (restart & RESHAPE_NO_BACKUP)
2676 return 0;
2677 goto started;
2678 }
2679 /* The container is frozen but the array may not be.
2680 * So freeze the array so spares don't get put to the wrong use
2681 * FIXME there should probably be a cleaner separation between
2682 * freeze_array and freeze_container.
2683 */
2684 sysfs_freeze_array(info);
2685 /* Check we have enough spares to not be degraded */
2686 added_disks = 0;
2687 for (dv = devlist; dv ; dv=dv->next)
2688 added_disks++;
2689 spares_needed = max(reshape.before.data_disks,
2690 reshape.after.data_disks)
2691 + reshape.parity - array.raid_disks;
2692
2693 if (!force &&
2694 info->new_level > 1 && info->array.level > 1 &&
2695 spares_needed > info->array.spare_disks + added_disks) {
2696 pr_err("Need %d spare%s to avoid degraded array,"
2697 " and only have %d.\n"
2698 " Use --force to over-ride this check.\n",
2699 spares_needed,
2700 spares_needed == 1 ? "" : "s",
2701 info->array.spare_disks + added_disks);
2702 goto release;
2703 }
2704 /* Check we have enough spares to not fail */
2705 spares_needed = max(reshape.before.data_disks,
2706 reshape.after.data_disks)
2707 - array.raid_disks;
2708 if ((info->new_level > 1 || info->new_level == 0) &&
2709 spares_needed > info->array.spare_disks +added_disks) {
2710 pr_err("Need %d spare%s to create working array,"
2711 " and only have %d.\n",
2712 spares_needed,
2713 spares_needed == 1 ? "" : "s",
2714 info->array.spare_disks + added_disks);
2715 goto release;
2716 }
2717
2718 if (reshape.level != array.level) {
2719 char *c = map_num(pers, reshape.level);
2720 int err;
2721 if (c == NULL)
2722 goto release;
2723
2724 err = sysfs_set_str(info, NULL, "level", c);
2725 if (err) {
2726 err = errno;
2727 pr_err("%s: could not set level to %s\n",
2728 devname, c);
2729 if (err == EBUSY &&
2730 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2731 cont_err("Bitmap must be removed"
2732 " before level can be changed\n");
2733 goto release;
2734 }
2735 if (verbose >= 0)
2736 pr_err("level of %s changed to %s\n",
2737 devname, c);
2738 orig_level = array.level;
2739 sysfs_freeze_array(info);
2740
2741 if (reshape.level > 0 && st->ss->external) {
2742 /* make sure mdmon is aware of the new level */
2743 if (mdmon_running(container))
2744 flush_mdmon(container);
2745
2746 if (!mdmon_running(container))
2747 start_mdmon(container);
2748 ping_monitor(container);
2749 if (mdmon_running(container) &&
2750 st->update_tail == NULL)
2751 st->update_tail = &st->updates;
2752 }
2753 }
2754 /* ->reshape_super might have chosen some spares from the
2755 * container that it wants to be part of the new array.
2756 * We can collect them with ->container_content and give
2757 * them to the kernel.
2758 */
2759 if (st->ss->reshape_super && st->ss->container_content) {
2760 char *subarray = strchr(info->text_version+1, '/')+1;
2761 struct mdinfo *info2 =
2762 st->ss->container_content(st, subarray);
2763 struct mdinfo *d;
2764
2765 if (info2) {
2766 sysfs_init(info2, fd, st->devnm);
2767 /* When increasing number of devices, we need to set
2768 * new raid_disks before adding these, or they might
2769 * be rejected.
2770 */
2771 if (reshape.backup_blocks &&
2772 reshape.after.data_disks > reshape.before.data_disks)
2773 subarray_set_num(container, info2, "raid_disks",
2774 reshape.after.data_disks +
2775 reshape.parity);
2776 for (d = info2->devs; d; d = d->next) {
2777 if (d->disk.state == 0 &&
2778 d->disk.raid_disk >= 0) {
2779 /* This is a spare that wants to
2780 * be part of the array.
2781 */
2782 add_disk(fd, st, info2, d);
2783 }
2784 }
2785 sysfs_free(info2);
2786 }
2787 }
2788 /* We might have been given some devices to add to the
2789 * array. Now that the array has been changed to the right
2790 * level and frozen, we can safely add them.
2791 */
2792 if (devlist)
2793 Manage_subdevs(devname, fd, devlist, verbose,
2794 0,NULL, 0);
2795
2796
2797 if (reshape.backup_blocks == 0 && data_offset)
2798 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
2799 if (reshape.backup_blocks == 0) {
2800 /* No restriping needed, but we might need to impose
2801 * some more changes: layout, raid_disks, chunk_size
2802 */
2803 /* read current array info */
2804 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2805 dprintf("Cannot get array information.\n");
2806 goto release;
2807 }
2808 /* compare current array info with new values and if
2809 * it is different update them to new */
2810 if (info->new_layout != UnSet &&
2811 info->new_layout != array.layout) {
2812 array.layout = info->new_layout;
2813 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2814 pr_err("failed to set new layout\n");
2815 goto release;
2816 } else if (verbose >= 0)
2817 printf("layout for %s set to %d\n",
2818 devname, array.layout);
2819 }
2820 if (info->delta_disks != UnSet &&
2821 info->delta_disks != 0 &&
2822 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
2823 array.raid_disks += info->delta_disks;
2824 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2825 pr_err("failed to set raid disks\n");
2826 goto release;
2827 } else if (verbose >= 0) {
2828 printf("raid_disks for %s set to %d\n",
2829 devname, array.raid_disks);
2830 }
2831 }
2832 if (info->new_chunk != 0 &&
2833 info->new_chunk != array.chunk_size) {
2834 if (sysfs_set_num(info, NULL,
2835 "chunk_size", info->new_chunk) != 0) {
2836 pr_err("failed to set chunk size\n");
2837 goto release;
2838 } else if (verbose >= 0)
2839 printf("chunk size for %s set to %d\n",
2840 devname, array.chunk_size);
2841 }
2842 unfreeze(st);
2843 return 0;
2844 }
2845
2846 /*
2847 * There are three possibilities.
2848 * 1/ The array will shrink.
2849 * We need to ensure the reshape will pause before reaching
2850 * the 'critical section'. We also need to fork and wait for
2851 * that to happen. When it does we
2852 * suspend/backup/complete/unfreeze
2853 *
2854 * 2/ The array will not change size.
2855 * This requires that we keep a backup of a sliding window
2856 * so that we can restore data after a crash. So we need
2857 * to fork and monitor progress.
2858 * In future we will allow the data_offset to change, so
2859 * a sliding backup becomes unnecessary.
2860 *
2861 * 3/ The array will grow. This is relatively easy.
2862 * However the kernel's restripe routines will cheerfully
2863 * overwrite some early data before it is safe. So we
2864 * need to make a backup of the early parts of the array
2865 * and be ready to restore it if rebuild aborts very early.
2866 * For externally managed metadata, we still need a forked
2867 * child to monitor the reshape and suspend IO over the region
2868 * that is being reshaped.
2869 *
2870 * We backup data by writing it to one spare, or to a
2871 * file which was given on command line.
2872 *
2873 * In each case, we first make sure that storage is available
2874 * for the required backup.
2875 * Then we:
2876 * - request the shape change.
2877 * - fork to handle backup etc.
2878 */
2879 /* Check that we can hold all the data */
2880 get_dev_size(fd, NULL, &array_size);
2881 if (reshape.new_size < (array_size/512)) {
2882 pr_err("this change will reduce the size of the array.\n"
2883 " use --grow --array-size first to truncate array.\n"
2884 " e.g. mdadm --grow %s --array-size %llu\n",
2885 devname, reshape.new_size/2);
2886 goto release;
2887 }
2888
2889 if (array.level == 10) {
2890 /* Reshaping RAID10 does not require any data backup by
2891 * user-space. Instead it requires that the data_offset
2892 * is changed to avoid the need for backup.
2893 * So this is handled very separately
2894 */
2895 if (restart)
2896 /* Nothing to do. */
2897 return 0;
2898 return raid10_reshape(container, fd, devname, st, info,
2899 &reshape, data_offset,
2900 force, verbose);
2901 }
2902 sra = sysfs_read(fd, NULL,
2903 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2904 GET_CACHE);
2905 if (!sra) {
2906 pr_err("%s: Cannot get array details from sysfs\n",
2907 devname);
2908 goto release;
2909 }
2910
2911 if (!backup_file)
2912 switch(set_new_data_offset(sra, st, devname,
2913 reshape.after.data_disks - reshape.before.data_disks,
2914 data_offset,
2915 reshape.min_offset_change)) {
2916 case -1:
2917 goto release;
2918 case 0:
2919 /* Updated data_offset, so it's easy now */
2920 update_cache_size(container, sra, info,
2921 min(reshape.before.data_disks,
2922 reshape.after.data_disks),
2923 reshape.backup_blocks);
2924
2925 /* Right, everything seems fine. Let's kick things off.
2926 */
2927 sync_metadata(st);
2928
2929 if (impose_reshape(sra, info, st, fd, restart,
2930 devname, container, &reshape) < 0)
2931 goto release;
2932 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
2933 pr_err("Failed to initiate reshape!\n");
2934 goto release;
2935 }
2936
2937 return 0;
2938 case 1: /* Couldn't set data_offset, try the old way */
2939 if (data_offset != INVALID_SECTORS) {
2940 pr_err("Cannot update data_offset on this array\n");
2941 goto release;
2942 }
2943 break;
2944 }
2945
2946 started:
2947 /* Decide how many blocks (sectors) for a reshape
2948 * unit. The number we have so far is just a minimum
2949 */
2950 blocks = reshape.backup_blocks;
2951 if (reshape.before.data_disks ==
2952 reshape.after.data_disks) {
2953 /* Make 'blocks' bigger for better throughput, but
2954 * not so big that we reject it below.
2955 * Try for 16 megabytes
2956 */
2957 while (blocks * 32 < sra->component_size &&
2958 blocks < 16*1024*2)
2959 blocks *= 2;
2960 } else
2961 pr_err("Need to backup %luK of critical "
2962 "section..\n", blocks/2);
2963
2964 if (blocks >= sra->component_size/2) {
2965 pr_err("%s: Something wrong"
2966 " - reshape aborted\n",
2967 devname);
2968 goto release;
2969 }
2970
2971 /* Now we need to open all these devices so we can read/write.
2972 */
2973 nrdisks = max(reshape.before.data_disks,
2974 reshape.after.data_disks) + reshape.parity
2975 + sra->array.spare_disks;
2976 fdlist = xcalloc((1+nrdisks), sizeof(int));
2977 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
2978
2979 odisks = reshape.before.data_disks + reshape.parity;
2980 d = reshape_prepare_fdlist(devname, sra, odisks,
2981 nrdisks, blocks, backup_file,
2982 fdlist, offsets);
2983 if (d < 0) {
2984 goto release;
2985 }
2986 if ((st->ss->manage_reshape == NULL) ||
2987 (st->ss->recover_backup == NULL)) {
2988 if (backup_file == NULL) {
2989 if (reshape.after.data_disks <=
2990 reshape.before.data_disks) {
2991 pr_err("%s: Cannot grow - "
2992 "need backup-file\n", devname);
2993 goto release;
2994 } else if (sra->array.spare_disks == 0) {
2995 pr_err("%s: Cannot grow - "
2996 "need a spare or backup-file to backup "
2997 "critical section\n", devname);
2998 goto release;
2999 }
3000 } else {
3001 if (!reshape_open_backup_file(backup_file, fd, devname,
3002 (signed)blocks,
3003 fdlist+d, offsets+d,
3004 restart)) {
3005 goto release;
3006 }
3007 d++;
3008 }
3009 }
3010
3011 update_cache_size(container, sra, info,
3012 min(reshape.before.data_disks, reshape.after.data_disks),
3013 blocks);
3014
3015 /* Right, everything seems fine. Let's kick things off.
3016 * If only changing raid_disks, use ioctl, else use
3017 * sysfs.
3018 */
3019 sync_metadata(st);
3020
3021 if (impose_reshape(sra, info, st, fd, restart,
3022 devname, container, &reshape) < 0)
3023 goto release;
3024
3025 err = start_reshape(sra, restart, reshape.before.data_disks,
3026 reshape.after.data_disks);
3027 if (err) {
3028 pr_err("Cannot %s reshape for %s\n",
3029 restart ? "continue" : "start",
3030 devname);
3031 goto release;
3032 }
3033 if (restart)
3034 sysfs_set_str(sra, NULL, "array_state", "active");
3035 if (freeze_reshape) {
3036 free(fdlist);
3037 free(offsets);
3038 sysfs_free(sra);
3039 pr_err("Reshape has to be continued from"
3040 " location %llu when root filesystem has been mounted.\n",
3041 sra->reshape_progress);
3042 return 1;
3043 }
3044
3045 /* Now we just need to kick off the reshape and watch, while
3046 * handling backups of the data...
3047 * This is all done by a forked background process.
3048 */
3049 switch(forked ? 0 : fork()) {
3050 case -1:
3051 pr_err("Cannot run child to monitor reshape: %s\n",
3052 strerror(errno));
3053 abort_reshape(sra);
3054 goto release;
3055 default:
3056 free(fdlist);
3057 free(offsets);
3058 sysfs_free(sra);
3059 return 0;
3060 case 0:
3061 map_fork();
3062 break;
3063 }
3064
3065 /* If another array on the same devices is busy, the
3066 * reshape will wait for them. This would mean that
3067 * the first section that we suspend will stay suspended
3068 * for a long time. So check on that possibility
3069 * by looking for "DELAYED" in /proc/mdstat, and if found,
3070 * wait a while
3071 */
3072 do {
3073 struct mdstat_ent *mds, *m;
3074 delayed = 0;
3075 mds = mdstat_read(0, 0);
3076 for (m = mds; m; m = m->next)
3077 if (strcmp(m->devnm, sra->sys_name) == 0) {
3078 if (m->resync &&
3079 m->percent == RESYNC_DELAYED)
3080 delayed = 1;
3081 if (m->resync == 0)
3082 /* Haven't started the reshape thread
3083 * yet, wait a bit
3084 */
3085 delayed = 2;
3086 break;
3087 }
3088 free_mdstat(mds);
3089 if (delayed == 1 && get_linux_version() < 3007000) {
3090 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3091 " You might experience problems until other reshapes complete.\n");
3092 delayed = 0;
3093 }
3094 if (delayed)
3095 sleep(30 - (delayed-1) * 25);
3096 } while (delayed);
3097
3098 close(fd);
3099 if (check_env("MDADM_GROW_VERIFY"))
3100 fd = open(devname, O_RDONLY | O_DIRECT);
3101 else
3102 fd = -1;
3103 mlockall(MCL_FUTURE);
3104
3105 if (st->ss->external) {
3106 /* metadata handler takes it from here */
3107 done = st->ss->manage_reshape(
3108 fd, sra, &reshape, st, blocks,
3109 fdlist, offsets,
3110 d - odisks, fdlist+odisks,
3111 offsets+odisks);
3112 } else
3113 done = child_monitor(
3114 fd, sra, &reshape, st, blocks,
3115 fdlist, offsets,
3116 d - odisks, fdlist+odisks,
3117 offsets+odisks);
3118
3119 free(fdlist);
3120 free(offsets);
3121
3122 if (backup_file && done)
3123 unlink(backup_file);
3124 if (!done) {
3125 abort_reshape(sra);
3126 goto out;
3127 }
3128
3129 if (!st->ss->external &&
3130 !(reshape.before.data_disks != reshape.after.data_disks
3131 && info->custom_array_size) &&
3132 info->new_level == reshape.level &&
3133 !forked) {
3134 /* no need to wait for the reshape to finish as
3135 * there is nothing more to do.
3136 */
3137 sysfs_free(sra);
3138 exit(0);
3139 }
3140 wait_reshape(sra);
3141
3142 if (st->ss->external) {
3143 /* Re-load the metadata as much could have changed */
3144 int cfd = open_dev(st->container_devnm);
3145 if (cfd >= 0) {
3146 flush_mdmon(container);
3147 st->ss->free_super(st);
3148 st->ss->load_container(st, cfd, container);
3149 close(cfd);
3150 }
3151 }
3152
3153 /* set new array size if required customer_array_size is used
3154 * by this metadata.
3155 */
3156 if (reshape.before.data_disks !=
3157 reshape.after.data_disks &&
3158 info->custom_array_size)
3159 set_array_size(st, info, info->text_version);
3160
3161 if (info->new_level != reshape.level) {
3162
3163 c = map_num(pers, info->new_level);
3164 if (c) {
3165 err = sysfs_set_str(sra, NULL, "level", c);
3166 if (err)
3167 pr_err("%s: could not set level "
3168 "to %s\n", devname, c);
3169 }
3170 if (info->new_level == 0)
3171 st->update_tail = NULL;
3172 }
3173 out:
3174 sysfs_free(sra);
3175 if (forked)
3176 return 0;
3177 unfreeze(st);
3178 exit(0);
3179
3180 release:
3181 free(fdlist);
3182 free(offsets);
3183 if (orig_level != UnSet && sra) {
3184 c = map_num(pers, orig_level);
3185 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3186 pr_err("aborting level change\n");
3187 }
3188 sysfs_free(sra);
3189 if (!forked)
3190 unfreeze(st);
3191 return 1;
3192 }
3193
3194 /* mdfd handle is passed to be closed in child process (after fork).
3195 */
3196 int reshape_container(char *container, char *devname,
3197 int mdfd,
3198 struct supertype *st,
3199 struct mdinfo *info,
3200 int force,
3201 char *backup_file,
3202 int verbose, int restart, int freeze_reshape)
3203 {
3204 struct mdinfo *cc = NULL;
3205 int rv = restart;
3206 char last_devnm[32] = "";
3207
3208 /* component_size is not meaningful for a container,
3209 * so pass '0' meaning 'no change'
3210 */
3211 if (!restart &&
3212 reshape_super(st, 0, info->new_level,
3213 info->new_layout, info->new_chunk,
3214 info->array.raid_disks, info->delta_disks,
3215 backup_file, devname, APPLY_METADATA_CHANGES,
3216 verbose)) {
3217 unfreeze(st);
3218 return 1;
3219 }
3220
3221 sync_metadata(st);
3222
3223 /* ping monitor to be sure that update is on disk
3224 */
3225 ping_monitor(container);
3226
3227 switch (fork()) {
3228 case -1: /* error */
3229 perror("Cannot fork to complete reshape\n");
3230 unfreeze(st);
3231 return 1;
3232 default: /* parent */
3233 if (!freeze_reshape)
3234 printf(Name ": multi-array reshape continues"
3235 " in background\n");
3236 return 0;
3237 case 0: /* child */
3238 map_fork();
3239 break;
3240 }
3241
3242 /* close unused handle in child process
3243 */
3244 if (mdfd > -1)
3245 close(mdfd);
3246
3247 while(1) {
3248 /* For each member array with reshape_active,
3249 * we need to perform the reshape.
3250 * We pick the first array that needs reshaping and
3251 * reshape it. reshape_array() will re-read the metadata
3252 * so the next time through a different array should be
3253 * ready for reshape.
3254 * It is possible that the 'different' array will not
3255 * be assembled yet. In that case we simple exit.
3256 * When it is assembled, the mdadm which assembles it
3257 * will take over the reshape.
3258 */
3259 struct mdinfo *content;
3260 int fd;
3261 struct mdstat_ent *mdstat;
3262 char *adev;
3263 int devid;
3264
3265 sysfs_free(cc);
3266
3267 cc = st->ss->container_content(st, NULL);
3268
3269 for (content = cc; content ; content = content->next) {
3270 char *subarray;
3271 if (!content->reshape_active)
3272 continue;
3273
3274 subarray = strchr(content->text_version+1, '/')+1;
3275 mdstat = mdstat_by_subdev(subarray, container);
3276 if (!mdstat)
3277 continue;
3278 if (mdstat->active == 0) {
3279 pr_err("Skipping inactive array %s.\n",
3280 mdstat->devnm);
3281 free_mdstat(mdstat);
3282 mdstat = NULL;
3283 continue;
3284 }
3285 break;
3286 }
3287 if (!content)
3288 break;
3289
3290 devid = devnm2devid(mdstat->devnm);
3291 adev = map_dev(major(devid), minor(devid), 0);
3292 if (!adev)
3293 adev = content->text_version;
3294
3295 fd = open_dev(mdstat->devnm);
3296 if (fd < 0) {
3297 printf(Name ": Device %s cannot be opened for reshape.",
3298 adev);
3299 break;
3300 }
3301
3302 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3303 /* Do not allow for multiple reshape_array() calls for
3304 * the same array.
3305 * It can happen when reshape_array() returns without
3306 * error, when reshape is not finished (wrong reshape
3307 * starting/continuation conditions). Mdmon doesn't
3308 * switch to next array in container and reentry
3309 * conditions for the same array occur.
3310 * This is possibly interim until the behaviour of
3311 * reshape_array is resolved().
3312 */
3313 printf(Name ": Multiple reshape execution detected for "
3314 "device %s.", adev);
3315 close(fd);
3316 break;
3317 }
3318 strcpy(last_devnm, mdstat->devnm);
3319
3320 sysfs_init(content, fd, mdstat->devnm);
3321
3322 if (mdmon_running(container))
3323 flush_mdmon(container);
3324
3325 rv = reshape_array(container, fd, adev, st,
3326 content, force, NULL, 0ULL,
3327 backup_file, verbose, 1, restart,
3328 freeze_reshape);
3329 close(fd);
3330
3331 if (freeze_reshape) {
3332 sysfs_free(cc);
3333 exit(0);
3334 }
3335
3336 restart = 0;
3337 if (rv)
3338 break;
3339
3340 if (mdmon_running(container))
3341 flush_mdmon(container);
3342 }
3343 if (!rv)
3344 unfreeze(st);
3345 sysfs_free(cc);
3346 exit(0);
3347 }
3348
3349 /*
3350 * We run a child process in the background which performs the following
3351 * steps:
3352 * - wait for resync to reach a certain point
3353 * - suspend io to the following section
3354 * - backup that section
3355 * - allow resync to proceed further
3356 * - resume io
3357 * - discard the backup.
3358 *
3359 * When are combined in slightly different ways in the three cases.
3360 * Grow:
3361 * - suspend/backup/allow/wait/resume/discard
3362 * Shrink:
3363 * - allow/wait/suspend/backup/allow/wait/resume/discard
3364 * same-size:
3365 * - wait/resume/discard/suspend/backup/allow
3366 *
3367 * suspend/backup/allow always come together
3368 * wait/resume/discard do too.
3369 * For the same-size case we have two backups to improve flow.
3370 *
3371 */
3372
3373 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3374 unsigned long long backup_point,
3375 unsigned long long wait_point,
3376 unsigned long long *suspend_point,
3377 unsigned long long *reshape_completed)
3378 {
3379 /* This function is called repeatedly by the reshape manager.
3380 * It determines how much progress can safely be made and allows
3381 * that progress.
3382 * - 'info' identifies the array and particularly records in
3383 * ->reshape_progress the metadata's knowledge of progress
3384 * This is a sector offset from the start of the array
3385 * of the next array block to be relocated. This number
3386 * may increase from 0 or decrease from array_size, depending
3387 * on the type of reshape that is happening.
3388 * Note that in contrast, 'sync_completed' is a block count of the
3389 * reshape so far. It gives the distance between the start point
3390 * (head or tail of device) and the next place that data will be
3391 * written. It always increases.
3392 * - 'reshape' is the structure created by analyse_change
3393 * - 'backup_point' shows how much the metadata manager has backed-up
3394 * data. For reshapes with increasing progress, it is the next address
3395 * to be backed up, previous addresses have been backed-up. For
3396 * decreasing progress, it is the earliest address that has been
3397 * backed up - later address are also backed up.
3398 * So addresses between reshape_progress and backup_point are
3399 * backed up providing those are in the 'correct' order.
3400 * - 'wait_point' is an array address. When reshape_completed
3401 * passes this point, progress_reshape should return. It might
3402 * return earlier if it determines that ->reshape_progress needs
3403 * to be updated or further backup is needed.
3404 * - suspend_point is maintained by progress_reshape and the caller
3405 * should not touch it except to initialise to zero.
3406 * It is an array address and it only increases in 2.6.37 and earlier.
3407 * This makes it difficult to handle reducing reshapes with
3408 * external metadata.
3409 * However: it is similar to backup_point in that it records the
3410 * other end of a suspended region from reshape_progress.
3411 * it is moved to extend the region that is safe to backup and/or
3412 * reshape
3413 * - reshape_completed is read from sysfs and returned. The caller
3414 * should copy this into ->reshape_progress when it has reason to
3415 * believe that the metadata knows this, and any backup outside this
3416 * has been erased.
3417 *
3418 * Return value is:
3419 * 1 if more data from backup_point - but only as far as suspend_point,
3420 * should be backed up
3421 * 0 if things are progressing smoothly
3422 * -1 if the reshape is finished because it is all done,
3423 * -2 if the reshape is finished due to an error.
3424 */
3425
3426 int advancing = (reshape->after.data_disks
3427 >= reshape->before.data_disks);
3428 unsigned long long need_backup; /* All data between start of array and
3429 * here will at some point need to
3430 * be backed up.
3431 */
3432 unsigned long long read_offset, write_offset;
3433 unsigned long long write_range;
3434 unsigned long long max_progress, target, completed;
3435 unsigned long long array_size = (info->component_size
3436 * reshape->before.data_disks);
3437 int fd;
3438 char buf[20];
3439
3440 /* First, we unsuspend any region that is now known to be safe.
3441 * If suspend_point is on the 'wrong' side of reshape_progress, then
3442 * we don't have or need suspension at the moment. This is true for
3443 * native metadata when we don't need to back-up.
3444 */
3445 if (advancing) {
3446 if (info->reshape_progress <= *suspend_point)
3447 sysfs_set_num(info, NULL, "suspend_lo",
3448 info->reshape_progress);
3449 } else {
3450 /* Note: this won't work in 2.6.37 and before.
3451 * Something somewhere should make sure we don't need it!
3452 */
3453 if (info->reshape_progress >= *suspend_point)
3454 sysfs_set_num(info, NULL, "suspend_hi",
3455 info->reshape_progress);
3456 }
3457
3458 /* Now work out how far it is safe to progress.
3459 * If the read_offset for ->reshape_progress is less than
3460 * 'blocks' beyond the write_offset, we can only progress as far
3461 * as a backup.
3462 * Otherwise we can progress until the write_offset for the new location
3463 * reaches (within 'blocks' of) the read_offset at the current location.
3464 * However that region must be suspended unless we are using native
3465 * metadata.
3466 * If we need to suspend more, we limit it to 128M per device, which is
3467 * rather arbitrary and should be some time-based calculation.
3468 */
3469 read_offset = info->reshape_progress / reshape->before.data_disks;
3470 write_offset = info->reshape_progress / reshape->after.data_disks;
3471 write_range = info->new_chunk/512;
3472 if (reshape->before.data_disks == reshape->after.data_disks)
3473 need_backup = array_size;
3474 else
3475 need_backup = reshape->backup_blocks;
3476 if (advancing) {
3477 if (read_offset < write_offset + write_range)
3478 max_progress = backup_point;
3479 else
3480 max_progress =
3481 read_offset *
3482 reshape->after.data_disks;
3483 } else {
3484 if (read_offset > write_offset - write_range)
3485 /* Can only progress as far as has been backed up,
3486 * which must be suspended */
3487 max_progress = backup_point;
3488 else if (info->reshape_progress <= need_backup)
3489 max_progress = backup_point;
3490 else {
3491 if (info->array.major_version >= 0)
3492 /* Can progress until backup is needed */
3493 max_progress = need_backup;
3494 else {
3495 /* Can progress until metadata update is required */
3496 max_progress =
3497 read_offset *
3498 reshape->after.data_disks;
3499 /* but data must be suspended */
3500 if (max_progress < *suspend_point)
3501 max_progress = *suspend_point;
3502 }
3503 }
3504 }
3505
3506 /* We know it is safe to progress to 'max_progress' providing
3507 * it is suspended or we are using native metadata.
3508 * Consider extending suspend_point 128M per device if it
3509 * is less than 64M per device beyond reshape_progress.
3510 * But always do a multiple of 'blocks'
3511 * FIXME this is too big - it takes to long to complete
3512 * this much.
3513 */
3514 target = 64*1024*2 * min(reshape->before.data_disks,
3515 reshape->after.data_disks);
3516 target /= reshape->backup_blocks;
3517 if (target < 2)
3518 target = 2;
3519 target *= reshape->backup_blocks;
3520
3521 /* For externally managed metadata we always need to suspend IO to
3522 * the area being reshaped so we regularly push suspend_point forward.
3523 * For native metadata we only need the suspend if we are going to do
3524 * a backup.
3525 */
3526 if (advancing) {
3527 if ((need_backup > info->reshape_progress
3528 || info->array.major_version < 0) &&
3529 *suspend_point < info->reshape_progress + target) {
3530 if (need_backup < *suspend_point + 2 * target)
3531 *suspend_point = need_backup;
3532 else if (*suspend_point + 2 * target < array_size)
3533 *suspend_point += 2 * target;
3534 else
3535 *suspend_point = array_size;
3536 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3537 if (max_progress > *suspend_point)
3538 max_progress = *suspend_point;
3539 }
3540 } else {
3541 if (info->array.major_version >= 0) {
3542 /* Only need to suspend when about to backup */
3543 if (info->reshape_progress < need_backup * 2 &&
3544 *suspend_point > 0) {
3545 *suspend_point = 0;
3546 sysfs_set_num(info, NULL, "suspend_lo", 0);
3547 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3548 }
3549 } else {
3550 /* Need to suspend continually */
3551 if (info->reshape_progress < *suspend_point)
3552 *suspend_point = info->reshape_progress;
3553 if (*suspend_point + target < info->reshape_progress)
3554 /* No need to move suspend region yet */;
3555 else {
3556 if (*suspend_point >= 2 * target)
3557 *suspend_point -= 2 * target;
3558 else
3559 *suspend_point = 0;
3560 sysfs_set_num(info, NULL, "suspend_lo",
3561 *suspend_point);
3562 }
3563 if (max_progress < *suspend_point)
3564 max_progress = *suspend_point;
3565 }
3566 }
3567
3568 /* now set sync_max to allow that progress. sync_max, like
3569 * sync_completed is a count of sectors written per device, so
3570 * we find the difference between max_progress and the start point,
3571 * and divide that by after.data_disks to get a sync_max
3572 * number.
3573 * At the same time we convert wait_point to a similar number
3574 * for comparing against sync_completed.
3575 */
3576 /* scale down max_progress to per_disk */
3577 max_progress /= reshape->after.data_disks;
3578 /* Round to chunk size as some kernels give an erroneously high number */
3579 max_progress /= info->new_chunk/512;
3580 max_progress *= info->new_chunk/512;
3581 /* And round to old chunk size as the kernel wants that */
3582 max_progress /= info->array.chunk_size/512;
3583 max_progress *= info->array.chunk_size/512;
3584 /* Limit progress to the whole device */
3585 if (max_progress > info->component_size)
3586 max_progress = info->component_size;
3587 wait_point /= reshape->after.data_disks;
3588 if (!advancing) {
3589 /* switch from 'device offset' to 'processed block count' */
3590 max_progress = info->component_size - max_progress;
3591 wait_point = info->component_size - wait_point;
3592 }
3593
3594 sysfs_set_num(info, NULL, "sync_max", max_progress);
3595
3596 /* Now wait. If we have already reached the point that we were
3597 * asked to wait to, don't wait at all, else wait for any change.
3598 * We need to select on 'sync_completed' as that is the place that
3599 * notifications happen, but we are really interested in
3600 * 'reshape_position'
3601 */
3602 fd = sysfs_get_fd(info, NULL, "sync_completed");
3603 if (fd < 0)
3604 goto check_progress;
3605
3606 if (sysfs_fd_get_ll(fd, &completed) < 0)
3607 goto check_progress;
3608
3609 while (completed < max_progress && completed < wait_point) {
3610 /* Check that sync_action is still 'reshape' to avoid
3611 * waiting forever on a dead array
3612 */
3613 char action[20];
3614 fd_set rfds;
3615 if (sysfs_get_str(info, NULL, "sync_action",
3616 action, 20) <= 0 ||
3617 strncmp(action, "reshape", 7) != 0)
3618 break;
3619 /* Some kernels reset 'sync_completed' to zero
3620 * before setting 'sync_action' to 'idle'.
3621 * So we need these extra tests.
3622 */
3623 if (completed == 0 && advancing
3624 && info->reshape_progress > 0)
3625 break;
3626 if (completed == 0 && !advancing
3627 && info->reshape_progress < (info->component_size
3628 * reshape->after.data_disks))
3629 break;
3630 FD_ZERO(&rfds);
3631 FD_SET(fd, &rfds);
3632 select(fd+1, NULL, NULL, &rfds, NULL);
3633 if (sysfs_fd_get_ll(fd, &completed) < 0)
3634 goto check_progress;
3635 }
3636 /* Some kernels reset 'sync_completed' to zero,
3637 * we need to have real point we are in md
3638 */
3639 if (completed == 0)
3640 completed = max_progress;
3641
3642 /* some kernels can give an incorrectly high 'completed' number */
3643 completed /= (info->new_chunk/512);
3644 completed *= (info->new_chunk/512);
3645 /* Convert 'completed' back in to a 'progress' number */
3646 completed *= reshape->after.data_disks;
3647 if (!advancing) {
3648 completed = info->component_size * reshape->after.data_disks
3649 - completed;
3650 }
3651 *reshape_completed = completed;
3652
3653 close(fd);
3654
3655 /* We return the need_backup flag. Caller will decide
3656 * how much - a multiple of ->backup_blocks up to *suspend_point
3657 */
3658 if (advancing)
3659 return need_backup > info->reshape_progress;
3660 else
3661 return need_backup >= info->reshape_progress;
3662
3663 check_progress:
3664 /* if we couldn't read a number from sync_completed, then
3665 * either the reshape did complete, or it aborted.
3666 * We can tell which by checking for 'none' in reshape_position.
3667 * If it did abort, then it might immediately restart if it
3668 * it was just a device failure that leaves us degraded but
3669 * functioning.
3670 */
3671 strcpy(buf, "hi");
3672 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3673 || strncmp(buf, "none", 4) != 0) {
3674 /* The abort might only be temporary. Wait up to 10
3675 * seconds for fd to contain a valid number again.
3676 */
3677 struct timeval tv;
3678 int rv = -2;
3679 tv.tv_sec = 10;
3680 tv.tv_usec = 0;
3681 while (fd >= 0 && rv < 0 && tv.tv_sec > 0) {
3682 fd_set rfds;
3683 FD_ZERO(&rfds);
3684 FD_SET(fd, &rfds);
3685 if (select(fd+1, NULL, NULL, &rfds, &tv) != 1)
3686 break;
3687 switch (sysfs_fd_get_ll(fd, &completed)) {
3688 case 0:
3689 /* all good again */
3690 rv = 1;
3691 break;
3692 case -2: /* read error - abort */
3693 tv.tv_sec = 0;
3694 break;
3695 }
3696 }
3697 if (fd >= 0)
3698 close(fd);
3699 return rv; /* abort */
3700 } else {
3701 /* Maybe racing with array shutdown - check state */
3702 if (fd >= 0)
3703 close(fd);
3704 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3705 || strncmp(buf, "inactive", 8) == 0
3706 || strncmp(buf, "clear",5) == 0)
3707 return -2; /* abort */
3708 return -1; /* complete */
3709 }
3710 }
3711
3712 /* FIXME return status is never checked */
3713 static int grow_backup(struct mdinfo *sra,
3714 unsigned long long offset, /* per device */
3715 unsigned long stripes, /* per device, in old chunks */
3716 int *sources, unsigned long long *offsets,
3717 int disks, int chunk, int level, int layout,
3718 int dests, int *destfd, unsigned long long *destoffsets,
3719 int part, int *degraded,
3720 char *buf)
3721 {
3722 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3723 * to storage 'destfd' (offset 'destoffsets'), after first
3724 * suspending IO. Then allow resync to continue
3725 * over the suspended section.
3726 * Use part 'part' of the backup-super-block.
3727 */
3728 int odata = disks;
3729 int rv = 0;
3730 int i;
3731 unsigned long long ll;
3732 int new_degraded;
3733 //printf("offset %llu\n", offset);
3734 if (level >= 4)
3735 odata--;
3736 if (level == 6)
3737 odata--;
3738
3739 /* Check that array hasn't become degraded, else we might backup the wrong data */
3740 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3741 return -1; /* FIXME this error is ignored */
3742 new_degraded = (int)ll;
3743 if (new_degraded != *degraded) {
3744 /* check each device to ensure it is still working */
3745 struct mdinfo *sd;
3746 for (sd = sra->devs ; sd ; sd = sd->next) {
3747 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3748 continue;
3749 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3750 char sbuf[20];
3751 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3752 strstr(sbuf, "faulty") ||
3753 strstr(sbuf, "in_sync") == NULL) {
3754 /* this device is dead */
3755 sd->disk.state = (1<<MD_DISK_FAULTY);
3756 if (sd->disk.raid_disk >= 0 &&
3757 sources[sd->disk.raid_disk] >= 0) {
3758 close(sources[sd->disk.raid_disk]);
3759 sources[sd->disk.raid_disk] = -1;
3760 }
3761 }
3762 }
3763 }
3764 *degraded = new_degraded;
3765 }
3766 if (part) {
3767 bsb.arraystart2 = __cpu_to_le64(offset * odata);
3768 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
3769 } else {
3770 bsb.arraystart = __cpu_to_le64(offset * odata);
3771 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
3772 }
3773 if (part)
3774 bsb.magic[15] = '2';
3775 for (i = 0; i < dests; i++)
3776 if (part)
3777 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
3778 else
3779 lseek64(destfd[i], destoffsets[i], 0);
3780
3781 rv = save_stripes(sources, offsets,
3782 disks, chunk, level, layout,
3783 dests, destfd,
3784 offset*512*odata, stripes * chunk * odata,
3785 buf);
3786
3787 if (rv)
3788 return rv;
3789 bsb.mtime = __cpu_to_le64(time(0));
3790 for (i = 0; i < dests; i++) {
3791 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3792
3793 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3794 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3795 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3796 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3797
3798 rv = -1;
3799 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
3800 != destoffsets[i] - 4096)
3801 break;
3802 if (write(destfd[i], &bsb, 512) != 512)
3803 break;
3804 if (destoffsets[i] > 4096) {
3805 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
3806 destoffsets[i]+stripes*chunk*odata)
3807 break;
3808 if (write(destfd[i], &bsb, 512) != 512)
3809 break;
3810 }
3811 fsync(destfd[i]);
3812 rv = 0;
3813 }
3814
3815 return rv;
3816 }
3817
3818 /* in 2.6.30, the value reported by sync_completed can be
3819 * less that it should be by one stripe.
3820 * This only happens when reshape hits sync_max and pauses.
3821 * So allow wait_backup to either extent sync_max further
3822 * than strictly necessary, or return before the
3823 * sync has got quite as far as we would really like.
3824 * This is what 'blocks2' is for.
3825 * The various caller give appropriate values so that
3826 * every works.
3827 */
3828 /* FIXME return value is often ignored */
3829 static int forget_backup(int dests, int *destfd,
3830 unsigned long long *destoffsets,
3831 int part)
3832 {
3833 /*
3834 * Erase backup 'part' (which is 0 or 1)
3835 */
3836 int i;
3837 int rv;
3838
3839 if (part) {
3840 bsb.arraystart2 = __cpu_to_le64(0);
3841 bsb.length2 = __cpu_to_le64(0);
3842 } else {
3843 bsb.arraystart = __cpu_to_le64(0);
3844 bsb.length = __cpu_to_le64(0);
3845 }
3846 bsb.mtime = __cpu_to_le64(time(0));
3847 rv = 0;
3848 for (i = 0; i < dests; i++) {
3849 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
3850 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
3851 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
3852 bsb.sb_csum2 = bsb_csum((char*)&bsb,
3853 ((char*)&bsb.sb_csum2)-((char*)&bsb));
3854 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
3855 destoffsets[i]-4096)
3856 rv = -1;
3857 if (rv == 0 &&
3858 write(destfd[i], &bsb, 512) != 512)
3859 rv = -1;
3860 fsync(destfd[i]);
3861 }
3862 return rv;
3863 }
3864
3865 static void fail(char *msg)
3866 {
3867 int rv;
3868 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
3869 rv |= (write(2, "\n", 1) != 1);
3870 exit(rv ? 1 : 2);
3871 }
3872
3873 static char *abuf, *bbuf;
3874 static unsigned long long abuflen;
3875 static void validate(int afd, int bfd, unsigned long long offset)
3876 {
3877 /* check that the data in the backup against the array.
3878 * This is only used for regression testing and should not
3879 * be used while the array is active
3880 */
3881 if (afd < 0)
3882 return;
3883 lseek64(bfd, offset - 4096, 0);
3884 if (read(bfd, &bsb2, 512) != 512)
3885 fail("cannot read bsb");
3886 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
3887 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
3888 fail("first csum bad");
3889 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
3890 fail("magic is bad");
3891 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
3892 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
3893 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
3894 fail("second csum bad");
3895
3896 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
3897 fail("devstart is wrong");
3898
3899 if (bsb2.length) {
3900 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
3901
3902 if (abuflen < len) {
3903 free(abuf);
3904 free(bbuf);
3905 abuflen = len;
3906 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
3907 posix_memalign((void**)&bbuf, 4096, abuflen)) {
3908 abuflen = 0;
3909 /* just stop validating on mem-alloc failure */
3910 return;
3911 }
3912 }
3913
3914 lseek64(bfd, offset, 0);
3915 if ((unsigned long long)read(bfd, bbuf, len) != len) {
3916 //printf("len %llu\n", len);
3917 fail("read first backup failed");
3918 }
3919 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
3920 if ((unsigned long long)read(afd, abuf, len) != len)
3921 fail("read first from array failed");
3922 if (memcmp(bbuf, abuf, len) != 0) {
3923 #if 0
3924 int i;
3925 printf("offset=%llu len=%llu\n",
3926 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
3927 for (i=0; i<len; i++)
3928 if (bbuf[i] != abuf[i]) {
3929 printf("first diff byte %d\n", i);
3930 break;
3931 }
3932 #endif
3933 fail("data1 compare failed");
3934 }
3935 }
3936 if (bsb2.length2) {
3937 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
3938
3939 if (abuflen < len) {
3940 free(abuf);
3941 free(bbuf);
3942 abuflen = len;
3943 abuf = xmalloc(abuflen);
3944 bbuf = xmalloc(abuflen);
3945 }
3946
3947 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
3948 if ((unsigned long long)read(bfd, bbuf, len) != len)
3949 fail("read second backup failed");
3950 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
3951 if ((unsigned long long)read(afd, abuf, len) != len)
3952 fail("read second from array failed");
3953 if (memcmp(bbuf, abuf, len) != 0)
3954 fail("data2 compare failed");
3955 }
3956 }
3957
3958 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
3959 struct supertype *st, unsigned long blocks,
3960 int *fds, unsigned long long *offsets,
3961 int dests, int *destfd, unsigned long long *destoffsets)
3962 {
3963 /* Monitor a reshape where backup is being performed using
3964 * 'native' mechanism - either to a backup file, or
3965 * to some space in a spare.
3966 */
3967 char *buf;
3968 int degraded = -1;
3969 unsigned long long speed;
3970 unsigned long long suspend_point, array_size;
3971 unsigned long long backup_point, wait_point;
3972 unsigned long long reshape_completed;
3973 int done = 0;
3974 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
3975 int part = 0; /* The next part of the backup area to fill. It may already
3976 * be full, so we need to check */
3977 int level = reshape->level;
3978 int layout = reshape->before.layout;
3979 int data = reshape->before.data_disks;
3980 int disks = reshape->before.data_disks + reshape->parity;
3981 int chunk = sra->array.chunk_size;
3982 struct mdinfo *sd;
3983 unsigned long stripes;
3984 int uuid[4];
3985
3986 /* set up the backup-super-block. This requires the
3987 * uuid from the array.
3988 */
3989 /* Find a superblock */
3990 for (sd = sra->devs; sd; sd = sd->next) {
3991 char *dn;
3992 int devfd;
3993 int ok;
3994 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3995 continue;
3996 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
3997 devfd = dev_open(dn, O_RDONLY);
3998 if (devfd < 0)
3999 continue;
4000 ok = st->ss->load_super(st, devfd, NULL);
4001 close(devfd);
4002 if (ok == 0)
4003 break;
4004 }
4005 if (!sd) {
4006 pr_err("Cannot find a superblock\n");
4007 return 0;
4008 }
4009
4010 memset(&bsb, 0, 512);
4011 memcpy(bsb.magic, "md_backup_data-1", 16);
4012 st->ss->uuid_from_super(st, uuid);
4013 memcpy(bsb.set_uuid, uuid, 16);
4014 bsb.mtime = __cpu_to_le64(time(0));
4015 bsb.devstart2 = blocks;
4016
4017 stripes = blocks / (sra->array.chunk_size/512) /
4018 reshape->before.data_disks;
4019
4020 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4021 /* Don't start the 'reshape' */
4022 return 0;
4023 if (reshape->before.data_disks == reshape->after.data_disks) {
4024 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4025 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4026 }
4027
4028 if (increasing) {
4029 array_size = sra->component_size * reshape->after.data_disks;
4030 backup_point = sra->reshape_progress;
4031 suspend_point = 0;
4032 } else {
4033 array_size = sra->component_size * reshape->before.data_disks;
4034 backup_point = reshape->backup_blocks;
4035 suspend_point = array_size;
4036 }
4037
4038 while (!done) {
4039 int rv;
4040
4041 /* Want to return as soon the oldest backup slot can
4042 * be released as that allows us to start backing up
4043 * some more, providing suspend_point has been
4044 * advanced, which it should have.
4045 */
4046 if (increasing) {
4047 wait_point = array_size;
4048 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4049 wait_point = (__le64_to_cpu(bsb.arraystart) +
4050 __le64_to_cpu(bsb.length));
4051 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4052 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4053 __le64_to_cpu(bsb.length2));
4054 } else {
4055 wait_point = 0;
4056 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4057 wait_point = __le64_to_cpu(bsb.arraystart);
4058 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4059 wait_point = __le64_to_cpu(bsb.arraystart2);
4060 }
4061
4062 rv = progress_reshape(sra, reshape,
4063 backup_point, wait_point,
4064 &suspend_point, &reshape_completed);
4065 /* external metadata would need to ping_monitor here */
4066 sra->reshape_progress = reshape_completed;
4067
4068 /* Clear any backup region that is before 'here' */
4069 if (increasing) {
4070 if (__le64_to_cpu(bsb.length) > 0 &&
4071 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4072 __le64_to_cpu(bsb.length)))
4073 forget_backup(dests, destfd,
4074 destoffsets, 0);
4075 if (__le64_to_cpu(bsb.length2) > 0 &&
4076 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4077 __le64_to_cpu(bsb.length2)))
4078 forget_backup(dests, destfd,
4079 destoffsets, 1);
4080 } else {
4081 if (__le64_to_cpu(bsb.length) > 0 &&
4082 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4083 forget_backup(dests, destfd,
4084 destoffsets, 0);
4085 if (__le64_to_cpu(bsb.length2) > 0 &&
4086 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4087 forget_backup(dests, destfd,
4088 destoffsets, 1);
4089 }
4090
4091 if (rv < 0) {
4092 if (rv == -1)
4093 done = 1;
4094 break;
4095 }
4096 if (rv == 0 && increasing && !st->ss->external) {
4097 /* No longer need to monitor this reshape */
4098 done = 1;
4099 break;
4100 }
4101
4102 while (rv) {
4103 unsigned long long offset;
4104 unsigned long actual_stripes;
4105 /* Need to backup some data.
4106 * If 'part' is not used and the desired
4107 * backup size is suspended, do a backup,
4108 * then consider the next part.
4109 */
4110 /* Check that 'part' is unused */
4111 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4112 break;
4113 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4114 break;
4115
4116 offset = backup_point / data;
4117 actual_stripes = stripes;
4118 if (increasing) {
4119 if (offset + actual_stripes * (chunk/512) >
4120 sra->component_size)
4121 actual_stripes = ((sra->component_size - offset)
4122 / (chunk/512));
4123 if (offset + actual_stripes * (chunk/512) >
4124 suspend_point/data)
4125 break;
4126 } else {
4127 if (offset < actual_stripes * (chunk/512))
4128 actual_stripes = offset / (chunk/512);
4129 offset -= actual_stripes * (chunk/512);
4130 if (offset < suspend_point/data)
4131 break;
4132 }
4133 if (actual_stripes == 0)
4134 break;
4135 grow_backup(sra, offset, actual_stripes,
4136 fds, offsets,
4137 disks, chunk, level, layout,
4138 dests, destfd, destoffsets,
4139 part, &degraded, buf);
4140 validate(afd, destfd[0], destoffsets[0]);
4141 /* record where 'part' is up to */
4142 part = !part;
4143 if (increasing)
4144 backup_point += actual_stripes * (chunk/512) * data;
4145 else
4146 backup_point -= actual_stripes * (chunk/512) * data;
4147 }
4148 }
4149
4150 /* FIXME maybe call progress_reshape one more time instead */
4151 abort_reshape(sra); /* remove any remaining suspension */
4152 if (reshape->before.data_disks == reshape->after.data_disks)
4153 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4154 free(buf);
4155 return done;
4156 }
4157
4158 /*
4159 * If any spare contains md_back_data-1 which is recent wrt mtime,
4160 * write that data into the array and update the super blocks with
4161 * the new reshape_progress
4162 */
4163 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4164 char *backup_file, int verbose)
4165 {
4166 int i, j;
4167 int old_disks;
4168 unsigned long long *offsets;
4169 unsigned long long nstripe, ostripe;
4170 int ndata, odata;
4171
4172 odata = info->array.raid_disks - info->delta_disks - 1;
4173 if (info->array.level == 6) odata--; /* number of data disks */
4174 ndata = info->array.raid_disks - 1;
4175 if (info->new_level == 6) ndata--;
4176
4177 old_disks = info->array.raid_disks - info->delta_disks;
4178
4179 if (info->delta_disks <= 0)
4180 /* Didn't grow, so the backup file must have
4181 * been used
4182 */
4183 old_disks = cnt;
4184 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4185 struct mdinfo dinfo;
4186 int fd;
4187 int bsbsize;
4188 char *devname, namebuf[20];
4189 unsigned long long lo, hi;
4190
4191 /* This was a spare and may have some saved data on it.
4192 * Load the superblock, find and load the
4193 * backup_super_block.
4194 * If either fail, go on to next device.
4195 * If the backup contains no new info, just return
4196 * else restore data and update all superblocks
4197 */
4198 if (i == old_disks-1) {
4199 fd = open(backup_file, O_RDONLY);
4200 if (fd<0) {
4201 pr_err("backup file %s inaccessible: %s\n",
4202 backup_file, strerror(errno));
4203 continue;
4204 }
4205 devname = backup_file;
4206 } else {
4207 fd = fdlist[i];
4208 if (fd < 0)
4209 continue;
4210 if (st->ss->load_super(st, fd, NULL))
4211 continue;
4212
4213 st->ss->getinfo_super(st, &dinfo, NULL);
4214 st->ss->free_super(st);
4215
4216 if (lseek64(fd,
4217 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4218 0) < 0) {
4219 pr_err("Cannot seek on device %d\n", i);
4220 continue; /* Cannot seek */
4221 }
4222 sprintf(namebuf, "device-%d", i);
4223 devname = namebuf;
4224 }
4225 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4226 if (verbose)
4227 pr_err("Cannot read from %s\n", devname);
4228 continue; /* Cannot read */
4229 }
4230 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4231 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4232 if (verbose)
4233 pr_err("No backup metadata on %s\n", devname);
4234 continue;
4235 }
4236 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4237 if (verbose)
4238 pr_err("Bad backup-metadata checksum on %s\n", devname);
4239 continue; /* bad checksum */
4240 }
4241 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4242 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4243 if (verbose)
4244 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4245 continue; /* Bad second checksum */
4246 }
4247 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4248 if (verbose)
4249 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4250 continue; /* Wrong uuid */
4251 }
4252
4253 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4254 * sometimes they aren't... So allow considerable flexability in matching, and allow
4255 * this test to be overridden by an environment variable.
4256 */
4257 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4258 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4259 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4260 pr_err("accepting backup with timestamp %lu "
4261 "for array with timestamp %lu\n",
4262 (unsigned long)__le64_to_cpu(bsb.mtime),
4263 (unsigned long)info->array.utime);
4264 } else {
4265 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4266 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4267 continue; /* time stamp is too bad */
4268 }
4269 }
4270
4271 if (bsb.magic[15] == '1') {
4272 if (bsb.length == 0)
4273 continue;
4274 if (info->delta_disks >= 0) {
4275 /* reshape_progress is increasing */
4276 if (__le64_to_cpu(bsb.arraystart)
4277 + __le64_to_cpu(bsb.length)
4278 < info->reshape_progress) {
4279 nonew:
4280 if (verbose)
4281 pr_err("backup-metadata found on %s but is not needed\n", devname);
4282 continue; /* No new data here */
4283 }
4284 } else {
4285 /* reshape_progress is decreasing */
4286 if (__le64_to_cpu(bsb.arraystart) >=
4287 info->reshape_progress)
4288 goto nonew; /* No new data here */
4289 }
4290 } else {
4291 if (bsb.length == 0 && bsb.length2 == 0)
4292 continue;
4293 if (info->delta_disks >= 0) {
4294 /* reshape_progress is increasing */
4295 if ((__le64_to_cpu(bsb.arraystart)
4296 + __le64_to_cpu(bsb.length)
4297 < info->reshape_progress)
4298 &&
4299 (__le64_to_cpu(bsb.arraystart2)
4300 + __le64_to_cpu(bsb.length2)
4301 < info->reshape_progress))
4302 goto nonew; /* No new data here */
4303 } else {
4304 /* reshape_progress is decreasing */
4305 if (__le64_to_cpu(bsb.arraystart) >=
4306 info->reshape_progress &&
4307 __le64_to_cpu(bsb.arraystart2) >=
4308 info->reshape_progress)
4309 goto nonew; /* No new data here */
4310 }
4311 }
4312 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4313 second_fail:
4314 if (verbose)
4315 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4316 devname);
4317 continue; /* Cannot seek */
4318 }
4319 /* There should be a duplicate backup superblock 4k before here */
4320 if (lseek64(fd, -4096, 1) < 0 ||
4321 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4322 goto second_fail; /* Cannot find leading superblock */
4323 if (bsb.magic[15] == '1')
4324 bsbsize = offsetof(struct mdp_backup_super, pad1);
4325 else
4326 bsbsize = offsetof(struct mdp_backup_super, pad);
4327 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4328 goto second_fail; /* Cannot find leading superblock */
4329
4330 /* Now need the data offsets for all devices. */
4331 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4332 for(j=0; j<info->array.raid_disks; j++) {
4333 if (fdlist[j] < 0)
4334 continue;
4335 if (st->ss->load_super(st, fdlist[j], NULL))
4336 /* FIXME should be this be an error */
4337 continue;
4338 st->ss->getinfo_super(st, &dinfo, NULL);
4339 st->ss->free_super(st);
4340 offsets[j] = dinfo.data_offset * 512;
4341 }
4342 printf(Name ": restoring critical section\n");
4343
4344 if (restore_stripes(fdlist, offsets,
4345 info->array.raid_disks,
4346 info->new_chunk,
4347 info->new_level,
4348 info->new_layout,
4349 fd, __le64_to_cpu(bsb.devstart)*512,
4350 __le64_to_cpu(bsb.arraystart)*512,
4351 __le64_to_cpu(bsb.length)*512, NULL)) {
4352 /* didn't succeed, so giveup */
4353 if (verbose)
4354 pr_err("Error restoring backup from %s\n",
4355 devname);
4356 free(offsets);
4357 return 1;
4358 }
4359
4360 if (bsb.magic[15] == '2' &&
4361 restore_stripes(fdlist, offsets,
4362 info->array.raid_disks,
4363 info->new_chunk,
4364 info->new_level,
4365 info->new_layout,
4366 fd, __le64_to_cpu(bsb.devstart)*512 +
4367 __le64_to_cpu(bsb.devstart2)*512,
4368 __le64_to_cpu(bsb.arraystart2)*512,
4369 __le64_to_cpu(bsb.length2)*512, NULL)) {
4370 /* didn't succeed, so giveup */
4371 if (verbose)
4372 pr_err("Error restoring second backup from %s\n",
4373 devname);
4374 free(offsets);
4375 return 1;
4376 }
4377
4378 free(offsets);
4379
4380 /* Ok, so the data is restored. Let's update those superblocks. */
4381
4382 lo = hi = 0;
4383 if (bsb.length) {
4384 lo = __le64_to_cpu(bsb.arraystart);
4385 hi = lo + __le64_to_cpu(bsb.length);
4386 }
4387 if (bsb.magic[15] == '2' && bsb.length2) {
4388 unsigned long long lo1, hi1;
4389 lo1 = __le64_to_cpu(bsb.arraystart2);
4390 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4391 if (lo == hi) {
4392 lo = lo1;
4393 hi = hi1;
4394 } else if (lo < lo1)
4395 hi = hi1;
4396 else
4397 lo = lo1;
4398 }
4399 if (lo < hi &&
4400 (info->reshape_progress < lo ||
4401 info->reshape_progress > hi))
4402 /* backup does not affect reshape_progress*/ ;
4403 else if (info->delta_disks >= 0) {
4404 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4405 __le64_to_cpu(bsb.length);
4406 if (bsb.magic[15] == '2') {
4407 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4408 __le64_to_cpu(bsb.length2);
4409 if (p2 > info->reshape_progress)
4410 info->reshape_progress = p2;
4411 }
4412 } else {
4413 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4414 if (bsb.magic[15] == '2') {
4415 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4416 if (p2 < info->reshape_progress)
4417 info->reshape_progress = p2;
4418 }
4419 }
4420 for (j=0; j<info->array.raid_disks; j++) {
4421 if (fdlist[j] < 0)
4422 continue;
4423 if (st->ss->load_super(st, fdlist[j], NULL))
4424 continue;
4425 st->ss->getinfo_super(st, &dinfo, NULL);
4426 dinfo.reshape_progress = info->reshape_progress;
4427 st->ss->update_super(st, &dinfo,
4428 "_reshape_progress",
4429 NULL,0, 0, NULL);
4430 st->ss->store_super(st, fdlist[j]);
4431 st->ss->free_super(st);
4432 }
4433 return 0;
4434 }
4435 /* Didn't find any backup data, try to see if any
4436 * was needed.
4437 */
4438 if (info->delta_disks < 0) {
4439 /* When shrinking, the critical section is at the end.
4440 * So see if we are before the critical section.
4441 */
4442 unsigned long long first_block;
4443 nstripe = ostripe = 0;
4444 first_block = 0;
4445 while (ostripe >= nstripe) {
4446 ostripe += info->array.chunk_size / 512;
4447 first_block = ostripe * odata;
4448 nstripe = first_block / ndata / (info->new_chunk/512) *
4449 (info->new_chunk/512);
4450 }
4451
4452 if (info->reshape_progress >= first_block)
4453 return 0;
4454 }
4455 if (info->delta_disks > 0) {
4456 /* See if we are beyond the critical section. */
4457 unsigned long long last_block;
4458 nstripe = ostripe = 0;
4459 last_block = 0;
4460 while (nstripe >= ostripe) {
4461 nstripe += info->new_chunk / 512;
4462 last_block = nstripe * ndata;
4463 ostripe = last_block / odata / (info->array.chunk_size/512) *
4464 (info->array.chunk_size/512);
4465 }
4466
4467 if (info->reshape_progress >= last_block)
4468 return 0;
4469 }
4470 /* needed to recover critical section! */
4471 if (verbose)
4472 pr_err("Failed to find backup of critical section\n");
4473 return 1;
4474 }
4475
4476 int Grow_continue_command(char *devname, int fd,
4477 char *backup_file, int verbose)
4478 {
4479 int ret_val = 0;
4480 struct supertype *st = NULL;
4481 struct mdinfo *content = NULL;
4482 struct mdinfo array;
4483 char *subarray = NULL;
4484 struct mdinfo *cc = NULL;
4485 struct mdstat_ent *mdstat = NULL;
4486 int cfd = -1;
4487 int fd2 = -1;
4488
4489 dprintf("Grow continue from command line called for %s\n",
4490 devname);
4491
4492 st = super_by_fd(fd, &subarray);
4493 if (!st || !st->ss) {
4494 pr_err("Unable to determine metadata format for %s\n",
4495 devname);
4496 return 1;
4497 }
4498 dprintf("Grow continue is run for ");
4499 if (st->ss->external == 0) {
4500 int d;
4501 dprintf("native array (%s)\n", devname);
4502 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4503 pr_err("%s is not an active md array -"
4504 " aborting\n", devname);
4505 ret_val = 1;
4506 goto Grow_continue_command_exit;
4507 }
4508 content = &array;
4509 /* Need to load a superblock.
4510 * FIXME we should really get what we need from
4511 * sysfs
4512 */
4513 for (d = 0; d < MAX_DISKS; d++) {
4514 mdu_disk_info_t disk;
4515 char *dv;
4516 int err;
4517 disk.number = d;
4518 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4519 continue;
4520 if (disk.major == 0 && disk.minor == 0)
4521 continue;
4522 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4523 continue;
4524 dv = map_dev(disk.major, disk.minor, 1);
4525 if (!dv)
4526 continue;
4527 fd2 = dev_open(dv, O_RDONLY);
4528 if (fd2 < 0)
4529 continue;
4530 err = st->ss->load_super(st, fd2, NULL);
4531 close(fd2);
4532 if (err)
4533 continue;
4534 break;
4535 }
4536 if (d == MAX_DISKS) {
4537 pr_err("Unable to load metadata for %s\n",
4538 devname);
4539 ret_val = 1;
4540 goto Grow_continue_command_exit;
4541 }
4542 st->ss->getinfo_super(st, content, NULL);
4543 } else {
4544 char *container;
4545
4546 if (subarray) {
4547 dprintf("subarray (%s)\n", subarray);
4548 container = st->container_devnm;
4549 cfd = open_dev_excl(st->container_devnm);
4550 } else {
4551 container = st->devnm;
4552 close(fd);
4553 cfd = open_dev_excl(st->devnm);
4554 dprintf("container (%s)\n", container);
4555 fd = cfd;
4556 }
4557 if (cfd < 0) {
4558 pr_err("Unable to open container "
4559 "for %s\n", devname);
4560 ret_val = 1;
4561 goto Grow_continue_command_exit;
4562 }
4563
4564 /* find in container array under reshape
4565 */
4566 ret_val = st->ss->load_container(st, cfd, NULL);
4567 if (ret_val) {
4568 pr_err("Cannot read superblock for %s\n",
4569 devname);
4570 ret_val = 1;
4571 goto Grow_continue_command_exit;
4572 }
4573
4574 cc = st->ss->container_content(st, subarray);
4575 for (content = cc; content ; content = content->next) {
4576 char *array;
4577 int allow_reshape = 1;
4578
4579 if (content->reshape_active == 0)
4580 continue;
4581 /* The decision about array or container wide
4582 * reshape is taken in Grow_continue based
4583 * content->reshape_active state, therefore we
4584 * need to check_reshape based on
4585 * reshape_active and subarray name
4586 */
4587 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4588 allow_reshape = 0;
4589 if (content->reshape_active == CONTAINER_RESHAPE &&
4590 (content->array.state
4591 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4592 allow_reshape = 0;
4593
4594 if (!allow_reshape) {
4595 pr_err("cannot continue reshape of an array"
4596 " in container with unsupported"
4597 " metadata: %s(%s)\n",
4598 devname, container);
4599 ret_val = 1;
4600 goto Grow_continue_command_exit;
4601 }
4602
4603 array = strchr(content->text_version+1, '/')+1;
4604 mdstat = mdstat_by_subdev(array, container);
4605 if (!mdstat)
4606 continue;
4607 if (mdstat->active == 0) {
4608 pr_err("Skipping inactive array %s.\n",
4609 mdstat->devnm);
4610 free_mdstat(mdstat);
4611 mdstat = NULL;
4612 continue;
4613 }
4614 break;
4615 }
4616 if (!content) {
4617 pr_err("Unable to determine reshaped "
4618 "array for %s\n", devname);
4619 ret_val = 1;
4620 goto Grow_continue_command_exit;
4621 }
4622 fd2 = open_dev(mdstat->devnm);
4623 if (fd2 < 0) {
4624 pr_err("cannot open (%s)\n", mdstat->devnm);
4625 ret_val = 1;
4626 goto Grow_continue_command_exit;
4627 }
4628
4629 sysfs_init(content, fd2, mdstat->devnm);
4630
4631 /* start mdmon in case it is not running
4632 */
4633 if (!mdmon_running(container))
4634 start_mdmon(container);
4635 ping_monitor(container);
4636
4637 if (mdmon_running(container))
4638 st->update_tail = &st->updates;
4639 else {
4640 pr_err("No mdmon found. "
4641 "Grow cannot continue.\n");
4642 ret_val = 1;
4643 goto Grow_continue_command_exit;
4644 }
4645 }
4646
4647 /* verify that array under reshape is started from
4648 * correct position
4649 */
4650 if (verify_reshape_position(content, content->array.level) < 0) {
4651 ret_val = 1;
4652 goto Grow_continue_command_exit;
4653 }
4654
4655 /* continue reshape
4656 */
4657 ret_val = Grow_continue(fd, st, content, backup_file, 0);
4658
4659 Grow_continue_command_exit:
4660 if (fd2 > -1)
4661 close(fd2);
4662 if (cfd > -1)
4663 close(cfd);
4664 st->ss->free_super(st);
4665 free_mdstat(mdstat);
4666 sysfs_free(cc);
4667 free(subarray);
4668
4669 return ret_val;
4670 }
4671
4672 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4673 char *backup_file, int freeze_reshape)
4674 {
4675 int ret_val = 2;
4676
4677 if (!info->reshape_active)
4678 return ret_val;
4679
4680 if (st->ss->external) {
4681 int cfd = open_dev(st->container_devnm);
4682
4683 if (cfd < 0)
4684 return 1;
4685
4686 st->ss->load_container(st, cfd, st->container_devnm);
4687 close(cfd);
4688 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4689 st, info, 0, backup_file,
4690 0,
4691 1 | info->reshape_active,
4692 freeze_reshape);
4693 } else
4694 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4695 NULL, 0ULL, backup_file, 0, 0,
4696 1 | info->reshape_active,
4697 freeze_reshape);
4698
4699 return ret_val;
4700 }
Unnamed repository; edit this file 'description' to name the repository.